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 zero = 0;
  24 static int one = 1;
  25 static int ushort_max = USHRT_MAX;
  26 static int min_sndbuf = SOCK_MIN_SNDBUF;
  27 static int min_rcvbuf = SOCK_MIN_RCVBUF;
  28
  29 #ifdef CONFIG_RPS
  30 static int rps_sock_flow_sysctl(ctl_table *table, int write,
  31                                 void __user *buffer, size_t *lenp, loff_t *ppos)
  32 {
  33         unsigned int orig_size, size;
  34         int ret, i;
  35         ctl_table tmp = {
  36                 .data = &size,
  37                 .maxlen = sizeof(size),
  38                 .mode = table->mode
  39         };
  40         struct rps_sock_flow_table *orig_sock_table, *sock_table;
  41         static DEFINE_MUTEX(sock_flow_mutex);
  42
  43         mutex_lock(&sock_flow_mutex);
  44
  45         orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
  46                                         lockdep_is_held(&sock_flow_mutex));
  47         size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
  48
  49         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
  50
  51         if (write) {
  52                 if (size) {
  53                         if (size > 1<<30) {
  54                                 /* Enforce limit to prevent overflow */
  55                                 mutex_unlock(&sock_flow_mutex);
  56                                 return -EINVAL;
  57                         }
  58                         size = roundup_pow_of_two(size);
  59                         if (size != orig_size) {
  60                                 sock_table =
  61                                     vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
  62                                 if (!sock_table) {
  63                                         mutex_unlock(&sock_flow_mutex);
  64                                         return -ENOMEM;
  65                                 }
  66
  67                                 sock_table->mask = size - 1;
  68                         } else
  69                                 sock_table = orig_sock_table;
  70
  71                         for (i = 0; i < size; i++)
  72                                 sock_table->ents[i] = RPS_NO_CPU;
  73                 } else
  74                         sock_table = NULL;
  75
  76                 if (sock_table != orig_sock_table) {
  77                         rcu_assign_pointer(rps_sock_flow_table, sock_table);
  78                         if (sock_table)
  79                                 static_key_slow_inc(&rps_needed);
  80                         if (orig_sock_table) {
  81                                 static_key_slow_dec(&rps_needed);
  82                                 synchronize_rcu();
  83                                 vfree(orig_sock_table);
  84                         }
  85                 }
  86         }
  87
  88         mutex_unlock(&sock_flow_mutex);
  89
  90         return ret;
  91 }
  92 #endif /* CONFIG_RPS */
  93
  94 static struct ctl_table net_core_table[] = {
  95 #ifdef CONFIG_NET
  96         {
  97                 .procname       = "wmem_max",
  98                 .data           = &sysctl_wmem_max,
  99                 .maxlen         = sizeof(int),
 100                 .mode           = 0644,
 101                 .proc_handler   = proc_dointvec_minmax,
 102                 .extra1         = &min_sndbuf,
 103         },
 104         {
 105                 .procname       = "rmem_max",
 106                 .data           = &sysctl_rmem_max,
 107                 .maxlen         = sizeof(int),
 108                 .mode           = 0644,
 109                 .proc_handler   = proc_dointvec_minmax,
 110                 .extra1         = &min_rcvbuf,
 111         },
 112         {
 113                 .procname       = "wmem_default",
 114                 .data           = &sysctl_wmem_default,
 115                 .maxlen         = sizeof(int),
 116                 .mode           = 0644,
 117                 .proc_handler   = proc_dointvec_minmax,
 118                 .extra1         = &min_sndbuf,
 119         },
 120         {
 121                 .procname       = "rmem_default",
 122                 .data           = &sysctl_rmem_default,
 123                 .maxlen         = sizeof(int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &min_rcvbuf,
 127         },
 128         {
 129                 .procname       = "dev_weight",
 130                 .data           = &weight_p,
 131                 .maxlen         = sizeof(int),
 132                 .mode           = 0644,
 133                 .proc_handler   = proc_dointvec
 134         },
 135         {
 136                 .procname       = "netdev_max_backlog",
 137                 .data           = &netdev_max_backlog,
 138                 .maxlen         = sizeof(int),
 139                 .mode           = 0644,
 140                 .proc_handler   = proc_dointvec
 141         },
 142 #ifdef CONFIG_BPF_JIT
 143         {
 144                 .procname       = "bpf_jit_enable",
 145                 .data           = &bpf_jit_enable,
 146                 .maxlen         = sizeof(int),
 147                 .mode           = 0644,
 148                 .proc_handler   = proc_dointvec
 149         },
 150 #endif
 151         {
 152                 .procname       = "netdev_tstamp_prequeue",
 153                 .data           = &netdev_tstamp_prequeue,
 154                 .maxlen         = sizeof(int),
 155                 .mode           = 0644,
 156                 .proc_handler   = proc_dointvec
 157         },
 158         {
 159                 .procname       = "message_cost",
 160                 .data           = &net_ratelimit_state.interval,
 161                 .maxlen         = sizeof(int),
 162                 .mode           = 0644,
 163                 .proc_handler   = proc_dointvec_jiffies,
 164         },
 165         {
 166                 .procname       = "message_burst",
 167                 .data           = &net_ratelimit_state.burst,
 168                 .maxlen         = sizeof(int),
 169                 .mode           = 0644,
 170                 .proc_handler   = proc_dointvec,
 171         },
 172         {
 173                 .procname       = "optmem_max",
 174                 .data           = &sysctl_optmem_max,
 175                 .maxlen         = sizeof(int),
 176                 .mode           = 0644,
 177                 .proc_handler   = proc_dointvec
 178         },
 179 #ifdef CONFIG_RPS
 180         {
 181                 .procname       = "rps_sock_flow_entries",
 182                 .maxlen         = sizeof(int),
 183                 .mode           = 0644,
 184                 .proc_handler   = rps_sock_flow_sysctl
 185         },
 186 #endif
 187 #endif /* CONFIG_NET */
 188         {
 189                 .procname       = "netdev_budget",
 190                 .data           = &netdev_budget,
 191                 .maxlen         = sizeof(int),
 192                 .mode           = 0644,
 193                 .proc_handler   = proc_dointvec
 194         },
 195         {
 196                 .procname       = "warnings",
 197                 .data           = &net_msg_warn,
 198                 .maxlen         = sizeof(int),
 199                 .mode           = 0644,
 200                 .proc_handler   = proc_dointvec
 201         },
 202         { }
 203 };
 204
 205 static struct ctl_table netns_core_table[] = {
 206         {
 207                 .procname       = "somaxconn",
 208                 .data           = &init_net.core.sysctl_somaxconn,
 209                 .maxlen         = sizeof(int),
 210                 .mode           = 0644,
 211                 .extra1         = &zero,
 212                 .extra2         = &ushort_max,
 213                 .proc_handler   = proc_dointvec_minmax
 214         },
 215         { }
 216 };
 217
 218 static __net_init int sysctl_core_net_init(struct net *net)
 219 {
 220         struct ctl_table *tbl;
 221
 222         net->core.sysctl_somaxconn = SOMAXCONN;
 223
 224         tbl = netns_core_table;
 225         if (!net_eq(net, &init_net)) {
 226                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 227                 if (tbl == NULL)
 228                         goto err_dup;
 229
 230                 tbl[0].data = &net->core.sysctl_somaxconn;
 231
 232                 /* Don't export any sysctls to unprivileged users */
 233                 if (net->user_ns != &init_user_ns) {
 234                         tbl[0].procname = NULL;
 235                 }
 236         }
 237
 238         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 239         if (net->core.sysctl_hdr == NULL)
 240                 goto err_reg;
 241
 242         return 0;
 243
 244 err_reg:
 245         if (tbl != netns_core_table)
 246                 kfree(tbl);
 247 err_dup:
 248         return -ENOMEM;
 249 }
 250
 251 static __net_exit void sysctl_core_net_exit(struct net *net)
 252 {
 253         struct ctl_table *tbl;
 254
 255         tbl = net->core.sysctl_hdr->ctl_table_arg;
 256         unregister_net_sysctl_table(net->core.sysctl_hdr);
 257         BUG_ON(tbl == netns_core_table);
 258         kfree(tbl);
 259 }
 260
 261 static __net_initdata struct pernet_operations sysctl_core_ops = {
 262         .init = sysctl_core_net_init,
 263         .exit = sysctl_core_net_exit,
 264 };
 265
 266 static __init int sysctl_core_init(void)
 267 {
 268         register_net_sysctl(&init_net, "net/core", net_core_table);
 269         return register_pernet_subsys(&sysctl_core_ops);
 270 }
 271
 272 fs_initcall(sysctl_core_init);