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/busy_poll.h>
  23 #include <net/pkt_sched.h>
  24
  25 static int zero = 0;
  26 static int one = 1;
  27 static int ushort_max = USHRT_MAX;
  28
  29 #ifdef CONFIG_RPS
  30 static int rps_sock_flow_sysctl(struct 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         struct 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 #ifdef CONFIG_NET_FLOW_LIMIT
  95 static DEFINE_MUTEX(flow_limit_update_mutex);
  96
  97 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
  98                                  void __user *buffer, size_t *lenp,
  99                                  loff_t *ppos)
 100 {
 101         struct sd_flow_limit *cur;
 102         struct softnet_data *sd;
 103         cpumask_var_t mask;
 104         int i, len, ret = 0;
 105
 106         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 107                 return -ENOMEM;
 108
 109         if (write) {
 110                 ret = cpumask_parse_user(buffer, *lenp, mask);
 111                 if (ret)
 112                         goto done;
 113
 114                 mutex_lock(&flow_limit_update_mutex);
 115                 len = sizeof(*cur) + netdev_flow_limit_table_len;
 116                 for_each_possible_cpu(i) {
 117                         sd = &per_cpu(softnet_data, i);
 118                         cur = rcu_dereference_protected(sd->flow_limit,
 119                                      lockdep_is_held(&flow_limit_update_mutex));
 120                         if (cur && !cpumask_test_cpu(i, mask)) {
 121                                 RCU_INIT_POINTER(sd->flow_limit, NULL);
 122                                 synchronize_rcu();
 123                                 kfree(cur);
 124                         } else if (!cur && cpumask_test_cpu(i, mask)) {
 125                                 cur = kzalloc_node(len, GFP_KERNEL,
 126                                                    cpu_to_node(i));
 127                                 if (!cur) {
 128                                         /* not unwinding previous changes */
 129                                         ret = -ENOMEM;
 130                                         goto write_unlock;
 131                                 }
 132                                 cur->num_buckets = netdev_flow_limit_table_len;
 133                                 rcu_assign_pointer(sd->flow_limit, cur);
 134                         }
 135                 }
 136 write_unlock:
 137                 mutex_unlock(&flow_limit_update_mutex);
 138         } else {
 139                 char kbuf[128];
 140
 141                 if (*ppos || !*lenp) {
 142                         *lenp = 0;
 143                         goto done;
 144                 }
 145
 146                 cpumask_clear(mask);
 147                 rcu_read_lock();
 148                 for_each_possible_cpu(i) {
 149                         sd = &per_cpu(softnet_data, i);
 150                         if (rcu_dereference(sd->flow_limit))
 151                                 cpumask_set_cpu(i, mask);
 152                 }
 153                 rcu_read_unlock();
 154
 155                 len = min(sizeof(kbuf) - 1, *lenp);
 156                 len = cpumask_scnprintf(kbuf, len, mask);
 157                 if (!len) {
 158                         *lenp = 0;
 159                         goto done;
 160                 }
 161                 if (len < *lenp)
 162                         kbuf[len++] = '\n';
 163                 if (copy_to_user(buffer, kbuf, len)) {
 164                         ret = -EFAULT;
 165                         goto done;
 166                 }
 167                 *lenp = len;
 168                 *ppos += len;
 169         }
 170
 171 done:
 172         free_cpumask_var(mask);
 173         return ret;
 174 }
 175
 176 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
 177                                        void __user *buffer, size_t *lenp,
 178                                        loff_t *ppos)
 179 {
 180         unsigned int old, *ptr;
 181         int ret;
 182
 183         mutex_lock(&flow_limit_update_mutex);
 184
 185         ptr = table->data;
 186         old = *ptr;
 187         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 188         if (!ret && write && !is_power_of_2(*ptr)) {
 189                 *ptr = old;
 190                 ret = -EINVAL;
 191         }
 192
 193         mutex_unlock(&flow_limit_update_mutex);
 194         return ret;
 195 }
 196 #endif /* CONFIG_NET_FLOW_LIMIT */
 197
 198 #ifdef CONFIG_NET_SCHED
 199 static int set_default_qdisc(struct ctl_table *table, int write,
 200                              void __user *buffer, size_t *lenp, loff_t *ppos)
 201 {
 202         char id[IFNAMSIZ];
 203         struct ctl_table tbl = {
 204                 .data = id,
 205                 .maxlen = IFNAMSIZ,
 206         };
 207         int ret;
 208
 209         qdisc_get_default(id, IFNAMSIZ);
 210
 211         ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
 212         if (write && ret == 0)
 213                 ret = qdisc_set_default(id);
 214         return ret;
 215 }
 216 #endif
 217
 218 static struct ctl_table net_core_table[] = {
 219 #ifdef CONFIG_NET
 220         {
 221                 .procname       = "wmem_max",
 222                 .data           = &sysctl_wmem_max,
 223                 .maxlen         = sizeof(int),
 224                 .mode           = 0644,
 225                 .proc_handler   = proc_dointvec_minmax,
 226                 .extra1         = &one,
 227         },
 228         {
 229                 .procname       = "rmem_max",
 230                 .data           = &sysctl_rmem_max,
 231                 .maxlen         = sizeof(int),
 232                 .mode           = 0644,
 233                 .proc_handler   = proc_dointvec_minmax,
 234                 .extra1         = &one,
 235         },
 236         {
 237                 .procname       = "wmem_default",
 238                 .data           = &sysctl_wmem_default,
 239                 .maxlen         = sizeof(int),
 240                 .mode           = 0644,
 241                 .proc_handler   = proc_dointvec_minmax,
 242                 .extra1         = &one,
 243         },
 244         {
 245                 .procname       = "rmem_default",
 246                 .data           = &sysctl_rmem_default,
 247                 .maxlen         = sizeof(int),
 248                 .mode           = 0644,
 249                 .proc_handler   = proc_dointvec_minmax,
 250                 .extra1         = &one,
 251         },
 252         {
 253                 .procname       = "dev_weight",
 254                 .data           = &weight_p,
 255                 .maxlen         = sizeof(int),
 256                 .mode           = 0644,
 257                 .proc_handler   = proc_dointvec
 258         },
 259         {
 260                 .procname       = "netdev_max_backlog",
 261                 .data           = &netdev_max_backlog,
 262                 .maxlen         = sizeof(int),
 263                 .mode           = 0644,
 264                 .proc_handler   = proc_dointvec
 265         },
 266 #ifdef CONFIG_BPF_JIT
 267         {
 268                 .procname       = "bpf_jit_enable",
 269                 .data           = &bpf_jit_enable,
 270                 .maxlen         = sizeof(int),
 271                 .mode           = 0644,
 272                 .proc_handler   = proc_dointvec
 273         },
 274 #endif
 275         {
 276                 .procname       = "netdev_tstamp_prequeue",
 277                 .data           = &netdev_tstamp_prequeue,
 278                 .maxlen         = sizeof(int),
 279                 .mode           = 0644,
 280                 .proc_handler   = proc_dointvec
 281         },
 282         {
 283                 .procname       = "message_cost",
 284                 .data           = &net_ratelimit_state.interval,
 285                 .maxlen         = sizeof(int),
 286                 .mode           = 0644,
 287                 .proc_handler   = proc_dointvec_jiffies,
 288         },
 289         {
 290                 .procname       = "message_burst",
 291                 .data           = &net_ratelimit_state.burst,
 292                 .maxlen         = sizeof(int),
 293                 .mode           = 0644,
 294                 .proc_handler   = proc_dointvec,
 295         },
 296         {
 297                 .procname       = "optmem_max",
 298                 .data           = &sysctl_optmem_max,
 299                 .maxlen         = sizeof(int),
 300                 .mode           = 0644,
 301                 .proc_handler   = proc_dointvec
 302         },
 303 #ifdef CONFIG_RPS
 304         {
 305                 .procname       = "rps_sock_flow_entries",
 306                 .maxlen         = sizeof(int),
 307                 .mode           = 0644,
 308                 .proc_handler   = rps_sock_flow_sysctl
 309         },
 310 #endif
 311 #ifdef CONFIG_NET_FLOW_LIMIT
 312         {
 313                 .procname       = "flow_limit_cpu_bitmap",
 314                 .mode           = 0644,
 315                 .proc_handler   = flow_limit_cpu_sysctl
 316         },
 317         {
 318                 .procname       = "flow_limit_table_len",
 319                 .data           = &netdev_flow_limit_table_len,
 320                 .maxlen         = sizeof(int),
 321                 .mode           = 0644,
 322                 .proc_handler   = flow_limit_table_len_sysctl
 323         },
 324 #endif /* CONFIG_NET_FLOW_LIMIT */
 325 #ifdef CONFIG_NET_RX_BUSY_POLL
 326         {
 327                 .procname       = "busy_poll",
 328                 .data           = &sysctl_net_busy_poll,
 329                 .maxlen         = sizeof(unsigned int),
 330                 .mode           = 0644,
 331                 .proc_handler   = proc_dointvec
 332         },
 333         {
 334                 .procname       = "busy_read",
 335                 .data           = &sysctl_net_busy_read,
 336                 .maxlen         = sizeof(unsigned int),
 337                 .mode           = 0644,
 338                 .proc_handler   = proc_dointvec
 339         },
 340 #endif
 341 #ifdef CONFIG_NET_SCHED
 342         {
 343                 .procname       = "default_qdisc",
 344                 .mode           = 0644,
 345                 .maxlen         = IFNAMSIZ,
 346                 .proc_handler   = set_default_qdisc
 347         },
 348 #endif
 349 #endif /* CONFIG_NET */
 350         {
 351                 .procname       = "netdev_budget",
 352                 .data           = &netdev_budget,
 353                 .maxlen         = sizeof(int),
 354                 .mode           = 0644,
 355                 .proc_handler   = proc_dointvec
 356         },
 357         {
 358                 .procname       = "warnings",
 359                 .data           = &net_msg_warn,
 360                 .maxlen         = sizeof(int),
 361                 .mode           = 0644,
 362                 .proc_handler   = proc_dointvec
 363         },
 364         { }
 365 };
 366
 367 static struct ctl_table netns_core_table[] = {
 368         {
 369                 .procname       = "somaxconn",
 370                 .data           = &init_net.core.sysctl_somaxconn,
 371                 .maxlen         = sizeof(int),
 372                 .mode           = 0644,
 373                 .extra1         = &zero,
 374                 .extra2         = &ushort_max,
 375                 .proc_handler   = proc_dointvec_minmax
 376         },
 377         { }
 378 };
 379
 380 static __net_init int sysctl_core_net_init(struct net *net)
 381 {
 382         struct ctl_table *tbl;
 383
 384         net->core.sysctl_somaxconn = SOMAXCONN;
 385
 386         tbl = netns_core_table;
 387         if (!net_eq(net, &init_net)) {
 388                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 389                 if (tbl == NULL)
 390                         goto err_dup;
 391
 392                 tbl[0].data = &net->core.sysctl_somaxconn;
 393
 394                 /* Don't export any sysctls to unprivileged users */
 395                 if (net->user_ns != &init_user_ns) {
 396                         tbl[0].procname = NULL;
 397                 }
 398         }
 399
 400         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 401         if (net->core.sysctl_hdr == NULL)
 402                 goto err_reg;
 403
 404         return 0;
 405
 406 err_reg:
 407         if (tbl != netns_core_table)
 408                 kfree(tbl);
 409 err_dup:
 410         return -ENOMEM;
 411 }
 412
 413 static __net_exit void sysctl_core_net_exit(struct net *net)
 414 {
 415         struct ctl_table *tbl;
 416
 417         tbl = net->core.sysctl_hdr->ctl_table_arg;
 418         unregister_net_sysctl_table(net->core.sysctl_hdr);
 419         BUG_ON(tbl == netns_core_table);
 420         kfree(tbl);
 421 }
 422
 423 static __net_initdata struct pernet_operations sysctl_core_ops = {
 424         .init = sysctl_core_net_init,
 425         .exit = sysctl_core_net_exit,
 426 };
 427
 428 static __init int sysctl_core_init(void)
 429 {
 430         register_net_sysctl(&init_net, "net/core", net_core_table);
 431         return register_pernet_subsys(&sysctl_core_ops);
 432 }
 433
 434 fs_initcall(sysctl_core_init);