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