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 two __maybe_unused = 2;
  28 static int min_sndbuf = SOCK_MIN_SNDBUF;
  29 static int min_rcvbuf = SOCK_MIN_RCVBUF;
  30 static int max_skb_frags = MAX_SKB_FRAGS;
  31 static long long_one __maybe_unused = 1;
  32 static long long_max __maybe_unused = LONG_MAX;
  33
  34 static int net_msg_warn;        /* Unused, but still a sysctl */
  35
  36 #ifdef CONFIG_RPS
  37 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
  38                                 void __user *buffer, size_t *lenp, loff_t *ppos)
  39 {
  40         unsigned int orig_size, size;
  41         int ret, i;
  42         struct ctl_table tmp = {
  43                 .data = &size,
  44                 .maxlen = sizeof(size),
  45                 .mode = table->mode
  46         };
  47         struct rps_sock_flow_table *orig_sock_table, *sock_table;
  48         static DEFINE_MUTEX(sock_flow_mutex);
  49
  50         mutex_lock(&sock_flow_mutex);
  51
  52         orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
  53                                         lockdep_is_held(&sock_flow_mutex));
  54         size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
  55
  56         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
  57
  58         if (write) {
  59                 if (size) {
  60                         if (size > 1<<29) {
  61                                 /* Enforce limit to prevent overflow */
  62                                 mutex_unlock(&sock_flow_mutex);
  63                                 return -EINVAL;
  64                         }
  65                         size = roundup_pow_of_two(size);
  66                         if (size != orig_size) {
  67                                 sock_table =
  68                                     vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
  69                                 if (!sock_table) {
  70                                         mutex_unlock(&sock_flow_mutex);
  71                                         return -ENOMEM;
  72                                 }
  73                                 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
  74                                 sock_table->mask = size - 1;
  75                         } else
  76                                 sock_table = orig_sock_table;
  77
  78                         for (i = 0; i < size; i++)
  79                                 sock_table->ents[i] = RPS_NO_CPU;
  80                 } else
  81                         sock_table = NULL;
  82
  83                 if (sock_table != orig_sock_table) {
  84                         rcu_assign_pointer(rps_sock_flow_table, sock_table);
  85                         if (sock_table)
  86                                 static_key_slow_inc(&rps_needed);
  87                         if (orig_sock_table) {
  88                                 static_key_slow_dec(&rps_needed);
  89                                 synchronize_rcu();
  90                                 vfree(orig_sock_table);
  91                         }
  92                 }
  93         }
  94
  95         mutex_unlock(&sock_flow_mutex);
  96
  97         return ret;
  98 }
  99 #endif /* CONFIG_RPS */
 100
 101 #ifdef CONFIG_NET_FLOW_LIMIT
 102 static DEFINE_MUTEX(flow_limit_update_mutex);
 103
 104 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
 105                                  void __user *buffer, size_t *lenp,
 106                                  loff_t *ppos)
 107 {
 108         struct sd_flow_limit *cur;
 109         struct softnet_data *sd;
 110         cpumask_var_t mask;
 111         int i, len, ret = 0;
 112
 113         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 114                 return -ENOMEM;
 115
 116         if (write) {
 117                 ret = cpumask_parse_user(buffer, *lenp, mask);
 118                 if (ret)
 119                         goto done;
 120
 121                 mutex_lock(&flow_limit_update_mutex);
 122                 len = sizeof(*cur) + netdev_flow_limit_table_len;
 123                 for_each_possible_cpu(i) {
 124                         sd = &per_cpu(softnet_data, i);
 125                         cur = rcu_dereference_protected(sd->flow_limit,
 126                                      lockdep_is_held(&flow_limit_update_mutex));
 127                         if (cur && !cpumask_test_cpu(i, mask)) {
 128                                 RCU_INIT_POINTER(sd->flow_limit, NULL);
 129                                 synchronize_rcu();
 130                                 kfree(cur);
 131                         } else if (!cur && cpumask_test_cpu(i, mask)) {
 132                                 cur = kzalloc_node(len, GFP_KERNEL,
 133                                                    cpu_to_node(i));
 134                                 if (!cur) {
 135                                         /* not unwinding previous changes */
 136                                         ret = -ENOMEM;
 137                                         goto write_unlock;
 138                                 }
 139                                 cur->num_buckets = netdev_flow_limit_table_len;
 140                                 rcu_assign_pointer(sd->flow_limit, cur);
 141                         }
 142                 }
 143 write_unlock:
 144                 mutex_unlock(&flow_limit_update_mutex);
 145         } else {
 146                 char kbuf[128];
 147
 148                 if (*ppos || !*lenp) {
 149                         *lenp = 0;
 150                         goto done;
 151                 }
 152
 153                 cpumask_clear(mask);
 154                 rcu_read_lock();
 155                 for_each_possible_cpu(i) {
 156                         sd = &per_cpu(softnet_data, i);
 157                         if (rcu_dereference(sd->flow_limit))
 158                                 cpumask_set_cpu(i, mask);
 159                 }
 160                 rcu_read_unlock();
 161
 162                 len = min(sizeof(kbuf) - 1, *lenp);
 163                 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
 164                 if (!len) {
 165                         *lenp = 0;
 166                         goto done;
 167                 }
 168                 if (len < *lenp)
 169                         kbuf[len++] = '\n';
 170                 if (copy_to_user(buffer, kbuf, len)) {
 171                         ret = -EFAULT;
 172                         goto done;
 173                 }
 174                 *lenp = len;
 175                 *ppos += len;
 176         }
 177
 178 done:
 179         free_cpumask_var(mask);
 180         return ret;
 181 }
 182
 183 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
 184                                        void __user *buffer, size_t *lenp,
 185                                        loff_t *ppos)
 186 {
 187         unsigned int old, *ptr;
 188         int ret;
 189
 190         mutex_lock(&flow_limit_update_mutex);
 191
 192         ptr = table->data;
 193         old = *ptr;
 194         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 195         if (!ret && write && !is_power_of_2(*ptr)) {
 196                 *ptr = old;
 197                 ret = -EINVAL;
 198         }
 199
 200         mutex_unlock(&flow_limit_update_mutex);
 201         return ret;
 202 }
 203 #endif /* CONFIG_NET_FLOW_LIMIT */
 204
 205 #ifdef CONFIG_NET_SCHED
 206 static int set_default_qdisc(struct ctl_table *table, int write,
 207                              void __user *buffer, size_t *lenp, loff_t *ppos)
 208 {
 209         char id[IFNAMSIZ];
 210         struct ctl_table tbl = {
 211                 .data = id,
 212                 .maxlen = IFNAMSIZ,
 213         };
 214         int ret;
 215
 216         qdisc_get_default(id, IFNAMSIZ);
 217
 218         ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
 219         if (write && ret == 0)
 220                 ret = qdisc_set_default(id);
 221         return ret;
 222 }
 223 #endif
 224
 225 static int proc_do_rss_key(struct ctl_table *table, int write,
 226                            void __user *buffer, size_t *lenp, loff_t *ppos)
 227 {
 228         struct ctl_table fake_table;
 229         char buf[NETDEV_RSS_KEY_LEN * 3];
 230
 231         snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
 232         fake_table.data = buf;
 233         fake_table.maxlen = sizeof(buf);
 234         return proc_dostring(&fake_table, write, buffer, lenp, ppos);
 235 }
 236
 237 #ifdef CONFIG_BPF_JIT
 238 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
 239                                            void __user *buffer, size_t *lenp,
 240                                            loff_t *ppos)
 241 {
 242         int ret, jit_enable = *(int *)table->data;
 243         struct ctl_table tmp = *table;
 244
 245         if (write && !capable(CAP_SYS_ADMIN))
 246                 return -EPERM;
 247
 248         tmp.data = &jit_enable;
 249         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 250         if (write && !ret) {
 251                 *(int *)table->data = jit_enable;
 252                 if (jit_enable == 2)
 253                         pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
 254         }
 255         return ret;
 256 }
 257
 258 # ifdef CONFIG_HAVE_EBPF_JIT
 259 static int
 260 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
 261                                     void __user *buffer, size_t *lenp,
 262                                     loff_t *ppos)
 263 {
 264         if (!capable(CAP_SYS_ADMIN))
 265                 return -EPERM;
 266
 267         return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
 268 }
 269 # endif /* CONFIG_HAVE_EBPF_JIT */
 270
 271 static int
 272 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
 273                                      void __user *buffer, size_t *lenp,
 274                                      loff_t *ppos)
 275 {
 276         if (!capable(CAP_SYS_ADMIN))
 277                 return -EPERM;
 278
 279         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 280 }
 281 #endif
 282
 283 static struct ctl_table net_core_table[] = {
 284 #ifdef CONFIG_NET
 285         {
 286                 .procname       = "wmem_max",
 287                 .data           = &sysctl_wmem_max,
 288                 .maxlen         = sizeof(int),
 289                 .mode           = 0644,
 290                 .proc_handler   = proc_dointvec_minmax,
 291                 .extra1         = &min_sndbuf,
 292         },
 293         {
 294                 .procname       = "rmem_max",
 295                 .data           = &sysctl_rmem_max,
 296                 .maxlen         = sizeof(int),
 297                 .mode           = 0644,
 298                 .proc_handler   = proc_dointvec_minmax,
 299                 .extra1         = &min_rcvbuf,
 300         },
 301         {
 302                 .procname       = "wmem_default",
 303                 .data           = &sysctl_wmem_default,
 304                 .maxlen         = sizeof(int),
 305                 .mode           = 0644,
 306                 .proc_handler   = proc_dointvec_minmax,
 307                 .extra1         = &min_sndbuf,
 308         },
 309         {
 310                 .procname       = "rmem_default",
 311                 .data           = &sysctl_rmem_default,
 312                 .maxlen         = sizeof(int),
 313                 .mode           = 0644,
 314                 .proc_handler   = proc_dointvec_minmax,
 315                 .extra1         = &min_rcvbuf,
 316         },
 317         {
 318                 .procname       = "dev_weight",
 319                 .data           = &weight_p,
 320                 .maxlen         = sizeof(int),
 321                 .mode           = 0644,
 322                 .proc_handler   = proc_dointvec
 323         },
 324         {
 325                 .procname       = "netdev_max_backlog",
 326                 .data           = &netdev_max_backlog,
 327                 .maxlen         = sizeof(int),
 328                 .mode           = 0644,
 329                 .proc_handler   = proc_dointvec
 330         },
 331         {
 332                 .procname       = "netdev_rss_key",
 333                 .data           = &netdev_rss_key,
 334                 .maxlen         = sizeof(int),
 335                 .mode           = 0444,
 336                 .proc_handler   = proc_do_rss_key,
 337         },
 338 #ifdef CONFIG_BPF_JIT
 339         {
 340                 .procname       = "bpf_jit_enable",
 341                 .data           = &bpf_jit_enable,
 342                 .maxlen         = sizeof(int),
 343                 .mode           = 0644,
 344                 .proc_handler   = proc_dointvec_minmax_bpf_enable,
 345 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
 346                 .extra1         = &one,
 347                 .extra2         = &one,
 348 # else
 349                 .extra1         = &zero,
 350                 .extra2         = &two,
 351 # endif
 352         },
 353 # ifdef CONFIG_HAVE_EBPF_JIT
 354         {
 355                 .procname       = "bpf_jit_harden",
 356                 .data           = &bpf_jit_harden,
 357                 .maxlen         = sizeof(int),
 358                 .mode           = 0600,
 359                 .proc_handler   = proc_dointvec_minmax_bpf_restricted,
 360                 .extra1         = &zero,
 361                 .extra2         = &two,
 362         },
 363 # endif
 364         {
 365                 .procname       = "bpf_jit_limit",
 366                 .data           = &bpf_jit_limit,
 367                 .maxlen         = sizeof(long),
 368                 .mode           = 0600,
 369                 .proc_handler   = proc_dolongvec_minmax_bpf_restricted,
 370                 .extra1         = &long_one,
 371                 .extra2         = &bpf_jit_limit_max,
 372         },
 373 #endif
 374         {
 375                 .procname       = "netdev_tstamp_prequeue",
 376                 .data           = &netdev_tstamp_prequeue,
 377                 .maxlen         = sizeof(int),
 378                 .mode           = 0644,
 379                 .proc_handler   = proc_dointvec
 380         },
 381         {
 382                 .procname       = "message_cost",
 383                 .data           = &net_ratelimit_state.interval,
 384                 .maxlen         = sizeof(int),
 385                 .mode           = 0644,
 386                 .proc_handler   = proc_dointvec_jiffies,
 387         },
 388         {
 389                 .procname       = "message_burst",
 390                 .data           = &net_ratelimit_state.burst,
 391                 .maxlen         = sizeof(int),
 392                 .mode           = 0644,
 393                 .proc_handler   = proc_dointvec,
 394         },
 395         {
 396                 .procname       = "optmem_max",
 397                 .data           = &sysctl_optmem_max,
 398                 .maxlen         = sizeof(int),
 399                 .mode           = 0644,
 400                 .proc_handler   = proc_dointvec
 401         },
 402         {
 403                 .procname       = "tstamp_allow_data",
 404                 .data           = &sysctl_tstamp_allow_data,
 405                 .maxlen         = sizeof(int),
 406                 .mode           = 0644,
 407                 .proc_handler   = proc_dointvec_minmax,
 408                 .extra1         = &zero,
 409                 .extra2         = &one
 410         },
 411 #ifdef CONFIG_RPS
 412         {
 413                 .procname       = "rps_sock_flow_entries",
 414                 .maxlen         = sizeof(int),
 415                 .mode           = 0644,
 416                 .proc_handler   = rps_sock_flow_sysctl
 417         },
 418 #endif
 419 #ifdef CONFIG_NET_FLOW_LIMIT
 420         {
 421                 .procname       = "flow_limit_cpu_bitmap",
 422                 .mode           = 0644,
 423                 .proc_handler   = flow_limit_cpu_sysctl
 424         },
 425         {
 426                 .procname       = "flow_limit_table_len",
 427                 .data           = &netdev_flow_limit_table_len,
 428                 .maxlen         = sizeof(int),
 429                 .mode           = 0644,
 430                 .proc_handler   = flow_limit_table_len_sysctl
 431         },
 432 #endif /* CONFIG_NET_FLOW_LIMIT */
 433 #ifdef CONFIG_NET_RX_BUSY_POLL
 434         {
 435                 .procname       = "busy_poll",
 436                 .data           = &sysctl_net_busy_poll,
 437                 .maxlen         = sizeof(unsigned int),
 438                 .mode           = 0644,
 439                 .proc_handler   = proc_dointvec_minmax,
 440                 .extra1         = &zero,
 441         },
 442         {
 443                 .procname       = "busy_read",
 444                 .data           = &sysctl_net_busy_read,
 445                 .maxlen         = sizeof(unsigned int),
 446                 .mode           = 0644,
 447                 .proc_handler   = proc_dointvec_minmax,
 448                 .extra1         = &zero,
 449         },
 450 #endif
 451 #ifdef CONFIG_NET_SCHED
 452         {
 453                 .procname       = "default_qdisc",
 454                 .mode           = 0644,
 455                 .maxlen         = IFNAMSIZ,
 456                 .proc_handler   = set_default_qdisc
 457         },
 458 #endif
 459 #endif /* CONFIG_NET */
 460         {
 461                 .procname       = "netdev_budget",
 462                 .data           = &netdev_budget,
 463                 .maxlen         = sizeof(int),
 464                 .mode           = 0644,
 465                 .proc_handler   = proc_dointvec
 466         },
 467         {
 468                 .procname       = "warnings",
 469                 .data           = &net_msg_warn,
 470                 .maxlen         = sizeof(int),
 471                 .mode           = 0644,
 472                 .proc_handler   = proc_dointvec
 473         },
 474         {
 475                 .procname       = "max_skb_frags",
 476                 .data           = &sysctl_max_skb_frags,
 477                 .maxlen         = sizeof(int),
 478                 .mode           = 0644,
 479                 .proc_handler   = proc_dointvec_minmax,
 480                 .extra1         = &one,
 481                 .extra2         = &max_skb_frags,
 482         },
 483         { }
 484 };
 485
 486 static struct ctl_table netns_core_table[] = {
 487         {
 488                 .procname       = "somaxconn",
 489                 .data           = &init_net.core.sysctl_somaxconn,
 490                 .maxlen         = sizeof(int),
 491                 .mode           = 0644,
 492                 .extra1         = &zero,
 493                 .proc_handler   = proc_dointvec_minmax
 494         },
 495         { }
 496 };
 497
 498 static __net_init int sysctl_core_net_init(struct net *net)
 499 {
 500         struct ctl_table *tbl;
 501
 502         tbl = netns_core_table;
 503         if (!net_eq(net, &init_net)) {
 504                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 505                 if (tbl == NULL)
 506                         goto err_dup;
 507
 508                 tbl[0].data = &net->core.sysctl_somaxconn;
 509
 510                 /* Don't export any sysctls to unprivileged users */
 511                 if (net->user_ns != &init_user_ns) {
 512                         tbl[0].procname = NULL;
 513                 }
 514         }
 515
 516         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 517         if (net->core.sysctl_hdr == NULL)
 518                 goto err_reg;
 519
 520         return 0;
 521
 522 err_reg:
 523         if (tbl != netns_core_table)
 524                 kfree(tbl);
 525 err_dup:
 526         return -ENOMEM;
 527 }
 528
 529 static __net_exit void sysctl_core_net_exit(struct net *net)
 530 {
 531         struct ctl_table *tbl;
 532
 533         tbl = net->core.sysctl_hdr->ctl_table_arg;
 534         unregister_net_sysctl_table(net->core.sysctl_hdr);
 535         BUG_ON(tbl == netns_core_table);
 536         kfree(tbl);
 537 }
 538
 539 static __net_initdata struct pernet_operations sysctl_core_ops = {
 540         .init = sysctl_core_net_init,
 541         .exit = sysctl_core_net_exit,
 542 };
 543
 544 static __init int sysctl_core_init(void)
 545 {
 546         register_net_sysctl(&init_net, "net/core", net_core_table);
 547         return register_pernet_subsys(&sysctl_core_ops);
 548 }
 549
 550 fs_initcall(sysctl_core_init);