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 min_sndbuf = SOCK_MIN_SNDBUF; 28 static int min_rcvbuf = SOCK_MIN_RCVBUF; 29 static int max_skb_frags = MAX_SKB_FRAGS; 30 31 static int net_msg_warn; /* Unused, but still a sysctl */ 32 33 #ifdef CONFIG_RPS 34 static int rps_sock_flow_sysctl(struct ctl_table *table, int write, 35 void __user *buffer, size_t *lenp, loff_t *ppos) 36 { 37 unsigned int orig_size, size; 38 int ret, i; 39 struct ctl_table tmp = { 40 .data = &size, 41 .maxlen = sizeof(size), 42 .mode = table->mode 43 }; 44 struct rps_sock_flow_table *orig_sock_table, *sock_table; 45 static DEFINE_MUTEX(sock_flow_mutex); 46 47 mutex_lock(&sock_flow_mutex); 48 49 orig_sock_table = rcu_dereference_protected(rps_sock_flow_table, 50 lockdep_is_held(&sock_flow_mutex)); 51 size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0; 52 53 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos); 54 55 if (write) { 56 if (size) { 57 if (size > 1<<29) { 58 /* Enforce limit to prevent overflow */ 59 mutex_unlock(&sock_flow_mutex); 60 return -EINVAL; 61 } 62 size = roundup_pow_of_two(size); 63 if (size != orig_size) { 64 sock_table = 65 vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size)); 66 if (!sock_table) { 67 mutex_unlock(&sock_flow_mutex); 68 return -ENOMEM; 69 } 70 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1; 71 sock_table->mask = size - 1; 72 } else 73 sock_table = orig_sock_table; 74 75 for (i = 0; i < size; i++) 76 sock_table->ents[i] = RPS_NO_CPU; 77 } else 78 sock_table = NULL; 79 80 if (sock_table != orig_sock_table) { 81 rcu_assign_pointer(rps_sock_flow_table, sock_table); 82 if (sock_table) { 83 static_key_slow_inc(&rps_needed); 84 static_key_slow_inc(&rfs_needed); 85 } 86 if (orig_sock_table) { 87 static_key_slow_dec(&rps_needed); 88 static_key_slow_dec(&rfs_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 static struct ctl_table net_core_table[] = { 238 #ifdef CONFIG_NET 239 { 240 .procname = "wmem_max", 241 .data = &sysctl_wmem_max, 242 .maxlen = sizeof(int), 243 .mode = 0644, 244 .proc_handler = proc_dointvec_minmax, 245 .extra1 = &min_sndbuf, 246 }, 247 { 248 .procname = "rmem_max", 249 .data = &sysctl_rmem_max, 250 .maxlen = sizeof(int), 251 .mode = 0644, 252 .proc_handler = proc_dointvec_minmax, 253 .extra1 = &min_rcvbuf, 254 }, 255 { 256 .procname = "wmem_default", 257 .data = &sysctl_wmem_default, 258 .maxlen = sizeof(int), 259 .mode = 0644, 260 .proc_handler = proc_dointvec_minmax, 261 .extra1 = &min_sndbuf, 262 }, 263 { 264 .procname = "rmem_default", 265 .data = &sysctl_rmem_default, 266 .maxlen = sizeof(int), 267 .mode = 0644, 268 .proc_handler = proc_dointvec_minmax, 269 .extra1 = &min_rcvbuf, 270 }, 271 { 272 .procname = "dev_weight", 273 .data = &weight_p, 274 .maxlen = sizeof(int), 275 .mode = 0644, 276 .proc_handler = proc_dointvec 277 }, 278 { 279 .procname = "netdev_max_backlog", 280 .data = &netdev_max_backlog, 281 .maxlen = sizeof(int), 282 .mode = 0644, 283 .proc_handler = proc_dointvec 284 }, 285 { 286 .procname = "netdev_rss_key", 287 .data = &netdev_rss_key, 288 .maxlen = sizeof(int), 289 .mode = 0444, 290 .proc_handler = proc_do_rss_key, 291 }, 292 #ifdef CONFIG_BPF_JIT 293 { 294 .procname = "bpf_jit_enable", 295 .data = &bpf_jit_enable, 296 .maxlen = sizeof(int), 297 .mode = 0644, 298 .proc_handler = proc_dointvec 299 }, 300 # ifdef CONFIG_HAVE_EBPF_JIT 301 { 302 .procname = "bpf_jit_harden", 303 .data = &bpf_jit_harden, 304 .maxlen = sizeof(int), 305 .mode = 0600, 306 .proc_handler = proc_dointvec, 307 }, 308 # endif 309 #endif 310 { 311 .procname = "netdev_tstamp_prequeue", 312 .data = &netdev_tstamp_prequeue, 313 .maxlen = sizeof(int), 314 .mode = 0644, 315 .proc_handler = proc_dointvec 316 }, 317 { 318 .procname = "message_cost", 319 .data = &net_ratelimit_state.interval, 320 .maxlen = sizeof(int), 321 .mode = 0644, 322 .proc_handler = proc_dointvec_jiffies, 323 }, 324 { 325 .procname = "message_burst", 326 .data = &net_ratelimit_state.burst, 327 .maxlen = sizeof(int), 328 .mode = 0644, 329 .proc_handler = proc_dointvec, 330 }, 331 { 332 .procname = "optmem_max", 333 .data = &sysctl_optmem_max, 334 .maxlen = sizeof(int), 335 .mode = 0644, 336 .proc_handler = proc_dointvec 337 }, 338 { 339 .procname = "tstamp_allow_data", 340 .data = &sysctl_tstamp_allow_data, 341 .maxlen = sizeof(int), 342 .mode = 0644, 343 .proc_handler = proc_dointvec_minmax, 344 .extra1 = &zero, 345 .extra2 = &one 346 }, 347 #ifdef CONFIG_RPS 348 { 349 .procname = "rps_sock_flow_entries", 350 .maxlen = sizeof(int), 351 .mode = 0644, 352 .proc_handler = rps_sock_flow_sysctl 353 }, 354 #endif 355 #ifdef CONFIG_NET_FLOW_LIMIT 356 { 357 .procname = "flow_limit_cpu_bitmap", 358 .mode = 0644, 359 .proc_handler = flow_limit_cpu_sysctl 360 }, 361 { 362 .procname = "flow_limit_table_len", 363 .data = &netdev_flow_limit_table_len, 364 .maxlen = sizeof(int), 365 .mode = 0644, 366 .proc_handler = flow_limit_table_len_sysctl 367 }, 368 #endif /* CONFIG_NET_FLOW_LIMIT */ 369 #ifdef CONFIG_NET_RX_BUSY_POLL 370 { 371 .procname = "busy_poll", 372 .data = &sysctl_net_busy_poll, 373 .maxlen = sizeof(unsigned int), 374 .mode = 0644, 375 .proc_handler = proc_dointvec 376 }, 377 { 378 .procname = "busy_read", 379 .data = &sysctl_net_busy_read, 380 .maxlen = sizeof(unsigned int), 381 .mode = 0644, 382 .proc_handler = proc_dointvec 383 }, 384 #endif 385 #ifdef CONFIG_NET_SCHED 386 { 387 .procname = "default_qdisc", 388 .mode = 0644, 389 .maxlen = IFNAMSIZ, 390 .proc_handler = set_default_qdisc 391 }, 392 #endif 393 #endif /* CONFIG_NET */ 394 { 395 .procname = "netdev_budget", 396 .data = &netdev_budget, 397 .maxlen = sizeof(int), 398 .mode = 0644, 399 .proc_handler = proc_dointvec 400 }, 401 { 402 .procname = "warnings", 403 .data = &net_msg_warn, 404 .maxlen = sizeof(int), 405 .mode = 0644, 406 .proc_handler = proc_dointvec 407 }, 408 { 409 .procname = "max_skb_frags", 410 .data = &sysctl_max_skb_frags, 411 .maxlen = sizeof(int), 412 .mode = 0644, 413 .proc_handler = proc_dointvec_minmax, 414 .extra1 = &one, 415 .extra2 = &max_skb_frags, 416 }, 417 { } 418 }; 419 420 static struct ctl_table netns_core_table[] = { 421 { 422 .procname = "somaxconn", 423 .data = &init_net.core.sysctl_somaxconn, 424 .maxlen = sizeof(int), 425 .mode = 0644, 426 .extra1 = &zero, 427 .proc_handler = proc_dointvec_minmax 428 }, 429 { } 430 }; 431 432 static __net_init int sysctl_core_net_init(struct net *net) 433 { 434 struct ctl_table *tbl; 435 436 net->core.sysctl_somaxconn = SOMAXCONN; 437 438 tbl = netns_core_table; 439 if (!net_eq(net, &init_net)) { 440 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL); 441 if (tbl == NULL) 442 goto err_dup; 443 444 tbl[0].data = &net->core.sysctl_somaxconn; 445 446 /* Don't export any sysctls to unprivileged users */ 447 if (net->user_ns != &init_user_ns) { 448 tbl[0].procname = NULL; 449 } 450 } 451 452 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl); 453 if (net->core.sysctl_hdr == NULL) 454 goto err_reg; 455 456 return 0; 457 458 err_reg: 459 if (tbl != netns_core_table) 460 kfree(tbl); 461 err_dup: 462 return -ENOMEM; 463 } 464 465 static __net_exit void sysctl_core_net_exit(struct net *net) 466 { 467 struct ctl_table *tbl; 468 469 tbl = net->core.sysctl_hdr->ctl_table_arg; 470 unregister_net_sysctl_table(net->core.sysctl_hdr); 471 BUG_ON(tbl == netns_core_table); 472 kfree(tbl); 473 } 474 475 static __net_initdata struct pernet_operations sysctl_core_ops = { 476 .init = sysctl_core_net_init, 477 .exit = sysctl_core_net_exit, 478 }; 479 480 static __init int sysctl_core_init(void) 481 { 482 register_net_sysctl(&init_net, "net/core", net_core_table); 483 return register_pernet_subsys(&sysctl_core_ops); 484 } 485 486 fs_initcall(sysctl_core_init); 487