1 /* 2 * NETLINK Kernel-user communication protocol. 3 * 4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> 5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 6 * Patrick McHardy <kaber@trash.net> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 * 13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith 14 * added netlink_proto_exit 15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br> 16 * use nlk_sk, as sk->protinfo is on a diet 8) 17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org> 18 * - inc module use count of module that owns 19 * the kernel socket in case userspace opens 20 * socket of same protocol 21 * - remove all module support, since netlink is 22 * mandatory if CONFIG_NET=y these days 23 */ 24 25 #include <linux/module.h> 26 27 #include <linux/capability.h> 28 #include <linux/kernel.h> 29 #include <linux/init.h> 30 #include <linux/signal.h> 31 #include <linux/sched.h> 32 #include <linux/errno.h> 33 #include <linux/string.h> 34 #include <linux/stat.h> 35 #include <linux/socket.h> 36 #include <linux/un.h> 37 #include <linux/fcntl.h> 38 #include <linux/termios.h> 39 #include <linux/sockios.h> 40 #include <linux/net.h> 41 #include <linux/fs.h> 42 #include <linux/slab.h> 43 #include <asm/uaccess.h> 44 #include <linux/skbuff.h> 45 #include <linux/netdevice.h> 46 #include <linux/rtnetlink.h> 47 #include <linux/proc_fs.h> 48 #include <linux/seq_file.h> 49 #include <linux/notifier.h> 50 #include <linux/security.h> 51 #include <linux/jhash.h> 52 #include <linux/jiffies.h> 53 #include <linux/random.h> 54 #include <linux/bitops.h> 55 #include <linux/mm.h> 56 #include <linux/types.h> 57 #include <linux/audit.h> 58 #include <linux/mutex.h> 59 #include <linux/vmalloc.h> 60 #include <linux/if_arp.h> 61 #include <linux/rhashtable.h> 62 #include <asm/cacheflush.h> 63 #include <linux/hash.h> 64 #include <linux/genetlink.h> 65 66 #include <net/net_namespace.h> 67 #include <net/sock.h> 68 #include <net/scm.h> 69 #include <net/netlink.h> 70 71 #include "af_netlink.h" 72 73 struct listeners { 74 struct rcu_head rcu; 75 unsigned long masks[0]; 76 }; 77 78 /* state bits */ 79 #define NETLINK_CONGESTED 0x0 80 81 /* flags */ 82 #define NETLINK_KERNEL_SOCKET 0x1 83 #define NETLINK_RECV_PKTINFO 0x2 84 #define NETLINK_BROADCAST_SEND_ERROR 0x4 85 #define NETLINK_RECV_NO_ENOBUFS 0x8 86 87 static inline int netlink_is_kernel(struct sock *sk) 88 { 89 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET; 90 } 91 92 struct netlink_table *nl_table __read_mostly; 93 EXPORT_SYMBOL_GPL(nl_table); 94 95 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait); 96 97 static int netlink_dump(struct sock *sk); 98 static void netlink_skb_destructor(struct sk_buff *skb); 99 100 /* nl_table locking explained: 101 * Lookup and traversal are protected with an RCU read-side lock. Insertion 102 * and removal are protected with per bucket lock while using RCU list 103 * modification primitives and may run in parallel to RCU protected lookups. 104 * Destruction of the Netlink socket may only occur *after* nl_table_lock has 105 * been acquired * either during or after the socket has been removed from 106 * the list and after an RCU grace period. 107 */ 108 DEFINE_RWLOCK(nl_table_lock); 109 EXPORT_SYMBOL_GPL(nl_table_lock); 110 static atomic_t nl_table_users = ATOMIC_INIT(0); 111 112 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock)); 113 114 static ATOMIC_NOTIFIER_HEAD(netlink_chain); 115 116 static DEFINE_SPINLOCK(netlink_tap_lock); 117 static struct list_head netlink_tap_all __read_mostly; 118 119 static const struct rhashtable_params netlink_rhashtable_params; 120 121 static inline u32 netlink_group_mask(u32 group) 122 { 123 return group ? 1 << (group - 1) : 0; 124 } 125 126 int netlink_add_tap(struct netlink_tap *nt) 127 { 128 if (unlikely(nt->dev->type != ARPHRD_NETLINK)) 129 return -EINVAL; 130 131 spin_lock(&netlink_tap_lock); 132 list_add_rcu(&nt->list, &netlink_tap_all); 133 spin_unlock(&netlink_tap_lock); 134 135 __module_get(nt->module); 136 137 return 0; 138 } 139 EXPORT_SYMBOL_GPL(netlink_add_tap); 140 141 static int __netlink_remove_tap(struct netlink_tap *nt) 142 { 143 bool found = false; 144 struct netlink_tap *tmp; 145 146 spin_lock(&netlink_tap_lock); 147 148 list_for_each_entry(tmp, &netlink_tap_all, list) { 149 if (nt == tmp) { 150 list_del_rcu(&nt->list); 151 found = true; 152 goto out; 153 } 154 } 155 156 pr_warn("__netlink_remove_tap: %p not found\n", nt); 157 out: 158 spin_unlock(&netlink_tap_lock); 159 160 if (found && nt->module) 161 module_put(nt->module); 162 163 return found ? 0 : -ENODEV; 164 } 165 166 int netlink_remove_tap(struct netlink_tap *nt) 167 { 168 int ret; 169 170 ret = __netlink_remove_tap(nt); 171 synchronize_net(); 172 173 return ret; 174 } 175 EXPORT_SYMBOL_GPL(netlink_remove_tap); 176 177 static bool netlink_filter_tap(const struct sk_buff *skb) 178 { 179 struct sock *sk = skb->sk; 180 181 /* We take the more conservative approach and 182 * whitelist socket protocols that may pass. 183 */ 184 switch (sk->sk_protocol) { 185 case NETLINK_ROUTE: 186 case NETLINK_USERSOCK: 187 case NETLINK_SOCK_DIAG: 188 case NETLINK_NFLOG: 189 case NETLINK_XFRM: 190 case NETLINK_FIB_LOOKUP: 191 case NETLINK_NETFILTER: 192 case NETLINK_GENERIC: 193 return true; 194 } 195 196 return false; 197 } 198 199 static int __netlink_deliver_tap_skb(struct sk_buff *skb, 200 struct net_device *dev) 201 { 202 struct sk_buff *nskb; 203 struct sock *sk = skb->sk; 204 int ret = -ENOMEM; 205 206 dev_hold(dev); 207 nskb = skb_clone(skb, GFP_ATOMIC); 208 if (nskb) { 209 nskb->dev = dev; 210 nskb->protocol = htons((u16) sk->sk_protocol); 211 nskb->pkt_type = netlink_is_kernel(sk) ? 212 PACKET_KERNEL : PACKET_USER; 213 skb_reset_network_header(nskb); 214 ret = dev_queue_xmit(nskb); 215 if (unlikely(ret > 0)) 216 ret = net_xmit_errno(ret); 217 } 218 219 dev_put(dev); 220 return ret; 221 } 222 223 static void __netlink_deliver_tap(struct sk_buff *skb) 224 { 225 int ret; 226 struct netlink_tap *tmp; 227 228 if (!netlink_filter_tap(skb)) 229 return; 230 231 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) { 232 ret = __netlink_deliver_tap_skb(skb, tmp->dev); 233 if (unlikely(ret)) 234 break; 235 } 236 } 237 238 static void netlink_deliver_tap(struct sk_buff *skb) 239 { 240 rcu_read_lock(); 241 242 if (unlikely(!list_empty(&netlink_tap_all))) 243 __netlink_deliver_tap(skb); 244 245 rcu_read_unlock(); 246 } 247 248 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src, 249 struct sk_buff *skb) 250 { 251 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src))) 252 netlink_deliver_tap(skb); 253 } 254 255 static void netlink_overrun(struct sock *sk) 256 { 257 struct netlink_sock *nlk = nlk_sk(sk); 258 259 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) { 260 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) { 261 sk->sk_err = ENOBUFS; 262 sk->sk_error_report(sk); 263 } 264 } 265 atomic_inc(&sk->sk_drops); 266 } 267 268 static void netlink_rcv_wake(struct sock *sk) 269 { 270 struct netlink_sock *nlk = nlk_sk(sk); 271 272 if (skb_queue_empty(&sk->sk_receive_queue)) 273 clear_bit(NETLINK_CONGESTED, &nlk->state); 274 if (!test_bit(NETLINK_CONGESTED, &nlk->state)) 275 wake_up_interruptible(&nlk->wait); 276 } 277 278 #ifdef CONFIG_NETLINK_MMAP 279 static bool netlink_skb_is_mmaped(const struct sk_buff *skb) 280 { 281 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED; 282 } 283 284 static bool netlink_rx_is_mmaped(struct sock *sk) 285 { 286 return nlk_sk(sk)->rx_ring.pg_vec != NULL; 287 } 288 289 static bool netlink_tx_is_mmaped(struct sock *sk) 290 { 291 return nlk_sk(sk)->tx_ring.pg_vec != NULL; 292 } 293 294 static __pure struct page *pgvec_to_page(const void *addr) 295 { 296 if (is_vmalloc_addr(addr)) 297 return vmalloc_to_page(addr); 298 else 299 return virt_to_page(addr); 300 } 301 302 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len) 303 { 304 unsigned int i; 305 306 for (i = 0; i < len; i++) { 307 if (pg_vec[i] != NULL) { 308 if (is_vmalloc_addr(pg_vec[i])) 309 vfree(pg_vec[i]); 310 else 311 free_pages((unsigned long)pg_vec[i], order); 312 } 313 } 314 kfree(pg_vec); 315 } 316 317 static void *alloc_one_pg_vec_page(unsigned long order) 318 { 319 void *buffer; 320 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO | 321 __GFP_NOWARN | __GFP_NORETRY; 322 323 buffer = (void *)__get_free_pages(gfp_flags, order); 324 if (buffer != NULL) 325 return buffer; 326 327 buffer = vzalloc((1 << order) * PAGE_SIZE); 328 if (buffer != NULL) 329 return buffer; 330 331 gfp_flags &= ~__GFP_NORETRY; 332 return (void *)__get_free_pages(gfp_flags, order); 333 } 334 335 static void **alloc_pg_vec(struct netlink_sock *nlk, 336 struct nl_mmap_req *req, unsigned int order) 337 { 338 unsigned int block_nr = req->nm_block_nr; 339 unsigned int i; 340 void **pg_vec; 341 342 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL); 343 if (pg_vec == NULL) 344 return NULL; 345 346 for (i = 0; i < block_nr; i++) { 347 pg_vec[i] = alloc_one_pg_vec_page(order); 348 if (pg_vec[i] == NULL) 349 goto err1; 350 } 351 352 return pg_vec; 353 err1: 354 free_pg_vec(pg_vec, order, block_nr); 355 return NULL; 356 } 357 358 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req, 359 bool closing, bool tx_ring) 360 { 361 struct netlink_sock *nlk = nlk_sk(sk); 362 struct netlink_ring *ring; 363 struct sk_buff_head *queue; 364 void **pg_vec = NULL; 365 unsigned int order = 0; 366 int err; 367 368 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring; 369 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 370 371 if (!closing) { 372 if (atomic_read(&nlk->mapped)) 373 return -EBUSY; 374 if (atomic_read(&ring->pending)) 375 return -EBUSY; 376 } 377 378 if (req->nm_block_nr) { 379 if (ring->pg_vec != NULL) 380 return -EBUSY; 381 382 if ((int)req->nm_block_size <= 0) 383 return -EINVAL; 384 if (!PAGE_ALIGNED(req->nm_block_size)) 385 return -EINVAL; 386 if (req->nm_frame_size < NL_MMAP_HDRLEN) 387 return -EINVAL; 388 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT)) 389 return -EINVAL; 390 391 ring->frames_per_block = req->nm_block_size / 392 req->nm_frame_size; 393 if (ring->frames_per_block == 0) 394 return -EINVAL; 395 if (ring->frames_per_block * req->nm_block_nr != 396 req->nm_frame_nr) 397 return -EINVAL; 398 399 order = get_order(req->nm_block_size); 400 pg_vec = alloc_pg_vec(nlk, req, order); 401 if (pg_vec == NULL) 402 return -ENOMEM; 403 } else { 404 if (req->nm_frame_nr) 405 return -EINVAL; 406 } 407 408 err = -EBUSY; 409 mutex_lock(&nlk->pg_vec_lock); 410 if (closing || atomic_read(&nlk->mapped) == 0) { 411 err = 0; 412 spin_lock_bh(&queue->lock); 413 414 ring->frame_max = req->nm_frame_nr - 1; 415 ring->head = 0; 416 ring->frame_size = req->nm_frame_size; 417 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE; 418 419 swap(ring->pg_vec_len, req->nm_block_nr); 420 swap(ring->pg_vec_order, order); 421 swap(ring->pg_vec, pg_vec); 422 423 __skb_queue_purge(queue); 424 spin_unlock_bh(&queue->lock); 425 426 WARN_ON(atomic_read(&nlk->mapped)); 427 } 428 mutex_unlock(&nlk->pg_vec_lock); 429 430 if (pg_vec) 431 free_pg_vec(pg_vec, order, req->nm_block_nr); 432 return err; 433 } 434 435 static void netlink_mm_open(struct vm_area_struct *vma) 436 { 437 struct file *file = vma->vm_file; 438 struct socket *sock = file->private_data; 439 struct sock *sk = sock->sk; 440 441 if (sk) 442 atomic_inc(&nlk_sk(sk)->mapped); 443 } 444 445 static void netlink_mm_close(struct vm_area_struct *vma) 446 { 447 struct file *file = vma->vm_file; 448 struct socket *sock = file->private_data; 449 struct sock *sk = sock->sk; 450 451 if (sk) 452 atomic_dec(&nlk_sk(sk)->mapped); 453 } 454 455 static const struct vm_operations_struct netlink_mmap_ops = { 456 .open = netlink_mm_open, 457 .close = netlink_mm_close, 458 }; 459 460 static int netlink_mmap(struct file *file, struct socket *sock, 461 struct vm_area_struct *vma) 462 { 463 struct sock *sk = sock->sk; 464 struct netlink_sock *nlk = nlk_sk(sk); 465 struct netlink_ring *ring; 466 unsigned long start, size, expected; 467 unsigned int i; 468 int err = -EINVAL; 469 470 if (vma->vm_pgoff) 471 return -EINVAL; 472 473 mutex_lock(&nlk->pg_vec_lock); 474 475 expected = 0; 476 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) { 477 if (ring->pg_vec == NULL) 478 continue; 479 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE; 480 } 481 482 if (expected == 0) 483 goto out; 484 485 size = vma->vm_end - vma->vm_start; 486 if (size != expected) 487 goto out; 488 489 start = vma->vm_start; 490 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) { 491 if (ring->pg_vec == NULL) 492 continue; 493 494 for (i = 0; i < ring->pg_vec_len; i++) { 495 struct page *page; 496 void *kaddr = ring->pg_vec[i]; 497 unsigned int pg_num; 498 499 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) { 500 page = pgvec_to_page(kaddr); 501 err = vm_insert_page(vma, start, page); 502 if (err < 0) 503 goto out; 504 start += PAGE_SIZE; 505 kaddr += PAGE_SIZE; 506 } 507 } 508 } 509 510 atomic_inc(&nlk->mapped); 511 vma->vm_ops = &netlink_mmap_ops; 512 err = 0; 513 out: 514 mutex_unlock(&nlk->pg_vec_lock); 515 return err; 516 } 517 518 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr, unsigned int nm_len) 519 { 520 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 521 struct page *p_start, *p_end; 522 523 /* First page is flushed through netlink_{get,set}_status */ 524 p_start = pgvec_to_page(hdr + PAGE_SIZE); 525 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + nm_len - 1); 526 while (p_start <= p_end) { 527 flush_dcache_page(p_start); 528 p_start++; 529 } 530 #endif 531 } 532 533 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr) 534 { 535 smp_rmb(); 536 flush_dcache_page(pgvec_to_page(hdr)); 537 return hdr->nm_status; 538 } 539 540 static void netlink_set_status(struct nl_mmap_hdr *hdr, 541 enum nl_mmap_status status) 542 { 543 smp_mb(); 544 hdr->nm_status = status; 545 flush_dcache_page(pgvec_to_page(hdr)); 546 } 547 548 static struct nl_mmap_hdr * 549 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos) 550 { 551 unsigned int pg_vec_pos, frame_off; 552 553 pg_vec_pos = pos / ring->frames_per_block; 554 frame_off = pos % ring->frames_per_block; 555 556 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size); 557 } 558 559 static struct nl_mmap_hdr * 560 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos, 561 enum nl_mmap_status status) 562 { 563 struct nl_mmap_hdr *hdr; 564 565 hdr = __netlink_lookup_frame(ring, pos); 566 if (netlink_get_status(hdr) != status) 567 return NULL; 568 569 return hdr; 570 } 571 572 static struct nl_mmap_hdr * 573 netlink_current_frame(const struct netlink_ring *ring, 574 enum nl_mmap_status status) 575 { 576 return netlink_lookup_frame(ring, ring->head, status); 577 } 578 579 static struct nl_mmap_hdr * 580 netlink_previous_frame(const struct netlink_ring *ring, 581 enum nl_mmap_status status) 582 { 583 unsigned int prev; 584 585 prev = ring->head ? ring->head - 1 : ring->frame_max; 586 return netlink_lookup_frame(ring, prev, status); 587 } 588 589 static void netlink_increment_head(struct netlink_ring *ring) 590 { 591 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0; 592 } 593 594 static void netlink_forward_ring(struct netlink_ring *ring) 595 { 596 unsigned int head = ring->head, pos = head; 597 const struct nl_mmap_hdr *hdr; 598 599 do { 600 hdr = __netlink_lookup_frame(ring, pos); 601 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED) 602 break; 603 if (hdr->nm_status != NL_MMAP_STATUS_SKIP) 604 break; 605 netlink_increment_head(ring); 606 } while (ring->head != head); 607 } 608 609 static bool netlink_dump_space(struct netlink_sock *nlk) 610 { 611 struct netlink_ring *ring = &nlk->rx_ring; 612 struct nl_mmap_hdr *hdr; 613 unsigned int n; 614 615 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED); 616 if (hdr == NULL) 617 return false; 618 619 n = ring->head + ring->frame_max / 2; 620 if (n > ring->frame_max) 621 n -= ring->frame_max; 622 623 hdr = __netlink_lookup_frame(ring, n); 624 625 return hdr->nm_status == NL_MMAP_STATUS_UNUSED; 626 } 627 628 static unsigned int netlink_poll(struct file *file, struct socket *sock, 629 poll_table *wait) 630 { 631 struct sock *sk = sock->sk; 632 struct netlink_sock *nlk = nlk_sk(sk); 633 unsigned int mask; 634 int err; 635 636 if (nlk->rx_ring.pg_vec != NULL) { 637 /* Memory mapped sockets don't call recvmsg(), so flow control 638 * for dumps is performed here. A dump is allowed to continue 639 * if at least half the ring is unused. 640 */ 641 while (nlk->cb_running && netlink_dump_space(nlk)) { 642 err = netlink_dump(sk); 643 if (err < 0) { 644 sk->sk_err = -err; 645 sk->sk_error_report(sk); 646 break; 647 } 648 } 649 netlink_rcv_wake(sk); 650 } 651 652 mask = datagram_poll(file, sock, wait); 653 654 spin_lock_bh(&sk->sk_receive_queue.lock); 655 if (nlk->rx_ring.pg_vec) { 656 netlink_forward_ring(&nlk->rx_ring); 657 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED)) 658 mask |= POLLIN | POLLRDNORM; 659 } 660 spin_unlock_bh(&sk->sk_receive_queue.lock); 661 662 spin_lock_bh(&sk->sk_write_queue.lock); 663 if (nlk->tx_ring.pg_vec) { 664 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED)) 665 mask |= POLLOUT | POLLWRNORM; 666 } 667 spin_unlock_bh(&sk->sk_write_queue.lock); 668 669 return mask; 670 } 671 672 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb) 673 { 674 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN); 675 } 676 677 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk, 678 struct netlink_ring *ring, 679 struct nl_mmap_hdr *hdr) 680 { 681 unsigned int size; 682 void *data; 683 684 size = ring->frame_size - NL_MMAP_HDRLEN; 685 data = (void *)hdr + NL_MMAP_HDRLEN; 686 687 skb->head = data; 688 skb->data = data; 689 skb_reset_tail_pointer(skb); 690 skb->end = skb->tail + size; 691 skb->len = 0; 692 693 skb->destructor = netlink_skb_destructor; 694 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED; 695 NETLINK_CB(skb).sk = sk; 696 } 697 698 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg, 699 u32 dst_portid, u32 dst_group, 700 struct scm_cookie *scm) 701 { 702 struct netlink_sock *nlk = nlk_sk(sk); 703 struct netlink_ring *ring; 704 struct nl_mmap_hdr *hdr; 705 struct sk_buff *skb; 706 unsigned int maxlen; 707 int err = 0, len = 0; 708 709 mutex_lock(&nlk->pg_vec_lock); 710 711 ring = &nlk->tx_ring; 712 maxlen = ring->frame_size - NL_MMAP_HDRLEN; 713 714 do { 715 unsigned int nm_len; 716 717 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID); 718 if (hdr == NULL) { 719 if (!(msg->msg_flags & MSG_DONTWAIT) && 720 atomic_read(&nlk->tx_ring.pending)) 721 schedule(); 722 continue; 723 } 724 725 nm_len = ACCESS_ONCE(hdr->nm_len); 726 if (nm_len > maxlen) { 727 err = -EINVAL; 728 goto out; 729 } 730 731 netlink_frame_flush_dcache(hdr, nm_len); 732 733 skb = alloc_skb(nm_len, GFP_KERNEL); 734 if (skb == NULL) { 735 err = -ENOBUFS; 736 goto out; 737 } 738 __skb_put(skb, nm_len); 739 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, nm_len); 740 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED); 741 742 netlink_increment_head(ring); 743 744 NETLINK_CB(skb).portid = nlk->portid; 745 NETLINK_CB(skb).dst_group = dst_group; 746 NETLINK_CB(skb).creds = scm->creds; 747 748 err = security_netlink_send(sk, skb); 749 if (err) { 750 kfree_skb(skb); 751 goto out; 752 } 753 754 if (unlikely(dst_group)) { 755 atomic_inc(&skb->users); 756 netlink_broadcast(sk, skb, dst_portid, dst_group, 757 GFP_KERNEL); 758 } 759 err = netlink_unicast(sk, skb, dst_portid, 760 msg->msg_flags & MSG_DONTWAIT); 761 if (err < 0) 762 goto out; 763 len += err; 764 765 } while (hdr != NULL || 766 (!(msg->msg_flags & MSG_DONTWAIT) && 767 atomic_read(&nlk->tx_ring.pending))); 768 769 if (len > 0) 770 err = len; 771 out: 772 mutex_unlock(&nlk->pg_vec_lock); 773 return err; 774 } 775 776 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb) 777 { 778 struct nl_mmap_hdr *hdr; 779 780 hdr = netlink_mmap_hdr(skb); 781 hdr->nm_len = skb->len; 782 hdr->nm_group = NETLINK_CB(skb).dst_group; 783 hdr->nm_pid = NETLINK_CB(skb).creds.pid; 784 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid); 785 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid); 786 netlink_frame_flush_dcache(hdr, hdr->nm_len); 787 netlink_set_status(hdr, NL_MMAP_STATUS_VALID); 788 789 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED; 790 kfree_skb(skb); 791 } 792 793 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb) 794 { 795 struct netlink_sock *nlk = nlk_sk(sk); 796 struct netlink_ring *ring = &nlk->rx_ring; 797 struct nl_mmap_hdr *hdr; 798 799 spin_lock_bh(&sk->sk_receive_queue.lock); 800 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED); 801 if (hdr == NULL) { 802 spin_unlock_bh(&sk->sk_receive_queue.lock); 803 kfree_skb(skb); 804 netlink_overrun(sk); 805 return; 806 } 807 netlink_increment_head(ring); 808 __skb_queue_tail(&sk->sk_receive_queue, skb); 809 spin_unlock_bh(&sk->sk_receive_queue.lock); 810 811 hdr->nm_len = skb->len; 812 hdr->nm_group = NETLINK_CB(skb).dst_group; 813 hdr->nm_pid = NETLINK_CB(skb).creds.pid; 814 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid); 815 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid); 816 netlink_set_status(hdr, NL_MMAP_STATUS_COPY); 817 } 818 819 #else /* CONFIG_NETLINK_MMAP */ 820 #define netlink_skb_is_mmaped(skb) false 821 #define netlink_rx_is_mmaped(sk) false 822 #define netlink_tx_is_mmaped(sk) false 823 #define netlink_mmap sock_no_mmap 824 #define netlink_poll datagram_poll 825 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, scm) 0 826 #endif /* CONFIG_NETLINK_MMAP */ 827 828 static void netlink_skb_destructor(struct sk_buff *skb) 829 { 830 #ifdef CONFIG_NETLINK_MMAP 831 struct nl_mmap_hdr *hdr; 832 struct netlink_ring *ring; 833 struct sock *sk; 834 835 /* If a packet from the kernel to userspace was freed because of an 836 * error without being delivered to userspace, the kernel must reset 837 * the status. In the direction userspace to kernel, the status is 838 * always reset here after the packet was processed and freed. 839 */ 840 if (netlink_skb_is_mmaped(skb)) { 841 hdr = netlink_mmap_hdr(skb); 842 sk = NETLINK_CB(skb).sk; 843 844 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) { 845 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED); 846 ring = &nlk_sk(sk)->tx_ring; 847 } else { 848 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) { 849 hdr->nm_len = 0; 850 netlink_set_status(hdr, NL_MMAP_STATUS_VALID); 851 } 852 ring = &nlk_sk(sk)->rx_ring; 853 } 854 855 WARN_ON(atomic_read(&ring->pending) == 0); 856 atomic_dec(&ring->pending); 857 sock_put(sk); 858 859 skb->head = NULL; 860 } 861 #endif 862 if (is_vmalloc_addr(skb->head)) { 863 if (!skb->cloned || 864 !atomic_dec_return(&(skb_shinfo(skb)->dataref))) 865 vfree(skb->head); 866 867 skb->head = NULL; 868 } 869 if (skb->sk != NULL) 870 sock_rfree(skb); 871 } 872 873 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 874 { 875 WARN_ON(skb->sk != NULL); 876 skb->sk = sk; 877 skb->destructor = netlink_skb_destructor; 878 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 879 sk_mem_charge(sk, skb->truesize); 880 } 881 882 static void netlink_sock_destruct(struct sock *sk) 883 { 884 struct netlink_sock *nlk = nlk_sk(sk); 885 886 if (nlk->cb_running) { 887 if (nlk->cb.done) 888 nlk->cb.done(&nlk->cb); 889 890 module_put(nlk->cb.module); 891 kfree_skb(nlk->cb.skb); 892 } 893 894 skb_queue_purge(&sk->sk_receive_queue); 895 #ifdef CONFIG_NETLINK_MMAP 896 if (1) { 897 struct nl_mmap_req req; 898 899 memset(&req, 0, sizeof(req)); 900 if (nlk->rx_ring.pg_vec) 901 netlink_set_ring(sk, &req, true, false); 902 memset(&req, 0, sizeof(req)); 903 if (nlk->tx_ring.pg_vec) 904 netlink_set_ring(sk, &req, true, true); 905 } 906 #endif /* CONFIG_NETLINK_MMAP */ 907 908 if (!sock_flag(sk, SOCK_DEAD)) { 909 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk); 910 return; 911 } 912 913 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 914 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 915 WARN_ON(nlk_sk(sk)->groups); 916 } 917 918 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on 919 * SMP. Look, when several writers sleep and reader wakes them up, all but one 920 * immediately hit write lock and grab all the cpus. Exclusive sleep solves 921 * this, _but_ remember, it adds useless work on UP machines. 922 */ 923 924 void netlink_table_grab(void) 925 __acquires(nl_table_lock) 926 { 927 might_sleep(); 928 929 write_lock_irq(&nl_table_lock); 930 931 if (atomic_read(&nl_table_users)) { 932 DECLARE_WAITQUEUE(wait, current); 933 934 add_wait_queue_exclusive(&nl_table_wait, &wait); 935 for (;;) { 936 set_current_state(TASK_UNINTERRUPTIBLE); 937 if (atomic_read(&nl_table_users) == 0) 938 break; 939 write_unlock_irq(&nl_table_lock); 940 schedule(); 941 write_lock_irq(&nl_table_lock); 942 } 943 944 __set_current_state(TASK_RUNNING); 945 remove_wait_queue(&nl_table_wait, &wait); 946 } 947 } 948 949 void netlink_table_ungrab(void) 950 __releases(nl_table_lock) 951 { 952 write_unlock_irq(&nl_table_lock); 953 wake_up(&nl_table_wait); 954 } 955 956 static inline void 957 netlink_lock_table(void) 958 { 959 /* read_lock() synchronizes us to netlink_table_grab */ 960 961 read_lock(&nl_table_lock); 962 atomic_inc(&nl_table_users); 963 read_unlock(&nl_table_lock); 964 } 965 966 static inline void 967 netlink_unlock_table(void) 968 { 969 if (atomic_dec_and_test(&nl_table_users)) 970 wake_up(&nl_table_wait); 971 } 972 973 struct netlink_compare_arg 974 { 975 possible_net_t pnet; 976 u32 portid; 977 }; 978 979 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */ 980 #define netlink_compare_arg_len \ 981 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32)) 982 983 static inline int netlink_compare(struct rhashtable_compare_arg *arg, 984 const void *ptr) 985 { 986 const struct netlink_compare_arg *x = arg->key; 987 const struct netlink_sock *nlk = ptr; 988 989 return nlk->portid != x->portid || 990 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet)); 991 } 992 993 static void netlink_compare_arg_init(struct netlink_compare_arg *arg, 994 struct net *net, u32 portid) 995 { 996 memset(arg, 0, sizeof(*arg)); 997 write_pnet(&arg->pnet, net); 998 arg->portid = portid; 999 } 1000 1001 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid, 1002 struct net *net) 1003 { 1004 struct netlink_compare_arg arg; 1005 1006 netlink_compare_arg_init(&arg, net, portid); 1007 return rhashtable_lookup_fast(&table->hash, &arg, 1008 netlink_rhashtable_params); 1009 } 1010 1011 static int __netlink_insert(struct netlink_table *table, struct sock *sk) 1012 { 1013 struct netlink_compare_arg arg; 1014 1015 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid); 1016 return rhashtable_lookup_insert_key(&table->hash, &arg, 1017 &nlk_sk(sk)->node, 1018 netlink_rhashtable_params); 1019 } 1020 1021 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid) 1022 { 1023 struct netlink_table *table = &nl_table[protocol]; 1024 struct sock *sk; 1025 1026 rcu_read_lock(); 1027 sk = __netlink_lookup(table, portid, net); 1028 if (sk) 1029 sock_hold(sk); 1030 rcu_read_unlock(); 1031 1032 return sk; 1033 } 1034 1035 static const struct proto_ops netlink_ops; 1036 1037 static void 1038 netlink_update_listeners(struct sock *sk) 1039 { 1040 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 1041 unsigned long mask; 1042 unsigned int i; 1043 struct listeners *listeners; 1044 1045 listeners = nl_deref_protected(tbl->listeners); 1046 if (!listeners) 1047 return; 1048 1049 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) { 1050 mask = 0; 1051 sk_for_each_bound(sk, &tbl->mc_list) { 1052 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups)) 1053 mask |= nlk_sk(sk)->groups[i]; 1054 } 1055 listeners->masks[i] = mask; 1056 } 1057 /* this function is only called with the netlink table "grabbed", which 1058 * makes sure updates are visible before bind or setsockopt return. */ 1059 } 1060 1061 static int netlink_insert(struct sock *sk, u32 portid) 1062 { 1063 struct netlink_table *table = &nl_table[sk->sk_protocol]; 1064 int err; 1065 1066 lock_sock(sk); 1067 1068 err = -EBUSY; 1069 if (nlk_sk(sk)->portid) 1070 goto err; 1071 1072 err = -ENOMEM; 1073 if (BITS_PER_LONG > 32 && 1074 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX)) 1075 goto err; 1076 1077 nlk_sk(sk)->portid = portid; 1078 sock_hold(sk); 1079 1080 err = __netlink_insert(table, sk); 1081 if (err) { 1082 if (err == -EEXIST) 1083 err = -EADDRINUSE; 1084 nlk_sk(sk)->portid = 0; 1085 sock_put(sk); 1086 } 1087 1088 err: 1089 release_sock(sk); 1090 return err; 1091 } 1092 1093 static void netlink_remove(struct sock *sk) 1094 { 1095 struct netlink_table *table; 1096 1097 table = &nl_table[sk->sk_protocol]; 1098 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node, 1099 netlink_rhashtable_params)) { 1100 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 1101 __sock_put(sk); 1102 } 1103 1104 netlink_table_grab(); 1105 if (nlk_sk(sk)->subscriptions) { 1106 __sk_del_bind_node(sk); 1107 netlink_update_listeners(sk); 1108 } 1109 if (sk->sk_protocol == NETLINK_GENERIC) 1110 atomic_inc(&genl_sk_destructing_cnt); 1111 netlink_table_ungrab(); 1112 } 1113 1114 static struct proto netlink_proto = { 1115 .name = "NETLINK", 1116 .owner = THIS_MODULE, 1117 .obj_size = sizeof(struct netlink_sock), 1118 }; 1119 1120 static int __netlink_create(struct net *net, struct socket *sock, 1121 struct mutex *cb_mutex, int protocol) 1122 { 1123 struct sock *sk; 1124 struct netlink_sock *nlk; 1125 1126 sock->ops = &netlink_ops; 1127 1128 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto); 1129 if (!sk) 1130 return -ENOMEM; 1131 1132 sock_init_data(sock, sk); 1133 1134 nlk = nlk_sk(sk); 1135 if (cb_mutex) { 1136 nlk->cb_mutex = cb_mutex; 1137 } else { 1138 nlk->cb_mutex = &nlk->cb_def_mutex; 1139 mutex_init(nlk->cb_mutex); 1140 } 1141 init_waitqueue_head(&nlk->wait); 1142 #ifdef CONFIG_NETLINK_MMAP 1143 mutex_init(&nlk->pg_vec_lock); 1144 #endif 1145 1146 sk->sk_destruct = netlink_sock_destruct; 1147 sk->sk_protocol = protocol; 1148 return 0; 1149 } 1150 1151 static int netlink_create(struct net *net, struct socket *sock, int protocol, 1152 int kern) 1153 { 1154 struct module *module = NULL; 1155 struct mutex *cb_mutex; 1156 struct netlink_sock *nlk; 1157 int (*bind)(struct net *net, int group); 1158 void (*unbind)(struct net *net, int group); 1159 int err = 0; 1160 1161 sock->state = SS_UNCONNECTED; 1162 1163 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 1164 return -ESOCKTNOSUPPORT; 1165 1166 if (protocol < 0 || protocol >= MAX_LINKS) 1167 return -EPROTONOSUPPORT; 1168 1169 netlink_lock_table(); 1170 #ifdef CONFIG_MODULES 1171 if (!nl_table[protocol].registered) { 1172 netlink_unlock_table(); 1173 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol); 1174 netlink_lock_table(); 1175 } 1176 #endif 1177 if (nl_table[protocol].registered && 1178 try_module_get(nl_table[protocol].module)) 1179 module = nl_table[protocol].module; 1180 else 1181 err = -EPROTONOSUPPORT; 1182 cb_mutex = nl_table[protocol].cb_mutex; 1183 bind = nl_table[protocol].bind; 1184 unbind = nl_table[protocol].unbind; 1185 netlink_unlock_table(); 1186 1187 if (err < 0) 1188 goto out; 1189 1190 err = __netlink_create(net, sock, cb_mutex, protocol); 1191 if (err < 0) 1192 goto out_module; 1193 1194 local_bh_disable(); 1195 sock_prot_inuse_add(net, &netlink_proto, 1); 1196 local_bh_enable(); 1197 1198 nlk = nlk_sk(sock->sk); 1199 nlk->module = module; 1200 nlk->netlink_bind = bind; 1201 nlk->netlink_unbind = unbind; 1202 out: 1203 return err; 1204 1205 out_module: 1206 module_put(module); 1207 goto out; 1208 } 1209 1210 static void deferred_put_nlk_sk(struct rcu_head *head) 1211 { 1212 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu); 1213 1214 sock_put(&nlk->sk); 1215 } 1216 1217 static int netlink_release(struct socket *sock) 1218 { 1219 struct sock *sk = sock->sk; 1220 struct netlink_sock *nlk; 1221 1222 if (!sk) 1223 return 0; 1224 1225 netlink_remove(sk); 1226 sock_orphan(sk); 1227 nlk = nlk_sk(sk); 1228 1229 /* 1230 * OK. Socket is unlinked, any packets that arrive now 1231 * will be purged. 1232 */ 1233 1234 /* must not acquire netlink_table_lock in any way again before unbind 1235 * and notifying genetlink is done as otherwise it might deadlock 1236 */ 1237 if (nlk->netlink_unbind) { 1238 int i; 1239 1240 for (i = 0; i < nlk->ngroups; i++) 1241 if (test_bit(i, nlk->groups)) 1242 nlk->netlink_unbind(sock_net(sk), i + 1); 1243 } 1244 if (sk->sk_protocol == NETLINK_GENERIC && 1245 atomic_dec_return(&genl_sk_destructing_cnt) == 0) 1246 wake_up(&genl_sk_destructing_waitq); 1247 1248 sock->sk = NULL; 1249 wake_up_interruptible_all(&nlk->wait); 1250 1251 skb_queue_purge(&sk->sk_write_queue); 1252 1253 if (nlk->portid) { 1254 struct netlink_notify n = { 1255 .net = sock_net(sk), 1256 .protocol = sk->sk_protocol, 1257 .portid = nlk->portid, 1258 }; 1259 atomic_notifier_call_chain(&netlink_chain, 1260 NETLINK_URELEASE, &n); 1261 } 1262 1263 module_put(nlk->module); 1264 1265 if (netlink_is_kernel(sk)) { 1266 netlink_table_grab(); 1267 BUG_ON(nl_table[sk->sk_protocol].registered == 0); 1268 if (--nl_table[sk->sk_protocol].registered == 0) { 1269 struct listeners *old; 1270 1271 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners); 1272 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL); 1273 kfree_rcu(old, rcu); 1274 nl_table[sk->sk_protocol].module = NULL; 1275 nl_table[sk->sk_protocol].bind = NULL; 1276 nl_table[sk->sk_protocol].unbind = NULL; 1277 nl_table[sk->sk_protocol].flags = 0; 1278 nl_table[sk->sk_protocol].registered = 0; 1279 } 1280 netlink_table_ungrab(); 1281 } 1282 1283 kfree(nlk->groups); 1284 nlk->groups = NULL; 1285 1286 local_bh_disable(); 1287 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1); 1288 local_bh_enable(); 1289 call_rcu(&nlk->rcu, deferred_put_nlk_sk); 1290 return 0; 1291 } 1292 1293 static int netlink_autobind(struct socket *sock) 1294 { 1295 struct sock *sk = sock->sk; 1296 struct net *net = sock_net(sk); 1297 struct netlink_table *table = &nl_table[sk->sk_protocol]; 1298 s32 portid = task_tgid_vnr(current); 1299 int err; 1300 static s32 rover = -4097; 1301 1302 retry: 1303 cond_resched(); 1304 rcu_read_lock(); 1305 if (__netlink_lookup(table, portid, net)) { 1306 /* Bind collision, search negative portid values. */ 1307 portid = rover--; 1308 if (rover > -4097) 1309 rover = -4097; 1310 rcu_read_unlock(); 1311 goto retry; 1312 } 1313 rcu_read_unlock(); 1314 1315 err = netlink_insert(sk, portid); 1316 if (err == -EADDRINUSE) 1317 goto retry; 1318 1319 /* If 2 threads race to autobind, that is fine. */ 1320 if (err == -EBUSY) 1321 err = 0; 1322 1323 return err; 1324 } 1325 1326 /** 1327 * __netlink_ns_capable - General netlink message capability test 1328 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace. 1329 * @user_ns: The user namespace of the capability to use 1330 * @cap: The capability to use 1331 * 1332 * Test to see if the opener of the socket we received the message 1333 * from had when the netlink socket was created and the sender of the 1334 * message has has the capability @cap in the user namespace @user_ns. 1335 */ 1336 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp, 1337 struct user_namespace *user_ns, int cap) 1338 { 1339 return ((nsp->flags & NETLINK_SKB_DST) || 1340 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) && 1341 ns_capable(user_ns, cap); 1342 } 1343 EXPORT_SYMBOL(__netlink_ns_capable); 1344 1345 /** 1346 * netlink_ns_capable - General netlink message capability test 1347 * @skb: socket buffer holding a netlink command from userspace 1348 * @user_ns: The user namespace of the capability to use 1349 * @cap: The capability to use 1350 * 1351 * Test to see if the opener of the socket we received the message 1352 * from had when the netlink socket was created and the sender of the 1353 * message has has the capability @cap in the user namespace @user_ns. 1354 */ 1355 bool netlink_ns_capable(const struct sk_buff *skb, 1356 struct user_namespace *user_ns, int cap) 1357 { 1358 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap); 1359 } 1360 EXPORT_SYMBOL(netlink_ns_capable); 1361 1362 /** 1363 * netlink_capable - Netlink global message capability test 1364 * @skb: socket buffer holding a netlink command from userspace 1365 * @cap: The capability to use 1366 * 1367 * Test to see if the opener of the socket we received the message 1368 * from had when the netlink socket was created and the sender of the 1369 * message has has the capability @cap in all user namespaces. 1370 */ 1371 bool netlink_capable(const struct sk_buff *skb, int cap) 1372 { 1373 return netlink_ns_capable(skb, &init_user_ns, cap); 1374 } 1375 EXPORT_SYMBOL(netlink_capable); 1376 1377 /** 1378 * netlink_net_capable - Netlink network namespace message capability test 1379 * @skb: socket buffer holding a netlink command from userspace 1380 * @cap: The capability to use 1381 * 1382 * Test to see if the opener of the socket we received the message 1383 * from had when the netlink socket was created and the sender of the 1384 * message has has the capability @cap over the network namespace of 1385 * the socket we received the message from. 1386 */ 1387 bool netlink_net_capable(const struct sk_buff *skb, int cap) 1388 { 1389 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap); 1390 } 1391 EXPORT_SYMBOL(netlink_net_capable); 1392 1393 static inline int netlink_allowed(const struct socket *sock, unsigned int flag) 1394 { 1395 return (nl_table[sock->sk->sk_protocol].flags & flag) || 1396 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN); 1397 } 1398 1399 static void 1400 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions) 1401 { 1402 struct netlink_sock *nlk = nlk_sk(sk); 1403 1404 if (nlk->subscriptions && !subscriptions) 1405 __sk_del_bind_node(sk); 1406 else if (!nlk->subscriptions && subscriptions) 1407 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); 1408 nlk->subscriptions = subscriptions; 1409 } 1410 1411 static int netlink_realloc_groups(struct sock *sk) 1412 { 1413 struct netlink_sock *nlk = nlk_sk(sk); 1414 unsigned int groups; 1415 unsigned long *new_groups; 1416 int err = 0; 1417 1418 netlink_table_grab(); 1419 1420 groups = nl_table[sk->sk_protocol].groups; 1421 if (!nl_table[sk->sk_protocol].registered) { 1422 err = -ENOENT; 1423 goto out_unlock; 1424 } 1425 1426 if (nlk->ngroups >= groups) 1427 goto out_unlock; 1428 1429 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC); 1430 if (new_groups == NULL) { 1431 err = -ENOMEM; 1432 goto out_unlock; 1433 } 1434 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0, 1435 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups)); 1436 1437 nlk->groups = new_groups; 1438 nlk->ngroups = groups; 1439 out_unlock: 1440 netlink_table_ungrab(); 1441 return err; 1442 } 1443 1444 static void netlink_undo_bind(int group, long unsigned int groups, 1445 struct sock *sk) 1446 { 1447 struct netlink_sock *nlk = nlk_sk(sk); 1448 int undo; 1449 1450 if (!nlk->netlink_unbind) 1451 return; 1452 1453 for (undo = 0; undo < group; undo++) 1454 if (test_bit(undo, &groups)) 1455 nlk->netlink_unbind(sock_net(sk), undo + 1); 1456 } 1457 1458 static int netlink_bind(struct socket *sock, struct sockaddr *addr, 1459 int addr_len) 1460 { 1461 struct sock *sk = sock->sk; 1462 struct net *net = sock_net(sk); 1463 struct netlink_sock *nlk = nlk_sk(sk); 1464 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 1465 int err; 1466 long unsigned int groups = nladdr->nl_groups; 1467 1468 if (addr_len < sizeof(struct sockaddr_nl)) 1469 return -EINVAL; 1470 1471 if (nladdr->nl_family != AF_NETLINK) 1472 return -EINVAL; 1473 1474 /* Only superuser is allowed to listen multicasts */ 1475 if (groups) { 1476 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV)) 1477 return -EPERM; 1478 err = netlink_realloc_groups(sk); 1479 if (err) 1480 return err; 1481 } 1482 1483 if (nlk->portid) 1484 if (nladdr->nl_pid != nlk->portid) 1485 return -EINVAL; 1486 1487 if (nlk->netlink_bind && groups) { 1488 int group; 1489 1490 for (group = 0; group < nlk->ngroups; group++) { 1491 if (!test_bit(group, &groups)) 1492 continue; 1493 err = nlk->netlink_bind(net, group + 1); 1494 if (!err) 1495 continue; 1496 netlink_undo_bind(group, groups, sk); 1497 return err; 1498 } 1499 } 1500 1501 if (!nlk->portid) { 1502 err = nladdr->nl_pid ? 1503 netlink_insert(sk, nladdr->nl_pid) : 1504 netlink_autobind(sock); 1505 if (err) { 1506 netlink_undo_bind(nlk->ngroups, groups, sk); 1507 return err; 1508 } 1509 } 1510 1511 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0])) 1512 return 0; 1513 1514 netlink_table_grab(); 1515 netlink_update_subscriptions(sk, nlk->subscriptions + 1516 hweight32(groups) - 1517 hweight32(nlk->groups[0])); 1518 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups; 1519 netlink_update_listeners(sk); 1520 netlink_table_ungrab(); 1521 1522 return 0; 1523 } 1524 1525 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 1526 int alen, int flags) 1527 { 1528 int err = 0; 1529 struct sock *sk = sock->sk; 1530 struct netlink_sock *nlk = nlk_sk(sk); 1531 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 1532 1533 if (alen < sizeof(addr->sa_family)) 1534 return -EINVAL; 1535 1536 if (addr->sa_family == AF_UNSPEC) { 1537 sk->sk_state = NETLINK_UNCONNECTED; 1538 nlk->dst_portid = 0; 1539 nlk->dst_group = 0; 1540 return 0; 1541 } 1542 if (addr->sa_family != AF_NETLINK) 1543 return -EINVAL; 1544 1545 if ((nladdr->nl_groups || nladdr->nl_pid) && 1546 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND)) 1547 return -EPERM; 1548 1549 if (!nlk->portid) 1550 err = netlink_autobind(sock); 1551 1552 if (err == 0) { 1553 sk->sk_state = NETLINK_CONNECTED; 1554 nlk->dst_portid = nladdr->nl_pid; 1555 nlk->dst_group = ffs(nladdr->nl_groups); 1556 } 1557 1558 return err; 1559 } 1560 1561 static int netlink_getname(struct socket *sock, struct sockaddr *addr, 1562 int *addr_len, int peer) 1563 { 1564 struct sock *sk = sock->sk; 1565 struct netlink_sock *nlk = nlk_sk(sk); 1566 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr); 1567 1568 nladdr->nl_family = AF_NETLINK; 1569 nladdr->nl_pad = 0; 1570 *addr_len = sizeof(*nladdr); 1571 1572 if (peer) { 1573 nladdr->nl_pid = nlk->dst_portid; 1574 nladdr->nl_groups = netlink_group_mask(nlk->dst_group); 1575 } else { 1576 nladdr->nl_pid = nlk->portid; 1577 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0; 1578 } 1579 return 0; 1580 } 1581 1582 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid) 1583 { 1584 struct sock *sock; 1585 struct netlink_sock *nlk; 1586 1587 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid); 1588 if (!sock) 1589 return ERR_PTR(-ECONNREFUSED); 1590 1591 /* Don't bother queuing skb if kernel socket has no input function */ 1592 nlk = nlk_sk(sock); 1593 if (sock->sk_state == NETLINK_CONNECTED && 1594 nlk->dst_portid != nlk_sk(ssk)->portid) { 1595 sock_put(sock); 1596 return ERR_PTR(-ECONNREFUSED); 1597 } 1598 return sock; 1599 } 1600 1601 struct sock *netlink_getsockbyfilp(struct file *filp) 1602 { 1603 struct inode *inode = file_inode(filp); 1604 struct sock *sock; 1605 1606 if (!S_ISSOCK(inode->i_mode)) 1607 return ERR_PTR(-ENOTSOCK); 1608 1609 sock = SOCKET_I(inode)->sk; 1610 if (sock->sk_family != AF_NETLINK) 1611 return ERR_PTR(-EINVAL); 1612 1613 sock_hold(sock); 1614 return sock; 1615 } 1616 1617 static struct sk_buff *netlink_alloc_large_skb(unsigned int size, 1618 int broadcast) 1619 { 1620 struct sk_buff *skb; 1621 void *data; 1622 1623 if (size <= NLMSG_GOODSIZE || broadcast) 1624 return alloc_skb(size, GFP_KERNEL); 1625 1626 size = SKB_DATA_ALIGN(size) + 1627 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 1628 1629 data = vmalloc(size); 1630 if (data == NULL) 1631 return NULL; 1632 1633 skb = __build_skb(data, size); 1634 if (skb == NULL) 1635 vfree(data); 1636 else 1637 skb->destructor = netlink_skb_destructor; 1638 1639 return skb; 1640 } 1641 1642 /* 1643 * Attach a skb to a netlink socket. 1644 * The caller must hold a reference to the destination socket. On error, the 1645 * reference is dropped. The skb is not send to the destination, just all 1646 * all error checks are performed and memory in the queue is reserved. 1647 * Return values: 1648 * < 0: error. skb freed, reference to sock dropped. 1649 * 0: continue 1650 * 1: repeat lookup - reference dropped while waiting for socket memory. 1651 */ 1652 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, 1653 long *timeo, struct sock *ssk) 1654 { 1655 struct netlink_sock *nlk; 1656 1657 nlk = nlk_sk(sk); 1658 1659 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 1660 test_bit(NETLINK_CONGESTED, &nlk->state)) && 1661 !netlink_skb_is_mmaped(skb)) { 1662 DECLARE_WAITQUEUE(wait, current); 1663 if (!*timeo) { 1664 if (!ssk || netlink_is_kernel(ssk)) 1665 netlink_overrun(sk); 1666 sock_put(sk); 1667 kfree_skb(skb); 1668 return -EAGAIN; 1669 } 1670 1671 __set_current_state(TASK_INTERRUPTIBLE); 1672 add_wait_queue(&nlk->wait, &wait); 1673 1674 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 1675 test_bit(NETLINK_CONGESTED, &nlk->state)) && 1676 !sock_flag(sk, SOCK_DEAD)) 1677 *timeo = schedule_timeout(*timeo); 1678 1679 __set_current_state(TASK_RUNNING); 1680 remove_wait_queue(&nlk->wait, &wait); 1681 sock_put(sk); 1682 1683 if (signal_pending(current)) { 1684 kfree_skb(skb); 1685 return sock_intr_errno(*timeo); 1686 } 1687 return 1; 1688 } 1689 netlink_skb_set_owner_r(skb, sk); 1690 return 0; 1691 } 1692 1693 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb) 1694 { 1695 int len = skb->len; 1696 1697 netlink_deliver_tap(skb); 1698 1699 #ifdef CONFIG_NETLINK_MMAP 1700 if (netlink_skb_is_mmaped(skb)) 1701 netlink_queue_mmaped_skb(sk, skb); 1702 else if (netlink_rx_is_mmaped(sk)) 1703 netlink_ring_set_copied(sk, skb); 1704 else 1705 #endif /* CONFIG_NETLINK_MMAP */ 1706 skb_queue_tail(&sk->sk_receive_queue, skb); 1707 sk->sk_data_ready(sk); 1708 return len; 1709 } 1710 1711 int netlink_sendskb(struct sock *sk, struct sk_buff *skb) 1712 { 1713 int len = __netlink_sendskb(sk, skb); 1714 1715 sock_put(sk); 1716 return len; 1717 } 1718 1719 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 1720 { 1721 kfree_skb(skb); 1722 sock_put(sk); 1723 } 1724 1725 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation) 1726 { 1727 int delta; 1728 1729 WARN_ON(skb->sk != NULL); 1730 if (netlink_skb_is_mmaped(skb)) 1731 return skb; 1732 1733 delta = skb->end - skb->tail; 1734 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize) 1735 return skb; 1736 1737 if (skb_shared(skb)) { 1738 struct sk_buff *nskb = skb_clone(skb, allocation); 1739 if (!nskb) 1740 return skb; 1741 consume_skb(skb); 1742 skb = nskb; 1743 } 1744 1745 if (!pskb_expand_head(skb, 0, -delta, allocation)) 1746 skb->truesize -= delta; 1747 1748 return skb; 1749 } 1750 1751 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb, 1752 struct sock *ssk) 1753 { 1754 int ret; 1755 struct netlink_sock *nlk = nlk_sk(sk); 1756 1757 ret = -ECONNREFUSED; 1758 if (nlk->netlink_rcv != NULL) { 1759 ret = skb->len; 1760 netlink_skb_set_owner_r(skb, sk); 1761 NETLINK_CB(skb).sk = ssk; 1762 netlink_deliver_tap_kernel(sk, ssk, skb); 1763 nlk->netlink_rcv(skb); 1764 consume_skb(skb); 1765 } else { 1766 kfree_skb(skb); 1767 } 1768 sock_put(sk); 1769 return ret; 1770 } 1771 1772 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, 1773 u32 portid, int nonblock) 1774 { 1775 struct sock *sk; 1776 int err; 1777 long timeo; 1778 1779 skb = netlink_trim(skb, gfp_any()); 1780 1781 timeo = sock_sndtimeo(ssk, nonblock); 1782 retry: 1783 sk = netlink_getsockbyportid(ssk, portid); 1784 if (IS_ERR(sk)) { 1785 kfree_skb(skb); 1786 return PTR_ERR(sk); 1787 } 1788 if (netlink_is_kernel(sk)) 1789 return netlink_unicast_kernel(sk, skb, ssk); 1790 1791 if (sk_filter(sk, skb)) { 1792 err = skb->len; 1793 kfree_skb(skb); 1794 sock_put(sk); 1795 return err; 1796 } 1797 1798 err = netlink_attachskb(sk, skb, &timeo, ssk); 1799 if (err == 1) 1800 goto retry; 1801 if (err) 1802 return err; 1803 1804 return netlink_sendskb(sk, skb); 1805 } 1806 EXPORT_SYMBOL(netlink_unicast); 1807 1808 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size, 1809 u32 dst_portid, gfp_t gfp_mask) 1810 { 1811 #ifdef CONFIG_NETLINK_MMAP 1812 struct sock *sk = NULL; 1813 struct sk_buff *skb; 1814 struct netlink_ring *ring; 1815 struct nl_mmap_hdr *hdr; 1816 unsigned int maxlen; 1817 1818 sk = netlink_getsockbyportid(ssk, dst_portid); 1819 if (IS_ERR(sk)) 1820 goto out; 1821 1822 ring = &nlk_sk(sk)->rx_ring; 1823 /* fast-path without atomic ops for common case: non-mmaped receiver */ 1824 if (ring->pg_vec == NULL) 1825 goto out_put; 1826 1827 if (ring->frame_size - NL_MMAP_HDRLEN < size) 1828 goto out_put; 1829 1830 skb = alloc_skb_head(gfp_mask); 1831 if (skb == NULL) 1832 goto err1; 1833 1834 spin_lock_bh(&sk->sk_receive_queue.lock); 1835 /* check again under lock */ 1836 if (ring->pg_vec == NULL) 1837 goto out_free; 1838 1839 /* check again under lock */ 1840 maxlen = ring->frame_size - NL_MMAP_HDRLEN; 1841 if (maxlen < size) 1842 goto out_free; 1843 1844 netlink_forward_ring(ring); 1845 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED); 1846 if (hdr == NULL) 1847 goto err2; 1848 netlink_ring_setup_skb(skb, sk, ring, hdr); 1849 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED); 1850 atomic_inc(&ring->pending); 1851 netlink_increment_head(ring); 1852 1853 spin_unlock_bh(&sk->sk_receive_queue.lock); 1854 return skb; 1855 1856 err2: 1857 kfree_skb(skb); 1858 spin_unlock_bh(&sk->sk_receive_queue.lock); 1859 netlink_overrun(sk); 1860 err1: 1861 sock_put(sk); 1862 return NULL; 1863 1864 out_free: 1865 kfree_skb(skb); 1866 spin_unlock_bh(&sk->sk_receive_queue.lock); 1867 out_put: 1868 sock_put(sk); 1869 out: 1870 #endif 1871 return alloc_skb(size, gfp_mask); 1872 } 1873 EXPORT_SYMBOL_GPL(netlink_alloc_skb); 1874 1875 int netlink_has_listeners(struct sock *sk, unsigned int group) 1876 { 1877 int res = 0; 1878 struct listeners *listeners; 1879 1880 BUG_ON(!netlink_is_kernel(sk)); 1881 1882 rcu_read_lock(); 1883 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners); 1884 1885 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups) 1886 res = test_bit(group - 1, listeners->masks); 1887 1888 rcu_read_unlock(); 1889 1890 return res; 1891 } 1892 EXPORT_SYMBOL_GPL(netlink_has_listeners); 1893 1894 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 1895 { 1896 struct netlink_sock *nlk = nlk_sk(sk); 1897 1898 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 1899 !test_bit(NETLINK_CONGESTED, &nlk->state)) { 1900 netlink_skb_set_owner_r(skb, sk); 1901 __netlink_sendskb(sk, skb); 1902 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1); 1903 } 1904 return -1; 1905 } 1906 1907 struct netlink_broadcast_data { 1908 struct sock *exclude_sk; 1909 struct net *net; 1910 u32 portid; 1911 u32 group; 1912 int failure; 1913 int delivery_failure; 1914 int congested; 1915 int delivered; 1916 gfp_t allocation; 1917 struct sk_buff *skb, *skb2; 1918 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data); 1919 void *tx_data; 1920 }; 1921 1922 static void do_one_broadcast(struct sock *sk, 1923 struct netlink_broadcast_data *p) 1924 { 1925 struct netlink_sock *nlk = nlk_sk(sk); 1926 int val; 1927 1928 if (p->exclude_sk == sk) 1929 return; 1930 1931 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || 1932 !test_bit(p->group - 1, nlk->groups)) 1933 return; 1934 1935 if (!net_eq(sock_net(sk), p->net)) 1936 return; 1937 1938 if (p->failure) { 1939 netlink_overrun(sk); 1940 return; 1941 } 1942 1943 sock_hold(sk); 1944 if (p->skb2 == NULL) { 1945 if (skb_shared(p->skb)) { 1946 p->skb2 = skb_clone(p->skb, p->allocation); 1947 } else { 1948 p->skb2 = skb_get(p->skb); 1949 /* 1950 * skb ownership may have been set when 1951 * delivered to a previous socket. 1952 */ 1953 skb_orphan(p->skb2); 1954 } 1955 } 1956 if (p->skb2 == NULL) { 1957 netlink_overrun(sk); 1958 /* Clone failed. Notify ALL listeners. */ 1959 p->failure = 1; 1960 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR) 1961 p->delivery_failure = 1; 1962 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) { 1963 kfree_skb(p->skb2); 1964 p->skb2 = NULL; 1965 } else if (sk_filter(sk, p->skb2)) { 1966 kfree_skb(p->skb2); 1967 p->skb2 = NULL; 1968 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) { 1969 netlink_overrun(sk); 1970 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR) 1971 p->delivery_failure = 1; 1972 } else { 1973 p->congested |= val; 1974 p->delivered = 1; 1975 p->skb2 = NULL; 1976 } 1977 sock_put(sk); 1978 } 1979 1980 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid, 1981 u32 group, gfp_t allocation, 1982 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data), 1983 void *filter_data) 1984 { 1985 struct net *net = sock_net(ssk); 1986 struct netlink_broadcast_data info; 1987 struct sock *sk; 1988 1989 skb = netlink_trim(skb, allocation); 1990 1991 info.exclude_sk = ssk; 1992 info.net = net; 1993 info.portid = portid; 1994 info.group = group; 1995 info.failure = 0; 1996 info.delivery_failure = 0; 1997 info.congested = 0; 1998 info.delivered = 0; 1999 info.allocation = allocation; 2000 info.skb = skb; 2001 info.skb2 = NULL; 2002 info.tx_filter = filter; 2003 info.tx_data = filter_data; 2004 2005 /* While we sleep in clone, do not allow to change socket list */ 2006 2007 netlink_lock_table(); 2008 2009 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list) 2010 do_one_broadcast(sk, &info); 2011 2012 consume_skb(skb); 2013 2014 netlink_unlock_table(); 2015 2016 if (info.delivery_failure) { 2017 kfree_skb(info.skb2); 2018 return -ENOBUFS; 2019 } 2020 consume_skb(info.skb2); 2021 2022 if (info.delivered) { 2023 if (info.congested && (allocation & __GFP_WAIT)) 2024 yield(); 2025 return 0; 2026 } 2027 return -ESRCH; 2028 } 2029 EXPORT_SYMBOL(netlink_broadcast_filtered); 2030 2031 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid, 2032 u32 group, gfp_t allocation) 2033 { 2034 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation, 2035 NULL, NULL); 2036 } 2037 EXPORT_SYMBOL(netlink_broadcast); 2038 2039 struct netlink_set_err_data { 2040 struct sock *exclude_sk; 2041 u32 portid; 2042 u32 group; 2043 int code; 2044 }; 2045 2046 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p) 2047 { 2048 struct netlink_sock *nlk = nlk_sk(sk); 2049 int ret = 0; 2050 2051 if (sk == p->exclude_sk) 2052 goto out; 2053 2054 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk))) 2055 goto out; 2056 2057 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || 2058 !test_bit(p->group - 1, nlk->groups)) 2059 goto out; 2060 2061 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) { 2062 ret = 1; 2063 goto out; 2064 } 2065 2066 sk->sk_err = p->code; 2067 sk->sk_error_report(sk); 2068 out: 2069 return ret; 2070 } 2071 2072 /** 2073 * netlink_set_err - report error to broadcast listeners 2074 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create() 2075 * @portid: the PORTID of a process that we want to skip (if any) 2076 * @group: the broadcast group that will notice the error 2077 * @code: error code, must be negative (as usual in kernelspace) 2078 * 2079 * This function returns the number of broadcast listeners that have set the 2080 * NETLINK_RECV_NO_ENOBUFS socket option. 2081 */ 2082 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code) 2083 { 2084 struct netlink_set_err_data info; 2085 struct sock *sk; 2086 int ret = 0; 2087 2088 info.exclude_sk = ssk; 2089 info.portid = portid; 2090 info.group = group; 2091 /* sk->sk_err wants a positive error value */ 2092 info.code = -code; 2093 2094 read_lock(&nl_table_lock); 2095 2096 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list) 2097 ret += do_one_set_err(sk, &info); 2098 2099 read_unlock(&nl_table_lock); 2100 return ret; 2101 } 2102 EXPORT_SYMBOL(netlink_set_err); 2103 2104 /* must be called with netlink table grabbed */ 2105 static void netlink_update_socket_mc(struct netlink_sock *nlk, 2106 unsigned int group, 2107 int is_new) 2108 { 2109 int old, new = !!is_new, subscriptions; 2110 2111 old = test_bit(group - 1, nlk->groups); 2112 subscriptions = nlk->subscriptions - old + new; 2113 if (new) 2114 __set_bit(group - 1, nlk->groups); 2115 else 2116 __clear_bit(group - 1, nlk->groups); 2117 netlink_update_subscriptions(&nlk->sk, subscriptions); 2118 netlink_update_listeners(&nlk->sk); 2119 } 2120 2121 static int netlink_setsockopt(struct socket *sock, int level, int optname, 2122 char __user *optval, unsigned int optlen) 2123 { 2124 struct sock *sk = sock->sk; 2125 struct netlink_sock *nlk = nlk_sk(sk); 2126 unsigned int val = 0; 2127 int err; 2128 2129 if (level != SOL_NETLINK) 2130 return -ENOPROTOOPT; 2131 2132 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING && 2133 optlen >= sizeof(int) && 2134 get_user(val, (unsigned int __user *)optval)) 2135 return -EFAULT; 2136 2137 switch (optname) { 2138 case NETLINK_PKTINFO: 2139 if (val) 2140 nlk->flags |= NETLINK_RECV_PKTINFO; 2141 else 2142 nlk->flags &= ~NETLINK_RECV_PKTINFO; 2143 err = 0; 2144 break; 2145 case NETLINK_ADD_MEMBERSHIP: 2146 case NETLINK_DROP_MEMBERSHIP: { 2147 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV)) 2148 return -EPERM; 2149 err = netlink_realloc_groups(sk); 2150 if (err) 2151 return err; 2152 if (!val || val - 1 >= nlk->ngroups) 2153 return -EINVAL; 2154 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) { 2155 err = nlk->netlink_bind(sock_net(sk), val); 2156 if (err) 2157 return err; 2158 } 2159 netlink_table_grab(); 2160 netlink_update_socket_mc(nlk, val, 2161 optname == NETLINK_ADD_MEMBERSHIP); 2162 netlink_table_ungrab(); 2163 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind) 2164 nlk->netlink_unbind(sock_net(sk), val); 2165 2166 err = 0; 2167 break; 2168 } 2169 case NETLINK_BROADCAST_ERROR: 2170 if (val) 2171 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR; 2172 else 2173 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR; 2174 err = 0; 2175 break; 2176 case NETLINK_NO_ENOBUFS: 2177 if (val) { 2178 nlk->flags |= NETLINK_RECV_NO_ENOBUFS; 2179 clear_bit(NETLINK_CONGESTED, &nlk->state); 2180 wake_up_interruptible(&nlk->wait); 2181 } else { 2182 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS; 2183 } 2184 err = 0; 2185 break; 2186 #ifdef CONFIG_NETLINK_MMAP 2187 case NETLINK_RX_RING: 2188 case NETLINK_TX_RING: { 2189 struct nl_mmap_req req; 2190 2191 /* Rings might consume more memory than queue limits, require 2192 * CAP_NET_ADMIN. 2193 */ 2194 if (!capable(CAP_NET_ADMIN)) 2195 return -EPERM; 2196 if (optlen < sizeof(req)) 2197 return -EINVAL; 2198 if (copy_from_user(&req, optval, sizeof(req))) 2199 return -EFAULT; 2200 err = netlink_set_ring(sk, &req, false, 2201 optname == NETLINK_TX_RING); 2202 break; 2203 } 2204 #endif /* CONFIG_NETLINK_MMAP */ 2205 default: 2206 err = -ENOPROTOOPT; 2207 } 2208 return err; 2209 } 2210 2211 static int netlink_getsockopt(struct socket *sock, int level, int optname, 2212 char __user *optval, int __user *optlen) 2213 { 2214 struct sock *sk = sock->sk; 2215 struct netlink_sock *nlk = nlk_sk(sk); 2216 int len, val, err; 2217 2218 if (level != SOL_NETLINK) 2219 return -ENOPROTOOPT; 2220 2221 if (get_user(len, optlen)) 2222 return -EFAULT; 2223 if (len < 0) 2224 return -EINVAL; 2225 2226 switch (optname) { 2227 case NETLINK_PKTINFO: 2228 if (len < sizeof(int)) 2229 return -EINVAL; 2230 len = sizeof(int); 2231 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0; 2232 if (put_user(len, optlen) || 2233 put_user(val, optval)) 2234 return -EFAULT; 2235 err = 0; 2236 break; 2237 case NETLINK_BROADCAST_ERROR: 2238 if (len < sizeof(int)) 2239 return -EINVAL; 2240 len = sizeof(int); 2241 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0; 2242 if (put_user(len, optlen) || 2243 put_user(val, optval)) 2244 return -EFAULT; 2245 err = 0; 2246 break; 2247 case NETLINK_NO_ENOBUFS: 2248 if (len < sizeof(int)) 2249 return -EINVAL; 2250 len = sizeof(int); 2251 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0; 2252 if (put_user(len, optlen) || 2253 put_user(val, optval)) 2254 return -EFAULT; 2255 err = 0; 2256 break; 2257 default: 2258 err = -ENOPROTOOPT; 2259 } 2260 return err; 2261 } 2262 2263 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb) 2264 { 2265 struct nl_pktinfo info; 2266 2267 info.group = NETLINK_CB(skb).dst_group; 2268 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info); 2269 } 2270 2271 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2272 { 2273 struct sock *sk = sock->sk; 2274 struct netlink_sock *nlk = nlk_sk(sk); 2275 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name); 2276 u32 dst_portid; 2277 u32 dst_group; 2278 struct sk_buff *skb; 2279 int err; 2280 struct scm_cookie scm; 2281 u32 netlink_skb_flags = 0; 2282 2283 if (msg->msg_flags&MSG_OOB) 2284 return -EOPNOTSUPP; 2285 2286 err = scm_send(sock, msg, &scm, true); 2287 if (err < 0) 2288 return err; 2289 2290 if (msg->msg_namelen) { 2291 err = -EINVAL; 2292 if (addr->nl_family != AF_NETLINK) 2293 goto out; 2294 dst_portid = addr->nl_pid; 2295 dst_group = ffs(addr->nl_groups); 2296 err = -EPERM; 2297 if ((dst_group || dst_portid) && 2298 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND)) 2299 goto out; 2300 netlink_skb_flags |= NETLINK_SKB_DST; 2301 } else { 2302 dst_portid = nlk->dst_portid; 2303 dst_group = nlk->dst_group; 2304 } 2305 2306 if (!nlk->portid) { 2307 err = netlink_autobind(sock); 2308 if (err) 2309 goto out; 2310 } 2311 2312 /* It's a really convoluted way for userland to ask for mmaped 2313 * sendmsg(), but that's what we've got... 2314 */ 2315 if (netlink_tx_is_mmaped(sk) && 2316 msg->msg_iter.type == ITER_IOVEC && 2317 msg->msg_iter.nr_segs == 1 && 2318 msg->msg_iter.iov->iov_base == NULL) { 2319 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, 2320 &scm); 2321 goto out; 2322 } 2323 2324 err = -EMSGSIZE; 2325 if (len > sk->sk_sndbuf - 32) 2326 goto out; 2327 err = -ENOBUFS; 2328 skb = netlink_alloc_large_skb(len, dst_group); 2329 if (skb == NULL) 2330 goto out; 2331 2332 NETLINK_CB(skb).portid = nlk->portid; 2333 NETLINK_CB(skb).dst_group = dst_group; 2334 NETLINK_CB(skb).creds = scm.creds; 2335 NETLINK_CB(skb).flags = netlink_skb_flags; 2336 2337 err = -EFAULT; 2338 if (memcpy_from_msg(skb_put(skb, len), msg, len)) { 2339 kfree_skb(skb); 2340 goto out; 2341 } 2342 2343 err = security_netlink_send(sk, skb); 2344 if (err) { 2345 kfree_skb(skb); 2346 goto out; 2347 } 2348 2349 if (dst_group) { 2350 atomic_inc(&skb->users); 2351 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL); 2352 } 2353 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT); 2354 2355 out: 2356 scm_destroy(&scm); 2357 return err; 2358 } 2359 2360 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2361 int flags) 2362 { 2363 struct scm_cookie scm; 2364 struct sock *sk = sock->sk; 2365 struct netlink_sock *nlk = nlk_sk(sk); 2366 int noblock = flags&MSG_DONTWAIT; 2367 size_t copied; 2368 struct sk_buff *skb, *data_skb; 2369 int err, ret; 2370 2371 if (flags&MSG_OOB) 2372 return -EOPNOTSUPP; 2373 2374 copied = 0; 2375 2376 skb = skb_recv_datagram(sk, flags, noblock, &err); 2377 if (skb == NULL) 2378 goto out; 2379 2380 data_skb = skb; 2381 2382 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES 2383 if (unlikely(skb_shinfo(skb)->frag_list)) { 2384 /* 2385 * If this skb has a frag_list, then here that means that we 2386 * will have to use the frag_list skb's data for compat tasks 2387 * and the regular skb's data for normal (non-compat) tasks. 2388 * 2389 * If we need to send the compat skb, assign it to the 2390 * 'data_skb' variable so that it will be used below for data 2391 * copying. We keep 'skb' for everything else, including 2392 * freeing both later. 2393 */ 2394 if (flags & MSG_CMSG_COMPAT) 2395 data_skb = skb_shinfo(skb)->frag_list; 2396 } 2397 #endif 2398 2399 /* Record the max length of recvmsg() calls for future allocations */ 2400 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len); 2401 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len, 2402 16384); 2403 2404 copied = data_skb->len; 2405 if (len < copied) { 2406 msg->msg_flags |= MSG_TRUNC; 2407 copied = len; 2408 } 2409 2410 skb_reset_transport_header(data_skb); 2411 err = skb_copy_datagram_msg(data_skb, 0, msg, copied); 2412 2413 if (msg->msg_name) { 2414 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name); 2415 addr->nl_family = AF_NETLINK; 2416 addr->nl_pad = 0; 2417 addr->nl_pid = NETLINK_CB(skb).portid; 2418 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group); 2419 msg->msg_namelen = sizeof(*addr); 2420 } 2421 2422 if (nlk->flags & NETLINK_RECV_PKTINFO) 2423 netlink_cmsg_recv_pktinfo(msg, skb); 2424 2425 memset(&scm, 0, sizeof(scm)); 2426 scm.creds = *NETLINK_CREDS(skb); 2427 if (flags & MSG_TRUNC) 2428 copied = data_skb->len; 2429 2430 skb_free_datagram(sk, skb); 2431 2432 if (nlk->cb_running && 2433 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) { 2434 ret = netlink_dump(sk); 2435 if (ret) { 2436 sk->sk_err = -ret; 2437 sk->sk_error_report(sk); 2438 } 2439 } 2440 2441 scm_recv(sock, msg, &scm, flags); 2442 out: 2443 netlink_rcv_wake(sk); 2444 return err ? : copied; 2445 } 2446 2447 static void netlink_data_ready(struct sock *sk) 2448 { 2449 BUG(); 2450 } 2451 2452 /* 2453 * We export these functions to other modules. They provide a 2454 * complete set of kernel non-blocking support for message 2455 * queueing. 2456 */ 2457 2458 struct sock * 2459 __netlink_kernel_create(struct net *net, int unit, struct module *module, 2460 struct netlink_kernel_cfg *cfg) 2461 { 2462 struct socket *sock; 2463 struct sock *sk; 2464 struct netlink_sock *nlk; 2465 struct listeners *listeners = NULL; 2466 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL; 2467 unsigned int groups; 2468 2469 BUG_ON(!nl_table); 2470 2471 if (unit < 0 || unit >= MAX_LINKS) 2472 return NULL; 2473 2474 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 2475 return NULL; 2476 2477 /* 2478 * We have to just have a reference on the net from sk, but don't 2479 * get_net it. Besides, we cannot get and then put the net here. 2480 * So we create one inside init_net and the move it to net. 2481 */ 2482 2483 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0) 2484 goto out_sock_release_nosk; 2485 2486 sk = sock->sk; 2487 sk_change_net(sk, net); 2488 2489 if (!cfg || cfg->groups < 32) 2490 groups = 32; 2491 else 2492 groups = cfg->groups; 2493 2494 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 2495 if (!listeners) 2496 goto out_sock_release; 2497 2498 sk->sk_data_ready = netlink_data_ready; 2499 if (cfg && cfg->input) 2500 nlk_sk(sk)->netlink_rcv = cfg->input; 2501 2502 if (netlink_insert(sk, 0)) 2503 goto out_sock_release; 2504 2505 nlk = nlk_sk(sk); 2506 nlk->flags |= NETLINK_KERNEL_SOCKET; 2507 2508 netlink_table_grab(); 2509 if (!nl_table[unit].registered) { 2510 nl_table[unit].groups = groups; 2511 rcu_assign_pointer(nl_table[unit].listeners, listeners); 2512 nl_table[unit].cb_mutex = cb_mutex; 2513 nl_table[unit].module = module; 2514 if (cfg) { 2515 nl_table[unit].bind = cfg->bind; 2516 nl_table[unit].unbind = cfg->unbind; 2517 nl_table[unit].flags = cfg->flags; 2518 if (cfg->compare) 2519 nl_table[unit].compare = cfg->compare; 2520 } 2521 nl_table[unit].registered = 1; 2522 } else { 2523 kfree(listeners); 2524 nl_table[unit].registered++; 2525 } 2526 netlink_table_ungrab(); 2527 return sk; 2528 2529 out_sock_release: 2530 kfree(listeners); 2531 netlink_kernel_release(sk); 2532 return NULL; 2533 2534 out_sock_release_nosk: 2535 sock_release(sock); 2536 return NULL; 2537 } 2538 EXPORT_SYMBOL(__netlink_kernel_create); 2539 2540 void 2541 netlink_kernel_release(struct sock *sk) 2542 { 2543 sk_release_kernel(sk); 2544 } 2545 EXPORT_SYMBOL(netlink_kernel_release); 2546 2547 int __netlink_change_ngroups(struct sock *sk, unsigned int groups) 2548 { 2549 struct listeners *new, *old; 2550 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 2551 2552 if (groups < 32) 2553 groups = 32; 2554 2555 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) { 2556 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC); 2557 if (!new) 2558 return -ENOMEM; 2559 old = nl_deref_protected(tbl->listeners); 2560 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups)); 2561 rcu_assign_pointer(tbl->listeners, new); 2562 2563 kfree_rcu(old, rcu); 2564 } 2565 tbl->groups = groups; 2566 2567 return 0; 2568 } 2569 2570 /** 2571 * netlink_change_ngroups - change number of multicast groups 2572 * 2573 * This changes the number of multicast groups that are available 2574 * on a certain netlink family. Note that it is not possible to 2575 * change the number of groups to below 32. Also note that it does 2576 * not implicitly call netlink_clear_multicast_users() when the 2577 * number of groups is reduced. 2578 * 2579 * @sk: The kernel netlink socket, as returned by netlink_kernel_create(). 2580 * @groups: The new number of groups. 2581 */ 2582 int netlink_change_ngroups(struct sock *sk, unsigned int groups) 2583 { 2584 int err; 2585 2586 netlink_table_grab(); 2587 err = __netlink_change_ngroups(sk, groups); 2588 netlink_table_ungrab(); 2589 2590 return err; 2591 } 2592 2593 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group) 2594 { 2595 struct sock *sk; 2596 struct netlink_table *tbl = &nl_table[ksk->sk_protocol]; 2597 2598 sk_for_each_bound(sk, &tbl->mc_list) 2599 netlink_update_socket_mc(nlk_sk(sk), group, 0); 2600 } 2601 2602 struct nlmsghdr * 2603 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags) 2604 { 2605 struct nlmsghdr *nlh; 2606 int size = nlmsg_msg_size(len); 2607 2608 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size)); 2609 nlh->nlmsg_type = type; 2610 nlh->nlmsg_len = size; 2611 nlh->nlmsg_flags = flags; 2612 nlh->nlmsg_pid = portid; 2613 nlh->nlmsg_seq = seq; 2614 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0) 2615 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size); 2616 return nlh; 2617 } 2618 EXPORT_SYMBOL(__nlmsg_put); 2619 2620 /* 2621 * It looks a bit ugly. 2622 * It would be better to create kernel thread. 2623 */ 2624 2625 static int netlink_dump(struct sock *sk) 2626 { 2627 struct netlink_sock *nlk = nlk_sk(sk); 2628 struct netlink_callback *cb; 2629 struct sk_buff *skb = NULL; 2630 struct nlmsghdr *nlh; 2631 int len, err = -ENOBUFS; 2632 int alloc_size; 2633 2634 mutex_lock(nlk->cb_mutex); 2635 if (!nlk->cb_running) { 2636 err = -EINVAL; 2637 goto errout_skb; 2638 } 2639 2640 cb = &nlk->cb; 2641 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE); 2642 2643 if (!netlink_rx_is_mmaped(sk) && 2644 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2645 goto errout_skb; 2646 2647 /* NLMSG_GOODSIZE is small to avoid high order allocations being 2648 * required, but it makes sense to _attempt_ a 16K bytes allocation 2649 * to reduce number of system calls on dump operations, if user 2650 * ever provided a big enough buffer. 2651 */ 2652 if (alloc_size < nlk->max_recvmsg_len) { 2653 skb = netlink_alloc_skb(sk, 2654 nlk->max_recvmsg_len, 2655 nlk->portid, 2656 GFP_KERNEL | 2657 __GFP_NOWARN | 2658 __GFP_NORETRY); 2659 /* available room should be exact amount to avoid MSG_TRUNC */ 2660 if (skb) 2661 skb_reserve(skb, skb_tailroom(skb) - 2662 nlk->max_recvmsg_len); 2663 } 2664 if (!skb) 2665 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid, 2666 GFP_KERNEL); 2667 if (!skb) 2668 goto errout_skb; 2669 netlink_skb_set_owner_r(skb, sk); 2670 2671 len = cb->dump(skb, cb); 2672 2673 if (len > 0) { 2674 mutex_unlock(nlk->cb_mutex); 2675 2676 if (sk_filter(sk, skb)) 2677 kfree_skb(skb); 2678 else 2679 __netlink_sendskb(sk, skb); 2680 return 0; 2681 } 2682 2683 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI); 2684 if (!nlh) 2685 goto errout_skb; 2686 2687 nl_dump_check_consistent(cb, nlh); 2688 2689 memcpy(nlmsg_data(nlh), &len, sizeof(len)); 2690 2691 if (sk_filter(sk, skb)) 2692 kfree_skb(skb); 2693 else 2694 __netlink_sendskb(sk, skb); 2695 2696 if (cb->done) 2697 cb->done(cb); 2698 2699 nlk->cb_running = false; 2700 mutex_unlock(nlk->cb_mutex); 2701 module_put(cb->module); 2702 consume_skb(cb->skb); 2703 return 0; 2704 2705 errout_skb: 2706 mutex_unlock(nlk->cb_mutex); 2707 kfree_skb(skb); 2708 return err; 2709 } 2710 2711 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 2712 const struct nlmsghdr *nlh, 2713 struct netlink_dump_control *control) 2714 { 2715 struct netlink_callback *cb; 2716 struct sock *sk; 2717 struct netlink_sock *nlk; 2718 int ret; 2719 2720 /* Memory mapped dump requests need to be copied to avoid looping 2721 * on the pending state in netlink_mmap_sendmsg() while the CB hold 2722 * a reference to the skb. 2723 */ 2724 if (netlink_skb_is_mmaped(skb)) { 2725 skb = skb_copy(skb, GFP_KERNEL); 2726 if (skb == NULL) 2727 return -ENOBUFS; 2728 } else 2729 atomic_inc(&skb->users); 2730 2731 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid); 2732 if (sk == NULL) { 2733 ret = -ECONNREFUSED; 2734 goto error_free; 2735 } 2736 2737 nlk = nlk_sk(sk); 2738 mutex_lock(nlk->cb_mutex); 2739 /* A dump is in progress... */ 2740 if (nlk->cb_running) { 2741 ret = -EBUSY; 2742 goto error_unlock; 2743 } 2744 /* add reference of module which cb->dump belongs to */ 2745 if (!try_module_get(control->module)) { 2746 ret = -EPROTONOSUPPORT; 2747 goto error_unlock; 2748 } 2749 2750 cb = &nlk->cb; 2751 memset(cb, 0, sizeof(*cb)); 2752 cb->dump = control->dump; 2753 cb->done = control->done; 2754 cb->nlh = nlh; 2755 cb->data = control->data; 2756 cb->module = control->module; 2757 cb->min_dump_alloc = control->min_dump_alloc; 2758 cb->skb = skb; 2759 2760 nlk->cb_running = true; 2761 2762 mutex_unlock(nlk->cb_mutex); 2763 2764 ret = netlink_dump(sk); 2765 sock_put(sk); 2766 2767 if (ret) 2768 return ret; 2769 2770 /* We successfully started a dump, by returning -EINTR we 2771 * signal not to send ACK even if it was requested. 2772 */ 2773 return -EINTR; 2774 2775 error_unlock: 2776 sock_put(sk); 2777 mutex_unlock(nlk->cb_mutex); 2778 error_free: 2779 kfree_skb(skb); 2780 return ret; 2781 } 2782 EXPORT_SYMBOL(__netlink_dump_start); 2783 2784 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err) 2785 { 2786 struct sk_buff *skb; 2787 struct nlmsghdr *rep; 2788 struct nlmsgerr *errmsg; 2789 size_t payload = sizeof(*errmsg); 2790 2791 /* error messages get the original request appened */ 2792 if (err) 2793 payload += nlmsg_len(nlh); 2794 2795 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload), 2796 NETLINK_CB(in_skb).portid, GFP_KERNEL); 2797 if (!skb) { 2798 struct sock *sk; 2799 2800 sk = netlink_lookup(sock_net(in_skb->sk), 2801 in_skb->sk->sk_protocol, 2802 NETLINK_CB(in_skb).portid); 2803 if (sk) { 2804 sk->sk_err = ENOBUFS; 2805 sk->sk_error_report(sk); 2806 sock_put(sk); 2807 } 2808 return; 2809 } 2810 2811 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, 2812 NLMSG_ERROR, payload, 0); 2813 errmsg = nlmsg_data(rep); 2814 errmsg->error = err; 2815 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh)); 2816 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT); 2817 } 2818 EXPORT_SYMBOL(netlink_ack); 2819 2820 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *, 2821 struct nlmsghdr *)) 2822 { 2823 struct nlmsghdr *nlh; 2824 int err; 2825 2826 while (skb->len >= nlmsg_total_size(0)) { 2827 int msglen; 2828 2829 nlh = nlmsg_hdr(skb); 2830 err = 0; 2831 2832 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len) 2833 return 0; 2834 2835 /* Only requests are handled by the kernel */ 2836 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) 2837 goto ack; 2838 2839 /* Skip control messages */ 2840 if (nlh->nlmsg_type < NLMSG_MIN_TYPE) 2841 goto ack; 2842 2843 err = cb(skb, nlh); 2844 if (err == -EINTR) 2845 goto skip; 2846 2847 ack: 2848 if (nlh->nlmsg_flags & NLM_F_ACK || err) 2849 netlink_ack(skb, nlh, err); 2850 2851 skip: 2852 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 2853 if (msglen > skb->len) 2854 msglen = skb->len; 2855 skb_pull(skb, msglen); 2856 } 2857 2858 return 0; 2859 } 2860 EXPORT_SYMBOL(netlink_rcv_skb); 2861 2862 /** 2863 * nlmsg_notify - send a notification netlink message 2864 * @sk: netlink socket to use 2865 * @skb: notification message 2866 * @portid: destination netlink portid for reports or 0 2867 * @group: destination multicast group or 0 2868 * @report: 1 to report back, 0 to disable 2869 * @flags: allocation flags 2870 */ 2871 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid, 2872 unsigned int group, int report, gfp_t flags) 2873 { 2874 int err = 0; 2875 2876 if (group) { 2877 int exclude_portid = 0; 2878 2879 if (report) { 2880 atomic_inc(&skb->users); 2881 exclude_portid = portid; 2882 } 2883 2884 /* errors reported via destination sk->sk_err, but propagate 2885 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */ 2886 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags); 2887 } 2888 2889 if (report) { 2890 int err2; 2891 2892 err2 = nlmsg_unicast(sk, skb, portid); 2893 if (!err || err == -ESRCH) 2894 err = err2; 2895 } 2896 2897 return err; 2898 } 2899 EXPORT_SYMBOL(nlmsg_notify); 2900 2901 #ifdef CONFIG_PROC_FS 2902 struct nl_seq_iter { 2903 struct seq_net_private p; 2904 struct rhashtable_iter hti; 2905 int link; 2906 }; 2907 2908 static int netlink_walk_start(struct nl_seq_iter *iter) 2909 { 2910 int err; 2911 2912 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti); 2913 if (err) { 2914 iter->link = MAX_LINKS; 2915 return err; 2916 } 2917 2918 err = rhashtable_walk_start(&iter->hti); 2919 return err == -EAGAIN ? 0 : err; 2920 } 2921 2922 static void netlink_walk_stop(struct nl_seq_iter *iter) 2923 { 2924 rhashtable_walk_stop(&iter->hti); 2925 rhashtable_walk_exit(&iter->hti); 2926 } 2927 2928 static void *__netlink_seq_next(struct seq_file *seq) 2929 { 2930 struct nl_seq_iter *iter = seq->private; 2931 struct netlink_sock *nlk; 2932 2933 do { 2934 for (;;) { 2935 int err; 2936 2937 nlk = rhashtable_walk_next(&iter->hti); 2938 2939 if (IS_ERR(nlk)) { 2940 if (PTR_ERR(nlk) == -EAGAIN) 2941 continue; 2942 2943 return nlk; 2944 } 2945 2946 if (nlk) 2947 break; 2948 2949 netlink_walk_stop(iter); 2950 if (++iter->link >= MAX_LINKS) 2951 return NULL; 2952 2953 err = netlink_walk_start(iter); 2954 if (err) 2955 return ERR_PTR(err); 2956 } 2957 } while (sock_net(&nlk->sk) != seq_file_net(seq)); 2958 2959 return nlk; 2960 } 2961 2962 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp) 2963 { 2964 struct nl_seq_iter *iter = seq->private; 2965 void *obj = SEQ_START_TOKEN; 2966 loff_t pos; 2967 int err; 2968 2969 iter->link = 0; 2970 2971 err = netlink_walk_start(iter); 2972 if (err) 2973 return ERR_PTR(err); 2974 2975 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--) 2976 obj = __netlink_seq_next(seq); 2977 2978 return obj; 2979 } 2980 2981 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2982 { 2983 ++*pos; 2984 return __netlink_seq_next(seq); 2985 } 2986 2987 static void netlink_seq_stop(struct seq_file *seq, void *v) 2988 { 2989 struct nl_seq_iter *iter = seq->private; 2990 2991 if (iter->link >= MAX_LINKS) 2992 return; 2993 2994 netlink_walk_stop(iter); 2995 } 2996 2997 2998 static int netlink_seq_show(struct seq_file *seq, void *v) 2999 { 3000 if (v == SEQ_START_TOKEN) { 3001 seq_puts(seq, 3002 "sk Eth Pid Groups " 3003 "Rmem Wmem Dump Locks Drops Inode\n"); 3004 } else { 3005 struct sock *s = v; 3006 struct netlink_sock *nlk = nlk_sk(s); 3007 3008 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n", 3009 s, 3010 s->sk_protocol, 3011 nlk->portid, 3012 nlk->groups ? (u32)nlk->groups[0] : 0, 3013 sk_rmem_alloc_get(s), 3014 sk_wmem_alloc_get(s), 3015 nlk->cb_running, 3016 atomic_read(&s->sk_refcnt), 3017 atomic_read(&s->sk_drops), 3018 sock_i_ino(s) 3019 ); 3020 3021 } 3022 return 0; 3023 } 3024 3025 static const struct seq_operations netlink_seq_ops = { 3026 .start = netlink_seq_start, 3027 .next = netlink_seq_next, 3028 .stop = netlink_seq_stop, 3029 .show = netlink_seq_show, 3030 }; 3031 3032 3033 static int netlink_seq_open(struct inode *inode, struct file *file) 3034 { 3035 return seq_open_net(inode, file, &netlink_seq_ops, 3036 sizeof(struct nl_seq_iter)); 3037 } 3038 3039 static const struct file_operations netlink_seq_fops = { 3040 .owner = THIS_MODULE, 3041 .open = netlink_seq_open, 3042 .read = seq_read, 3043 .llseek = seq_lseek, 3044 .release = seq_release_net, 3045 }; 3046 3047 #endif 3048 3049 int netlink_register_notifier(struct notifier_block *nb) 3050 { 3051 return atomic_notifier_chain_register(&netlink_chain, nb); 3052 } 3053 EXPORT_SYMBOL(netlink_register_notifier); 3054 3055 int netlink_unregister_notifier(struct notifier_block *nb) 3056 { 3057 return atomic_notifier_chain_unregister(&netlink_chain, nb); 3058 } 3059 EXPORT_SYMBOL(netlink_unregister_notifier); 3060 3061 static const struct proto_ops netlink_ops = { 3062 .family = PF_NETLINK, 3063 .owner = THIS_MODULE, 3064 .release = netlink_release, 3065 .bind = netlink_bind, 3066 .connect = netlink_connect, 3067 .socketpair = sock_no_socketpair, 3068 .accept = sock_no_accept, 3069 .getname = netlink_getname, 3070 .poll = netlink_poll, 3071 .ioctl = sock_no_ioctl, 3072 .listen = sock_no_listen, 3073 .shutdown = sock_no_shutdown, 3074 .setsockopt = netlink_setsockopt, 3075 .getsockopt = netlink_getsockopt, 3076 .sendmsg = netlink_sendmsg, 3077 .recvmsg = netlink_recvmsg, 3078 .mmap = netlink_mmap, 3079 .sendpage = sock_no_sendpage, 3080 }; 3081 3082 static const struct net_proto_family netlink_family_ops = { 3083 .family = PF_NETLINK, 3084 .create = netlink_create, 3085 .owner = THIS_MODULE, /* for consistency 8) */ 3086 }; 3087 3088 static int __net_init netlink_net_init(struct net *net) 3089 { 3090 #ifdef CONFIG_PROC_FS 3091 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops)) 3092 return -ENOMEM; 3093 #endif 3094 return 0; 3095 } 3096 3097 static void __net_exit netlink_net_exit(struct net *net) 3098 { 3099 #ifdef CONFIG_PROC_FS 3100 remove_proc_entry("netlink", net->proc_net); 3101 #endif 3102 } 3103 3104 static void __init netlink_add_usersock_entry(void) 3105 { 3106 struct listeners *listeners; 3107 int groups = 32; 3108 3109 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 3110 if (!listeners) 3111 panic("netlink_add_usersock_entry: Cannot allocate listeners\n"); 3112 3113 netlink_table_grab(); 3114 3115 nl_table[NETLINK_USERSOCK].groups = groups; 3116 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners); 3117 nl_table[NETLINK_USERSOCK].module = THIS_MODULE; 3118 nl_table[NETLINK_USERSOCK].registered = 1; 3119 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND; 3120 3121 netlink_table_ungrab(); 3122 } 3123 3124 static struct pernet_operations __net_initdata netlink_net_ops = { 3125 .init = netlink_net_init, 3126 .exit = netlink_net_exit, 3127 }; 3128 3129 static inline u32 netlink_hash(const void *data, u32 len, u32 seed) 3130 { 3131 const struct netlink_sock *nlk = data; 3132 struct netlink_compare_arg arg; 3133 3134 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid); 3135 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed); 3136 } 3137 3138 static const struct rhashtable_params netlink_rhashtable_params = { 3139 .head_offset = offsetof(struct netlink_sock, node), 3140 .key_len = netlink_compare_arg_len, 3141 .obj_hashfn = netlink_hash, 3142 .obj_cmpfn = netlink_compare, 3143 .automatic_shrinking = true, 3144 }; 3145 3146 static int __init netlink_proto_init(void) 3147 { 3148 int i; 3149 int err = proto_register(&netlink_proto, 0); 3150 3151 if (err != 0) 3152 goto out; 3153 3154 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb)); 3155 3156 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL); 3157 if (!nl_table) 3158 goto panic; 3159 3160 for (i = 0; i < MAX_LINKS; i++) { 3161 if (rhashtable_init(&nl_table[i].hash, 3162 &netlink_rhashtable_params) < 0) { 3163 while (--i > 0) 3164 rhashtable_destroy(&nl_table[i].hash); 3165 kfree(nl_table); 3166 goto panic; 3167 } 3168 } 3169 3170 INIT_LIST_HEAD(&netlink_tap_all); 3171 3172 netlink_add_usersock_entry(); 3173 3174 sock_register(&netlink_family_ops); 3175 register_pernet_subsys(&netlink_net_ops); 3176 /* The netlink device handler may be needed early. */ 3177 rtnetlink_init(); 3178 out: 3179 return err; 3180 panic: 3181 panic("netlink_init: Cannot allocate nl_table\n"); 3182 } 3183 3184 core_initcall(netlink_proto_init); 3185