1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * PACKET - implements raw packet sockets. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Alan Cox, <gw4pts@gw4pts.ampr.org> 11 * 12 * Fixes: 13 * Alan Cox : verify_area() now used correctly 14 * Alan Cox : new skbuff lists, look ma no backlogs! 15 * Alan Cox : tidied skbuff lists. 16 * Alan Cox : Now uses generic datagram routines I 17 * added. Also fixed the peek/read crash 18 * from all old Linux datagram code. 19 * Alan Cox : Uses the improved datagram code. 20 * Alan Cox : Added NULL's for socket options. 21 * Alan Cox : Re-commented the code. 22 * Alan Cox : Use new kernel side addressing 23 * Rob Janssen : Correct MTU usage. 24 * Dave Platt : Counter leaks caused by incorrect 25 * interrupt locking and some slightly 26 * dubious gcc output. Can you read 27 * compiler: it said _VOLATILE_ 28 * Richard Kooijman : Timestamp fixes. 29 * Alan Cox : New buffers. Use sk->mac.raw. 30 * Alan Cox : sendmsg/recvmsg support. 31 * Alan Cox : Protocol setting support 32 * Alexey Kuznetsov : Untied from IPv4 stack. 33 * Cyrus Durgin : Fixed kerneld for kmod. 34 * Michal Ostrowski : Module initialization cleanup. 35 * Ulises Alonso : Frame number limit removal and 36 * packet_set_ring memory leak. 37 * Eric Biederman : Allow for > 8 byte hardware addresses. 38 * The convention is that longer addresses 39 * will simply extend the hardware address 40 * byte arrays at the end of sockaddr_ll 41 * and packet_mreq. 42 * Johann Baudy : Added TX RING. 43 * Chetan Loke : Implemented TPACKET_V3 block abstraction 44 * layer. 45 * Copyright (C) 2011, <lokec@ccs.neu.edu> 46 * 47 * 48 * This program is free software; you can redistribute it and/or 49 * modify it under the terms of the GNU General Public License 50 * as published by the Free Software Foundation; either version 51 * 2 of the License, or (at your option) any later version. 52 * 53 */ 54 55 #include <linux/types.h> 56 #include <linux/mm.h> 57 #include <linux/capability.h> 58 #include <linux/fcntl.h> 59 #include <linux/socket.h> 60 #include <linux/in.h> 61 #include <linux/inet.h> 62 #include <linux/netdevice.h> 63 #include <linux/if_packet.h> 64 #include <linux/wireless.h> 65 #include <linux/kernel.h> 66 #include <linux/kmod.h> 67 #include <linux/slab.h> 68 #include <linux/vmalloc.h> 69 #include <net/net_namespace.h> 70 #include <net/ip.h> 71 #include <net/protocol.h> 72 #include <linux/skbuff.h> 73 #include <net/sock.h> 74 #include <linux/errno.h> 75 #include <linux/timer.h> 76 #include <asm/uaccess.h> 77 #include <asm/ioctls.h> 78 #include <asm/page.h> 79 #include <asm/cacheflush.h> 80 #include <asm/io.h> 81 #include <linux/proc_fs.h> 82 #include <linux/seq_file.h> 83 #include <linux/poll.h> 84 #include <linux/module.h> 85 #include <linux/init.h> 86 #include <linux/mutex.h> 87 #include <linux/if_vlan.h> 88 #include <linux/virtio_net.h> 89 #include <linux/errqueue.h> 90 #include <linux/net_tstamp.h> 91 92 #ifdef CONFIG_INET 93 #include <net/inet_common.h> 94 #endif 95 96 #include "internal.h" 97 98 /* 99 Assumptions: 100 - if device has no dev->hard_header routine, it adds and removes ll header 101 inside itself. In this case ll header is invisible outside of device, 102 but higher levels still should reserve dev->hard_header_len. 103 Some devices are enough clever to reallocate skb, when header 104 will not fit to reserved space (tunnel), another ones are silly 105 (PPP). 106 - packet socket receives packets with pulled ll header, 107 so that SOCK_RAW should push it back. 108 109 On receive: 110 ----------- 111 112 Incoming, dev->hard_header!=NULL 113 mac_header -> ll header 114 data -> data 115 116 Outgoing, dev->hard_header!=NULL 117 mac_header -> ll header 118 data -> ll header 119 120 Incoming, dev->hard_header==NULL 121 mac_header -> UNKNOWN position. It is very likely, that it points to ll 122 header. PPP makes it, that is wrong, because introduce 123 assymetry between rx and tx paths. 124 data -> data 125 126 Outgoing, dev->hard_header==NULL 127 mac_header -> data. ll header is still not built! 128 data -> data 129 130 Resume 131 If dev->hard_header==NULL we are unlikely to restore sensible ll header. 132 133 134 On transmit: 135 ------------ 136 137 dev->hard_header != NULL 138 mac_header -> ll header 139 data -> ll header 140 141 dev->hard_header == NULL (ll header is added by device, we cannot control it) 142 mac_header -> data 143 data -> data 144 145 We should set nh.raw on output to correct posistion, 146 packet classifier depends on it. 147 */ 148 149 /* Private packet socket structures. */ 150 151 /* identical to struct packet_mreq except it has 152 * a longer address field. 153 */ 154 struct packet_mreq_max { 155 int mr_ifindex; 156 unsigned short mr_type; 157 unsigned short mr_alen; 158 unsigned char mr_address[MAX_ADDR_LEN]; 159 }; 160 161 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 162 int closing, int tx_ring); 163 164 165 #define V3_ALIGNMENT (8) 166 167 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) 168 169 #define BLK_PLUS_PRIV(sz_of_priv) \ 170 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) 171 172 #define PGV_FROM_VMALLOC 1 173 174 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) 175 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) 176 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) 177 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) 178 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) 179 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) 180 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x))) 181 182 struct packet_sock; 183 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg); 184 185 static void *packet_previous_frame(struct packet_sock *po, 186 struct packet_ring_buffer *rb, 187 int status); 188 static void packet_increment_head(struct packet_ring_buffer *buff); 189 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *, 190 struct tpacket_block_desc *); 191 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 192 struct packet_sock *); 193 static void prb_retire_current_block(struct tpacket_kbdq_core *, 194 struct packet_sock *, unsigned int status); 195 static int prb_queue_frozen(struct tpacket_kbdq_core *); 196 static void prb_open_block(struct tpacket_kbdq_core *, 197 struct tpacket_block_desc *); 198 static void prb_retire_rx_blk_timer_expired(unsigned long); 199 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); 200 static void prb_init_blk_timer(struct packet_sock *, 201 struct tpacket_kbdq_core *, 202 void (*func) (unsigned long)); 203 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 204 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 205 struct tpacket3_hdr *); 206 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 207 struct tpacket3_hdr *); 208 static void packet_flush_mclist(struct sock *sk); 209 210 struct packet_skb_cb { 211 unsigned int origlen; 212 union { 213 struct sockaddr_pkt pkt; 214 struct sockaddr_ll ll; 215 } sa; 216 }; 217 218 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 219 220 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 221 #define GET_PBLOCK_DESC(x, bid) \ 222 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 223 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 224 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 225 #define GET_NEXT_PRB_BLK_NUM(x) \ 226 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 227 ((x)->kactive_blk_num+1) : 0) 228 229 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 230 static void __fanout_link(struct sock *sk, struct packet_sock *po); 231 232 /* register_prot_hook must be invoked with the po->bind_lock held, 233 * or from a context in which asynchronous accesses to the packet 234 * socket is not possible (packet_create()). 235 */ 236 static void register_prot_hook(struct sock *sk) 237 { 238 struct packet_sock *po = pkt_sk(sk); 239 if (!po->running) { 240 if (po->fanout) 241 __fanout_link(sk, po); 242 else 243 dev_add_pack(&po->prot_hook); 244 sock_hold(sk); 245 po->running = 1; 246 } 247 } 248 249 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock 250 * held. If the sync parameter is true, we will temporarily drop 251 * the po->bind_lock and do a synchronize_net to make sure no 252 * asynchronous packet processing paths still refer to the elements 253 * of po->prot_hook. If the sync parameter is false, it is the 254 * callers responsibility to take care of this. 255 */ 256 static void __unregister_prot_hook(struct sock *sk, bool sync) 257 { 258 struct packet_sock *po = pkt_sk(sk); 259 260 po->running = 0; 261 if (po->fanout) 262 __fanout_unlink(sk, po); 263 else 264 __dev_remove_pack(&po->prot_hook); 265 __sock_put(sk); 266 267 if (sync) { 268 spin_unlock(&po->bind_lock); 269 synchronize_net(); 270 spin_lock(&po->bind_lock); 271 } 272 } 273 274 static void unregister_prot_hook(struct sock *sk, bool sync) 275 { 276 struct packet_sock *po = pkt_sk(sk); 277 278 if (po->running) 279 __unregister_prot_hook(sk, sync); 280 } 281 282 static inline __pure struct page *pgv_to_page(void *addr) 283 { 284 if (is_vmalloc_addr(addr)) 285 return vmalloc_to_page(addr); 286 return virt_to_page(addr); 287 } 288 289 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 290 { 291 union { 292 struct tpacket_hdr *h1; 293 struct tpacket2_hdr *h2; 294 void *raw; 295 } h; 296 297 h.raw = frame; 298 switch (po->tp_version) { 299 case TPACKET_V1: 300 h.h1->tp_status = status; 301 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 302 break; 303 case TPACKET_V2: 304 h.h2->tp_status = status; 305 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 306 break; 307 case TPACKET_V3: 308 default: 309 WARN(1, "TPACKET version not supported.\n"); 310 BUG(); 311 } 312 313 smp_wmb(); 314 } 315 316 static int __packet_get_status(struct packet_sock *po, void *frame) 317 { 318 union { 319 struct tpacket_hdr *h1; 320 struct tpacket2_hdr *h2; 321 void *raw; 322 } h; 323 324 smp_rmb(); 325 326 h.raw = frame; 327 switch (po->tp_version) { 328 case TPACKET_V1: 329 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 330 return h.h1->tp_status; 331 case TPACKET_V2: 332 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 333 return h.h2->tp_status; 334 case TPACKET_V3: 335 default: 336 WARN(1, "TPACKET version not supported.\n"); 337 BUG(); 338 return 0; 339 } 340 } 341 342 static void *packet_lookup_frame(struct packet_sock *po, 343 struct packet_ring_buffer *rb, 344 unsigned int position, 345 int status) 346 { 347 unsigned int pg_vec_pos, frame_offset; 348 union { 349 struct tpacket_hdr *h1; 350 struct tpacket2_hdr *h2; 351 void *raw; 352 } h; 353 354 pg_vec_pos = position / rb->frames_per_block; 355 frame_offset = position % rb->frames_per_block; 356 357 h.raw = rb->pg_vec[pg_vec_pos].buffer + 358 (frame_offset * rb->frame_size); 359 360 if (status != __packet_get_status(po, h.raw)) 361 return NULL; 362 363 return h.raw; 364 } 365 366 static void *packet_current_frame(struct packet_sock *po, 367 struct packet_ring_buffer *rb, 368 int status) 369 { 370 return packet_lookup_frame(po, rb, rb->head, status); 371 } 372 373 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 374 { 375 del_timer_sync(&pkc->retire_blk_timer); 376 } 377 378 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 379 int tx_ring, 380 struct sk_buff_head *rb_queue) 381 { 382 struct tpacket_kbdq_core *pkc; 383 384 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc; 385 386 spin_lock(&rb_queue->lock); 387 pkc->delete_blk_timer = 1; 388 spin_unlock(&rb_queue->lock); 389 390 prb_del_retire_blk_timer(pkc); 391 } 392 393 static void prb_init_blk_timer(struct packet_sock *po, 394 struct tpacket_kbdq_core *pkc, 395 void (*func) (unsigned long)) 396 { 397 init_timer(&pkc->retire_blk_timer); 398 pkc->retire_blk_timer.data = (long)po; 399 pkc->retire_blk_timer.function = func; 400 pkc->retire_blk_timer.expires = jiffies; 401 } 402 403 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring) 404 { 405 struct tpacket_kbdq_core *pkc; 406 407 if (tx_ring) 408 BUG(); 409 410 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc; 411 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired); 412 } 413 414 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 415 int blk_size_in_bytes) 416 { 417 struct net_device *dev; 418 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; 419 struct ethtool_cmd ecmd; 420 int err; 421 u32 speed; 422 423 rtnl_lock(); 424 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 425 if (unlikely(!dev)) { 426 rtnl_unlock(); 427 return DEFAULT_PRB_RETIRE_TOV; 428 } 429 err = __ethtool_get_settings(dev, &ecmd); 430 speed = ethtool_cmd_speed(&ecmd); 431 rtnl_unlock(); 432 if (!err) { 433 /* 434 * If the link speed is so slow you don't really 435 * need to worry about perf anyways 436 */ 437 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) { 438 return DEFAULT_PRB_RETIRE_TOV; 439 } else { 440 msec = 1; 441 div = speed / 1000; 442 } 443 } 444 445 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 446 447 if (div) 448 mbits /= div; 449 450 tmo = mbits * msec; 451 452 if (div) 453 return tmo+1; 454 return tmo; 455 } 456 457 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 458 union tpacket_req_u *req_u) 459 { 460 p1->feature_req_word = req_u->req3.tp_feature_req_word; 461 } 462 463 static void init_prb_bdqc(struct packet_sock *po, 464 struct packet_ring_buffer *rb, 465 struct pgv *pg_vec, 466 union tpacket_req_u *req_u, int tx_ring) 467 { 468 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc; 469 struct tpacket_block_desc *pbd; 470 471 memset(p1, 0x0, sizeof(*p1)); 472 473 p1->knxt_seq_num = 1; 474 p1->pkbdq = pg_vec; 475 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 476 p1->pkblk_start = pg_vec[0].buffer; 477 p1->kblk_size = req_u->req3.tp_block_size; 478 p1->knum_blocks = req_u->req3.tp_block_nr; 479 p1->hdrlen = po->tp_hdrlen; 480 p1->version = po->tp_version; 481 p1->last_kactive_blk_num = 0; 482 po->stats_u.stats3.tp_freeze_q_cnt = 0; 483 if (req_u->req3.tp_retire_blk_tov) 484 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 485 else 486 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 487 req_u->req3.tp_block_size); 488 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 489 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 490 491 prb_init_ft_ops(p1, req_u); 492 prb_setup_retire_blk_timer(po, tx_ring); 493 prb_open_block(p1, pbd); 494 } 495 496 /* Do NOT update the last_blk_num first. 497 * Assumes sk_buff_head lock is held. 498 */ 499 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 500 { 501 mod_timer(&pkc->retire_blk_timer, 502 jiffies + pkc->tov_in_jiffies); 503 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 504 } 505 506 /* 507 * Timer logic: 508 * 1) We refresh the timer only when we open a block. 509 * By doing this we don't waste cycles refreshing the timer 510 * on packet-by-packet basis. 511 * 512 * With a 1MB block-size, on a 1Gbps line, it will take 513 * i) ~8 ms to fill a block + ii) memcpy etc. 514 * In this cut we are not accounting for the memcpy time. 515 * 516 * So, if the user sets the 'tmo' to 10ms then the timer 517 * will never fire while the block is still getting filled 518 * (which is what we want). However, the user could choose 519 * to close a block early and that's fine. 520 * 521 * But when the timer does fire, we check whether or not to refresh it. 522 * Since the tmo granularity is in msecs, it is not too expensive 523 * to refresh the timer, lets say every '8' msecs. 524 * Either the user can set the 'tmo' or we can derive it based on 525 * a) line-speed and b) block-size. 526 * prb_calc_retire_blk_tmo() calculates the tmo. 527 * 528 */ 529 static void prb_retire_rx_blk_timer_expired(unsigned long data) 530 { 531 struct packet_sock *po = (struct packet_sock *)data; 532 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc; 533 unsigned int frozen; 534 struct tpacket_block_desc *pbd; 535 536 spin_lock(&po->sk.sk_receive_queue.lock); 537 538 frozen = prb_queue_frozen(pkc); 539 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 540 541 if (unlikely(pkc->delete_blk_timer)) 542 goto out; 543 544 /* We only need to plug the race when the block is partially filled. 545 * tpacket_rcv: 546 * lock(); increment BLOCK_NUM_PKTS; unlock() 547 * copy_bits() is in progress ... 548 * timer fires on other cpu: 549 * we can't retire the current block because copy_bits 550 * is in progress. 551 * 552 */ 553 if (BLOCK_NUM_PKTS(pbd)) { 554 while (atomic_read(&pkc->blk_fill_in_prog)) { 555 /* Waiting for skb_copy_bits to finish... */ 556 cpu_relax(); 557 } 558 } 559 560 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 561 if (!frozen) { 562 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 563 if (!prb_dispatch_next_block(pkc, po)) 564 goto refresh_timer; 565 else 566 goto out; 567 } else { 568 /* Case 1. Queue was frozen because user-space was 569 * lagging behind. 570 */ 571 if (prb_curr_blk_in_use(pkc, pbd)) { 572 /* 573 * Ok, user-space is still behind. 574 * So just refresh the timer. 575 */ 576 goto refresh_timer; 577 } else { 578 /* Case 2. queue was frozen,user-space caught up, 579 * now the link went idle && the timer fired. 580 * We don't have a block to close.So we open this 581 * block and restart the timer. 582 * opening a block thaws the queue,restarts timer 583 * Thawing/timer-refresh is a side effect. 584 */ 585 prb_open_block(pkc, pbd); 586 goto out; 587 } 588 } 589 } 590 591 refresh_timer: 592 _prb_refresh_rx_retire_blk_timer(pkc); 593 594 out: 595 spin_unlock(&po->sk.sk_receive_queue.lock); 596 } 597 598 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 599 struct tpacket_block_desc *pbd1, __u32 status) 600 { 601 /* Flush everything minus the block header */ 602 603 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 604 u8 *start, *end; 605 606 start = (u8 *)pbd1; 607 608 /* Skip the block header(we know header WILL fit in 4K) */ 609 start += PAGE_SIZE; 610 611 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 612 for (; start < end; start += PAGE_SIZE) 613 flush_dcache_page(pgv_to_page(start)); 614 615 smp_wmb(); 616 #endif 617 618 /* Now update the block status. */ 619 620 BLOCK_STATUS(pbd1) = status; 621 622 /* Flush the block header */ 623 624 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 625 start = (u8 *)pbd1; 626 flush_dcache_page(pgv_to_page(start)); 627 628 smp_wmb(); 629 #endif 630 } 631 632 /* 633 * Side effect: 634 * 635 * 1) flush the block 636 * 2) Increment active_blk_num 637 * 638 * Note:We DONT refresh the timer on purpose. 639 * Because almost always the next block will be opened. 640 */ 641 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 642 struct tpacket_block_desc *pbd1, 643 struct packet_sock *po, unsigned int stat) 644 { 645 __u32 status = TP_STATUS_USER | stat; 646 647 struct tpacket3_hdr *last_pkt; 648 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 649 650 if (po->stats.tp_drops) 651 status |= TP_STATUS_LOSING; 652 653 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 654 last_pkt->tp_next_offset = 0; 655 656 /* Get the ts of the last pkt */ 657 if (BLOCK_NUM_PKTS(pbd1)) { 658 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 659 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 660 } else { 661 /* Ok, we tmo'd - so get the current time */ 662 struct timespec ts; 663 getnstimeofday(&ts); 664 h1->ts_last_pkt.ts_sec = ts.tv_sec; 665 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 666 } 667 668 smp_wmb(); 669 670 /* Flush the block */ 671 prb_flush_block(pkc1, pbd1, status); 672 673 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 674 } 675 676 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 677 { 678 pkc->reset_pending_on_curr_blk = 0; 679 } 680 681 /* 682 * Side effect of opening a block: 683 * 684 * 1) prb_queue is thawed. 685 * 2) retire_blk_timer is refreshed. 686 * 687 */ 688 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 689 struct tpacket_block_desc *pbd1) 690 { 691 struct timespec ts; 692 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 693 694 smp_rmb(); 695 696 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) { 697 698 /* We could have just memset this but we will lose the 699 * flexibility of making the priv area sticky 700 */ 701 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 702 BLOCK_NUM_PKTS(pbd1) = 0; 703 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 704 getnstimeofday(&ts); 705 h1->ts_first_pkt.ts_sec = ts.tv_sec; 706 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 707 pkc1->pkblk_start = (char *)pbd1; 708 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 709 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 710 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 711 pbd1->version = pkc1->version; 712 pkc1->prev = pkc1->nxt_offset; 713 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 714 prb_thaw_queue(pkc1); 715 _prb_refresh_rx_retire_blk_timer(pkc1); 716 717 smp_wmb(); 718 719 return; 720 } 721 722 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n", 723 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num); 724 dump_stack(); 725 BUG(); 726 } 727 728 /* 729 * Queue freeze logic: 730 * 1) Assume tp_block_nr = 8 blocks. 731 * 2) At time 't0', user opens Rx ring. 732 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 733 * 4) user-space is either sleeping or processing block '0'. 734 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 735 * it will close block-7,loop around and try to fill block '0'. 736 * call-flow: 737 * __packet_lookup_frame_in_block 738 * prb_retire_current_block() 739 * prb_dispatch_next_block() 740 * |->(BLOCK_STATUS == USER) evaluates to true 741 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 742 * 6) Now there are two cases: 743 * 6.1) Link goes idle right after the queue is frozen. 744 * But remember, the last open_block() refreshed the timer. 745 * When this timer expires,it will refresh itself so that we can 746 * re-open block-0 in near future. 747 * 6.2) Link is busy and keeps on receiving packets. This is a simple 748 * case and __packet_lookup_frame_in_block will check if block-0 749 * is free and can now be re-used. 750 */ 751 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 752 struct packet_sock *po) 753 { 754 pkc->reset_pending_on_curr_blk = 1; 755 po->stats_u.stats3.tp_freeze_q_cnt++; 756 } 757 758 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 759 760 /* 761 * If the next block is free then we will dispatch it 762 * and return a good offset. 763 * Else, we will freeze the queue. 764 * So, caller must check the return value. 765 */ 766 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 767 struct packet_sock *po) 768 { 769 struct tpacket_block_desc *pbd; 770 771 smp_rmb(); 772 773 /* 1. Get current block num */ 774 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 775 776 /* 2. If this block is currently in_use then freeze the queue */ 777 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 778 prb_freeze_queue(pkc, po); 779 return NULL; 780 } 781 782 /* 783 * 3. 784 * open this block and return the offset where the first packet 785 * needs to get stored. 786 */ 787 prb_open_block(pkc, pbd); 788 return (void *)pkc->nxt_offset; 789 } 790 791 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 792 struct packet_sock *po, unsigned int status) 793 { 794 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 795 796 /* retire/close the current block */ 797 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 798 /* 799 * Plug the case where copy_bits() is in progress on 800 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 801 * have space to copy the pkt in the current block and 802 * called prb_retire_current_block() 803 * 804 * We don't need to worry about the TMO case because 805 * the timer-handler already handled this case. 806 */ 807 if (!(status & TP_STATUS_BLK_TMO)) { 808 while (atomic_read(&pkc->blk_fill_in_prog)) { 809 /* Waiting for skb_copy_bits to finish... */ 810 cpu_relax(); 811 } 812 } 813 prb_close_block(pkc, pbd, po, status); 814 return; 815 } 816 817 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd); 818 dump_stack(); 819 BUG(); 820 } 821 822 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc, 823 struct tpacket_block_desc *pbd) 824 { 825 return TP_STATUS_USER & BLOCK_STATUS(pbd); 826 } 827 828 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 829 { 830 return pkc->reset_pending_on_curr_blk; 831 } 832 833 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 834 { 835 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 836 atomic_dec(&pkc->blk_fill_in_prog); 837 } 838 839 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 840 struct tpacket3_hdr *ppd) 841 { 842 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb); 843 } 844 845 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 846 struct tpacket3_hdr *ppd) 847 { 848 ppd->hv1.tp_rxhash = 0; 849 } 850 851 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 852 struct tpacket3_hdr *ppd) 853 { 854 if (vlan_tx_tag_present(pkc->skb)) { 855 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb); 856 ppd->tp_status = TP_STATUS_VLAN_VALID; 857 } else { 858 ppd->hv1.tp_vlan_tci = 0; 859 ppd->tp_status = TP_STATUS_AVAILABLE; 860 } 861 } 862 863 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 864 struct tpacket3_hdr *ppd) 865 { 866 prb_fill_vlan_info(pkc, ppd); 867 868 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 869 prb_fill_rxhash(pkc, ppd); 870 else 871 prb_clear_rxhash(pkc, ppd); 872 } 873 874 static void prb_fill_curr_block(char *curr, 875 struct tpacket_kbdq_core *pkc, 876 struct tpacket_block_desc *pbd, 877 unsigned int len) 878 { 879 struct tpacket3_hdr *ppd; 880 881 ppd = (struct tpacket3_hdr *)curr; 882 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 883 pkc->prev = curr; 884 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 885 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 886 BLOCK_NUM_PKTS(pbd) += 1; 887 atomic_inc(&pkc->blk_fill_in_prog); 888 prb_run_all_ft_ops(pkc, ppd); 889 } 890 891 /* Assumes caller has the sk->rx_queue.lock */ 892 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 893 struct sk_buff *skb, 894 int status, 895 unsigned int len 896 ) 897 { 898 struct tpacket_kbdq_core *pkc; 899 struct tpacket_block_desc *pbd; 900 char *curr, *end; 901 902 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 903 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 904 905 /* Queue is frozen when user space is lagging behind */ 906 if (prb_queue_frozen(pkc)) { 907 /* 908 * Check if that last block which caused the queue to freeze, 909 * is still in_use by user-space. 910 */ 911 if (prb_curr_blk_in_use(pkc, pbd)) { 912 /* Can't record this packet */ 913 return NULL; 914 } else { 915 /* 916 * Ok, the block was released by user-space. 917 * Now let's open that block. 918 * opening a block also thaws the queue. 919 * Thawing is a side effect. 920 */ 921 prb_open_block(pkc, pbd); 922 } 923 } 924 925 smp_mb(); 926 curr = pkc->nxt_offset; 927 pkc->skb = skb; 928 end = (char *)pbd + pkc->kblk_size; 929 930 /* first try the current block */ 931 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 932 prb_fill_curr_block(curr, pkc, pbd, len); 933 return (void *)curr; 934 } 935 936 /* Ok, close the current block */ 937 prb_retire_current_block(pkc, po, 0); 938 939 /* Now, try to dispatch the next block */ 940 curr = (char *)prb_dispatch_next_block(pkc, po); 941 if (curr) { 942 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 943 prb_fill_curr_block(curr, pkc, pbd, len); 944 return (void *)curr; 945 } 946 947 /* 948 * No free blocks are available.user_space hasn't caught up yet. 949 * Queue was just frozen and now this packet will get dropped. 950 */ 951 return NULL; 952 } 953 954 static void *packet_current_rx_frame(struct packet_sock *po, 955 struct sk_buff *skb, 956 int status, unsigned int len) 957 { 958 char *curr = NULL; 959 switch (po->tp_version) { 960 case TPACKET_V1: 961 case TPACKET_V2: 962 curr = packet_lookup_frame(po, &po->rx_ring, 963 po->rx_ring.head, status); 964 return curr; 965 case TPACKET_V3: 966 return __packet_lookup_frame_in_block(po, skb, status, len); 967 default: 968 WARN(1, "TPACKET version not supported\n"); 969 BUG(); 970 return NULL; 971 } 972 } 973 974 static void *prb_lookup_block(struct packet_sock *po, 975 struct packet_ring_buffer *rb, 976 unsigned int previous, 977 int status) 978 { 979 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 980 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous); 981 982 if (status != BLOCK_STATUS(pbd)) 983 return NULL; 984 return pbd; 985 } 986 987 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 988 { 989 unsigned int prev; 990 if (rb->prb_bdqc.kactive_blk_num) 991 prev = rb->prb_bdqc.kactive_blk_num-1; 992 else 993 prev = rb->prb_bdqc.knum_blocks-1; 994 return prev; 995 } 996 997 /* Assumes caller has held the rx_queue.lock */ 998 static void *__prb_previous_block(struct packet_sock *po, 999 struct packet_ring_buffer *rb, 1000 int status) 1001 { 1002 unsigned int previous = prb_previous_blk_num(rb); 1003 return prb_lookup_block(po, rb, previous, status); 1004 } 1005 1006 static void *packet_previous_rx_frame(struct packet_sock *po, 1007 struct packet_ring_buffer *rb, 1008 int status) 1009 { 1010 if (po->tp_version <= TPACKET_V2) 1011 return packet_previous_frame(po, rb, status); 1012 1013 return __prb_previous_block(po, rb, status); 1014 } 1015 1016 static void packet_increment_rx_head(struct packet_sock *po, 1017 struct packet_ring_buffer *rb) 1018 { 1019 switch (po->tp_version) { 1020 case TPACKET_V1: 1021 case TPACKET_V2: 1022 return packet_increment_head(rb); 1023 case TPACKET_V3: 1024 default: 1025 WARN(1, "TPACKET version not supported.\n"); 1026 BUG(); 1027 return; 1028 } 1029 } 1030 1031 static void *packet_previous_frame(struct packet_sock *po, 1032 struct packet_ring_buffer *rb, 1033 int status) 1034 { 1035 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1036 return packet_lookup_frame(po, rb, previous, status); 1037 } 1038 1039 static void packet_increment_head(struct packet_ring_buffer *buff) 1040 { 1041 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1042 } 1043 1044 static void packet_sock_destruct(struct sock *sk) 1045 { 1046 skb_queue_purge(&sk->sk_error_queue); 1047 1048 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1049 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 1050 1051 if (!sock_flag(sk, SOCK_DEAD)) { 1052 pr_err("Attempt to release alive packet socket: %p\n", sk); 1053 return; 1054 } 1055 1056 sk_refcnt_debug_dec(sk); 1057 } 1058 1059 static int fanout_rr_next(struct packet_fanout *f, unsigned int num) 1060 { 1061 int x = atomic_read(&f->rr_cur) + 1; 1062 1063 if (x >= num) 1064 x = 0; 1065 1066 return x; 1067 } 1068 1069 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num) 1070 { 1071 u32 idx, hash = skb->rxhash; 1072 1073 idx = ((u64)hash * num) >> 32; 1074 1075 return f->arr[idx]; 1076 } 1077 1078 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num) 1079 { 1080 int cur, old; 1081 1082 cur = atomic_read(&f->rr_cur); 1083 while ((old = atomic_cmpxchg(&f->rr_cur, cur, 1084 fanout_rr_next(f, num))) != cur) 1085 cur = old; 1086 return f->arr[cur]; 1087 } 1088 1089 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num) 1090 { 1091 unsigned int cpu = smp_processor_id(); 1092 1093 return f->arr[cpu % num]; 1094 } 1095 1096 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1097 struct packet_type *pt, struct net_device *orig_dev) 1098 { 1099 struct packet_fanout *f = pt->af_packet_priv; 1100 unsigned int num = f->num_members; 1101 struct packet_sock *po; 1102 struct sock *sk; 1103 1104 if (!net_eq(dev_net(dev), read_pnet(&f->net)) || 1105 !num) { 1106 kfree_skb(skb); 1107 return 0; 1108 } 1109 1110 switch (f->type) { 1111 case PACKET_FANOUT_HASH: 1112 default: 1113 if (f->defrag) { 1114 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET); 1115 if (!skb) 1116 return 0; 1117 } 1118 skb_get_rxhash(skb); 1119 sk = fanout_demux_hash(f, skb, num); 1120 break; 1121 case PACKET_FANOUT_LB: 1122 sk = fanout_demux_lb(f, skb, num); 1123 break; 1124 case PACKET_FANOUT_CPU: 1125 sk = fanout_demux_cpu(f, skb, num); 1126 break; 1127 } 1128 1129 po = pkt_sk(sk); 1130 1131 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1132 } 1133 1134 DEFINE_MUTEX(fanout_mutex); 1135 EXPORT_SYMBOL_GPL(fanout_mutex); 1136 static LIST_HEAD(fanout_list); 1137 1138 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1139 { 1140 struct packet_fanout *f = po->fanout; 1141 1142 spin_lock(&f->lock); 1143 f->arr[f->num_members] = sk; 1144 smp_wmb(); 1145 f->num_members++; 1146 spin_unlock(&f->lock); 1147 } 1148 1149 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1150 { 1151 struct packet_fanout *f = po->fanout; 1152 int i; 1153 1154 spin_lock(&f->lock); 1155 for (i = 0; i < f->num_members; i++) { 1156 if (f->arr[i] == sk) 1157 break; 1158 } 1159 BUG_ON(i >= f->num_members); 1160 f->arr[i] = f->arr[f->num_members - 1]; 1161 f->num_members--; 1162 spin_unlock(&f->lock); 1163 } 1164 1165 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk) 1166 { 1167 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout) 1168 return true; 1169 1170 return false; 1171 } 1172 1173 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1174 { 1175 struct packet_sock *po = pkt_sk(sk); 1176 struct packet_fanout *f, *match; 1177 u8 type = type_flags & 0xff; 1178 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0; 1179 int err; 1180 1181 switch (type) { 1182 case PACKET_FANOUT_HASH: 1183 case PACKET_FANOUT_LB: 1184 case PACKET_FANOUT_CPU: 1185 break; 1186 default: 1187 return -EINVAL; 1188 } 1189 1190 if (!po->running) 1191 return -EINVAL; 1192 1193 if (po->fanout) 1194 return -EALREADY; 1195 1196 mutex_lock(&fanout_mutex); 1197 match = NULL; 1198 list_for_each_entry(f, &fanout_list, list) { 1199 if (f->id == id && 1200 read_pnet(&f->net) == sock_net(sk)) { 1201 match = f; 1202 break; 1203 } 1204 } 1205 err = -EINVAL; 1206 if (match && match->defrag != defrag) 1207 goto out; 1208 if (!match) { 1209 err = -ENOMEM; 1210 match = kzalloc(sizeof(*match), GFP_KERNEL); 1211 if (!match) 1212 goto out; 1213 write_pnet(&match->net, sock_net(sk)); 1214 match->id = id; 1215 match->type = type; 1216 match->defrag = defrag; 1217 atomic_set(&match->rr_cur, 0); 1218 INIT_LIST_HEAD(&match->list); 1219 spin_lock_init(&match->lock); 1220 atomic_set(&match->sk_ref, 0); 1221 match->prot_hook.type = po->prot_hook.type; 1222 match->prot_hook.dev = po->prot_hook.dev; 1223 match->prot_hook.func = packet_rcv_fanout; 1224 match->prot_hook.af_packet_priv = match; 1225 match->prot_hook.id_match = match_fanout_group; 1226 dev_add_pack(&match->prot_hook); 1227 list_add(&match->list, &fanout_list); 1228 } 1229 err = -EINVAL; 1230 if (match->type == type && 1231 match->prot_hook.type == po->prot_hook.type && 1232 match->prot_hook.dev == po->prot_hook.dev) { 1233 err = -ENOSPC; 1234 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1235 __dev_remove_pack(&po->prot_hook); 1236 po->fanout = match; 1237 atomic_inc(&match->sk_ref); 1238 __fanout_link(sk, po); 1239 err = 0; 1240 } 1241 } 1242 out: 1243 mutex_unlock(&fanout_mutex); 1244 return err; 1245 } 1246 1247 static void fanout_release(struct sock *sk) 1248 { 1249 struct packet_sock *po = pkt_sk(sk); 1250 struct packet_fanout *f; 1251 1252 f = po->fanout; 1253 if (!f) 1254 return; 1255 1256 mutex_lock(&fanout_mutex); 1257 po->fanout = NULL; 1258 1259 if (atomic_dec_and_test(&f->sk_ref)) { 1260 list_del(&f->list); 1261 dev_remove_pack(&f->prot_hook); 1262 kfree(f); 1263 } 1264 mutex_unlock(&fanout_mutex); 1265 } 1266 1267 static const struct proto_ops packet_ops; 1268 1269 static const struct proto_ops packet_ops_spkt; 1270 1271 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1272 struct packet_type *pt, struct net_device *orig_dev) 1273 { 1274 struct sock *sk; 1275 struct sockaddr_pkt *spkt; 1276 1277 /* 1278 * When we registered the protocol we saved the socket in the data 1279 * field for just this event. 1280 */ 1281 1282 sk = pt->af_packet_priv; 1283 1284 /* 1285 * Yank back the headers [hope the device set this 1286 * right or kerboom...] 1287 * 1288 * Incoming packets have ll header pulled, 1289 * push it back. 1290 * 1291 * For outgoing ones skb->data == skb_mac_header(skb) 1292 * so that this procedure is noop. 1293 */ 1294 1295 if (skb->pkt_type == PACKET_LOOPBACK) 1296 goto out; 1297 1298 if (!net_eq(dev_net(dev), sock_net(sk))) 1299 goto out; 1300 1301 skb = skb_share_check(skb, GFP_ATOMIC); 1302 if (skb == NULL) 1303 goto oom; 1304 1305 /* drop any routing info */ 1306 skb_dst_drop(skb); 1307 1308 /* drop conntrack reference */ 1309 nf_reset(skb); 1310 1311 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1312 1313 skb_push(skb, skb->data - skb_mac_header(skb)); 1314 1315 /* 1316 * The SOCK_PACKET socket receives _all_ frames. 1317 */ 1318 1319 spkt->spkt_family = dev->type; 1320 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1321 spkt->spkt_protocol = skb->protocol; 1322 1323 /* 1324 * Charge the memory to the socket. This is done specifically 1325 * to prevent sockets using all the memory up. 1326 */ 1327 1328 if (sock_queue_rcv_skb(sk, skb) == 0) 1329 return 0; 1330 1331 out: 1332 kfree_skb(skb); 1333 oom: 1334 return 0; 1335 } 1336 1337 1338 /* 1339 * Output a raw packet to a device layer. This bypasses all the other 1340 * protocol layers and you must therefore supply it with a complete frame 1341 */ 1342 1343 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, 1344 struct msghdr *msg, size_t len) 1345 { 1346 struct sock *sk = sock->sk; 1347 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name; 1348 struct sk_buff *skb = NULL; 1349 struct net_device *dev; 1350 __be16 proto = 0; 1351 int err; 1352 int extra_len = 0; 1353 1354 /* 1355 * Get and verify the address. 1356 */ 1357 1358 if (saddr) { 1359 if (msg->msg_namelen < sizeof(struct sockaddr)) 1360 return -EINVAL; 1361 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1362 proto = saddr->spkt_protocol; 1363 } else 1364 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1365 1366 /* 1367 * Find the device first to size check it 1368 */ 1369 1370 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1371 retry: 1372 rcu_read_lock(); 1373 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1374 err = -ENODEV; 1375 if (dev == NULL) 1376 goto out_unlock; 1377 1378 err = -ENETDOWN; 1379 if (!(dev->flags & IFF_UP)) 1380 goto out_unlock; 1381 1382 /* 1383 * You may not queue a frame bigger than the mtu. This is the lowest level 1384 * raw protocol and you must do your own fragmentation at this level. 1385 */ 1386 1387 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1388 if (!netif_supports_nofcs(dev)) { 1389 err = -EPROTONOSUPPORT; 1390 goto out_unlock; 1391 } 1392 extra_len = 4; /* We're doing our own CRC */ 1393 } 1394 1395 err = -EMSGSIZE; 1396 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1397 goto out_unlock; 1398 1399 if (!skb) { 1400 size_t reserved = LL_RESERVED_SPACE(dev); 1401 int tlen = dev->needed_tailroom; 1402 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1403 1404 rcu_read_unlock(); 1405 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1406 if (skb == NULL) 1407 return -ENOBUFS; 1408 /* FIXME: Save some space for broken drivers that write a hard 1409 * header at transmission time by themselves. PPP is the notable 1410 * one here. This should really be fixed at the driver level. 1411 */ 1412 skb_reserve(skb, reserved); 1413 skb_reset_network_header(skb); 1414 1415 /* Try to align data part correctly */ 1416 if (hhlen) { 1417 skb->data -= hhlen; 1418 skb->tail -= hhlen; 1419 if (len < hhlen) 1420 skb_reset_network_header(skb); 1421 } 1422 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len); 1423 if (err) 1424 goto out_free; 1425 goto retry; 1426 } 1427 1428 if (len > (dev->mtu + dev->hard_header_len + extra_len)) { 1429 /* Earlier code assumed this would be a VLAN pkt, 1430 * double-check this now that we have the actual 1431 * packet in hand. 1432 */ 1433 struct ethhdr *ehdr; 1434 skb_reset_mac_header(skb); 1435 ehdr = eth_hdr(skb); 1436 if (ehdr->h_proto != htons(ETH_P_8021Q)) { 1437 err = -EMSGSIZE; 1438 goto out_unlock; 1439 } 1440 } 1441 1442 skb->protocol = proto; 1443 skb->dev = dev; 1444 skb->priority = sk->sk_priority; 1445 skb->mark = sk->sk_mark; 1446 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags); 1447 if (err < 0) 1448 goto out_unlock; 1449 1450 if (unlikely(extra_len == 4)) 1451 skb->no_fcs = 1; 1452 1453 dev_queue_xmit(skb); 1454 rcu_read_unlock(); 1455 return len; 1456 1457 out_unlock: 1458 rcu_read_unlock(); 1459 out_free: 1460 kfree_skb(skb); 1461 return err; 1462 } 1463 1464 static unsigned int run_filter(const struct sk_buff *skb, 1465 const struct sock *sk, 1466 unsigned int res) 1467 { 1468 struct sk_filter *filter; 1469 1470 rcu_read_lock(); 1471 filter = rcu_dereference(sk->sk_filter); 1472 if (filter != NULL) 1473 res = SK_RUN_FILTER(filter, skb); 1474 rcu_read_unlock(); 1475 1476 return res; 1477 } 1478 1479 /* 1480 * This function makes lazy skb cloning in hope that most of packets 1481 * are discarded by BPF. 1482 * 1483 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 1484 * and skb->cb are mangled. It works because (and until) packets 1485 * falling here are owned by current CPU. Output packets are cloned 1486 * by dev_queue_xmit_nit(), input packets are processed by net_bh 1487 * sequencially, so that if we return skb to original state on exit, 1488 * we will not harm anyone. 1489 */ 1490 1491 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 1492 struct packet_type *pt, struct net_device *orig_dev) 1493 { 1494 struct sock *sk; 1495 struct sockaddr_ll *sll; 1496 struct packet_sock *po; 1497 u8 *skb_head = skb->data; 1498 int skb_len = skb->len; 1499 unsigned int snaplen, res; 1500 1501 if (skb->pkt_type == PACKET_LOOPBACK) 1502 goto drop; 1503 1504 sk = pt->af_packet_priv; 1505 po = pkt_sk(sk); 1506 1507 if (!net_eq(dev_net(dev), sock_net(sk))) 1508 goto drop; 1509 1510 skb->dev = dev; 1511 1512 if (dev->header_ops) { 1513 /* The device has an explicit notion of ll header, 1514 * exported to higher levels. 1515 * 1516 * Otherwise, the device hides details of its frame 1517 * structure, so that corresponding packet head is 1518 * never delivered to user. 1519 */ 1520 if (sk->sk_type != SOCK_DGRAM) 1521 skb_push(skb, skb->data - skb_mac_header(skb)); 1522 else if (skb->pkt_type == PACKET_OUTGOING) { 1523 /* Special case: outgoing packets have ll header at head */ 1524 skb_pull(skb, skb_network_offset(skb)); 1525 } 1526 } 1527 1528 snaplen = skb->len; 1529 1530 res = run_filter(skb, sk, snaplen); 1531 if (!res) 1532 goto drop_n_restore; 1533 if (snaplen > res) 1534 snaplen = res; 1535 1536 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 1537 goto drop_n_acct; 1538 1539 if (skb_shared(skb)) { 1540 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 1541 if (nskb == NULL) 1542 goto drop_n_acct; 1543 1544 if (skb_head != skb->data) { 1545 skb->data = skb_head; 1546 skb->len = skb_len; 1547 } 1548 consume_skb(skb); 1549 skb = nskb; 1550 } 1551 1552 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 > 1553 sizeof(skb->cb)); 1554 1555 sll = &PACKET_SKB_CB(skb)->sa.ll; 1556 sll->sll_family = AF_PACKET; 1557 sll->sll_hatype = dev->type; 1558 sll->sll_protocol = skb->protocol; 1559 sll->sll_pkttype = skb->pkt_type; 1560 if (unlikely(po->origdev)) 1561 sll->sll_ifindex = orig_dev->ifindex; 1562 else 1563 sll->sll_ifindex = dev->ifindex; 1564 1565 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 1566 1567 PACKET_SKB_CB(skb)->origlen = skb->len; 1568 1569 if (pskb_trim(skb, snaplen)) 1570 goto drop_n_acct; 1571 1572 skb_set_owner_r(skb, sk); 1573 skb->dev = NULL; 1574 skb_dst_drop(skb); 1575 1576 /* drop conntrack reference */ 1577 nf_reset(skb); 1578 1579 spin_lock(&sk->sk_receive_queue.lock); 1580 po->stats.tp_packets++; 1581 skb->dropcount = atomic_read(&sk->sk_drops); 1582 __skb_queue_tail(&sk->sk_receive_queue, skb); 1583 spin_unlock(&sk->sk_receive_queue.lock); 1584 sk->sk_data_ready(sk, skb->len); 1585 return 0; 1586 1587 drop_n_acct: 1588 spin_lock(&sk->sk_receive_queue.lock); 1589 po->stats.tp_drops++; 1590 atomic_inc(&sk->sk_drops); 1591 spin_unlock(&sk->sk_receive_queue.lock); 1592 1593 drop_n_restore: 1594 if (skb_head != skb->data && skb_shared(skb)) { 1595 skb->data = skb_head; 1596 skb->len = skb_len; 1597 } 1598 drop: 1599 consume_skb(skb); 1600 return 0; 1601 } 1602 1603 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 1604 struct packet_type *pt, struct net_device *orig_dev) 1605 { 1606 struct sock *sk; 1607 struct packet_sock *po; 1608 struct sockaddr_ll *sll; 1609 union { 1610 struct tpacket_hdr *h1; 1611 struct tpacket2_hdr *h2; 1612 struct tpacket3_hdr *h3; 1613 void *raw; 1614 } h; 1615 u8 *skb_head = skb->data; 1616 int skb_len = skb->len; 1617 unsigned int snaplen, res; 1618 unsigned long status = TP_STATUS_USER; 1619 unsigned short macoff, netoff, hdrlen; 1620 struct sk_buff *copy_skb = NULL; 1621 struct timeval tv; 1622 struct timespec ts; 1623 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 1624 1625 if (skb->pkt_type == PACKET_LOOPBACK) 1626 goto drop; 1627 1628 sk = pt->af_packet_priv; 1629 po = pkt_sk(sk); 1630 1631 if (!net_eq(dev_net(dev), sock_net(sk))) 1632 goto drop; 1633 1634 if (dev->header_ops) { 1635 if (sk->sk_type != SOCK_DGRAM) 1636 skb_push(skb, skb->data - skb_mac_header(skb)); 1637 else if (skb->pkt_type == PACKET_OUTGOING) { 1638 /* Special case: outgoing packets have ll header at head */ 1639 skb_pull(skb, skb_network_offset(skb)); 1640 } 1641 } 1642 1643 if (skb->ip_summed == CHECKSUM_PARTIAL) 1644 status |= TP_STATUS_CSUMNOTREADY; 1645 1646 snaplen = skb->len; 1647 1648 res = run_filter(skb, sk, snaplen); 1649 if (!res) 1650 goto drop_n_restore; 1651 if (snaplen > res) 1652 snaplen = res; 1653 1654 if (sk->sk_type == SOCK_DGRAM) { 1655 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 1656 po->tp_reserve; 1657 } else { 1658 unsigned int maclen = skb_network_offset(skb); 1659 netoff = TPACKET_ALIGN(po->tp_hdrlen + 1660 (maclen < 16 ? 16 : maclen)) + 1661 po->tp_reserve; 1662 macoff = netoff - maclen; 1663 } 1664 if (po->tp_version <= TPACKET_V2) { 1665 if (macoff + snaplen > po->rx_ring.frame_size) { 1666 if (po->copy_thresh && 1667 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1668 if (skb_shared(skb)) { 1669 copy_skb = skb_clone(skb, GFP_ATOMIC); 1670 } else { 1671 copy_skb = skb_get(skb); 1672 skb_head = skb->data; 1673 } 1674 if (copy_skb) 1675 skb_set_owner_r(copy_skb, sk); 1676 } 1677 snaplen = po->rx_ring.frame_size - macoff; 1678 if ((int)snaplen < 0) 1679 snaplen = 0; 1680 } 1681 } 1682 spin_lock(&sk->sk_receive_queue.lock); 1683 h.raw = packet_current_rx_frame(po, skb, 1684 TP_STATUS_KERNEL, (macoff+snaplen)); 1685 if (!h.raw) 1686 goto ring_is_full; 1687 if (po->tp_version <= TPACKET_V2) { 1688 packet_increment_rx_head(po, &po->rx_ring); 1689 /* 1690 * LOSING will be reported till you read the stats, 1691 * because it's COR - Clear On Read. 1692 * Anyways, moving it for V1/V2 only as V3 doesn't need this 1693 * at packet level. 1694 */ 1695 if (po->stats.tp_drops) 1696 status |= TP_STATUS_LOSING; 1697 } 1698 po->stats.tp_packets++; 1699 if (copy_skb) { 1700 status |= TP_STATUS_COPY; 1701 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 1702 } 1703 spin_unlock(&sk->sk_receive_queue.lock); 1704 1705 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 1706 1707 switch (po->tp_version) { 1708 case TPACKET_V1: 1709 h.h1->tp_len = skb->len; 1710 h.h1->tp_snaplen = snaplen; 1711 h.h1->tp_mac = macoff; 1712 h.h1->tp_net = netoff; 1713 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE) 1714 && shhwtstamps->syststamp.tv64) 1715 tv = ktime_to_timeval(shhwtstamps->syststamp); 1716 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE) 1717 && shhwtstamps->hwtstamp.tv64) 1718 tv = ktime_to_timeval(shhwtstamps->hwtstamp); 1719 else if (skb->tstamp.tv64) 1720 tv = ktime_to_timeval(skb->tstamp); 1721 else 1722 do_gettimeofday(&tv); 1723 h.h1->tp_sec = tv.tv_sec; 1724 h.h1->tp_usec = tv.tv_usec; 1725 hdrlen = sizeof(*h.h1); 1726 break; 1727 case TPACKET_V2: 1728 h.h2->tp_len = skb->len; 1729 h.h2->tp_snaplen = snaplen; 1730 h.h2->tp_mac = macoff; 1731 h.h2->tp_net = netoff; 1732 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE) 1733 && shhwtstamps->syststamp.tv64) 1734 ts = ktime_to_timespec(shhwtstamps->syststamp); 1735 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE) 1736 && shhwtstamps->hwtstamp.tv64) 1737 ts = ktime_to_timespec(shhwtstamps->hwtstamp); 1738 else if (skb->tstamp.tv64) 1739 ts = ktime_to_timespec(skb->tstamp); 1740 else 1741 getnstimeofday(&ts); 1742 h.h2->tp_sec = ts.tv_sec; 1743 h.h2->tp_nsec = ts.tv_nsec; 1744 if (vlan_tx_tag_present(skb)) { 1745 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb); 1746 status |= TP_STATUS_VLAN_VALID; 1747 } else { 1748 h.h2->tp_vlan_tci = 0; 1749 } 1750 h.h2->tp_padding = 0; 1751 hdrlen = sizeof(*h.h2); 1752 break; 1753 case TPACKET_V3: 1754 /* tp_nxt_offset,vlan are already populated above. 1755 * So DONT clear those fields here 1756 */ 1757 h.h3->tp_status |= status; 1758 h.h3->tp_len = skb->len; 1759 h.h3->tp_snaplen = snaplen; 1760 h.h3->tp_mac = macoff; 1761 h.h3->tp_net = netoff; 1762 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE) 1763 && shhwtstamps->syststamp.tv64) 1764 ts = ktime_to_timespec(shhwtstamps->syststamp); 1765 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE) 1766 && shhwtstamps->hwtstamp.tv64) 1767 ts = ktime_to_timespec(shhwtstamps->hwtstamp); 1768 else if (skb->tstamp.tv64) 1769 ts = ktime_to_timespec(skb->tstamp); 1770 else 1771 getnstimeofday(&ts); 1772 h.h3->tp_sec = ts.tv_sec; 1773 h.h3->tp_nsec = ts.tv_nsec; 1774 hdrlen = sizeof(*h.h3); 1775 break; 1776 default: 1777 BUG(); 1778 } 1779 1780 sll = h.raw + TPACKET_ALIGN(hdrlen); 1781 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 1782 sll->sll_family = AF_PACKET; 1783 sll->sll_hatype = dev->type; 1784 sll->sll_protocol = skb->protocol; 1785 sll->sll_pkttype = skb->pkt_type; 1786 if (unlikely(po->origdev)) 1787 sll->sll_ifindex = orig_dev->ifindex; 1788 else 1789 sll->sll_ifindex = dev->ifindex; 1790 1791 smp_mb(); 1792 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 1793 { 1794 u8 *start, *end; 1795 1796 if (po->tp_version <= TPACKET_V2) { 1797 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw 1798 + macoff + snaplen); 1799 for (start = h.raw; start < end; start += PAGE_SIZE) 1800 flush_dcache_page(pgv_to_page(start)); 1801 } 1802 smp_wmb(); 1803 } 1804 #endif 1805 if (po->tp_version <= TPACKET_V2) 1806 __packet_set_status(po, h.raw, status); 1807 else 1808 prb_clear_blk_fill_status(&po->rx_ring); 1809 1810 sk->sk_data_ready(sk, 0); 1811 1812 drop_n_restore: 1813 if (skb_head != skb->data && skb_shared(skb)) { 1814 skb->data = skb_head; 1815 skb->len = skb_len; 1816 } 1817 drop: 1818 kfree_skb(skb); 1819 return 0; 1820 1821 ring_is_full: 1822 po->stats.tp_drops++; 1823 spin_unlock(&sk->sk_receive_queue.lock); 1824 1825 sk->sk_data_ready(sk, 0); 1826 kfree_skb(copy_skb); 1827 goto drop_n_restore; 1828 } 1829 1830 static void tpacket_destruct_skb(struct sk_buff *skb) 1831 { 1832 struct packet_sock *po = pkt_sk(skb->sk); 1833 void *ph; 1834 1835 if (likely(po->tx_ring.pg_vec)) { 1836 ph = skb_shinfo(skb)->destructor_arg; 1837 BUG_ON(atomic_read(&po->tx_ring.pending) == 0); 1838 atomic_dec(&po->tx_ring.pending); 1839 __packet_set_status(po, ph, TP_STATUS_AVAILABLE); 1840 } 1841 1842 sock_wfree(skb); 1843 } 1844 1845 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 1846 void *frame, struct net_device *dev, int size_max, 1847 __be16 proto, unsigned char *addr, int hlen) 1848 { 1849 union { 1850 struct tpacket_hdr *h1; 1851 struct tpacket2_hdr *h2; 1852 void *raw; 1853 } ph; 1854 int to_write, offset, len, tp_len, nr_frags, len_max; 1855 struct socket *sock = po->sk.sk_socket; 1856 struct page *page; 1857 void *data; 1858 int err; 1859 1860 ph.raw = frame; 1861 1862 skb->protocol = proto; 1863 skb->dev = dev; 1864 skb->priority = po->sk.sk_priority; 1865 skb->mark = po->sk.sk_mark; 1866 skb_shinfo(skb)->destructor_arg = ph.raw; 1867 1868 switch (po->tp_version) { 1869 case TPACKET_V2: 1870 tp_len = ph.h2->tp_len; 1871 break; 1872 default: 1873 tp_len = ph.h1->tp_len; 1874 break; 1875 } 1876 if (unlikely(tp_len > size_max)) { 1877 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 1878 return -EMSGSIZE; 1879 } 1880 1881 skb_reserve(skb, hlen); 1882 skb_reset_network_header(skb); 1883 1884 if (po->tp_tx_has_off) { 1885 int off_min, off_max, off; 1886 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 1887 off_max = po->tx_ring.frame_size - tp_len; 1888 if (sock->type == SOCK_DGRAM) { 1889 switch (po->tp_version) { 1890 case TPACKET_V2: 1891 off = ph.h2->tp_net; 1892 break; 1893 default: 1894 off = ph.h1->tp_net; 1895 break; 1896 } 1897 } else { 1898 switch (po->tp_version) { 1899 case TPACKET_V2: 1900 off = ph.h2->tp_mac; 1901 break; 1902 default: 1903 off = ph.h1->tp_mac; 1904 break; 1905 } 1906 } 1907 if (unlikely((off < off_min) || (off_max < off))) 1908 return -EINVAL; 1909 data = ph.raw + off; 1910 } else { 1911 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll); 1912 } 1913 to_write = tp_len; 1914 1915 if (sock->type == SOCK_DGRAM) { 1916 err = dev_hard_header(skb, dev, ntohs(proto), addr, 1917 NULL, tp_len); 1918 if (unlikely(err < 0)) 1919 return -EINVAL; 1920 } else if (dev->hard_header_len) { 1921 /* net device doesn't like empty head */ 1922 if (unlikely(tp_len <= dev->hard_header_len)) { 1923 pr_err("packet size is too short (%d < %d)\n", 1924 tp_len, dev->hard_header_len); 1925 return -EINVAL; 1926 } 1927 1928 skb_push(skb, dev->hard_header_len); 1929 err = skb_store_bits(skb, 0, data, 1930 dev->hard_header_len); 1931 if (unlikely(err)) 1932 return err; 1933 1934 data += dev->hard_header_len; 1935 to_write -= dev->hard_header_len; 1936 } 1937 1938 offset = offset_in_page(data); 1939 len_max = PAGE_SIZE - offset; 1940 len = ((to_write > len_max) ? len_max : to_write); 1941 1942 skb->data_len = to_write; 1943 skb->len += to_write; 1944 skb->truesize += to_write; 1945 atomic_add(to_write, &po->sk.sk_wmem_alloc); 1946 1947 while (likely(to_write)) { 1948 nr_frags = skb_shinfo(skb)->nr_frags; 1949 1950 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 1951 pr_err("Packet exceed the number of skb frags(%lu)\n", 1952 MAX_SKB_FRAGS); 1953 return -EFAULT; 1954 } 1955 1956 page = pgv_to_page(data); 1957 data += len; 1958 flush_dcache_page(page); 1959 get_page(page); 1960 skb_fill_page_desc(skb, nr_frags, page, offset, len); 1961 to_write -= len; 1962 offset = 0; 1963 len_max = PAGE_SIZE; 1964 len = ((to_write > len_max) ? len_max : to_write); 1965 } 1966 1967 return tp_len; 1968 } 1969 1970 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 1971 { 1972 struct sk_buff *skb; 1973 struct net_device *dev; 1974 __be16 proto; 1975 bool need_rls_dev = false; 1976 int err, reserve = 0; 1977 void *ph; 1978 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name; 1979 int tp_len, size_max; 1980 unsigned char *addr; 1981 int len_sum = 0; 1982 int status = TP_STATUS_AVAILABLE; 1983 int hlen, tlen; 1984 1985 mutex_lock(&po->pg_vec_lock); 1986 1987 if (saddr == NULL) { 1988 dev = po->prot_hook.dev; 1989 proto = po->num; 1990 addr = NULL; 1991 } else { 1992 err = -EINVAL; 1993 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 1994 goto out; 1995 if (msg->msg_namelen < (saddr->sll_halen 1996 + offsetof(struct sockaddr_ll, 1997 sll_addr))) 1998 goto out; 1999 proto = saddr->sll_protocol; 2000 addr = saddr->sll_addr; 2001 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2002 need_rls_dev = true; 2003 } 2004 2005 err = -ENXIO; 2006 if (unlikely(dev == NULL)) 2007 goto out; 2008 2009 reserve = dev->hard_header_len; 2010 2011 err = -ENETDOWN; 2012 if (unlikely(!(dev->flags & IFF_UP))) 2013 goto out_put; 2014 2015 size_max = po->tx_ring.frame_size 2016 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2017 2018 if (size_max > dev->mtu + reserve) 2019 size_max = dev->mtu + reserve; 2020 2021 do { 2022 ph = packet_current_frame(po, &po->tx_ring, 2023 TP_STATUS_SEND_REQUEST); 2024 2025 if (unlikely(ph == NULL)) { 2026 schedule(); 2027 continue; 2028 } 2029 2030 status = TP_STATUS_SEND_REQUEST; 2031 hlen = LL_RESERVED_SPACE(dev); 2032 tlen = dev->needed_tailroom; 2033 skb = sock_alloc_send_skb(&po->sk, 2034 hlen + tlen + sizeof(struct sockaddr_ll), 2035 0, &err); 2036 2037 if (unlikely(skb == NULL)) 2038 goto out_status; 2039 2040 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto, 2041 addr, hlen); 2042 2043 if (unlikely(tp_len < 0)) { 2044 if (po->tp_loss) { 2045 __packet_set_status(po, ph, 2046 TP_STATUS_AVAILABLE); 2047 packet_increment_head(&po->tx_ring); 2048 kfree_skb(skb); 2049 continue; 2050 } else { 2051 status = TP_STATUS_WRONG_FORMAT; 2052 err = tp_len; 2053 goto out_status; 2054 } 2055 } 2056 2057 skb->destructor = tpacket_destruct_skb; 2058 __packet_set_status(po, ph, TP_STATUS_SENDING); 2059 atomic_inc(&po->tx_ring.pending); 2060 2061 status = TP_STATUS_SEND_REQUEST; 2062 err = dev_queue_xmit(skb); 2063 if (unlikely(err > 0)) { 2064 err = net_xmit_errno(err); 2065 if (err && __packet_get_status(po, ph) == 2066 TP_STATUS_AVAILABLE) { 2067 /* skb was destructed already */ 2068 skb = NULL; 2069 goto out_status; 2070 } 2071 /* 2072 * skb was dropped but not destructed yet; 2073 * let's treat it like congestion or err < 0 2074 */ 2075 err = 0; 2076 } 2077 packet_increment_head(&po->tx_ring); 2078 len_sum += tp_len; 2079 } while (likely((ph != NULL) || 2080 ((!(msg->msg_flags & MSG_DONTWAIT)) && 2081 (atomic_read(&po->tx_ring.pending)))) 2082 ); 2083 2084 err = len_sum; 2085 goto out_put; 2086 2087 out_status: 2088 __packet_set_status(po, ph, status); 2089 kfree_skb(skb); 2090 out_put: 2091 if (need_rls_dev) 2092 dev_put(dev); 2093 out: 2094 mutex_unlock(&po->pg_vec_lock); 2095 return err; 2096 } 2097 2098 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2099 size_t reserve, size_t len, 2100 size_t linear, int noblock, 2101 int *err) 2102 { 2103 struct sk_buff *skb; 2104 2105 /* Under a page? Don't bother with paged skb. */ 2106 if (prepad + len < PAGE_SIZE || !linear) 2107 linear = len; 2108 2109 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2110 err); 2111 if (!skb) 2112 return NULL; 2113 2114 skb_reserve(skb, reserve); 2115 skb_put(skb, linear); 2116 skb->data_len = len - linear; 2117 skb->len += len - linear; 2118 2119 return skb; 2120 } 2121 2122 static int packet_snd(struct socket *sock, 2123 struct msghdr *msg, size_t len) 2124 { 2125 struct sock *sk = sock->sk; 2126 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name; 2127 struct sk_buff *skb; 2128 struct net_device *dev; 2129 __be16 proto; 2130 bool need_rls_dev = false; 2131 unsigned char *addr; 2132 int err, reserve = 0; 2133 struct virtio_net_hdr vnet_hdr = { 0 }; 2134 int offset = 0; 2135 int vnet_hdr_len; 2136 struct packet_sock *po = pkt_sk(sk); 2137 unsigned short gso_type = 0; 2138 int hlen, tlen; 2139 int extra_len = 0; 2140 2141 /* 2142 * Get and verify the address. 2143 */ 2144 2145 if (saddr == NULL) { 2146 dev = po->prot_hook.dev; 2147 proto = po->num; 2148 addr = NULL; 2149 } else { 2150 err = -EINVAL; 2151 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2152 goto out; 2153 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2154 goto out; 2155 proto = saddr->sll_protocol; 2156 addr = saddr->sll_addr; 2157 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2158 need_rls_dev = true; 2159 } 2160 2161 err = -ENXIO; 2162 if (dev == NULL) 2163 goto out_unlock; 2164 if (sock->type == SOCK_RAW) 2165 reserve = dev->hard_header_len; 2166 2167 err = -ENETDOWN; 2168 if (!(dev->flags & IFF_UP)) 2169 goto out_unlock; 2170 2171 if (po->has_vnet_hdr) { 2172 vnet_hdr_len = sizeof(vnet_hdr); 2173 2174 err = -EINVAL; 2175 if (len < vnet_hdr_len) 2176 goto out_unlock; 2177 2178 len -= vnet_hdr_len; 2179 2180 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov, 2181 vnet_hdr_len); 2182 if (err < 0) 2183 goto out_unlock; 2184 2185 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2186 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > 2187 vnet_hdr.hdr_len)) 2188 vnet_hdr.hdr_len = vnet_hdr.csum_start + 2189 vnet_hdr.csum_offset + 2; 2190 2191 err = -EINVAL; 2192 if (vnet_hdr.hdr_len > len) 2193 goto out_unlock; 2194 2195 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { 2196 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 2197 case VIRTIO_NET_HDR_GSO_TCPV4: 2198 gso_type = SKB_GSO_TCPV4; 2199 break; 2200 case VIRTIO_NET_HDR_GSO_TCPV6: 2201 gso_type = SKB_GSO_TCPV6; 2202 break; 2203 case VIRTIO_NET_HDR_GSO_UDP: 2204 gso_type = SKB_GSO_UDP; 2205 break; 2206 default: 2207 goto out_unlock; 2208 } 2209 2210 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) 2211 gso_type |= SKB_GSO_TCP_ECN; 2212 2213 if (vnet_hdr.gso_size == 0) 2214 goto out_unlock; 2215 2216 } 2217 } 2218 2219 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2220 if (!netif_supports_nofcs(dev)) { 2221 err = -EPROTONOSUPPORT; 2222 goto out_unlock; 2223 } 2224 extra_len = 4; /* We're doing our own CRC */ 2225 } 2226 2227 err = -EMSGSIZE; 2228 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2229 goto out_unlock; 2230 2231 err = -ENOBUFS; 2232 hlen = LL_RESERVED_SPACE(dev); 2233 tlen = dev->needed_tailroom; 2234 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len, 2235 msg->msg_flags & MSG_DONTWAIT, &err); 2236 if (skb == NULL) 2237 goto out_unlock; 2238 2239 skb_set_network_header(skb, reserve); 2240 2241 err = -EINVAL; 2242 if (sock->type == SOCK_DGRAM && 2243 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0) 2244 goto out_free; 2245 2246 /* Returns -EFAULT on error */ 2247 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len); 2248 if (err) 2249 goto out_free; 2250 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags); 2251 if (err < 0) 2252 goto out_free; 2253 2254 if (!gso_type && (len > dev->mtu + reserve + extra_len)) { 2255 /* Earlier code assumed this would be a VLAN pkt, 2256 * double-check this now that we have the actual 2257 * packet in hand. 2258 */ 2259 struct ethhdr *ehdr; 2260 skb_reset_mac_header(skb); 2261 ehdr = eth_hdr(skb); 2262 if (ehdr->h_proto != htons(ETH_P_8021Q)) { 2263 err = -EMSGSIZE; 2264 goto out_free; 2265 } 2266 } 2267 2268 skb->protocol = proto; 2269 skb->dev = dev; 2270 skb->priority = sk->sk_priority; 2271 skb->mark = sk->sk_mark; 2272 2273 if (po->has_vnet_hdr) { 2274 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 2275 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start, 2276 vnet_hdr.csum_offset)) { 2277 err = -EINVAL; 2278 goto out_free; 2279 } 2280 } 2281 2282 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size; 2283 skb_shinfo(skb)->gso_type = gso_type; 2284 2285 /* Header must be checked, and gso_segs computed. */ 2286 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 2287 skb_shinfo(skb)->gso_segs = 0; 2288 2289 len += vnet_hdr_len; 2290 } 2291 2292 if (unlikely(extra_len == 4)) 2293 skb->no_fcs = 1; 2294 2295 /* 2296 * Now send it 2297 */ 2298 2299 err = dev_queue_xmit(skb); 2300 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2301 goto out_unlock; 2302 2303 if (need_rls_dev) 2304 dev_put(dev); 2305 2306 return len; 2307 2308 out_free: 2309 kfree_skb(skb); 2310 out_unlock: 2311 if (dev && need_rls_dev) 2312 dev_put(dev); 2313 out: 2314 return err; 2315 } 2316 2317 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock, 2318 struct msghdr *msg, size_t len) 2319 { 2320 struct sock *sk = sock->sk; 2321 struct packet_sock *po = pkt_sk(sk); 2322 if (po->tx_ring.pg_vec) 2323 return tpacket_snd(po, msg); 2324 else 2325 return packet_snd(sock, msg, len); 2326 } 2327 2328 /* 2329 * Close a PACKET socket. This is fairly simple. We immediately go 2330 * to 'closed' state and remove our protocol entry in the device list. 2331 */ 2332 2333 static int packet_release(struct socket *sock) 2334 { 2335 struct sock *sk = sock->sk; 2336 struct packet_sock *po; 2337 struct net *net; 2338 union tpacket_req_u req_u; 2339 2340 if (!sk) 2341 return 0; 2342 2343 net = sock_net(sk); 2344 po = pkt_sk(sk); 2345 2346 mutex_lock(&net->packet.sklist_lock); 2347 sk_del_node_init_rcu(sk); 2348 mutex_unlock(&net->packet.sklist_lock); 2349 2350 preempt_disable(); 2351 sock_prot_inuse_add(net, sk->sk_prot, -1); 2352 preempt_enable(); 2353 2354 spin_lock(&po->bind_lock); 2355 unregister_prot_hook(sk, false); 2356 if (po->prot_hook.dev) { 2357 dev_put(po->prot_hook.dev); 2358 po->prot_hook.dev = NULL; 2359 } 2360 spin_unlock(&po->bind_lock); 2361 2362 packet_flush_mclist(sk); 2363 2364 memset(&req_u, 0, sizeof(req_u)); 2365 2366 if (po->rx_ring.pg_vec) 2367 packet_set_ring(sk, &req_u, 1, 0); 2368 2369 if (po->tx_ring.pg_vec) 2370 packet_set_ring(sk, &req_u, 1, 1); 2371 2372 fanout_release(sk); 2373 2374 synchronize_net(); 2375 /* 2376 * Now the socket is dead. No more input will appear. 2377 */ 2378 sock_orphan(sk); 2379 sock->sk = NULL; 2380 2381 /* Purge queues */ 2382 2383 skb_queue_purge(&sk->sk_receive_queue); 2384 sk_refcnt_debug_release(sk); 2385 2386 sock_put(sk); 2387 return 0; 2388 } 2389 2390 /* 2391 * Attach a packet hook. 2392 */ 2393 2394 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol) 2395 { 2396 struct packet_sock *po = pkt_sk(sk); 2397 2398 if (po->fanout) { 2399 if (dev) 2400 dev_put(dev); 2401 2402 return -EINVAL; 2403 } 2404 2405 lock_sock(sk); 2406 2407 spin_lock(&po->bind_lock); 2408 unregister_prot_hook(sk, true); 2409 po->num = protocol; 2410 po->prot_hook.type = protocol; 2411 if (po->prot_hook.dev) 2412 dev_put(po->prot_hook.dev); 2413 po->prot_hook.dev = dev; 2414 2415 po->ifindex = dev ? dev->ifindex : 0; 2416 2417 if (protocol == 0) 2418 goto out_unlock; 2419 2420 if (!dev || (dev->flags & IFF_UP)) { 2421 register_prot_hook(sk); 2422 } else { 2423 sk->sk_err = ENETDOWN; 2424 if (!sock_flag(sk, SOCK_DEAD)) 2425 sk->sk_error_report(sk); 2426 } 2427 2428 out_unlock: 2429 spin_unlock(&po->bind_lock); 2430 release_sock(sk); 2431 return 0; 2432 } 2433 2434 /* 2435 * Bind a packet socket to a device 2436 */ 2437 2438 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 2439 int addr_len) 2440 { 2441 struct sock *sk = sock->sk; 2442 char name[15]; 2443 struct net_device *dev; 2444 int err = -ENODEV; 2445 2446 /* 2447 * Check legality 2448 */ 2449 2450 if (addr_len != sizeof(struct sockaddr)) 2451 return -EINVAL; 2452 strlcpy(name, uaddr->sa_data, sizeof(name)); 2453 2454 dev = dev_get_by_name(sock_net(sk), name); 2455 if (dev) 2456 err = packet_do_bind(sk, dev, pkt_sk(sk)->num); 2457 return err; 2458 } 2459 2460 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 2461 { 2462 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 2463 struct sock *sk = sock->sk; 2464 struct net_device *dev = NULL; 2465 int err; 2466 2467 2468 /* 2469 * Check legality 2470 */ 2471 2472 if (addr_len < sizeof(struct sockaddr_ll)) 2473 return -EINVAL; 2474 if (sll->sll_family != AF_PACKET) 2475 return -EINVAL; 2476 2477 if (sll->sll_ifindex) { 2478 err = -ENODEV; 2479 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex); 2480 if (dev == NULL) 2481 goto out; 2482 } 2483 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num); 2484 2485 out: 2486 return err; 2487 } 2488 2489 static struct proto packet_proto = { 2490 .name = "PACKET", 2491 .owner = THIS_MODULE, 2492 .obj_size = sizeof(struct packet_sock), 2493 }; 2494 2495 /* 2496 * Create a packet of type SOCK_PACKET. 2497 */ 2498 2499 static int packet_create(struct net *net, struct socket *sock, int protocol, 2500 int kern) 2501 { 2502 struct sock *sk; 2503 struct packet_sock *po; 2504 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 2505 int err; 2506 2507 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 2508 return -EPERM; 2509 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 2510 sock->type != SOCK_PACKET) 2511 return -ESOCKTNOSUPPORT; 2512 2513 sock->state = SS_UNCONNECTED; 2514 2515 err = -ENOBUFS; 2516 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto); 2517 if (sk == NULL) 2518 goto out; 2519 2520 sock->ops = &packet_ops; 2521 if (sock->type == SOCK_PACKET) 2522 sock->ops = &packet_ops_spkt; 2523 2524 sock_init_data(sock, sk); 2525 2526 po = pkt_sk(sk); 2527 sk->sk_family = PF_PACKET; 2528 po->num = proto; 2529 2530 sk->sk_destruct = packet_sock_destruct; 2531 sk_refcnt_debug_inc(sk); 2532 2533 /* 2534 * Attach a protocol block 2535 */ 2536 2537 spin_lock_init(&po->bind_lock); 2538 mutex_init(&po->pg_vec_lock); 2539 po->prot_hook.func = packet_rcv; 2540 2541 if (sock->type == SOCK_PACKET) 2542 po->prot_hook.func = packet_rcv_spkt; 2543 2544 po->prot_hook.af_packet_priv = sk; 2545 2546 if (proto) { 2547 po->prot_hook.type = proto; 2548 register_prot_hook(sk); 2549 } 2550 2551 mutex_lock(&net->packet.sklist_lock); 2552 sk_add_node_rcu(sk, &net->packet.sklist); 2553 mutex_unlock(&net->packet.sklist_lock); 2554 2555 preempt_disable(); 2556 sock_prot_inuse_add(net, &packet_proto, 1); 2557 preempt_enable(); 2558 2559 return 0; 2560 out: 2561 return err; 2562 } 2563 2564 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len) 2565 { 2566 struct sock_exterr_skb *serr; 2567 struct sk_buff *skb, *skb2; 2568 int copied, err; 2569 2570 err = -EAGAIN; 2571 skb = skb_dequeue(&sk->sk_error_queue); 2572 if (skb == NULL) 2573 goto out; 2574 2575 copied = skb->len; 2576 if (copied > len) { 2577 msg->msg_flags |= MSG_TRUNC; 2578 copied = len; 2579 } 2580 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 2581 if (err) 2582 goto out_free_skb; 2583 2584 sock_recv_timestamp(msg, sk, skb); 2585 2586 serr = SKB_EXT_ERR(skb); 2587 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP, 2588 sizeof(serr->ee), &serr->ee); 2589 2590 msg->msg_flags |= MSG_ERRQUEUE; 2591 err = copied; 2592 2593 /* Reset and regenerate socket error */ 2594 spin_lock_bh(&sk->sk_error_queue.lock); 2595 sk->sk_err = 0; 2596 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) { 2597 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno; 2598 spin_unlock_bh(&sk->sk_error_queue.lock); 2599 sk->sk_error_report(sk); 2600 } else 2601 spin_unlock_bh(&sk->sk_error_queue.lock); 2602 2603 out_free_skb: 2604 kfree_skb(skb); 2605 out: 2606 return err; 2607 } 2608 2609 /* 2610 * Pull a packet from our receive queue and hand it to the user. 2611 * If necessary we block. 2612 */ 2613 2614 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock, 2615 struct msghdr *msg, size_t len, int flags) 2616 { 2617 struct sock *sk = sock->sk; 2618 struct sk_buff *skb; 2619 int copied, err; 2620 struct sockaddr_ll *sll; 2621 int vnet_hdr_len = 0; 2622 2623 err = -EINVAL; 2624 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 2625 goto out; 2626 2627 #if 0 2628 /* What error should we return now? EUNATTACH? */ 2629 if (pkt_sk(sk)->ifindex < 0) 2630 return -ENODEV; 2631 #endif 2632 2633 if (flags & MSG_ERRQUEUE) { 2634 err = packet_recv_error(sk, msg, len); 2635 goto out; 2636 } 2637 2638 /* 2639 * Call the generic datagram receiver. This handles all sorts 2640 * of horrible races and re-entrancy so we can forget about it 2641 * in the protocol layers. 2642 * 2643 * Now it will return ENETDOWN, if device have just gone down, 2644 * but then it will block. 2645 */ 2646 2647 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 2648 2649 /* 2650 * An error occurred so return it. Because skb_recv_datagram() 2651 * handles the blocking we don't see and worry about blocking 2652 * retries. 2653 */ 2654 2655 if (skb == NULL) 2656 goto out; 2657 2658 if (pkt_sk(sk)->has_vnet_hdr) { 2659 struct virtio_net_hdr vnet_hdr = { 0 }; 2660 2661 err = -EINVAL; 2662 vnet_hdr_len = sizeof(vnet_hdr); 2663 if (len < vnet_hdr_len) 2664 goto out_free; 2665 2666 len -= vnet_hdr_len; 2667 2668 if (skb_is_gso(skb)) { 2669 struct skb_shared_info *sinfo = skb_shinfo(skb); 2670 2671 /* This is a hint as to how much should be linear. */ 2672 vnet_hdr.hdr_len = skb_headlen(skb); 2673 vnet_hdr.gso_size = sinfo->gso_size; 2674 if (sinfo->gso_type & SKB_GSO_TCPV4) 2675 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 2676 else if (sinfo->gso_type & SKB_GSO_TCPV6) 2677 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 2678 else if (sinfo->gso_type & SKB_GSO_UDP) 2679 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP; 2680 else if (sinfo->gso_type & SKB_GSO_FCOE) 2681 goto out_free; 2682 else 2683 BUG(); 2684 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 2685 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; 2686 } else 2687 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; 2688 2689 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2690 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 2691 vnet_hdr.csum_start = skb_checksum_start_offset(skb); 2692 vnet_hdr.csum_offset = skb->csum_offset; 2693 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2694 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID; 2695 } /* else everything is zero */ 2696 2697 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr, 2698 vnet_hdr_len); 2699 if (err < 0) 2700 goto out_free; 2701 } 2702 2703 /* 2704 * If the address length field is there to be filled in, we fill 2705 * it in now. 2706 */ 2707 2708 sll = &PACKET_SKB_CB(skb)->sa.ll; 2709 if (sock->type == SOCK_PACKET) 2710 msg->msg_namelen = sizeof(struct sockaddr_pkt); 2711 else 2712 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr); 2713 2714 /* 2715 * You lose any data beyond the buffer you gave. If it worries a 2716 * user program they can ask the device for its MTU anyway. 2717 */ 2718 2719 copied = skb->len; 2720 if (copied > len) { 2721 copied = len; 2722 msg->msg_flags |= MSG_TRUNC; 2723 } 2724 2725 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 2726 if (err) 2727 goto out_free; 2728 2729 sock_recv_ts_and_drops(msg, sk, skb); 2730 2731 if (msg->msg_name) 2732 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, 2733 msg->msg_namelen); 2734 2735 if (pkt_sk(sk)->auxdata) { 2736 struct tpacket_auxdata aux; 2737 2738 aux.tp_status = TP_STATUS_USER; 2739 if (skb->ip_summed == CHECKSUM_PARTIAL) 2740 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 2741 aux.tp_len = PACKET_SKB_CB(skb)->origlen; 2742 aux.tp_snaplen = skb->len; 2743 aux.tp_mac = 0; 2744 aux.tp_net = skb_network_offset(skb); 2745 if (vlan_tx_tag_present(skb)) { 2746 aux.tp_vlan_tci = vlan_tx_tag_get(skb); 2747 aux.tp_status |= TP_STATUS_VLAN_VALID; 2748 } else { 2749 aux.tp_vlan_tci = 0; 2750 } 2751 aux.tp_padding = 0; 2752 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 2753 } 2754 2755 /* 2756 * Free or return the buffer as appropriate. Again this 2757 * hides all the races and re-entrancy issues from us. 2758 */ 2759 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 2760 2761 out_free: 2762 skb_free_datagram(sk, skb); 2763 out: 2764 return err; 2765 } 2766 2767 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 2768 int *uaddr_len, int peer) 2769 { 2770 struct net_device *dev; 2771 struct sock *sk = sock->sk; 2772 2773 if (peer) 2774 return -EOPNOTSUPP; 2775 2776 uaddr->sa_family = AF_PACKET; 2777 rcu_read_lock(); 2778 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 2779 if (dev) 2780 strncpy(uaddr->sa_data, dev->name, 14); 2781 else 2782 memset(uaddr->sa_data, 0, 14); 2783 rcu_read_unlock(); 2784 *uaddr_len = sizeof(*uaddr); 2785 2786 return 0; 2787 } 2788 2789 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 2790 int *uaddr_len, int peer) 2791 { 2792 struct net_device *dev; 2793 struct sock *sk = sock->sk; 2794 struct packet_sock *po = pkt_sk(sk); 2795 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 2796 2797 if (peer) 2798 return -EOPNOTSUPP; 2799 2800 sll->sll_family = AF_PACKET; 2801 sll->sll_ifindex = po->ifindex; 2802 sll->sll_protocol = po->num; 2803 sll->sll_pkttype = 0; 2804 rcu_read_lock(); 2805 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 2806 if (dev) { 2807 sll->sll_hatype = dev->type; 2808 sll->sll_halen = dev->addr_len; 2809 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 2810 } else { 2811 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 2812 sll->sll_halen = 0; 2813 } 2814 rcu_read_unlock(); 2815 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 2816 2817 return 0; 2818 } 2819 2820 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 2821 int what) 2822 { 2823 switch (i->type) { 2824 case PACKET_MR_MULTICAST: 2825 if (i->alen != dev->addr_len) 2826 return -EINVAL; 2827 if (what > 0) 2828 return dev_mc_add(dev, i->addr); 2829 else 2830 return dev_mc_del(dev, i->addr); 2831 break; 2832 case PACKET_MR_PROMISC: 2833 return dev_set_promiscuity(dev, what); 2834 break; 2835 case PACKET_MR_ALLMULTI: 2836 return dev_set_allmulti(dev, what); 2837 break; 2838 case PACKET_MR_UNICAST: 2839 if (i->alen != dev->addr_len) 2840 return -EINVAL; 2841 if (what > 0) 2842 return dev_uc_add(dev, i->addr); 2843 else 2844 return dev_uc_del(dev, i->addr); 2845 break; 2846 default: 2847 break; 2848 } 2849 return 0; 2850 } 2851 2852 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what) 2853 { 2854 for ( ; i; i = i->next) { 2855 if (i->ifindex == dev->ifindex) 2856 packet_dev_mc(dev, i, what); 2857 } 2858 } 2859 2860 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 2861 { 2862 struct packet_sock *po = pkt_sk(sk); 2863 struct packet_mclist *ml, *i; 2864 struct net_device *dev; 2865 int err; 2866 2867 rtnl_lock(); 2868 2869 err = -ENODEV; 2870 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 2871 if (!dev) 2872 goto done; 2873 2874 err = -EINVAL; 2875 if (mreq->mr_alen > dev->addr_len) 2876 goto done; 2877 2878 err = -ENOBUFS; 2879 i = kmalloc(sizeof(*i), GFP_KERNEL); 2880 if (i == NULL) 2881 goto done; 2882 2883 err = 0; 2884 for (ml = po->mclist; ml; ml = ml->next) { 2885 if (ml->ifindex == mreq->mr_ifindex && 2886 ml->type == mreq->mr_type && 2887 ml->alen == mreq->mr_alen && 2888 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 2889 ml->count++; 2890 /* Free the new element ... */ 2891 kfree(i); 2892 goto done; 2893 } 2894 } 2895 2896 i->type = mreq->mr_type; 2897 i->ifindex = mreq->mr_ifindex; 2898 i->alen = mreq->mr_alen; 2899 memcpy(i->addr, mreq->mr_address, i->alen); 2900 i->count = 1; 2901 i->next = po->mclist; 2902 po->mclist = i; 2903 err = packet_dev_mc(dev, i, 1); 2904 if (err) { 2905 po->mclist = i->next; 2906 kfree(i); 2907 } 2908 2909 done: 2910 rtnl_unlock(); 2911 return err; 2912 } 2913 2914 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 2915 { 2916 struct packet_mclist *ml, **mlp; 2917 2918 rtnl_lock(); 2919 2920 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 2921 if (ml->ifindex == mreq->mr_ifindex && 2922 ml->type == mreq->mr_type && 2923 ml->alen == mreq->mr_alen && 2924 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 2925 if (--ml->count == 0) { 2926 struct net_device *dev; 2927 *mlp = ml->next; 2928 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 2929 if (dev) 2930 packet_dev_mc(dev, ml, -1); 2931 kfree(ml); 2932 } 2933 rtnl_unlock(); 2934 return 0; 2935 } 2936 } 2937 rtnl_unlock(); 2938 return -EADDRNOTAVAIL; 2939 } 2940 2941 static void packet_flush_mclist(struct sock *sk) 2942 { 2943 struct packet_sock *po = pkt_sk(sk); 2944 struct packet_mclist *ml; 2945 2946 if (!po->mclist) 2947 return; 2948 2949 rtnl_lock(); 2950 while ((ml = po->mclist) != NULL) { 2951 struct net_device *dev; 2952 2953 po->mclist = ml->next; 2954 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 2955 if (dev != NULL) 2956 packet_dev_mc(dev, ml, -1); 2957 kfree(ml); 2958 } 2959 rtnl_unlock(); 2960 } 2961 2962 static int 2963 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 2964 { 2965 struct sock *sk = sock->sk; 2966 struct packet_sock *po = pkt_sk(sk); 2967 int ret; 2968 2969 if (level != SOL_PACKET) 2970 return -ENOPROTOOPT; 2971 2972 switch (optname) { 2973 case PACKET_ADD_MEMBERSHIP: 2974 case PACKET_DROP_MEMBERSHIP: 2975 { 2976 struct packet_mreq_max mreq; 2977 int len = optlen; 2978 memset(&mreq, 0, sizeof(mreq)); 2979 if (len < sizeof(struct packet_mreq)) 2980 return -EINVAL; 2981 if (len > sizeof(mreq)) 2982 len = sizeof(mreq); 2983 if (copy_from_user(&mreq, optval, len)) 2984 return -EFAULT; 2985 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 2986 return -EINVAL; 2987 if (optname == PACKET_ADD_MEMBERSHIP) 2988 ret = packet_mc_add(sk, &mreq); 2989 else 2990 ret = packet_mc_drop(sk, &mreq); 2991 return ret; 2992 } 2993 2994 case PACKET_RX_RING: 2995 case PACKET_TX_RING: 2996 { 2997 union tpacket_req_u req_u; 2998 int len; 2999 3000 switch (po->tp_version) { 3001 case TPACKET_V1: 3002 case TPACKET_V2: 3003 len = sizeof(req_u.req); 3004 break; 3005 case TPACKET_V3: 3006 default: 3007 len = sizeof(req_u.req3); 3008 break; 3009 } 3010 if (optlen < len) 3011 return -EINVAL; 3012 if (pkt_sk(sk)->has_vnet_hdr) 3013 return -EINVAL; 3014 if (copy_from_user(&req_u.req, optval, len)) 3015 return -EFAULT; 3016 return packet_set_ring(sk, &req_u, 0, 3017 optname == PACKET_TX_RING); 3018 } 3019 case PACKET_COPY_THRESH: 3020 { 3021 int val; 3022 3023 if (optlen != sizeof(val)) 3024 return -EINVAL; 3025 if (copy_from_user(&val, optval, sizeof(val))) 3026 return -EFAULT; 3027 3028 pkt_sk(sk)->copy_thresh = val; 3029 return 0; 3030 } 3031 case PACKET_VERSION: 3032 { 3033 int val; 3034 3035 if (optlen != sizeof(val)) 3036 return -EINVAL; 3037 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3038 return -EBUSY; 3039 if (copy_from_user(&val, optval, sizeof(val))) 3040 return -EFAULT; 3041 switch (val) { 3042 case TPACKET_V1: 3043 case TPACKET_V2: 3044 case TPACKET_V3: 3045 po->tp_version = val; 3046 return 0; 3047 default: 3048 return -EINVAL; 3049 } 3050 } 3051 case PACKET_RESERVE: 3052 { 3053 unsigned int val; 3054 3055 if (optlen != sizeof(val)) 3056 return -EINVAL; 3057 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3058 return -EBUSY; 3059 if (copy_from_user(&val, optval, sizeof(val))) 3060 return -EFAULT; 3061 po->tp_reserve = val; 3062 return 0; 3063 } 3064 case PACKET_LOSS: 3065 { 3066 unsigned int val; 3067 3068 if (optlen != sizeof(val)) 3069 return -EINVAL; 3070 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3071 return -EBUSY; 3072 if (copy_from_user(&val, optval, sizeof(val))) 3073 return -EFAULT; 3074 po->tp_loss = !!val; 3075 return 0; 3076 } 3077 case PACKET_AUXDATA: 3078 { 3079 int val; 3080 3081 if (optlen < sizeof(val)) 3082 return -EINVAL; 3083 if (copy_from_user(&val, optval, sizeof(val))) 3084 return -EFAULT; 3085 3086 po->auxdata = !!val; 3087 return 0; 3088 } 3089 case PACKET_ORIGDEV: 3090 { 3091 int val; 3092 3093 if (optlen < sizeof(val)) 3094 return -EINVAL; 3095 if (copy_from_user(&val, optval, sizeof(val))) 3096 return -EFAULT; 3097 3098 po->origdev = !!val; 3099 return 0; 3100 } 3101 case PACKET_VNET_HDR: 3102 { 3103 int val; 3104 3105 if (sock->type != SOCK_RAW) 3106 return -EINVAL; 3107 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3108 return -EBUSY; 3109 if (optlen < sizeof(val)) 3110 return -EINVAL; 3111 if (copy_from_user(&val, optval, sizeof(val))) 3112 return -EFAULT; 3113 3114 po->has_vnet_hdr = !!val; 3115 return 0; 3116 } 3117 case PACKET_TIMESTAMP: 3118 { 3119 int val; 3120 3121 if (optlen != sizeof(val)) 3122 return -EINVAL; 3123 if (copy_from_user(&val, optval, sizeof(val))) 3124 return -EFAULT; 3125 3126 po->tp_tstamp = val; 3127 return 0; 3128 } 3129 case PACKET_FANOUT: 3130 { 3131 int val; 3132 3133 if (optlen != sizeof(val)) 3134 return -EINVAL; 3135 if (copy_from_user(&val, optval, sizeof(val))) 3136 return -EFAULT; 3137 3138 return fanout_add(sk, val & 0xffff, val >> 16); 3139 } 3140 case PACKET_TX_HAS_OFF: 3141 { 3142 unsigned int val; 3143 3144 if (optlen != sizeof(val)) 3145 return -EINVAL; 3146 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3147 return -EBUSY; 3148 if (copy_from_user(&val, optval, sizeof(val))) 3149 return -EFAULT; 3150 po->tp_tx_has_off = !!val; 3151 return 0; 3152 } 3153 default: 3154 return -ENOPROTOOPT; 3155 } 3156 } 3157 3158 static int packet_getsockopt(struct socket *sock, int level, int optname, 3159 char __user *optval, int __user *optlen) 3160 { 3161 int len; 3162 int val, lv = sizeof(val); 3163 struct sock *sk = sock->sk; 3164 struct packet_sock *po = pkt_sk(sk); 3165 void *data = &val; 3166 struct tpacket_stats st; 3167 union tpacket_stats_u st_u; 3168 3169 if (level != SOL_PACKET) 3170 return -ENOPROTOOPT; 3171 3172 if (get_user(len, optlen)) 3173 return -EFAULT; 3174 3175 if (len < 0) 3176 return -EINVAL; 3177 3178 switch (optname) { 3179 case PACKET_STATISTICS: 3180 spin_lock_bh(&sk->sk_receive_queue.lock); 3181 if (po->tp_version == TPACKET_V3) { 3182 lv = sizeof(struct tpacket_stats_v3); 3183 memcpy(&st_u.stats3, &po->stats, 3184 sizeof(struct tpacket_stats)); 3185 st_u.stats3.tp_freeze_q_cnt = 3186 po->stats_u.stats3.tp_freeze_q_cnt; 3187 st_u.stats3.tp_packets += po->stats.tp_drops; 3188 data = &st_u.stats3; 3189 } else { 3190 lv = sizeof(struct tpacket_stats); 3191 st = po->stats; 3192 st.tp_packets += st.tp_drops; 3193 data = &st; 3194 } 3195 memset(&po->stats, 0, sizeof(st)); 3196 spin_unlock_bh(&sk->sk_receive_queue.lock); 3197 break; 3198 case PACKET_AUXDATA: 3199 val = po->auxdata; 3200 break; 3201 case PACKET_ORIGDEV: 3202 val = po->origdev; 3203 break; 3204 case PACKET_VNET_HDR: 3205 val = po->has_vnet_hdr; 3206 break; 3207 case PACKET_VERSION: 3208 val = po->tp_version; 3209 break; 3210 case PACKET_HDRLEN: 3211 if (len > sizeof(int)) 3212 len = sizeof(int); 3213 if (copy_from_user(&val, optval, len)) 3214 return -EFAULT; 3215 switch (val) { 3216 case TPACKET_V1: 3217 val = sizeof(struct tpacket_hdr); 3218 break; 3219 case TPACKET_V2: 3220 val = sizeof(struct tpacket2_hdr); 3221 break; 3222 case TPACKET_V3: 3223 val = sizeof(struct tpacket3_hdr); 3224 break; 3225 default: 3226 return -EINVAL; 3227 } 3228 break; 3229 case PACKET_RESERVE: 3230 val = po->tp_reserve; 3231 break; 3232 case PACKET_LOSS: 3233 val = po->tp_loss; 3234 break; 3235 case PACKET_TIMESTAMP: 3236 val = po->tp_tstamp; 3237 break; 3238 case PACKET_FANOUT: 3239 val = (po->fanout ? 3240 ((u32)po->fanout->id | 3241 ((u32)po->fanout->type << 16)) : 3242 0); 3243 break; 3244 case PACKET_TX_HAS_OFF: 3245 val = po->tp_tx_has_off; 3246 break; 3247 default: 3248 return -ENOPROTOOPT; 3249 } 3250 3251 if (len > lv) 3252 len = lv; 3253 if (put_user(len, optlen)) 3254 return -EFAULT; 3255 if (copy_to_user(optval, data, len)) 3256 return -EFAULT; 3257 return 0; 3258 } 3259 3260 3261 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data) 3262 { 3263 struct sock *sk; 3264 struct hlist_node *node; 3265 struct net_device *dev = data; 3266 struct net *net = dev_net(dev); 3267 3268 rcu_read_lock(); 3269 sk_for_each_rcu(sk, node, &net->packet.sklist) { 3270 struct packet_sock *po = pkt_sk(sk); 3271 3272 switch (msg) { 3273 case NETDEV_UNREGISTER: 3274 if (po->mclist) 3275 packet_dev_mclist(dev, po->mclist, -1); 3276 /* fallthrough */ 3277 3278 case NETDEV_DOWN: 3279 if (dev->ifindex == po->ifindex) { 3280 spin_lock(&po->bind_lock); 3281 if (po->running) { 3282 __unregister_prot_hook(sk, false); 3283 sk->sk_err = ENETDOWN; 3284 if (!sock_flag(sk, SOCK_DEAD)) 3285 sk->sk_error_report(sk); 3286 } 3287 if (msg == NETDEV_UNREGISTER) { 3288 po->ifindex = -1; 3289 if (po->prot_hook.dev) 3290 dev_put(po->prot_hook.dev); 3291 po->prot_hook.dev = NULL; 3292 } 3293 spin_unlock(&po->bind_lock); 3294 } 3295 break; 3296 case NETDEV_UP: 3297 if (dev->ifindex == po->ifindex) { 3298 spin_lock(&po->bind_lock); 3299 if (po->num) 3300 register_prot_hook(sk); 3301 spin_unlock(&po->bind_lock); 3302 } 3303 break; 3304 } 3305 } 3306 rcu_read_unlock(); 3307 return NOTIFY_DONE; 3308 } 3309 3310 3311 static int packet_ioctl(struct socket *sock, unsigned int cmd, 3312 unsigned long arg) 3313 { 3314 struct sock *sk = sock->sk; 3315 3316 switch (cmd) { 3317 case SIOCOUTQ: 3318 { 3319 int amount = sk_wmem_alloc_get(sk); 3320 3321 return put_user(amount, (int __user *)arg); 3322 } 3323 case SIOCINQ: 3324 { 3325 struct sk_buff *skb; 3326 int amount = 0; 3327 3328 spin_lock_bh(&sk->sk_receive_queue.lock); 3329 skb = skb_peek(&sk->sk_receive_queue); 3330 if (skb) 3331 amount = skb->len; 3332 spin_unlock_bh(&sk->sk_receive_queue.lock); 3333 return put_user(amount, (int __user *)arg); 3334 } 3335 case SIOCGSTAMP: 3336 return sock_get_timestamp(sk, (struct timeval __user *)arg); 3337 case SIOCGSTAMPNS: 3338 return sock_get_timestampns(sk, (struct timespec __user *)arg); 3339 3340 #ifdef CONFIG_INET 3341 case SIOCADDRT: 3342 case SIOCDELRT: 3343 case SIOCDARP: 3344 case SIOCGARP: 3345 case SIOCSARP: 3346 case SIOCGIFADDR: 3347 case SIOCSIFADDR: 3348 case SIOCGIFBRDADDR: 3349 case SIOCSIFBRDADDR: 3350 case SIOCGIFNETMASK: 3351 case SIOCSIFNETMASK: 3352 case SIOCGIFDSTADDR: 3353 case SIOCSIFDSTADDR: 3354 case SIOCSIFFLAGS: 3355 return inet_dgram_ops.ioctl(sock, cmd, arg); 3356 #endif 3357 3358 default: 3359 return -ENOIOCTLCMD; 3360 } 3361 return 0; 3362 } 3363 3364 static unsigned int packet_poll(struct file *file, struct socket *sock, 3365 poll_table *wait) 3366 { 3367 struct sock *sk = sock->sk; 3368 struct packet_sock *po = pkt_sk(sk); 3369 unsigned int mask = datagram_poll(file, sock, wait); 3370 3371 spin_lock_bh(&sk->sk_receive_queue.lock); 3372 if (po->rx_ring.pg_vec) { 3373 if (!packet_previous_rx_frame(po, &po->rx_ring, 3374 TP_STATUS_KERNEL)) 3375 mask |= POLLIN | POLLRDNORM; 3376 } 3377 spin_unlock_bh(&sk->sk_receive_queue.lock); 3378 spin_lock_bh(&sk->sk_write_queue.lock); 3379 if (po->tx_ring.pg_vec) { 3380 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 3381 mask |= POLLOUT | POLLWRNORM; 3382 } 3383 spin_unlock_bh(&sk->sk_write_queue.lock); 3384 return mask; 3385 } 3386 3387 3388 /* Dirty? Well, I still did not learn better way to account 3389 * for user mmaps. 3390 */ 3391 3392 static void packet_mm_open(struct vm_area_struct *vma) 3393 { 3394 struct file *file = vma->vm_file; 3395 struct socket *sock = file->private_data; 3396 struct sock *sk = sock->sk; 3397 3398 if (sk) 3399 atomic_inc(&pkt_sk(sk)->mapped); 3400 } 3401 3402 static void packet_mm_close(struct vm_area_struct *vma) 3403 { 3404 struct file *file = vma->vm_file; 3405 struct socket *sock = file->private_data; 3406 struct sock *sk = sock->sk; 3407 3408 if (sk) 3409 atomic_dec(&pkt_sk(sk)->mapped); 3410 } 3411 3412 static const struct vm_operations_struct packet_mmap_ops = { 3413 .open = packet_mm_open, 3414 .close = packet_mm_close, 3415 }; 3416 3417 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 3418 unsigned int len) 3419 { 3420 int i; 3421 3422 for (i = 0; i < len; i++) { 3423 if (likely(pg_vec[i].buffer)) { 3424 if (is_vmalloc_addr(pg_vec[i].buffer)) 3425 vfree(pg_vec[i].buffer); 3426 else 3427 free_pages((unsigned long)pg_vec[i].buffer, 3428 order); 3429 pg_vec[i].buffer = NULL; 3430 } 3431 } 3432 kfree(pg_vec); 3433 } 3434 3435 static char *alloc_one_pg_vec_page(unsigned long order) 3436 { 3437 char *buffer = NULL; 3438 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 3439 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 3440 3441 buffer = (char *) __get_free_pages(gfp_flags, order); 3442 3443 if (buffer) 3444 return buffer; 3445 3446 /* 3447 * __get_free_pages failed, fall back to vmalloc 3448 */ 3449 buffer = vzalloc((1 << order) * PAGE_SIZE); 3450 3451 if (buffer) 3452 return buffer; 3453 3454 /* 3455 * vmalloc failed, lets dig into swap here 3456 */ 3457 gfp_flags &= ~__GFP_NORETRY; 3458 buffer = (char *)__get_free_pages(gfp_flags, order); 3459 if (buffer) 3460 return buffer; 3461 3462 /* 3463 * complete and utter failure 3464 */ 3465 return NULL; 3466 } 3467 3468 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 3469 { 3470 unsigned int block_nr = req->tp_block_nr; 3471 struct pgv *pg_vec; 3472 int i; 3473 3474 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL); 3475 if (unlikely(!pg_vec)) 3476 goto out; 3477 3478 for (i = 0; i < block_nr; i++) { 3479 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 3480 if (unlikely(!pg_vec[i].buffer)) 3481 goto out_free_pgvec; 3482 } 3483 3484 out: 3485 return pg_vec; 3486 3487 out_free_pgvec: 3488 free_pg_vec(pg_vec, order, block_nr); 3489 pg_vec = NULL; 3490 goto out; 3491 } 3492 3493 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 3494 int closing, int tx_ring) 3495 { 3496 struct pgv *pg_vec = NULL; 3497 struct packet_sock *po = pkt_sk(sk); 3498 int was_running, order = 0; 3499 struct packet_ring_buffer *rb; 3500 struct sk_buff_head *rb_queue; 3501 __be16 num; 3502 int err = -EINVAL; 3503 /* Added to avoid minimal code churn */ 3504 struct tpacket_req *req = &req_u->req; 3505 3506 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */ 3507 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) { 3508 WARN(1, "Tx-ring is not supported.\n"); 3509 goto out; 3510 } 3511 3512 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 3513 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 3514 3515 err = -EBUSY; 3516 if (!closing) { 3517 if (atomic_read(&po->mapped)) 3518 goto out; 3519 if (atomic_read(&rb->pending)) 3520 goto out; 3521 } 3522 3523 if (req->tp_block_nr) { 3524 /* Sanity tests and some calculations */ 3525 err = -EBUSY; 3526 if (unlikely(rb->pg_vec)) 3527 goto out; 3528 3529 switch (po->tp_version) { 3530 case TPACKET_V1: 3531 po->tp_hdrlen = TPACKET_HDRLEN; 3532 break; 3533 case TPACKET_V2: 3534 po->tp_hdrlen = TPACKET2_HDRLEN; 3535 break; 3536 case TPACKET_V3: 3537 po->tp_hdrlen = TPACKET3_HDRLEN; 3538 break; 3539 } 3540 3541 err = -EINVAL; 3542 if (unlikely((int)req->tp_block_size <= 0)) 3543 goto out; 3544 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1))) 3545 goto out; 3546 if (unlikely(req->tp_frame_size < po->tp_hdrlen + 3547 po->tp_reserve)) 3548 goto out; 3549 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 3550 goto out; 3551 3552 rb->frames_per_block = req->tp_block_size/req->tp_frame_size; 3553 if (unlikely(rb->frames_per_block <= 0)) 3554 goto out; 3555 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 3556 req->tp_frame_nr)) 3557 goto out; 3558 3559 err = -ENOMEM; 3560 order = get_order(req->tp_block_size); 3561 pg_vec = alloc_pg_vec(req, order); 3562 if (unlikely(!pg_vec)) 3563 goto out; 3564 switch (po->tp_version) { 3565 case TPACKET_V3: 3566 /* Transmit path is not supported. We checked 3567 * it above but just being paranoid 3568 */ 3569 if (!tx_ring) 3570 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring); 3571 break; 3572 default: 3573 break; 3574 } 3575 } 3576 /* Done */ 3577 else { 3578 err = -EINVAL; 3579 if (unlikely(req->tp_frame_nr)) 3580 goto out; 3581 } 3582 3583 lock_sock(sk); 3584 3585 /* Detach socket from network */ 3586 spin_lock(&po->bind_lock); 3587 was_running = po->running; 3588 num = po->num; 3589 if (was_running) { 3590 po->num = 0; 3591 __unregister_prot_hook(sk, false); 3592 } 3593 spin_unlock(&po->bind_lock); 3594 3595 synchronize_net(); 3596 3597 err = -EBUSY; 3598 mutex_lock(&po->pg_vec_lock); 3599 if (closing || atomic_read(&po->mapped) == 0) { 3600 err = 0; 3601 spin_lock_bh(&rb_queue->lock); 3602 swap(rb->pg_vec, pg_vec); 3603 rb->frame_max = (req->tp_frame_nr - 1); 3604 rb->head = 0; 3605 rb->frame_size = req->tp_frame_size; 3606 spin_unlock_bh(&rb_queue->lock); 3607 3608 swap(rb->pg_vec_order, order); 3609 swap(rb->pg_vec_len, req->tp_block_nr); 3610 3611 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 3612 po->prot_hook.func = (po->rx_ring.pg_vec) ? 3613 tpacket_rcv : packet_rcv; 3614 skb_queue_purge(rb_queue); 3615 if (atomic_read(&po->mapped)) 3616 pr_err("packet_mmap: vma is busy: %d\n", 3617 atomic_read(&po->mapped)); 3618 } 3619 mutex_unlock(&po->pg_vec_lock); 3620 3621 spin_lock(&po->bind_lock); 3622 if (was_running) { 3623 po->num = num; 3624 register_prot_hook(sk); 3625 } 3626 spin_unlock(&po->bind_lock); 3627 if (closing && (po->tp_version > TPACKET_V2)) { 3628 /* Because we don't support block-based V3 on tx-ring */ 3629 if (!tx_ring) 3630 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue); 3631 } 3632 release_sock(sk); 3633 3634 if (pg_vec) 3635 free_pg_vec(pg_vec, order, req->tp_block_nr); 3636 out: 3637 return err; 3638 } 3639 3640 static int packet_mmap(struct file *file, struct socket *sock, 3641 struct vm_area_struct *vma) 3642 { 3643 struct sock *sk = sock->sk; 3644 struct packet_sock *po = pkt_sk(sk); 3645 unsigned long size, expected_size; 3646 struct packet_ring_buffer *rb; 3647 unsigned long start; 3648 int err = -EINVAL; 3649 int i; 3650 3651 if (vma->vm_pgoff) 3652 return -EINVAL; 3653 3654 mutex_lock(&po->pg_vec_lock); 3655 3656 expected_size = 0; 3657 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 3658 if (rb->pg_vec) { 3659 expected_size += rb->pg_vec_len 3660 * rb->pg_vec_pages 3661 * PAGE_SIZE; 3662 } 3663 } 3664 3665 if (expected_size == 0) 3666 goto out; 3667 3668 size = vma->vm_end - vma->vm_start; 3669 if (size != expected_size) 3670 goto out; 3671 3672 start = vma->vm_start; 3673 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 3674 if (rb->pg_vec == NULL) 3675 continue; 3676 3677 for (i = 0; i < rb->pg_vec_len; i++) { 3678 struct page *page; 3679 void *kaddr = rb->pg_vec[i].buffer; 3680 int pg_num; 3681 3682 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 3683 page = pgv_to_page(kaddr); 3684 err = vm_insert_page(vma, start, page); 3685 if (unlikely(err)) 3686 goto out; 3687 start += PAGE_SIZE; 3688 kaddr += PAGE_SIZE; 3689 } 3690 } 3691 } 3692 3693 atomic_inc(&po->mapped); 3694 vma->vm_ops = &packet_mmap_ops; 3695 err = 0; 3696 3697 out: 3698 mutex_unlock(&po->pg_vec_lock); 3699 return err; 3700 } 3701 3702 static const struct proto_ops packet_ops_spkt = { 3703 .family = PF_PACKET, 3704 .owner = THIS_MODULE, 3705 .release = packet_release, 3706 .bind = packet_bind_spkt, 3707 .connect = sock_no_connect, 3708 .socketpair = sock_no_socketpair, 3709 .accept = sock_no_accept, 3710 .getname = packet_getname_spkt, 3711 .poll = datagram_poll, 3712 .ioctl = packet_ioctl, 3713 .listen = sock_no_listen, 3714 .shutdown = sock_no_shutdown, 3715 .setsockopt = sock_no_setsockopt, 3716 .getsockopt = sock_no_getsockopt, 3717 .sendmsg = packet_sendmsg_spkt, 3718 .recvmsg = packet_recvmsg, 3719 .mmap = sock_no_mmap, 3720 .sendpage = sock_no_sendpage, 3721 }; 3722 3723 static const struct proto_ops packet_ops = { 3724 .family = PF_PACKET, 3725 .owner = THIS_MODULE, 3726 .release = packet_release, 3727 .bind = packet_bind, 3728 .connect = sock_no_connect, 3729 .socketpair = sock_no_socketpair, 3730 .accept = sock_no_accept, 3731 .getname = packet_getname, 3732 .poll = packet_poll, 3733 .ioctl = packet_ioctl, 3734 .listen = sock_no_listen, 3735 .shutdown = sock_no_shutdown, 3736 .setsockopt = packet_setsockopt, 3737 .getsockopt = packet_getsockopt, 3738 .sendmsg = packet_sendmsg, 3739 .recvmsg = packet_recvmsg, 3740 .mmap = packet_mmap, 3741 .sendpage = sock_no_sendpage, 3742 }; 3743 3744 static const struct net_proto_family packet_family_ops = { 3745 .family = PF_PACKET, 3746 .create = packet_create, 3747 .owner = THIS_MODULE, 3748 }; 3749 3750 static struct notifier_block packet_netdev_notifier = { 3751 .notifier_call = packet_notifier, 3752 }; 3753 3754 #ifdef CONFIG_PROC_FS 3755 3756 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 3757 __acquires(RCU) 3758 { 3759 struct net *net = seq_file_net(seq); 3760 3761 rcu_read_lock(); 3762 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 3763 } 3764 3765 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3766 { 3767 struct net *net = seq_file_net(seq); 3768 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 3769 } 3770 3771 static void packet_seq_stop(struct seq_file *seq, void *v) 3772 __releases(RCU) 3773 { 3774 rcu_read_unlock(); 3775 } 3776 3777 static int packet_seq_show(struct seq_file *seq, void *v) 3778 { 3779 if (v == SEQ_START_TOKEN) 3780 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 3781 else { 3782 struct sock *s = sk_entry(v); 3783 const struct packet_sock *po = pkt_sk(s); 3784 3785 seq_printf(seq, 3786 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 3787 s, 3788 atomic_read(&s->sk_refcnt), 3789 s->sk_type, 3790 ntohs(po->num), 3791 po->ifindex, 3792 po->running, 3793 atomic_read(&s->sk_rmem_alloc), 3794 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 3795 sock_i_ino(s)); 3796 } 3797 3798 return 0; 3799 } 3800 3801 static const struct seq_operations packet_seq_ops = { 3802 .start = packet_seq_start, 3803 .next = packet_seq_next, 3804 .stop = packet_seq_stop, 3805 .show = packet_seq_show, 3806 }; 3807 3808 static int packet_seq_open(struct inode *inode, struct file *file) 3809 { 3810 return seq_open_net(inode, file, &packet_seq_ops, 3811 sizeof(struct seq_net_private)); 3812 } 3813 3814 static const struct file_operations packet_seq_fops = { 3815 .owner = THIS_MODULE, 3816 .open = packet_seq_open, 3817 .read = seq_read, 3818 .llseek = seq_lseek, 3819 .release = seq_release_net, 3820 }; 3821 3822 #endif 3823 3824 static int __net_init packet_net_init(struct net *net) 3825 { 3826 mutex_init(&net->packet.sklist_lock); 3827 INIT_HLIST_HEAD(&net->packet.sklist); 3828 3829 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops)) 3830 return -ENOMEM; 3831 3832 return 0; 3833 } 3834 3835 static void __net_exit packet_net_exit(struct net *net) 3836 { 3837 proc_net_remove(net, "packet"); 3838 } 3839 3840 static struct pernet_operations packet_net_ops = { 3841 .init = packet_net_init, 3842 .exit = packet_net_exit, 3843 }; 3844 3845 3846 static void __exit packet_exit(void) 3847 { 3848 unregister_netdevice_notifier(&packet_netdev_notifier); 3849 unregister_pernet_subsys(&packet_net_ops); 3850 sock_unregister(PF_PACKET); 3851 proto_unregister(&packet_proto); 3852 } 3853 3854 static int __init packet_init(void) 3855 { 3856 int rc = proto_register(&packet_proto, 0); 3857 3858 if (rc != 0) 3859 goto out; 3860 3861 sock_register(&packet_family_ops); 3862 register_pernet_subsys(&packet_net_ops); 3863 register_netdevice_notifier(&packet_netdev_notifier); 3864 out: 3865 return rc; 3866 } 3867 3868 module_init(packet_init); 3869 module_exit(packet_exit); 3870 MODULE_LICENSE("GPL"); 3871 MODULE_ALIAS_NETPROTO(PF_PACKET); 3872