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 #include <linux/percpu.h> 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 union tpacket_uhdr { 162 struct tpacket_hdr *h1; 163 struct tpacket2_hdr *h2; 164 struct tpacket3_hdr *h3; 165 void *raw; 166 }; 167 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 169 int closing, int tx_ring); 170 171 #define V3_ALIGNMENT (8) 172 173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) 174 175 #define BLK_PLUS_PRIV(sz_of_priv) \ 176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) 177 178 #define PGV_FROM_VMALLOC 1 179 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) 181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) 182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) 183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) 184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) 186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x))) 187 188 struct packet_sock; 189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg); 190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 191 struct packet_type *pt, struct net_device *orig_dev); 192 193 static void *packet_previous_frame(struct packet_sock *po, 194 struct packet_ring_buffer *rb, 195 int status); 196 static void packet_increment_head(struct packet_ring_buffer *buff); 197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *, 198 struct tpacket_block_desc *); 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 200 struct packet_sock *); 201 static void prb_retire_current_block(struct tpacket_kbdq_core *, 202 struct packet_sock *, unsigned int status); 203 static int prb_queue_frozen(struct tpacket_kbdq_core *); 204 static void prb_open_block(struct tpacket_kbdq_core *, 205 struct tpacket_block_desc *); 206 static void prb_retire_rx_blk_timer_expired(unsigned long); 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); 208 static void prb_init_blk_timer(struct packet_sock *, 209 struct tpacket_kbdq_core *, 210 void (*func) (unsigned long)); 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 213 struct tpacket3_hdr *); 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 215 struct tpacket3_hdr *); 216 static void packet_flush_mclist(struct sock *sk); 217 218 struct packet_skb_cb { 219 unsigned int origlen; 220 union { 221 struct sockaddr_pkt pkt; 222 struct sockaddr_ll ll; 223 } sa; 224 }; 225 226 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 227 228 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 229 #define GET_PBLOCK_DESC(x, bid) \ 230 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 232 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 233 #define GET_NEXT_PRB_BLK_NUM(x) \ 234 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 235 ((x)->kactive_blk_num+1) : 0) 236 237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 238 static void __fanout_link(struct sock *sk, struct packet_sock *po); 239 240 static int packet_direct_xmit(struct sk_buff *skb) 241 { 242 struct net_device *dev = skb->dev; 243 const struct net_device_ops *ops = dev->netdev_ops; 244 netdev_features_t features; 245 struct netdev_queue *txq; 246 int ret = NETDEV_TX_BUSY; 247 u16 queue_map; 248 249 if (unlikely(!netif_running(dev) || 250 !netif_carrier_ok(dev))) 251 goto drop; 252 253 features = netif_skb_features(skb); 254 if (skb_needs_linearize(skb, features) && 255 __skb_linearize(skb)) 256 goto drop; 257 258 queue_map = skb_get_queue_mapping(skb); 259 txq = netdev_get_tx_queue(dev, queue_map); 260 261 local_bh_disable(); 262 263 HARD_TX_LOCK(dev, txq, smp_processor_id()); 264 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 265 ret = ops->ndo_start_xmit(skb, dev); 266 if (ret == NETDEV_TX_OK) 267 txq_trans_update(txq); 268 } 269 HARD_TX_UNLOCK(dev, txq); 270 271 local_bh_enable(); 272 273 if (!dev_xmit_complete(ret)) 274 kfree_skb(skb); 275 276 return ret; 277 drop: 278 atomic_long_inc(&dev->tx_dropped); 279 kfree_skb(skb); 280 return NET_XMIT_DROP; 281 } 282 283 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 284 { 285 struct net_device *dev; 286 287 rcu_read_lock(); 288 dev = rcu_dereference(po->cached_dev); 289 if (likely(dev)) 290 dev_hold(dev); 291 rcu_read_unlock(); 292 293 return dev; 294 } 295 296 static void packet_cached_dev_assign(struct packet_sock *po, 297 struct net_device *dev) 298 { 299 rcu_assign_pointer(po->cached_dev, dev); 300 } 301 302 static void packet_cached_dev_reset(struct packet_sock *po) 303 { 304 RCU_INIT_POINTER(po->cached_dev, NULL); 305 } 306 307 static bool packet_use_direct_xmit(const struct packet_sock *po) 308 { 309 return po->xmit == packet_direct_xmit; 310 } 311 312 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 313 { 314 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues; 315 } 316 317 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 318 { 319 const struct net_device_ops *ops = dev->netdev_ops; 320 u16 queue_index; 321 322 if (ops->ndo_select_queue) { 323 queue_index = ops->ndo_select_queue(dev, skb, NULL, 324 __packet_pick_tx_queue); 325 queue_index = netdev_cap_txqueue(dev, queue_index); 326 } else { 327 queue_index = __packet_pick_tx_queue(dev, skb); 328 } 329 330 skb_set_queue_mapping(skb, queue_index); 331 } 332 333 /* register_prot_hook must be invoked with the po->bind_lock held, 334 * or from a context in which asynchronous accesses to the packet 335 * socket is not possible (packet_create()). 336 */ 337 static void register_prot_hook(struct sock *sk) 338 { 339 struct packet_sock *po = pkt_sk(sk); 340 341 if (!po->running) { 342 if (po->fanout) 343 __fanout_link(sk, po); 344 else 345 dev_add_pack(&po->prot_hook); 346 347 sock_hold(sk); 348 po->running = 1; 349 } 350 } 351 352 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock 353 * held. If the sync parameter is true, we will temporarily drop 354 * the po->bind_lock and do a synchronize_net to make sure no 355 * asynchronous packet processing paths still refer to the elements 356 * of po->prot_hook. If the sync parameter is false, it is the 357 * callers responsibility to take care of this. 358 */ 359 static void __unregister_prot_hook(struct sock *sk, bool sync) 360 { 361 struct packet_sock *po = pkt_sk(sk); 362 363 po->running = 0; 364 365 if (po->fanout) 366 __fanout_unlink(sk, po); 367 else 368 __dev_remove_pack(&po->prot_hook); 369 370 __sock_put(sk); 371 372 if (sync) { 373 spin_unlock(&po->bind_lock); 374 synchronize_net(); 375 spin_lock(&po->bind_lock); 376 } 377 } 378 379 static void unregister_prot_hook(struct sock *sk, bool sync) 380 { 381 struct packet_sock *po = pkt_sk(sk); 382 383 if (po->running) 384 __unregister_prot_hook(sk, sync); 385 } 386 387 static inline __pure struct page *pgv_to_page(void *addr) 388 { 389 if (is_vmalloc_addr(addr)) 390 return vmalloc_to_page(addr); 391 return virt_to_page(addr); 392 } 393 394 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 395 { 396 union tpacket_uhdr h; 397 398 h.raw = frame; 399 switch (po->tp_version) { 400 case TPACKET_V1: 401 h.h1->tp_status = status; 402 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 403 break; 404 case TPACKET_V2: 405 h.h2->tp_status = status; 406 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 407 break; 408 case TPACKET_V3: 409 default: 410 WARN(1, "TPACKET version not supported.\n"); 411 BUG(); 412 } 413 414 smp_wmb(); 415 } 416 417 static int __packet_get_status(struct packet_sock *po, void *frame) 418 { 419 union tpacket_uhdr h; 420 421 smp_rmb(); 422 423 h.raw = frame; 424 switch (po->tp_version) { 425 case TPACKET_V1: 426 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 427 return h.h1->tp_status; 428 case TPACKET_V2: 429 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 430 return h.h2->tp_status; 431 case TPACKET_V3: 432 default: 433 WARN(1, "TPACKET version not supported.\n"); 434 BUG(); 435 return 0; 436 } 437 } 438 439 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, 440 unsigned int flags) 441 { 442 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 443 444 if (shhwtstamps) { 445 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) && 446 ktime_to_timespec_cond(shhwtstamps->syststamp, ts)) 447 return TP_STATUS_TS_SYS_HARDWARE; 448 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 449 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) 450 return TP_STATUS_TS_RAW_HARDWARE; 451 } 452 453 if (ktime_to_timespec_cond(skb->tstamp, ts)) 454 return TP_STATUS_TS_SOFTWARE; 455 456 return 0; 457 } 458 459 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 460 struct sk_buff *skb) 461 { 462 union tpacket_uhdr h; 463 struct timespec ts; 464 __u32 ts_status; 465 466 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 467 return 0; 468 469 h.raw = frame; 470 switch (po->tp_version) { 471 case TPACKET_V1: 472 h.h1->tp_sec = ts.tv_sec; 473 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 474 break; 475 case TPACKET_V2: 476 h.h2->tp_sec = ts.tv_sec; 477 h.h2->tp_nsec = ts.tv_nsec; 478 break; 479 case TPACKET_V3: 480 default: 481 WARN(1, "TPACKET version not supported.\n"); 482 BUG(); 483 } 484 485 /* one flush is safe, as both fields always lie on the same cacheline */ 486 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 487 smp_wmb(); 488 489 return ts_status; 490 } 491 492 static void *packet_lookup_frame(struct packet_sock *po, 493 struct packet_ring_buffer *rb, 494 unsigned int position, 495 int status) 496 { 497 unsigned int pg_vec_pos, frame_offset; 498 union tpacket_uhdr h; 499 500 pg_vec_pos = position / rb->frames_per_block; 501 frame_offset = position % rb->frames_per_block; 502 503 h.raw = rb->pg_vec[pg_vec_pos].buffer + 504 (frame_offset * rb->frame_size); 505 506 if (status != __packet_get_status(po, h.raw)) 507 return NULL; 508 509 return h.raw; 510 } 511 512 static void *packet_current_frame(struct packet_sock *po, 513 struct packet_ring_buffer *rb, 514 int status) 515 { 516 return packet_lookup_frame(po, rb, rb->head, status); 517 } 518 519 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 520 { 521 del_timer_sync(&pkc->retire_blk_timer); 522 } 523 524 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 525 int tx_ring, 526 struct sk_buff_head *rb_queue) 527 { 528 struct tpacket_kbdq_core *pkc; 529 530 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) : 531 GET_PBDQC_FROM_RB(&po->rx_ring); 532 533 spin_lock_bh(&rb_queue->lock); 534 pkc->delete_blk_timer = 1; 535 spin_unlock_bh(&rb_queue->lock); 536 537 prb_del_retire_blk_timer(pkc); 538 } 539 540 static void prb_init_blk_timer(struct packet_sock *po, 541 struct tpacket_kbdq_core *pkc, 542 void (*func) (unsigned long)) 543 { 544 init_timer(&pkc->retire_blk_timer); 545 pkc->retire_blk_timer.data = (long)po; 546 pkc->retire_blk_timer.function = func; 547 pkc->retire_blk_timer.expires = jiffies; 548 } 549 550 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring) 551 { 552 struct tpacket_kbdq_core *pkc; 553 554 if (tx_ring) 555 BUG(); 556 557 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) : 558 GET_PBDQC_FROM_RB(&po->rx_ring); 559 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired); 560 } 561 562 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 563 int blk_size_in_bytes) 564 { 565 struct net_device *dev; 566 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; 567 struct ethtool_cmd ecmd; 568 int err; 569 u32 speed; 570 571 rtnl_lock(); 572 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 573 if (unlikely(!dev)) { 574 rtnl_unlock(); 575 return DEFAULT_PRB_RETIRE_TOV; 576 } 577 err = __ethtool_get_settings(dev, &ecmd); 578 speed = ethtool_cmd_speed(&ecmd); 579 rtnl_unlock(); 580 if (!err) { 581 /* 582 * If the link speed is so slow you don't really 583 * need to worry about perf anyways 584 */ 585 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) { 586 return DEFAULT_PRB_RETIRE_TOV; 587 } else { 588 msec = 1; 589 div = speed / 1000; 590 } 591 } 592 593 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 594 595 if (div) 596 mbits /= div; 597 598 tmo = mbits * msec; 599 600 if (div) 601 return tmo+1; 602 return tmo; 603 } 604 605 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 606 union tpacket_req_u *req_u) 607 { 608 p1->feature_req_word = req_u->req3.tp_feature_req_word; 609 } 610 611 static void init_prb_bdqc(struct packet_sock *po, 612 struct packet_ring_buffer *rb, 613 struct pgv *pg_vec, 614 union tpacket_req_u *req_u, int tx_ring) 615 { 616 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 617 struct tpacket_block_desc *pbd; 618 619 memset(p1, 0x0, sizeof(*p1)); 620 621 p1->knxt_seq_num = 1; 622 p1->pkbdq = pg_vec; 623 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 624 p1->pkblk_start = pg_vec[0].buffer; 625 p1->kblk_size = req_u->req3.tp_block_size; 626 p1->knum_blocks = req_u->req3.tp_block_nr; 627 p1->hdrlen = po->tp_hdrlen; 628 p1->version = po->tp_version; 629 p1->last_kactive_blk_num = 0; 630 po->stats.stats3.tp_freeze_q_cnt = 0; 631 if (req_u->req3.tp_retire_blk_tov) 632 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 633 else 634 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 635 req_u->req3.tp_block_size); 636 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 637 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 638 639 prb_init_ft_ops(p1, req_u); 640 prb_setup_retire_blk_timer(po, tx_ring); 641 prb_open_block(p1, pbd); 642 } 643 644 /* Do NOT update the last_blk_num first. 645 * Assumes sk_buff_head lock is held. 646 */ 647 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 648 { 649 mod_timer(&pkc->retire_blk_timer, 650 jiffies + pkc->tov_in_jiffies); 651 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 652 } 653 654 /* 655 * Timer logic: 656 * 1) We refresh the timer only when we open a block. 657 * By doing this we don't waste cycles refreshing the timer 658 * on packet-by-packet basis. 659 * 660 * With a 1MB block-size, on a 1Gbps line, it will take 661 * i) ~8 ms to fill a block + ii) memcpy etc. 662 * In this cut we are not accounting for the memcpy time. 663 * 664 * So, if the user sets the 'tmo' to 10ms then the timer 665 * will never fire while the block is still getting filled 666 * (which is what we want). However, the user could choose 667 * to close a block early and that's fine. 668 * 669 * But when the timer does fire, we check whether or not to refresh it. 670 * Since the tmo granularity is in msecs, it is not too expensive 671 * to refresh the timer, lets say every '8' msecs. 672 * Either the user can set the 'tmo' or we can derive it based on 673 * a) line-speed and b) block-size. 674 * prb_calc_retire_blk_tmo() calculates the tmo. 675 * 676 */ 677 static void prb_retire_rx_blk_timer_expired(unsigned long data) 678 { 679 struct packet_sock *po = (struct packet_sock *)data; 680 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 681 unsigned int frozen; 682 struct tpacket_block_desc *pbd; 683 684 spin_lock(&po->sk.sk_receive_queue.lock); 685 686 frozen = prb_queue_frozen(pkc); 687 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 688 689 if (unlikely(pkc->delete_blk_timer)) 690 goto out; 691 692 /* We only need to plug the race when the block is partially filled. 693 * tpacket_rcv: 694 * lock(); increment BLOCK_NUM_PKTS; unlock() 695 * copy_bits() is in progress ... 696 * timer fires on other cpu: 697 * we can't retire the current block because copy_bits 698 * is in progress. 699 * 700 */ 701 if (BLOCK_NUM_PKTS(pbd)) { 702 while (atomic_read(&pkc->blk_fill_in_prog)) { 703 /* Waiting for skb_copy_bits to finish... */ 704 cpu_relax(); 705 } 706 } 707 708 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 709 if (!frozen) { 710 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 711 if (!prb_dispatch_next_block(pkc, po)) 712 goto refresh_timer; 713 else 714 goto out; 715 } else { 716 /* Case 1. Queue was frozen because user-space was 717 * lagging behind. 718 */ 719 if (prb_curr_blk_in_use(pkc, pbd)) { 720 /* 721 * Ok, user-space is still behind. 722 * So just refresh the timer. 723 */ 724 goto refresh_timer; 725 } else { 726 /* Case 2. queue was frozen,user-space caught up, 727 * now the link went idle && the timer fired. 728 * We don't have a block to close.So we open this 729 * block and restart the timer. 730 * opening a block thaws the queue,restarts timer 731 * Thawing/timer-refresh is a side effect. 732 */ 733 prb_open_block(pkc, pbd); 734 goto out; 735 } 736 } 737 } 738 739 refresh_timer: 740 _prb_refresh_rx_retire_blk_timer(pkc); 741 742 out: 743 spin_unlock(&po->sk.sk_receive_queue.lock); 744 } 745 746 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 747 struct tpacket_block_desc *pbd1, __u32 status) 748 { 749 /* Flush everything minus the block header */ 750 751 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 752 u8 *start, *end; 753 754 start = (u8 *)pbd1; 755 756 /* Skip the block header(we know header WILL fit in 4K) */ 757 start += PAGE_SIZE; 758 759 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 760 for (; start < end; start += PAGE_SIZE) 761 flush_dcache_page(pgv_to_page(start)); 762 763 smp_wmb(); 764 #endif 765 766 /* Now update the block status. */ 767 768 BLOCK_STATUS(pbd1) = status; 769 770 /* Flush the block header */ 771 772 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 773 start = (u8 *)pbd1; 774 flush_dcache_page(pgv_to_page(start)); 775 776 smp_wmb(); 777 #endif 778 } 779 780 /* 781 * Side effect: 782 * 783 * 1) flush the block 784 * 2) Increment active_blk_num 785 * 786 * Note:We DONT refresh the timer on purpose. 787 * Because almost always the next block will be opened. 788 */ 789 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 790 struct tpacket_block_desc *pbd1, 791 struct packet_sock *po, unsigned int stat) 792 { 793 __u32 status = TP_STATUS_USER | stat; 794 795 struct tpacket3_hdr *last_pkt; 796 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 797 798 if (po->stats.stats3.tp_drops) 799 status |= TP_STATUS_LOSING; 800 801 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 802 last_pkt->tp_next_offset = 0; 803 804 /* Get the ts of the last pkt */ 805 if (BLOCK_NUM_PKTS(pbd1)) { 806 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 807 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 808 } else { 809 /* Ok, we tmo'd - so get the current time */ 810 struct timespec ts; 811 getnstimeofday(&ts); 812 h1->ts_last_pkt.ts_sec = ts.tv_sec; 813 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 814 } 815 816 smp_wmb(); 817 818 /* Flush the block */ 819 prb_flush_block(pkc1, pbd1, status); 820 821 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 822 } 823 824 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 825 { 826 pkc->reset_pending_on_curr_blk = 0; 827 } 828 829 /* 830 * Side effect of opening a block: 831 * 832 * 1) prb_queue is thawed. 833 * 2) retire_blk_timer is refreshed. 834 * 835 */ 836 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 837 struct tpacket_block_desc *pbd1) 838 { 839 struct timespec ts; 840 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 841 842 smp_rmb(); 843 844 /* We could have just memset this but we will lose the 845 * flexibility of making the priv area sticky 846 */ 847 848 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 849 BLOCK_NUM_PKTS(pbd1) = 0; 850 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 851 852 getnstimeofday(&ts); 853 854 h1->ts_first_pkt.ts_sec = ts.tv_sec; 855 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 856 857 pkc1->pkblk_start = (char *)pbd1; 858 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 859 860 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 861 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 862 863 pbd1->version = pkc1->version; 864 pkc1->prev = pkc1->nxt_offset; 865 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 866 867 prb_thaw_queue(pkc1); 868 _prb_refresh_rx_retire_blk_timer(pkc1); 869 870 smp_wmb(); 871 } 872 873 /* 874 * Queue freeze logic: 875 * 1) Assume tp_block_nr = 8 blocks. 876 * 2) At time 't0', user opens Rx ring. 877 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 878 * 4) user-space is either sleeping or processing block '0'. 879 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 880 * it will close block-7,loop around and try to fill block '0'. 881 * call-flow: 882 * __packet_lookup_frame_in_block 883 * prb_retire_current_block() 884 * prb_dispatch_next_block() 885 * |->(BLOCK_STATUS == USER) evaluates to true 886 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 887 * 6) Now there are two cases: 888 * 6.1) Link goes idle right after the queue is frozen. 889 * But remember, the last open_block() refreshed the timer. 890 * When this timer expires,it will refresh itself so that we can 891 * re-open block-0 in near future. 892 * 6.2) Link is busy and keeps on receiving packets. This is a simple 893 * case and __packet_lookup_frame_in_block will check if block-0 894 * is free and can now be re-used. 895 */ 896 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 897 struct packet_sock *po) 898 { 899 pkc->reset_pending_on_curr_blk = 1; 900 po->stats.stats3.tp_freeze_q_cnt++; 901 } 902 903 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 904 905 /* 906 * If the next block is free then we will dispatch it 907 * and return a good offset. 908 * Else, we will freeze the queue. 909 * So, caller must check the return value. 910 */ 911 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 912 struct packet_sock *po) 913 { 914 struct tpacket_block_desc *pbd; 915 916 smp_rmb(); 917 918 /* 1. Get current block num */ 919 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 920 921 /* 2. If this block is currently in_use then freeze the queue */ 922 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 923 prb_freeze_queue(pkc, po); 924 return NULL; 925 } 926 927 /* 928 * 3. 929 * open this block and return the offset where the first packet 930 * needs to get stored. 931 */ 932 prb_open_block(pkc, pbd); 933 return (void *)pkc->nxt_offset; 934 } 935 936 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 937 struct packet_sock *po, unsigned int status) 938 { 939 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 940 941 /* retire/close the current block */ 942 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 943 /* 944 * Plug the case where copy_bits() is in progress on 945 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 946 * have space to copy the pkt in the current block and 947 * called prb_retire_current_block() 948 * 949 * We don't need to worry about the TMO case because 950 * the timer-handler already handled this case. 951 */ 952 if (!(status & TP_STATUS_BLK_TMO)) { 953 while (atomic_read(&pkc->blk_fill_in_prog)) { 954 /* Waiting for skb_copy_bits to finish... */ 955 cpu_relax(); 956 } 957 } 958 prb_close_block(pkc, pbd, po, status); 959 return; 960 } 961 } 962 963 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc, 964 struct tpacket_block_desc *pbd) 965 { 966 return TP_STATUS_USER & BLOCK_STATUS(pbd); 967 } 968 969 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 970 { 971 return pkc->reset_pending_on_curr_blk; 972 } 973 974 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 975 { 976 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 977 atomic_dec(&pkc->blk_fill_in_prog); 978 } 979 980 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 981 struct tpacket3_hdr *ppd) 982 { 983 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 984 } 985 986 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 987 struct tpacket3_hdr *ppd) 988 { 989 ppd->hv1.tp_rxhash = 0; 990 } 991 992 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 993 struct tpacket3_hdr *ppd) 994 { 995 if (vlan_tx_tag_present(pkc->skb)) { 996 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb); 997 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 998 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 999 } else { 1000 ppd->hv1.tp_vlan_tci = 0; 1001 ppd->hv1.tp_vlan_tpid = 0; 1002 ppd->tp_status = TP_STATUS_AVAILABLE; 1003 } 1004 } 1005 1006 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 1007 struct tpacket3_hdr *ppd) 1008 { 1009 ppd->hv1.tp_padding = 0; 1010 prb_fill_vlan_info(pkc, ppd); 1011 1012 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 1013 prb_fill_rxhash(pkc, ppd); 1014 else 1015 prb_clear_rxhash(pkc, ppd); 1016 } 1017 1018 static void prb_fill_curr_block(char *curr, 1019 struct tpacket_kbdq_core *pkc, 1020 struct tpacket_block_desc *pbd, 1021 unsigned int len) 1022 { 1023 struct tpacket3_hdr *ppd; 1024 1025 ppd = (struct tpacket3_hdr *)curr; 1026 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 1027 pkc->prev = curr; 1028 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 1029 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1030 BLOCK_NUM_PKTS(pbd) += 1; 1031 atomic_inc(&pkc->blk_fill_in_prog); 1032 prb_run_all_ft_ops(pkc, ppd); 1033 } 1034 1035 /* Assumes caller has the sk->rx_queue.lock */ 1036 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1037 struct sk_buff *skb, 1038 int status, 1039 unsigned int len 1040 ) 1041 { 1042 struct tpacket_kbdq_core *pkc; 1043 struct tpacket_block_desc *pbd; 1044 char *curr, *end; 1045 1046 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1047 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1048 1049 /* Queue is frozen when user space is lagging behind */ 1050 if (prb_queue_frozen(pkc)) { 1051 /* 1052 * Check if that last block which caused the queue to freeze, 1053 * is still in_use by user-space. 1054 */ 1055 if (prb_curr_blk_in_use(pkc, pbd)) { 1056 /* Can't record this packet */ 1057 return NULL; 1058 } else { 1059 /* 1060 * Ok, the block was released by user-space. 1061 * Now let's open that block. 1062 * opening a block also thaws the queue. 1063 * Thawing is a side effect. 1064 */ 1065 prb_open_block(pkc, pbd); 1066 } 1067 } 1068 1069 smp_mb(); 1070 curr = pkc->nxt_offset; 1071 pkc->skb = skb; 1072 end = (char *)pbd + pkc->kblk_size; 1073 1074 /* first try the current block */ 1075 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1076 prb_fill_curr_block(curr, pkc, pbd, len); 1077 return (void *)curr; 1078 } 1079 1080 /* Ok, close the current block */ 1081 prb_retire_current_block(pkc, po, 0); 1082 1083 /* Now, try to dispatch the next block */ 1084 curr = (char *)prb_dispatch_next_block(pkc, po); 1085 if (curr) { 1086 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1087 prb_fill_curr_block(curr, pkc, pbd, len); 1088 return (void *)curr; 1089 } 1090 1091 /* 1092 * No free blocks are available.user_space hasn't caught up yet. 1093 * Queue was just frozen and now this packet will get dropped. 1094 */ 1095 return NULL; 1096 } 1097 1098 static void *packet_current_rx_frame(struct packet_sock *po, 1099 struct sk_buff *skb, 1100 int status, unsigned int len) 1101 { 1102 char *curr = NULL; 1103 switch (po->tp_version) { 1104 case TPACKET_V1: 1105 case TPACKET_V2: 1106 curr = packet_lookup_frame(po, &po->rx_ring, 1107 po->rx_ring.head, status); 1108 return curr; 1109 case TPACKET_V3: 1110 return __packet_lookup_frame_in_block(po, skb, status, len); 1111 default: 1112 WARN(1, "TPACKET version not supported\n"); 1113 BUG(); 1114 return NULL; 1115 } 1116 } 1117 1118 static void *prb_lookup_block(struct packet_sock *po, 1119 struct packet_ring_buffer *rb, 1120 unsigned int idx, 1121 int status) 1122 { 1123 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1124 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1125 1126 if (status != BLOCK_STATUS(pbd)) 1127 return NULL; 1128 return pbd; 1129 } 1130 1131 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1132 { 1133 unsigned int prev; 1134 if (rb->prb_bdqc.kactive_blk_num) 1135 prev = rb->prb_bdqc.kactive_blk_num-1; 1136 else 1137 prev = rb->prb_bdqc.knum_blocks-1; 1138 return prev; 1139 } 1140 1141 /* Assumes caller has held the rx_queue.lock */ 1142 static void *__prb_previous_block(struct packet_sock *po, 1143 struct packet_ring_buffer *rb, 1144 int status) 1145 { 1146 unsigned int previous = prb_previous_blk_num(rb); 1147 return prb_lookup_block(po, rb, previous, status); 1148 } 1149 1150 static void *packet_previous_rx_frame(struct packet_sock *po, 1151 struct packet_ring_buffer *rb, 1152 int status) 1153 { 1154 if (po->tp_version <= TPACKET_V2) 1155 return packet_previous_frame(po, rb, status); 1156 1157 return __prb_previous_block(po, rb, status); 1158 } 1159 1160 static void packet_increment_rx_head(struct packet_sock *po, 1161 struct packet_ring_buffer *rb) 1162 { 1163 switch (po->tp_version) { 1164 case TPACKET_V1: 1165 case TPACKET_V2: 1166 return packet_increment_head(rb); 1167 case TPACKET_V3: 1168 default: 1169 WARN(1, "TPACKET version not supported.\n"); 1170 BUG(); 1171 return; 1172 } 1173 } 1174 1175 static void *packet_previous_frame(struct packet_sock *po, 1176 struct packet_ring_buffer *rb, 1177 int status) 1178 { 1179 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1180 return packet_lookup_frame(po, rb, previous, status); 1181 } 1182 1183 static void packet_increment_head(struct packet_ring_buffer *buff) 1184 { 1185 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1186 } 1187 1188 static void packet_inc_pending(struct packet_ring_buffer *rb) 1189 { 1190 this_cpu_inc(*rb->pending_refcnt); 1191 } 1192 1193 static void packet_dec_pending(struct packet_ring_buffer *rb) 1194 { 1195 this_cpu_dec(*rb->pending_refcnt); 1196 } 1197 1198 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1199 { 1200 unsigned int refcnt = 0; 1201 int cpu; 1202 1203 /* We don't use pending refcount in rx_ring. */ 1204 if (rb->pending_refcnt == NULL) 1205 return 0; 1206 1207 for_each_possible_cpu(cpu) 1208 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1209 1210 return refcnt; 1211 } 1212 1213 static int packet_alloc_pending(struct packet_sock *po) 1214 { 1215 po->rx_ring.pending_refcnt = NULL; 1216 1217 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1218 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1219 return -ENOBUFS; 1220 1221 return 0; 1222 } 1223 1224 static void packet_free_pending(struct packet_sock *po) 1225 { 1226 free_percpu(po->tx_ring.pending_refcnt); 1227 } 1228 1229 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1230 { 1231 struct sock *sk = &po->sk; 1232 bool has_room; 1233 1234 if (po->prot_hook.func != tpacket_rcv) 1235 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize) 1236 <= sk->sk_rcvbuf; 1237 1238 spin_lock(&sk->sk_receive_queue.lock); 1239 if (po->tp_version == TPACKET_V3) 1240 has_room = prb_lookup_block(po, &po->rx_ring, 1241 po->rx_ring.prb_bdqc.kactive_blk_num, 1242 TP_STATUS_KERNEL); 1243 else 1244 has_room = packet_lookup_frame(po, &po->rx_ring, 1245 po->rx_ring.head, 1246 TP_STATUS_KERNEL); 1247 spin_unlock(&sk->sk_receive_queue.lock); 1248 1249 return has_room; 1250 } 1251 1252 static void packet_sock_destruct(struct sock *sk) 1253 { 1254 skb_queue_purge(&sk->sk_error_queue); 1255 1256 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1257 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 1258 1259 if (!sock_flag(sk, SOCK_DEAD)) { 1260 pr_err("Attempt to release alive packet socket: %p\n", sk); 1261 return; 1262 } 1263 1264 sk_refcnt_debug_dec(sk); 1265 } 1266 1267 static int fanout_rr_next(struct packet_fanout *f, unsigned int num) 1268 { 1269 int x = atomic_read(&f->rr_cur) + 1; 1270 1271 if (x >= num) 1272 x = 0; 1273 1274 return x; 1275 } 1276 1277 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1278 struct sk_buff *skb, 1279 unsigned int num) 1280 { 1281 return reciprocal_scale(skb_get_hash(skb), num); 1282 } 1283 1284 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1285 struct sk_buff *skb, 1286 unsigned int num) 1287 { 1288 int cur, old; 1289 1290 cur = atomic_read(&f->rr_cur); 1291 while ((old = atomic_cmpxchg(&f->rr_cur, cur, 1292 fanout_rr_next(f, num))) != cur) 1293 cur = old; 1294 return cur; 1295 } 1296 1297 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1298 struct sk_buff *skb, 1299 unsigned int num) 1300 { 1301 return smp_processor_id() % num; 1302 } 1303 1304 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1305 struct sk_buff *skb, 1306 unsigned int num) 1307 { 1308 return prandom_u32_max(num); 1309 } 1310 1311 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1312 struct sk_buff *skb, 1313 unsigned int idx, unsigned int skip, 1314 unsigned int num) 1315 { 1316 unsigned int i, j; 1317 1318 i = j = min_t(int, f->next[idx], num - 1); 1319 do { 1320 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) { 1321 if (i != j) 1322 f->next[idx] = i; 1323 return i; 1324 } 1325 if (++i == num) 1326 i = 0; 1327 } while (i != j); 1328 1329 return idx; 1330 } 1331 1332 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1333 struct sk_buff *skb, 1334 unsigned int num) 1335 { 1336 return skb_get_queue_mapping(skb) % num; 1337 } 1338 1339 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1340 { 1341 return f->flags & (flag >> 8); 1342 } 1343 1344 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1345 struct packet_type *pt, struct net_device *orig_dev) 1346 { 1347 struct packet_fanout *f = pt->af_packet_priv; 1348 unsigned int num = f->num_members; 1349 struct packet_sock *po; 1350 unsigned int idx; 1351 1352 if (!net_eq(dev_net(dev), read_pnet(&f->net)) || 1353 !num) { 1354 kfree_skb(skb); 1355 return 0; 1356 } 1357 1358 switch (f->type) { 1359 case PACKET_FANOUT_HASH: 1360 default: 1361 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1362 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET); 1363 if (!skb) 1364 return 0; 1365 } 1366 idx = fanout_demux_hash(f, skb, num); 1367 break; 1368 case PACKET_FANOUT_LB: 1369 idx = fanout_demux_lb(f, skb, num); 1370 break; 1371 case PACKET_FANOUT_CPU: 1372 idx = fanout_demux_cpu(f, skb, num); 1373 break; 1374 case PACKET_FANOUT_RND: 1375 idx = fanout_demux_rnd(f, skb, num); 1376 break; 1377 case PACKET_FANOUT_QM: 1378 idx = fanout_demux_qm(f, skb, num); 1379 break; 1380 case PACKET_FANOUT_ROLLOVER: 1381 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num); 1382 break; 1383 } 1384 1385 po = pkt_sk(f->arr[idx]); 1386 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) && 1387 unlikely(!packet_rcv_has_room(po, skb))) { 1388 idx = fanout_demux_rollover(f, skb, idx, idx, num); 1389 po = pkt_sk(f->arr[idx]); 1390 } 1391 1392 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1393 } 1394 1395 DEFINE_MUTEX(fanout_mutex); 1396 EXPORT_SYMBOL_GPL(fanout_mutex); 1397 static LIST_HEAD(fanout_list); 1398 1399 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1400 { 1401 struct packet_fanout *f = po->fanout; 1402 1403 spin_lock(&f->lock); 1404 f->arr[f->num_members] = sk; 1405 smp_wmb(); 1406 f->num_members++; 1407 spin_unlock(&f->lock); 1408 } 1409 1410 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1411 { 1412 struct packet_fanout *f = po->fanout; 1413 int i; 1414 1415 spin_lock(&f->lock); 1416 for (i = 0; i < f->num_members; i++) { 1417 if (f->arr[i] == sk) 1418 break; 1419 } 1420 BUG_ON(i >= f->num_members); 1421 f->arr[i] = f->arr[f->num_members - 1]; 1422 f->num_members--; 1423 spin_unlock(&f->lock); 1424 } 1425 1426 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1427 { 1428 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout) 1429 return true; 1430 1431 return false; 1432 } 1433 1434 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1435 { 1436 struct packet_sock *po = pkt_sk(sk); 1437 struct packet_fanout *f, *match; 1438 u8 type = type_flags & 0xff; 1439 u8 flags = type_flags >> 8; 1440 int err; 1441 1442 switch (type) { 1443 case PACKET_FANOUT_ROLLOVER: 1444 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1445 return -EINVAL; 1446 case PACKET_FANOUT_HASH: 1447 case PACKET_FANOUT_LB: 1448 case PACKET_FANOUT_CPU: 1449 case PACKET_FANOUT_RND: 1450 case PACKET_FANOUT_QM: 1451 break; 1452 default: 1453 return -EINVAL; 1454 } 1455 1456 if (!po->running) 1457 return -EINVAL; 1458 1459 if (po->fanout) 1460 return -EALREADY; 1461 1462 mutex_lock(&fanout_mutex); 1463 match = NULL; 1464 list_for_each_entry(f, &fanout_list, list) { 1465 if (f->id == id && 1466 read_pnet(&f->net) == sock_net(sk)) { 1467 match = f; 1468 break; 1469 } 1470 } 1471 err = -EINVAL; 1472 if (match && match->flags != flags) 1473 goto out; 1474 if (!match) { 1475 err = -ENOMEM; 1476 match = kzalloc(sizeof(*match), GFP_KERNEL); 1477 if (!match) 1478 goto out; 1479 write_pnet(&match->net, sock_net(sk)); 1480 match->id = id; 1481 match->type = type; 1482 match->flags = flags; 1483 atomic_set(&match->rr_cur, 0); 1484 INIT_LIST_HEAD(&match->list); 1485 spin_lock_init(&match->lock); 1486 atomic_set(&match->sk_ref, 0); 1487 match->prot_hook.type = po->prot_hook.type; 1488 match->prot_hook.dev = po->prot_hook.dev; 1489 match->prot_hook.func = packet_rcv_fanout; 1490 match->prot_hook.af_packet_priv = match; 1491 match->prot_hook.id_match = match_fanout_group; 1492 dev_add_pack(&match->prot_hook); 1493 list_add(&match->list, &fanout_list); 1494 } 1495 err = -EINVAL; 1496 if (match->type == type && 1497 match->prot_hook.type == po->prot_hook.type && 1498 match->prot_hook.dev == po->prot_hook.dev) { 1499 err = -ENOSPC; 1500 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1501 __dev_remove_pack(&po->prot_hook); 1502 po->fanout = match; 1503 atomic_inc(&match->sk_ref); 1504 __fanout_link(sk, po); 1505 err = 0; 1506 } 1507 } 1508 out: 1509 mutex_unlock(&fanout_mutex); 1510 return err; 1511 } 1512 1513 static void fanout_release(struct sock *sk) 1514 { 1515 struct packet_sock *po = pkt_sk(sk); 1516 struct packet_fanout *f; 1517 1518 f = po->fanout; 1519 if (!f) 1520 return; 1521 1522 mutex_lock(&fanout_mutex); 1523 po->fanout = NULL; 1524 1525 if (atomic_dec_and_test(&f->sk_ref)) { 1526 list_del(&f->list); 1527 dev_remove_pack(&f->prot_hook); 1528 kfree(f); 1529 } 1530 mutex_unlock(&fanout_mutex); 1531 } 1532 1533 static const struct proto_ops packet_ops; 1534 1535 static const struct proto_ops packet_ops_spkt; 1536 1537 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1538 struct packet_type *pt, struct net_device *orig_dev) 1539 { 1540 struct sock *sk; 1541 struct sockaddr_pkt *spkt; 1542 1543 /* 1544 * When we registered the protocol we saved the socket in the data 1545 * field for just this event. 1546 */ 1547 1548 sk = pt->af_packet_priv; 1549 1550 /* 1551 * Yank back the headers [hope the device set this 1552 * right or kerboom...] 1553 * 1554 * Incoming packets have ll header pulled, 1555 * push it back. 1556 * 1557 * For outgoing ones skb->data == skb_mac_header(skb) 1558 * so that this procedure is noop. 1559 */ 1560 1561 if (skb->pkt_type == PACKET_LOOPBACK) 1562 goto out; 1563 1564 if (!net_eq(dev_net(dev), sock_net(sk))) 1565 goto out; 1566 1567 skb = skb_share_check(skb, GFP_ATOMIC); 1568 if (skb == NULL) 1569 goto oom; 1570 1571 /* drop any routing info */ 1572 skb_dst_drop(skb); 1573 1574 /* drop conntrack reference */ 1575 nf_reset(skb); 1576 1577 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1578 1579 skb_push(skb, skb->data - skb_mac_header(skb)); 1580 1581 /* 1582 * The SOCK_PACKET socket receives _all_ frames. 1583 */ 1584 1585 spkt->spkt_family = dev->type; 1586 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1587 spkt->spkt_protocol = skb->protocol; 1588 1589 /* 1590 * Charge the memory to the socket. This is done specifically 1591 * to prevent sockets using all the memory up. 1592 */ 1593 1594 if (sock_queue_rcv_skb(sk, skb) == 0) 1595 return 0; 1596 1597 out: 1598 kfree_skb(skb); 1599 oom: 1600 return 0; 1601 } 1602 1603 1604 /* 1605 * Output a raw packet to a device layer. This bypasses all the other 1606 * protocol layers and you must therefore supply it with a complete frame 1607 */ 1608 1609 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, 1610 struct msghdr *msg, size_t len) 1611 { 1612 struct sock *sk = sock->sk; 1613 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1614 struct sk_buff *skb = NULL; 1615 struct net_device *dev; 1616 __be16 proto = 0; 1617 int err; 1618 int extra_len = 0; 1619 1620 /* 1621 * Get and verify the address. 1622 */ 1623 1624 if (saddr) { 1625 if (msg->msg_namelen < sizeof(struct sockaddr)) 1626 return -EINVAL; 1627 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1628 proto = saddr->spkt_protocol; 1629 } else 1630 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1631 1632 /* 1633 * Find the device first to size check it 1634 */ 1635 1636 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1637 retry: 1638 rcu_read_lock(); 1639 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1640 err = -ENODEV; 1641 if (dev == NULL) 1642 goto out_unlock; 1643 1644 err = -ENETDOWN; 1645 if (!(dev->flags & IFF_UP)) 1646 goto out_unlock; 1647 1648 /* 1649 * You may not queue a frame bigger than the mtu. This is the lowest level 1650 * raw protocol and you must do your own fragmentation at this level. 1651 */ 1652 1653 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1654 if (!netif_supports_nofcs(dev)) { 1655 err = -EPROTONOSUPPORT; 1656 goto out_unlock; 1657 } 1658 extra_len = 4; /* We're doing our own CRC */ 1659 } 1660 1661 err = -EMSGSIZE; 1662 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1663 goto out_unlock; 1664 1665 if (!skb) { 1666 size_t reserved = LL_RESERVED_SPACE(dev); 1667 int tlen = dev->needed_tailroom; 1668 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1669 1670 rcu_read_unlock(); 1671 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1672 if (skb == NULL) 1673 return -ENOBUFS; 1674 /* FIXME: Save some space for broken drivers that write a hard 1675 * header at transmission time by themselves. PPP is the notable 1676 * one here. This should really be fixed at the driver level. 1677 */ 1678 skb_reserve(skb, reserved); 1679 skb_reset_network_header(skb); 1680 1681 /* Try to align data part correctly */ 1682 if (hhlen) { 1683 skb->data -= hhlen; 1684 skb->tail -= hhlen; 1685 if (len < hhlen) 1686 skb_reset_network_header(skb); 1687 } 1688 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len); 1689 if (err) 1690 goto out_free; 1691 goto retry; 1692 } 1693 1694 if (len > (dev->mtu + dev->hard_header_len + extra_len)) { 1695 /* Earlier code assumed this would be a VLAN pkt, 1696 * double-check this now that we have the actual 1697 * packet in hand. 1698 */ 1699 struct ethhdr *ehdr; 1700 skb_reset_mac_header(skb); 1701 ehdr = eth_hdr(skb); 1702 if (ehdr->h_proto != htons(ETH_P_8021Q)) { 1703 err = -EMSGSIZE; 1704 goto out_unlock; 1705 } 1706 } 1707 1708 skb->protocol = proto; 1709 skb->dev = dev; 1710 skb->priority = sk->sk_priority; 1711 skb->mark = sk->sk_mark; 1712 1713 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags); 1714 1715 if (unlikely(extra_len == 4)) 1716 skb->no_fcs = 1; 1717 1718 skb_probe_transport_header(skb, 0); 1719 1720 dev_queue_xmit(skb); 1721 rcu_read_unlock(); 1722 return len; 1723 1724 out_unlock: 1725 rcu_read_unlock(); 1726 out_free: 1727 kfree_skb(skb); 1728 return err; 1729 } 1730 1731 static unsigned int run_filter(const struct sk_buff *skb, 1732 const struct sock *sk, 1733 unsigned int res) 1734 { 1735 struct sk_filter *filter; 1736 1737 rcu_read_lock(); 1738 filter = rcu_dereference(sk->sk_filter); 1739 if (filter != NULL) 1740 res = SK_RUN_FILTER(filter, skb); 1741 rcu_read_unlock(); 1742 1743 return res; 1744 } 1745 1746 /* 1747 * This function makes lazy skb cloning in hope that most of packets 1748 * are discarded by BPF. 1749 * 1750 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 1751 * and skb->cb are mangled. It works because (and until) packets 1752 * falling here are owned by current CPU. Output packets are cloned 1753 * by dev_queue_xmit_nit(), input packets are processed by net_bh 1754 * sequencially, so that if we return skb to original state on exit, 1755 * we will not harm anyone. 1756 */ 1757 1758 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 1759 struct packet_type *pt, struct net_device *orig_dev) 1760 { 1761 struct sock *sk; 1762 struct sockaddr_ll *sll; 1763 struct packet_sock *po; 1764 u8 *skb_head = skb->data; 1765 int skb_len = skb->len; 1766 unsigned int snaplen, res; 1767 1768 if (skb->pkt_type == PACKET_LOOPBACK) 1769 goto drop; 1770 1771 sk = pt->af_packet_priv; 1772 po = pkt_sk(sk); 1773 1774 if (!net_eq(dev_net(dev), sock_net(sk))) 1775 goto drop; 1776 1777 skb->dev = dev; 1778 1779 if (dev->header_ops) { 1780 /* The device has an explicit notion of ll header, 1781 * exported to higher levels. 1782 * 1783 * Otherwise, the device hides details of its frame 1784 * structure, so that corresponding packet head is 1785 * never delivered to user. 1786 */ 1787 if (sk->sk_type != SOCK_DGRAM) 1788 skb_push(skb, skb->data - skb_mac_header(skb)); 1789 else if (skb->pkt_type == PACKET_OUTGOING) { 1790 /* Special case: outgoing packets have ll header at head */ 1791 skb_pull(skb, skb_network_offset(skb)); 1792 } 1793 } 1794 1795 snaplen = skb->len; 1796 1797 res = run_filter(skb, sk, snaplen); 1798 if (!res) 1799 goto drop_n_restore; 1800 if (snaplen > res) 1801 snaplen = res; 1802 1803 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 1804 goto drop_n_acct; 1805 1806 if (skb_shared(skb)) { 1807 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 1808 if (nskb == NULL) 1809 goto drop_n_acct; 1810 1811 if (skb_head != skb->data) { 1812 skb->data = skb_head; 1813 skb->len = skb_len; 1814 } 1815 consume_skb(skb); 1816 skb = nskb; 1817 } 1818 1819 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 > 1820 sizeof(skb->cb)); 1821 1822 sll = &PACKET_SKB_CB(skb)->sa.ll; 1823 sll->sll_family = AF_PACKET; 1824 sll->sll_hatype = dev->type; 1825 sll->sll_protocol = skb->protocol; 1826 sll->sll_pkttype = skb->pkt_type; 1827 if (unlikely(po->origdev)) 1828 sll->sll_ifindex = orig_dev->ifindex; 1829 else 1830 sll->sll_ifindex = dev->ifindex; 1831 1832 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 1833 1834 PACKET_SKB_CB(skb)->origlen = skb->len; 1835 1836 if (pskb_trim(skb, snaplen)) 1837 goto drop_n_acct; 1838 1839 skb_set_owner_r(skb, sk); 1840 skb->dev = NULL; 1841 skb_dst_drop(skb); 1842 1843 /* drop conntrack reference */ 1844 nf_reset(skb); 1845 1846 spin_lock(&sk->sk_receive_queue.lock); 1847 po->stats.stats1.tp_packets++; 1848 skb->dropcount = atomic_read(&sk->sk_drops); 1849 __skb_queue_tail(&sk->sk_receive_queue, skb); 1850 spin_unlock(&sk->sk_receive_queue.lock); 1851 sk->sk_data_ready(sk); 1852 return 0; 1853 1854 drop_n_acct: 1855 spin_lock(&sk->sk_receive_queue.lock); 1856 po->stats.stats1.tp_drops++; 1857 atomic_inc(&sk->sk_drops); 1858 spin_unlock(&sk->sk_receive_queue.lock); 1859 1860 drop_n_restore: 1861 if (skb_head != skb->data && skb_shared(skb)) { 1862 skb->data = skb_head; 1863 skb->len = skb_len; 1864 } 1865 drop: 1866 consume_skb(skb); 1867 return 0; 1868 } 1869 1870 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 1871 struct packet_type *pt, struct net_device *orig_dev) 1872 { 1873 struct sock *sk; 1874 struct packet_sock *po; 1875 struct sockaddr_ll *sll; 1876 union tpacket_uhdr h; 1877 u8 *skb_head = skb->data; 1878 int skb_len = skb->len; 1879 unsigned int snaplen, res; 1880 unsigned long status = TP_STATUS_USER; 1881 unsigned short macoff, netoff, hdrlen; 1882 struct sk_buff *copy_skb = NULL; 1883 struct timespec ts; 1884 __u32 ts_status; 1885 1886 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 1887 * We may add members to them until current aligned size without forcing 1888 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 1889 */ 1890 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 1891 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 1892 1893 if (skb->pkt_type == PACKET_LOOPBACK) 1894 goto drop; 1895 1896 sk = pt->af_packet_priv; 1897 po = pkt_sk(sk); 1898 1899 if (!net_eq(dev_net(dev), sock_net(sk))) 1900 goto drop; 1901 1902 if (dev->header_ops) { 1903 if (sk->sk_type != SOCK_DGRAM) 1904 skb_push(skb, skb->data - skb_mac_header(skb)); 1905 else if (skb->pkt_type == PACKET_OUTGOING) { 1906 /* Special case: outgoing packets have ll header at head */ 1907 skb_pull(skb, skb_network_offset(skb)); 1908 } 1909 } 1910 1911 if (skb->ip_summed == CHECKSUM_PARTIAL) 1912 status |= TP_STATUS_CSUMNOTREADY; 1913 1914 snaplen = skb->len; 1915 1916 res = run_filter(skb, sk, snaplen); 1917 if (!res) 1918 goto drop_n_restore; 1919 if (snaplen > res) 1920 snaplen = res; 1921 1922 if (sk->sk_type == SOCK_DGRAM) { 1923 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 1924 po->tp_reserve; 1925 } else { 1926 unsigned int maclen = skb_network_offset(skb); 1927 netoff = TPACKET_ALIGN(po->tp_hdrlen + 1928 (maclen < 16 ? 16 : maclen)) + 1929 po->tp_reserve; 1930 macoff = netoff - maclen; 1931 } 1932 if (po->tp_version <= TPACKET_V2) { 1933 if (macoff + snaplen > po->rx_ring.frame_size) { 1934 if (po->copy_thresh && 1935 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1936 if (skb_shared(skb)) { 1937 copy_skb = skb_clone(skb, GFP_ATOMIC); 1938 } else { 1939 copy_skb = skb_get(skb); 1940 skb_head = skb->data; 1941 } 1942 if (copy_skb) 1943 skb_set_owner_r(copy_skb, sk); 1944 } 1945 snaplen = po->rx_ring.frame_size - macoff; 1946 if ((int)snaplen < 0) 1947 snaplen = 0; 1948 } 1949 } 1950 spin_lock(&sk->sk_receive_queue.lock); 1951 h.raw = packet_current_rx_frame(po, skb, 1952 TP_STATUS_KERNEL, (macoff+snaplen)); 1953 if (!h.raw) 1954 goto ring_is_full; 1955 if (po->tp_version <= TPACKET_V2) { 1956 packet_increment_rx_head(po, &po->rx_ring); 1957 /* 1958 * LOSING will be reported till you read the stats, 1959 * because it's COR - Clear On Read. 1960 * Anyways, moving it for V1/V2 only as V3 doesn't need this 1961 * at packet level. 1962 */ 1963 if (po->stats.stats1.tp_drops) 1964 status |= TP_STATUS_LOSING; 1965 } 1966 po->stats.stats1.tp_packets++; 1967 if (copy_skb) { 1968 status |= TP_STATUS_COPY; 1969 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 1970 } 1971 spin_unlock(&sk->sk_receive_queue.lock); 1972 1973 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 1974 1975 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 1976 getnstimeofday(&ts); 1977 1978 status |= ts_status; 1979 1980 switch (po->tp_version) { 1981 case TPACKET_V1: 1982 h.h1->tp_len = skb->len; 1983 h.h1->tp_snaplen = snaplen; 1984 h.h1->tp_mac = macoff; 1985 h.h1->tp_net = netoff; 1986 h.h1->tp_sec = ts.tv_sec; 1987 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 1988 hdrlen = sizeof(*h.h1); 1989 break; 1990 case TPACKET_V2: 1991 h.h2->tp_len = skb->len; 1992 h.h2->tp_snaplen = snaplen; 1993 h.h2->tp_mac = macoff; 1994 h.h2->tp_net = netoff; 1995 h.h2->tp_sec = ts.tv_sec; 1996 h.h2->tp_nsec = ts.tv_nsec; 1997 if (vlan_tx_tag_present(skb)) { 1998 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb); 1999 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2000 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2001 } else { 2002 h.h2->tp_vlan_tci = 0; 2003 h.h2->tp_vlan_tpid = 0; 2004 } 2005 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2006 hdrlen = sizeof(*h.h2); 2007 break; 2008 case TPACKET_V3: 2009 /* tp_nxt_offset,vlan are already populated above. 2010 * So DONT clear those fields here 2011 */ 2012 h.h3->tp_status |= status; 2013 h.h3->tp_len = skb->len; 2014 h.h3->tp_snaplen = snaplen; 2015 h.h3->tp_mac = macoff; 2016 h.h3->tp_net = netoff; 2017 h.h3->tp_sec = ts.tv_sec; 2018 h.h3->tp_nsec = ts.tv_nsec; 2019 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2020 hdrlen = sizeof(*h.h3); 2021 break; 2022 default: 2023 BUG(); 2024 } 2025 2026 sll = h.raw + TPACKET_ALIGN(hdrlen); 2027 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2028 sll->sll_family = AF_PACKET; 2029 sll->sll_hatype = dev->type; 2030 sll->sll_protocol = skb->protocol; 2031 sll->sll_pkttype = skb->pkt_type; 2032 if (unlikely(po->origdev)) 2033 sll->sll_ifindex = orig_dev->ifindex; 2034 else 2035 sll->sll_ifindex = dev->ifindex; 2036 2037 smp_mb(); 2038 2039 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2040 if (po->tp_version <= TPACKET_V2) { 2041 u8 *start, *end; 2042 2043 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2044 macoff + snaplen); 2045 2046 for (start = h.raw; start < end; start += PAGE_SIZE) 2047 flush_dcache_page(pgv_to_page(start)); 2048 } 2049 smp_wmb(); 2050 #endif 2051 2052 if (po->tp_version <= TPACKET_V2) 2053 __packet_set_status(po, h.raw, status); 2054 else 2055 prb_clear_blk_fill_status(&po->rx_ring); 2056 2057 sk->sk_data_ready(sk); 2058 2059 drop_n_restore: 2060 if (skb_head != skb->data && skb_shared(skb)) { 2061 skb->data = skb_head; 2062 skb->len = skb_len; 2063 } 2064 drop: 2065 kfree_skb(skb); 2066 return 0; 2067 2068 ring_is_full: 2069 po->stats.stats1.tp_drops++; 2070 spin_unlock(&sk->sk_receive_queue.lock); 2071 2072 sk->sk_data_ready(sk); 2073 kfree_skb(copy_skb); 2074 goto drop_n_restore; 2075 } 2076 2077 static void tpacket_destruct_skb(struct sk_buff *skb) 2078 { 2079 struct packet_sock *po = pkt_sk(skb->sk); 2080 2081 if (likely(po->tx_ring.pg_vec)) { 2082 void *ph; 2083 __u32 ts; 2084 2085 ph = skb_shinfo(skb)->destructor_arg; 2086 packet_dec_pending(&po->tx_ring); 2087 2088 ts = __packet_set_timestamp(po, ph, skb); 2089 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2090 } 2091 2092 sock_wfree(skb); 2093 } 2094 2095 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2096 void *frame, struct net_device *dev, int size_max, 2097 __be16 proto, unsigned char *addr, int hlen) 2098 { 2099 union tpacket_uhdr ph; 2100 int to_write, offset, len, tp_len, nr_frags, len_max; 2101 struct socket *sock = po->sk.sk_socket; 2102 struct page *page; 2103 void *data; 2104 int err; 2105 2106 ph.raw = frame; 2107 2108 skb->protocol = proto; 2109 skb->dev = dev; 2110 skb->priority = po->sk.sk_priority; 2111 skb->mark = po->sk.sk_mark; 2112 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags); 2113 skb_shinfo(skb)->destructor_arg = ph.raw; 2114 2115 switch (po->tp_version) { 2116 case TPACKET_V2: 2117 tp_len = ph.h2->tp_len; 2118 break; 2119 default: 2120 tp_len = ph.h1->tp_len; 2121 break; 2122 } 2123 if (unlikely(tp_len > size_max)) { 2124 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2125 return -EMSGSIZE; 2126 } 2127 2128 skb_reserve(skb, hlen); 2129 skb_reset_network_header(skb); 2130 2131 if (!packet_use_direct_xmit(po)) 2132 skb_probe_transport_header(skb, 0); 2133 if (unlikely(po->tp_tx_has_off)) { 2134 int off_min, off_max, off; 2135 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2136 off_max = po->tx_ring.frame_size - tp_len; 2137 if (sock->type == SOCK_DGRAM) { 2138 switch (po->tp_version) { 2139 case TPACKET_V2: 2140 off = ph.h2->tp_net; 2141 break; 2142 default: 2143 off = ph.h1->tp_net; 2144 break; 2145 } 2146 } else { 2147 switch (po->tp_version) { 2148 case TPACKET_V2: 2149 off = ph.h2->tp_mac; 2150 break; 2151 default: 2152 off = ph.h1->tp_mac; 2153 break; 2154 } 2155 } 2156 if (unlikely((off < off_min) || (off_max < off))) 2157 return -EINVAL; 2158 data = ph.raw + off; 2159 } else { 2160 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll); 2161 } 2162 to_write = tp_len; 2163 2164 if (sock->type == SOCK_DGRAM) { 2165 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2166 NULL, tp_len); 2167 if (unlikely(err < 0)) 2168 return -EINVAL; 2169 } else if (dev->hard_header_len) { 2170 /* net device doesn't like empty head */ 2171 if (unlikely(tp_len <= dev->hard_header_len)) { 2172 pr_err("packet size is too short (%d < %d)\n", 2173 tp_len, dev->hard_header_len); 2174 return -EINVAL; 2175 } 2176 2177 skb_push(skb, dev->hard_header_len); 2178 err = skb_store_bits(skb, 0, data, 2179 dev->hard_header_len); 2180 if (unlikely(err)) 2181 return err; 2182 2183 data += dev->hard_header_len; 2184 to_write -= dev->hard_header_len; 2185 } 2186 2187 offset = offset_in_page(data); 2188 len_max = PAGE_SIZE - offset; 2189 len = ((to_write > len_max) ? len_max : to_write); 2190 2191 skb->data_len = to_write; 2192 skb->len += to_write; 2193 skb->truesize += to_write; 2194 atomic_add(to_write, &po->sk.sk_wmem_alloc); 2195 2196 while (likely(to_write)) { 2197 nr_frags = skb_shinfo(skb)->nr_frags; 2198 2199 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2200 pr_err("Packet exceed the number of skb frags(%lu)\n", 2201 MAX_SKB_FRAGS); 2202 return -EFAULT; 2203 } 2204 2205 page = pgv_to_page(data); 2206 data += len; 2207 flush_dcache_page(page); 2208 get_page(page); 2209 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2210 to_write -= len; 2211 offset = 0; 2212 len_max = PAGE_SIZE; 2213 len = ((to_write > len_max) ? len_max : to_write); 2214 } 2215 2216 return tp_len; 2217 } 2218 2219 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2220 { 2221 struct sk_buff *skb; 2222 struct net_device *dev; 2223 __be16 proto; 2224 int err, reserve = 0; 2225 void *ph; 2226 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2227 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2228 int tp_len, size_max; 2229 unsigned char *addr; 2230 int len_sum = 0; 2231 int status = TP_STATUS_AVAILABLE; 2232 int hlen, tlen; 2233 2234 mutex_lock(&po->pg_vec_lock); 2235 2236 if (likely(saddr == NULL)) { 2237 dev = packet_cached_dev_get(po); 2238 proto = po->num; 2239 addr = NULL; 2240 } else { 2241 err = -EINVAL; 2242 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2243 goto out; 2244 if (msg->msg_namelen < (saddr->sll_halen 2245 + offsetof(struct sockaddr_ll, 2246 sll_addr))) 2247 goto out; 2248 proto = saddr->sll_protocol; 2249 addr = saddr->sll_addr; 2250 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2251 } 2252 2253 err = -ENXIO; 2254 if (unlikely(dev == NULL)) 2255 goto out; 2256 err = -ENETDOWN; 2257 if (unlikely(!(dev->flags & IFF_UP))) 2258 goto out_put; 2259 2260 reserve = dev->hard_header_len + VLAN_HLEN; 2261 size_max = po->tx_ring.frame_size 2262 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2263 2264 if (size_max > dev->mtu + reserve) 2265 size_max = dev->mtu + reserve; 2266 2267 do { 2268 ph = packet_current_frame(po, &po->tx_ring, 2269 TP_STATUS_SEND_REQUEST); 2270 if (unlikely(ph == NULL)) { 2271 if (need_wait && need_resched()) 2272 schedule(); 2273 continue; 2274 } 2275 2276 status = TP_STATUS_SEND_REQUEST; 2277 hlen = LL_RESERVED_SPACE(dev); 2278 tlen = dev->needed_tailroom; 2279 skb = sock_alloc_send_skb(&po->sk, 2280 hlen + tlen + sizeof(struct sockaddr_ll), 2281 0, &err); 2282 2283 if (unlikely(skb == NULL)) 2284 goto out_status; 2285 2286 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto, 2287 addr, hlen); 2288 if (tp_len > dev->mtu + dev->hard_header_len) { 2289 struct ethhdr *ehdr; 2290 /* Earlier code assumed this would be a VLAN pkt, 2291 * double-check this now that we have the actual 2292 * packet in hand. 2293 */ 2294 2295 skb_reset_mac_header(skb); 2296 ehdr = eth_hdr(skb); 2297 if (ehdr->h_proto != htons(ETH_P_8021Q)) 2298 tp_len = -EMSGSIZE; 2299 } 2300 if (unlikely(tp_len < 0)) { 2301 if (po->tp_loss) { 2302 __packet_set_status(po, ph, 2303 TP_STATUS_AVAILABLE); 2304 packet_increment_head(&po->tx_ring); 2305 kfree_skb(skb); 2306 continue; 2307 } else { 2308 status = TP_STATUS_WRONG_FORMAT; 2309 err = tp_len; 2310 goto out_status; 2311 } 2312 } 2313 2314 packet_pick_tx_queue(dev, skb); 2315 2316 skb->destructor = tpacket_destruct_skb; 2317 __packet_set_status(po, ph, TP_STATUS_SENDING); 2318 packet_inc_pending(&po->tx_ring); 2319 2320 status = TP_STATUS_SEND_REQUEST; 2321 err = po->xmit(skb); 2322 if (unlikely(err > 0)) { 2323 err = net_xmit_errno(err); 2324 if (err && __packet_get_status(po, ph) == 2325 TP_STATUS_AVAILABLE) { 2326 /* skb was destructed already */ 2327 skb = NULL; 2328 goto out_status; 2329 } 2330 /* 2331 * skb was dropped but not destructed yet; 2332 * let's treat it like congestion or err < 0 2333 */ 2334 err = 0; 2335 } 2336 packet_increment_head(&po->tx_ring); 2337 len_sum += tp_len; 2338 } while (likely((ph != NULL) || 2339 /* Note: packet_read_pending() might be slow if we have 2340 * to call it as it's per_cpu variable, but in fast-path 2341 * we already short-circuit the loop with the first 2342 * condition, and luckily don't have to go that path 2343 * anyway. 2344 */ 2345 (need_wait && packet_read_pending(&po->tx_ring)))); 2346 2347 err = len_sum; 2348 goto out_put; 2349 2350 out_status: 2351 __packet_set_status(po, ph, status); 2352 kfree_skb(skb); 2353 out_put: 2354 dev_put(dev); 2355 out: 2356 mutex_unlock(&po->pg_vec_lock); 2357 return err; 2358 } 2359 2360 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2361 size_t reserve, size_t len, 2362 size_t linear, int noblock, 2363 int *err) 2364 { 2365 struct sk_buff *skb; 2366 2367 /* Under a page? Don't bother with paged skb. */ 2368 if (prepad + len < PAGE_SIZE || !linear) 2369 linear = len; 2370 2371 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2372 err, 0); 2373 if (!skb) 2374 return NULL; 2375 2376 skb_reserve(skb, reserve); 2377 skb_put(skb, linear); 2378 skb->data_len = len - linear; 2379 skb->len += len - linear; 2380 2381 return skb; 2382 } 2383 2384 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2385 { 2386 struct sock *sk = sock->sk; 2387 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2388 struct sk_buff *skb; 2389 struct net_device *dev; 2390 __be16 proto; 2391 unsigned char *addr; 2392 int err, reserve = 0; 2393 struct virtio_net_hdr vnet_hdr = { 0 }; 2394 int offset = 0; 2395 int vnet_hdr_len; 2396 struct packet_sock *po = pkt_sk(sk); 2397 unsigned short gso_type = 0; 2398 int hlen, tlen; 2399 int extra_len = 0; 2400 2401 /* 2402 * Get and verify the address. 2403 */ 2404 2405 if (likely(saddr == NULL)) { 2406 dev = packet_cached_dev_get(po); 2407 proto = po->num; 2408 addr = NULL; 2409 } else { 2410 err = -EINVAL; 2411 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2412 goto out; 2413 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2414 goto out; 2415 proto = saddr->sll_protocol; 2416 addr = saddr->sll_addr; 2417 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2418 } 2419 2420 err = -ENXIO; 2421 if (unlikely(dev == NULL)) 2422 goto out_unlock; 2423 err = -ENETDOWN; 2424 if (unlikely(!(dev->flags & IFF_UP))) 2425 goto out_unlock; 2426 2427 if (sock->type == SOCK_RAW) 2428 reserve = dev->hard_header_len; 2429 if (po->has_vnet_hdr) { 2430 vnet_hdr_len = sizeof(vnet_hdr); 2431 2432 err = -EINVAL; 2433 if (len < vnet_hdr_len) 2434 goto out_unlock; 2435 2436 len -= vnet_hdr_len; 2437 2438 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov, 2439 vnet_hdr_len); 2440 if (err < 0) 2441 goto out_unlock; 2442 2443 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2444 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > 2445 vnet_hdr.hdr_len)) 2446 vnet_hdr.hdr_len = vnet_hdr.csum_start + 2447 vnet_hdr.csum_offset + 2; 2448 2449 err = -EINVAL; 2450 if (vnet_hdr.hdr_len > len) 2451 goto out_unlock; 2452 2453 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { 2454 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 2455 case VIRTIO_NET_HDR_GSO_TCPV4: 2456 gso_type = SKB_GSO_TCPV4; 2457 break; 2458 case VIRTIO_NET_HDR_GSO_TCPV6: 2459 gso_type = SKB_GSO_TCPV6; 2460 break; 2461 case VIRTIO_NET_HDR_GSO_UDP: 2462 gso_type = SKB_GSO_UDP; 2463 break; 2464 default: 2465 goto out_unlock; 2466 } 2467 2468 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) 2469 gso_type |= SKB_GSO_TCP_ECN; 2470 2471 if (vnet_hdr.gso_size == 0) 2472 goto out_unlock; 2473 2474 } 2475 } 2476 2477 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2478 if (!netif_supports_nofcs(dev)) { 2479 err = -EPROTONOSUPPORT; 2480 goto out_unlock; 2481 } 2482 extra_len = 4; /* We're doing our own CRC */ 2483 } 2484 2485 err = -EMSGSIZE; 2486 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2487 goto out_unlock; 2488 2489 err = -ENOBUFS; 2490 hlen = LL_RESERVED_SPACE(dev); 2491 tlen = dev->needed_tailroom; 2492 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len, 2493 msg->msg_flags & MSG_DONTWAIT, &err); 2494 if (skb == NULL) 2495 goto out_unlock; 2496 2497 skb_set_network_header(skb, reserve); 2498 2499 err = -EINVAL; 2500 if (sock->type == SOCK_DGRAM && 2501 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0) 2502 goto out_free; 2503 2504 /* Returns -EFAULT on error */ 2505 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len); 2506 if (err) 2507 goto out_free; 2508 2509 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags); 2510 2511 if (!gso_type && (len > dev->mtu + reserve + extra_len)) { 2512 /* Earlier code assumed this would be a VLAN pkt, 2513 * double-check this now that we have the actual 2514 * packet in hand. 2515 */ 2516 struct ethhdr *ehdr; 2517 skb_reset_mac_header(skb); 2518 ehdr = eth_hdr(skb); 2519 if (ehdr->h_proto != htons(ETH_P_8021Q)) { 2520 err = -EMSGSIZE; 2521 goto out_free; 2522 } 2523 } 2524 2525 skb->protocol = proto; 2526 skb->dev = dev; 2527 skb->priority = sk->sk_priority; 2528 skb->mark = sk->sk_mark; 2529 2530 packet_pick_tx_queue(dev, skb); 2531 2532 if (po->has_vnet_hdr) { 2533 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 2534 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start, 2535 vnet_hdr.csum_offset)) { 2536 err = -EINVAL; 2537 goto out_free; 2538 } 2539 } 2540 2541 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size; 2542 skb_shinfo(skb)->gso_type = gso_type; 2543 2544 /* Header must be checked, and gso_segs computed. */ 2545 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 2546 skb_shinfo(skb)->gso_segs = 0; 2547 2548 len += vnet_hdr_len; 2549 } 2550 2551 if (!packet_use_direct_xmit(po)) 2552 skb_probe_transport_header(skb, reserve); 2553 if (unlikely(extra_len == 4)) 2554 skb->no_fcs = 1; 2555 2556 err = po->xmit(skb); 2557 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2558 goto out_unlock; 2559 2560 dev_put(dev); 2561 2562 return len; 2563 2564 out_free: 2565 kfree_skb(skb); 2566 out_unlock: 2567 if (dev) 2568 dev_put(dev); 2569 out: 2570 return err; 2571 } 2572 2573 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock, 2574 struct msghdr *msg, size_t len) 2575 { 2576 struct sock *sk = sock->sk; 2577 struct packet_sock *po = pkt_sk(sk); 2578 2579 if (po->tx_ring.pg_vec) 2580 return tpacket_snd(po, msg); 2581 else 2582 return packet_snd(sock, msg, len); 2583 } 2584 2585 /* 2586 * Close a PACKET socket. This is fairly simple. We immediately go 2587 * to 'closed' state and remove our protocol entry in the device list. 2588 */ 2589 2590 static int packet_release(struct socket *sock) 2591 { 2592 struct sock *sk = sock->sk; 2593 struct packet_sock *po; 2594 struct net *net; 2595 union tpacket_req_u req_u; 2596 2597 if (!sk) 2598 return 0; 2599 2600 net = sock_net(sk); 2601 po = pkt_sk(sk); 2602 2603 mutex_lock(&net->packet.sklist_lock); 2604 sk_del_node_init_rcu(sk); 2605 mutex_unlock(&net->packet.sklist_lock); 2606 2607 preempt_disable(); 2608 sock_prot_inuse_add(net, sk->sk_prot, -1); 2609 preempt_enable(); 2610 2611 spin_lock(&po->bind_lock); 2612 unregister_prot_hook(sk, false); 2613 packet_cached_dev_reset(po); 2614 2615 if (po->prot_hook.dev) { 2616 dev_put(po->prot_hook.dev); 2617 po->prot_hook.dev = NULL; 2618 } 2619 spin_unlock(&po->bind_lock); 2620 2621 packet_flush_mclist(sk); 2622 2623 if (po->rx_ring.pg_vec) { 2624 memset(&req_u, 0, sizeof(req_u)); 2625 packet_set_ring(sk, &req_u, 1, 0); 2626 } 2627 2628 if (po->tx_ring.pg_vec) { 2629 memset(&req_u, 0, sizeof(req_u)); 2630 packet_set_ring(sk, &req_u, 1, 1); 2631 } 2632 2633 fanout_release(sk); 2634 2635 synchronize_net(); 2636 /* 2637 * Now the socket is dead. No more input will appear. 2638 */ 2639 sock_orphan(sk); 2640 sock->sk = NULL; 2641 2642 /* Purge queues */ 2643 2644 skb_queue_purge(&sk->sk_receive_queue); 2645 packet_free_pending(po); 2646 sk_refcnt_debug_release(sk); 2647 2648 sock_put(sk); 2649 return 0; 2650 } 2651 2652 /* 2653 * Attach a packet hook. 2654 */ 2655 2656 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto) 2657 { 2658 struct packet_sock *po = pkt_sk(sk); 2659 const struct net_device *dev_curr; 2660 __be16 proto_curr; 2661 bool need_rehook; 2662 2663 if (po->fanout) { 2664 if (dev) 2665 dev_put(dev); 2666 2667 return -EINVAL; 2668 } 2669 2670 lock_sock(sk); 2671 spin_lock(&po->bind_lock); 2672 2673 proto_curr = po->prot_hook.type; 2674 dev_curr = po->prot_hook.dev; 2675 2676 need_rehook = proto_curr != proto || dev_curr != dev; 2677 2678 if (need_rehook) { 2679 unregister_prot_hook(sk, true); 2680 2681 po->num = proto; 2682 po->prot_hook.type = proto; 2683 2684 if (po->prot_hook.dev) 2685 dev_put(po->prot_hook.dev); 2686 2687 po->prot_hook.dev = dev; 2688 2689 po->ifindex = dev ? dev->ifindex : 0; 2690 packet_cached_dev_assign(po, dev); 2691 } 2692 2693 if (proto == 0 || !need_rehook) 2694 goto out_unlock; 2695 2696 if (!dev || (dev->flags & IFF_UP)) { 2697 register_prot_hook(sk); 2698 } else { 2699 sk->sk_err = ENETDOWN; 2700 if (!sock_flag(sk, SOCK_DEAD)) 2701 sk->sk_error_report(sk); 2702 } 2703 2704 out_unlock: 2705 spin_unlock(&po->bind_lock); 2706 release_sock(sk); 2707 return 0; 2708 } 2709 2710 /* 2711 * Bind a packet socket to a device 2712 */ 2713 2714 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 2715 int addr_len) 2716 { 2717 struct sock *sk = sock->sk; 2718 char name[15]; 2719 struct net_device *dev; 2720 int err = -ENODEV; 2721 2722 /* 2723 * Check legality 2724 */ 2725 2726 if (addr_len != sizeof(struct sockaddr)) 2727 return -EINVAL; 2728 strlcpy(name, uaddr->sa_data, sizeof(name)); 2729 2730 dev = dev_get_by_name(sock_net(sk), name); 2731 if (dev) 2732 err = packet_do_bind(sk, dev, pkt_sk(sk)->num); 2733 return err; 2734 } 2735 2736 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 2737 { 2738 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 2739 struct sock *sk = sock->sk; 2740 struct net_device *dev = NULL; 2741 int err; 2742 2743 2744 /* 2745 * Check legality 2746 */ 2747 2748 if (addr_len < sizeof(struct sockaddr_ll)) 2749 return -EINVAL; 2750 if (sll->sll_family != AF_PACKET) 2751 return -EINVAL; 2752 2753 if (sll->sll_ifindex) { 2754 err = -ENODEV; 2755 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex); 2756 if (dev == NULL) 2757 goto out; 2758 } 2759 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num); 2760 2761 out: 2762 return err; 2763 } 2764 2765 static struct proto packet_proto = { 2766 .name = "PACKET", 2767 .owner = THIS_MODULE, 2768 .obj_size = sizeof(struct packet_sock), 2769 }; 2770 2771 /* 2772 * Create a packet of type SOCK_PACKET. 2773 */ 2774 2775 static int packet_create(struct net *net, struct socket *sock, int protocol, 2776 int kern) 2777 { 2778 struct sock *sk; 2779 struct packet_sock *po; 2780 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 2781 int err; 2782 2783 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 2784 return -EPERM; 2785 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 2786 sock->type != SOCK_PACKET) 2787 return -ESOCKTNOSUPPORT; 2788 2789 sock->state = SS_UNCONNECTED; 2790 2791 err = -ENOBUFS; 2792 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto); 2793 if (sk == NULL) 2794 goto out; 2795 2796 sock->ops = &packet_ops; 2797 if (sock->type == SOCK_PACKET) 2798 sock->ops = &packet_ops_spkt; 2799 2800 sock_init_data(sock, sk); 2801 2802 po = pkt_sk(sk); 2803 sk->sk_family = PF_PACKET; 2804 po->num = proto; 2805 po->xmit = dev_queue_xmit; 2806 2807 err = packet_alloc_pending(po); 2808 if (err) 2809 goto out2; 2810 2811 packet_cached_dev_reset(po); 2812 2813 sk->sk_destruct = packet_sock_destruct; 2814 sk_refcnt_debug_inc(sk); 2815 2816 /* 2817 * Attach a protocol block 2818 */ 2819 2820 spin_lock_init(&po->bind_lock); 2821 mutex_init(&po->pg_vec_lock); 2822 po->prot_hook.func = packet_rcv; 2823 2824 if (sock->type == SOCK_PACKET) 2825 po->prot_hook.func = packet_rcv_spkt; 2826 2827 po->prot_hook.af_packet_priv = sk; 2828 2829 if (proto) { 2830 po->prot_hook.type = proto; 2831 register_prot_hook(sk); 2832 } 2833 2834 mutex_lock(&net->packet.sklist_lock); 2835 sk_add_node_rcu(sk, &net->packet.sklist); 2836 mutex_unlock(&net->packet.sklist_lock); 2837 2838 preempt_disable(); 2839 sock_prot_inuse_add(net, &packet_proto, 1); 2840 preempt_enable(); 2841 2842 return 0; 2843 out2: 2844 sk_free(sk); 2845 out: 2846 return err; 2847 } 2848 2849 /* 2850 * Pull a packet from our receive queue and hand it to the user. 2851 * If necessary we block. 2852 */ 2853 2854 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock, 2855 struct msghdr *msg, size_t len, int flags) 2856 { 2857 struct sock *sk = sock->sk; 2858 struct sk_buff *skb; 2859 int copied, err; 2860 int vnet_hdr_len = 0; 2861 2862 err = -EINVAL; 2863 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 2864 goto out; 2865 2866 #if 0 2867 /* What error should we return now? EUNATTACH? */ 2868 if (pkt_sk(sk)->ifindex < 0) 2869 return -ENODEV; 2870 #endif 2871 2872 if (flags & MSG_ERRQUEUE) { 2873 err = sock_recv_errqueue(sk, msg, len, 2874 SOL_PACKET, PACKET_TX_TIMESTAMP); 2875 goto out; 2876 } 2877 2878 /* 2879 * Call the generic datagram receiver. This handles all sorts 2880 * of horrible races and re-entrancy so we can forget about it 2881 * in the protocol layers. 2882 * 2883 * Now it will return ENETDOWN, if device have just gone down, 2884 * but then it will block. 2885 */ 2886 2887 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 2888 2889 /* 2890 * An error occurred so return it. Because skb_recv_datagram() 2891 * handles the blocking we don't see and worry about blocking 2892 * retries. 2893 */ 2894 2895 if (skb == NULL) 2896 goto out; 2897 2898 if (pkt_sk(sk)->has_vnet_hdr) { 2899 struct virtio_net_hdr vnet_hdr = { 0 }; 2900 2901 err = -EINVAL; 2902 vnet_hdr_len = sizeof(vnet_hdr); 2903 if (len < vnet_hdr_len) 2904 goto out_free; 2905 2906 len -= vnet_hdr_len; 2907 2908 if (skb_is_gso(skb)) { 2909 struct skb_shared_info *sinfo = skb_shinfo(skb); 2910 2911 /* This is a hint as to how much should be linear. */ 2912 vnet_hdr.hdr_len = skb_headlen(skb); 2913 vnet_hdr.gso_size = sinfo->gso_size; 2914 if (sinfo->gso_type & SKB_GSO_TCPV4) 2915 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 2916 else if (sinfo->gso_type & SKB_GSO_TCPV6) 2917 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 2918 else if (sinfo->gso_type & SKB_GSO_UDP) 2919 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP; 2920 else if (sinfo->gso_type & SKB_GSO_FCOE) 2921 goto out_free; 2922 else 2923 BUG(); 2924 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 2925 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; 2926 } else 2927 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; 2928 2929 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2930 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 2931 vnet_hdr.csum_start = skb_checksum_start_offset(skb); 2932 vnet_hdr.csum_offset = skb->csum_offset; 2933 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2934 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID; 2935 } /* else everything is zero */ 2936 2937 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr, 2938 vnet_hdr_len); 2939 if (err < 0) 2940 goto out_free; 2941 } 2942 2943 /* You lose any data beyond the buffer you gave. If it worries 2944 * a user program they can ask the device for its MTU 2945 * anyway. 2946 */ 2947 copied = skb->len; 2948 if (copied > len) { 2949 copied = len; 2950 msg->msg_flags |= MSG_TRUNC; 2951 } 2952 2953 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 2954 if (err) 2955 goto out_free; 2956 2957 sock_recv_ts_and_drops(msg, sk, skb); 2958 2959 if (msg->msg_name) { 2960 /* If the address length field is there to be filled 2961 * in, we fill it in now. 2962 */ 2963 if (sock->type == SOCK_PACKET) { 2964 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 2965 msg->msg_namelen = sizeof(struct sockaddr_pkt); 2966 } else { 2967 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 2968 msg->msg_namelen = sll->sll_halen + 2969 offsetof(struct sockaddr_ll, sll_addr); 2970 } 2971 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, 2972 msg->msg_namelen); 2973 } 2974 2975 if (pkt_sk(sk)->auxdata) { 2976 struct tpacket_auxdata aux; 2977 2978 aux.tp_status = TP_STATUS_USER; 2979 if (skb->ip_summed == CHECKSUM_PARTIAL) 2980 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 2981 aux.tp_len = PACKET_SKB_CB(skb)->origlen; 2982 aux.tp_snaplen = skb->len; 2983 aux.tp_mac = 0; 2984 aux.tp_net = skb_network_offset(skb); 2985 if (vlan_tx_tag_present(skb)) { 2986 aux.tp_vlan_tci = vlan_tx_tag_get(skb); 2987 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 2988 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2989 } else { 2990 aux.tp_vlan_tci = 0; 2991 aux.tp_vlan_tpid = 0; 2992 } 2993 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 2994 } 2995 2996 /* 2997 * Free or return the buffer as appropriate. Again this 2998 * hides all the races and re-entrancy issues from us. 2999 */ 3000 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3001 3002 out_free: 3003 skb_free_datagram(sk, skb); 3004 out: 3005 return err; 3006 } 3007 3008 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3009 int *uaddr_len, int peer) 3010 { 3011 struct net_device *dev; 3012 struct sock *sk = sock->sk; 3013 3014 if (peer) 3015 return -EOPNOTSUPP; 3016 3017 uaddr->sa_family = AF_PACKET; 3018 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3019 rcu_read_lock(); 3020 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3021 if (dev) 3022 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3023 rcu_read_unlock(); 3024 *uaddr_len = sizeof(*uaddr); 3025 3026 return 0; 3027 } 3028 3029 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3030 int *uaddr_len, int peer) 3031 { 3032 struct net_device *dev; 3033 struct sock *sk = sock->sk; 3034 struct packet_sock *po = pkt_sk(sk); 3035 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3036 3037 if (peer) 3038 return -EOPNOTSUPP; 3039 3040 sll->sll_family = AF_PACKET; 3041 sll->sll_ifindex = po->ifindex; 3042 sll->sll_protocol = po->num; 3043 sll->sll_pkttype = 0; 3044 rcu_read_lock(); 3045 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 3046 if (dev) { 3047 sll->sll_hatype = dev->type; 3048 sll->sll_halen = dev->addr_len; 3049 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3050 } else { 3051 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3052 sll->sll_halen = 0; 3053 } 3054 rcu_read_unlock(); 3055 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3056 3057 return 0; 3058 } 3059 3060 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3061 int what) 3062 { 3063 switch (i->type) { 3064 case PACKET_MR_MULTICAST: 3065 if (i->alen != dev->addr_len) 3066 return -EINVAL; 3067 if (what > 0) 3068 return dev_mc_add(dev, i->addr); 3069 else 3070 return dev_mc_del(dev, i->addr); 3071 break; 3072 case PACKET_MR_PROMISC: 3073 return dev_set_promiscuity(dev, what); 3074 break; 3075 case PACKET_MR_ALLMULTI: 3076 return dev_set_allmulti(dev, what); 3077 break; 3078 case PACKET_MR_UNICAST: 3079 if (i->alen != dev->addr_len) 3080 return -EINVAL; 3081 if (what > 0) 3082 return dev_uc_add(dev, i->addr); 3083 else 3084 return dev_uc_del(dev, i->addr); 3085 break; 3086 default: 3087 break; 3088 } 3089 return 0; 3090 } 3091 3092 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what) 3093 { 3094 for ( ; i; i = i->next) { 3095 if (i->ifindex == dev->ifindex) 3096 packet_dev_mc(dev, i, what); 3097 } 3098 } 3099 3100 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3101 { 3102 struct packet_sock *po = pkt_sk(sk); 3103 struct packet_mclist *ml, *i; 3104 struct net_device *dev; 3105 int err; 3106 3107 rtnl_lock(); 3108 3109 err = -ENODEV; 3110 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3111 if (!dev) 3112 goto done; 3113 3114 err = -EINVAL; 3115 if (mreq->mr_alen > dev->addr_len) 3116 goto done; 3117 3118 err = -ENOBUFS; 3119 i = kmalloc(sizeof(*i), GFP_KERNEL); 3120 if (i == NULL) 3121 goto done; 3122 3123 err = 0; 3124 for (ml = po->mclist; ml; ml = ml->next) { 3125 if (ml->ifindex == mreq->mr_ifindex && 3126 ml->type == mreq->mr_type && 3127 ml->alen == mreq->mr_alen && 3128 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3129 ml->count++; 3130 /* Free the new element ... */ 3131 kfree(i); 3132 goto done; 3133 } 3134 } 3135 3136 i->type = mreq->mr_type; 3137 i->ifindex = mreq->mr_ifindex; 3138 i->alen = mreq->mr_alen; 3139 memcpy(i->addr, mreq->mr_address, i->alen); 3140 i->count = 1; 3141 i->next = po->mclist; 3142 po->mclist = i; 3143 err = packet_dev_mc(dev, i, 1); 3144 if (err) { 3145 po->mclist = i->next; 3146 kfree(i); 3147 } 3148 3149 done: 3150 rtnl_unlock(); 3151 return err; 3152 } 3153 3154 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3155 { 3156 struct packet_mclist *ml, **mlp; 3157 3158 rtnl_lock(); 3159 3160 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3161 if (ml->ifindex == mreq->mr_ifindex && 3162 ml->type == mreq->mr_type && 3163 ml->alen == mreq->mr_alen && 3164 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3165 if (--ml->count == 0) { 3166 struct net_device *dev; 3167 *mlp = ml->next; 3168 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3169 if (dev) 3170 packet_dev_mc(dev, ml, -1); 3171 kfree(ml); 3172 } 3173 rtnl_unlock(); 3174 return 0; 3175 } 3176 } 3177 rtnl_unlock(); 3178 return -EADDRNOTAVAIL; 3179 } 3180 3181 static void packet_flush_mclist(struct sock *sk) 3182 { 3183 struct packet_sock *po = pkt_sk(sk); 3184 struct packet_mclist *ml; 3185 3186 if (!po->mclist) 3187 return; 3188 3189 rtnl_lock(); 3190 while ((ml = po->mclist) != NULL) { 3191 struct net_device *dev; 3192 3193 po->mclist = ml->next; 3194 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3195 if (dev != NULL) 3196 packet_dev_mc(dev, ml, -1); 3197 kfree(ml); 3198 } 3199 rtnl_unlock(); 3200 } 3201 3202 static int 3203 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3204 { 3205 struct sock *sk = sock->sk; 3206 struct packet_sock *po = pkt_sk(sk); 3207 int ret; 3208 3209 if (level != SOL_PACKET) 3210 return -ENOPROTOOPT; 3211 3212 switch (optname) { 3213 case PACKET_ADD_MEMBERSHIP: 3214 case PACKET_DROP_MEMBERSHIP: 3215 { 3216 struct packet_mreq_max mreq; 3217 int len = optlen; 3218 memset(&mreq, 0, sizeof(mreq)); 3219 if (len < sizeof(struct packet_mreq)) 3220 return -EINVAL; 3221 if (len > sizeof(mreq)) 3222 len = sizeof(mreq); 3223 if (copy_from_user(&mreq, optval, len)) 3224 return -EFAULT; 3225 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3226 return -EINVAL; 3227 if (optname == PACKET_ADD_MEMBERSHIP) 3228 ret = packet_mc_add(sk, &mreq); 3229 else 3230 ret = packet_mc_drop(sk, &mreq); 3231 return ret; 3232 } 3233 3234 case PACKET_RX_RING: 3235 case PACKET_TX_RING: 3236 { 3237 union tpacket_req_u req_u; 3238 int len; 3239 3240 switch (po->tp_version) { 3241 case TPACKET_V1: 3242 case TPACKET_V2: 3243 len = sizeof(req_u.req); 3244 break; 3245 case TPACKET_V3: 3246 default: 3247 len = sizeof(req_u.req3); 3248 break; 3249 } 3250 if (optlen < len) 3251 return -EINVAL; 3252 if (pkt_sk(sk)->has_vnet_hdr) 3253 return -EINVAL; 3254 if (copy_from_user(&req_u.req, optval, len)) 3255 return -EFAULT; 3256 return packet_set_ring(sk, &req_u, 0, 3257 optname == PACKET_TX_RING); 3258 } 3259 case PACKET_COPY_THRESH: 3260 { 3261 int val; 3262 3263 if (optlen != sizeof(val)) 3264 return -EINVAL; 3265 if (copy_from_user(&val, optval, sizeof(val))) 3266 return -EFAULT; 3267 3268 pkt_sk(sk)->copy_thresh = val; 3269 return 0; 3270 } 3271 case PACKET_VERSION: 3272 { 3273 int val; 3274 3275 if (optlen != sizeof(val)) 3276 return -EINVAL; 3277 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3278 return -EBUSY; 3279 if (copy_from_user(&val, optval, sizeof(val))) 3280 return -EFAULT; 3281 switch (val) { 3282 case TPACKET_V1: 3283 case TPACKET_V2: 3284 case TPACKET_V3: 3285 po->tp_version = val; 3286 return 0; 3287 default: 3288 return -EINVAL; 3289 } 3290 } 3291 case PACKET_RESERVE: 3292 { 3293 unsigned int val; 3294 3295 if (optlen != sizeof(val)) 3296 return -EINVAL; 3297 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3298 return -EBUSY; 3299 if (copy_from_user(&val, optval, sizeof(val))) 3300 return -EFAULT; 3301 po->tp_reserve = val; 3302 return 0; 3303 } 3304 case PACKET_LOSS: 3305 { 3306 unsigned int val; 3307 3308 if (optlen != sizeof(val)) 3309 return -EINVAL; 3310 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3311 return -EBUSY; 3312 if (copy_from_user(&val, optval, sizeof(val))) 3313 return -EFAULT; 3314 po->tp_loss = !!val; 3315 return 0; 3316 } 3317 case PACKET_AUXDATA: 3318 { 3319 int val; 3320 3321 if (optlen < sizeof(val)) 3322 return -EINVAL; 3323 if (copy_from_user(&val, optval, sizeof(val))) 3324 return -EFAULT; 3325 3326 po->auxdata = !!val; 3327 return 0; 3328 } 3329 case PACKET_ORIGDEV: 3330 { 3331 int val; 3332 3333 if (optlen < sizeof(val)) 3334 return -EINVAL; 3335 if (copy_from_user(&val, optval, sizeof(val))) 3336 return -EFAULT; 3337 3338 po->origdev = !!val; 3339 return 0; 3340 } 3341 case PACKET_VNET_HDR: 3342 { 3343 int val; 3344 3345 if (sock->type != SOCK_RAW) 3346 return -EINVAL; 3347 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3348 return -EBUSY; 3349 if (optlen < sizeof(val)) 3350 return -EINVAL; 3351 if (copy_from_user(&val, optval, sizeof(val))) 3352 return -EFAULT; 3353 3354 po->has_vnet_hdr = !!val; 3355 return 0; 3356 } 3357 case PACKET_TIMESTAMP: 3358 { 3359 int val; 3360 3361 if (optlen != sizeof(val)) 3362 return -EINVAL; 3363 if (copy_from_user(&val, optval, sizeof(val))) 3364 return -EFAULT; 3365 3366 po->tp_tstamp = val; 3367 return 0; 3368 } 3369 case PACKET_FANOUT: 3370 { 3371 int val; 3372 3373 if (optlen != sizeof(val)) 3374 return -EINVAL; 3375 if (copy_from_user(&val, optval, sizeof(val))) 3376 return -EFAULT; 3377 3378 return fanout_add(sk, val & 0xffff, val >> 16); 3379 } 3380 case PACKET_TX_HAS_OFF: 3381 { 3382 unsigned int val; 3383 3384 if (optlen != sizeof(val)) 3385 return -EINVAL; 3386 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3387 return -EBUSY; 3388 if (copy_from_user(&val, optval, sizeof(val))) 3389 return -EFAULT; 3390 po->tp_tx_has_off = !!val; 3391 return 0; 3392 } 3393 case PACKET_QDISC_BYPASS: 3394 { 3395 int val; 3396 3397 if (optlen != sizeof(val)) 3398 return -EINVAL; 3399 if (copy_from_user(&val, optval, sizeof(val))) 3400 return -EFAULT; 3401 3402 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; 3403 return 0; 3404 } 3405 default: 3406 return -ENOPROTOOPT; 3407 } 3408 } 3409 3410 static int packet_getsockopt(struct socket *sock, int level, int optname, 3411 char __user *optval, int __user *optlen) 3412 { 3413 int len; 3414 int val, lv = sizeof(val); 3415 struct sock *sk = sock->sk; 3416 struct packet_sock *po = pkt_sk(sk); 3417 void *data = &val; 3418 union tpacket_stats_u st; 3419 3420 if (level != SOL_PACKET) 3421 return -ENOPROTOOPT; 3422 3423 if (get_user(len, optlen)) 3424 return -EFAULT; 3425 3426 if (len < 0) 3427 return -EINVAL; 3428 3429 switch (optname) { 3430 case PACKET_STATISTICS: 3431 spin_lock_bh(&sk->sk_receive_queue.lock); 3432 memcpy(&st, &po->stats, sizeof(st)); 3433 memset(&po->stats, 0, sizeof(po->stats)); 3434 spin_unlock_bh(&sk->sk_receive_queue.lock); 3435 3436 if (po->tp_version == TPACKET_V3) { 3437 lv = sizeof(struct tpacket_stats_v3); 3438 st.stats3.tp_packets += st.stats3.tp_drops; 3439 data = &st.stats3; 3440 } else { 3441 lv = sizeof(struct tpacket_stats); 3442 st.stats1.tp_packets += st.stats1.tp_drops; 3443 data = &st.stats1; 3444 } 3445 3446 break; 3447 case PACKET_AUXDATA: 3448 val = po->auxdata; 3449 break; 3450 case PACKET_ORIGDEV: 3451 val = po->origdev; 3452 break; 3453 case PACKET_VNET_HDR: 3454 val = po->has_vnet_hdr; 3455 break; 3456 case PACKET_VERSION: 3457 val = po->tp_version; 3458 break; 3459 case PACKET_HDRLEN: 3460 if (len > sizeof(int)) 3461 len = sizeof(int); 3462 if (copy_from_user(&val, optval, len)) 3463 return -EFAULT; 3464 switch (val) { 3465 case TPACKET_V1: 3466 val = sizeof(struct tpacket_hdr); 3467 break; 3468 case TPACKET_V2: 3469 val = sizeof(struct tpacket2_hdr); 3470 break; 3471 case TPACKET_V3: 3472 val = sizeof(struct tpacket3_hdr); 3473 break; 3474 default: 3475 return -EINVAL; 3476 } 3477 break; 3478 case PACKET_RESERVE: 3479 val = po->tp_reserve; 3480 break; 3481 case PACKET_LOSS: 3482 val = po->tp_loss; 3483 break; 3484 case PACKET_TIMESTAMP: 3485 val = po->tp_tstamp; 3486 break; 3487 case PACKET_FANOUT: 3488 val = (po->fanout ? 3489 ((u32)po->fanout->id | 3490 ((u32)po->fanout->type << 16) | 3491 ((u32)po->fanout->flags << 24)) : 3492 0); 3493 break; 3494 case PACKET_TX_HAS_OFF: 3495 val = po->tp_tx_has_off; 3496 break; 3497 case PACKET_QDISC_BYPASS: 3498 val = packet_use_direct_xmit(po); 3499 break; 3500 default: 3501 return -ENOPROTOOPT; 3502 } 3503 3504 if (len > lv) 3505 len = lv; 3506 if (put_user(len, optlen)) 3507 return -EFAULT; 3508 if (copy_to_user(optval, data, len)) 3509 return -EFAULT; 3510 return 0; 3511 } 3512 3513 3514 static int packet_notifier(struct notifier_block *this, 3515 unsigned long msg, void *ptr) 3516 { 3517 struct sock *sk; 3518 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 3519 struct net *net = dev_net(dev); 3520 3521 rcu_read_lock(); 3522 sk_for_each_rcu(sk, &net->packet.sklist) { 3523 struct packet_sock *po = pkt_sk(sk); 3524 3525 switch (msg) { 3526 case NETDEV_UNREGISTER: 3527 if (po->mclist) 3528 packet_dev_mclist(dev, po->mclist, -1); 3529 /* fallthrough */ 3530 3531 case NETDEV_DOWN: 3532 if (dev->ifindex == po->ifindex) { 3533 spin_lock(&po->bind_lock); 3534 if (po->running) { 3535 __unregister_prot_hook(sk, false); 3536 sk->sk_err = ENETDOWN; 3537 if (!sock_flag(sk, SOCK_DEAD)) 3538 sk->sk_error_report(sk); 3539 } 3540 if (msg == NETDEV_UNREGISTER) { 3541 packet_cached_dev_reset(po); 3542 po->ifindex = -1; 3543 if (po->prot_hook.dev) 3544 dev_put(po->prot_hook.dev); 3545 po->prot_hook.dev = NULL; 3546 } 3547 spin_unlock(&po->bind_lock); 3548 } 3549 break; 3550 case NETDEV_UP: 3551 if (dev->ifindex == po->ifindex) { 3552 spin_lock(&po->bind_lock); 3553 if (po->num) 3554 register_prot_hook(sk); 3555 spin_unlock(&po->bind_lock); 3556 } 3557 break; 3558 } 3559 } 3560 rcu_read_unlock(); 3561 return NOTIFY_DONE; 3562 } 3563 3564 3565 static int packet_ioctl(struct socket *sock, unsigned int cmd, 3566 unsigned long arg) 3567 { 3568 struct sock *sk = sock->sk; 3569 3570 switch (cmd) { 3571 case SIOCOUTQ: 3572 { 3573 int amount = sk_wmem_alloc_get(sk); 3574 3575 return put_user(amount, (int __user *)arg); 3576 } 3577 case SIOCINQ: 3578 { 3579 struct sk_buff *skb; 3580 int amount = 0; 3581 3582 spin_lock_bh(&sk->sk_receive_queue.lock); 3583 skb = skb_peek(&sk->sk_receive_queue); 3584 if (skb) 3585 amount = skb->len; 3586 spin_unlock_bh(&sk->sk_receive_queue.lock); 3587 return put_user(amount, (int __user *)arg); 3588 } 3589 case SIOCGSTAMP: 3590 return sock_get_timestamp(sk, (struct timeval __user *)arg); 3591 case SIOCGSTAMPNS: 3592 return sock_get_timestampns(sk, (struct timespec __user *)arg); 3593 3594 #ifdef CONFIG_INET 3595 case SIOCADDRT: 3596 case SIOCDELRT: 3597 case SIOCDARP: 3598 case SIOCGARP: 3599 case SIOCSARP: 3600 case SIOCGIFADDR: 3601 case SIOCSIFADDR: 3602 case SIOCGIFBRDADDR: 3603 case SIOCSIFBRDADDR: 3604 case SIOCGIFNETMASK: 3605 case SIOCSIFNETMASK: 3606 case SIOCGIFDSTADDR: 3607 case SIOCSIFDSTADDR: 3608 case SIOCSIFFLAGS: 3609 return inet_dgram_ops.ioctl(sock, cmd, arg); 3610 #endif 3611 3612 default: 3613 return -ENOIOCTLCMD; 3614 } 3615 return 0; 3616 } 3617 3618 static unsigned int packet_poll(struct file *file, struct socket *sock, 3619 poll_table *wait) 3620 { 3621 struct sock *sk = sock->sk; 3622 struct packet_sock *po = pkt_sk(sk); 3623 unsigned int mask = datagram_poll(file, sock, wait); 3624 3625 spin_lock_bh(&sk->sk_receive_queue.lock); 3626 if (po->rx_ring.pg_vec) { 3627 if (!packet_previous_rx_frame(po, &po->rx_ring, 3628 TP_STATUS_KERNEL)) 3629 mask |= POLLIN | POLLRDNORM; 3630 } 3631 spin_unlock_bh(&sk->sk_receive_queue.lock); 3632 spin_lock_bh(&sk->sk_write_queue.lock); 3633 if (po->tx_ring.pg_vec) { 3634 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 3635 mask |= POLLOUT | POLLWRNORM; 3636 } 3637 spin_unlock_bh(&sk->sk_write_queue.lock); 3638 return mask; 3639 } 3640 3641 3642 /* Dirty? Well, I still did not learn better way to account 3643 * for user mmaps. 3644 */ 3645 3646 static void packet_mm_open(struct vm_area_struct *vma) 3647 { 3648 struct file *file = vma->vm_file; 3649 struct socket *sock = file->private_data; 3650 struct sock *sk = sock->sk; 3651 3652 if (sk) 3653 atomic_inc(&pkt_sk(sk)->mapped); 3654 } 3655 3656 static void packet_mm_close(struct vm_area_struct *vma) 3657 { 3658 struct file *file = vma->vm_file; 3659 struct socket *sock = file->private_data; 3660 struct sock *sk = sock->sk; 3661 3662 if (sk) 3663 atomic_dec(&pkt_sk(sk)->mapped); 3664 } 3665 3666 static const struct vm_operations_struct packet_mmap_ops = { 3667 .open = packet_mm_open, 3668 .close = packet_mm_close, 3669 }; 3670 3671 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 3672 unsigned int len) 3673 { 3674 int i; 3675 3676 for (i = 0; i < len; i++) { 3677 if (likely(pg_vec[i].buffer)) { 3678 if (is_vmalloc_addr(pg_vec[i].buffer)) 3679 vfree(pg_vec[i].buffer); 3680 else 3681 free_pages((unsigned long)pg_vec[i].buffer, 3682 order); 3683 pg_vec[i].buffer = NULL; 3684 } 3685 } 3686 kfree(pg_vec); 3687 } 3688 3689 static char *alloc_one_pg_vec_page(unsigned long order) 3690 { 3691 char *buffer; 3692 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 3693 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 3694 3695 buffer = (char *) __get_free_pages(gfp_flags, order); 3696 if (buffer) 3697 return buffer; 3698 3699 /* __get_free_pages failed, fall back to vmalloc */ 3700 buffer = vzalloc((1 << order) * PAGE_SIZE); 3701 if (buffer) 3702 return buffer; 3703 3704 /* vmalloc failed, lets dig into swap here */ 3705 gfp_flags &= ~__GFP_NORETRY; 3706 buffer = (char *) __get_free_pages(gfp_flags, order); 3707 if (buffer) 3708 return buffer; 3709 3710 /* complete and utter failure */ 3711 return NULL; 3712 } 3713 3714 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 3715 { 3716 unsigned int block_nr = req->tp_block_nr; 3717 struct pgv *pg_vec; 3718 int i; 3719 3720 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL); 3721 if (unlikely(!pg_vec)) 3722 goto out; 3723 3724 for (i = 0; i < block_nr; i++) { 3725 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 3726 if (unlikely(!pg_vec[i].buffer)) 3727 goto out_free_pgvec; 3728 } 3729 3730 out: 3731 return pg_vec; 3732 3733 out_free_pgvec: 3734 free_pg_vec(pg_vec, order, block_nr); 3735 pg_vec = NULL; 3736 goto out; 3737 } 3738 3739 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 3740 int closing, int tx_ring) 3741 { 3742 struct pgv *pg_vec = NULL; 3743 struct packet_sock *po = pkt_sk(sk); 3744 int was_running, order = 0; 3745 struct packet_ring_buffer *rb; 3746 struct sk_buff_head *rb_queue; 3747 __be16 num; 3748 int err = -EINVAL; 3749 /* Added to avoid minimal code churn */ 3750 struct tpacket_req *req = &req_u->req; 3751 3752 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */ 3753 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) { 3754 WARN(1, "Tx-ring is not supported.\n"); 3755 goto out; 3756 } 3757 3758 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 3759 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 3760 3761 err = -EBUSY; 3762 if (!closing) { 3763 if (atomic_read(&po->mapped)) 3764 goto out; 3765 if (packet_read_pending(rb)) 3766 goto out; 3767 } 3768 3769 if (req->tp_block_nr) { 3770 /* Sanity tests and some calculations */ 3771 err = -EBUSY; 3772 if (unlikely(rb->pg_vec)) 3773 goto out; 3774 3775 switch (po->tp_version) { 3776 case TPACKET_V1: 3777 po->tp_hdrlen = TPACKET_HDRLEN; 3778 break; 3779 case TPACKET_V2: 3780 po->tp_hdrlen = TPACKET2_HDRLEN; 3781 break; 3782 case TPACKET_V3: 3783 po->tp_hdrlen = TPACKET3_HDRLEN; 3784 break; 3785 } 3786 3787 err = -EINVAL; 3788 if (unlikely((int)req->tp_block_size <= 0)) 3789 goto out; 3790 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1))) 3791 goto out; 3792 if (unlikely(req->tp_frame_size < po->tp_hdrlen + 3793 po->tp_reserve)) 3794 goto out; 3795 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 3796 goto out; 3797 3798 rb->frames_per_block = req->tp_block_size/req->tp_frame_size; 3799 if (unlikely(rb->frames_per_block <= 0)) 3800 goto out; 3801 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 3802 req->tp_frame_nr)) 3803 goto out; 3804 3805 err = -ENOMEM; 3806 order = get_order(req->tp_block_size); 3807 pg_vec = alloc_pg_vec(req, order); 3808 if (unlikely(!pg_vec)) 3809 goto out; 3810 switch (po->tp_version) { 3811 case TPACKET_V3: 3812 /* Transmit path is not supported. We checked 3813 * it above but just being paranoid 3814 */ 3815 if (!tx_ring) 3816 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring); 3817 break; 3818 default: 3819 break; 3820 } 3821 } 3822 /* Done */ 3823 else { 3824 err = -EINVAL; 3825 if (unlikely(req->tp_frame_nr)) 3826 goto out; 3827 } 3828 3829 lock_sock(sk); 3830 3831 /* Detach socket from network */ 3832 spin_lock(&po->bind_lock); 3833 was_running = po->running; 3834 num = po->num; 3835 if (was_running) { 3836 po->num = 0; 3837 __unregister_prot_hook(sk, false); 3838 } 3839 spin_unlock(&po->bind_lock); 3840 3841 synchronize_net(); 3842 3843 err = -EBUSY; 3844 mutex_lock(&po->pg_vec_lock); 3845 if (closing || atomic_read(&po->mapped) == 0) { 3846 err = 0; 3847 spin_lock_bh(&rb_queue->lock); 3848 swap(rb->pg_vec, pg_vec); 3849 rb->frame_max = (req->tp_frame_nr - 1); 3850 rb->head = 0; 3851 rb->frame_size = req->tp_frame_size; 3852 spin_unlock_bh(&rb_queue->lock); 3853 3854 swap(rb->pg_vec_order, order); 3855 swap(rb->pg_vec_len, req->tp_block_nr); 3856 3857 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 3858 po->prot_hook.func = (po->rx_ring.pg_vec) ? 3859 tpacket_rcv : packet_rcv; 3860 skb_queue_purge(rb_queue); 3861 if (atomic_read(&po->mapped)) 3862 pr_err("packet_mmap: vma is busy: %d\n", 3863 atomic_read(&po->mapped)); 3864 } 3865 mutex_unlock(&po->pg_vec_lock); 3866 3867 spin_lock(&po->bind_lock); 3868 if (was_running) { 3869 po->num = num; 3870 register_prot_hook(sk); 3871 } 3872 spin_unlock(&po->bind_lock); 3873 if (closing && (po->tp_version > TPACKET_V2)) { 3874 /* Because we don't support block-based V3 on tx-ring */ 3875 if (!tx_ring) 3876 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue); 3877 } 3878 release_sock(sk); 3879 3880 if (pg_vec) 3881 free_pg_vec(pg_vec, order, req->tp_block_nr); 3882 out: 3883 return err; 3884 } 3885 3886 static int packet_mmap(struct file *file, struct socket *sock, 3887 struct vm_area_struct *vma) 3888 { 3889 struct sock *sk = sock->sk; 3890 struct packet_sock *po = pkt_sk(sk); 3891 unsigned long size, expected_size; 3892 struct packet_ring_buffer *rb; 3893 unsigned long start; 3894 int err = -EINVAL; 3895 int i; 3896 3897 if (vma->vm_pgoff) 3898 return -EINVAL; 3899 3900 mutex_lock(&po->pg_vec_lock); 3901 3902 expected_size = 0; 3903 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 3904 if (rb->pg_vec) { 3905 expected_size += rb->pg_vec_len 3906 * rb->pg_vec_pages 3907 * PAGE_SIZE; 3908 } 3909 } 3910 3911 if (expected_size == 0) 3912 goto out; 3913 3914 size = vma->vm_end - vma->vm_start; 3915 if (size != expected_size) 3916 goto out; 3917 3918 start = vma->vm_start; 3919 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 3920 if (rb->pg_vec == NULL) 3921 continue; 3922 3923 for (i = 0; i < rb->pg_vec_len; i++) { 3924 struct page *page; 3925 void *kaddr = rb->pg_vec[i].buffer; 3926 int pg_num; 3927 3928 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 3929 page = pgv_to_page(kaddr); 3930 err = vm_insert_page(vma, start, page); 3931 if (unlikely(err)) 3932 goto out; 3933 start += PAGE_SIZE; 3934 kaddr += PAGE_SIZE; 3935 } 3936 } 3937 } 3938 3939 atomic_inc(&po->mapped); 3940 vma->vm_ops = &packet_mmap_ops; 3941 err = 0; 3942 3943 out: 3944 mutex_unlock(&po->pg_vec_lock); 3945 return err; 3946 } 3947 3948 static const struct proto_ops packet_ops_spkt = { 3949 .family = PF_PACKET, 3950 .owner = THIS_MODULE, 3951 .release = packet_release, 3952 .bind = packet_bind_spkt, 3953 .connect = sock_no_connect, 3954 .socketpair = sock_no_socketpair, 3955 .accept = sock_no_accept, 3956 .getname = packet_getname_spkt, 3957 .poll = datagram_poll, 3958 .ioctl = packet_ioctl, 3959 .listen = sock_no_listen, 3960 .shutdown = sock_no_shutdown, 3961 .setsockopt = sock_no_setsockopt, 3962 .getsockopt = sock_no_getsockopt, 3963 .sendmsg = packet_sendmsg_spkt, 3964 .recvmsg = packet_recvmsg, 3965 .mmap = sock_no_mmap, 3966 .sendpage = sock_no_sendpage, 3967 }; 3968 3969 static const struct proto_ops packet_ops = { 3970 .family = PF_PACKET, 3971 .owner = THIS_MODULE, 3972 .release = packet_release, 3973 .bind = packet_bind, 3974 .connect = sock_no_connect, 3975 .socketpair = sock_no_socketpair, 3976 .accept = sock_no_accept, 3977 .getname = packet_getname, 3978 .poll = packet_poll, 3979 .ioctl = packet_ioctl, 3980 .listen = sock_no_listen, 3981 .shutdown = sock_no_shutdown, 3982 .setsockopt = packet_setsockopt, 3983 .getsockopt = packet_getsockopt, 3984 .sendmsg = packet_sendmsg, 3985 .recvmsg = packet_recvmsg, 3986 .mmap = packet_mmap, 3987 .sendpage = sock_no_sendpage, 3988 }; 3989 3990 static const struct net_proto_family packet_family_ops = { 3991 .family = PF_PACKET, 3992 .create = packet_create, 3993 .owner = THIS_MODULE, 3994 }; 3995 3996 static struct notifier_block packet_netdev_notifier = { 3997 .notifier_call = packet_notifier, 3998 }; 3999 4000 #ifdef CONFIG_PROC_FS 4001 4002 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4003 __acquires(RCU) 4004 { 4005 struct net *net = seq_file_net(seq); 4006 4007 rcu_read_lock(); 4008 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4009 } 4010 4011 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4012 { 4013 struct net *net = seq_file_net(seq); 4014 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4015 } 4016 4017 static void packet_seq_stop(struct seq_file *seq, void *v) 4018 __releases(RCU) 4019 { 4020 rcu_read_unlock(); 4021 } 4022 4023 static int packet_seq_show(struct seq_file *seq, void *v) 4024 { 4025 if (v == SEQ_START_TOKEN) 4026 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4027 else { 4028 struct sock *s = sk_entry(v); 4029 const struct packet_sock *po = pkt_sk(s); 4030 4031 seq_printf(seq, 4032 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4033 s, 4034 atomic_read(&s->sk_refcnt), 4035 s->sk_type, 4036 ntohs(po->num), 4037 po->ifindex, 4038 po->running, 4039 atomic_read(&s->sk_rmem_alloc), 4040 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4041 sock_i_ino(s)); 4042 } 4043 4044 return 0; 4045 } 4046 4047 static const struct seq_operations packet_seq_ops = { 4048 .start = packet_seq_start, 4049 .next = packet_seq_next, 4050 .stop = packet_seq_stop, 4051 .show = packet_seq_show, 4052 }; 4053 4054 static int packet_seq_open(struct inode *inode, struct file *file) 4055 { 4056 return seq_open_net(inode, file, &packet_seq_ops, 4057 sizeof(struct seq_net_private)); 4058 } 4059 4060 static const struct file_operations packet_seq_fops = { 4061 .owner = THIS_MODULE, 4062 .open = packet_seq_open, 4063 .read = seq_read, 4064 .llseek = seq_lseek, 4065 .release = seq_release_net, 4066 }; 4067 4068 #endif 4069 4070 static int __net_init packet_net_init(struct net *net) 4071 { 4072 mutex_init(&net->packet.sklist_lock); 4073 INIT_HLIST_HEAD(&net->packet.sklist); 4074 4075 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops)) 4076 return -ENOMEM; 4077 4078 return 0; 4079 } 4080 4081 static void __net_exit packet_net_exit(struct net *net) 4082 { 4083 remove_proc_entry("packet", net->proc_net); 4084 } 4085 4086 static struct pernet_operations packet_net_ops = { 4087 .init = packet_net_init, 4088 .exit = packet_net_exit, 4089 }; 4090 4091 4092 static void __exit packet_exit(void) 4093 { 4094 unregister_netdevice_notifier(&packet_netdev_notifier); 4095 unregister_pernet_subsys(&packet_net_ops); 4096 sock_unregister(PF_PACKET); 4097 proto_unregister(&packet_proto); 4098 } 4099 4100 static int __init packet_init(void) 4101 { 4102 int rc = proto_register(&packet_proto, 0); 4103 4104 if (rc != 0) 4105 goto out; 4106 4107 sock_register(&packet_family_ops); 4108 register_pernet_subsys(&packet_net_ops); 4109 register_netdevice_notifier(&packet_netdev_notifier); 4110 out: 4111 return rc; 4112 } 4113 4114 module_init(packet_init); 4115 module_exit(packet_exit); 4116 MODULE_LICENSE("GPL"); 4117 MODULE_ALIAS_NETPROTO(PF_PACKET); 4118