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 <linux/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 #include <linux/bpf.h> 96 #include <net/compat.h> 97 98 #include "internal.h" 99 100 /* 101 Assumptions: 102 - if device has no dev->hard_header routine, it adds and removes ll header 103 inside itself. In this case ll header is invisible outside of device, 104 but higher levels still should reserve dev->hard_header_len. 105 Some devices are enough clever to reallocate skb, when header 106 will not fit to reserved space (tunnel), another ones are silly 107 (PPP). 108 - packet socket receives packets with pulled ll header, 109 so that SOCK_RAW should push it back. 110 111 On receive: 112 ----------- 113 114 Incoming, dev->hard_header!=NULL 115 mac_header -> ll header 116 data -> data 117 118 Outgoing, dev->hard_header!=NULL 119 mac_header -> ll header 120 data -> ll header 121 122 Incoming, dev->hard_header==NULL 123 mac_header -> UNKNOWN position. It is very likely, that it points to ll 124 header. PPP makes it, that is wrong, because introduce 125 assymetry between rx and tx paths. 126 data -> data 127 128 Outgoing, dev->hard_header==NULL 129 mac_header -> data. ll header is still not built! 130 data -> data 131 132 Resume 133 If dev->hard_header==NULL we are unlikely to restore sensible ll header. 134 135 136 On transmit: 137 ------------ 138 139 dev->hard_header != NULL 140 mac_header -> ll header 141 data -> ll header 142 143 dev->hard_header == NULL (ll header is added by device, we cannot control it) 144 mac_header -> data 145 data -> data 146 147 We should set nh.raw on output to correct posistion, 148 packet classifier depends on it. 149 */ 150 151 /* Private packet socket structures. */ 152 153 /* identical to struct packet_mreq except it has 154 * a longer address field. 155 */ 156 struct packet_mreq_max { 157 int mr_ifindex; 158 unsigned short mr_type; 159 unsigned short mr_alen; 160 unsigned char mr_address[MAX_ADDR_LEN]; 161 }; 162 163 union tpacket_uhdr { 164 struct tpacket_hdr *h1; 165 struct tpacket2_hdr *h2; 166 struct tpacket3_hdr *h3; 167 void *raw; 168 }; 169 170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 171 int closing, int tx_ring); 172 173 #define V3_ALIGNMENT (8) 174 175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) 176 177 #define BLK_PLUS_PRIV(sz_of_priv) \ 178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) 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_rcv(struct sk_buff *skb, struct net_device *dev, 190 struct packet_type *pt, struct net_device *orig_dev); 191 192 static void *packet_previous_frame(struct packet_sock *po, 193 struct packet_ring_buffer *rb, 194 int status); 195 static void packet_increment_head(struct packet_ring_buffer *buff); 196 static int prb_curr_blk_in_use(struct tpacket_block_desc *); 197 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 198 struct packet_sock *); 199 static void prb_retire_current_block(struct tpacket_kbdq_core *, 200 struct packet_sock *, unsigned int status); 201 static int prb_queue_frozen(struct tpacket_kbdq_core *); 202 static void prb_open_block(struct tpacket_kbdq_core *, 203 struct tpacket_block_desc *); 204 static void prb_retire_rx_blk_timer_expired(struct timer_list *); 205 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); 206 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 207 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 208 struct tpacket3_hdr *); 209 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 210 struct tpacket3_hdr *); 211 static void packet_flush_mclist(struct sock *sk); 212 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb); 213 214 struct packet_skb_cb { 215 union { 216 struct sockaddr_pkt pkt; 217 union { 218 /* Trick: alias skb original length with 219 * ll.sll_family and ll.protocol in order 220 * to save room. 221 */ 222 unsigned int origlen; 223 struct sockaddr_ll ll; 224 }; 225 } sa; 226 }; 227 228 #define vio_le() virtio_legacy_is_little_endian() 229 230 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 231 232 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 233 #define GET_PBLOCK_DESC(x, bid) \ 234 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 235 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 236 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 237 #define GET_NEXT_PRB_BLK_NUM(x) \ 238 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 239 ((x)->kactive_blk_num+1) : 0) 240 241 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 242 static void __fanout_link(struct sock *sk, struct packet_sock *po); 243 244 static int packet_direct_xmit(struct sk_buff *skb) 245 { 246 struct net_device *dev = skb->dev; 247 struct sk_buff *orig_skb = skb; 248 struct netdev_queue *txq; 249 int ret = NETDEV_TX_BUSY; 250 bool again = false; 251 252 if (unlikely(!netif_running(dev) || 253 !netif_carrier_ok(dev))) 254 goto drop; 255 256 skb = validate_xmit_skb_list(skb, dev, &again); 257 if (skb != orig_skb) 258 goto drop; 259 260 packet_pick_tx_queue(dev, skb); 261 txq = skb_get_tx_queue(dev, skb); 262 263 local_bh_disable(); 264 265 HARD_TX_LOCK(dev, txq, smp_processor_id()); 266 if (!netif_xmit_frozen_or_drv_stopped(txq)) 267 ret = netdev_start_xmit(skb, dev, txq, false); 268 HARD_TX_UNLOCK(dev, txq); 269 270 local_bh_enable(); 271 272 if (!dev_xmit_complete(ret)) 273 kfree_skb(skb); 274 275 return ret; 276 drop: 277 atomic_long_inc(&dev->tx_dropped); 278 kfree_skb_list(skb); 279 return NET_XMIT_DROP; 280 } 281 282 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 283 { 284 struct net_device *dev; 285 286 rcu_read_lock(); 287 dev = rcu_dereference(po->cached_dev); 288 if (likely(dev)) 289 dev_hold(dev); 290 rcu_read_unlock(); 291 292 return dev; 293 } 294 295 static void packet_cached_dev_assign(struct packet_sock *po, 296 struct net_device *dev) 297 { 298 rcu_assign_pointer(po->cached_dev, dev); 299 } 300 301 static void packet_cached_dev_reset(struct packet_sock *po) 302 { 303 RCU_INIT_POINTER(po->cached_dev, NULL); 304 } 305 306 static bool packet_use_direct_xmit(const struct packet_sock *po) 307 { 308 return po->xmit == packet_direct_xmit; 309 } 310 311 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 312 { 313 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues; 314 } 315 316 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 317 { 318 const struct net_device_ops *ops = dev->netdev_ops; 319 u16 queue_index; 320 321 if (ops->ndo_select_queue) { 322 queue_index = ops->ndo_select_queue(dev, skb, NULL, 323 __packet_pick_tx_queue); 324 queue_index = netdev_cap_txqueue(dev, queue_index); 325 } else { 326 queue_index = __packet_pick_tx_queue(dev, skb); 327 } 328 329 skb_set_queue_mapping(skb, queue_index); 330 } 331 332 /* __register_prot_hook must be invoked through register_prot_hook 333 * or from a context in which asynchronous accesses to the packet 334 * socket is not possible (packet_create()). 335 */ 336 static void __register_prot_hook(struct sock *sk) 337 { 338 struct packet_sock *po = pkt_sk(sk); 339 340 if (!po->running) { 341 if (po->fanout) 342 __fanout_link(sk, po); 343 else 344 dev_add_pack(&po->prot_hook); 345 346 sock_hold(sk); 347 po->running = 1; 348 } 349 } 350 351 static void register_prot_hook(struct sock *sk) 352 { 353 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); 354 __register_prot_hook(sk); 355 } 356 357 /* If the sync parameter is true, we will temporarily drop 358 * the po->bind_lock and do a synchronize_net to make sure no 359 * asynchronous packet processing paths still refer to the elements 360 * of po->prot_hook. If the sync parameter is false, it is the 361 * callers responsibility to take care of this. 362 */ 363 static void __unregister_prot_hook(struct sock *sk, bool sync) 364 { 365 struct packet_sock *po = pkt_sk(sk); 366 367 lockdep_assert_held_once(&po->bind_lock); 368 369 po->running = 0; 370 371 if (po->fanout) 372 __fanout_unlink(sk, po); 373 else 374 __dev_remove_pack(&po->prot_hook); 375 376 __sock_put(sk); 377 378 if (sync) { 379 spin_unlock(&po->bind_lock); 380 synchronize_net(); 381 spin_lock(&po->bind_lock); 382 } 383 } 384 385 static void unregister_prot_hook(struct sock *sk, bool sync) 386 { 387 struct packet_sock *po = pkt_sk(sk); 388 389 if (po->running) 390 __unregister_prot_hook(sk, sync); 391 } 392 393 static inline struct page * __pure pgv_to_page(void *addr) 394 { 395 if (is_vmalloc_addr(addr)) 396 return vmalloc_to_page(addr); 397 return virt_to_page(addr); 398 } 399 400 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 401 { 402 union tpacket_uhdr h; 403 404 h.raw = frame; 405 switch (po->tp_version) { 406 case TPACKET_V1: 407 h.h1->tp_status = status; 408 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 409 break; 410 case TPACKET_V2: 411 h.h2->tp_status = status; 412 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 413 break; 414 case TPACKET_V3: 415 h.h3->tp_status = status; 416 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 417 break; 418 default: 419 WARN(1, "TPACKET version not supported.\n"); 420 BUG(); 421 } 422 423 smp_wmb(); 424 } 425 426 static int __packet_get_status(struct packet_sock *po, void *frame) 427 { 428 union tpacket_uhdr h; 429 430 smp_rmb(); 431 432 h.raw = frame; 433 switch (po->tp_version) { 434 case TPACKET_V1: 435 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 436 return h.h1->tp_status; 437 case TPACKET_V2: 438 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 439 return h.h2->tp_status; 440 case TPACKET_V3: 441 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 442 return h.h3->tp_status; 443 default: 444 WARN(1, "TPACKET version not supported.\n"); 445 BUG(); 446 return 0; 447 } 448 } 449 450 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, 451 unsigned int flags) 452 { 453 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 454 455 if (shhwtstamps && 456 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 457 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) 458 return TP_STATUS_TS_RAW_HARDWARE; 459 460 if (ktime_to_timespec_cond(skb->tstamp, ts)) 461 return TP_STATUS_TS_SOFTWARE; 462 463 return 0; 464 } 465 466 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 467 struct sk_buff *skb) 468 { 469 union tpacket_uhdr h; 470 struct timespec ts; 471 __u32 ts_status; 472 473 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 474 return 0; 475 476 h.raw = frame; 477 switch (po->tp_version) { 478 case TPACKET_V1: 479 h.h1->tp_sec = ts.tv_sec; 480 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 481 break; 482 case TPACKET_V2: 483 h.h2->tp_sec = ts.tv_sec; 484 h.h2->tp_nsec = ts.tv_nsec; 485 break; 486 case TPACKET_V3: 487 h.h3->tp_sec = ts.tv_sec; 488 h.h3->tp_nsec = ts.tv_nsec; 489 break; 490 default: 491 WARN(1, "TPACKET version not supported.\n"); 492 BUG(); 493 } 494 495 /* one flush is safe, as both fields always lie on the same cacheline */ 496 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 497 smp_wmb(); 498 499 return ts_status; 500 } 501 502 static void *packet_lookup_frame(struct packet_sock *po, 503 struct packet_ring_buffer *rb, 504 unsigned int position, 505 int status) 506 { 507 unsigned int pg_vec_pos, frame_offset; 508 union tpacket_uhdr h; 509 510 pg_vec_pos = position / rb->frames_per_block; 511 frame_offset = position % rb->frames_per_block; 512 513 h.raw = rb->pg_vec[pg_vec_pos].buffer + 514 (frame_offset * rb->frame_size); 515 516 if (status != __packet_get_status(po, h.raw)) 517 return NULL; 518 519 return h.raw; 520 } 521 522 static void *packet_current_frame(struct packet_sock *po, 523 struct packet_ring_buffer *rb, 524 int status) 525 { 526 return packet_lookup_frame(po, rb, rb->head, status); 527 } 528 529 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 530 { 531 del_timer_sync(&pkc->retire_blk_timer); 532 } 533 534 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 535 struct sk_buff_head *rb_queue) 536 { 537 struct tpacket_kbdq_core *pkc; 538 539 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 540 541 spin_lock_bh(&rb_queue->lock); 542 pkc->delete_blk_timer = 1; 543 spin_unlock_bh(&rb_queue->lock); 544 545 prb_del_retire_blk_timer(pkc); 546 } 547 548 static void prb_setup_retire_blk_timer(struct packet_sock *po) 549 { 550 struct tpacket_kbdq_core *pkc; 551 552 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 553 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, 554 0); 555 pkc->retire_blk_timer.expires = jiffies; 556 } 557 558 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 559 int blk_size_in_bytes) 560 { 561 struct net_device *dev; 562 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; 563 struct ethtool_link_ksettings ecmd; 564 int err; 565 566 rtnl_lock(); 567 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 568 if (unlikely(!dev)) { 569 rtnl_unlock(); 570 return DEFAULT_PRB_RETIRE_TOV; 571 } 572 err = __ethtool_get_link_ksettings(dev, &ecmd); 573 rtnl_unlock(); 574 if (!err) { 575 /* 576 * If the link speed is so slow you don't really 577 * need to worry about perf anyways 578 */ 579 if (ecmd.base.speed < SPEED_1000 || 580 ecmd.base.speed == SPEED_UNKNOWN) { 581 return DEFAULT_PRB_RETIRE_TOV; 582 } else { 583 msec = 1; 584 div = ecmd.base.speed / 1000; 585 } 586 } 587 588 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 589 590 if (div) 591 mbits /= div; 592 593 tmo = mbits * msec; 594 595 if (div) 596 return tmo+1; 597 return tmo; 598 } 599 600 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 601 union tpacket_req_u *req_u) 602 { 603 p1->feature_req_word = req_u->req3.tp_feature_req_word; 604 } 605 606 static void init_prb_bdqc(struct packet_sock *po, 607 struct packet_ring_buffer *rb, 608 struct pgv *pg_vec, 609 union tpacket_req_u *req_u) 610 { 611 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 612 struct tpacket_block_desc *pbd; 613 614 memset(p1, 0x0, sizeof(*p1)); 615 616 p1->knxt_seq_num = 1; 617 p1->pkbdq = pg_vec; 618 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 619 p1->pkblk_start = pg_vec[0].buffer; 620 p1->kblk_size = req_u->req3.tp_block_size; 621 p1->knum_blocks = req_u->req3.tp_block_nr; 622 p1->hdrlen = po->tp_hdrlen; 623 p1->version = po->tp_version; 624 p1->last_kactive_blk_num = 0; 625 po->stats.stats3.tp_freeze_q_cnt = 0; 626 if (req_u->req3.tp_retire_blk_tov) 627 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 628 else 629 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 630 req_u->req3.tp_block_size); 631 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 632 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 633 634 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 635 prb_init_ft_ops(p1, req_u); 636 prb_setup_retire_blk_timer(po); 637 prb_open_block(p1, pbd); 638 } 639 640 /* Do NOT update the last_blk_num first. 641 * Assumes sk_buff_head lock is held. 642 */ 643 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 644 { 645 mod_timer(&pkc->retire_blk_timer, 646 jiffies + pkc->tov_in_jiffies); 647 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 648 } 649 650 /* 651 * Timer logic: 652 * 1) We refresh the timer only when we open a block. 653 * By doing this we don't waste cycles refreshing the timer 654 * on packet-by-packet basis. 655 * 656 * With a 1MB block-size, on a 1Gbps line, it will take 657 * i) ~8 ms to fill a block + ii) memcpy etc. 658 * In this cut we are not accounting for the memcpy time. 659 * 660 * So, if the user sets the 'tmo' to 10ms then the timer 661 * will never fire while the block is still getting filled 662 * (which is what we want). However, the user could choose 663 * to close a block early and that's fine. 664 * 665 * But when the timer does fire, we check whether or not to refresh it. 666 * Since the tmo granularity is in msecs, it is not too expensive 667 * to refresh the timer, lets say every '8' msecs. 668 * Either the user can set the 'tmo' or we can derive it based on 669 * a) line-speed and b) block-size. 670 * prb_calc_retire_blk_tmo() calculates the tmo. 671 * 672 */ 673 static void prb_retire_rx_blk_timer_expired(struct timer_list *t) 674 { 675 struct packet_sock *po = 676 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer); 677 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 678 unsigned int frozen; 679 struct tpacket_block_desc *pbd; 680 681 spin_lock(&po->sk.sk_receive_queue.lock); 682 683 frozen = prb_queue_frozen(pkc); 684 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 685 686 if (unlikely(pkc->delete_blk_timer)) 687 goto out; 688 689 /* We only need to plug the race when the block is partially filled. 690 * tpacket_rcv: 691 * lock(); increment BLOCK_NUM_PKTS; unlock() 692 * copy_bits() is in progress ... 693 * timer fires on other cpu: 694 * we can't retire the current block because copy_bits 695 * is in progress. 696 * 697 */ 698 if (BLOCK_NUM_PKTS(pbd)) { 699 while (atomic_read(&pkc->blk_fill_in_prog)) { 700 /* Waiting for skb_copy_bits to finish... */ 701 cpu_relax(); 702 } 703 } 704 705 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 706 if (!frozen) { 707 if (!BLOCK_NUM_PKTS(pbd)) { 708 /* An empty block. Just refresh the timer. */ 709 goto refresh_timer; 710 } 711 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 712 if (!prb_dispatch_next_block(pkc, po)) 713 goto refresh_timer; 714 else 715 goto out; 716 } else { 717 /* Case 1. Queue was frozen because user-space was 718 * lagging behind. 719 */ 720 if (prb_curr_blk_in_use(pbd)) { 721 /* 722 * Ok, user-space is still behind. 723 * So just refresh the timer. 724 */ 725 goto refresh_timer; 726 } else { 727 /* Case 2. queue was frozen,user-space caught up, 728 * now the link went idle && the timer fired. 729 * We don't have a block to close.So we open this 730 * block and restart the timer. 731 * opening a block thaws the queue,restarts timer 732 * Thawing/timer-refresh is a side effect. 733 */ 734 prb_open_block(pkc, pbd); 735 goto out; 736 } 737 } 738 } 739 740 refresh_timer: 741 _prb_refresh_rx_retire_blk_timer(pkc); 742 743 out: 744 spin_unlock(&po->sk.sk_receive_queue.lock); 745 } 746 747 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 748 struct tpacket_block_desc *pbd1, __u32 status) 749 { 750 /* Flush everything minus the block header */ 751 752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 753 u8 *start, *end; 754 755 start = (u8 *)pbd1; 756 757 /* Skip the block header(we know header WILL fit in 4K) */ 758 start += PAGE_SIZE; 759 760 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 761 for (; start < end; start += PAGE_SIZE) 762 flush_dcache_page(pgv_to_page(start)); 763 764 smp_wmb(); 765 #endif 766 767 /* Now update the block status. */ 768 769 BLOCK_STATUS(pbd1) = status; 770 771 /* Flush the block header */ 772 773 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 774 start = (u8 *)pbd1; 775 flush_dcache_page(pgv_to_page(start)); 776 777 smp_wmb(); 778 #endif 779 } 780 781 /* 782 * Side effect: 783 * 784 * 1) flush the block 785 * 2) Increment active_blk_num 786 * 787 * Note:We DONT refresh the timer on purpose. 788 * Because almost always the next block will be opened. 789 */ 790 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 791 struct tpacket_block_desc *pbd1, 792 struct packet_sock *po, unsigned int stat) 793 { 794 __u32 status = TP_STATUS_USER | stat; 795 796 struct tpacket3_hdr *last_pkt; 797 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 798 struct sock *sk = &po->sk; 799 800 if (po->stats.stats3.tp_drops) 801 status |= TP_STATUS_LOSING; 802 803 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 804 last_pkt->tp_next_offset = 0; 805 806 /* Get the ts of the last pkt */ 807 if (BLOCK_NUM_PKTS(pbd1)) { 808 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 809 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 810 } else { 811 /* Ok, we tmo'd - so get the current time. 812 * 813 * It shouldn't really happen as we don't close empty 814 * blocks. See prb_retire_rx_blk_timer_expired(). 815 */ 816 struct timespec ts; 817 getnstimeofday(&ts); 818 h1->ts_last_pkt.ts_sec = ts.tv_sec; 819 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 820 } 821 822 smp_wmb(); 823 824 /* Flush the block */ 825 prb_flush_block(pkc1, pbd1, status); 826 827 sk->sk_data_ready(sk); 828 829 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 830 } 831 832 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 833 { 834 pkc->reset_pending_on_curr_blk = 0; 835 } 836 837 /* 838 * Side effect of opening a block: 839 * 840 * 1) prb_queue is thawed. 841 * 2) retire_blk_timer is refreshed. 842 * 843 */ 844 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 845 struct tpacket_block_desc *pbd1) 846 { 847 struct timespec ts; 848 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 849 850 smp_rmb(); 851 852 /* We could have just memset this but we will lose the 853 * flexibility of making the priv area sticky 854 */ 855 856 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 857 BLOCK_NUM_PKTS(pbd1) = 0; 858 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 859 860 getnstimeofday(&ts); 861 862 h1->ts_first_pkt.ts_sec = ts.tv_sec; 863 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 864 865 pkc1->pkblk_start = (char *)pbd1; 866 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 867 868 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 869 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 870 871 pbd1->version = pkc1->version; 872 pkc1->prev = pkc1->nxt_offset; 873 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 874 875 prb_thaw_queue(pkc1); 876 _prb_refresh_rx_retire_blk_timer(pkc1); 877 878 smp_wmb(); 879 } 880 881 /* 882 * Queue freeze logic: 883 * 1) Assume tp_block_nr = 8 blocks. 884 * 2) At time 't0', user opens Rx ring. 885 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 886 * 4) user-space is either sleeping or processing block '0'. 887 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 888 * it will close block-7,loop around and try to fill block '0'. 889 * call-flow: 890 * __packet_lookup_frame_in_block 891 * prb_retire_current_block() 892 * prb_dispatch_next_block() 893 * |->(BLOCK_STATUS == USER) evaluates to true 894 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 895 * 6) Now there are two cases: 896 * 6.1) Link goes idle right after the queue is frozen. 897 * But remember, the last open_block() refreshed the timer. 898 * When this timer expires,it will refresh itself so that we can 899 * re-open block-0 in near future. 900 * 6.2) Link is busy and keeps on receiving packets. This is a simple 901 * case and __packet_lookup_frame_in_block will check if block-0 902 * is free and can now be re-used. 903 */ 904 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 905 struct packet_sock *po) 906 { 907 pkc->reset_pending_on_curr_blk = 1; 908 po->stats.stats3.tp_freeze_q_cnt++; 909 } 910 911 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 912 913 /* 914 * If the next block is free then we will dispatch it 915 * and return a good offset. 916 * Else, we will freeze the queue. 917 * So, caller must check the return value. 918 */ 919 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 920 struct packet_sock *po) 921 { 922 struct tpacket_block_desc *pbd; 923 924 smp_rmb(); 925 926 /* 1. Get current block num */ 927 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 928 929 /* 2. If this block is currently in_use then freeze the queue */ 930 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 931 prb_freeze_queue(pkc, po); 932 return NULL; 933 } 934 935 /* 936 * 3. 937 * open this block and return the offset where the first packet 938 * needs to get stored. 939 */ 940 prb_open_block(pkc, pbd); 941 return (void *)pkc->nxt_offset; 942 } 943 944 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 945 struct packet_sock *po, unsigned int status) 946 { 947 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 948 949 /* retire/close the current block */ 950 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 951 /* 952 * Plug the case where copy_bits() is in progress on 953 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 954 * have space to copy the pkt in the current block and 955 * called prb_retire_current_block() 956 * 957 * We don't need to worry about the TMO case because 958 * the timer-handler already handled this case. 959 */ 960 if (!(status & TP_STATUS_BLK_TMO)) { 961 while (atomic_read(&pkc->blk_fill_in_prog)) { 962 /* Waiting for skb_copy_bits to finish... */ 963 cpu_relax(); 964 } 965 } 966 prb_close_block(pkc, pbd, po, status); 967 return; 968 } 969 } 970 971 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) 972 { 973 return TP_STATUS_USER & BLOCK_STATUS(pbd); 974 } 975 976 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 977 { 978 return pkc->reset_pending_on_curr_blk; 979 } 980 981 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 982 { 983 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 984 atomic_dec(&pkc->blk_fill_in_prog); 985 } 986 987 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 988 struct tpacket3_hdr *ppd) 989 { 990 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 991 } 992 993 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 994 struct tpacket3_hdr *ppd) 995 { 996 ppd->hv1.tp_rxhash = 0; 997 } 998 999 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 1000 struct tpacket3_hdr *ppd) 1001 { 1002 if (skb_vlan_tag_present(pkc->skb)) { 1003 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 1004 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 1005 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 1006 } else { 1007 ppd->hv1.tp_vlan_tci = 0; 1008 ppd->hv1.tp_vlan_tpid = 0; 1009 ppd->tp_status = TP_STATUS_AVAILABLE; 1010 } 1011 } 1012 1013 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 1014 struct tpacket3_hdr *ppd) 1015 { 1016 ppd->hv1.tp_padding = 0; 1017 prb_fill_vlan_info(pkc, ppd); 1018 1019 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 1020 prb_fill_rxhash(pkc, ppd); 1021 else 1022 prb_clear_rxhash(pkc, ppd); 1023 } 1024 1025 static void prb_fill_curr_block(char *curr, 1026 struct tpacket_kbdq_core *pkc, 1027 struct tpacket_block_desc *pbd, 1028 unsigned int len) 1029 { 1030 struct tpacket3_hdr *ppd; 1031 1032 ppd = (struct tpacket3_hdr *)curr; 1033 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 1034 pkc->prev = curr; 1035 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 1036 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1037 BLOCK_NUM_PKTS(pbd) += 1; 1038 atomic_inc(&pkc->blk_fill_in_prog); 1039 prb_run_all_ft_ops(pkc, ppd); 1040 } 1041 1042 /* Assumes caller has the sk->rx_queue.lock */ 1043 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1044 struct sk_buff *skb, 1045 int status, 1046 unsigned int len 1047 ) 1048 { 1049 struct tpacket_kbdq_core *pkc; 1050 struct tpacket_block_desc *pbd; 1051 char *curr, *end; 1052 1053 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1054 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1055 1056 /* Queue is frozen when user space is lagging behind */ 1057 if (prb_queue_frozen(pkc)) { 1058 /* 1059 * Check if that last block which caused the queue to freeze, 1060 * is still in_use by user-space. 1061 */ 1062 if (prb_curr_blk_in_use(pbd)) { 1063 /* Can't record this packet */ 1064 return NULL; 1065 } else { 1066 /* 1067 * Ok, the block was released by user-space. 1068 * Now let's open that block. 1069 * opening a block also thaws the queue. 1070 * Thawing is a side effect. 1071 */ 1072 prb_open_block(pkc, pbd); 1073 } 1074 } 1075 1076 smp_mb(); 1077 curr = pkc->nxt_offset; 1078 pkc->skb = skb; 1079 end = (char *)pbd + pkc->kblk_size; 1080 1081 /* first try the current block */ 1082 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1083 prb_fill_curr_block(curr, pkc, pbd, len); 1084 return (void *)curr; 1085 } 1086 1087 /* Ok, close the current block */ 1088 prb_retire_current_block(pkc, po, 0); 1089 1090 /* Now, try to dispatch the next block */ 1091 curr = (char *)prb_dispatch_next_block(pkc, po); 1092 if (curr) { 1093 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1094 prb_fill_curr_block(curr, pkc, pbd, len); 1095 return (void *)curr; 1096 } 1097 1098 /* 1099 * No free blocks are available.user_space hasn't caught up yet. 1100 * Queue was just frozen and now this packet will get dropped. 1101 */ 1102 return NULL; 1103 } 1104 1105 static void *packet_current_rx_frame(struct packet_sock *po, 1106 struct sk_buff *skb, 1107 int status, unsigned int len) 1108 { 1109 char *curr = NULL; 1110 switch (po->tp_version) { 1111 case TPACKET_V1: 1112 case TPACKET_V2: 1113 curr = packet_lookup_frame(po, &po->rx_ring, 1114 po->rx_ring.head, status); 1115 return curr; 1116 case TPACKET_V3: 1117 return __packet_lookup_frame_in_block(po, skb, status, len); 1118 default: 1119 WARN(1, "TPACKET version not supported\n"); 1120 BUG(); 1121 return NULL; 1122 } 1123 } 1124 1125 static void *prb_lookup_block(struct packet_sock *po, 1126 struct packet_ring_buffer *rb, 1127 unsigned int idx, 1128 int status) 1129 { 1130 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1131 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1132 1133 if (status != BLOCK_STATUS(pbd)) 1134 return NULL; 1135 return pbd; 1136 } 1137 1138 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1139 { 1140 unsigned int prev; 1141 if (rb->prb_bdqc.kactive_blk_num) 1142 prev = rb->prb_bdqc.kactive_blk_num-1; 1143 else 1144 prev = rb->prb_bdqc.knum_blocks-1; 1145 return prev; 1146 } 1147 1148 /* Assumes caller has held the rx_queue.lock */ 1149 static void *__prb_previous_block(struct packet_sock *po, 1150 struct packet_ring_buffer *rb, 1151 int status) 1152 { 1153 unsigned int previous = prb_previous_blk_num(rb); 1154 return prb_lookup_block(po, rb, previous, status); 1155 } 1156 1157 static void *packet_previous_rx_frame(struct packet_sock *po, 1158 struct packet_ring_buffer *rb, 1159 int status) 1160 { 1161 if (po->tp_version <= TPACKET_V2) 1162 return packet_previous_frame(po, rb, status); 1163 1164 return __prb_previous_block(po, rb, status); 1165 } 1166 1167 static void packet_increment_rx_head(struct packet_sock *po, 1168 struct packet_ring_buffer *rb) 1169 { 1170 switch (po->tp_version) { 1171 case TPACKET_V1: 1172 case TPACKET_V2: 1173 return packet_increment_head(rb); 1174 case TPACKET_V3: 1175 default: 1176 WARN(1, "TPACKET version not supported.\n"); 1177 BUG(); 1178 return; 1179 } 1180 } 1181 1182 static void *packet_previous_frame(struct packet_sock *po, 1183 struct packet_ring_buffer *rb, 1184 int status) 1185 { 1186 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1187 return packet_lookup_frame(po, rb, previous, status); 1188 } 1189 1190 static void packet_increment_head(struct packet_ring_buffer *buff) 1191 { 1192 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1193 } 1194 1195 static void packet_inc_pending(struct packet_ring_buffer *rb) 1196 { 1197 this_cpu_inc(*rb->pending_refcnt); 1198 } 1199 1200 static void packet_dec_pending(struct packet_ring_buffer *rb) 1201 { 1202 this_cpu_dec(*rb->pending_refcnt); 1203 } 1204 1205 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1206 { 1207 unsigned int refcnt = 0; 1208 int cpu; 1209 1210 /* We don't use pending refcount in rx_ring. */ 1211 if (rb->pending_refcnt == NULL) 1212 return 0; 1213 1214 for_each_possible_cpu(cpu) 1215 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1216 1217 return refcnt; 1218 } 1219 1220 static int packet_alloc_pending(struct packet_sock *po) 1221 { 1222 po->rx_ring.pending_refcnt = NULL; 1223 1224 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1225 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1226 return -ENOBUFS; 1227 1228 return 0; 1229 } 1230 1231 static void packet_free_pending(struct packet_sock *po) 1232 { 1233 free_percpu(po->tx_ring.pending_refcnt); 1234 } 1235 1236 #define ROOM_POW_OFF 2 1237 #define ROOM_NONE 0x0 1238 #define ROOM_LOW 0x1 1239 #define ROOM_NORMAL 0x2 1240 1241 static bool __tpacket_has_room(struct packet_sock *po, int pow_off) 1242 { 1243 int idx, len; 1244 1245 len = po->rx_ring.frame_max + 1; 1246 idx = po->rx_ring.head; 1247 if (pow_off) 1248 idx += len >> pow_off; 1249 if (idx >= len) 1250 idx -= len; 1251 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1252 } 1253 1254 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off) 1255 { 1256 int idx, len; 1257 1258 len = po->rx_ring.prb_bdqc.knum_blocks; 1259 idx = po->rx_ring.prb_bdqc.kactive_blk_num; 1260 if (pow_off) 1261 idx += len >> pow_off; 1262 if (idx >= len) 1263 idx -= len; 1264 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1265 } 1266 1267 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1268 { 1269 struct sock *sk = &po->sk; 1270 int ret = ROOM_NONE; 1271 1272 if (po->prot_hook.func != tpacket_rcv) { 1273 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1274 - (skb ? skb->truesize : 0); 1275 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF)) 1276 return ROOM_NORMAL; 1277 else if (avail > 0) 1278 return ROOM_LOW; 1279 else 1280 return ROOM_NONE; 1281 } 1282 1283 if (po->tp_version == TPACKET_V3) { 1284 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1285 ret = ROOM_NORMAL; 1286 else if (__tpacket_v3_has_room(po, 0)) 1287 ret = ROOM_LOW; 1288 } else { 1289 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1290 ret = ROOM_NORMAL; 1291 else if (__tpacket_has_room(po, 0)) 1292 ret = ROOM_LOW; 1293 } 1294 1295 return ret; 1296 } 1297 1298 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1299 { 1300 int ret; 1301 bool has_room; 1302 1303 spin_lock_bh(&po->sk.sk_receive_queue.lock); 1304 ret = __packet_rcv_has_room(po, skb); 1305 has_room = ret == ROOM_NORMAL; 1306 if (po->pressure == has_room) 1307 po->pressure = !has_room; 1308 spin_unlock_bh(&po->sk.sk_receive_queue.lock); 1309 1310 return ret; 1311 } 1312 1313 static void packet_sock_destruct(struct sock *sk) 1314 { 1315 skb_queue_purge(&sk->sk_error_queue); 1316 1317 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1318 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 1319 1320 if (!sock_flag(sk, SOCK_DEAD)) { 1321 pr_err("Attempt to release alive packet socket: %p\n", sk); 1322 return; 1323 } 1324 1325 sk_refcnt_debug_dec(sk); 1326 } 1327 1328 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1329 { 1330 u32 rxhash; 1331 int i, count = 0; 1332 1333 rxhash = skb_get_hash(skb); 1334 for (i = 0; i < ROLLOVER_HLEN; i++) 1335 if (po->rollover->history[i] == rxhash) 1336 count++; 1337 1338 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash; 1339 return count > (ROLLOVER_HLEN >> 1); 1340 } 1341 1342 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1343 struct sk_buff *skb, 1344 unsigned int num) 1345 { 1346 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1347 } 1348 1349 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1350 struct sk_buff *skb, 1351 unsigned int num) 1352 { 1353 unsigned int val = atomic_inc_return(&f->rr_cur); 1354 1355 return val % num; 1356 } 1357 1358 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1359 struct sk_buff *skb, 1360 unsigned int num) 1361 { 1362 return smp_processor_id() % num; 1363 } 1364 1365 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1366 struct sk_buff *skb, 1367 unsigned int num) 1368 { 1369 return prandom_u32_max(num); 1370 } 1371 1372 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1373 struct sk_buff *skb, 1374 unsigned int idx, bool try_self, 1375 unsigned int num) 1376 { 1377 struct packet_sock *po, *po_next, *po_skip = NULL; 1378 unsigned int i, j, room = ROOM_NONE; 1379 1380 po = pkt_sk(f->arr[idx]); 1381 1382 if (try_self) { 1383 room = packet_rcv_has_room(po, skb); 1384 if (room == ROOM_NORMAL || 1385 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1386 return idx; 1387 po_skip = po; 1388 } 1389 1390 i = j = min_t(int, po->rollover->sock, num - 1); 1391 do { 1392 po_next = pkt_sk(f->arr[i]); 1393 if (po_next != po_skip && !po_next->pressure && 1394 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1395 if (i != j) 1396 po->rollover->sock = i; 1397 atomic_long_inc(&po->rollover->num); 1398 if (room == ROOM_LOW) 1399 atomic_long_inc(&po->rollover->num_huge); 1400 return i; 1401 } 1402 1403 if (++i == num) 1404 i = 0; 1405 } while (i != j); 1406 1407 atomic_long_inc(&po->rollover->num_failed); 1408 return idx; 1409 } 1410 1411 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1412 struct sk_buff *skb, 1413 unsigned int num) 1414 { 1415 return skb_get_queue_mapping(skb) % num; 1416 } 1417 1418 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1419 struct sk_buff *skb, 1420 unsigned int num) 1421 { 1422 struct bpf_prog *prog; 1423 unsigned int ret = 0; 1424 1425 rcu_read_lock(); 1426 prog = rcu_dereference(f->bpf_prog); 1427 if (prog) 1428 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1429 rcu_read_unlock(); 1430 1431 return ret; 1432 } 1433 1434 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1435 { 1436 return f->flags & (flag >> 8); 1437 } 1438 1439 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1440 struct packet_type *pt, struct net_device *orig_dev) 1441 { 1442 struct packet_fanout *f = pt->af_packet_priv; 1443 unsigned int num = READ_ONCE(f->num_members); 1444 struct net *net = read_pnet(&f->net); 1445 struct packet_sock *po; 1446 unsigned int idx; 1447 1448 if (!net_eq(dev_net(dev), net) || !num) { 1449 kfree_skb(skb); 1450 return 0; 1451 } 1452 1453 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1454 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1455 if (!skb) 1456 return 0; 1457 } 1458 switch (f->type) { 1459 case PACKET_FANOUT_HASH: 1460 default: 1461 idx = fanout_demux_hash(f, skb, num); 1462 break; 1463 case PACKET_FANOUT_LB: 1464 idx = fanout_demux_lb(f, skb, num); 1465 break; 1466 case PACKET_FANOUT_CPU: 1467 idx = fanout_demux_cpu(f, skb, num); 1468 break; 1469 case PACKET_FANOUT_RND: 1470 idx = fanout_demux_rnd(f, skb, num); 1471 break; 1472 case PACKET_FANOUT_QM: 1473 idx = fanout_demux_qm(f, skb, num); 1474 break; 1475 case PACKET_FANOUT_ROLLOVER: 1476 idx = fanout_demux_rollover(f, skb, 0, false, num); 1477 break; 1478 case PACKET_FANOUT_CBPF: 1479 case PACKET_FANOUT_EBPF: 1480 idx = fanout_demux_bpf(f, skb, num); 1481 break; 1482 } 1483 1484 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1485 idx = fanout_demux_rollover(f, skb, idx, true, num); 1486 1487 po = pkt_sk(f->arr[idx]); 1488 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1489 } 1490 1491 DEFINE_MUTEX(fanout_mutex); 1492 EXPORT_SYMBOL_GPL(fanout_mutex); 1493 static LIST_HEAD(fanout_list); 1494 static u16 fanout_next_id; 1495 1496 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1497 { 1498 struct packet_fanout *f = po->fanout; 1499 1500 spin_lock(&f->lock); 1501 f->arr[f->num_members] = sk; 1502 smp_wmb(); 1503 f->num_members++; 1504 if (f->num_members == 1) 1505 dev_add_pack(&f->prot_hook); 1506 spin_unlock(&f->lock); 1507 } 1508 1509 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1510 { 1511 struct packet_fanout *f = po->fanout; 1512 int i; 1513 1514 spin_lock(&f->lock); 1515 for (i = 0; i < f->num_members; i++) { 1516 if (f->arr[i] == sk) 1517 break; 1518 } 1519 BUG_ON(i >= f->num_members); 1520 f->arr[i] = f->arr[f->num_members - 1]; 1521 f->num_members--; 1522 if (f->num_members == 0) 1523 __dev_remove_pack(&f->prot_hook); 1524 spin_unlock(&f->lock); 1525 } 1526 1527 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1528 { 1529 if (sk->sk_family != PF_PACKET) 1530 return false; 1531 1532 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1533 } 1534 1535 static void fanout_init_data(struct packet_fanout *f) 1536 { 1537 switch (f->type) { 1538 case PACKET_FANOUT_LB: 1539 atomic_set(&f->rr_cur, 0); 1540 break; 1541 case PACKET_FANOUT_CBPF: 1542 case PACKET_FANOUT_EBPF: 1543 RCU_INIT_POINTER(f->bpf_prog, NULL); 1544 break; 1545 } 1546 } 1547 1548 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1549 { 1550 struct bpf_prog *old; 1551 1552 spin_lock(&f->lock); 1553 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1554 rcu_assign_pointer(f->bpf_prog, new); 1555 spin_unlock(&f->lock); 1556 1557 if (old) { 1558 synchronize_net(); 1559 bpf_prog_destroy(old); 1560 } 1561 } 1562 1563 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data, 1564 unsigned int len) 1565 { 1566 struct bpf_prog *new; 1567 struct sock_fprog fprog; 1568 int ret; 1569 1570 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1571 return -EPERM; 1572 if (len != sizeof(fprog)) 1573 return -EINVAL; 1574 if (copy_from_user(&fprog, data, len)) 1575 return -EFAULT; 1576 1577 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1578 if (ret) 1579 return ret; 1580 1581 __fanout_set_data_bpf(po->fanout, new); 1582 return 0; 1583 } 1584 1585 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data, 1586 unsigned int len) 1587 { 1588 struct bpf_prog *new; 1589 u32 fd; 1590 1591 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1592 return -EPERM; 1593 if (len != sizeof(fd)) 1594 return -EINVAL; 1595 if (copy_from_user(&fd, data, len)) 1596 return -EFAULT; 1597 1598 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1599 if (IS_ERR(new)) 1600 return PTR_ERR(new); 1601 1602 __fanout_set_data_bpf(po->fanout, new); 1603 return 0; 1604 } 1605 1606 static int fanout_set_data(struct packet_sock *po, char __user *data, 1607 unsigned int len) 1608 { 1609 switch (po->fanout->type) { 1610 case PACKET_FANOUT_CBPF: 1611 return fanout_set_data_cbpf(po, data, len); 1612 case PACKET_FANOUT_EBPF: 1613 return fanout_set_data_ebpf(po, data, len); 1614 default: 1615 return -EINVAL; 1616 }; 1617 } 1618 1619 static void fanout_release_data(struct packet_fanout *f) 1620 { 1621 switch (f->type) { 1622 case PACKET_FANOUT_CBPF: 1623 case PACKET_FANOUT_EBPF: 1624 __fanout_set_data_bpf(f, NULL); 1625 }; 1626 } 1627 1628 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) 1629 { 1630 struct packet_fanout *f; 1631 1632 list_for_each_entry(f, &fanout_list, list) { 1633 if (f->id == candidate_id && 1634 read_pnet(&f->net) == sock_net(sk)) { 1635 return false; 1636 } 1637 } 1638 return true; 1639 } 1640 1641 static bool fanout_find_new_id(struct sock *sk, u16 *new_id) 1642 { 1643 u16 id = fanout_next_id; 1644 1645 do { 1646 if (__fanout_id_is_free(sk, id)) { 1647 *new_id = id; 1648 fanout_next_id = id + 1; 1649 return true; 1650 } 1651 1652 id++; 1653 } while (id != fanout_next_id); 1654 1655 return false; 1656 } 1657 1658 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1659 { 1660 struct packet_rollover *rollover = NULL; 1661 struct packet_sock *po = pkt_sk(sk); 1662 struct packet_fanout *f, *match; 1663 u8 type = type_flags & 0xff; 1664 u8 flags = type_flags >> 8; 1665 int err; 1666 1667 switch (type) { 1668 case PACKET_FANOUT_ROLLOVER: 1669 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1670 return -EINVAL; 1671 case PACKET_FANOUT_HASH: 1672 case PACKET_FANOUT_LB: 1673 case PACKET_FANOUT_CPU: 1674 case PACKET_FANOUT_RND: 1675 case PACKET_FANOUT_QM: 1676 case PACKET_FANOUT_CBPF: 1677 case PACKET_FANOUT_EBPF: 1678 break; 1679 default: 1680 return -EINVAL; 1681 } 1682 1683 mutex_lock(&fanout_mutex); 1684 1685 err = -EALREADY; 1686 if (po->fanout) 1687 goto out; 1688 1689 if (type == PACKET_FANOUT_ROLLOVER || 1690 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1691 err = -ENOMEM; 1692 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); 1693 if (!rollover) 1694 goto out; 1695 atomic_long_set(&rollover->num, 0); 1696 atomic_long_set(&rollover->num_huge, 0); 1697 atomic_long_set(&rollover->num_failed, 0); 1698 } 1699 1700 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { 1701 if (id != 0) { 1702 err = -EINVAL; 1703 goto out; 1704 } 1705 if (!fanout_find_new_id(sk, &id)) { 1706 err = -ENOMEM; 1707 goto out; 1708 } 1709 /* ephemeral flag for the first socket in the group: drop it */ 1710 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); 1711 } 1712 1713 match = NULL; 1714 list_for_each_entry(f, &fanout_list, list) { 1715 if (f->id == id && 1716 read_pnet(&f->net) == sock_net(sk)) { 1717 match = f; 1718 break; 1719 } 1720 } 1721 err = -EINVAL; 1722 if (match && match->flags != flags) 1723 goto out; 1724 if (!match) { 1725 err = -ENOMEM; 1726 match = kzalloc(sizeof(*match), GFP_KERNEL); 1727 if (!match) 1728 goto out; 1729 write_pnet(&match->net, sock_net(sk)); 1730 match->id = id; 1731 match->type = type; 1732 match->flags = flags; 1733 INIT_LIST_HEAD(&match->list); 1734 spin_lock_init(&match->lock); 1735 refcount_set(&match->sk_ref, 0); 1736 fanout_init_data(match); 1737 match->prot_hook.type = po->prot_hook.type; 1738 match->prot_hook.dev = po->prot_hook.dev; 1739 match->prot_hook.func = packet_rcv_fanout; 1740 match->prot_hook.af_packet_priv = match; 1741 match->prot_hook.id_match = match_fanout_group; 1742 list_add(&match->list, &fanout_list); 1743 } 1744 err = -EINVAL; 1745 1746 spin_lock(&po->bind_lock); 1747 if (po->running && 1748 match->type == type && 1749 match->prot_hook.type == po->prot_hook.type && 1750 match->prot_hook.dev == po->prot_hook.dev) { 1751 err = -ENOSPC; 1752 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1753 __dev_remove_pack(&po->prot_hook); 1754 po->fanout = match; 1755 po->rollover = rollover; 1756 rollover = NULL; 1757 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); 1758 __fanout_link(sk, po); 1759 err = 0; 1760 } 1761 } 1762 spin_unlock(&po->bind_lock); 1763 1764 if (err && !refcount_read(&match->sk_ref)) { 1765 list_del(&match->list); 1766 kfree(match); 1767 } 1768 1769 out: 1770 kfree(rollover); 1771 mutex_unlock(&fanout_mutex); 1772 return err; 1773 } 1774 1775 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes 1776 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. 1777 * It is the responsibility of the caller to call fanout_release_data() and 1778 * free the returned packet_fanout (after synchronize_net()) 1779 */ 1780 static struct packet_fanout *fanout_release(struct sock *sk) 1781 { 1782 struct packet_sock *po = pkt_sk(sk); 1783 struct packet_fanout *f; 1784 1785 mutex_lock(&fanout_mutex); 1786 f = po->fanout; 1787 if (f) { 1788 po->fanout = NULL; 1789 1790 if (refcount_dec_and_test(&f->sk_ref)) 1791 list_del(&f->list); 1792 else 1793 f = NULL; 1794 } 1795 mutex_unlock(&fanout_mutex); 1796 1797 return f; 1798 } 1799 1800 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1801 struct sk_buff *skb) 1802 { 1803 /* Earlier code assumed this would be a VLAN pkt, double-check 1804 * this now that we have the actual packet in hand. We can only 1805 * do this check on Ethernet devices. 1806 */ 1807 if (unlikely(dev->type != ARPHRD_ETHER)) 1808 return false; 1809 1810 skb_reset_mac_header(skb); 1811 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1812 } 1813 1814 static const struct proto_ops packet_ops; 1815 1816 static const struct proto_ops packet_ops_spkt; 1817 1818 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1819 struct packet_type *pt, struct net_device *orig_dev) 1820 { 1821 struct sock *sk; 1822 struct sockaddr_pkt *spkt; 1823 1824 /* 1825 * When we registered the protocol we saved the socket in the data 1826 * field for just this event. 1827 */ 1828 1829 sk = pt->af_packet_priv; 1830 1831 /* 1832 * Yank back the headers [hope the device set this 1833 * right or kerboom...] 1834 * 1835 * Incoming packets have ll header pulled, 1836 * push it back. 1837 * 1838 * For outgoing ones skb->data == skb_mac_header(skb) 1839 * so that this procedure is noop. 1840 */ 1841 1842 if (skb->pkt_type == PACKET_LOOPBACK) 1843 goto out; 1844 1845 if (!net_eq(dev_net(dev), sock_net(sk))) 1846 goto out; 1847 1848 skb = skb_share_check(skb, GFP_ATOMIC); 1849 if (skb == NULL) 1850 goto oom; 1851 1852 /* drop any routing info */ 1853 skb_dst_drop(skb); 1854 1855 /* drop conntrack reference */ 1856 nf_reset(skb); 1857 1858 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1859 1860 skb_push(skb, skb->data - skb_mac_header(skb)); 1861 1862 /* 1863 * The SOCK_PACKET socket receives _all_ frames. 1864 */ 1865 1866 spkt->spkt_family = dev->type; 1867 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1868 spkt->spkt_protocol = skb->protocol; 1869 1870 /* 1871 * Charge the memory to the socket. This is done specifically 1872 * to prevent sockets using all the memory up. 1873 */ 1874 1875 if (sock_queue_rcv_skb(sk, skb) == 0) 1876 return 0; 1877 1878 out: 1879 kfree_skb(skb); 1880 oom: 1881 return 0; 1882 } 1883 1884 1885 /* 1886 * Output a raw packet to a device layer. This bypasses all the other 1887 * protocol layers and you must therefore supply it with a complete frame 1888 */ 1889 1890 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1891 size_t len) 1892 { 1893 struct sock *sk = sock->sk; 1894 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1895 struct sk_buff *skb = NULL; 1896 struct net_device *dev; 1897 struct sockcm_cookie sockc; 1898 __be16 proto = 0; 1899 int err; 1900 int extra_len = 0; 1901 1902 /* 1903 * Get and verify the address. 1904 */ 1905 1906 if (saddr) { 1907 if (msg->msg_namelen < sizeof(struct sockaddr)) 1908 return -EINVAL; 1909 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1910 proto = saddr->spkt_protocol; 1911 } else 1912 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1913 1914 /* 1915 * Find the device first to size check it 1916 */ 1917 1918 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1919 retry: 1920 rcu_read_lock(); 1921 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1922 err = -ENODEV; 1923 if (dev == NULL) 1924 goto out_unlock; 1925 1926 err = -ENETDOWN; 1927 if (!(dev->flags & IFF_UP)) 1928 goto out_unlock; 1929 1930 /* 1931 * You may not queue a frame bigger than the mtu. This is the lowest level 1932 * raw protocol and you must do your own fragmentation at this level. 1933 */ 1934 1935 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1936 if (!netif_supports_nofcs(dev)) { 1937 err = -EPROTONOSUPPORT; 1938 goto out_unlock; 1939 } 1940 extra_len = 4; /* We're doing our own CRC */ 1941 } 1942 1943 err = -EMSGSIZE; 1944 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1945 goto out_unlock; 1946 1947 if (!skb) { 1948 size_t reserved = LL_RESERVED_SPACE(dev); 1949 int tlen = dev->needed_tailroom; 1950 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1951 1952 rcu_read_unlock(); 1953 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1954 if (skb == NULL) 1955 return -ENOBUFS; 1956 /* FIXME: Save some space for broken drivers that write a hard 1957 * header at transmission time by themselves. PPP is the notable 1958 * one here. This should really be fixed at the driver level. 1959 */ 1960 skb_reserve(skb, reserved); 1961 skb_reset_network_header(skb); 1962 1963 /* Try to align data part correctly */ 1964 if (hhlen) { 1965 skb->data -= hhlen; 1966 skb->tail -= hhlen; 1967 if (len < hhlen) 1968 skb_reset_network_header(skb); 1969 } 1970 err = memcpy_from_msg(skb_put(skb, len), msg, len); 1971 if (err) 1972 goto out_free; 1973 goto retry; 1974 } 1975 1976 if (!dev_validate_header(dev, skb->data, len)) { 1977 err = -EINVAL; 1978 goto out_unlock; 1979 } 1980 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 1981 !packet_extra_vlan_len_allowed(dev, skb)) { 1982 err = -EMSGSIZE; 1983 goto out_unlock; 1984 } 1985 1986 sockc.tsflags = sk->sk_tsflags; 1987 if (msg->msg_controllen) { 1988 err = sock_cmsg_send(sk, msg, &sockc); 1989 if (unlikely(err)) 1990 goto out_unlock; 1991 } 1992 1993 skb->protocol = proto; 1994 skb->dev = dev; 1995 skb->priority = sk->sk_priority; 1996 skb->mark = sk->sk_mark; 1997 1998 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags); 1999 2000 if (unlikely(extra_len == 4)) 2001 skb->no_fcs = 1; 2002 2003 skb_probe_transport_header(skb, 0); 2004 2005 dev_queue_xmit(skb); 2006 rcu_read_unlock(); 2007 return len; 2008 2009 out_unlock: 2010 rcu_read_unlock(); 2011 out_free: 2012 kfree_skb(skb); 2013 return err; 2014 } 2015 2016 static unsigned int run_filter(struct sk_buff *skb, 2017 const struct sock *sk, 2018 unsigned int res) 2019 { 2020 struct sk_filter *filter; 2021 2022 rcu_read_lock(); 2023 filter = rcu_dereference(sk->sk_filter); 2024 if (filter != NULL) 2025 res = bpf_prog_run_clear_cb(filter->prog, skb); 2026 rcu_read_unlock(); 2027 2028 return res; 2029 } 2030 2031 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2032 size_t *len) 2033 { 2034 struct virtio_net_hdr vnet_hdr; 2035 2036 if (*len < sizeof(vnet_hdr)) 2037 return -EINVAL; 2038 *len -= sizeof(vnet_hdr); 2039 2040 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true)) 2041 return -EINVAL; 2042 2043 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 2044 } 2045 2046 /* 2047 * This function makes lazy skb cloning in hope that most of packets 2048 * are discarded by BPF. 2049 * 2050 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2051 * and skb->cb are mangled. It works because (and until) packets 2052 * falling here are owned by current CPU. Output packets are cloned 2053 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2054 * sequencially, so that if we return skb to original state on exit, 2055 * we will not harm anyone. 2056 */ 2057 2058 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2059 struct packet_type *pt, struct net_device *orig_dev) 2060 { 2061 struct sock *sk; 2062 struct sockaddr_ll *sll; 2063 struct packet_sock *po; 2064 u8 *skb_head = skb->data; 2065 int skb_len = skb->len; 2066 unsigned int snaplen, res; 2067 bool is_drop_n_account = false; 2068 2069 if (skb->pkt_type == PACKET_LOOPBACK) 2070 goto drop; 2071 2072 sk = pt->af_packet_priv; 2073 po = pkt_sk(sk); 2074 2075 if (!net_eq(dev_net(dev), sock_net(sk))) 2076 goto drop; 2077 2078 skb->dev = dev; 2079 2080 if (dev->header_ops) { 2081 /* The device has an explicit notion of ll header, 2082 * exported to higher levels. 2083 * 2084 * Otherwise, the device hides details of its frame 2085 * structure, so that corresponding packet head is 2086 * never delivered to user. 2087 */ 2088 if (sk->sk_type != SOCK_DGRAM) 2089 skb_push(skb, skb->data - skb_mac_header(skb)); 2090 else if (skb->pkt_type == PACKET_OUTGOING) { 2091 /* Special case: outgoing packets have ll header at head */ 2092 skb_pull(skb, skb_network_offset(skb)); 2093 } 2094 } 2095 2096 snaplen = skb->len; 2097 2098 res = run_filter(skb, sk, snaplen); 2099 if (!res) 2100 goto drop_n_restore; 2101 if (snaplen > res) 2102 snaplen = res; 2103 2104 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2105 goto drop_n_acct; 2106 2107 if (skb_shared(skb)) { 2108 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2109 if (nskb == NULL) 2110 goto drop_n_acct; 2111 2112 if (skb_head != skb->data) { 2113 skb->data = skb_head; 2114 skb->len = skb_len; 2115 } 2116 consume_skb(skb); 2117 skb = nskb; 2118 } 2119 2120 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2121 2122 sll = &PACKET_SKB_CB(skb)->sa.ll; 2123 sll->sll_hatype = dev->type; 2124 sll->sll_pkttype = skb->pkt_type; 2125 if (unlikely(po->origdev)) 2126 sll->sll_ifindex = orig_dev->ifindex; 2127 else 2128 sll->sll_ifindex = dev->ifindex; 2129 2130 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2131 2132 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2133 * Use their space for storing the original skb length. 2134 */ 2135 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2136 2137 if (pskb_trim(skb, snaplen)) 2138 goto drop_n_acct; 2139 2140 skb_set_owner_r(skb, sk); 2141 skb->dev = NULL; 2142 skb_dst_drop(skb); 2143 2144 /* drop conntrack reference */ 2145 nf_reset(skb); 2146 2147 spin_lock(&sk->sk_receive_queue.lock); 2148 po->stats.stats1.tp_packets++; 2149 sock_skb_set_dropcount(sk, skb); 2150 __skb_queue_tail(&sk->sk_receive_queue, skb); 2151 spin_unlock(&sk->sk_receive_queue.lock); 2152 sk->sk_data_ready(sk); 2153 return 0; 2154 2155 drop_n_acct: 2156 is_drop_n_account = true; 2157 spin_lock(&sk->sk_receive_queue.lock); 2158 po->stats.stats1.tp_drops++; 2159 atomic_inc(&sk->sk_drops); 2160 spin_unlock(&sk->sk_receive_queue.lock); 2161 2162 drop_n_restore: 2163 if (skb_head != skb->data && skb_shared(skb)) { 2164 skb->data = skb_head; 2165 skb->len = skb_len; 2166 } 2167 drop: 2168 if (!is_drop_n_account) 2169 consume_skb(skb); 2170 else 2171 kfree_skb(skb); 2172 return 0; 2173 } 2174 2175 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2176 struct packet_type *pt, struct net_device *orig_dev) 2177 { 2178 struct sock *sk; 2179 struct packet_sock *po; 2180 struct sockaddr_ll *sll; 2181 union tpacket_uhdr h; 2182 u8 *skb_head = skb->data; 2183 int skb_len = skb->len; 2184 unsigned int snaplen, res; 2185 unsigned long status = TP_STATUS_USER; 2186 unsigned short macoff, netoff, hdrlen; 2187 struct sk_buff *copy_skb = NULL; 2188 struct timespec ts; 2189 __u32 ts_status; 2190 bool is_drop_n_account = false; 2191 bool do_vnet = false; 2192 2193 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2194 * We may add members to them until current aligned size without forcing 2195 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2196 */ 2197 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2198 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2199 2200 if (skb->pkt_type == PACKET_LOOPBACK) 2201 goto drop; 2202 2203 sk = pt->af_packet_priv; 2204 po = pkt_sk(sk); 2205 2206 if (!net_eq(dev_net(dev), sock_net(sk))) 2207 goto drop; 2208 2209 if (dev->header_ops) { 2210 if (sk->sk_type != SOCK_DGRAM) 2211 skb_push(skb, skb->data - skb_mac_header(skb)); 2212 else if (skb->pkt_type == PACKET_OUTGOING) { 2213 /* Special case: outgoing packets have ll header at head */ 2214 skb_pull(skb, skb_network_offset(skb)); 2215 } 2216 } 2217 2218 snaplen = skb->len; 2219 2220 res = run_filter(skb, sk, snaplen); 2221 if (!res) 2222 goto drop_n_restore; 2223 2224 if (skb->ip_summed == CHECKSUM_PARTIAL) 2225 status |= TP_STATUS_CSUMNOTREADY; 2226 else if (skb->pkt_type != PACKET_OUTGOING && 2227 (skb->ip_summed == CHECKSUM_COMPLETE || 2228 skb_csum_unnecessary(skb))) 2229 status |= TP_STATUS_CSUM_VALID; 2230 2231 if (snaplen > res) 2232 snaplen = res; 2233 2234 if (sk->sk_type == SOCK_DGRAM) { 2235 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2236 po->tp_reserve; 2237 } else { 2238 unsigned int maclen = skb_network_offset(skb); 2239 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2240 (maclen < 16 ? 16 : maclen)) + 2241 po->tp_reserve; 2242 if (po->has_vnet_hdr) { 2243 netoff += sizeof(struct virtio_net_hdr); 2244 do_vnet = true; 2245 } 2246 macoff = netoff - maclen; 2247 } 2248 if (po->tp_version <= TPACKET_V2) { 2249 if (macoff + snaplen > po->rx_ring.frame_size) { 2250 if (po->copy_thresh && 2251 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2252 if (skb_shared(skb)) { 2253 copy_skb = skb_clone(skb, GFP_ATOMIC); 2254 } else { 2255 copy_skb = skb_get(skb); 2256 skb_head = skb->data; 2257 } 2258 if (copy_skb) 2259 skb_set_owner_r(copy_skb, sk); 2260 } 2261 snaplen = po->rx_ring.frame_size - macoff; 2262 if ((int)snaplen < 0) { 2263 snaplen = 0; 2264 do_vnet = false; 2265 } 2266 } 2267 } else if (unlikely(macoff + snaplen > 2268 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2269 u32 nval; 2270 2271 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2272 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2273 snaplen, nval, macoff); 2274 snaplen = nval; 2275 if (unlikely((int)snaplen < 0)) { 2276 snaplen = 0; 2277 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2278 do_vnet = false; 2279 } 2280 } 2281 spin_lock(&sk->sk_receive_queue.lock); 2282 h.raw = packet_current_rx_frame(po, skb, 2283 TP_STATUS_KERNEL, (macoff+snaplen)); 2284 if (!h.raw) 2285 goto drop_n_account; 2286 if (po->tp_version <= TPACKET_V2) { 2287 packet_increment_rx_head(po, &po->rx_ring); 2288 /* 2289 * LOSING will be reported till you read the stats, 2290 * because it's COR - Clear On Read. 2291 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2292 * at packet level. 2293 */ 2294 if (po->stats.stats1.tp_drops) 2295 status |= TP_STATUS_LOSING; 2296 } 2297 po->stats.stats1.tp_packets++; 2298 if (copy_skb) { 2299 status |= TP_STATUS_COPY; 2300 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2301 } 2302 spin_unlock(&sk->sk_receive_queue.lock); 2303 2304 if (do_vnet) { 2305 if (virtio_net_hdr_from_skb(skb, h.raw + macoff - 2306 sizeof(struct virtio_net_hdr), 2307 vio_le(), true)) { 2308 spin_lock(&sk->sk_receive_queue.lock); 2309 goto drop_n_account; 2310 } 2311 } 2312 2313 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2314 2315 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 2316 getnstimeofday(&ts); 2317 2318 status |= ts_status; 2319 2320 switch (po->tp_version) { 2321 case TPACKET_V1: 2322 h.h1->tp_len = skb->len; 2323 h.h1->tp_snaplen = snaplen; 2324 h.h1->tp_mac = macoff; 2325 h.h1->tp_net = netoff; 2326 h.h1->tp_sec = ts.tv_sec; 2327 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2328 hdrlen = sizeof(*h.h1); 2329 break; 2330 case TPACKET_V2: 2331 h.h2->tp_len = skb->len; 2332 h.h2->tp_snaplen = snaplen; 2333 h.h2->tp_mac = macoff; 2334 h.h2->tp_net = netoff; 2335 h.h2->tp_sec = ts.tv_sec; 2336 h.h2->tp_nsec = ts.tv_nsec; 2337 if (skb_vlan_tag_present(skb)) { 2338 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2339 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2340 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2341 } else { 2342 h.h2->tp_vlan_tci = 0; 2343 h.h2->tp_vlan_tpid = 0; 2344 } 2345 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2346 hdrlen = sizeof(*h.h2); 2347 break; 2348 case TPACKET_V3: 2349 /* tp_nxt_offset,vlan are already populated above. 2350 * So DONT clear those fields here 2351 */ 2352 h.h3->tp_status |= status; 2353 h.h3->tp_len = skb->len; 2354 h.h3->tp_snaplen = snaplen; 2355 h.h3->tp_mac = macoff; 2356 h.h3->tp_net = netoff; 2357 h.h3->tp_sec = ts.tv_sec; 2358 h.h3->tp_nsec = ts.tv_nsec; 2359 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2360 hdrlen = sizeof(*h.h3); 2361 break; 2362 default: 2363 BUG(); 2364 } 2365 2366 sll = h.raw + TPACKET_ALIGN(hdrlen); 2367 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2368 sll->sll_family = AF_PACKET; 2369 sll->sll_hatype = dev->type; 2370 sll->sll_protocol = skb->protocol; 2371 sll->sll_pkttype = skb->pkt_type; 2372 if (unlikely(po->origdev)) 2373 sll->sll_ifindex = orig_dev->ifindex; 2374 else 2375 sll->sll_ifindex = dev->ifindex; 2376 2377 smp_mb(); 2378 2379 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2380 if (po->tp_version <= TPACKET_V2) { 2381 u8 *start, *end; 2382 2383 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2384 macoff + snaplen); 2385 2386 for (start = h.raw; start < end; start += PAGE_SIZE) 2387 flush_dcache_page(pgv_to_page(start)); 2388 } 2389 smp_wmb(); 2390 #endif 2391 2392 if (po->tp_version <= TPACKET_V2) { 2393 __packet_set_status(po, h.raw, status); 2394 sk->sk_data_ready(sk); 2395 } else { 2396 prb_clear_blk_fill_status(&po->rx_ring); 2397 } 2398 2399 drop_n_restore: 2400 if (skb_head != skb->data && skb_shared(skb)) { 2401 skb->data = skb_head; 2402 skb->len = skb_len; 2403 } 2404 drop: 2405 if (!is_drop_n_account) 2406 consume_skb(skb); 2407 else 2408 kfree_skb(skb); 2409 return 0; 2410 2411 drop_n_account: 2412 is_drop_n_account = true; 2413 po->stats.stats1.tp_drops++; 2414 spin_unlock(&sk->sk_receive_queue.lock); 2415 2416 sk->sk_data_ready(sk); 2417 kfree_skb(copy_skb); 2418 goto drop_n_restore; 2419 } 2420 2421 static void tpacket_destruct_skb(struct sk_buff *skb) 2422 { 2423 struct packet_sock *po = pkt_sk(skb->sk); 2424 2425 if (likely(po->tx_ring.pg_vec)) { 2426 void *ph; 2427 __u32 ts; 2428 2429 ph = skb_shinfo(skb)->destructor_arg; 2430 packet_dec_pending(&po->tx_ring); 2431 2432 ts = __packet_set_timestamp(po, ph, skb); 2433 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2434 } 2435 2436 sock_wfree(skb); 2437 } 2438 2439 static void tpacket_set_protocol(const struct net_device *dev, 2440 struct sk_buff *skb) 2441 { 2442 if (dev->type == ARPHRD_ETHER) { 2443 skb_reset_mac_header(skb); 2444 skb->protocol = eth_hdr(skb)->h_proto; 2445 } 2446 } 2447 2448 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2449 { 2450 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2451 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2452 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2453 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2454 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2455 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2456 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2457 2458 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2459 return -EINVAL; 2460 2461 return 0; 2462 } 2463 2464 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2465 struct virtio_net_hdr *vnet_hdr) 2466 { 2467 if (*len < sizeof(*vnet_hdr)) 2468 return -EINVAL; 2469 *len -= sizeof(*vnet_hdr); 2470 2471 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2472 return -EFAULT; 2473 2474 return __packet_snd_vnet_parse(vnet_hdr, *len); 2475 } 2476 2477 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2478 void *frame, struct net_device *dev, void *data, int tp_len, 2479 __be16 proto, unsigned char *addr, int hlen, int copylen, 2480 const struct sockcm_cookie *sockc) 2481 { 2482 union tpacket_uhdr ph; 2483 int to_write, offset, len, nr_frags, len_max; 2484 struct socket *sock = po->sk.sk_socket; 2485 struct page *page; 2486 int err; 2487 2488 ph.raw = frame; 2489 2490 skb->protocol = proto; 2491 skb->dev = dev; 2492 skb->priority = po->sk.sk_priority; 2493 skb->mark = po->sk.sk_mark; 2494 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags); 2495 skb_shinfo(skb)->destructor_arg = ph.raw; 2496 2497 skb_reserve(skb, hlen); 2498 skb_reset_network_header(skb); 2499 2500 to_write = tp_len; 2501 2502 if (sock->type == SOCK_DGRAM) { 2503 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2504 NULL, tp_len); 2505 if (unlikely(err < 0)) 2506 return -EINVAL; 2507 } else if (copylen) { 2508 int hdrlen = min_t(int, copylen, tp_len); 2509 2510 skb_push(skb, dev->hard_header_len); 2511 skb_put(skb, copylen - dev->hard_header_len); 2512 err = skb_store_bits(skb, 0, data, hdrlen); 2513 if (unlikely(err)) 2514 return err; 2515 if (!dev_validate_header(dev, skb->data, hdrlen)) 2516 return -EINVAL; 2517 if (!skb->protocol) 2518 tpacket_set_protocol(dev, skb); 2519 2520 data += hdrlen; 2521 to_write -= hdrlen; 2522 } 2523 2524 offset = offset_in_page(data); 2525 len_max = PAGE_SIZE - offset; 2526 len = ((to_write > len_max) ? len_max : to_write); 2527 2528 skb->data_len = to_write; 2529 skb->len += to_write; 2530 skb->truesize += to_write; 2531 refcount_add(to_write, &po->sk.sk_wmem_alloc); 2532 2533 while (likely(to_write)) { 2534 nr_frags = skb_shinfo(skb)->nr_frags; 2535 2536 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2537 pr_err("Packet exceed the number of skb frags(%lu)\n", 2538 MAX_SKB_FRAGS); 2539 return -EFAULT; 2540 } 2541 2542 page = pgv_to_page(data); 2543 data += len; 2544 flush_dcache_page(page); 2545 get_page(page); 2546 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2547 to_write -= len; 2548 offset = 0; 2549 len_max = PAGE_SIZE; 2550 len = ((to_write > len_max) ? len_max : to_write); 2551 } 2552 2553 skb_probe_transport_header(skb, 0); 2554 2555 return tp_len; 2556 } 2557 2558 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2559 int size_max, void **data) 2560 { 2561 union tpacket_uhdr ph; 2562 int tp_len, off; 2563 2564 ph.raw = frame; 2565 2566 switch (po->tp_version) { 2567 case TPACKET_V3: 2568 if (ph.h3->tp_next_offset != 0) { 2569 pr_warn_once("variable sized slot not supported"); 2570 return -EINVAL; 2571 } 2572 tp_len = ph.h3->tp_len; 2573 break; 2574 case TPACKET_V2: 2575 tp_len = ph.h2->tp_len; 2576 break; 2577 default: 2578 tp_len = ph.h1->tp_len; 2579 break; 2580 } 2581 if (unlikely(tp_len > size_max)) { 2582 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2583 return -EMSGSIZE; 2584 } 2585 2586 if (unlikely(po->tp_tx_has_off)) { 2587 int off_min, off_max; 2588 2589 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2590 off_max = po->tx_ring.frame_size - tp_len; 2591 if (po->sk.sk_type == SOCK_DGRAM) { 2592 switch (po->tp_version) { 2593 case TPACKET_V3: 2594 off = ph.h3->tp_net; 2595 break; 2596 case TPACKET_V2: 2597 off = ph.h2->tp_net; 2598 break; 2599 default: 2600 off = ph.h1->tp_net; 2601 break; 2602 } 2603 } else { 2604 switch (po->tp_version) { 2605 case TPACKET_V3: 2606 off = ph.h3->tp_mac; 2607 break; 2608 case TPACKET_V2: 2609 off = ph.h2->tp_mac; 2610 break; 2611 default: 2612 off = ph.h1->tp_mac; 2613 break; 2614 } 2615 } 2616 if (unlikely((off < off_min) || (off_max < off))) 2617 return -EINVAL; 2618 } else { 2619 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2620 } 2621 2622 *data = frame + off; 2623 return tp_len; 2624 } 2625 2626 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2627 { 2628 struct sk_buff *skb; 2629 struct net_device *dev; 2630 struct virtio_net_hdr *vnet_hdr = NULL; 2631 struct sockcm_cookie sockc; 2632 __be16 proto; 2633 int err, reserve = 0; 2634 void *ph; 2635 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2636 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2637 int tp_len, size_max; 2638 unsigned char *addr; 2639 void *data; 2640 int len_sum = 0; 2641 int status = TP_STATUS_AVAILABLE; 2642 int hlen, tlen, copylen = 0; 2643 2644 mutex_lock(&po->pg_vec_lock); 2645 2646 if (likely(saddr == NULL)) { 2647 dev = packet_cached_dev_get(po); 2648 proto = po->num; 2649 addr = NULL; 2650 } else { 2651 err = -EINVAL; 2652 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2653 goto out; 2654 if (msg->msg_namelen < (saddr->sll_halen 2655 + offsetof(struct sockaddr_ll, 2656 sll_addr))) 2657 goto out; 2658 proto = saddr->sll_protocol; 2659 addr = saddr->sll_addr; 2660 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2661 } 2662 2663 err = -ENXIO; 2664 if (unlikely(dev == NULL)) 2665 goto out; 2666 err = -ENETDOWN; 2667 if (unlikely(!(dev->flags & IFF_UP))) 2668 goto out_put; 2669 2670 sockc.tsflags = po->sk.sk_tsflags; 2671 if (msg->msg_controllen) { 2672 err = sock_cmsg_send(&po->sk, msg, &sockc); 2673 if (unlikely(err)) 2674 goto out_put; 2675 } 2676 2677 if (po->sk.sk_socket->type == SOCK_RAW) 2678 reserve = dev->hard_header_len; 2679 size_max = po->tx_ring.frame_size 2680 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2681 2682 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) 2683 size_max = dev->mtu + reserve + VLAN_HLEN; 2684 2685 do { 2686 ph = packet_current_frame(po, &po->tx_ring, 2687 TP_STATUS_SEND_REQUEST); 2688 if (unlikely(ph == NULL)) { 2689 if (need_wait && need_resched()) 2690 schedule(); 2691 continue; 2692 } 2693 2694 skb = NULL; 2695 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2696 if (tp_len < 0) 2697 goto tpacket_error; 2698 2699 status = TP_STATUS_SEND_REQUEST; 2700 hlen = LL_RESERVED_SPACE(dev); 2701 tlen = dev->needed_tailroom; 2702 if (po->has_vnet_hdr) { 2703 vnet_hdr = data; 2704 data += sizeof(*vnet_hdr); 2705 tp_len -= sizeof(*vnet_hdr); 2706 if (tp_len < 0 || 2707 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2708 tp_len = -EINVAL; 2709 goto tpacket_error; 2710 } 2711 copylen = __virtio16_to_cpu(vio_le(), 2712 vnet_hdr->hdr_len); 2713 } 2714 copylen = max_t(int, copylen, dev->hard_header_len); 2715 skb = sock_alloc_send_skb(&po->sk, 2716 hlen + tlen + sizeof(struct sockaddr_ll) + 2717 (copylen - dev->hard_header_len), 2718 !need_wait, &err); 2719 2720 if (unlikely(skb == NULL)) { 2721 /* we assume the socket was initially writeable ... */ 2722 if (likely(len_sum > 0)) 2723 err = len_sum; 2724 goto out_status; 2725 } 2726 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2727 addr, hlen, copylen, &sockc); 2728 if (likely(tp_len >= 0) && 2729 tp_len > dev->mtu + reserve && 2730 !po->has_vnet_hdr && 2731 !packet_extra_vlan_len_allowed(dev, skb)) 2732 tp_len = -EMSGSIZE; 2733 2734 if (unlikely(tp_len < 0)) { 2735 tpacket_error: 2736 if (po->tp_loss) { 2737 __packet_set_status(po, ph, 2738 TP_STATUS_AVAILABLE); 2739 packet_increment_head(&po->tx_ring); 2740 kfree_skb(skb); 2741 continue; 2742 } else { 2743 status = TP_STATUS_WRONG_FORMAT; 2744 err = tp_len; 2745 goto out_status; 2746 } 2747 } 2748 2749 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr, 2750 vio_le())) { 2751 tp_len = -EINVAL; 2752 goto tpacket_error; 2753 } 2754 2755 skb->destructor = tpacket_destruct_skb; 2756 __packet_set_status(po, ph, TP_STATUS_SENDING); 2757 packet_inc_pending(&po->tx_ring); 2758 2759 status = TP_STATUS_SEND_REQUEST; 2760 err = po->xmit(skb); 2761 if (unlikely(err > 0)) { 2762 err = net_xmit_errno(err); 2763 if (err && __packet_get_status(po, ph) == 2764 TP_STATUS_AVAILABLE) { 2765 /* skb was destructed already */ 2766 skb = NULL; 2767 goto out_status; 2768 } 2769 /* 2770 * skb was dropped but not destructed yet; 2771 * let's treat it like congestion or err < 0 2772 */ 2773 err = 0; 2774 } 2775 packet_increment_head(&po->tx_ring); 2776 len_sum += tp_len; 2777 } while (likely((ph != NULL) || 2778 /* Note: packet_read_pending() might be slow if we have 2779 * to call it as it's per_cpu variable, but in fast-path 2780 * we already short-circuit the loop with the first 2781 * condition, and luckily don't have to go that path 2782 * anyway. 2783 */ 2784 (need_wait && packet_read_pending(&po->tx_ring)))); 2785 2786 err = len_sum; 2787 goto out_put; 2788 2789 out_status: 2790 __packet_set_status(po, ph, status); 2791 kfree_skb(skb); 2792 out_put: 2793 dev_put(dev); 2794 out: 2795 mutex_unlock(&po->pg_vec_lock); 2796 return err; 2797 } 2798 2799 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2800 size_t reserve, size_t len, 2801 size_t linear, int noblock, 2802 int *err) 2803 { 2804 struct sk_buff *skb; 2805 2806 /* Under a page? Don't bother with paged skb. */ 2807 if (prepad + len < PAGE_SIZE || !linear) 2808 linear = len; 2809 2810 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2811 err, 0); 2812 if (!skb) 2813 return NULL; 2814 2815 skb_reserve(skb, reserve); 2816 skb_put(skb, linear); 2817 skb->data_len = len - linear; 2818 skb->len += len - linear; 2819 2820 return skb; 2821 } 2822 2823 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2824 { 2825 struct sock *sk = sock->sk; 2826 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2827 struct sk_buff *skb; 2828 struct net_device *dev; 2829 __be16 proto; 2830 unsigned char *addr; 2831 int err, reserve = 0; 2832 struct sockcm_cookie sockc; 2833 struct virtio_net_hdr vnet_hdr = { 0 }; 2834 int offset = 0; 2835 struct packet_sock *po = pkt_sk(sk); 2836 bool has_vnet_hdr = false; 2837 int hlen, tlen, linear; 2838 int extra_len = 0; 2839 2840 /* 2841 * Get and verify the address. 2842 */ 2843 2844 if (likely(saddr == NULL)) { 2845 dev = packet_cached_dev_get(po); 2846 proto = po->num; 2847 addr = NULL; 2848 } else { 2849 err = -EINVAL; 2850 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2851 goto out; 2852 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2853 goto out; 2854 proto = saddr->sll_protocol; 2855 addr = saddr->sll_addr; 2856 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2857 } 2858 2859 err = -ENXIO; 2860 if (unlikely(dev == NULL)) 2861 goto out_unlock; 2862 err = -ENETDOWN; 2863 if (unlikely(!(dev->flags & IFF_UP))) 2864 goto out_unlock; 2865 2866 sockc.tsflags = sk->sk_tsflags; 2867 sockc.mark = sk->sk_mark; 2868 if (msg->msg_controllen) { 2869 err = sock_cmsg_send(sk, msg, &sockc); 2870 if (unlikely(err)) 2871 goto out_unlock; 2872 } 2873 2874 if (sock->type == SOCK_RAW) 2875 reserve = dev->hard_header_len; 2876 if (po->has_vnet_hdr) { 2877 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2878 if (err) 2879 goto out_unlock; 2880 has_vnet_hdr = true; 2881 } 2882 2883 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2884 if (!netif_supports_nofcs(dev)) { 2885 err = -EPROTONOSUPPORT; 2886 goto out_unlock; 2887 } 2888 extra_len = 4; /* We're doing our own CRC */ 2889 } 2890 2891 err = -EMSGSIZE; 2892 if (!vnet_hdr.gso_type && 2893 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2894 goto out_unlock; 2895 2896 err = -ENOBUFS; 2897 hlen = LL_RESERVED_SPACE(dev); 2898 tlen = dev->needed_tailroom; 2899 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); 2900 linear = max(linear, min_t(int, len, dev->hard_header_len)); 2901 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, 2902 msg->msg_flags & MSG_DONTWAIT, &err); 2903 if (skb == NULL) 2904 goto out_unlock; 2905 2906 skb_set_network_header(skb, reserve); 2907 2908 err = -EINVAL; 2909 if (sock->type == SOCK_DGRAM) { 2910 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 2911 if (unlikely(offset < 0)) 2912 goto out_free; 2913 } 2914 2915 /* Returns -EFAULT on error */ 2916 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 2917 if (err) 2918 goto out_free; 2919 2920 if (sock->type == SOCK_RAW && 2921 !dev_validate_header(dev, skb->data, len)) { 2922 err = -EINVAL; 2923 goto out_free; 2924 } 2925 2926 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags); 2927 2928 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 2929 !packet_extra_vlan_len_allowed(dev, skb)) { 2930 err = -EMSGSIZE; 2931 goto out_free; 2932 } 2933 2934 skb->protocol = proto; 2935 skb->dev = dev; 2936 skb->priority = sk->sk_priority; 2937 skb->mark = sockc.mark; 2938 2939 if (has_vnet_hdr) { 2940 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 2941 if (err) 2942 goto out_free; 2943 len += sizeof(vnet_hdr); 2944 } 2945 2946 skb_probe_transport_header(skb, reserve); 2947 2948 if (unlikely(extra_len == 4)) 2949 skb->no_fcs = 1; 2950 2951 err = po->xmit(skb); 2952 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2953 goto out_unlock; 2954 2955 dev_put(dev); 2956 2957 return len; 2958 2959 out_free: 2960 kfree_skb(skb); 2961 out_unlock: 2962 if (dev) 2963 dev_put(dev); 2964 out: 2965 return err; 2966 } 2967 2968 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2969 { 2970 struct sock *sk = sock->sk; 2971 struct packet_sock *po = pkt_sk(sk); 2972 2973 if (po->tx_ring.pg_vec) 2974 return tpacket_snd(po, msg); 2975 else 2976 return packet_snd(sock, msg, len); 2977 } 2978 2979 /* 2980 * Close a PACKET socket. This is fairly simple. We immediately go 2981 * to 'closed' state and remove our protocol entry in the device list. 2982 */ 2983 2984 static int packet_release(struct socket *sock) 2985 { 2986 struct sock *sk = sock->sk; 2987 struct packet_sock *po; 2988 struct packet_fanout *f; 2989 struct net *net; 2990 union tpacket_req_u req_u; 2991 2992 if (!sk) 2993 return 0; 2994 2995 net = sock_net(sk); 2996 po = pkt_sk(sk); 2997 2998 mutex_lock(&net->packet.sklist_lock); 2999 sk_del_node_init_rcu(sk); 3000 mutex_unlock(&net->packet.sklist_lock); 3001 3002 preempt_disable(); 3003 sock_prot_inuse_add(net, sk->sk_prot, -1); 3004 preempt_enable(); 3005 3006 spin_lock(&po->bind_lock); 3007 unregister_prot_hook(sk, false); 3008 packet_cached_dev_reset(po); 3009 3010 if (po->prot_hook.dev) { 3011 dev_put(po->prot_hook.dev); 3012 po->prot_hook.dev = NULL; 3013 } 3014 spin_unlock(&po->bind_lock); 3015 3016 packet_flush_mclist(sk); 3017 3018 lock_sock(sk); 3019 if (po->rx_ring.pg_vec) { 3020 memset(&req_u, 0, sizeof(req_u)); 3021 packet_set_ring(sk, &req_u, 1, 0); 3022 } 3023 3024 if (po->tx_ring.pg_vec) { 3025 memset(&req_u, 0, sizeof(req_u)); 3026 packet_set_ring(sk, &req_u, 1, 1); 3027 } 3028 release_sock(sk); 3029 3030 f = fanout_release(sk); 3031 3032 synchronize_net(); 3033 3034 if (f) { 3035 kfree(po->rollover); 3036 fanout_release_data(f); 3037 kfree(f); 3038 } 3039 /* 3040 * Now the socket is dead. No more input will appear. 3041 */ 3042 sock_orphan(sk); 3043 sock->sk = NULL; 3044 3045 /* Purge queues */ 3046 3047 skb_queue_purge(&sk->sk_receive_queue); 3048 packet_free_pending(po); 3049 sk_refcnt_debug_release(sk); 3050 3051 sock_put(sk); 3052 return 0; 3053 } 3054 3055 /* 3056 * Attach a packet hook. 3057 */ 3058 3059 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3060 __be16 proto) 3061 { 3062 struct packet_sock *po = pkt_sk(sk); 3063 struct net_device *dev_curr; 3064 __be16 proto_curr; 3065 bool need_rehook; 3066 struct net_device *dev = NULL; 3067 int ret = 0; 3068 bool unlisted = false; 3069 3070 lock_sock(sk); 3071 spin_lock(&po->bind_lock); 3072 rcu_read_lock(); 3073 3074 if (po->fanout) { 3075 ret = -EINVAL; 3076 goto out_unlock; 3077 } 3078 3079 if (name) { 3080 dev = dev_get_by_name_rcu(sock_net(sk), name); 3081 if (!dev) { 3082 ret = -ENODEV; 3083 goto out_unlock; 3084 } 3085 } else if (ifindex) { 3086 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3087 if (!dev) { 3088 ret = -ENODEV; 3089 goto out_unlock; 3090 } 3091 } 3092 3093 if (dev) 3094 dev_hold(dev); 3095 3096 proto_curr = po->prot_hook.type; 3097 dev_curr = po->prot_hook.dev; 3098 3099 need_rehook = proto_curr != proto || dev_curr != dev; 3100 3101 if (need_rehook) { 3102 if (po->running) { 3103 rcu_read_unlock(); 3104 /* prevents packet_notifier() from calling 3105 * register_prot_hook() 3106 */ 3107 po->num = 0; 3108 __unregister_prot_hook(sk, true); 3109 rcu_read_lock(); 3110 dev_curr = po->prot_hook.dev; 3111 if (dev) 3112 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3113 dev->ifindex); 3114 } 3115 3116 BUG_ON(po->running); 3117 po->num = proto; 3118 po->prot_hook.type = proto; 3119 3120 if (unlikely(unlisted)) { 3121 dev_put(dev); 3122 po->prot_hook.dev = NULL; 3123 po->ifindex = -1; 3124 packet_cached_dev_reset(po); 3125 } else { 3126 po->prot_hook.dev = dev; 3127 po->ifindex = dev ? dev->ifindex : 0; 3128 packet_cached_dev_assign(po, dev); 3129 } 3130 } 3131 if (dev_curr) 3132 dev_put(dev_curr); 3133 3134 if (proto == 0 || !need_rehook) 3135 goto out_unlock; 3136 3137 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3138 register_prot_hook(sk); 3139 } else { 3140 sk->sk_err = ENETDOWN; 3141 if (!sock_flag(sk, SOCK_DEAD)) 3142 sk->sk_error_report(sk); 3143 } 3144 3145 out_unlock: 3146 rcu_read_unlock(); 3147 spin_unlock(&po->bind_lock); 3148 release_sock(sk); 3149 return ret; 3150 } 3151 3152 /* 3153 * Bind a packet socket to a device 3154 */ 3155 3156 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3157 int addr_len) 3158 { 3159 struct sock *sk = sock->sk; 3160 char name[sizeof(uaddr->sa_data) + 1]; 3161 3162 /* 3163 * Check legality 3164 */ 3165 3166 if (addr_len != sizeof(struct sockaddr)) 3167 return -EINVAL; 3168 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be 3169 * zero-terminated. 3170 */ 3171 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); 3172 name[sizeof(uaddr->sa_data)] = 0; 3173 3174 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3175 } 3176 3177 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3178 { 3179 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3180 struct sock *sk = sock->sk; 3181 3182 /* 3183 * Check legality 3184 */ 3185 3186 if (addr_len < sizeof(struct sockaddr_ll)) 3187 return -EINVAL; 3188 if (sll->sll_family != AF_PACKET) 3189 return -EINVAL; 3190 3191 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3192 sll->sll_protocol ? : pkt_sk(sk)->num); 3193 } 3194 3195 static struct proto packet_proto = { 3196 .name = "PACKET", 3197 .owner = THIS_MODULE, 3198 .obj_size = sizeof(struct packet_sock), 3199 }; 3200 3201 /* 3202 * Create a packet of type SOCK_PACKET. 3203 */ 3204 3205 static int packet_create(struct net *net, struct socket *sock, int protocol, 3206 int kern) 3207 { 3208 struct sock *sk; 3209 struct packet_sock *po; 3210 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3211 int err; 3212 3213 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3214 return -EPERM; 3215 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3216 sock->type != SOCK_PACKET) 3217 return -ESOCKTNOSUPPORT; 3218 3219 sock->state = SS_UNCONNECTED; 3220 3221 err = -ENOBUFS; 3222 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3223 if (sk == NULL) 3224 goto out; 3225 3226 sock->ops = &packet_ops; 3227 if (sock->type == SOCK_PACKET) 3228 sock->ops = &packet_ops_spkt; 3229 3230 sock_init_data(sock, sk); 3231 3232 po = pkt_sk(sk); 3233 sk->sk_family = PF_PACKET; 3234 po->num = proto; 3235 po->xmit = dev_queue_xmit; 3236 3237 err = packet_alloc_pending(po); 3238 if (err) 3239 goto out2; 3240 3241 packet_cached_dev_reset(po); 3242 3243 sk->sk_destruct = packet_sock_destruct; 3244 sk_refcnt_debug_inc(sk); 3245 3246 /* 3247 * Attach a protocol block 3248 */ 3249 3250 spin_lock_init(&po->bind_lock); 3251 mutex_init(&po->pg_vec_lock); 3252 po->rollover = NULL; 3253 po->prot_hook.func = packet_rcv; 3254 3255 if (sock->type == SOCK_PACKET) 3256 po->prot_hook.func = packet_rcv_spkt; 3257 3258 po->prot_hook.af_packet_priv = sk; 3259 3260 if (proto) { 3261 po->prot_hook.type = proto; 3262 __register_prot_hook(sk); 3263 } 3264 3265 mutex_lock(&net->packet.sklist_lock); 3266 sk_add_node_rcu(sk, &net->packet.sklist); 3267 mutex_unlock(&net->packet.sklist_lock); 3268 3269 preempt_disable(); 3270 sock_prot_inuse_add(net, &packet_proto, 1); 3271 preempt_enable(); 3272 3273 return 0; 3274 out2: 3275 sk_free(sk); 3276 out: 3277 return err; 3278 } 3279 3280 /* 3281 * Pull a packet from our receive queue and hand it to the user. 3282 * If necessary we block. 3283 */ 3284 3285 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3286 int flags) 3287 { 3288 struct sock *sk = sock->sk; 3289 struct sk_buff *skb; 3290 int copied, err; 3291 int vnet_hdr_len = 0; 3292 unsigned int origlen = 0; 3293 3294 err = -EINVAL; 3295 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3296 goto out; 3297 3298 #if 0 3299 /* What error should we return now? EUNATTACH? */ 3300 if (pkt_sk(sk)->ifindex < 0) 3301 return -ENODEV; 3302 #endif 3303 3304 if (flags & MSG_ERRQUEUE) { 3305 err = sock_recv_errqueue(sk, msg, len, 3306 SOL_PACKET, PACKET_TX_TIMESTAMP); 3307 goto out; 3308 } 3309 3310 /* 3311 * Call the generic datagram receiver. This handles all sorts 3312 * of horrible races and re-entrancy so we can forget about it 3313 * in the protocol layers. 3314 * 3315 * Now it will return ENETDOWN, if device have just gone down, 3316 * but then it will block. 3317 */ 3318 3319 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 3320 3321 /* 3322 * An error occurred so return it. Because skb_recv_datagram() 3323 * handles the blocking we don't see and worry about blocking 3324 * retries. 3325 */ 3326 3327 if (skb == NULL) 3328 goto out; 3329 3330 if (pkt_sk(sk)->pressure) 3331 packet_rcv_has_room(pkt_sk(sk), NULL); 3332 3333 if (pkt_sk(sk)->has_vnet_hdr) { 3334 err = packet_rcv_vnet(msg, skb, &len); 3335 if (err) 3336 goto out_free; 3337 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3338 } 3339 3340 /* You lose any data beyond the buffer you gave. If it worries 3341 * a user program they can ask the device for its MTU 3342 * anyway. 3343 */ 3344 copied = skb->len; 3345 if (copied > len) { 3346 copied = len; 3347 msg->msg_flags |= MSG_TRUNC; 3348 } 3349 3350 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3351 if (err) 3352 goto out_free; 3353 3354 if (sock->type != SOCK_PACKET) { 3355 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3356 3357 /* Original length was stored in sockaddr_ll fields */ 3358 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3359 sll->sll_family = AF_PACKET; 3360 sll->sll_protocol = skb->protocol; 3361 } 3362 3363 sock_recv_ts_and_drops(msg, sk, skb); 3364 3365 if (msg->msg_name) { 3366 /* If the address length field is there to be filled 3367 * in, we fill it in now. 3368 */ 3369 if (sock->type == SOCK_PACKET) { 3370 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3371 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3372 } else { 3373 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3374 3375 msg->msg_namelen = sll->sll_halen + 3376 offsetof(struct sockaddr_ll, sll_addr); 3377 } 3378 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, 3379 msg->msg_namelen); 3380 } 3381 3382 if (pkt_sk(sk)->auxdata) { 3383 struct tpacket_auxdata aux; 3384 3385 aux.tp_status = TP_STATUS_USER; 3386 if (skb->ip_summed == CHECKSUM_PARTIAL) 3387 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3388 else if (skb->pkt_type != PACKET_OUTGOING && 3389 (skb->ip_summed == CHECKSUM_COMPLETE || 3390 skb_csum_unnecessary(skb))) 3391 aux.tp_status |= TP_STATUS_CSUM_VALID; 3392 3393 aux.tp_len = origlen; 3394 aux.tp_snaplen = skb->len; 3395 aux.tp_mac = 0; 3396 aux.tp_net = skb_network_offset(skb); 3397 if (skb_vlan_tag_present(skb)) { 3398 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3399 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3400 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3401 } else { 3402 aux.tp_vlan_tci = 0; 3403 aux.tp_vlan_tpid = 0; 3404 } 3405 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3406 } 3407 3408 /* 3409 * Free or return the buffer as appropriate. Again this 3410 * hides all the races and re-entrancy issues from us. 3411 */ 3412 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3413 3414 out_free: 3415 skb_free_datagram(sk, skb); 3416 out: 3417 return err; 3418 } 3419 3420 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3421 int peer) 3422 { 3423 struct net_device *dev; 3424 struct sock *sk = sock->sk; 3425 3426 if (peer) 3427 return -EOPNOTSUPP; 3428 3429 uaddr->sa_family = AF_PACKET; 3430 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3431 rcu_read_lock(); 3432 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3433 if (dev) 3434 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3435 rcu_read_unlock(); 3436 3437 return sizeof(*uaddr); 3438 } 3439 3440 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3441 int peer) 3442 { 3443 struct net_device *dev; 3444 struct sock *sk = sock->sk; 3445 struct packet_sock *po = pkt_sk(sk); 3446 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3447 3448 if (peer) 3449 return -EOPNOTSUPP; 3450 3451 sll->sll_family = AF_PACKET; 3452 sll->sll_ifindex = po->ifindex; 3453 sll->sll_protocol = po->num; 3454 sll->sll_pkttype = 0; 3455 rcu_read_lock(); 3456 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 3457 if (dev) { 3458 sll->sll_hatype = dev->type; 3459 sll->sll_halen = dev->addr_len; 3460 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3461 } else { 3462 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3463 sll->sll_halen = 0; 3464 } 3465 rcu_read_unlock(); 3466 3467 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3468 } 3469 3470 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3471 int what) 3472 { 3473 switch (i->type) { 3474 case PACKET_MR_MULTICAST: 3475 if (i->alen != dev->addr_len) 3476 return -EINVAL; 3477 if (what > 0) 3478 return dev_mc_add(dev, i->addr); 3479 else 3480 return dev_mc_del(dev, i->addr); 3481 break; 3482 case PACKET_MR_PROMISC: 3483 return dev_set_promiscuity(dev, what); 3484 case PACKET_MR_ALLMULTI: 3485 return dev_set_allmulti(dev, what); 3486 case PACKET_MR_UNICAST: 3487 if (i->alen != dev->addr_len) 3488 return -EINVAL; 3489 if (what > 0) 3490 return dev_uc_add(dev, i->addr); 3491 else 3492 return dev_uc_del(dev, i->addr); 3493 break; 3494 default: 3495 break; 3496 } 3497 return 0; 3498 } 3499 3500 static void packet_dev_mclist_delete(struct net_device *dev, 3501 struct packet_mclist **mlp) 3502 { 3503 struct packet_mclist *ml; 3504 3505 while ((ml = *mlp) != NULL) { 3506 if (ml->ifindex == dev->ifindex) { 3507 packet_dev_mc(dev, ml, -1); 3508 *mlp = ml->next; 3509 kfree(ml); 3510 } else 3511 mlp = &ml->next; 3512 } 3513 } 3514 3515 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3516 { 3517 struct packet_sock *po = pkt_sk(sk); 3518 struct packet_mclist *ml, *i; 3519 struct net_device *dev; 3520 int err; 3521 3522 rtnl_lock(); 3523 3524 err = -ENODEV; 3525 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3526 if (!dev) 3527 goto done; 3528 3529 err = -EINVAL; 3530 if (mreq->mr_alen > dev->addr_len) 3531 goto done; 3532 3533 err = -ENOBUFS; 3534 i = kmalloc(sizeof(*i), GFP_KERNEL); 3535 if (i == NULL) 3536 goto done; 3537 3538 err = 0; 3539 for (ml = po->mclist; ml; ml = ml->next) { 3540 if (ml->ifindex == mreq->mr_ifindex && 3541 ml->type == mreq->mr_type && 3542 ml->alen == mreq->mr_alen && 3543 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3544 ml->count++; 3545 /* Free the new element ... */ 3546 kfree(i); 3547 goto done; 3548 } 3549 } 3550 3551 i->type = mreq->mr_type; 3552 i->ifindex = mreq->mr_ifindex; 3553 i->alen = mreq->mr_alen; 3554 memcpy(i->addr, mreq->mr_address, i->alen); 3555 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3556 i->count = 1; 3557 i->next = po->mclist; 3558 po->mclist = i; 3559 err = packet_dev_mc(dev, i, 1); 3560 if (err) { 3561 po->mclist = i->next; 3562 kfree(i); 3563 } 3564 3565 done: 3566 rtnl_unlock(); 3567 return err; 3568 } 3569 3570 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3571 { 3572 struct packet_mclist *ml, **mlp; 3573 3574 rtnl_lock(); 3575 3576 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3577 if (ml->ifindex == mreq->mr_ifindex && 3578 ml->type == mreq->mr_type && 3579 ml->alen == mreq->mr_alen && 3580 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3581 if (--ml->count == 0) { 3582 struct net_device *dev; 3583 *mlp = ml->next; 3584 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3585 if (dev) 3586 packet_dev_mc(dev, ml, -1); 3587 kfree(ml); 3588 } 3589 break; 3590 } 3591 } 3592 rtnl_unlock(); 3593 return 0; 3594 } 3595 3596 static void packet_flush_mclist(struct sock *sk) 3597 { 3598 struct packet_sock *po = pkt_sk(sk); 3599 struct packet_mclist *ml; 3600 3601 if (!po->mclist) 3602 return; 3603 3604 rtnl_lock(); 3605 while ((ml = po->mclist) != NULL) { 3606 struct net_device *dev; 3607 3608 po->mclist = ml->next; 3609 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3610 if (dev != NULL) 3611 packet_dev_mc(dev, ml, -1); 3612 kfree(ml); 3613 } 3614 rtnl_unlock(); 3615 } 3616 3617 static int 3618 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3619 { 3620 struct sock *sk = sock->sk; 3621 struct packet_sock *po = pkt_sk(sk); 3622 int ret; 3623 3624 if (level != SOL_PACKET) 3625 return -ENOPROTOOPT; 3626 3627 switch (optname) { 3628 case PACKET_ADD_MEMBERSHIP: 3629 case PACKET_DROP_MEMBERSHIP: 3630 { 3631 struct packet_mreq_max mreq; 3632 int len = optlen; 3633 memset(&mreq, 0, sizeof(mreq)); 3634 if (len < sizeof(struct packet_mreq)) 3635 return -EINVAL; 3636 if (len > sizeof(mreq)) 3637 len = sizeof(mreq); 3638 if (copy_from_user(&mreq, optval, len)) 3639 return -EFAULT; 3640 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3641 return -EINVAL; 3642 if (optname == PACKET_ADD_MEMBERSHIP) 3643 ret = packet_mc_add(sk, &mreq); 3644 else 3645 ret = packet_mc_drop(sk, &mreq); 3646 return ret; 3647 } 3648 3649 case PACKET_RX_RING: 3650 case PACKET_TX_RING: 3651 { 3652 union tpacket_req_u req_u; 3653 int len; 3654 3655 lock_sock(sk); 3656 switch (po->tp_version) { 3657 case TPACKET_V1: 3658 case TPACKET_V2: 3659 len = sizeof(req_u.req); 3660 break; 3661 case TPACKET_V3: 3662 default: 3663 len = sizeof(req_u.req3); 3664 break; 3665 } 3666 if (optlen < len) { 3667 ret = -EINVAL; 3668 } else { 3669 if (copy_from_user(&req_u.req, optval, len)) 3670 ret = -EFAULT; 3671 else 3672 ret = packet_set_ring(sk, &req_u, 0, 3673 optname == PACKET_TX_RING); 3674 } 3675 release_sock(sk); 3676 return ret; 3677 } 3678 case PACKET_COPY_THRESH: 3679 { 3680 int val; 3681 3682 if (optlen != sizeof(val)) 3683 return -EINVAL; 3684 if (copy_from_user(&val, optval, sizeof(val))) 3685 return -EFAULT; 3686 3687 pkt_sk(sk)->copy_thresh = val; 3688 return 0; 3689 } 3690 case PACKET_VERSION: 3691 { 3692 int val; 3693 3694 if (optlen != sizeof(val)) 3695 return -EINVAL; 3696 if (copy_from_user(&val, optval, sizeof(val))) 3697 return -EFAULT; 3698 switch (val) { 3699 case TPACKET_V1: 3700 case TPACKET_V2: 3701 case TPACKET_V3: 3702 break; 3703 default: 3704 return -EINVAL; 3705 } 3706 lock_sock(sk); 3707 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3708 ret = -EBUSY; 3709 } else { 3710 po->tp_version = val; 3711 ret = 0; 3712 } 3713 release_sock(sk); 3714 return ret; 3715 } 3716 case PACKET_RESERVE: 3717 { 3718 unsigned int val; 3719 3720 if (optlen != sizeof(val)) 3721 return -EINVAL; 3722 if (copy_from_user(&val, optval, sizeof(val))) 3723 return -EFAULT; 3724 if (val > INT_MAX) 3725 return -EINVAL; 3726 lock_sock(sk); 3727 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3728 ret = -EBUSY; 3729 } else { 3730 po->tp_reserve = val; 3731 ret = 0; 3732 } 3733 release_sock(sk); 3734 return ret; 3735 } 3736 case PACKET_LOSS: 3737 { 3738 unsigned int val; 3739 3740 if (optlen != sizeof(val)) 3741 return -EINVAL; 3742 if (copy_from_user(&val, optval, sizeof(val))) 3743 return -EFAULT; 3744 3745 lock_sock(sk); 3746 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3747 ret = -EBUSY; 3748 } else { 3749 po->tp_loss = !!val; 3750 ret = 0; 3751 } 3752 release_sock(sk); 3753 return ret; 3754 } 3755 case PACKET_AUXDATA: 3756 { 3757 int val; 3758 3759 if (optlen < sizeof(val)) 3760 return -EINVAL; 3761 if (copy_from_user(&val, optval, sizeof(val))) 3762 return -EFAULT; 3763 3764 lock_sock(sk); 3765 po->auxdata = !!val; 3766 release_sock(sk); 3767 return 0; 3768 } 3769 case PACKET_ORIGDEV: 3770 { 3771 int val; 3772 3773 if (optlen < sizeof(val)) 3774 return -EINVAL; 3775 if (copy_from_user(&val, optval, sizeof(val))) 3776 return -EFAULT; 3777 3778 lock_sock(sk); 3779 po->origdev = !!val; 3780 release_sock(sk); 3781 return 0; 3782 } 3783 case PACKET_VNET_HDR: 3784 { 3785 int val; 3786 3787 if (sock->type != SOCK_RAW) 3788 return -EINVAL; 3789 if (optlen < sizeof(val)) 3790 return -EINVAL; 3791 if (copy_from_user(&val, optval, sizeof(val))) 3792 return -EFAULT; 3793 3794 lock_sock(sk); 3795 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3796 ret = -EBUSY; 3797 } else { 3798 po->has_vnet_hdr = !!val; 3799 ret = 0; 3800 } 3801 release_sock(sk); 3802 return ret; 3803 } 3804 case PACKET_TIMESTAMP: 3805 { 3806 int val; 3807 3808 if (optlen != sizeof(val)) 3809 return -EINVAL; 3810 if (copy_from_user(&val, optval, sizeof(val))) 3811 return -EFAULT; 3812 3813 po->tp_tstamp = val; 3814 return 0; 3815 } 3816 case PACKET_FANOUT: 3817 { 3818 int val; 3819 3820 if (optlen != sizeof(val)) 3821 return -EINVAL; 3822 if (copy_from_user(&val, optval, sizeof(val))) 3823 return -EFAULT; 3824 3825 return fanout_add(sk, val & 0xffff, val >> 16); 3826 } 3827 case PACKET_FANOUT_DATA: 3828 { 3829 if (!po->fanout) 3830 return -EINVAL; 3831 3832 return fanout_set_data(po, optval, optlen); 3833 } 3834 case PACKET_TX_HAS_OFF: 3835 { 3836 unsigned int val; 3837 3838 if (optlen != sizeof(val)) 3839 return -EINVAL; 3840 if (copy_from_user(&val, optval, sizeof(val))) 3841 return -EFAULT; 3842 3843 lock_sock(sk); 3844 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3845 ret = -EBUSY; 3846 } else { 3847 po->tp_tx_has_off = !!val; 3848 ret = 0; 3849 } 3850 release_sock(sk); 3851 return 0; 3852 } 3853 case PACKET_QDISC_BYPASS: 3854 { 3855 int val; 3856 3857 if (optlen != sizeof(val)) 3858 return -EINVAL; 3859 if (copy_from_user(&val, optval, sizeof(val))) 3860 return -EFAULT; 3861 3862 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; 3863 return 0; 3864 } 3865 default: 3866 return -ENOPROTOOPT; 3867 } 3868 } 3869 3870 static int packet_getsockopt(struct socket *sock, int level, int optname, 3871 char __user *optval, int __user *optlen) 3872 { 3873 int len; 3874 int val, lv = sizeof(val); 3875 struct sock *sk = sock->sk; 3876 struct packet_sock *po = pkt_sk(sk); 3877 void *data = &val; 3878 union tpacket_stats_u st; 3879 struct tpacket_rollover_stats rstats; 3880 3881 if (level != SOL_PACKET) 3882 return -ENOPROTOOPT; 3883 3884 if (get_user(len, optlen)) 3885 return -EFAULT; 3886 3887 if (len < 0) 3888 return -EINVAL; 3889 3890 switch (optname) { 3891 case PACKET_STATISTICS: 3892 spin_lock_bh(&sk->sk_receive_queue.lock); 3893 memcpy(&st, &po->stats, sizeof(st)); 3894 memset(&po->stats, 0, sizeof(po->stats)); 3895 spin_unlock_bh(&sk->sk_receive_queue.lock); 3896 3897 if (po->tp_version == TPACKET_V3) { 3898 lv = sizeof(struct tpacket_stats_v3); 3899 st.stats3.tp_packets += st.stats3.tp_drops; 3900 data = &st.stats3; 3901 } else { 3902 lv = sizeof(struct tpacket_stats); 3903 st.stats1.tp_packets += st.stats1.tp_drops; 3904 data = &st.stats1; 3905 } 3906 3907 break; 3908 case PACKET_AUXDATA: 3909 val = po->auxdata; 3910 break; 3911 case PACKET_ORIGDEV: 3912 val = po->origdev; 3913 break; 3914 case PACKET_VNET_HDR: 3915 val = po->has_vnet_hdr; 3916 break; 3917 case PACKET_VERSION: 3918 val = po->tp_version; 3919 break; 3920 case PACKET_HDRLEN: 3921 if (len > sizeof(int)) 3922 len = sizeof(int); 3923 if (len < sizeof(int)) 3924 return -EINVAL; 3925 if (copy_from_user(&val, optval, len)) 3926 return -EFAULT; 3927 switch (val) { 3928 case TPACKET_V1: 3929 val = sizeof(struct tpacket_hdr); 3930 break; 3931 case TPACKET_V2: 3932 val = sizeof(struct tpacket2_hdr); 3933 break; 3934 case TPACKET_V3: 3935 val = sizeof(struct tpacket3_hdr); 3936 break; 3937 default: 3938 return -EINVAL; 3939 } 3940 break; 3941 case PACKET_RESERVE: 3942 val = po->tp_reserve; 3943 break; 3944 case PACKET_LOSS: 3945 val = po->tp_loss; 3946 break; 3947 case PACKET_TIMESTAMP: 3948 val = po->tp_tstamp; 3949 break; 3950 case PACKET_FANOUT: 3951 val = (po->fanout ? 3952 ((u32)po->fanout->id | 3953 ((u32)po->fanout->type << 16) | 3954 ((u32)po->fanout->flags << 24)) : 3955 0); 3956 break; 3957 case PACKET_ROLLOVER_STATS: 3958 if (!po->rollover) 3959 return -EINVAL; 3960 rstats.tp_all = atomic_long_read(&po->rollover->num); 3961 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 3962 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 3963 data = &rstats; 3964 lv = sizeof(rstats); 3965 break; 3966 case PACKET_TX_HAS_OFF: 3967 val = po->tp_tx_has_off; 3968 break; 3969 case PACKET_QDISC_BYPASS: 3970 val = packet_use_direct_xmit(po); 3971 break; 3972 default: 3973 return -ENOPROTOOPT; 3974 } 3975 3976 if (len > lv) 3977 len = lv; 3978 if (put_user(len, optlen)) 3979 return -EFAULT; 3980 if (copy_to_user(optval, data, len)) 3981 return -EFAULT; 3982 return 0; 3983 } 3984 3985 3986 #ifdef CONFIG_COMPAT 3987 static int compat_packet_setsockopt(struct socket *sock, int level, int optname, 3988 char __user *optval, unsigned int optlen) 3989 { 3990 struct packet_sock *po = pkt_sk(sock->sk); 3991 3992 if (level != SOL_PACKET) 3993 return -ENOPROTOOPT; 3994 3995 if (optname == PACKET_FANOUT_DATA && 3996 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) { 3997 optval = (char __user *)get_compat_bpf_fprog(optval); 3998 if (!optval) 3999 return -EFAULT; 4000 optlen = sizeof(struct sock_fprog); 4001 } 4002 4003 return packet_setsockopt(sock, level, optname, optval, optlen); 4004 } 4005 #endif 4006 4007 static int packet_notifier(struct notifier_block *this, 4008 unsigned long msg, void *ptr) 4009 { 4010 struct sock *sk; 4011 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4012 struct net *net = dev_net(dev); 4013 4014 rcu_read_lock(); 4015 sk_for_each_rcu(sk, &net->packet.sklist) { 4016 struct packet_sock *po = pkt_sk(sk); 4017 4018 switch (msg) { 4019 case NETDEV_UNREGISTER: 4020 if (po->mclist) 4021 packet_dev_mclist_delete(dev, &po->mclist); 4022 /* fallthrough */ 4023 4024 case NETDEV_DOWN: 4025 if (dev->ifindex == po->ifindex) { 4026 spin_lock(&po->bind_lock); 4027 if (po->running) { 4028 __unregister_prot_hook(sk, false); 4029 sk->sk_err = ENETDOWN; 4030 if (!sock_flag(sk, SOCK_DEAD)) 4031 sk->sk_error_report(sk); 4032 } 4033 if (msg == NETDEV_UNREGISTER) { 4034 packet_cached_dev_reset(po); 4035 po->ifindex = -1; 4036 if (po->prot_hook.dev) 4037 dev_put(po->prot_hook.dev); 4038 po->prot_hook.dev = NULL; 4039 } 4040 spin_unlock(&po->bind_lock); 4041 } 4042 break; 4043 case NETDEV_UP: 4044 if (dev->ifindex == po->ifindex) { 4045 spin_lock(&po->bind_lock); 4046 if (po->num) 4047 register_prot_hook(sk); 4048 spin_unlock(&po->bind_lock); 4049 } 4050 break; 4051 } 4052 } 4053 rcu_read_unlock(); 4054 return NOTIFY_DONE; 4055 } 4056 4057 4058 static int packet_ioctl(struct socket *sock, unsigned int cmd, 4059 unsigned long arg) 4060 { 4061 struct sock *sk = sock->sk; 4062 4063 switch (cmd) { 4064 case SIOCOUTQ: 4065 { 4066 int amount = sk_wmem_alloc_get(sk); 4067 4068 return put_user(amount, (int __user *)arg); 4069 } 4070 case SIOCINQ: 4071 { 4072 struct sk_buff *skb; 4073 int amount = 0; 4074 4075 spin_lock_bh(&sk->sk_receive_queue.lock); 4076 skb = skb_peek(&sk->sk_receive_queue); 4077 if (skb) 4078 amount = skb->len; 4079 spin_unlock_bh(&sk->sk_receive_queue.lock); 4080 return put_user(amount, (int __user *)arg); 4081 } 4082 case SIOCGSTAMP: 4083 return sock_get_timestamp(sk, (struct timeval __user *)arg); 4084 case SIOCGSTAMPNS: 4085 return sock_get_timestampns(sk, (struct timespec __user *)arg); 4086 4087 #ifdef CONFIG_INET 4088 case SIOCADDRT: 4089 case SIOCDELRT: 4090 case SIOCDARP: 4091 case SIOCGARP: 4092 case SIOCSARP: 4093 case SIOCGIFADDR: 4094 case SIOCSIFADDR: 4095 case SIOCGIFBRDADDR: 4096 case SIOCSIFBRDADDR: 4097 case SIOCGIFNETMASK: 4098 case SIOCSIFNETMASK: 4099 case SIOCGIFDSTADDR: 4100 case SIOCSIFDSTADDR: 4101 case SIOCSIFFLAGS: 4102 return inet_dgram_ops.ioctl(sock, cmd, arg); 4103 #endif 4104 4105 default: 4106 return -ENOIOCTLCMD; 4107 } 4108 return 0; 4109 } 4110 4111 static __poll_t packet_poll(struct file *file, struct socket *sock, 4112 poll_table *wait) 4113 { 4114 struct sock *sk = sock->sk; 4115 struct packet_sock *po = pkt_sk(sk); 4116 __poll_t mask = datagram_poll(file, sock, wait); 4117 4118 spin_lock_bh(&sk->sk_receive_queue.lock); 4119 if (po->rx_ring.pg_vec) { 4120 if (!packet_previous_rx_frame(po, &po->rx_ring, 4121 TP_STATUS_KERNEL)) 4122 mask |= EPOLLIN | EPOLLRDNORM; 4123 } 4124 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 4125 po->pressure = 0; 4126 spin_unlock_bh(&sk->sk_receive_queue.lock); 4127 spin_lock_bh(&sk->sk_write_queue.lock); 4128 if (po->tx_ring.pg_vec) { 4129 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4130 mask |= EPOLLOUT | EPOLLWRNORM; 4131 } 4132 spin_unlock_bh(&sk->sk_write_queue.lock); 4133 return mask; 4134 } 4135 4136 4137 /* Dirty? Well, I still did not learn better way to account 4138 * for user mmaps. 4139 */ 4140 4141 static void packet_mm_open(struct vm_area_struct *vma) 4142 { 4143 struct file *file = vma->vm_file; 4144 struct socket *sock = file->private_data; 4145 struct sock *sk = sock->sk; 4146 4147 if (sk) 4148 atomic_inc(&pkt_sk(sk)->mapped); 4149 } 4150 4151 static void packet_mm_close(struct vm_area_struct *vma) 4152 { 4153 struct file *file = vma->vm_file; 4154 struct socket *sock = file->private_data; 4155 struct sock *sk = sock->sk; 4156 4157 if (sk) 4158 atomic_dec(&pkt_sk(sk)->mapped); 4159 } 4160 4161 static const struct vm_operations_struct packet_mmap_ops = { 4162 .open = packet_mm_open, 4163 .close = packet_mm_close, 4164 }; 4165 4166 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4167 unsigned int len) 4168 { 4169 int i; 4170 4171 for (i = 0; i < len; i++) { 4172 if (likely(pg_vec[i].buffer)) { 4173 if (is_vmalloc_addr(pg_vec[i].buffer)) 4174 vfree(pg_vec[i].buffer); 4175 else 4176 free_pages((unsigned long)pg_vec[i].buffer, 4177 order); 4178 pg_vec[i].buffer = NULL; 4179 } 4180 } 4181 kfree(pg_vec); 4182 } 4183 4184 static char *alloc_one_pg_vec_page(unsigned long order) 4185 { 4186 char *buffer; 4187 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4188 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4189 4190 buffer = (char *) __get_free_pages(gfp_flags, order); 4191 if (buffer) 4192 return buffer; 4193 4194 /* __get_free_pages failed, fall back to vmalloc */ 4195 buffer = vzalloc((1 << order) * PAGE_SIZE); 4196 if (buffer) 4197 return buffer; 4198 4199 /* vmalloc failed, lets dig into swap here */ 4200 gfp_flags &= ~__GFP_NORETRY; 4201 buffer = (char *) __get_free_pages(gfp_flags, order); 4202 if (buffer) 4203 return buffer; 4204 4205 /* complete and utter failure */ 4206 return NULL; 4207 } 4208 4209 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4210 { 4211 unsigned int block_nr = req->tp_block_nr; 4212 struct pgv *pg_vec; 4213 int i; 4214 4215 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL); 4216 if (unlikely(!pg_vec)) 4217 goto out; 4218 4219 for (i = 0; i < block_nr; i++) { 4220 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4221 if (unlikely(!pg_vec[i].buffer)) 4222 goto out_free_pgvec; 4223 } 4224 4225 out: 4226 return pg_vec; 4227 4228 out_free_pgvec: 4229 free_pg_vec(pg_vec, order, block_nr); 4230 pg_vec = NULL; 4231 goto out; 4232 } 4233 4234 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4235 int closing, int tx_ring) 4236 { 4237 struct pgv *pg_vec = NULL; 4238 struct packet_sock *po = pkt_sk(sk); 4239 int was_running, order = 0; 4240 struct packet_ring_buffer *rb; 4241 struct sk_buff_head *rb_queue; 4242 __be16 num; 4243 int err = -EINVAL; 4244 /* Added to avoid minimal code churn */ 4245 struct tpacket_req *req = &req_u->req; 4246 4247 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4248 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4249 4250 err = -EBUSY; 4251 if (!closing) { 4252 if (atomic_read(&po->mapped)) 4253 goto out; 4254 if (packet_read_pending(rb)) 4255 goto out; 4256 } 4257 4258 if (req->tp_block_nr) { 4259 /* Sanity tests and some calculations */ 4260 err = -EBUSY; 4261 if (unlikely(rb->pg_vec)) 4262 goto out; 4263 4264 switch (po->tp_version) { 4265 case TPACKET_V1: 4266 po->tp_hdrlen = TPACKET_HDRLEN; 4267 break; 4268 case TPACKET_V2: 4269 po->tp_hdrlen = TPACKET2_HDRLEN; 4270 break; 4271 case TPACKET_V3: 4272 po->tp_hdrlen = TPACKET3_HDRLEN; 4273 break; 4274 } 4275 4276 err = -EINVAL; 4277 if (unlikely((int)req->tp_block_size <= 0)) 4278 goto out; 4279 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4280 goto out; 4281 if (po->tp_version >= TPACKET_V3 && 4282 req->tp_block_size <= 4283 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv)) 4284 goto out; 4285 if (unlikely(req->tp_frame_size < po->tp_hdrlen + 4286 po->tp_reserve)) 4287 goto out; 4288 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4289 goto out; 4290 4291 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4292 if (unlikely(rb->frames_per_block == 0)) 4293 goto out; 4294 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr)) 4295 goto out; 4296 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4297 req->tp_frame_nr)) 4298 goto out; 4299 4300 err = -ENOMEM; 4301 order = get_order(req->tp_block_size); 4302 pg_vec = alloc_pg_vec(req, order); 4303 if (unlikely(!pg_vec)) 4304 goto out; 4305 switch (po->tp_version) { 4306 case TPACKET_V3: 4307 /* Block transmit is not supported yet */ 4308 if (!tx_ring) { 4309 init_prb_bdqc(po, rb, pg_vec, req_u); 4310 } else { 4311 struct tpacket_req3 *req3 = &req_u->req3; 4312 4313 if (req3->tp_retire_blk_tov || 4314 req3->tp_sizeof_priv || 4315 req3->tp_feature_req_word) { 4316 err = -EINVAL; 4317 goto out; 4318 } 4319 } 4320 break; 4321 default: 4322 break; 4323 } 4324 } 4325 /* Done */ 4326 else { 4327 err = -EINVAL; 4328 if (unlikely(req->tp_frame_nr)) 4329 goto out; 4330 } 4331 4332 4333 /* Detach socket from network */ 4334 spin_lock(&po->bind_lock); 4335 was_running = po->running; 4336 num = po->num; 4337 if (was_running) { 4338 po->num = 0; 4339 __unregister_prot_hook(sk, false); 4340 } 4341 spin_unlock(&po->bind_lock); 4342 4343 synchronize_net(); 4344 4345 err = -EBUSY; 4346 mutex_lock(&po->pg_vec_lock); 4347 if (closing || atomic_read(&po->mapped) == 0) { 4348 err = 0; 4349 spin_lock_bh(&rb_queue->lock); 4350 swap(rb->pg_vec, pg_vec); 4351 rb->frame_max = (req->tp_frame_nr - 1); 4352 rb->head = 0; 4353 rb->frame_size = req->tp_frame_size; 4354 spin_unlock_bh(&rb_queue->lock); 4355 4356 swap(rb->pg_vec_order, order); 4357 swap(rb->pg_vec_len, req->tp_block_nr); 4358 4359 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4360 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4361 tpacket_rcv : packet_rcv; 4362 skb_queue_purge(rb_queue); 4363 if (atomic_read(&po->mapped)) 4364 pr_err("packet_mmap: vma is busy: %d\n", 4365 atomic_read(&po->mapped)); 4366 } 4367 mutex_unlock(&po->pg_vec_lock); 4368 4369 spin_lock(&po->bind_lock); 4370 if (was_running) { 4371 po->num = num; 4372 register_prot_hook(sk); 4373 } 4374 spin_unlock(&po->bind_lock); 4375 if (pg_vec && (po->tp_version > TPACKET_V2)) { 4376 /* Because we don't support block-based V3 on tx-ring */ 4377 if (!tx_ring) 4378 prb_shutdown_retire_blk_timer(po, rb_queue); 4379 } 4380 4381 if (pg_vec) 4382 free_pg_vec(pg_vec, order, req->tp_block_nr); 4383 out: 4384 return err; 4385 } 4386 4387 static int packet_mmap(struct file *file, struct socket *sock, 4388 struct vm_area_struct *vma) 4389 { 4390 struct sock *sk = sock->sk; 4391 struct packet_sock *po = pkt_sk(sk); 4392 unsigned long size, expected_size; 4393 struct packet_ring_buffer *rb; 4394 unsigned long start; 4395 int err = -EINVAL; 4396 int i; 4397 4398 if (vma->vm_pgoff) 4399 return -EINVAL; 4400 4401 mutex_lock(&po->pg_vec_lock); 4402 4403 expected_size = 0; 4404 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4405 if (rb->pg_vec) { 4406 expected_size += rb->pg_vec_len 4407 * rb->pg_vec_pages 4408 * PAGE_SIZE; 4409 } 4410 } 4411 4412 if (expected_size == 0) 4413 goto out; 4414 4415 size = vma->vm_end - vma->vm_start; 4416 if (size != expected_size) 4417 goto out; 4418 4419 start = vma->vm_start; 4420 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4421 if (rb->pg_vec == NULL) 4422 continue; 4423 4424 for (i = 0; i < rb->pg_vec_len; i++) { 4425 struct page *page; 4426 void *kaddr = rb->pg_vec[i].buffer; 4427 int pg_num; 4428 4429 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4430 page = pgv_to_page(kaddr); 4431 err = vm_insert_page(vma, start, page); 4432 if (unlikely(err)) 4433 goto out; 4434 start += PAGE_SIZE; 4435 kaddr += PAGE_SIZE; 4436 } 4437 } 4438 } 4439 4440 atomic_inc(&po->mapped); 4441 vma->vm_ops = &packet_mmap_ops; 4442 err = 0; 4443 4444 out: 4445 mutex_unlock(&po->pg_vec_lock); 4446 return err; 4447 } 4448 4449 static const struct proto_ops packet_ops_spkt = { 4450 .family = PF_PACKET, 4451 .owner = THIS_MODULE, 4452 .release = packet_release, 4453 .bind = packet_bind_spkt, 4454 .connect = sock_no_connect, 4455 .socketpair = sock_no_socketpair, 4456 .accept = sock_no_accept, 4457 .getname = packet_getname_spkt, 4458 .poll = datagram_poll, 4459 .ioctl = packet_ioctl, 4460 .listen = sock_no_listen, 4461 .shutdown = sock_no_shutdown, 4462 .setsockopt = sock_no_setsockopt, 4463 .getsockopt = sock_no_getsockopt, 4464 .sendmsg = packet_sendmsg_spkt, 4465 .recvmsg = packet_recvmsg, 4466 .mmap = sock_no_mmap, 4467 .sendpage = sock_no_sendpage, 4468 }; 4469 4470 static const struct proto_ops packet_ops = { 4471 .family = PF_PACKET, 4472 .owner = THIS_MODULE, 4473 .release = packet_release, 4474 .bind = packet_bind, 4475 .connect = sock_no_connect, 4476 .socketpair = sock_no_socketpair, 4477 .accept = sock_no_accept, 4478 .getname = packet_getname, 4479 .poll = packet_poll, 4480 .ioctl = packet_ioctl, 4481 .listen = sock_no_listen, 4482 .shutdown = sock_no_shutdown, 4483 .setsockopt = packet_setsockopt, 4484 .getsockopt = packet_getsockopt, 4485 #ifdef CONFIG_COMPAT 4486 .compat_setsockopt = compat_packet_setsockopt, 4487 #endif 4488 .sendmsg = packet_sendmsg, 4489 .recvmsg = packet_recvmsg, 4490 .mmap = packet_mmap, 4491 .sendpage = sock_no_sendpage, 4492 }; 4493 4494 static const struct net_proto_family packet_family_ops = { 4495 .family = PF_PACKET, 4496 .create = packet_create, 4497 .owner = THIS_MODULE, 4498 }; 4499 4500 static struct notifier_block packet_netdev_notifier = { 4501 .notifier_call = packet_notifier, 4502 }; 4503 4504 #ifdef CONFIG_PROC_FS 4505 4506 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4507 __acquires(RCU) 4508 { 4509 struct net *net = seq_file_net(seq); 4510 4511 rcu_read_lock(); 4512 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4513 } 4514 4515 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4516 { 4517 struct net *net = seq_file_net(seq); 4518 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4519 } 4520 4521 static void packet_seq_stop(struct seq_file *seq, void *v) 4522 __releases(RCU) 4523 { 4524 rcu_read_unlock(); 4525 } 4526 4527 static int packet_seq_show(struct seq_file *seq, void *v) 4528 { 4529 if (v == SEQ_START_TOKEN) 4530 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4531 else { 4532 struct sock *s = sk_entry(v); 4533 const struct packet_sock *po = pkt_sk(s); 4534 4535 seq_printf(seq, 4536 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4537 s, 4538 refcount_read(&s->sk_refcnt), 4539 s->sk_type, 4540 ntohs(po->num), 4541 po->ifindex, 4542 po->running, 4543 atomic_read(&s->sk_rmem_alloc), 4544 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4545 sock_i_ino(s)); 4546 } 4547 4548 return 0; 4549 } 4550 4551 static const struct seq_operations packet_seq_ops = { 4552 .start = packet_seq_start, 4553 .next = packet_seq_next, 4554 .stop = packet_seq_stop, 4555 .show = packet_seq_show, 4556 }; 4557 4558 static int packet_seq_open(struct inode *inode, struct file *file) 4559 { 4560 return seq_open_net(inode, file, &packet_seq_ops, 4561 sizeof(struct seq_net_private)); 4562 } 4563 4564 static const struct file_operations packet_seq_fops = { 4565 .open = packet_seq_open, 4566 .read = seq_read, 4567 .llseek = seq_lseek, 4568 .release = seq_release_net, 4569 }; 4570 4571 #endif 4572 4573 static int __net_init packet_net_init(struct net *net) 4574 { 4575 mutex_init(&net->packet.sklist_lock); 4576 INIT_HLIST_HEAD(&net->packet.sklist); 4577 4578 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops)) 4579 return -ENOMEM; 4580 4581 return 0; 4582 } 4583 4584 static void __net_exit packet_net_exit(struct net *net) 4585 { 4586 remove_proc_entry("packet", net->proc_net); 4587 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); 4588 } 4589 4590 static struct pernet_operations packet_net_ops = { 4591 .init = packet_net_init, 4592 .exit = packet_net_exit, 4593 }; 4594 4595 4596 static void __exit packet_exit(void) 4597 { 4598 unregister_netdevice_notifier(&packet_netdev_notifier); 4599 unregister_pernet_subsys(&packet_net_ops); 4600 sock_unregister(PF_PACKET); 4601 proto_unregister(&packet_proto); 4602 } 4603 4604 static int __init packet_init(void) 4605 { 4606 int rc = proto_register(&packet_proto, 0); 4607 4608 if (rc != 0) 4609 goto out; 4610 4611 sock_register(&packet_family_ops); 4612 register_pernet_subsys(&packet_net_ops); 4613 register_netdevice_notifier(&packet_netdev_notifier); 4614 out: 4615 return rc; 4616 } 4617 4618 module_init(packet_init); 4619 module_exit(packet_exit); 4620 MODULE_LICENSE("GPL"); 4621 MODULE_ALIAS_NETPROTO(PF_PACKET); 4622