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