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