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 u16 packet_pick_tx_queue(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 return dev_direct_xmit(skb, packet_pick_tx_queue(skb)); 247 } 248 249 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 250 { 251 struct net_device *dev; 252 253 rcu_read_lock(); 254 dev = rcu_dereference(po->cached_dev); 255 if (likely(dev)) 256 dev_hold(dev); 257 rcu_read_unlock(); 258 259 return dev; 260 } 261 262 static void packet_cached_dev_assign(struct packet_sock *po, 263 struct net_device *dev) 264 { 265 rcu_assign_pointer(po->cached_dev, dev); 266 } 267 268 static void packet_cached_dev_reset(struct packet_sock *po) 269 { 270 RCU_INIT_POINTER(po->cached_dev, NULL); 271 } 272 273 static bool packet_use_direct_xmit(const struct packet_sock *po) 274 { 275 return po->xmit == packet_direct_xmit; 276 } 277 278 static u16 packet_pick_tx_queue(struct sk_buff *skb) 279 { 280 struct net_device *dev = skb->dev; 281 const struct net_device_ops *ops = dev->netdev_ops; 282 int cpu = raw_smp_processor_id(); 283 u16 queue_index; 284 285 #ifdef CONFIG_XPS 286 skb->sender_cpu = cpu + 1; 287 #endif 288 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues); 289 if (ops->ndo_select_queue) { 290 queue_index = ops->ndo_select_queue(dev, skb, NULL); 291 queue_index = netdev_cap_txqueue(dev, queue_index); 292 } else { 293 queue_index = netdev_pick_tx(dev, skb, NULL); 294 } 295 296 return queue_index; 297 } 298 299 /* __register_prot_hook must be invoked through register_prot_hook 300 * or from a context in which asynchronous accesses to the packet 301 * socket is not possible (packet_create()). 302 */ 303 static void __register_prot_hook(struct sock *sk) 304 { 305 struct packet_sock *po = pkt_sk(sk); 306 307 if (!po->running) { 308 if (po->fanout) 309 __fanout_link(sk, po); 310 else 311 dev_add_pack(&po->prot_hook); 312 313 sock_hold(sk); 314 po->running = 1; 315 } 316 } 317 318 static void register_prot_hook(struct sock *sk) 319 { 320 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); 321 __register_prot_hook(sk); 322 } 323 324 /* If the sync parameter is true, we will temporarily drop 325 * the po->bind_lock and do a synchronize_net to make sure no 326 * asynchronous packet processing paths still refer to the elements 327 * of po->prot_hook. If the sync parameter is false, it is the 328 * callers responsibility to take care of this. 329 */ 330 static void __unregister_prot_hook(struct sock *sk, bool sync) 331 { 332 struct packet_sock *po = pkt_sk(sk); 333 334 lockdep_assert_held_once(&po->bind_lock); 335 336 po->running = 0; 337 338 if (po->fanout) 339 __fanout_unlink(sk, po); 340 else 341 __dev_remove_pack(&po->prot_hook); 342 343 __sock_put(sk); 344 345 if (sync) { 346 spin_unlock(&po->bind_lock); 347 synchronize_net(); 348 spin_lock(&po->bind_lock); 349 } 350 } 351 352 static void unregister_prot_hook(struct sock *sk, bool sync) 353 { 354 struct packet_sock *po = pkt_sk(sk); 355 356 if (po->running) 357 __unregister_prot_hook(sk, sync); 358 } 359 360 static inline struct page * __pure pgv_to_page(void *addr) 361 { 362 if (is_vmalloc_addr(addr)) 363 return vmalloc_to_page(addr); 364 return virt_to_page(addr); 365 } 366 367 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 368 { 369 union tpacket_uhdr h; 370 371 h.raw = frame; 372 switch (po->tp_version) { 373 case TPACKET_V1: 374 h.h1->tp_status = status; 375 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 376 break; 377 case TPACKET_V2: 378 h.h2->tp_status = status; 379 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 380 break; 381 case TPACKET_V3: 382 h.h3->tp_status = status; 383 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 384 break; 385 default: 386 WARN(1, "TPACKET version not supported.\n"); 387 BUG(); 388 } 389 390 smp_wmb(); 391 } 392 393 static int __packet_get_status(struct packet_sock *po, void *frame) 394 { 395 union tpacket_uhdr h; 396 397 smp_rmb(); 398 399 h.raw = frame; 400 switch (po->tp_version) { 401 case TPACKET_V1: 402 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 403 return h.h1->tp_status; 404 case TPACKET_V2: 405 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 406 return h.h2->tp_status; 407 case TPACKET_V3: 408 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 409 return h.h3->tp_status; 410 default: 411 WARN(1, "TPACKET version not supported.\n"); 412 BUG(); 413 return 0; 414 } 415 } 416 417 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, 418 unsigned int flags) 419 { 420 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 421 422 if (shhwtstamps && 423 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 424 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) 425 return TP_STATUS_TS_RAW_HARDWARE; 426 427 if (ktime_to_timespec_cond(skb->tstamp, ts)) 428 return TP_STATUS_TS_SOFTWARE; 429 430 return 0; 431 } 432 433 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 434 struct sk_buff *skb) 435 { 436 union tpacket_uhdr h; 437 struct timespec ts; 438 __u32 ts_status; 439 440 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 441 return 0; 442 443 h.raw = frame; 444 switch (po->tp_version) { 445 case TPACKET_V1: 446 h.h1->tp_sec = ts.tv_sec; 447 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 448 break; 449 case TPACKET_V2: 450 h.h2->tp_sec = ts.tv_sec; 451 h.h2->tp_nsec = ts.tv_nsec; 452 break; 453 case TPACKET_V3: 454 h.h3->tp_sec = ts.tv_sec; 455 h.h3->tp_nsec = ts.tv_nsec; 456 break; 457 default: 458 WARN(1, "TPACKET version not supported.\n"); 459 BUG(); 460 } 461 462 /* one flush is safe, as both fields always lie on the same cacheline */ 463 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 464 smp_wmb(); 465 466 return ts_status; 467 } 468 469 static void *packet_lookup_frame(struct packet_sock *po, 470 struct packet_ring_buffer *rb, 471 unsigned int position, 472 int status) 473 { 474 unsigned int pg_vec_pos, frame_offset; 475 union tpacket_uhdr h; 476 477 pg_vec_pos = position / rb->frames_per_block; 478 frame_offset = position % rb->frames_per_block; 479 480 h.raw = rb->pg_vec[pg_vec_pos].buffer + 481 (frame_offset * rb->frame_size); 482 483 if (status != __packet_get_status(po, h.raw)) 484 return NULL; 485 486 return h.raw; 487 } 488 489 static void *packet_current_frame(struct packet_sock *po, 490 struct packet_ring_buffer *rb, 491 int status) 492 { 493 return packet_lookup_frame(po, rb, rb->head, status); 494 } 495 496 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 497 { 498 del_timer_sync(&pkc->retire_blk_timer); 499 } 500 501 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 502 struct sk_buff_head *rb_queue) 503 { 504 struct tpacket_kbdq_core *pkc; 505 506 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 507 508 spin_lock_bh(&rb_queue->lock); 509 pkc->delete_blk_timer = 1; 510 spin_unlock_bh(&rb_queue->lock); 511 512 prb_del_retire_blk_timer(pkc); 513 } 514 515 static void prb_setup_retire_blk_timer(struct packet_sock *po) 516 { 517 struct tpacket_kbdq_core *pkc; 518 519 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 520 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, 521 0); 522 pkc->retire_blk_timer.expires = jiffies; 523 } 524 525 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 526 int blk_size_in_bytes) 527 { 528 struct net_device *dev; 529 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; 530 struct ethtool_link_ksettings ecmd; 531 int err; 532 533 rtnl_lock(); 534 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 535 if (unlikely(!dev)) { 536 rtnl_unlock(); 537 return DEFAULT_PRB_RETIRE_TOV; 538 } 539 err = __ethtool_get_link_ksettings(dev, &ecmd); 540 rtnl_unlock(); 541 if (!err) { 542 /* 543 * If the link speed is so slow you don't really 544 * need to worry about perf anyways 545 */ 546 if (ecmd.base.speed < SPEED_1000 || 547 ecmd.base.speed == SPEED_UNKNOWN) { 548 return DEFAULT_PRB_RETIRE_TOV; 549 } else { 550 msec = 1; 551 div = ecmd.base.speed / 1000; 552 } 553 } 554 555 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 556 557 if (div) 558 mbits /= div; 559 560 tmo = mbits * msec; 561 562 if (div) 563 return tmo+1; 564 return tmo; 565 } 566 567 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 568 union tpacket_req_u *req_u) 569 { 570 p1->feature_req_word = req_u->req3.tp_feature_req_word; 571 } 572 573 static void init_prb_bdqc(struct packet_sock *po, 574 struct packet_ring_buffer *rb, 575 struct pgv *pg_vec, 576 union tpacket_req_u *req_u) 577 { 578 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 579 struct tpacket_block_desc *pbd; 580 581 memset(p1, 0x0, sizeof(*p1)); 582 583 p1->knxt_seq_num = 1; 584 p1->pkbdq = pg_vec; 585 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 586 p1->pkblk_start = pg_vec[0].buffer; 587 p1->kblk_size = req_u->req3.tp_block_size; 588 p1->knum_blocks = req_u->req3.tp_block_nr; 589 p1->hdrlen = po->tp_hdrlen; 590 p1->version = po->tp_version; 591 p1->last_kactive_blk_num = 0; 592 po->stats.stats3.tp_freeze_q_cnt = 0; 593 if (req_u->req3.tp_retire_blk_tov) 594 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 595 else 596 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 597 req_u->req3.tp_block_size); 598 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 599 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 600 601 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 602 prb_init_ft_ops(p1, req_u); 603 prb_setup_retire_blk_timer(po); 604 prb_open_block(p1, pbd); 605 } 606 607 /* Do NOT update the last_blk_num first. 608 * Assumes sk_buff_head lock is held. 609 */ 610 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 611 { 612 mod_timer(&pkc->retire_blk_timer, 613 jiffies + pkc->tov_in_jiffies); 614 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 615 } 616 617 /* 618 * Timer logic: 619 * 1) We refresh the timer only when we open a block. 620 * By doing this we don't waste cycles refreshing the timer 621 * on packet-by-packet basis. 622 * 623 * With a 1MB block-size, on a 1Gbps line, it will take 624 * i) ~8 ms to fill a block + ii) memcpy etc. 625 * In this cut we are not accounting for the memcpy time. 626 * 627 * So, if the user sets the 'tmo' to 10ms then the timer 628 * will never fire while the block is still getting filled 629 * (which is what we want). However, the user could choose 630 * to close a block early and that's fine. 631 * 632 * But when the timer does fire, we check whether or not to refresh it. 633 * Since the tmo granularity is in msecs, it is not too expensive 634 * to refresh the timer, lets say every '8' msecs. 635 * Either the user can set the 'tmo' or we can derive it based on 636 * a) line-speed and b) block-size. 637 * prb_calc_retire_blk_tmo() calculates the tmo. 638 * 639 */ 640 static void prb_retire_rx_blk_timer_expired(struct timer_list *t) 641 { 642 struct packet_sock *po = 643 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer); 644 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 645 unsigned int frozen; 646 struct tpacket_block_desc *pbd; 647 648 spin_lock(&po->sk.sk_receive_queue.lock); 649 650 frozen = prb_queue_frozen(pkc); 651 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 652 653 if (unlikely(pkc->delete_blk_timer)) 654 goto out; 655 656 /* We only need to plug the race when the block is partially filled. 657 * tpacket_rcv: 658 * lock(); increment BLOCK_NUM_PKTS; unlock() 659 * copy_bits() is in progress ... 660 * timer fires on other cpu: 661 * we can't retire the current block because copy_bits 662 * is in progress. 663 * 664 */ 665 if (BLOCK_NUM_PKTS(pbd)) { 666 while (atomic_read(&pkc->blk_fill_in_prog)) { 667 /* Waiting for skb_copy_bits to finish... */ 668 cpu_relax(); 669 } 670 } 671 672 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 673 if (!frozen) { 674 if (!BLOCK_NUM_PKTS(pbd)) { 675 /* An empty block. Just refresh the timer. */ 676 goto refresh_timer; 677 } 678 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 679 if (!prb_dispatch_next_block(pkc, po)) 680 goto refresh_timer; 681 else 682 goto out; 683 } else { 684 /* Case 1. Queue was frozen because user-space was 685 * lagging behind. 686 */ 687 if (prb_curr_blk_in_use(pbd)) { 688 /* 689 * Ok, user-space is still behind. 690 * So just refresh the timer. 691 */ 692 goto refresh_timer; 693 } else { 694 /* Case 2. queue was frozen,user-space caught up, 695 * now the link went idle && the timer fired. 696 * We don't have a block to close.So we open this 697 * block and restart the timer. 698 * opening a block thaws the queue,restarts timer 699 * Thawing/timer-refresh is a side effect. 700 */ 701 prb_open_block(pkc, pbd); 702 goto out; 703 } 704 } 705 } 706 707 refresh_timer: 708 _prb_refresh_rx_retire_blk_timer(pkc); 709 710 out: 711 spin_unlock(&po->sk.sk_receive_queue.lock); 712 } 713 714 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 715 struct tpacket_block_desc *pbd1, __u32 status) 716 { 717 /* Flush everything minus the block header */ 718 719 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 720 u8 *start, *end; 721 722 start = (u8 *)pbd1; 723 724 /* Skip the block header(we know header WILL fit in 4K) */ 725 start += PAGE_SIZE; 726 727 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 728 for (; start < end; start += PAGE_SIZE) 729 flush_dcache_page(pgv_to_page(start)); 730 731 smp_wmb(); 732 #endif 733 734 /* Now update the block status. */ 735 736 BLOCK_STATUS(pbd1) = status; 737 738 /* Flush the block header */ 739 740 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 741 start = (u8 *)pbd1; 742 flush_dcache_page(pgv_to_page(start)); 743 744 smp_wmb(); 745 #endif 746 } 747 748 /* 749 * Side effect: 750 * 751 * 1) flush the block 752 * 2) Increment active_blk_num 753 * 754 * Note:We DONT refresh the timer on purpose. 755 * Because almost always the next block will be opened. 756 */ 757 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 758 struct tpacket_block_desc *pbd1, 759 struct packet_sock *po, unsigned int stat) 760 { 761 __u32 status = TP_STATUS_USER | stat; 762 763 struct tpacket3_hdr *last_pkt; 764 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 765 struct sock *sk = &po->sk; 766 767 if (po->stats.stats3.tp_drops) 768 status |= TP_STATUS_LOSING; 769 770 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 771 last_pkt->tp_next_offset = 0; 772 773 /* Get the ts of the last pkt */ 774 if (BLOCK_NUM_PKTS(pbd1)) { 775 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 776 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 777 } else { 778 /* Ok, we tmo'd - so get the current time. 779 * 780 * It shouldn't really happen as we don't close empty 781 * blocks. See prb_retire_rx_blk_timer_expired(). 782 */ 783 struct timespec ts; 784 getnstimeofday(&ts); 785 h1->ts_last_pkt.ts_sec = ts.tv_sec; 786 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 787 } 788 789 smp_wmb(); 790 791 /* Flush the block */ 792 prb_flush_block(pkc1, pbd1, status); 793 794 sk->sk_data_ready(sk); 795 796 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 797 } 798 799 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 800 { 801 pkc->reset_pending_on_curr_blk = 0; 802 } 803 804 /* 805 * Side effect of opening a block: 806 * 807 * 1) prb_queue is thawed. 808 * 2) retire_blk_timer is refreshed. 809 * 810 */ 811 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 812 struct tpacket_block_desc *pbd1) 813 { 814 struct timespec ts; 815 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 816 817 smp_rmb(); 818 819 /* We could have just memset this but we will lose the 820 * flexibility of making the priv area sticky 821 */ 822 823 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 824 BLOCK_NUM_PKTS(pbd1) = 0; 825 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 826 827 getnstimeofday(&ts); 828 829 h1->ts_first_pkt.ts_sec = ts.tv_sec; 830 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 831 832 pkc1->pkblk_start = (char *)pbd1; 833 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 834 835 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 836 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 837 838 pbd1->version = pkc1->version; 839 pkc1->prev = pkc1->nxt_offset; 840 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 841 842 prb_thaw_queue(pkc1); 843 _prb_refresh_rx_retire_blk_timer(pkc1); 844 845 smp_wmb(); 846 } 847 848 /* 849 * Queue freeze logic: 850 * 1) Assume tp_block_nr = 8 blocks. 851 * 2) At time 't0', user opens Rx ring. 852 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 853 * 4) user-space is either sleeping or processing block '0'. 854 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 855 * it will close block-7,loop around and try to fill block '0'. 856 * call-flow: 857 * __packet_lookup_frame_in_block 858 * prb_retire_current_block() 859 * prb_dispatch_next_block() 860 * |->(BLOCK_STATUS == USER) evaluates to true 861 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 862 * 6) Now there are two cases: 863 * 6.1) Link goes idle right after the queue is frozen. 864 * But remember, the last open_block() refreshed the timer. 865 * When this timer expires,it will refresh itself so that we can 866 * re-open block-0 in near future. 867 * 6.2) Link is busy and keeps on receiving packets. This is a simple 868 * case and __packet_lookup_frame_in_block will check if block-0 869 * is free and can now be re-used. 870 */ 871 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 872 struct packet_sock *po) 873 { 874 pkc->reset_pending_on_curr_blk = 1; 875 po->stats.stats3.tp_freeze_q_cnt++; 876 } 877 878 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 879 880 /* 881 * If the next block is free then we will dispatch it 882 * and return a good offset. 883 * Else, we will freeze the queue. 884 * So, caller must check the return value. 885 */ 886 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 887 struct packet_sock *po) 888 { 889 struct tpacket_block_desc *pbd; 890 891 smp_rmb(); 892 893 /* 1. Get current block num */ 894 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 895 896 /* 2. If this block is currently in_use then freeze the queue */ 897 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 898 prb_freeze_queue(pkc, po); 899 return NULL; 900 } 901 902 /* 903 * 3. 904 * open this block and return the offset where the first packet 905 * needs to get stored. 906 */ 907 prb_open_block(pkc, pbd); 908 return (void *)pkc->nxt_offset; 909 } 910 911 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 912 struct packet_sock *po, unsigned int status) 913 { 914 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 915 916 /* retire/close the current block */ 917 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 918 /* 919 * Plug the case where copy_bits() is in progress on 920 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 921 * have space to copy the pkt in the current block and 922 * called prb_retire_current_block() 923 * 924 * We don't need to worry about the TMO case because 925 * the timer-handler already handled this case. 926 */ 927 if (!(status & TP_STATUS_BLK_TMO)) { 928 while (atomic_read(&pkc->blk_fill_in_prog)) { 929 /* Waiting for skb_copy_bits to finish... */ 930 cpu_relax(); 931 } 932 } 933 prb_close_block(pkc, pbd, po, status); 934 return; 935 } 936 } 937 938 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) 939 { 940 return TP_STATUS_USER & BLOCK_STATUS(pbd); 941 } 942 943 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 944 { 945 return pkc->reset_pending_on_curr_blk; 946 } 947 948 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 949 { 950 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 951 atomic_dec(&pkc->blk_fill_in_prog); 952 } 953 954 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 955 struct tpacket3_hdr *ppd) 956 { 957 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 958 } 959 960 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 961 struct tpacket3_hdr *ppd) 962 { 963 ppd->hv1.tp_rxhash = 0; 964 } 965 966 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 967 struct tpacket3_hdr *ppd) 968 { 969 if (skb_vlan_tag_present(pkc->skb)) { 970 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 971 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 972 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 973 } else { 974 ppd->hv1.tp_vlan_tci = 0; 975 ppd->hv1.tp_vlan_tpid = 0; 976 ppd->tp_status = TP_STATUS_AVAILABLE; 977 } 978 } 979 980 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 981 struct tpacket3_hdr *ppd) 982 { 983 ppd->hv1.tp_padding = 0; 984 prb_fill_vlan_info(pkc, ppd); 985 986 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 987 prb_fill_rxhash(pkc, ppd); 988 else 989 prb_clear_rxhash(pkc, ppd); 990 } 991 992 static void prb_fill_curr_block(char *curr, 993 struct tpacket_kbdq_core *pkc, 994 struct tpacket_block_desc *pbd, 995 unsigned int len) 996 { 997 struct tpacket3_hdr *ppd; 998 999 ppd = (struct tpacket3_hdr *)curr; 1000 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 1001 pkc->prev = curr; 1002 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 1003 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1004 BLOCK_NUM_PKTS(pbd) += 1; 1005 atomic_inc(&pkc->blk_fill_in_prog); 1006 prb_run_all_ft_ops(pkc, ppd); 1007 } 1008 1009 /* Assumes caller has the sk->rx_queue.lock */ 1010 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1011 struct sk_buff *skb, 1012 int status, 1013 unsigned int len 1014 ) 1015 { 1016 struct tpacket_kbdq_core *pkc; 1017 struct tpacket_block_desc *pbd; 1018 char *curr, *end; 1019 1020 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1021 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1022 1023 /* Queue is frozen when user space is lagging behind */ 1024 if (prb_queue_frozen(pkc)) { 1025 /* 1026 * Check if that last block which caused the queue to freeze, 1027 * is still in_use by user-space. 1028 */ 1029 if (prb_curr_blk_in_use(pbd)) { 1030 /* Can't record this packet */ 1031 return NULL; 1032 } else { 1033 /* 1034 * Ok, the block was released by user-space. 1035 * Now let's open that block. 1036 * opening a block also thaws the queue. 1037 * Thawing is a side effect. 1038 */ 1039 prb_open_block(pkc, pbd); 1040 } 1041 } 1042 1043 smp_mb(); 1044 curr = pkc->nxt_offset; 1045 pkc->skb = skb; 1046 end = (char *)pbd + pkc->kblk_size; 1047 1048 /* first try the current block */ 1049 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1050 prb_fill_curr_block(curr, pkc, pbd, len); 1051 return (void *)curr; 1052 } 1053 1054 /* Ok, close the current block */ 1055 prb_retire_current_block(pkc, po, 0); 1056 1057 /* Now, try to dispatch the next block */ 1058 curr = (char *)prb_dispatch_next_block(pkc, po); 1059 if (curr) { 1060 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1061 prb_fill_curr_block(curr, pkc, pbd, len); 1062 return (void *)curr; 1063 } 1064 1065 /* 1066 * No free blocks are available.user_space hasn't caught up yet. 1067 * Queue was just frozen and now this packet will get dropped. 1068 */ 1069 return NULL; 1070 } 1071 1072 static void *packet_current_rx_frame(struct packet_sock *po, 1073 struct sk_buff *skb, 1074 int status, unsigned int len) 1075 { 1076 char *curr = NULL; 1077 switch (po->tp_version) { 1078 case TPACKET_V1: 1079 case TPACKET_V2: 1080 curr = packet_lookup_frame(po, &po->rx_ring, 1081 po->rx_ring.head, status); 1082 return curr; 1083 case TPACKET_V3: 1084 return __packet_lookup_frame_in_block(po, skb, status, len); 1085 default: 1086 WARN(1, "TPACKET version not supported\n"); 1087 BUG(); 1088 return NULL; 1089 } 1090 } 1091 1092 static void *prb_lookup_block(struct packet_sock *po, 1093 struct packet_ring_buffer *rb, 1094 unsigned int idx, 1095 int status) 1096 { 1097 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1098 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1099 1100 if (status != BLOCK_STATUS(pbd)) 1101 return NULL; 1102 return pbd; 1103 } 1104 1105 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1106 { 1107 unsigned int prev; 1108 if (rb->prb_bdqc.kactive_blk_num) 1109 prev = rb->prb_bdqc.kactive_blk_num-1; 1110 else 1111 prev = rb->prb_bdqc.knum_blocks-1; 1112 return prev; 1113 } 1114 1115 /* Assumes caller has held the rx_queue.lock */ 1116 static void *__prb_previous_block(struct packet_sock *po, 1117 struct packet_ring_buffer *rb, 1118 int status) 1119 { 1120 unsigned int previous = prb_previous_blk_num(rb); 1121 return prb_lookup_block(po, rb, previous, status); 1122 } 1123 1124 static void *packet_previous_rx_frame(struct packet_sock *po, 1125 struct packet_ring_buffer *rb, 1126 int status) 1127 { 1128 if (po->tp_version <= TPACKET_V2) 1129 return packet_previous_frame(po, rb, status); 1130 1131 return __prb_previous_block(po, rb, status); 1132 } 1133 1134 static void packet_increment_rx_head(struct packet_sock *po, 1135 struct packet_ring_buffer *rb) 1136 { 1137 switch (po->tp_version) { 1138 case TPACKET_V1: 1139 case TPACKET_V2: 1140 return packet_increment_head(rb); 1141 case TPACKET_V3: 1142 default: 1143 WARN(1, "TPACKET version not supported.\n"); 1144 BUG(); 1145 return; 1146 } 1147 } 1148 1149 static void *packet_previous_frame(struct packet_sock *po, 1150 struct packet_ring_buffer *rb, 1151 int status) 1152 { 1153 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1154 return packet_lookup_frame(po, rb, previous, status); 1155 } 1156 1157 static void packet_increment_head(struct packet_ring_buffer *buff) 1158 { 1159 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1160 } 1161 1162 static void packet_inc_pending(struct packet_ring_buffer *rb) 1163 { 1164 this_cpu_inc(*rb->pending_refcnt); 1165 } 1166 1167 static void packet_dec_pending(struct packet_ring_buffer *rb) 1168 { 1169 this_cpu_dec(*rb->pending_refcnt); 1170 } 1171 1172 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1173 { 1174 unsigned int refcnt = 0; 1175 int cpu; 1176 1177 /* We don't use pending refcount in rx_ring. */ 1178 if (rb->pending_refcnt == NULL) 1179 return 0; 1180 1181 for_each_possible_cpu(cpu) 1182 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1183 1184 return refcnt; 1185 } 1186 1187 static int packet_alloc_pending(struct packet_sock *po) 1188 { 1189 po->rx_ring.pending_refcnt = NULL; 1190 1191 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1192 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1193 return -ENOBUFS; 1194 1195 return 0; 1196 } 1197 1198 static void packet_free_pending(struct packet_sock *po) 1199 { 1200 free_percpu(po->tx_ring.pending_refcnt); 1201 } 1202 1203 #define ROOM_POW_OFF 2 1204 #define ROOM_NONE 0x0 1205 #define ROOM_LOW 0x1 1206 #define ROOM_NORMAL 0x2 1207 1208 static bool __tpacket_has_room(struct packet_sock *po, int pow_off) 1209 { 1210 int idx, len; 1211 1212 len = po->rx_ring.frame_max + 1; 1213 idx = po->rx_ring.head; 1214 if (pow_off) 1215 idx += len >> pow_off; 1216 if (idx >= len) 1217 idx -= len; 1218 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1219 } 1220 1221 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off) 1222 { 1223 int idx, len; 1224 1225 len = po->rx_ring.prb_bdqc.knum_blocks; 1226 idx = po->rx_ring.prb_bdqc.kactive_blk_num; 1227 if (pow_off) 1228 idx += len >> pow_off; 1229 if (idx >= len) 1230 idx -= len; 1231 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1232 } 1233 1234 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1235 { 1236 struct sock *sk = &po->sk; 1237 int ret = ROOM_NONE; 1238 1239 if (po->prot_hook.func != tpacket_rcv) { 1240 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1241 - (skb ? skb->truesize : 0); 1242 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF)) 1243 return ROOM_NORMAL; 1244 else if (avail > 0) 1245 return ROOM_LOW; 1246 else 1247 return ROOM_NONE; 1248 } 1249 1250 if (po->tp_version == TPACKET_V3) { 1251 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1252 ret = ROOM_NORMAL; 1253 else if (__tpacket_v3_has_room(po, 0)) 1254 ret = ROOM_LOW; 1255 } else { 1256 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1257 ret = ROOM_NORMAL; 1258 else if (__tpacket_has_room(po, 0)) 1259 ret = ROOM_LOW; 1260 } 1261 1262 return ret; 1263 } 1264 1265 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1266 { 1267 int ret; 1268 bool has_room; 1269 1270 spin_lock_bh(&po->sk.sk_receive_queue.lock); 1271 ret = __packet_rcv_has_room(po, skb); 1272 has_room = ret == ROOM_NORMAL; 1273 if (po->pressure == has_room) 1274 po->pressure = !has_room; 1275 spin_unlock_bh(&po->sk.sk_receive_queue.lock); 1276 1277 return ret; 1278 } 1279 1280 static void packet_sock_destruct(struct sock *sk) 1281 { 1282 skb_queue_purge(&sk->sk_error_queue); 1283 1284 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1285 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 1286 1287 if (!sock_flag(sk, SOCK_DEAD)) { 1288 pr_err("Attempt to release alive packet socket: %p\n", sk); 1289 return; 1290 } 1291 1292 sk_refcnt_debug_dec(sk); 1293 } 1294 1295 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1296 { 1297 u32 rxhash; 1298 int i, count = 0; 1299 1300 rxhash = skb_get_hash(skb); 1301 for (i = 0; i < ROLLOVER_HLEN; i++) 1302 if (po->rollover->history[i] == rxhash) 1303 count++; 1304 1305 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash; 1306 return count > (ROLLOVER_HLEN >> 1); 1307 } 1308 1309 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1310 struct sk_buff *skb, 1311 unsigned int num) 1312 { 1313 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1314 } 1315 1316 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1317 struct sk_buff *skb, 1318 unsigned int num) 1319 { 1320 unsigned int val = atomic_inc_return(&f->rr_cur); 1321 1322 return val % num; 1323 } 1324 1325 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1326 struct sk_buff *skb, 1327 unsigned int num) 1328 { 1329 return smp_processor_id() % num; 1330 } 1331 1332 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1333 struct sk_buff *skb, 1334 unsigned int num) 1335 { 1336 return prandom_u32_max(num); 1337 } 1338 1339 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1340 struct sk_buff *skb, 1341 unsigned int idx, bool try_self, 1342 unsigned int num) 1343 { 1344 struct packet_sock *po, *po_next, *po_skip = NULL; 1345 unsigned int i, j, room = ROOM_NONE; 1346 1347 po = pkt_sk(f->arr[idx]); 1348 1349 if (try_self) { 1350 room = packet_rcv_has_room(po, skb); 1351 if (room == ROOM_NORMAL || 1352 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1353 return idx; 1354 po_skip = po; 1355 } 1356 1357 i = j = min_t(int, po->rollover->sock, num - 1); 1358 do { 1359 po_next = pkt_sk(f->arr[i]); 1360 if (po_next != po_skip && !po_next->pressure && 1361 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1362 if (i != j) 1363 po->rollover->sock = i; 1364 atomic_long_inc(&po->rollover->num); 1365 if (room == ROOM_LOW) 1366 atomic_long_inc(&po->rollover->num_huge); 1367 return i; 1368 } 1369 1370 if (++i == num) 1371 i = 0; 1372 } while (i != j); 1373 1374 atomic_long_inc(&po->rollover->num_failed); 1375 return idx; 1376 } 1377 1378 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1379 struct sk_buff *skb, 1380 unsigned int num) 1381 { 1382 return skb_get_queue_mapping(skb) % num; 1383 } 1384 1385 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1386 struct sk_buff *skb, 1387 unsigned int num) 1388 { 1389 struct bpf_prog *prog; 1390 unsigned int ret = 0; 1391 1392 rcu_read_lock(); 1393 prog = rcu_dereference(f->bpf_prog); 1394 if (prog) 1395 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1396 rcu_read_unlock(); 1397 1398 return ret; 1399 } 1400 1401 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1402 { 1403 return f->flags & (flag >> 8); 1404 } 1405 1406 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1407 struct packet_type *pt, struct net_device *orig_dev) 1408 { 1409 struct packet_fanout *f = pt->af_packet_priv; 1410 unsigned int num = READ_ONCE(f->num_members); 1411 struct net *net = read_pnet(&f->net); 1412 struct packet_sock *po; 1413 unsigned int idx; 1414 1415 if (!net_eq(dev_net(dev), net) || !num) { 1416 kfree_skb(skb); 1417 return 0; 1418 } 1419 1420 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1421 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1422 if (!skb) 1423 return 0; 1424 } 1425 switch (f->type) { 1426 case PACKET_FANOUT_HASH: 1427 default: 1428 idx = fanout_demux_hash(f, skb, num); 1429 break; 1430 case PACKET_FANOUT_LB: 1431 idx = fanout_demux_lb(f, skb, num); 1432 break; 1433 case PACKET_FANOUT_CPU: 1434 idx = fanout_demux_cpu(f, skb, num); 1435 break; 1436 case PACKET_FANOUT_RND: 1437 idx = fanout_demux_rnd(f, skb, num); 1438 break; 1439 case PACKET_FANOUT_QM: 1440 idx = fanout_demux_qm(f, skb, num); 1441 break; 1442 case PACKET_FANOUT_ROLLOVER: 1443 idx = fanout_demux_rollover(f, skb, 0, false, num); 1444 break; 1445 case PACKET_FANOUT_CBPF: 1446 case PACKET_FANOUT_EBPF: 1447 idx = fanout_demux_bpf(f, skb, num); 1448 break; 1449 } 1450 1451 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1452 idx = fanout_demux_rollover(f, skb, idx, true, num); 1453 1454 po = pkt_sk(f->arr[idx]); 1455 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1456 } 1457 1458 DEFINE_MUTEX(fanout_mutex); 1459 EXPORT_SYMBOL_GPL(fanout_mutex); 1460 static LIST_HEAD(fanout_list); 1461 static u16 fanout_next_id; 1462 1463 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1464 { 1465 struct packet_fanout *f = po->fanout; 1466 1467 spin_lock(&f->lock); 1468 f->arr[f->num_members] = sk; 1469 smp_wmb(); 1470 f->num_members++; 1471 if (f->num_members == 1) 1472 dev_add_pack(&f->prot_hook); 1473 spin_unlock(&f->lock); 1474 } 1475 1476 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1477 { 1478 struct packet_fanout *f = po->fanout; 1479 int i; 1480 1481 spin_lock(&f->lock); 1482 for (i = 0; i < f->num_members; i++) { 1483 if (f->arr[i] == sk) 1484 break; 1485 } 1486 BUG_ON(i >= f->num_members); 1487 f->arr[i] = f->arr[f->num_members - 1]; 1488 f->num_members--; 1489 if (f->num_members == 0) 1490 __dev_remove_pack(&f->prot_hook); 1491 spin_unlock(&f->lock); 1492 } 1493 1494 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1495 { 1496 if (sk->sk_family != PF_PACKET) 1497 return false; 1498 1499 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1500 } 1501 1502 static void fanout_init_data(struct packet_fanout *f) 1503 { 1504 switch (f->type) { 1505 case PACKET_FANOUT_LB: 1506 atomic_set(&f->rr_cur, 0); 1507 break; 1508 case PACKET_FANOUT_CBPF: 1509 case PACKET_FANOUT_EBPF: 1510 RCU_INIT_POINTER(f->bpf_prog, NULL); 1511 break; 1512 } 1513 } 1514 1515 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1516 { 1517 struct bpf_prog *old; 1518 1519 spin_lock(&f->lock); 1520 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1521 rcu_assign_pointer(f->bpf_prog, new); 1522 spin_unlock(&f->lock); 1523 1524 if (old) { 1525 synchronize_net(); 1526 bpf_prog_destroy(old); 1527 } 1528 } 1529 1530 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data, 1531 unsigned int len) 1532 { 1533 struct bpf_prog *new; 1534 struct sock_fprog fprog; 1535 int ret; 1536 1537 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1538 return -EPERM; 1539 if (len != sizeof(fprog)) 1540 return -EINVAL; 1541 if (copy_from_user(&fprog, data, len)) 1542 return -EFAULT; 1543 1544 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1545 if (ret) 1546 return ret; 1547 1548 __fanout_set_data_bpf(po->fanout, new); 1549 return 0; 1550 } 1551 1552 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data, 1553 unsigned int len) 1554 { 1555 struct bpf_prog *new; 1556 u32 fd; 1557 1558 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1559 return -EPERM; 1560 if (len != sizeof(fd)) 1561 return -EINVAL; 1562 if (copy_from_user(&fd, data, len)) 1563 return -EFAULT; 1564 1565 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1566 if (IS_ERR(new)) 1567 return PTR_ERR(new); 1568 1569 __fanout_set_data_bpf(po->fanout, new); 1570 return 0; 1571 } 1572 1573 static int fanout_set_data(struct packet_sock *po, char __user *data, 1574 unsigned int len) 1575 { 1576 switch (po->fanout->type) { 1577 case PACKET_FANOUT_CBPF: 1578 return fanout_set_data_cbpf(po, data, len); 1579 case PACKET_FANOUT_EBPF: 1580 return fanout_set_data_ebpf(po, data, len); 1581 default: 1582 return -EINVAL; 1583 } 1584 } 1585 1586 static void fanout_release_data(struct packet_fanout *f) 1587 { 1588 switch (f->type) { 1589 case PACKET_FANOUT_CBPF: 1590 case PACKET_FANOUT_EBPF: 1591 __fanout_set_data_bpf(f, NULL); 1592 } 1593 } 1594 1595 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) 1596 { 1597 struct packet_fanout *f; 1598 1599 list_for_each_entry(f, &fanout_list, list) { 1600 if (f->id == candidate_id && 1601 read_pnet(&f->net) == sock_net(sk)) { 1602 return false; 1603 } 1604 } 1605 return true; 1606 } 1607 1608 static bool fanout_find_new_id(struct sock *sk, u16 *new_id) 1609 { 1610 u16 id = fanout_next_id; 1611 1612 do { 1613 if (__fanout_id_is_free(sk, id)) { 1614 *new_id = id; 1615 fanout_next_id = id + 1; 1616 return true; 1617 } 1618 1619 id++; 1620 } while (id != fanout_next_id); 1621 1622 return false; 1623 } 1624 1625 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1626 { 1627 struct packet_rollover *rollover = NULL; 1628 struct packet_sock *po = pkt_sk(sk); 1629 struct packet_fanout *f, *match; 1630 u8 type = type_flags & 0xff; 1631 u8 flags = type_flags >> 8; 1632 int err; 1633 1634 switch (type) { 1635 case PACKET_FANOUT_ROLLOVER: 1636 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1637 return -EINVAL; 1638 case PACKET_FANOUT_HASH: 1639 case PACKET_FANOUT_LB: 1640 case PACKET_FANOUT_CPU: 1641 case PACKET_FANOUT_RND: 1642 case PACKET_FANOUT_QM: 1643 case PACKET_FANOUT_CBPF: 1644 case PACKET_FANOUT_EBPF: 1645 break; 1646 default: 1647 return -EINVAL; 1648 } 1649 1650 mutex_lock(&fanout_mutex); 1651 1652 err = -EALREADY; 1653 if (po->fanout) 1654 goto out; 1655 1656 if (type == PACKET_FANOUT_ROLLOVER || 1657 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1658 err = -ENOMEM; 1659 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); 1660 if (!rollover) 1661 goto out; 1662 atomic_long_set(&rollover->num, 0); 1663 atomic_long_set(&rollover->num_huge, 0); 1664 atomic_long_set(&rollover->num_failed, 0); 1665 } 1666 1667 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { 1668 if (id != 0) { 1669 err = -EINVAL; 1670 goto out; 1671 } 1672 if (!fanout_find_new_id(sk, &id)) { 1673 err = -ENOMEM; 1674 goto out; 1675 } 1676 /* ephemeral flag for the first socket in the group: drop it */ 1677 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); 1678 } 1679 1680 match = NULL; 1681 list_for_each_entry(f, &fanout_list, list) { 1682 if (f->id == id && 1683 read_pnet(&f->net) == sock_net(sk)) { 1684 match = f; 1685 break; 1686 } 1687 } 1688 err = -EINVAL; 1689 if (match && match->flags != flags) 1690 goto out; 1691 if (!match) { 1692 err = -ENOMEM; 1693 match = kzalloc(sizeof(*match), GFP_KERNEL); 1694 if (!match) 1695 goto out; 1696 write_pnet(&match->net, sock_net(sk)); 1697 match->id = id; 1698 match->type = type; 1699 match->flags = flags; 1700 INIT_LIST_HEAD(&match->list); 1701 spin_lock_init(&match->lock); 1702 refcount_set(&match->sk_ref, 0); 1703 fanout_init_data(match); 1704 match->prot_hook.type = po->prot_hook.type; 1705 match->prot_hook.dev = po->prot_hook.dev; 1706 match->prot_hook.func = packet_rcv_fanout; 1707 match->prot_hook.af_packet_priv = match; 1708 match->prot_hook.id_match = match_fanout_group; 1709 list_add(&match->list, &fanout_list); 1710 } 1711 err = -EINVAL; 1712 1713 spin_lock(&po->bind_lock); 1714 if (po->running && 1715 match->type == type && 1716 match->prot_hook.type == po->prot_hook.type && 1717 match->prot_hook.dev == po->prot_hook.dev) { 1718 err = -ENOSPC; 1719 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1720 __dev_remove_pack(&po->prot_hook); 1721 po->fanout = match; 1722 po->rollover = rollover; 1723 rollover = NULL; 1724 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); 1725 __fanout_link(sk, po); 1726 err = 0; 1727 } 1728 } 1729 spin_unlock(&po->bind_lock); 1730 1731 if (err && !refcount_read(&match->sk_ref)) { 1732 list_del(&match->list); 1733 kfree(match); 1734 } 1735 1736 out: 1737 kfree(rollover); 1738 mutex_unlock(&fanout_mutex); 1739 return err; 1740 } 1741 1742 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes 1743 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. 1744 * It is the responsibility of the caller to call fanout_release_data() and 1745 * free the returned packet_fanout (after synchronize_net()) 1746 */ 1747 static struct packet_fanout *fanout_release(struct sock *sk) 1748 { 1749 struct packet_sock *po = pkt_sk(sk); 1750 struct packet_fanout *f; 1751 1752 mutex_lock(&fanout_mutex); 1753 f = po->fanout; 1754 if (f) { 1755 po->fanout = NULL; 1756 1757 if (refcount_dec_and_test(&f->sk_ref)) 1758 list_del(&f->list); 1759 else 1760 f = NULL; 1761 } 1762 mutex_unlock(&fanout_mutex); 1763 1764 return f; 1765 } 1766 1767 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1768 struct sk_buff *skb) 1769 { 1770 /* Earlier code assumed this would be a VLAN pkt, double-check 1771 * this now that we have the actual packet in hand. We can only 1772 * do this check on Ethernet devices. 1773 */ 1774 if (unlikely(dev->type != ARPHRD_ETHER)) 1775 return false; 1776 1777 skb_reset_mac_header(skb); 1778 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1779 } 1780 1781 static const struct proto_ops packet_ops; 1782 1783 static const struct proto_ops packet_ops_spkt; 1784 1785 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1786 struct packet_type *pt, struct net_device *orig_dev) 1787 { 1788 struct sock *sk; 1789 struct sockaddr_pkt *spkt; 1790 1791 /* 1792 * When we registered the protocol we saved the socket in the data 1793 * field for just this event. 1794 */ 1795 1796 sk = pt->af_packet_priv; 1797 1798 /* 1799 * Yank back the headers [hope the device set this 1800 * right or kerboom...] 1801 * 1802 * Incoming packets have ll header pulled, 1803 * push it back. 1804 * 1805 * For outgoing ones skb->data == skb_mac_header(skb) 1806 * so that this procedure is noop. 1807 */ 1808 1809 if (skb->pkt_type == PACKET_LOOPBACK) 1810 goto out; 1811 1812 if (!net_eq(dev_net(dev), sock_net(sk))) 1813 goto out; 1814 1815 skb = skb_share_check(skb, GFP_ATOMIC); 1816 if (skb == NULL) 1817 goto oom; 1818 1819 /* drop any routing info */ 1820 skb_dst_drop(skb); 1821 1822 /* drop conntrack reference */ 1823 nf_reset(skb); 1824 1825 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1826 1827 skb_push(skb, skb->data - skb_mac_header(skb)); 1828 1829 /* 1830 * The SOCK_PACKET socket receives _all_ frames. 1831 */ 1832 1833 spkt->spkt_family = dev->type; 1834 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1835 spkt->spkt_protocol = skb->protocol; 1836 1837 /* 1838 * Charge the memory to the socket. This is done specifically 1839 * to prevent sockets using all the memory up. 1840 */ 1841 1842 if (sock_queue_rcv_skb(sk, skb) == 0) 1843 return 0; 1844 1845 out: 1846 kfree_skb(skb); 1847 oom: 1848 return 0; 1849 } 1850 1851 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock) 1852 { 1853 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && 1854 sock->type == SOCK_RAW) { 1855 skb_reset_mac_header(skb); 1856 skb->protocol = dev_parse_header_protocol(skb); 1857 } 1858 1859 skb_probe_transport_header(skb); 1860 } 1861 1862 /* 1863 * Output a raw packet to a device layer. This bypasses all the other 1864 * protocol layers and you must therefore supply it with a complete frame 1865 */ 1866 1867 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1868 size_t len) 1869 { 1870 struct sock *sk = sock->sk; 1871 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1872 struct sk_buff *skb = NULL; 1873 struct net_device *dev; 1874 struct sockcm_cookie sockc; 1875 __be16 proto = 0; 1876 int err; 1877 int extra_len = 0; 1878 1879 /* 1880 * Get and verify the address. 1881 */ 1882 1883 if (saddr) { 1884 if (msg->msg_namelen < sizeof(struct sockaddr)) 1885 return -EINVAL; 1886 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1887 proto = saddr->spkt_protocol; 1888 } else 1889 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1890 1891 /* 1892 * Find the device first to size check it 1893 */ 1894 1895 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1896 retry: 1897 rcu_read_lock(); 1898 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1899 err = -ENODEV; 1900 if (dev == NULL) 1901 goto out_unlock; 1902 1903 err = -ENETDOWN; 1904 if (!(dev->flags & IFF_UP)) 1905 goto out_unlock; 1906 1907 /* 1908 * You may not queue a frame bigger than the mtu. This is the lowest level 1909 * raw protocol and you must do your own fragmentation at this level. 1910 */ 1911 1912 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1913 if (!netif_supports_nofcs(dev)) { 1914 err = -EPROTONOSUPPORT; 1915 goto out_unlock; 1916 } 1917 extra_len = 4; /* We're doing our own CRC */ 1918 } 1919 1920 err = -EMSGSIZE; 1921 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1922 goto out_unlock; 1923 1924 if (!skb) { 1925 size_t reserved = LL_RESERVED_SPACE(dev); 1926 int tlen = dev->needed_tailroom; 1927 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1928 1929 rcu_read_unlock(); 1930 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1931 if (skb == NULL) 1932 return -ENOBUFS; 1933 /* FIXME: Save some space for broken drivers that write a hard 1934 * header at transmission time by themselves. PPP is the notable 1935 * one here. This should really be fixed at the driver level. 1936 */ 1937 skb_reserve(skb, reserved); 1938 skb_reset_network_header(skb); 1939 1940 /* Try to align data part correctly */ 1941 if (hhlen) { 1942 skb->data -= hhlen; 1943 skb->tail -= hhlen; 1944 if (len < hhlen) 1945 skb_reset_network_header(skb); 1946 } 1947 err = memcpy_from_msg(skb_put(skb, len), msg, len); 1948 if (err) 1949 goto out_free; 1950 goto retry; 1951 } 1952 1953 if (!dev_validate_header(dev, skb->data, len)) { 1954 err = -EINVAL; 1955 goto out_unlock; 1956 } 1957 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 1958 !packet_extra_vlan_len_allowed(dev, skb)) { 1959 err = -EMSGSIZE; 1960 goto out_unlock; 1961 } 1962 1963 sockcm_init(&sockc, sk); 1964 if (msg->msg_controllen) { 1965 err = sock_cmsg_send(sk, msg, &sockc); 1966 if (unlikely(err)) 1967 goto out_unlock; 1968 } 1969 1970 skb->protocol = proto; 1971 skb->dev = dev; 1972 skb->priority = sk->sk_priority; 1973 skb->mark = sk->sk_mark; 1974 skb->tstamp = sockc.transmit_time; 1975 1976 skb_setup_tx_timestamp(skb, sockc.tsflags); 1977 1978 if (unlikely(extra_len == 4)) 1979 skb->no_fcs = 1; 1980 1981 packet_parse_headers(skb, sock); 1982 1983 dev_queue_xmit(skb); 1984 rcu_read_unlock(); 1985 return len; 1986 1987 out_unlock: 1988 rcu_read_unlock(); 1989 out_free: 1990 kfree_skb(skb); 1991 return err; 1992 } 1993 1994 static unsigned int run_filter(struct sk_buff *skb, 1995 const struct sock *sk, 1996 unsigned int res) 1997 { 1998 struct sk_filter *filter; 1999 2000 rcu_read_lock(); 2001 filter = rcu_dereference(sk->sk_filter); 2002 if (filter != NULL) 2003 res = bpf_prog_run_clear_cb(filter->prog, skb); 2004 rcu_read_unlock(); 2005 2006 return res; 2007 } 2008 2009 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2010 size_t *len) 2011 { 2012 struct virtio_net_hdr vnet_hdr; 2013 2014 if (*len < sizeof(vnet_hdr)) 2015 return -EINVAL; 2016 *len -= sizeof(vnet_hdr); 2017 2018 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0)) 2019 return -EINVAL; 2020 2021 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 2022 } 2023 2024 /* 2025 * This function makes lazy skb cloning in hope that most of packets 2026 * are discarded by BPF. 2027 * 2028 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2029 * and skb->cb are mangled. It works because (and until) packets 2030 * falling here are owned by current CPU. Output packets are cloned 2031 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2032 * sequencially, so that if we return skb to original state on exit, 2033 * we will not harm anyone. 2034 */ 2035 2036 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2037 struct packet_type *pt, struct net_device *orig_dev) 2038 { 2039 struct sock *sk; 2040 struct sockaddr_ll *sll; 2041 struct packet_sock *po; 2042 u8 *skb_head = skb->data; 2043 int skb_len = skb->len; 2044 unsigned int snaplen, res; 2045 bool is_drop_n_account = false; 2046 2047 if (skb->pkt_type == PACKET_LOOPBACK) 2048 goto drop; 2049 2050 sk = pt->af_packet_priv; 2051 po = pkt_sk(sk); 2052 2053 if (!net_eq(dev_net(dev), sock_net(sk))) 2054 goto drop; 2055 2056 skb->dev = dev; 2057 2058 if (dev->header_ops) { 2059 /* The device has an explicit notion of ll header, 2060 * exported to higher levels. 2061 * 2062 * Otherwise, the device hides details of its frame 2063 * structure, so that corresponding packet head is 2064 * never delivered to user. 2065 */ 2066 if (sk->sk_type != SOCK_DGRAM) 2067 skb_push(skb, skb->data - skb_mac_header(skb)); 2068 else if (skb->pkt_type == PACKET_OUTGOING) { 2069 /* Special case: outgoing packets have ll header at head */ 2070 skb_pull(skb, skb_network_offset(skb)); 2071 } 2072 } 2073 2074 snaplen = skb->len; 2075 2076 res = run_filter(skb, sk, snaplen); 2077 if (!res) 2078 goto drop_n_restore; 2079 if (snaplen > res) 2080 snaplen = res; 2081 2082 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2083 goto drop_n_acct; 2084 2085 if (skb_shared(skb)) { 2086 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2087 if (nskb == NULL) 2088 goto drop_n_acct; 2089 2090 if (skb_head != skb->data) { 2091 skb->data = skb_head; 2092 skb->len = skb_len; 2093 } 2094 consume_skb(skb); 2095 skb = nskb; 2096 } 2097 2098 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2099 2100 sll = &PACKET_SKB_CB(skb)->sa.ll; 2101 sll->sll_hatype = dev->type; 2102 sll->sll_pkttype = skb->pkt_type; 2103 if (unlikely(po->origdev)) 2104 sll->sll_ifindex = orig_dev->ifindex; 2105 else 2106 sll->sll_ifindex = dev->ifindex; 2107 2108 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2109 2110 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2111 * Use their space for storing the original skb length. 2112 */ 2113 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2114 2115 if (pskb_trim(skb, snaplen)) 2116 goto drop_n_acct; 2117 2118 skb_set_owner_r(skb, sk); 2119 skb->dev = NULL; 2120 skb_dst_drop(skb); 2121 2122 /* drop conntrack reference */ 2123 nf_reset(skb); 2124 2125 spin_lock(&sk->sk_receive_queue.lock); 2126 po->stats.stats1.tp_packets++; 2127 sock_skb_set_dropcount(sk, skb); 2128 __skb_queue_tail(&sk->sk_receive_queue, skb); 2129 spin_unlock(&sk->sk_receive_queue.lock); 2130 sk->sk_data_ready(sk); 2131 return 0; 2132 2133 drop_n_acct: 2134 is_drop_n_account = true; 2135 spin_lock(&sk->sk_receive_queue.lock); 2136 po->stats.stats1.tp_drops++; 2137 atomic_inc(&sk->sk_drops); 2138 spin_unlock(&sk->sk_receive_queue.lock); 2139 2140 drop_n_restore: 2141 if (skb_head != skb->data && skb_shared(skb)) { 2142 skb->data = skb_head; 2143 skb->len = skb_len; 2144 } 2145 drop: 2146 if (!is_drop_n_account) 2147 consume_skb(skb); 2148 else 2149 kfree_skb(skb); 2150 return 0; 2151 } 2152 2153 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2154 struct packet_type *pt, struct net_device *orig_dev) 2155 { 2156 struct sock *sk; 2157 struct packet_sock *po; 2158 struct sockaddr_ll *sll; 2159 union tpacket_uhdr h; 2160 u8 *skb_head = skb->data; 2161 int skb_len = skb->len; 2162 unsigned int snaplen, res; 2163 unsigned long status = TP_STATUS_USER; 2164 unsigned short macoff, netoff, hdrlen; 2165 struct sk_buff *copy_skb = NULL; 2166 struct timespec ts; 2167 __u32 ts_status; 2168 bool is_drop_n_account = false; 2169 bool do_vnet = false; 2170 2171 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2172 * We may add members to them until current aligned size without forcing 2173 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2174 */ 2175 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2176 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2177 2178 if (skb->pkt_type == PACKET_LOOPBACK) 2179 goto drop; 2180 2181 sk = pt->af_packet_priv; 2182 po = pkt_sk(sk); 2183 2184 if (!net_eq(dev_net(dev), sock_net(sk))) 2185 goto drop; 2186 2187 if (dev->header_ops) { 2188 if (sk->sk_type != SOCK_DGRAM) 2189 skb_push(skb, skb->data - skb_mac_header(skb)); 2190 else if (skb->pkt_type == PACKET_OUTGOING) { 2191 /* Special case: outgoing packets have ll header at head */ 2192 skb_pull(skb, skb_network_offset(skb)); 2193 } 2194 } 2195 2196 snaplen = skb->len; 2197 2198 res = run_filter(skb, sk, snaplen); 2199 if (!res) 2200 goto drop_n_restore; 2201 2202 if (skb->ip_summed == CHECKSUM_PARTIAL) 2203 status |= TP_STATUS_CSUMNOTREADY; 2204 else if (skb->pkt_type != PACKET_OUTGOING && 2205 (skb->ip_summed == CHECKSUM_COMPLETE || 2206 skb_csum_unnecessary(skb))) 2207 status |= TP_STATUS_CSUM_VALID; 2208 2209 if (snaplen > res) 2210 snaplen = res; 2211 2212 if (sk->sk_type == SOCK_DGRAM) { 2213 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2214 po->tp_reserve; 2215 } else { 2216 unsigned int maclen = skb_network_offset(skb); 2217 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2218 (maclen < 16 ? 16 : maclen)) + 2219 po->tp_reserve; 2220 if (po->has_vnet_hdr) { 2221 netoff += sizeof(struct virtio_net_hdr); 2222 do_vnet = true; 2223 } 2224 macoff = netoff - maclen; 2225 } 2226 if (po->tp_version <= TPACKET_V2) { 2227 if (macoff + snaplen > po->rx_ring.frame_size) { 2228 if (po->copy_thresh && 2229 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2230 if (skb_shared(skb)) { 2231 copy_skb = skb_clone(skb, GFP_ATOMIC); 2232 } else { 2233 copy_skb = skb_get(skb); 2234 skb_head = skb->data; 2235 } 2236 if (copy_skb) 2237 skb_set_owner_r(copy_skb, sk); 2238 } 2239 snaplen = po->rx_ring.frame_size - macoff; 2240 if ((int)snaplen < 0) { 2241 snaplen = 0; 2242 do_vnet = false; 2243 } 2244 } 2245 } else if (unlikely(macoff + snaplen > 2246 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2247 u32 nval; 2248 2249 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2250 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2251 snaplen, nval, macoff); 2252 snaplen = nval; 2253 if (unlikely((int)snaplen < 0)) { 2254 snaplen = 0; 2255 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2256 do_vnet = false; 2257 } 2258 } 2259 spin_lock(&sk->sk_receive_queue.lock); 2260 h.raw = packet_current_rx_frame(po, skb, 2261 TP_STATUS_KERNEL, (macoff+snaplen)); 2262 if (!h.raw) 2263 goto drop_n_account; 2264 if (po->tp_version <= TPACKET_V2) { 2265 packet_increment_rx_head(po, &po->rx_ring); 2266 /* 2267 * LOSING will be reported till you read the stats, 2268 * because it's COR - Clear On Read. 2269 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2270 * at packet level. 2271 */ 2272 if (po->stats.stats1.tp_drops) 2273 status |= TP_STATUS_LOSING; 2274 } 2275 2276 if (do_vnet && 2277 virtio_net_hdr_from_skb(skb, h.raw + macoff - 2278 sizeof(struct virtio_net_hdr), 2279 vio_le(), true, 0)) 2280 goto drop_n_account; 2281 2282 po->stats.stats1.tp_packets++; 2283 if (copy_skb) { 2284 status |= TP_STATUS_COPY; 2285 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2286 } 2287 spin_unlock(&sk->sk_receive_queue.lock); 2288 2289 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2290 2291 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 2292 getnstimeofday(&ts); 2293 2294 status |= ts_status; 2295 2296 switch (po->tp_version) { 2297 case TPACKET_V1: 2298 h.h1->tp_len = skb->len; 2299 h.h1->tp_snaplen = snaplen; 2300 h.h1->tp_mac = macoff; 2301 h.h1->tp_net = netoff; 2302 h.h1->tp_sec = ts.tv_sec; 2303 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2304 hdrlen = sizeof(*h.h1); 2305 break; 2306 case TPACKET_V2: 2307 h.h2->tp_len = skb->len; 2308 h.h2->tp_snaplen = snaplen; 2309 h.h2->tp_mac = macoff; 2310 h.h2->tp_net = netoff; 2311 h.h2->tp_sec = ts.tv_sec; 2312 h.h2->tp_nsec = ts.tv_nsec; 2313 if (skb_vlan_tag_present(skb)) { 2314 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2315 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2316 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2317 } else { 2318 h.h2->tp_vlan_tci = 0; 2319 h.h2->tp_vlan_tpid = 0; 2320 } 2321 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2322 hdrlen = sizeof(*h.h2); 2323 break; 2324 case TPACKET_V3: 2325 /* tp_nxt_offset,vlan are already populated above. 2326 * So DONT clear those fields here 2327 */ 2328 h.h3->tp_status |= status; 2329 h.h3->tp_len = skb->len; 2330 h.h3->tp_snaplen = snaplen; 2331 h.h3->tp_mac = macoff; 2332 h.h3->tp_net = netoff; 2333 h.h3->tp_sec = ts.tv_sec; 2334 h.h3->tp_nsec = ts.tv_nsec; 2335 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2336 hdrlen = sizeof(*h.h3); 2337 break; 2338 default: 2339 BUG(); 2340 } 2341 2342 sll = h.raw + TPACKET_ALIGN(hdrlen); 2343 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2344 sll->sll_family = AF_PACKET; 2345 sll->sll_hatype = dev->type; 2346 sll->sll_protocol = skb->protocol; 2347 sll->sll_pkttype = skb->pkt_type; 2348 if (unlikely(po->origdev)) 2349 sll->sll_ifindex = orig_dev->ifindex; 2350 else 2351 sll->sll_ifindex = dev->ifindex; 2352 2353 smp_mb(); 2354 2355 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2356 if (po->tp_version <= TPACKET_V2) { 2357 u8 *start, *end; 2358 2359 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2360 macoff + snaplen); 2361 2362 for (start = h.raw; start < end; start += PAGE_SIZE) 2363 flush_dcache_page(pgv_to_page(start)); 2364 } 2365 smp_wmb(); 2366 #endif 2367 2368 if (po->tp_version <= TPACKET_V2) { 2369 __packet_set_status(po, h.raw, status); 2370 sk->sk_data_ready(sk); 2371 } else { 2372 prb_clear_blk_fill_status(&po->rx_ring); 2373 } 2374 2375 drop_n_restore: 2376 if (skb_head != skb->data && skb_shared(skb)) { 2377 skb->data = skb_head; 2378 skb->len = skb_len; 2379 } 2380 drop: 2381 if (!is_drop_n_account) 2382 consume_skb(skb); 2383 else 2384 kfree_skb(skb); 2385 return 0; 2386 2387 drop_n_account: 2388 is_drop_n_account = true; 2389 po->stats.stats1.tp_drops++; 2390 spin_unlock(&sk->sk_receive_queue.lock); 2391 2392 sk->sk_data_ready(sk); 2393 kfree_skb(copy_skb); 2394 goto drop_n_restore; 2395 } 2396 2397 static void tpacket_destruct_skb(struct sk_buff *skb) 2398 { 2399 struct packet_sock *po = pkt_sk(skb->sk); 2400 2401 if (likely(po->tx_ring.pg_vec)) { 2402 void *ph; 2403 __u32 ts; 2404 2405 ph = skb_zcopy_get_nouarg(skb); 2406 packet_dec_pending(&po->tx_ring); 2407 2408 ts = __packet_set_timestamp(po, ph, skb); 2409 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2410 } 2411 2412 sock_wfree(skb); 2413 } 2414 2415 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2416 { 2417 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2418 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2419 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2420 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2421 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2422 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2423 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2424 2425 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2426 return -EINVAL; 2427 2428 return 0; 2429 } 2430 2431 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2432 struct virtio_net_hdr *vnet_hdr) 2433 { 2434 if (*len < sizeof(*vnet_hdr)) 2435 return -EINVAL; 2436 *len -= sizeof(*vnet_hdr); 2437 2438 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2439 return -EFAULT; 2440 2441 return __packet_snd_vnet_parse(vnet_hdr, *len); 2442 } 2443 2444 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2445 void *frame, struct net_device *dev, void *data, int tp_len, 2446 __be16 proto, unsigned char *addr, int hlen, int copylen, 2447 const struct sockcm_cookie *sockc) 2448 { 2449 union tpacket_uhdr ph; 2450 int to_write, offset, len, nr_frags, len_max; 2451 struct socket *sock = po->sk.sk_socket; 2452 struct page *page; 2453 int err; 2454 2455 ph.raw = frame; 2456 2457 skb->protocol = proto; 2458 skb->dev = dev; 2459 skb->priority = po->sk.sk_priority; 2460 skb->mark = po->sk.sk_mark; 2461 skb->tstamp = sockc->transmit_time; 2462 skb_setup_tx_timestamp(skb, sockc->tsflags); 2463 skb_zcopy_set_nouarg(skb, ph.raw); 2464 2465 skb_reserve(skb, hlen); 2466 skb_reset_network_header(skb); 2467 2468 to_write = tp_len; 2469 2470 if (sock->type == SOCK_DGRAM) { 2471 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2472 NULL, tp_len); 2473 if (unlikely(err < 0)) 2474 return -EINVAL; 2475 } else if (copylen) { 2476 int hdrlen = min_t(int, copylen, tp_len); 2477 2478 skb_push(skb, dev->hard_header_len); 2479 skb_put(skb, copylen - dev->hard_header_len); 2480 err = skb_store_bits(skb, 0, data, hdrlen); 2481 if (unlikely(err)) 2482 return err; 2483 if (!dev_validate_header(dev, skb->data, hdrlen)) 2484 return -EINVAL; 2485 2486 data += hdrlen; 2487 to_write -= hdrlen; 2488 } 2489 2490 offset = offset_in_page(data); 2491 len_max = PAGE_SIZE - offset; 2492 len = ((to_write > len_max) ? len_max : to_write); 2493 2494 skb->data_len = to_write; 2495 skb->len += to_write; 2496 skb->truesize += to_write; 2497 refcount_add(to_write, &po->sk.sk_wmem_alloc); 2498 2499 while (likely(to_write)) { 2500 nr_frags = skb_shinfo(skb)->nr_frags; 2501 2502 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2503 pr_err("Packet exceed the number of skb frags(%lu)\n", 2504 MAX_SKB_FRAGS); 2505 return -EFAULT; 2506 } 2507 2508 page = pgv_to_page(data); 2509 data += len; 2510 flush_dcache_page(page); 2511 get_page(page); 2512 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2513 to_write -= len; 2514 offset = 0; 2515 len_max = PAGE_SIZE; 2516 len = ((to_write > len_max) ? len_max : to_write); 2517 } 2518 2519 packet_parse_headers(skb, sock); 2520 2521 return tp_len; 2522 } 2523 2524 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2525 int size_max, void **data) 2526 { 2527 union tpacket_uhdr ph; 2528 int tp_len, off; 2529 2530 ph.raw = frame; 2531 2532 switch (po->tp_version) { 2533 case TPACKET_V3: 2534 if (ph.h3->tp_next_offset != 0) { 2535 pr_warn_once("variable sized slot not supported"); 2536 return -EINVAL; 2537 } 2538 tp_len = ph.h3->tp_len; 2539 break; 2540 case TPACKET_V2: 2541 tp_len = ph.h2->tp_len; 2542 break; 2543 default: 2544 tp_len = ph.h1->tp_len; 2545 break; 2546 } 2547 if (unlikely(tp_len > size_max)) { 2548 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2549 return -EMSGSIZE; 2550 } 2551 2552 if (unlikely(po->tp_tx_has_off)) { 2553 int off_min, off_max; 2554 2555 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2556 off_max = po->tx_ring.frame_size - tp_len; 2557 if (po->sk.sk_type == SOCK_DGRAM) { 2558 switch (po->tp_version) { 2559 case TPACKET_V3: 2560 off = ph.h3->tp_net; 2561 break; 2562 case TPACKET_V2: 2563 off = ph.h2->tp_net; 2564 break; 2565 default: 2566 off = ph.h1->tp_net; 2567 break; 2568 } 2569 } else { 2570 switch (po->tp_version) { 2571 case TPACKET_V3: 2572 off = ph.h3->tp_mac; 2573 break; 2574 case TPACKET_V2: 2575 off = ph.h2->tp_mac; 2576 break; 2577 default: 2578 off = ph.h1->tp_mac; 2579 break; 2580 } 2581 } 2582 if (unlikely((off < off_min) || (off_max < off))) 2583 return -EINVAL; 2584 } else { 2585 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2586 } 2587 2588 *data = frame + off; 2589 return tp_len; 2590 } 2591 2592 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2593 { 2594 struct sk_buff *skb; 2595 struct net_device *dev; 2596 struct virtio_net_hdr *vnet_hdr = NULL; 2597 struct sockcm_cookie sockc; 2598 __be16 proto; 2599 int err, reserve = 0; 2600 void *ph; 2601 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2602 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2603 unsigned char *addr = NULL; 2604 int tp_len, size_max; 2605 void *data; 2606 int len_sum = 0; 2607 int status = TP_STATUS_AVAILABLE; 2608 int hlen, tlen, copylen = 0; 2609 2610 mutex_lock(&po->pg_vec_lock); 2611 2612 if (likely(saddr == NULL)) { 2613 dev = packet_cached_dev_get(po); 2614 proto = po->num; 2615 } else { 2616 err = -EINVAL; 2617 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2618 goto out; 2619 if (msg->msg_namelen < (saddr->sll_halen 2620 + offsetof(struct sockaddr_ll, 2621 sll_addr))) 2622 goto out; 2623 proto = saddr->sll_protocol; 2624 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2625 if (po->sk.sk_socket->type == SOCK_DGRAM) { 2626 if (dev && msg->msg_namelen < dev->addr_len + 2627 offsetof(struct sockaddr_ll, sll_addr)) 2628 goto out_put; 2629 addr = saddr->sll_addr; 2630 } 2631 } 2632 2633 err = -ENXIO; 2634 if (unlikely(dev == NULL)) 2635 goto out; 2636 err = -ENETDOWN; 2637 if (unlikely(!(dev->flags & IFF_UP))) 2638 goto out_put; 2639 2640 sockcm_init(&sockc, &po->sk); 2641 if (msg->msg_controllen) { 2642 err = sock_cmsg_send(&po->sk, msg, &sockc); 2643 if (unlikely(err)) 2644 goto out_put; 2645 } 2646 2647 if (po->sk.sk_socket->type == SOCK_RAW) 2648 reserve = dev->hard_header_len; 2649 size_max = po->tx_ring.frame_size 2650 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2651 2652 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) 2653 size_max = dev->mtu + reserve + VLAN_HLEN; 2654 2655 do { 2656 ph = packet_current_frame(po, &po->tx_ring, 2657 TP_STATUS_SEND_REQUEST); 2658 if (unlikely(ph == NULL)) { 2659 if (need_wait && need_resched()) 2660 schedule(); 2661 continue; 2662 } 2663 2664 skb = NULL; 2665 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2666 if (tp_len < 0) 2667 goto tpacket_error; 2668 2669 status = TP_STATUS_SEND_REQUEST; 2670 hlen = LL_RESERVED_SPACE(dev); 2671 tlen = dev->needed_tailroom; 2672 if (po->has_vnet_hdr) { 2673 vnet_hdr = data; 2674 data += sizeof(*vnet_hdr); 2675 tp_len -= sizeof(*vnet_hdr); 2676 if (tp_len < 0 || 2677 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2678 tp_len = -EINVAL; 2679 goto tpacket_error; 2680 } 2681 copylen = __virtio16_to_cpu(vio_le(), 2682 vnet_hdr->hdr_len); 2683 } 2684 copylen = max_t(int, copylen, dev->hard_header_len); 2685 skb = sock_alloc_send_skb(&po->sk, 2686 hlen + tlen + sizeof(struct sockaddr_ll) + 2687 (copylen - dev->hard_header_len), 2688 !need_wait, &err); 2689 2690 if (unlikely(skb == NULL)) { 2691 /* we assume the socket was initially writeable ... */ 2692 if (likely(len_sum > 0)) 2693 err = len_sum; 2694 goto out_status; 2695 } 2696 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2697 addr, hlen, copylen, &sockc); 2698 if (likely(tp_len >= 0) && 2699 tp_len > dev->mtu + reserve && 2700 !po->has_vnet_hdr && 2701 !packet_extra_vlan_len_allowed(dev, skb)) 2702 tp_len = -EMSGSIZE; 2703 2704 if (unlikely(tp_len < 0)) { 2705 tpacket_error: 2706 if (po->tp_loss) { 2707 __packet_set_status(po, ph, 2708 TP_STATUS_AVAILABLE); 2709 packet_increment_head(&po->tx_ring); 2710 kfree_skb(skb); 2711 continue; 2712 } else { 2713 status = TP_STATUS_WRONG_FORMAT; 2714 err = tp_len; 2715 goto out_status; 2716 } 2717 } 2718 2719 if (po->has_vnet_hdr) { 2720 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) { 2721 tp_len = -EINVAL; 2722 goto tpacket_error; 2723 } 2724 virtio_net_hdr_set_proto(skb, vnet_hdr); 2725 } 2726 2727 skb->destructor = tpacket_destruct_skb; 2728 __packet_set_status(po, ph, TP_STATUS_SENDING); 2729 packet_inc_pending(&po->tx_ring); 2730 2731 status = TP_STATUS_SEND_REQUEST; 2732 err = po->xmit(skb); 2733 if (unlikely(err > 0)) { 2734 err = net_xmit_errno(err); 2735 if (err && __packet_get_status(po, ph) == 2736 TP_STATUS_AVAILABLE) { 2737 /* skb was destructed already */ 2738 skb = NULL; 2739 goto out_status; 2740 } 2741 /* 2742 * skb was dropped but not destructed yet; 2743 * let's treat it like congestion or err < 0 2744 */ 2745 err = 0; 2746 } 2747 packet_increment_head(&po->tx_ring); 2748 len_sum += tp_len; 2749 } while (likely((ph != NULL) || 2750 /* Note: packet_read_pending() might be slow if we have 2751 * to call it as it's per_cpu variable, but in fast-path 2752 * we already short-circuit the loop with the first 2753 * condition, and luckily don't have to go that path 2754 * anyway. 2755 */ 2756 (need_wait && packet_read_pending(&po->tx_ring)))); 2757 2758 err = len_sum; 2759 goto out_put; 2760 2761 out_status: 2762 __packet_set_status(po, ph, status); 2763 kfree_skb(skb); 2764 out_put: 2765 dev_put(dev); 2766 out: 2767 mutex_unlock(&po->pg_vec_lock); 2768 return err; 2769 } 2770 2771 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2772 size_t reserve, size_t len, 2773 size_t linear, int noblock, 2774 int *err) 2775 { 2776 struct sk_buff *skb; 2777 2778 /* Under a page? Don't bother with paged skb. */ 2779 if (prepad + len < PAGE_SIZE || !linear) 2780 linear = len; 2781 2782 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2783 err, 0); 2784 if (!skb) 2785 return NULL; 2786 2787 skb_reserve(skb, reserve); 2788 skb_put(skb, linear); 2789 skb->data_len = len - linear; 2790 skb->len += len - linear; 2791 2792 return skb; 2793 } 2794 2795 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2796 { 2797 struct sock *sk = sock->sk; 2798 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2799 struct sk_buff *skb; 2800 struct net_device *dev; 2801 __be16 proto; 2802 unsigned char *addr = NULL; 2803 int err, reserve = 0; 2804 struct sockcm_cookie sockc; 2805 struct virtio_net_hdr vnet_hdr = { 0 }; 2806 int offset = 0; 2807 struct packet_sock *po = pkt_sk(sk); 2808 bool has_vnet_hdr = false; 2809 int hlen, tlen, linear; 2810 int extra_len = 0; 2811 2812 /* 2813 * Get and verify the address. 2814 */ 2815 2816 if (likely(saddr == NULL)) { 2817 dev = packet_cached_dev_get(po); 2818 proto = po->num; 2819 } else { 2820 err = -EINVAL; 2821 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2822 goto out; 2823 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2824 goto out; 2825 proto = saddr->sll_protocol; 2826 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2827 if (sock->type == SOCK_DGRAM) { 2828 if (dev && msg->msg_namelen < dev->addr_len + 2829 offsetof(struct sockaddr_ll, sll_addr)) 2830 goto out_unlock; 2831 addr = saddr->sll_addr; 2832 } 2833 } 2834 2835 err = -ENXIO; 2836 if (unlikely(dev == NULL)) 2837 goto out_unlock; 2838 err = -ENETDOWN; 2839 if (unlikely(!(dev->flags & IFF_UP))) 2840 goto out_unlock; 2841 2842 sockcm_init(&sockc, sk); 2843 sockc.mark = sk->sk_mark; 2844 if (msg->msg_controllen) { 2845 err = sock_cmsg_send(sk, msg, &sockc); 2846 if (unlikely(err)) 2847 goto out_unlock; 2848 } 2849 2850 if (sock->type == SOCK_RAW) 2851 reserve = dev->hard_header_len; 2852 if (po->has_vnet_hdr) { 2853 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2854 if (err) 2855 goto out_unlock; 2856 has_vnet_hdr = true; 2857 } 2858 2859 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2860 if (!netif_supports_nofcs(dev)) { 2861 err = -EPROTONOSUPPORT; 2862 goto out_unlock; 2863 } 2864 extra_len = 4; /* We're doing our own CRC */ 2865 } 2866 2867 err = -EMSGSIZE; 2868 if (!vnet_hdr.gso_type && 2869 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2870 goto out_unlock; 2871 2872 err = -ENOBUFS; 2873 hlen = LL_RESERVED_SPACE(dev); 2874 tlen = dev->needed_tailroom; 2875 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); 2876 linear = max(linear, min_t(int, len, dev->hard_header_len)); 2877 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, 2878 msg->msg_flags & MSG_DONTWAIT, &err); 2879 if (skb == NULL) 2880 goto out_unlock; 2881 2882 skb_reset_network_header(skb); 2883 2884 err = -EINVAL; 2885 if (sock->type == SOCK_DGRAM) { 2886 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 2887 if (unlikely(offset < 0)) 2888 goto out_free; 2889 } else if (reserve) { 2890 skb_reserve(skb, -reserve); 2891 if (len < reserve + sizeof(struct ipv6hdr) && 2892 dev->min_header_len != dev->hard_header_len) 2893 skb_reset_network_header(skb); 2894 } 2895 2896 /* Returns -EFAULT on error */ 2897 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 2898 if (err) 2899 goto out_free; 2900 2901 if (sock->type == SOCK_RAW && 2902 !dev_validate_header(dev, skb->data, len)) { 2903 err = -EINVAL; 2904 goto out_free; 2905 } 2906 2907 skb_setup_tx_timestamp(skb, sockc.tsflags); 2908 2909 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 2910 !packet_extra_vlan_len_allowed(dev, skb)) { 2911 err = -EMSGSIZE; 2912 goto out_free; 2913 } 2914 2915 skb->protocol = proto; 2916 skb->dev = dev; 2917 skb->priority = sk->sk_priority; 2918 skb->mark = sockc.mark; 2919 skb->tstamp = sockc.transmit_time; 2920 2921 if (has_vnet_hdr) { 2922 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 2923 if (err) 2924 goto out_free; 2925 len += sizeof(vnet_hdr); 2926 virtio_net_hdr_set_proto(skb, &vnet_hdr); 2927 } 2928 2929 packet_parse_headers(skb, sock); 2930 2931 if (unlikely(extra_len == 4)) 2932 skb->no_fcs = 1; 2933 2934 err = po->xmit(skb); 2935 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2936 goto out_unlock; 2937 2938 dev_put(dev); 2939 2940 return len; 2941 2942 out_free: 2943 kfree_skb(skb); 2944 out_unlock: 2945 if (dev) 2946 dev_put(dev); 2947 out: 2948 return err; 2949 } 2950 2951 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2952 { 2953 struct sock *sk = sock->sk; 2954 struct packet_sock *po = pkt_sk(sk); 2955 2956 if (po->tx_ring.pg_vec) 2957 return tpacket_snd(po, msg); 2958 else 2959 return packet_snd(sock, msg, len); 2960 } 2961 2962 /* 2963 * Close a PACKET socket. This is fairly simple. We immediately go 2964 * to 'closed' state and remove our protocol entry in the device list. 2965 */ 2966 2967 static int packet_release(struct socket *sock) 2968 { 2969 struct sock *sk = sock->sk; 2970 struct packet_sock *po; 2971 struct packet_fanout *f; 2972 struct net *net; 2973 union tpacket_req_u req_u; 2974 2975 if (!sk) 2976 return 0; 2977 2978 net = sock_net(sk); 2979 po = pkt_sk(sk); 2980 2981 mutex_lock(&net->packet.sklist_lock); 2982 sk_del_node_init_rcu(sk); 2983 mutex_unlock(&net->packet.sklist_lock); 2984 2985 preempt_disable(); 2986 sock_prot_inuse_add(net, sk->sk_prot, -1); 2987 preempt_enable(); 2988 2989 spin_lock(&po->bind_lock); 2990 unregister_prot_hook(sk, false); 2991 packet_cached_dev_reset(po); 2992 2993 if (po->prot_hook.dev) { 2994 dev_put(po->prot_hook.dev); 2995 po->prot_hook.dev = NULL; 2996 } 2997 spin_unlock(&po->bind_lock); 2998 2999 packet_flush_mclist(sk); 3000 3001 lock_sock(sk); 3002 if (po->rx_ring.pg_vec) { 3003 memset(&req_u, 0, sizeof(req_u)); 3004 packet_set_ring(sk, &req_u, 1, 0); 3005 } 3006 3007 if (po->tx_ring.pg_vec) { 3008 memset(&req_u, 0, sizeof(req_u)); 3009 packet_set_ring(sk, &req_u, 1, 1); 3010 } 3011 release_sock(sk); 3012 3013 f = fanout_release(sk); 3014 3015 synchronize_net(); 3016 3017 if (f) { 3018 kfree(po->rollover); 3019 fanout_release_data(f); 3020 kfree(f); 3021 } 3022 /* 3023 * Now the socket is dead. No more input will appear. 3024 */ 3025 sock_orphan(sk); 3026 sock->sk = NULL; 3027 3028 /* Purge queues */ 3029 3030 skb_queue_purge(&sk->sk_receive_queue); 3031 packet_free_pending(po); 3032 sk_refcnt_debug_release(sk); 3033 3034 sock_put(sk); 3035 return 0; 3036 } 3037 3038 /* 3039 * Attach a packet hook. 3040 */ 3041 3042 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3043 __be16 proto) 3044 { 3045 struct packet_sock *po = pkt_sk(sk); 3046 struct net_device *dev_curr; 3047 __be16 proto_curr; 3048 bool need_rehook; 3049 struct net_device *dev = NULL; 3050 int ret = 0; 3051 bool unlisted = false; 3052 3053 lock_sock(sk); 3054 spin_lock(&po->bind_lock); 3055 rcu_read_lock(); 3056 3057 if (po->fanout) { 3058 ret = -EINVAL; 3059 goto out_unlock; 3060 } 3061 3062 if (name) { 3063 dev = dev_get_by_name_rcu(sock_net(sk), name); 3064 if (!dev) { 3065 ret = -ENODEV; 3066 goto out_unlock; 3067 } 3068 } else if (ifindex) { 3069 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3070 if (!dev) { 3071 ret = -ENODEV; 3072 goto out_unlock; 3073 } 3074 } 3075 3076 if (dev) 3077 dev_hold(dev); 3078 3079 proto_curr = po->prot_hook.type; 3080 dev_curr = po->prot_hook.dev; 3081 3082 need_rehook = proto_curr != proto || dev_curr != dev; 3083 3084 if (need_rehook) { 3085 if (po->running) { 3086 rcu_read_unlock(); 3087 /* prevents packet_notifier() from calling 3088 * register_prot_hook() 3089 */ 3090 po->num = 0; 3091 __unregister_prot_hook(sk, true); 3092 rcu_read_lock(); 3093 dev_curr = po->prot_hook.dev; 3094 if (dev) 3095 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3096 dev->ifindex); 3097 } 3098 3099 BUG_ON(po->running); 3100 po->num = proto; 3101 po->prot_hook.type = proto; 3102 3103 if (unlikely(unlisted)) { 3104 dev_put(dev); 3105 po->prot_hook.dev = NULL; 3106 po->ifindex = -1; 3107 packet_cached_dev_reset(po); 3108 } else { 3109 po->prot_hook.dev = dev; 3110 po->ifindex = dev ? dev->ifindex : 0; 3111 packet_cached_dev_assign(po, dev); 3112 } 3113 } 3114 if (dev_curr) 3115 dev_put(dev_curr); 3116 3117 if (proto == 0 || !need_rehook) 3118 goto out_unlock; 3119 3120 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3121 register_prot_hook(sk); 3122 } else { 3123 sk->sk_err = ENETDOWN; 3124 if (!sock_flag(sk, SOCK_DEAD)) 3125 sk->sk_error_report(sk); 3126 } 3127 3128 out_unlock: 3129 rcu_read_unlock(); 3130 spin_unlock(&po->bind_lock); 3131 release_sock(sk); 3132 return ret; 3133 } 3134 3135 /* 3136 * Bind a packet socket to a device 3137 */ 3138 3139 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3140 int addr_len) 3141 { 3142 struct sock *sk = sock->sk; 3143 char name[sizeof(uaddr->sa_data) + 1]; 3144 3145 /* 3146 * Check legality 3147 */ 3148 3149 if (addr_len != sizeof(struct sockaddr)) 3150 return -EINVAL; 3151 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be 3152 * zero-terminated. 3153 */ 3154 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); 3155 name[sizeof(uaddr->sa_data)] = 0; 3156 3157 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3158 } 3159 3160 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3161 { 3162 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3163 struct sock *sk = sock->sk; 3164 3165 /* 3166 * Check legality 3167 */ 3168 3169 if (addr_len < sizeof(struct sockaddr_ll)) 3170 return -EINVAL; 3171 if (sll->sll_family != AF_PACKET) 3172 return -EINVAL; 3173 3174 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3175 sll->sll_protocol ? : pkt_sk(sk)->num); 3176 } 3177 3178 static struct proto packet_proto = { 3179 .name = "PACKET", 3180 .owner = THIS_MODULE, 3181 .obj_size = sizeof(struct packet_sock), 3182 }; 3183 3184 /* 3185 * Create a packet of type SOCK_PACKET. 3186 */ 3187 3188 static int packet_create(struct net *net, struct socket *sock, int protocol, 3189 int kern) 3190 { 3191 struct sock *sk; 3192 struct packet_sock *po; 3193 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3194 int err; 3195 3196 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3197 return -EPERM; 3198 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3199 sock->type != SOCK_PACKET) 3200 return -ESOCKTNOSUPPORT; 3201 3202 sock->state = SS_UNCONNECTED; 3203 3204 err = -ENOBUFS; 3205 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3206 if (sk == NULL) 3207 goto out; 3208 3209 sock->ops = &packet_ops; 3210 if (sock->type == SOCK_PACKET) 3211 sock->ops = &packet_ops_spkt; 3212 3213 sock_init_data(sock, sk); 3214 3215 po = pkt_sk(sk); 3216 sk->sk_family = PF_PACKET; 3217 po->num = proto; 3218 po->xmit = dev_queue_xmit; 3219 3220 err = packet_alloc_pending(po); 3221 if (err) 3222 goto out2; 3223 3224 packet_cached_dev_reset(po); 3225 3226 sk->sk_destruct = packet_sock_destruct; 3227 sk_refcnt_debug_inc(sk); 3228 3229 /* 3230 * Attach a protocol block 3231 */ 3232 3233 spin_lock_init(&po->bind_lock); 3234 mutex_init(&po->pg_vec_lock); 3235 po->rollover = NULL; 3236 po->prot_hook.func = packet_rcv; 3237 3238 if (sock->type == SOCK_PACKET) 3239 po->prot_hook.func = packet_rcv_spkt; 3240 3241 po->prot_hook.af_packet_priv = sk; 3242 3243 if (proto) { 3244 po->prot_hook.type = proto; 3245 __register_prot_hook(sk); 3246 } 3247 3248 mutex_lock(&net->packet.sklist_lock); 3249 sk_add_node_tail_rcu(sk, &net->packet.sklist); 3250 mutex_unlock(&net->packet.sklist_lock); 3251 3252 preempt_disable(); 3253 sock_prot_inuse_add(net, &packet_proto, 1); 3254 preempt_enable(); 3255 3256 return 0; 3257 out2: 3258 sk_free(sk); 3259 out: 3260 return err; 3261 } 3262 3263 /* 3264 * Pull a packet from our receive queue and hand it to the user. 3265 * If necessary we block. 3266 */ 3267 3268 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3269 int flags) 3270 { 3271 struct sock *sk = sock->sk; 3272 struct sk_buff *skb; 3273 int copied, err; 3274 int vnet_hdr_len = 0; 3275 unsigned int origlen = 0; 3276 3277 err = -EINVAL; 3278 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3279 goto out; 3280 3281 #if 0 3282 /* What error should we return now? EUNATTACH? */ 3283 if (pkt_sk(sk)->ifindex < 0) 3284 return -ENODEV; 3285 #endif 3286 3287 if (flags & MSG_ERRQUEUE) { 3288 err = sock_recv_errqueue(sk, msg, len, 3289 SOL_PACKET, PACKET_TX_TIMESTAMP); 3290 goto out; 3291 } 3292 3293 /* 3294 * Call the generic datagram receiver. This handles all sorts 3295 * of horrible races and re-entrancy so we can forget about it 3296 * in the protocol layers. 3297 * 3298 * Now it will return ENETDOWN, if device have just gone down, 3299 * but then it will block. 3300 */ 3301 3302 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 3303 3304 /* 3305 * An error occurred so return it. Because skb_recv_datagram() 3306 * handles the blocking we don't see and worry about blocking 3307 * retries. 3308 */ 3309 3310 if (skb == NULL) 3311 goto out; 3312 3313 if (pkt_sk(sk)->pressure) 3314 packet_rcv_has_room(pkt_sk(sk), NULL); 3315 3316 if (pkt_sk(sk)->has_vnet_hdr) { 3317 err = packet_rcv_vnet(msg, skb, &len); 3318 if (err) 3319 goto out_free; 3320 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3321 } 3322 3323 /* You lose any data beyond the buffer you gave. If it worries 3324 * a user program they can ask the device for its MTU 3325 * anyway. 3326 */ 3327 copied = skb->len; 3328 if (copied > len) { 3329 copied = len; 3330 msg->msg_flags |= MSG_TRUNC; 3331 } 3332 3333 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3334 if (err) 3335 goto out_free; 3336 3337 if (sock->type != SOCK_PACKET) { 3338 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3339 3340 /* Original length was stored in sockaddr_ll fields */ 3341 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3342 sll->sll_family = AF_PACKET; 3343 sll->sll_protocol = skb->protocol; 3344 } 3345 3346 sock_recv_ts_and_drops(msg, sk, skb); 3347 3348 if (msg->msg_name) { 3349 int copy_len; 3350 3351 /* If the address length field is there to be filled 3352 * in, we fill it in now. 3353 */ 3354 if (sock->type == SOCK_PACKET) { 3355 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3356 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3357 copy_len = msg->msg_namelen; 3358 } else { 3359 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3360 3361 msg->msg_namelen = sll->sll_halen + 3362 offsetof(struct sockaddr_ll, sll_addr); 3363 copy_len = msg->msg_namelen; 3364 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { 3365 memset(msg->msg_name + 3366 offsetof(struct sockaddr_ll, sll_addr), 3367 0, sizeof(sll->sll_addr)); 3368 msg->msg_namelen = sizeof(struct sockaddr_ll); 3369 } 3370 } 3371 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); 3372 } 3373 3374 if (pkt_sk(sk)->auxdata) { 3375 struct tpacket_auxdata aux; 3376 3377 aux.tp_status = TP_STATUS_USER; 3378 if (skb->ip_summed == CHECKSUM_PARTIAL) 3379 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3380 else if (skb->pkt_type != PACKET_OUTGOING && 3381 (skb->ip_summed == CHECKSUM_COMPLETE || 3382 skb_csum_unnecessary(skb))) 3383 aux.tp_status |= TP_STATUS_CSUM_VALID; 3384 3385 aux.tp_len = origlen; 3386 aux.tp_snaplen = skb->len; 3387 aux.tp_mac = 0; 3388 aux.tp_net = skb_network_offset(skb); 3389 if (skb_vlan_tag_present(skb)) { 3390 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3391 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3392 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3393 } else { 3394 aux.tp_vlan_tci = 0; 3395 aux.tp_vlan_tpid = 0; 3396 } 3397 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3398 } 3399 3400 /* 3401 * Free or return the buffer as appropriate. Again this 3402 * hides all the races and re-entrancy issues from us. 3403 */ 3404 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3405 3406 out_free: 3407 skb_free_datagram(sk, skb); 3408 out: 3409 return err; 3410 } 3411 3412 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3413 int peer) 3414 { 3415 struct net_device *dev; 3416 struct sock *sk = sock->sk; 3417 3418 if (peer) 3419 return -EOPNOTSUPP; 3420 3421 uaddr->sa_family = AF_PACKET; 3422 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3423 rcu_read_lock(); 3424 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3425 if (dev) 3426 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3427 rcu_read_unlock(); 3428 3429 return sizeof(*uaddr); 3430 } 3431 3432 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3433 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 3459 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3460 } 3461 3462 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3463 int what) 3464 { 3465 switch (i->type) { 3466 case PACKET_MR_MULTICAST: 3467 if (i->alen != dev->addr_len) 3468 return -EINVAL; 3469 if (what > 0) 3470 return dev_mc_add(dev, i->addr); 3471 else 3472 return dev_mc_del(dev, i->addr); 3473 break; 3474 case PACKET_MR_PROMISC: 3475 return dev_set_promiscuity(dev, what); 3476 case PACKET_MR_ALLMULTI: 3477 return dev_set_allmulti(dev, what); 3478 case PACKET_MR_UNICAST: 3479 if (i->alen != dev->addr_len) 3480 return -EINVAL; 3481 if (what > 0) 3482 return dev_uc_add(dev, i->addr); 3483 else 3484 return dev_uc_del(dev, i->addr); 3485 break; 3486 default: 3487 break; 3488 } 3489 return 0; 3490 } 3491 3492 static void packet_dev_mclist_delete(struct net_device *dev, 3493 struct packet_mclist **mlp) 3494 { 3495 struct packet_mclist *ml; 3496 3497 while ((ml = *mlp) != NULL) { 3498 if (ml->ifindex == dev->ifindex) { 3499 packet_dev_mc(dev, ml, -1); 3500 *mlp = ml->next; 3501 kfree(ml); 3502 } else 3503 mlp = &ml->next; 3504 } 3505 } 3506 3507 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3508 { 3509 struct packet_sock *po = pkt_sk(sk); 3510 struct packet_mclist *ml, *i; 3511 struct net_device *dev; 3512 int err; 3513 3514 rtnl_lock(); 3515 3516 err = -ENODEV; 3517 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3518 if (!dev) 3519 goto done; 3520 3521 err = -EINVAL; 3522 if (mreq->mr_alen > dev->addr_len) 3523 goto done; 3524 3525 err = -ENOBUFS; 3526 i = kmalloc(sizeof(*i), GFP_KERNEL); 3527 if (i == NULL) 3528 goto done; 3529 3530 err = 0; 3531 for (ml = po->mclist; ml; ml = ml->next) { 3532 if (ml->ifindex == mreq->mr_ifindex && 3533 ml->type == mreq->mr_type && 3534 ml->alen == mreq->mr_alen && 3535 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3536 ml->count++; 3537 /* Free the new element ... */ 3538 kfree(i); 3539 goto done; 3540 } 3541 } 3542 3543 i->type = mreq->mr_type; 3544 i->ifindex = mreq->mr_ifindex; 3545 i->alen = mreq->mr_alen; 3546 memcpy(i->addr, mreq->mr_address, i->alen); 3547 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3548 i->count = 1; 3549 i->next = po->mclist; 3550 po->mclist = i; 3551 err = packet_dev_mc(dev, i, 1); 3552 if (err) { 3553 po->mclist = i->next; 3554 kfree(i); 3555 } 3556 3557 done: 3558 rtnl_unlock(); 3559 return err; 3560 } 3561 3562 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3563 { 3564 struct packet_mclist *ml, **mlp; 3565 3566 rtnl_lock(); 3567 3568 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3569 if (ml->ifindex == mreq->mr_ifindex && 3570 ml->type == mreq->mr_type && 3571 ml->alen == mreq->mr_alen && 3572 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3573 if (--ml->count == 0) { 3574 struct net_device *dev; 3575 *mlp = ml->next; 3576 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3577 if (dev) 3578 packet_dev_mc(dev, ml, -1); 3579 kfree(ml); 3580 } 3581 break; 3582 } 3583 } 3584 rtnl_unlock(); 3585 return 0; 3586 } 3587 3588 static void packet_flush_mclist(struct sock *sk) 3589 { 3590 struct packet_sock *po = pkt_sk(sk); 3591 struct packet_mclist *ml; 3592 3593 if (!po->mclist) 3594 return; 3595 3596 rtnl_lock(); 3597 while ((ml = po->mclist) != NULL) { 3598 struct net_device *dev; 3599 3600 po->mclist = ml->next; 3601 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3602 if (dev != NULL) 3603 packet_dev_mc(dev, ml, -1); 3604 kfree(ml); 3605 } 3606 rtnl_unlock(); 3607 } 3608 3609 static int 3610 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3611 { 3612 struct sock *sk = sock->sk; 3613 struct packet_sock *po = pkt_sk(sk); 3614 int ret; 3615 3616 if (level != SOL_PACKET) 3617 return -ENOPROTOOPT; 3618 3619 switch (optname) { 3620 case PACKET_ADD_MEMBERSHIP: 3621 case PACKET_DROP_MEMBERSHIP: 3622 { 3623 struct packet_mreq_max mreq; 3624 int len = optlen; 3625 memset(&mreq, 0, sizeof(mreq)); 3626 if (len < sizeof(struct packet_mreq)) 3627 return -EINVAL; 3628 if (len > sizeof(mreq)) 3629 len = sizeof(mreq); 3630 if (copy_from_user(&mreq, optval, len)) 3631 return -EFAULT; 3632 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3633 return -EINVAL; 3634 if (optname == PACKET_ADD_MEMBERSHIP) 3635 ret = packet_mc_add(sk, &mreq); 3636 else 3637 ret = packet_mc_drop(sk, &mreq); 3638 return ret; 3639 } 3640 3641 case PACKET_RX_RING: 3642 case PACKET_TX_RING: 3643 { 3644 union tpacket_req_u req_u; 3645 int len; 3646 3647 lock_sock(sk); 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 ret = -EINVAL; 3660 } else { 3661 if (copy_from_user(&req_u.req, optval, len)) 3662 ret = -EFAULT; 3663 else 3664 ret = packet_set_ring(sk, &req_u, 0, 3665 optname == PACKET_TX_RING); 3666 } 3667 release_sock(sk); 3668 return ret; 3669 } 3670 case PACKET_COPY_THRESH: 3671 { 3672 int val; 3673 3674 if (optlen != sizeof(val)) 3675 return -EINVAL; 3676 if (copy_from_user(&val, optval, sizeof(val))) 3677 return -EFAULT; 3678 3679 pkt_sk(sk)->copy_thresh = val; 3680 return 0; 3681 } 3682 case PACKET_VERSION: 3683 { 3684 int val; 3685 3686 if (optlen != sizeof(val)) 3687 return -EINVAL; 3688 if (copy_from_user(&val, optval, sizeof(val))) 3689 return -EFAULT; 3690 switch (val) { 3691 case TPACKET_V1: 3692 case TPACKET_V2: 3693 case TPACKET_V3: 3694 break; 3695 default: 3696 return -EINVAL; 3697 } 3698 lock_sock(sk); 3699 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3700 ret = -EBUSY; 3701 } else { 3702 po->tp_version = val; 3703 ret = 0; 3704 } 3705 release_sock(sk); 3706 return ret; 3707 } 3708 case PACKET_RESERVE: 3709 { 3710 unsigned int val; 3711 3712 if (optlen != sizeof(val)) 3713 return -EINVAL; 3714 if (copy_from_user(&val, optval, sizeof(val))) 3715 return -EFAULT; 3716 if (val > INT_MAX) 3717 return -EINVAL; 3718 lock_sock(sk); 3719 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3720 ret = -EBUSY; 3721 } else { 3722 po->tp_reserve = val; 3723 ret = 0; 3724 } 3725 release_sock(sk); 3726 return ret; 3727 } 3728 case PACKET_LOSS: 3729 { 3730 unsigned int val; 3731 3732 if (optlen != sizeof(val)) 3733 return -EINVAL; 3734 if (copy_from_user(&val, optval, sizeof(val))) 3735 return -EFAULT; 3736 3737 lock_sock(sk); 3738 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3739 ret = -EBUSY; 3740 } else { 3741 po->tp_loss = !!val; 3742 ret = 0; 3743 } 3744 release_sock(sk); 3745 return ret; 3746 } 3747 case PACKET_AUXDATA: 3748 { 3749 int val; 3750 3751 if (optlen < sizeof(val)) 3752 return -EINVAL; 3753 if (copy_from_user(&val, optval, sizeof(val))) 3754 return -EFAULT; 3755 3756 lock_sock(sk); 3757 po->auxdata = !!val; 3758 release_sock(sk); 3759 return 0; 3760 } 3761 case PACKET_ORIGDEV: 3762 { 3763 int val; 3764 3765 if (optlen < sizeof(val)) 3766 return -EINVAL; 3767 if (copy_from_user(&val, optval, sizeof(val))) 3768 return -EFAULT; 3769 3770 lock_sock(sk); 3771 po->origdev = !!val; 3772 release_sock(sk); 3773 return 0; 3774 } 3775 case PACKET_VNET_HDR: 3776 { 3777 int val; 3778 3779 if (sock->type != SOCK_RAW) 3780 return -EINVAL; 3781 if (optlen < sizeof(val)) 3782 return -EINVAL; 3783 if (copy_from_user(&val, optval, sizeof(val))) 3784 return -EFAULT; 3785 3786 lock_sock(sk); 3787 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3788 ret = -EBUSY; 3789 } else { 3790 po->has_vnet_hdr = !!val; 3791 ret = 0; 3792 } 3793 release_sock(sk); 3794 return ret; 3795 } 3796 case PACKET_TIMESTAMP: 3797 { 3798 int val; 3799 3800 if (optlen != sizeof(val)) 3801 return -EINVAL; 3802 if (copy_from_user(&val, optval, sizeof(val))) 3803 return -EFAULT; 3804 3805 po->tp_tstamp = val; 3806 return 0; 3807 } 3808 case PACKET_FANOUT: 3809 { 3810 int val; 3811 3812 if (optlen != sizeof(val)) 3813 return -EINVAL; 3814 if (copy_from_user(&val, optval, sizeof(val))) 3815 return -EFAULT; 3816 3817 return fanout_add(sk, val & 0xffff, val >> 16); 3818 } 3819 case PACKET_FANOUT_DATA: 3820 { 3821 if (!po->fanout) 3822 return -EINVAL; 3823 3824 return fanout_set_data(po, optval, optlen); 3825 } 3826 case PACKET_IGNORE_OUTGOING: 3827 { 3828 int val; 3829 3830 if (optlen != sizeof(val)) 3831 return -EINVAL; 3832 if (copy_from_user(&val, optval, sizeof(val))) 3833 return -EFAULT; 3834 if (val < 0 || val > 1) 3835 return -EINVAL; 3836 3837 po->prot_hook.ignore_outgoing = !!val; 3838 return 0; 3839 } 3840 case PACKET_TX_HAS_OFF: 3841 { 3842 unsigned int val; 3843 3844 if (optlen != sizeof(val)) 3845 return -EINVAL; 3846 if (copy_from_user(&val, optval, sizeof(val))) 3847 return -EFAULT; 3848 3849 lock_sock(sk); 3850 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3851 ret = -EBUSY; 3852 } else { 3853 po->tp_tx_has_off = !!val; 3854 ret = 0; 3855 } 3856 release_sock(sk); 3857 return 0; 3858 } 3859 case PACKET_QDISC_BYPASS: 3860 { 3861 int val; 3862 3863 if (optlen != sizeof(val)) 3864 return -EINVAL; 3865 if (copy_from_user(&val, optval, sizeof(val))) 3866 return -EFAULT; 3867 3868 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; 3869 return 0; 3870 } 3871 default: 3872 return -ENOPROTOOPT; 3873 } 3874 } 3875 3876 static int packet_getsockopt(struct socket *sock, int level, int optname, 3877 char __user *optval, int __user *optlen) 3878 { 3879 int len; 3880 int val, lv = sizeof(val); 3881 struct sock *sk = sock->sk; 3882 struct packet_sock *po = pkt_sk(sk); 3883 void *data = &val; 3884 union tpacket_stats_u st; 3885 struct tpacket_rollover_stats rstats; 3886 3887 if (level != SOL_PACKET) 3888 return -ENOPROTOOPT; 3889 3890 if (get_user(len, optlen)) 3891 return -EFAULT; 3892 3893 if (len < 0) 3894 return -EINVAL; 3895 3896 switch (optname) { 3897 case PACKET_STATISTICS: 3898 spin_lock_bh(&sk->sk_receive_queue.lock); 3899 memcpy(&st, &po->stats, sizeof(st)); 3900 memset(&po->stats, 0, sizeof(po->stats)); 3901 spin_unlock_bh(&sk->sk_receive_queue.lock); 3902 3903 if (po->tp_version == TPACKET_V3) { 3904 lv = sizeof(struct tpacket_stats_v3); 3905 st.stats3.tp_packets += st.stats3.tp_drops; 3906 data = &st.stats3; 3907 } else { 3908 lv = sizeof(struct tpacket_stats); 3909 st.stats1.tp_packets += st.stats1.tp_drops; 3910 data = &st.stats1; 3911 } 3912 3913 break; 3914 case PACKET_AUXDATA: 3915 val = po->auxdata; 3916 break; 3917 case PACKET_ORIGDEV: 3918 val = po->origdev; 3919 break; 3920 case PACKET_VNET_HDR: 3921 val = po->has_vnet_hdr; 3922 break; 3923 case PACKET_VERSION: 3924 val = po->tp_version; 3925 break; 3926 case PACKET_HDRLEN: 3927 if (len > sizeof(int)) 3928 len = sizeof(int); 3929 if (len < sizeof(int)) 3930 return -EINVAL; 3931 if (copy_from_user(&val, optval, len)) 3932 return -EFAULT; 3933 switch (val) { 3934 case TPACKET_V1: 3935 val = sizeof(struct tpacket_hdr); 3936 break; 3937 case TPACKET_V2: 3938 val = sizeof(struct tpacket2_hdr); 3939 break; 3940 case TPACKET_V3: 3941 val = sizeof(struct tpacket3_hdr); 3942 break; 3943 default: 3944 return -EINVAL; 3945 } 3946 break; 3947 case PACKET_RESERVE: 3948 val = po->tp_reserve; 3949 break; 3950 case PACKET_LOSS: 3951 val = po->tp_loss; 3952 break; 3953 case PACKET_TIMESTAMP: 3954 val = po->tp_tstamp; 3955 break; 3956 case PACKET_FANOUT: 3957 val = (po->fanout ? 3958 ((u32)po->fanout->id | 3959 ((u32)po->fanout->type << 16) | 3960 ((u32)po->fanout->flags << 24)) : 3961 0); 3962 break; 3963 case PACKET_IGNORE_OUTGOING: 3964 val = po->prot_hook.ignore_outgoing; 3965 break; 3966 case PACKET_ROLLOVER_STATS: 3967 if (!po->rollover) 3968 return -EINVAL; 3969 rstats.tp_all = atomic_long_read(&po->rollover->num); 3970 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 3971 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 3972 data = &rstats; 3973 lv = sizeof(rstats); 3974 break; 3975 case PACKET_TX_HAS_OFF: 3976 val = po->tp_tx_has_off; 3977 break; 3978 case PACKET_QDISC_BYPASS: 3979 val = packet_use_direct_xmit(po); 3980 break; 3981 default: 3982 return -ENOPROTOOPT; 3983 } 3984 3985 if (len > lv) 3986 len = lv; 3987 if (put_user(len, optlen)) 3988 return -EFAULT; 3989 if (copy_to_user(optval, data, len)) 3990 return -EFAULT; 3991 return 0; 3992 } 3993 3994 3995 #ifdef CONFIG_COMPAT 3996 static int compat_packet_setsockopt(struct socket *sock, int level, int optname, 3997 char __user *optval, unsigned int optlen) 3998 { 3999 struct packet_sock *po = pkt_sk(sock->sk); 4000 4001 if (level != SOL_PACKET) 4002 return -ENOPROTOOPT; 4003 4004 if (optname == PACKET_FANOUT_DATA && 4005 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) { 4006 optval = (char __user *)get_compat_bpf_fprog(optval); 4007 if (!optval) 4008 return -EFAULT; 4009 optlen = sizeof(struct sock_fprog); 4010 } 4011 4012 return packet_setsockopt(sock, level, optname, optval, optlen); 4013 } 4014 #endif 4015 4016 static int packet_notifier(struct notifier_block *this, 4017 unsigned long msg, void *ptr) 4018 { 4019 struct sock *sk; 4020 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4021 struct net *net = dev_net(dev); 4022 4023 rcu_read_lock(); 4024 sk_for_each_rcu(sk, &net->packet.sklist) { 4025 struct packet_sock *po = pkt_sk(sk); 4026 4027 switch (msg) { 4028 case NETDEV_UNREGISTER: 4029 if (po->mclist) 4030 packet_dev_mclist_delete(dev, &po->mclist); 4031 /* fallthrough */ 4032 4033 case NETDEV_DOWN: 4034 if (dev->ifindex == po->ifindex) { 4035 spin_lock(&po->bind_lock); 4036 if (po->running) { 4037 __unregister_prot_hook(sk, false); 4038 sk->sk_err = ENETDOWN; 4039 if (!sock_flag(sk, SOCK_DEAD)) 4040 sk->sk_error_report(sk); 4041 } 4042 if (msg == NETDEV_UNREGISTER) { 4043 packet_cached_dev_reset(po); 4044 po->ifindex = -1; 4045 if (po->prot_hook.dev) 4046 dev_put(po->prot_hook.dev); 4047 po->prot_hook.dev = NULL; 4048 } 4049 spin_unlock(&po->bind_lock); 4050 } 4051 break; 4052 case NETDEV_UP: 4053 if (dev->ifindex == po->ifindex) { 4054 spin_lock(&po->bind_lock); 4055 if (po->num) 4056 register_prot_hook(sk); 4057 spin_unlock(&po->bind_lock); 4058 } 4059 break; 4060 } 4061 } 4062 rcu_read_unlock(); 4063 return NOTIFY_DONE; 4064 } 4065 4066 4067 static int packet_ioctl(struct socket *sock, unsigned int cmd, 4068 unsigned long arg) 4069 { 4070 struct sock *sk = sock->sk; 4071 4072 switch (cmd) { 4073 case SIOCOUTQ: 4074 { 4075 int amount = sk_wmem_alloc_get(sk); 4076 4077 return put_user(amount, (int __user *)arg); 4078 } 4079 case SIOCINQ: 4080 { 4081 struct sk_buff *skb; 4082 int amount = 0; 4083 4084 spin_lock_bh(&sk->sk_receive_queue.lock); 4085 skb = skb_peek(&sk->sk_receive_queue); 4086 if (skb) 4087 amount = skb->len; 4088 spin_unlock_bh(&sk->sk_receive_queue.lock); 4089 return put_user(amount, (int __user *)arg); 4090 } 4091 #ifdef CONFIG_INET 4092 case SIOCADDRT: 4093 case SIOCDELRT: 4094 case SIOCDARP: 4095 case SIOCGARP: 4096 case SIOCSARP: 4097 case SIOCGIFADDR: 4098 case SIOCSIFADDR: 4099 case SIOCGIFBRDADDR: 4100 case SIOCSIFBRDADDR: 4101 case SIOCGIFNETMASK: 4102 case SIOCSIFNETMASK: 4103 case SIOCGIFDSTADDR: 4104 case SIOCSIFDSTADDR: 4105 case SIOCSIFFLAGS: 4106 return inet_dgram_ops.ioctl(sock, cmd, arg); 4107 #endif 4108 4109 default: 4110 return -ENOIOCTLCMD; 4111 } 4112 return 0; 4113 } 4114 4115 static __poll_t packet_poll(struct file *file, struct socket *sock, 4116 poll_table *wait) 4117 { 4118 struct sock *sk = sock->sk; 4119 struct packet_sock *po = pkt_sk(sk); 4120 __poll_t mask = datagram_poll(file, sock, wait); 4121 4122 spin_lock_bh(&sk->sk_receive_queue.lock); 4123 if (po->rx_ring.pg_vec) { 4124 if (!packet_previous_rx_frame(po, &po->rx_ring, 4125 TP_STATUS_KERNEL)) 4126 mask |= EPOLLIN | EPOLLRDNORM; 4127 } 4128 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 4129 po->pressure = 0; 4130 spin_unlock_bh(&sk->sk_receive_queue.lock); 4131 spin_lock_bh(&sk->sk_write_queue.lock); 4132 if (po->tx_ring.pg_vec) { 4133 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4134 mask |= EPOLLOUT | EPOLLWRNORM; 4135 } 4136 spin_unlock_bh(&sk->sk_write_queue.lock); 4137 return mask; 4138 } 4139 4140 4141 /* Dirty? Well, I still did not learn better way to account 4142 * for user mmaps. 4143 */ 4144 4145 static void packet_mm_open(struct vm_area_struct *vma) 4146 { 4147 struct file *file = vma->vm_file; 4148 struct socket *sock = file->private_data; 4149 struct sock *sk = sock->sk; 4150 4151 if (sk) 4152 atomic_inc(&pkt_sk(sk)->mapped); 4153 } 4154 4155 static void packet_mm_close(struct vm_area_struct *vma) 4156 { 4157 struct file *file = vma->vm_file; 4158 struct socket *sock = file->private_data; 4159 struct sock *sk = sock->sk; 4160 4161 if (sk) 4162 atomic_dec(&pkt_sk(sk)->mapped); 4163 } 4164 4165 static const struct vm_operations_struct packet_mmap_ops = { 4166 .open = packet_mm_open, 4167 .close = packet_mm_close, 4168 }; 4169 4170 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4171 unsigned int len) 4172 { 4173 int i; 4174 4175 for (i = 0; i < len; i++) { 4176 if (likely(pg_vec[i].buffer)) { 4177 if (is_vmalloc_addr(pg_vec[i].buffer)) 4178 vfree(pg_vec[i].buffer); 4179 else 4180 free_pages((unsigned long)pg_vec[i].buffer, 4181 order); 4182 pg_vec[i].buffer = NULL; 4183 } 4184 } 4185 kfree(pg_vec); 4186 } 4187 4188 static char *alloc_one_pg_vec_page(unsigned long order) 4189 { 4190 char *buffer; 4191 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4192 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4193 4194 buffer = (char *) __get_free_pages(gfp_flags, order); 4195 if (buffer) 4196 return buffer; 4197 4198 /* __get_free_pages failed, fall back to vmalloc */ 4199 buffer = vzalloc(array_size((1 << order), PAGE_SIZE)); 4200 if (buffer) 4201 return buffer; 4202 4203 /* vmalloc failed, lets dig into swap here */ 4204 gfp_flags &= ~__GFP_NORETRY; 4205 buffer = (char *) __get_free_pages(gfp_flags, order); 4206 if (buffer) 4207 return buffer; 4208 4209 /* complete and utter failure */ 4210 return NULL; 4211 } 4212 4213 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4214 { 4215 unsigned int block_nr = req->tp_block_nr; 4216 struct pgv *pg_vec; 4217 int i; 4218 4219 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN); 4220 if (unlikely(!pg_vec)) 4221 goto out; 4222 4223 for (i = 0; i < block_nr; i++) { 4224 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4225 if (unlikely(!pg_vec[i].buffer)) 4226 goto out_free_pgvec; 4227 } 4228 4229 out: 4230 return pg_vec; 4231 4232 out_free_pgvec: 4233 free_pg_vec(pg_vec, order, block_nr); 4234 pg_vec = NULL; 4235 goto out; 4236 } 4237 4238 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4239 int closing, int tx_ring) 4240 { 4241 struct pgv *pg_vec = NULL; 4242 struct packet_sock *po = pkt_sk(sk); 4243 int was_running, order = 0; 4244 struct packet_ring_buffer *rb; 4245 struct sk_buff_head *rb_queue; 4246 __be16 num; 4247 int err = -EINVAL; 4248 /* Added to avoid minimal code churn */ 4249 struct tpacket_req *req = &req_u->req; 4250 4251 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4252 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4253 4254 err = -EBUSY; 4255 if (!closing) { 4256 if (atomic_read(&po->mapped)) 4257 goto out; 4258 if (packet_read_pending(rb)) 4259 goto out; 4260 } 4261 4262 if (req->tp_block_nr) { 4263 unsigned int min_frame_size; 4264 4265 /* Sanity tests and some calculations */ 4266 err = -EBUSY; 4267 if (unlikely(rb->pg_vec)) 4268 goto out; 4269 4270 switch (po->tp_version) { 4271 case TPACKET_V1: 4272 po->tp_hdrlen = TPACKET_HDRLEN; 4273 break; 4274 case TPACKET_V2: 4275 po->tp_hdrlen = TPACKET2_HDRLEN; 4276 break; 4277 case TPACKET_V3: 4278 po->tp_hdrlen = TPACKET3_HDRLEN; 4279 break; 4280 } 4281 4282 err = -EINVAL; 4283 if (unlikely((int)req->tp_block_size <= 0)) 4284 goto out; 4285 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4286 goto out; 4287 min_frame_size = po->tp_hdrlen + po->tp_reserve; 4288 if (po->tp_version >= TPACKET_V3 && 4289 req->tp_block_size < 4290 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) 4291 goto out; 4292 if (unlikely(req->tp_frame_size < min_frame_size)) 4293 goto out; 4294 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4295 goto out; 4296 4297 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4298 if (unlikely(rb->frames_per_block == 0)) 4299 goto out; 4300 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) 4301 goto out; 4302 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4303 req->tp_frame_nr)) 4304 goto out; 4305 4306 err = -ENOMEM; 4307 order = get_order(req->tp_block_size); 4308 pg_vec = alloc_pg_vec(req, order); 4309 if (unlikely(!pg_vec)) 4310 goto out; 4311 switch (po->tp_version) { 4312 case TPACKET_V3: 4313 /* Block transmit is not supported yet */ 4314 if (!tx_ring) { 4315 init_prb_bdqc(po, rb, pg_vec, req_u); 4316 } else { 4317 struct tpacket_req3 *req3 = &req_u->req3; 4318 4319 if (req3->tp_retire_blk_tov || 4320 req3->tp_sizeof_priv || 4321 req3->tp_feature_req_word) { 4322 err = -EINVAL; 4323 goto out; 4324 } 4325 } 4326 break; 4327 default: 4328 break; 4329 } 4330 } 4331 /* Done */ 4332 else { 4333 err = -EINVAL; 4334 if (unlikely(req->tp_frame_nr)) 4335 goto out; 4336 } 4337 4338 4339 /* Detach socket from network */ 4340 spin_lock(&po->bind_lock); 4341 was_running = po->running; 4342 num = po->num; 4343 if (was_running) { 4344 po->num = 0; 4345 __unregister_prot_hook(sk, false); 4346 } 4347 spin_unlock(&po->bind_lock); 4348 4349 synchronize_net(); 4350 4351 err = -EBUSY; 4352 mutex_lock(&po->pg_vec_lock); 4353 if (closing || atomic_read(&po->mapped) == 0) { 4354 err = 0; 4355 spin_lock_bh(&rb_queue->lock); 4356 swap(rb->pg_vec, pg_vec); 4357 rb->frame_max = (req->tp_frame_nr - 1); 4358 rb->head = 0; 4359 rb->frame_size = req->tp_frame_size; 4360 spin_unlock_bh(&rb_queue->lock); 4361 4362 swap(rb->pg_vec_order, order); 4363 swap(rb->pg_vec_len, req->tp_block_nr); 4364 4365 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4366 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4367 tpacket_rcv : packet_rcv; 4368 skb_queue_purge(rb_queue); 4369 if (atomic_read(&po->mapped)) 4370 pr_err("packet_mmap: vma is busy: %d\n", 4371 atomic_read(&po->mapped)); 4372 } 4373 mutex_unlock(&po->pg_vec_lock); 4374 4375 spin_lock(&po->bind_lock); 4376 if (was_running) { 4377 po->num = num; 4378 register_prot_hook(sk); 4379 } 4380 spin_unlock(&po->bind_lock); 4381 if (pg_vec && (po->tp_version > TPACKET_V2)) { 4382 /* Because we don't support block-based V3 on tx-ring */ 4383 if (!tx_ring) 4384 prb_shutdown_retire_blk_timer(po, rb_queue); 4385 } 4386 4387 if (pg_vec) 4388 free_pg_vec(pg_vec, order, req->tp_block_nr); 4389 out: 4390 return err; 4391 } 4392 4393 static int packet_mmap(struct file *file, struct socket *sock, 4394 struct vm_area_struct *vma) 4395 { 4396 struct sock *sk = sock->sk; 4397 struct packet_sock *po = pkt_sk(sk); 4398 unsigned long size, expected_size; 4399 struct packet_ring_buffer *rb; 4400 unsigned long start; 4401 int err = -EINVAL; 4402 int i; 4403 4404 if (vma->vm_pgoff) 4405 return -EINVAL; 4406 4407 mutex_lock(&po->pg_vec_lock); 4408 4409 expected_size = 0; 4410 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4411 if (rb->pg_vec) { 4412 expected_size += rb->pg_vec_len 4413 * rb->pg_vec_pages 4414 * PAGE_SIZE; 4415 } 4416 } 4417 4418 if (expected_size == 0) 4419 goto out; 4420 4421 size = vma->vm_end - vma->vm_start; 4422 if (size != expected_size) 4423 goto out; 4424 4425 start = vma->vm_start; 4426 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4427 if (rb->pg_vec == NULL) 4428 continue; 4429 4430 for (i = 0; i < rb->pg_vec_len; i++) { 4431 struct page *page; 4432 void *kaddr = rb->pg_vec[i].buffer; 4433 int pg_num; 4434 4435 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4436 page = pgv_to_page(kaddr); 4437 err = vm_insert_page(vma, start, page); 4438 if (unlikely(err)) 4439 goto out; 4440 start += PAGE_SIZE; 4441 kaddr += PAGE_SIZE; 4442 } 4443 } 4444 } 4445 4446 atomic_inc(&po->mapped); 4447 vma->vm_ops = &packet_mmap_ops; 4448 err = 0; 4449 4450 out: 4451 mutex_unlock(&po->pg_vec_lock); 4452 return err; 4453 } 4454 4455 static const struct proto_ops packet_ops_spkt = { 4456 .family = PF_PACKET, 4457 .owner = THIS_MODULE, 4458 .release = packet_release, 4459 .bind = packet_bind_spkt, 4460 .connect = sock_no_connect, 4461 .socketpair = sock_no_socketpair, 4462 .accept = sock_no_accept, 4463 .getname = packet_getname_spkt, 4464 .poll = datagram_poll, 4465 .ioctl = packet_ioctl, 4466 .gettstamp = sock_gettstamp, 4467 .listen = sock_no_listen, 4468 .shutdown = sock_no_shutdown, 4469 .setsockopt = sock_no_setsockopt, 4470 .getsockopt = sock_no_getsockopt, 4471 .sendmsg = packet_sendmsg_spkt, 4472 .recvmsg = packet_recvmsg, 4473 .mmap = sock_no_mmap, 4474 .sendpage = sock_no_sendpage, 4475 }; 4476 4477 static const struct proto_ops packet_ops = { 4478 .family = PF_PACKET, 4479 .owner = THIS_MODULE, 4480 .release = packet_release, 4481 .bind = packet_bind, 4482 .connect = sock_no_connect, 4483 .socketpair = sock_no_socketpair, 4484 .accept = sock_no_accept, 4485 .getname = packet_getname, 4486 .poll = packet_poll, 4487 .ioctl = packet_ioctl, 4488 .gettstamp = sock_gettstamp, 4489 .listen = sock_no_listen, 4490 .shutdown = sock_no_shutdown, 4491 .setsockopt = packet_setsockopt, 4492 .getsockopt = packet_getsockopt, 4493 #ifdef CONFIG_COMPAT 4494 .compat_setsockopt = compat_packet_setsockopt, 4495 #endif 4496 .sendmsg = packet_sendmsg, 4497 .recvmsg = packet_recvmsg, 4498 .mmap = packet_mmap, 4499 .sendpage = sock_no_sendpage, 4500 }; 4501 4502 static const struct net_proto_family packet_family_ops = { 4503 .family = PF_PACKET, 4504 .create = packet_create, 4505 .owner = THIS_MODULE, 4506 }; 4507 4508 static struct notifier_block packet_netdev_notifier = { 4509 .notifier_call = packet_notifier, 4510 }; 4511 4512 #ifdef CONFIG_PROC_FS 4513 4514 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4515 __acquires(RCU) 4516 { 4517 struct net *net = seq_file_net(seq); 4518 4519 rcu_read_lock(); 4520 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4521 } 4522 4523 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4524 { 4525 struct net *net = seq_file_net(seq); 4526 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4527 } 4528 4529 static void packet_seq_stop(struct seq_file *seq, void *v) 4530 __releases(RCU) 4531 { 4532 rcu_read_unlock(); 4533 } 4534 4535 static int packet_seq_show(struct seq_file *seq, void *v) 4536 { 4537 if (v == SEQ_START_TOKEN) 4538 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4539 else { 4540 struct sock *s = sk_entry(v); 4541 const struct packet_sock *po = pkt_sk(s); 4542 4543 seq_printf(seq, 4544 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4545 s, 4546 refcount_read(&s->sk_refcnt), 4547 s->sk_type, 4548 ntohs(po->num), 4549 po->ifindex, 4550 po->running, 4551 atomic_read(&s->sk_rmem_alloc), 4552 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4553 sock_i_ino(s)); 4554 } 4555 4556 return 0; 4557 } 4558 4559 static const struct seq_operations packet_seq_ops = { 4560 .start = packet_seq_start, 4561 .next = packet_seq_next, 4562 .stop = packet_seq_stop, 4563 .show = packet_seq_show, 4564 }; 4565 #endif 4566 4567 static int __net_init packet_net_init(struct net *net) 4568 { 4569 mutex_init(&net->packet.sklist_lock); 4570 INIT_HLIST_HEAD(&net->packet.sklist); 4571 4572 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, 4573 sizeof(struct seq_net_private))) 4574 return -ENOMEM; 4575 4576 return 0; 4577 } 4578 4579 static void __net_exit packet_net_exit(struct net *net) 4580 { 4581 remove_proc_entry("packet", net->proc_net); 4582 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); 4583 } 4584 4585 static struct pernet_operations packet_net_ops = { 4586 .init = packet_net_init, 4587 .exit = packet_net_exit, 4588 }; 4589 4590 4591 static void __exit packet_exit(void) 4592 { 4593 unregister_netdevice_notifier(&packet_netdev_notifier); 4594 unregister_pernet_subsys(&packet_net_ops); 4595 sock_unregister(PF_PACKET); 4596 proto_unregister(&packet_proto); 4597 } 4598 4599 static int __init packet_init(void) 4600 { 4601 int rc; 4602 4603 rc = proto_register(&packet_proto, 0); 4604 if (rc) 4605 goto out; 4606 rc = sock_register(&packet_family_ops); 4607 if (rc) 4608 goto out_proto; 4609 rc = register_pernet_subsys(&packet_net_ops); 4610 if (rc) 4611 goto out_sock; 4612 rc = register_netdevice_notifier(&packet_netdev_notifier); 4613 if (rc) 4614 goto out_pernet; 4615 4616 return 0; 4617 4618 out_pernet: 4619 unregister_pernet_subsys(&packet_net_ops); 4620 out_sock: 4621 sock_unregister(PF_PACKET); 4622 out_proto: 4623 proto_unregister(&packet_proto); 4624 out: 4625 return rc; 4626 } 4627 4628 module_init(packet_init); 4629 module_exit(packet_exit); 4630 MODULE_LICENSE("GPL"); 4631 MODULE_ALIAS_NETPROTO(PF_PACKET); 4632