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 net_device *dev, struct sk_buff *skb, 279 struct net_device *sb_dev) 280 { 281 return dev_pick_tx_cpu_id(dev, skb, sb_dev, NULL); 282 } 283 284 static u16 packet_pick_tx_queue(struct sk_buff *skb) 285 { 286 struct net_device *dev = skb->dev; 287 const struct net_device_ops *ops = dev->netdev_ops; 288 u16 queue_index; 289 290 if (ops->ndo_select_queue) { 291 queue_index = ops->ndo_select_queue(dev, skb, NULL, 292 __packet_pick_tx_queue); 293 queue_index = netdev_cap_txqueue(dev, queue_index); 294 } else { 295 queue_index = __packet_pick_tx_queue(dev, skb, NULL); 296 } 297 298 return queue_index; 299 } 300 301 /* __register_prot_hook must be invoked through register_prot_hook 302 * or from a context in which asynchronous accesses to the packet 303 * socket is not possible (packet_create()). 304 */ 305 static void __register_prot_hook(struct sock *sk) 306 { 307 struct packet_sock *po = pkt_sk(sk); 308 309 if (!po->running) { 310 if (po->fanout) 311 __fanout_link(sk, po); 312 else 313 dev_add_pack(&po->prot_hook); 314 315 sock_hold(sk); 316 po->running = 1; 317 } 318 } 319 320 static void register_prot_hook(struct sock *sk) 321 { 322 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); 323 __register_prot_hook(sk); 324 } 325 326 /* If the sync parameter is true, we will temporarily drop 327 * the po->bind_lock and do a synchronize_net to make sure no 328 * asynchronous packet processing paths still refer to the elements 329 * of po->prot_hook. If the sync parameter is false, it is the 330 * callers responsibility to take care of this. 331 */ 332 static void __unregister_prot_hook(struct sock *sk, bool sync) 333 { 334 struct packet_sock *po = pkt_sk(sk); 335 336 lockdep_assert_held_once(&po->bind_lock); 337 338 po->running = 0; 339 340 if (po->fanout) 341 __fanout_unlink(sk, po); 342 else 343 __dev_remove_pack(&po->prot_hook); 344 345 __sock_put(sk); 346 347 if (sync) { 348 spin_unlock(&po->bind_lock); 349 synchronize_net(); 350 spin_lock(&po->bind_lock); 351 } 352 } 353 354 static void unregister_prot_hook(struct sock *sk, bool sync) 355 { 356 struct packet_sock *po = pkt_sk(sk); 357 358 if (po->running) 359 __unregister_prot_hook(sk, sync); 360 } 361 362 static inline struct page * __pure pgv_to_page(void *addr) 363 { 364 if (is_vmalloc_addr(addr)) 365 return vmalloc_to_page(addr); 366 return virt_to_page(addr); 367 } 368 369 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 370 { 371 union tpacket_uhdr h; 372 373 h.raw = frame; 374 switch (po->tp_version) { 375 case TPACKET_V1: 376 h.h1->tp_status = status; 377 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 378 break; 379 case TPACKET_V2: 380 h.h2->tp_status = status; 381 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 382 break; 383 case TPACKET_V3: 384 h.h3->tp_status = status; 385 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 386 break; 387 default: 388 WARN(1, "TPACKET version not supported.\n"); 389 BUG(); 390 } 391 392 smp_wmb(); 393 } 394 395 static int __packet_get_status(struct packet_sock *po, void *frame) 396 { 397 union tpacket_uhdr h; 398 399 smp_rmb(); 400 401 h.raw = frame; 402 switch (po->tp_version) { 403 case TPACKET_V1: 404 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 405 return h.h1->tp_status; 406 case TPACKET_V2: 407 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 408 return h.h2->tp_status; 409 case TPACKET_V3: 410 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 411 return h.h3->tp_status; 412 default: 413 WARN(1, "TPACKET version not supported.\n"); 414 BUG(); 415 return 0; 416 } 417 } 418 419 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, 420 unsigned int flags) 421 { 422 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 423 424 if (shhwtstamps && 425 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 426 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) 427 return TP_STATUS_TS_RAW_HARDWARE; 428 429 if (ktime_to_timespec_cond(skb->tstamp, ts)) 430 return TP_STATUS_TS_SOFTWARE; 431 432 return 0; 433 } 434 435 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 436 struct sk_buff *skb) 437 { 438 union tpacket_uhdr h; 439 struct timespec ts; 440 __u32 ts_status; 441 442 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 443 return 0; 444 445 h.raw = frame; 446 switch (po->tp_version) { 447 case TPACKET_V1: 448 h.h1->tp_sec = ts.tv_sec; 449 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 450 break; 451 case TPACKET_V2: 452 h.h2->tp_sec = ts.tv_sec; 453 h.h2->tp_nsec = ts.tv_nsec; 454 break; 455 case TPACKET_V3: 456 h.h3->tp_sec = ts.tv_sec; 457 h.h3->tp_nsec = ts.tv_nsec; 458 break; 459 default: 460 WARN(1, "TPACKET version not supported.\n"); 461 BUG(); 462 } 463 464 /* one flush is safe, as both fields always lie on the same cacheline */ 465 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 466 smp_wmb(); 467 468 return ts_status; 469 } 470 471 static void *packet_lookup_frame(struct packet_sock *po, 472 struct packet_ring_buffer *rb, 473 unsigned int position, 474 int status) 475 { 476 unsigned int pg_vec_pos, frame_offset; 477 union tpacket_uhdr h; 478 479 pg_vec_pos = position / rb->frames_per_block; 480 frame_offset = position % rb->frames_per_block; 481 482 h.raw = rb->pg_vec[pg_vec_pos].buffer + 483 (frame_offset * rb->frame_size); 484 485 if (status != __packet_get_status(po, h.raw)) 486 return NULL; 487 488 return h.raw; 489 } 490 491 static void *packet_current_frame(struct packet_sock *po, 492 struct packet_ring_buffer *rb, 493 int status) 494 { 495 return packet_lookup_frame(po, rb, rb->head, status); 496 } 497 498 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 499 { 500 del_timer_sync(&pkc->retire_blk_timer); 501 } 502 503 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 504 struct sk_buff_head *rb_queue) 505 { 506 struct tpacket_kbdq_core *pkc; 507 508 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 509 510 spin_lock_bh(&rb_queue->lock); 511 pkc->delete_blk_timer = 1; 512 spin_unlock_bh(&rb_queue->lock); 513 514 prb_del_retire_blk_timer(pkc); 515 } 516 517 static void prb_setup_retire_blk_timer(struct packet_sock *po) 518 { 519 struct tpacket_kbdq_core *pkc; 520 521 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 522 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, 523 0); 524 pkc->retire_blk_timer.expires = jiffies; 525 } 526 527 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 528 int blk_size_in_bytes) 529 { 530 struct net_device *dev; 531 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; 532 struct ethtool_link_ksettings ecmd; 533 int err; 534 535 rtnl_lock(); 536 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 537 if (unlikely(!dev)) { 538 rtnl_unlock(); 539 return DEFAULT_PRB_RETIRE_TOV; 540 } 541 err = __ethtool_get_link_ksettings(dev, &ecmd); 542 rtnl_unlock(); 543 if (!err) { 544 /* 545 * If the link speed is so slow you don't really 546 * need to worry about perf anyways 547 */ 548 if (ecmd.base.speed < SPEED_1000 || 549 ecmd.base.speed == SPEED_UNKNOWN) { 550 return DEFAULT_PRB_RETIRE_TOV; 551 } else { 552 msec = 1; 553 div = ecmd.base.speed / 1000; 554 } 555 } 556 557 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 558 559 if (div) 560 mbits /= div; 561 562 tmo = mbits * msec; 563 564 if (div) 565 return tmo+1; 566 return tmo; 567 } 568 569 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 570 union tpacket_req_u *req_u) 571 { 572 p1->feature_req_word = req_u->req3.tp_feature_req_word; 573 } 574 575 static void init_prb_bdqc(struct packet_sock *po, 576 struct packet_ring_buffer *rb, 577 struct pgv *pg_vec, 578 union tpacket_req_u *req_u) 579 { 580 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 581 struct tpacket_block_desc *pbd; 582 583 memset(p1, 0x0, sizeof(*p1)); 584 585 p1->knxt_seq_num = 1; 586 p1->pkbdq = pg_vec; 587 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 588 p1->pkblk_start = pg_vec[0].buffer; 589 p1->kblk_size = req_u->req3.tp_block_size; 590 p1->knum_blocks = req_u->req3.tp_block_nr; 591 p1->hdrlen = po->tp_hdrlen; 592 p1->version = po->tp_version; 593 p1->last_kactive_blk_num = 0; 594 po->stats.stats3.tp_freeze_q_cnt = 0; 595 if (req_u->req3.tp_retire_blk_tov) 596 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 597 else 598 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 599 req_u->req3.tp_block_size); 600 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 601 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 602 603 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 604 prb_init_ft_ops(p1, req_u); 605 prb_setup_retire_blk_timer(po); 606 prb_open_block(p1, pbd); 607 } 608 609 /* Do NOT update the last_blk_num first. 610 * Assumes sk_buff_head lock is held. 611 */ 612 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 613 { 614 mod_timer(&pkc->retire_blk_timer, 615 jiffies + pkc->tov_in_jiffies); 616 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 617 } 618 619 /* 620 * Timer logic: 621 * 1) We refresh the timer only when we open a block. 622 * By doing this we don't waste cycles refreshing the timer 623 * on packet-by-packet basis. 624 * 625 * With a 1MB block-size, on a 1Gbps line, it will take 626 * i) ~8 ms to fill a block + ii) memcpy etc. 627 * In this cut we are not accounting for the memcpy time. 628 * 629 * So, if the user sets the 'tmo' to 10ms then the timer 630 * will never fire while the block is still getting filled 631 * (which is what we want). However, the user could choose 632 * to close a block early and that's fine. 633 * 634 * But when the timer does fire, we check whether or not to refresh it. 635 * Since the tmo granularity is in msecs, it is not too expensive 636 * to refresh the timer, lets say every '8' msecs. 637 * Either the user can set the 'tmo' or we can derive it based on 638 * a) line-speed and b) block-size. 639 * prb_calc_retire_blk_tmo() calculates the tmo. 640 * 641 */ 642 static void prb_retire_rx_blk_timer_expired(struct timer_list *t) 643 { 644 struct packet_sock *po = 645 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer); 646 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 647 unsigned int frozen; 648 struct tpacket_block_desc *pbd; 649 650 spin_lock(&po->sk.sk_receive_queue.lock); 651 652 frozen = prb_queue_frozen(pkc); 653 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 654 655 if (unlikely(pkc->delete_blk_timer)) 656 goto out; 657 658 /* We only need to plug the race when the block is partially filled. 659 * tpacket_rcv: 660 * lock(); increment BLOCK_NUM_PKTS; unlock() 661 * copy_bits() is in progress ... 662 * timer fires on other cpu: 663 * we can't retire the current block because copy_bits 664 * is in progress. 665 * 666 */ 667 if (BLOCK_NUM_PKTS(pbd)) { 668 while (atomic_read(&pkc->blk_fill_in_prog)) { 669 /* Waiting for skb_copy_bits to finish... */ 670 cpu_relax(); 671 } 672 } 673 674 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 675 if (!frozen) { 676 if (!BLOCK_NUM_PKTS(pbd)) { 677 /* An empty block. Just refresh the timer. */ 678 goto refresh_timer; 679 } 680 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 681 if (!prb_dispatch_next_block(pkc, po)) 682 goto refresh_timer; 683 else 684 goto out; 685 } else { 686 /* Case 1. Queue was frozen because user-space was 687 * lagging behind. 688 */ 689 if (prb_curr_blk_in_use(pbd)) { 690 /* 691 * Ok, user-space is still behind. 692 * So just refresh the timer. 693 */ 694 goto refresh_timer; 695 } else { 696 /* Case 2. queue was frozen,user-space caught up, 697 * now the link went idle && the timer fired. 698 * We don't have a block to close.So we open this 699 * block and restart the timer. 700 * opening a block thaws the queue,restarts timer 701 * Thawing/timer-refresh is a side effect. 702 */ 703 prb_open_block(pkc, pbd); 704 goto out; 705 } 706 } 707 } 708 709 refresh_timer: 710 _prb_refresh_rx_retire_blk_timer(pkc); 711 712 out: 713 spin_unlock(&po->sk.sk_receive_queue.lock); 714 } 715 716 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 717 struct tpacket_block_desc *pbd1, __u32 status) 718 { 719 /* Flush everything minus the block header */ 720 721 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 722 u8 *start, *end; 723 724 start = (u8 *)pbd1; 725 726 /* Skip the block header(we know header WILL fit in 4K) */ 727 start += PAGE_SIZE; 728 729 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 730 for (; start < end; start += PAGE_SIZE) 731 flush_dcache_page(pgv_to_page(start)); 732 733 smp_wmb(); 734 #endif 735 736 /* Now update the block status. */ 737 738 BLOCK_STATUS(pbd1) = status; 739 740 /* Flush the block header */ 741 742 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 743 start = (u8 *)pbd1; 744 flush_dcache_page(pgv_to_page(start)); 745 746 smp_wmb(); 747 #endif 748 } 749 750 /* 751 * Side effect: 752 * 753 * 1) flush the block 754 * 2) Increment active_blk_num 755 * 756 * Note:We DONT refresh the timer on purpose. 757 * Because almost always the next block will be opened. 758 */ 759 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 760 struct tpacket_block_desc *pbd1, 761 struct packet_sock *po, unsigned int stat) 762 { 763 __u32 status = TP_STATUS_USER | stat; 764 765 struct tpacket3_hdr *last_pkt; 766 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 767 struct sock *sk = &po->sk; 768 769 if (po->stats.stats3.tp_drops) 770 status |= TP_STATUS_LOSING; 771 772 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 773 last_pkt->tp_next_offset = 0; 774 775 /* Get the ts of the last pkt */ 776 if (BLOCK_NUM_PKTS(pbd1)) { 777 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 778 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 779 } else { 780 /* Ok, we tmo'd - so get the current time. 781 * 782 * It shouldn't really happen as we don't close empty 783 * blocks. See prb_retire_rx_blk_timer_expired(). 784 */ 785 struct timespec ts; 786 getnstimeofday(&ts); 787 h1->ts_last_pkt.ts_sec = ts.tv_sec; 788 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 789 } 790 791 smp_wmb(); 792 793 /* Flush the block */ 794 prb_flush_block(pkc1, pbd1, status); 795 796 sk->sk_data_ready(sk); 797 798 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 799 } 800 801 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 802 { 803 pkc->reset_pending_on_curr_blk = 0; 804 } 805 806 /* 807 * Side effect of opening a block: 808 * 809 * 1) prb_queue is thawed. 810 * 2) retire_blk_timer is refreshed. 811 * 812 */ 813 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 814 struct tpacket_block_desc *pbd1) 815 { 816 struct timespec ts; 817 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 818 819 smp_rmb(); 820 821 /* We could have just memset this but we will lose the 822 * flexibility of making the priv area sticky 823 */ 824 825 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 826 BLOCK_NUM_PKTS(pbd1) = 0; 827 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 828 829 getnstimeofday(&ts); 830 831 h1->ts_first_pkt.ts_sec = ts.tv_sec; 832 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 833 834 pkc1->pkblk_start = (char *)pbd1; 835 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 836 837 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 838 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 839 840 pbd1->version = pkc1->version; 841 pkc1->prev = pkc1->nxt_offset; 842 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 843 844 prb_thaw_queue(pkc1); 845 _prb_refresh_rx_retire_blk_timer(pkc1); 846 847 smp_wmb(); 848 } 849 850 /* 851 * Queue freeze logic: 852 * 1) Assume tp_block_nr = 8 blocks. 853 * 2) At time 't0', user opens Rx ring. 854 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 855 * 4) user-space is either sleeping or processing block '0'. 856 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 857 * it will close block-7,loop around and try to fill block '0'. 858 * call-flow: 859 * __packet_lookup_frame_in_block 860 * prb_retire_current_block() 861 * prb_dispatch_next_block() 862 * |->(BLOCK_STATUS == USER) evaluates to true 863 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 864 * 6) Now there are two cases: 865 * 6.1) Link goes idle right after the queue is frozen. 866 * But remember, the last open_block() refreshed the timer. 867 * When this timer expires,it will refresh itself so that we can 868 * re-open block-0 in near future. 869 * 6.2) Link is busy and keeps on receiving packets. This is a simple 870 * case and __packet_lookup_frame_in_block will check if block-0 871 * is free and can now be re-used. 872 */ 873 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 874 struct packet_sock *po) 875 { 876 pkc->reset_pending_on_curr_blk = 1; 877 po->stats.stats3.tp_freeze_q_cnt++; 878 } 879 880 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 881 882 /* 883 * If the next block is free then we will dispatch it 884 * and return a good offset. 885 * Else, we will freeze the queue. 886 * So, caller must check the return value. 887 */ 888 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 889 struct packet_sock *po) 890 { 891 struct tpacket_block_desc *pbd; 892 893 smp_rmb(); 894 895 /* 1. Get current block num */ 896 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 897 898 /* 2. If this block is currently in_use then freeze the queue */ 899 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 900 prb_freeze_queue(pkc, po); 901 return NULL; 902 } 903 904 /* 905 * 3. 906 * open this block and return the offset where the first packet 907 * needs to get stored. 908 */ 909 prb_open_block(pkc, pbd); 910 return (void *)pkc->nxt_offset; 911 } 912 913 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 914 struct packet_sock *po, unsigned int status) 915 { 916 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 917 918 /* retire/close the current block */ 919 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 920 /* 921 * Plug the case where copy_bits() is in progress on 922 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 923 * have space to copy the pkt in the current block and 924 * called prb_retire_current_block() 925 * 926 * We don't need to worry about the TMO case because 927 * the timer-handler already handled this case. 928 */ 929 if (!(status & TP_STATUS_BLK_TMO)) { 930 while (atomic_read(&pkc->blk_fill_in_prog)) { 931 /* Waiting for skb_copy_bits to finish... */ 932 cpu_relax(); 933 } 934 } 935 prb_close_block(pkc, pbd, po, status); 936 return; 937 } 938 } 939 940 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) 941 { 942 return TP_STATUS_USER & BLOCK_STATUS(pbd); 943 } 944 945 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 946 { 947 return pkc->reset_pending_on_curr_blk; 948 } 949 950 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 951 { 952 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 953 atomic_dec(&pkc->blk_fill_in_prog); 954 } 955 956 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 957 struct tpacket3_hdr *ppd) 958 { 959 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 960 } 961 962 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 963 struct tpacket3_hdr *ppd) 964 { 965 ppd->hv1.tp_rxhash = 0; 966 } 967 968 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 969 struct tpacket3_hdr *ppd) 970 { 971 if (skb_vlan_tag_present(pkc->skb)) { 972 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 973 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 974 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 975 } else { 976 ppd->hv1.tp_vlan_tci = 0; 977 ppd->hv1.tp_vlan_tpid = 0; 978 ppd->tp_status = TP_STATUS_AVAILABLE; 979 } 980 } 981 982 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 983 struct tpacket3_hdr *ppd) 984 { 985 ppd->hv1.tp_padding = 0; 986 prb_fill_vlan_info(pkc, ppd); 987 988 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 989 prb_fill_rxhash(pkc, ppd); 990 else 991 prb_clear_rxhash(pkc, ppd); 992 } 993 994 static void prb_fill_curr_block(char *curr, 995 struct tpacket_kbdq_core *pkc, 996 struct tpacket_block_desc *pbd, 997 unsigned int len) 998 { 999 struct tpacket3_hdr *ppd; 1000 1001 ppd = (struct tpacket3_hdr *)curr; 1002 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 1003 pkc->prev = curr; 1004 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 1005 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1006 BLOCK_NUM_PKTS(pbd) += 1; 1007 atomic_inc(&pkc->blk_fill_in_prog); 1008 prb_run_all_ft_ops(pkc, ppd); 1009 } 1010 1011 /* Assumes caller has the sk->rx_queue.lock */ 1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1013 struct sk_buff *skb, 1014 int status, 1015 unsigned int len 1016 ) 1017 { 1018 struct tpacket_kbdq_core *pkc; 1019 struct tpacket_block_desc *pbd; 1020 char *curr, *end; 1021 1022 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1023 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1024 1025 /* Queue is frozen when user space is lagging behind */ 1026 if (prb_queue_frozen(pkc)) { 1027 /* 1028 * Check if that last block which caused the queue to freeze, 1029 * is still in_use by user-space. 1030 */ 1031 if (prb_curr_blk_in_use(pbd)) { 1032 /* Can't record this packet */ 1033 return NULL; 1034 } else { 1035 /* 1036 * Ok, the block was released by user-space. 1037 * Now let's open that block. 1038 * opening a block also thaws the queue. 1039 * Thawing is a side effect. 1040 */ 1041 prb_open_block(pkc, pbd); 1042 } 1043 } 1044 1045 smp_mb(); 1046 curr = pkc->nxt_offset; 1047 pkc->skb = skb; 1048 end = (char *)pbd + pkc->kblk_size; 1049 1050 /* first try the current block */ 1051 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1052 prb_fill_curr_block(curr, pkc, pbd, len); 1053 return (void *)curr; 1054 } 1055 1056 /* Ok, close the current block */ 1057 prb_retire_current_block(pkc, po, 0); 1058 1059 /* Now, try to dispatch the next block */ 1060 curr = (char *)prb_dispatch_next_block(pkc, po); 1061 if (curr) { 1062 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1063 prb_fill_curr_block(curr, pkc, pbd, len); 1064 return (void *)curr; 1065 } 1066 1067 /* 1068 * No free blocks are available.user_space hasn't caught up yet. 1069 * Queue was just frozen and now this packet will get dropped. 1070 */ 1071 return NULL; 1072 } 1073 1074 static void *packet_current_rx_frame(struct packet_sock *po, 1075 struct sk_buff *skb, 1076 int status, unsigned int len) 1077 { 1078 char *curr = NULL; 1079 switch (po->tp_version) { 1080 case TPACKET_V1: 1081 case TPACKET_V2: 1082 curr = packet_lookup_frame(po, &po->rx_ring, 1083 po->rx_ring.head, status); 1084 return curr; 1085 case TPACKET_V3: 1086 return __packet_lookup_frame_in_block(po, skb, status, len); 1087 default: 1088 WARN(1, "TPACKET version not supported\n"); 1089 BUG(); 1090 return NULL; 1091 } 1092 } 1093 1094 static void *prb_lookup_block(struct packet_sock *po, 1095 struct packet_ring_buffer *rb, 1096 unsigned int idx, 1097 int status) 1098 { 1099 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1100 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1101 1102 if (status != BLOCK_STATUS(pbd)) 1103 return NULL; 1104 return pbd; 1105 } 1106 1107 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1108 { 1109 unsigned int prev; 1110 if (rb->prb_bdqc.kactive_blk_num) 1111 prev = rb->prb_bdqc.kactive_blk_num-1; 1112 else 1113 prev = rb->prb_bdqc.knum_blocks-1; 1114 return prev; 1115 } 1116 1117 /* Assumes caller has held the rx_queue.lock */ 1118 static void *__prb_previous_block(struct packet_sock *po, 1119 struct packet_ring_buffer *rb, 1120 int status) 1121 { 1122 unsigned int previous = prb_previous_blk_num(rb); 1123 return prb_lookup_block(po, rb, previous, status); 1124 } 1125 1126 static void *packet_previous_rx_frame(struct packet_sock *po, 1127 struct packet_ring_buffer *rb, 1128 int status) 1129 { 1130 if (po->tp_version <= TPACKET_V2) 1131 return packet_previous_frame(po, rb, status); 1132 1133 return __prb_previous_block(po, rb, status); 1134 } 1135 1136 static void packet_increment_rx_head(struct packet_sock *po, 1137 struct packet_ring_buffer *rb) 1138 { 1139 switch (po->tp_version) { 1140 case TPACKET_V1: 1141 case TPACKET_V2: 1142 return packet_increment_head(rb); 1143 case TPACKET_V3: 1144 default: 1145 WARN(1, "TPACKET version not supported.\n"); 1146 BUG(); 1147 return; 1148 } 1149 } 1150 1151 static void *packet_previous_frame(struct packet_sock *po, 1152 struct packet_ring_buffer *rb, 1153 int status) 1154 { 1155 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1156 return packet_lookup_frame(po, rb, previous, status); 1157 } 1158 1159 static void packet_increment_head(struct packet_ring_buffer *buff) 1160 { 1161 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1162 } 1163 1164 static void packet_inc_pending(struct packet_ring_buffer *rb) 1165 { 1166 this_cpu_inc(*rb->pending_refcnt); 1167 } 1168 1169 static void packet_dec_pending(struct packet_ring_buffer *rb) 1170 { 1171 this_cpu_dec(*rb->pending_refcnt); 1172 } 1173 1174 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1175 { 1176 unsigned int refcnt = 0; 1177 int cpu; 1178 1179 /* We don't use pending refcount in rx_ring. */ 1180 if (rb->pending_refcnt == NULL) 1181 return 0; 1182 1183 for_each_possible_cpu(cpu) 1184 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1185 1186 return refcnt; 1187 } 1188 1189 static int packet_alloc_pending(struct packet_sock *po) 1190 { 1191 po->rx_ring.pending_refcnt = NULL; 1192 1193 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1194 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1195 return -ENOBUFS; 1196 1197 return 0; 1198 } 1199 1200 static void packet_free_pending(struct packet_sock *po) 1201 { 1202 free_percpu(po->tx_ring.pending_refcnt); 1203 } 1204 1205 #define ROOM_POW_OFF 2 1206 #define ROOM_NONE 0x0 1207 #define ROOM_LOW 0x1 1208 #define ROOM_NORMAL 0x2 1209 1210 static bool __tpacket_has_room(struct packet_sock *po, int pow_off) 1211 { 1212 int idx, len; 1213 1214 len = po->rx_ring.frame_max + 1; 1215 idx = po->rx_ring.head; 1216 if (pow_off) 1217 idx += len >> pow_off; 1218 if (idx >= len) 1219 idx -= len; 1220 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1221 } 1222 1223 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off) 1224 { 1225 int idx, len; 1226 1227 len = po->rx_ring.prb_bdqc.knum_blocks; 1228 idx = po->rx_ring.prb_bdqc.kactive_blk_num; 1229 if (pow_off) 1230 idx += len >> pow_off; 1231 if (idx >= len) 1232 idx -= len; 1233 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1234 } 1235 1236 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1237 { 1238 struct sock *sk = &po->sk; 1239 int ret = ROOM_NONE; 1240 1241 if (po->prot_hook.func != tpacket_rcv) { 1242 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1243 - (skb ? skb->truesize : 0); 1244 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF)) 1245 return ROOM_NORMAL; 1246 else if (avail > 0) 1247 return ROOM_LOW; 1248 else 1249 return ROOM_NONE; 1250 } 1251 1252 if (po->tp_version == TPACKET_V3) { 1253 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1254 ret = ROOM_NORMAL; 1255 else if (__tpacket_v3_has_room(po, 0)) 1256 ret = ROOM_LOW; 1257 } else { 1258 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1259 ret = ROOM_NORMAL; 1260 else if (__tpacket_has_room(po, 0)) 1261 ret = ROOM_LOW; 1262 } 1263 1264 return ret; 1265 } 1266 1267 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1268 { 1269 int ret; 1270 bool has_room; 1271 1272 spin_lock_bh(&po->sk.sk_receive_queue.lock); 1273 ret = __packet_rcv_has_room(po, skb); 1274 has_room = ret == ROOM_NORMAL; 1275 if (po->pressure == has_room) 1276 po->pressure = !has_room; 1277 spin_unlock_bh(&po->sk.sk_receive_queue.lock); 1278 1279 return ret; 1280 } 1281 1282 static void packet_sock_destruct(struct sock *sk) 1283 { 1284 skb_queue_purge(&sk->sk_error_queue); 1285 1286 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1287 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 1288 1289 if (!sock_flag(sk, SOCK_DEAD)) { 1290 pr_err("Attempt to release alive packet socket: %p\n", sk); 1291 return; 1292 } 1293 1294 sk_refcnt_debug_dec(sk); 1295 } 1296 1297 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1298 { 1299 u32 rxhash; 1300 int i, count = 0; 1301 1302 rxhash = skb_get_hash(skb); 1303 for (i = 0; i < ROLLOVER_HLEN; i++) 1304 if (po->rollover->history[i] == rxhash) 1305 count++; 1306 1307 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash; 1308 return count > (ROLLOVER_HLEN >> 1); 1309 } 1310 1311 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1312 struct sk_buff *skb, 1313 unsigned int num) 1314 { 1315 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1316 } 1317 1318 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1319 struct sk_buff *skb, 1320 unsigned int num) 1321 { 1322 unsigned int val = atomic_inc_return(&f->rr_cur); 1323 1324 return val % num; 1325 } 1326 1327 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1328 struct sk_buff *skb, 1329 unsigned int num) 1330 { 1331 return smp_processor_id() % num; 1332 } 1333 1334 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1335 struct sk_buff *skb, 1336 unsigned int num) 1337 { 1338 return prandom_u32_max(num); 1339 } 1340 1341 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1342 struct sk_buff *skb, 1343 unsigned int idx, bool try_self, 1344 unsigned int num) 1345 { 1346 struct packet_sock *po, *po_next, *po_skip = NULL; 1347 unsigned int i, j, room = ROOM_NONE; 1348 1349 po = pkt_sk(f->arr[idx]); 1350 1351 if (try_self) { 1352 room = packet_rcv_has_room(po, skb); 1353 if (room == ROOM_NORMAL || 1354 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1355 return idx; 1356 po_skip = po; 1357 } 1358 1359 i = j = min_t(int, po->rollover->sock, num - 1); 1360 do { 1361 po_next = pkt_sk(f->arr[i]); 1362 if (po_next != po_skip && !po_next->pressure && 1363 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1364 if (i != j) 1365 po->rollover->sock = i; 1366 atomic_long_inc(&po->rollover->num); 1367 if (room == ROOM_LOW) 1368 atomic_long_inc(&po->rollover->num_huge); 1369 return i; 1370 } 1371 1372 if (++i == num) 1373 i = 0; 1374 } while (i != j); 1375 1376 atomic_long_inc(&po->rollover->num_failed); 1377 return idx; 1378 } 1379 1380 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1381 struct sk_buff *skb, 1382 unsigned int num) 1383 { 1384 return skb_get_queue_mapping(skb) % num; 1385 } 1386 1387 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1388 struct sk_buff *skb, 1389 unsigned int num) 1390 { 1391 struct bpf_prog *prog; 1392 unsigned int ret = 0; 1393 1394 rcu_read_lock(); 1395 prog = rcu_dereference(f->bpf_prog); 1396 if (prog) 1397 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1398 rcu_read_unlock(); 1399 1400 return ret; 1401 } 1402 1403 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1404 { 1405 return f->flags & (flag >> 8); 1406 } 1407 1408 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1409 struct packet_type *pt, struct net_device *orig_dev) 1410 { 1411 struct packet_fanout *f = pt->af_packet_priv; 1412 unsigned int num = READ_ONCE(f->num_members); 1413 struct net *net = read_pnet(&f->net); 1414 struct packet_sock *po; 1415 unsigned int idx; 1416 1417 if (!net_eq(dev_net(dev), net) || !num) { 1418 kfree_skb(skb); 1419 return 0; 1420 } 1421 1422 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1423 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1424 if (!skb) 1425 return 0; 1426 } 1427 switch (f->type) { 1428 case PACKET_FANOUT_HASH: 1429 default: 1430 idx = fanout_demux_hash(f, skb, num); 1431 break; 1432 case PACKET_FANOUT_LB: 1433 idx = fanout_demux_lb(f, skb, num); 1434 break; 1435 case PACKET_FANOUT_CPU: 1436 idx = fanout_demux_cpu(f, skb, num); 1437 break; 1438 case PACKET_FANOUT_RND: 1439 idx = fanout_demux_rnd(f, skb, num); 1440 break; 1441 case PACKET_FANOUT_QM: 1442 idx = fanout_demux_qm(f, skb, num); 1443 break; 1444 case PACKET_FANOUT_ROLLOVER: 1445 idx = fanout_demux_rollover(f, skb, 0, false, num); 1446 break; 1447 case PACKET_FANOUT_CBPF: 1448 case PACKET_FANOUT_EBPF: 1449 idx = fanout_demux_bpf(f, skb, num); 1450 break; 1451 } 1452 1453 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1454 idx = fanout_demux_rollover(f, skb, idx, true, num); 1455 1456 po = pkt_sk(f->arr[idx]); 1457 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1458 } 1459 1460 DEFINE_MUTEX(fanout_mutex); 1461 EXPORT_SYMBOL_GPL(fanout_mutex); 1462 static LIST_HEAD(fanout_list); 1463 static u16 fanout_next_id; 1464 1465 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1466 { 1467 struct packet_fanout *f = po->fanout; 1468 1469 spin_lock(&f->lock); 1470 f->arr[f->num_members] = sk; 1471 smp_wmb(); 1472 f->num_members++; 1473 if (f->num_members == 1) 1474 dev_add_pack(&f->prot_hook); 1475 spin_unlock(&f->lock); 1476 } 1477 1478 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1479 { 1480 struct packet_fanout *f = po->fanout; 1481 int i; 1482 1483 spin_lock(&f->lock); 1484 for (i = 0; i < f->num_members; i++) { 1485 if (f->arr[i] == sk) 1486 break; 1487 } 1488 BUG_ON(i >= f->num_members); 1489 f->arr[i] = f->arr[f->num_members - 1]; 1490 f->num_members--; 1491 if (f->num_members == 0) 1492 __dev_remove_pack(&f->prot_hook); 1493 spin_unlock(&f->lock); 1494 } 1495 1496 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1497 { 1498 if (sk->sk_family != PF_PACKET) 1499 return false; 1500 1501 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1502 } 1503 1504 static void fanout_init_data(struct packet_fanout *f) 1505 { 1506 switch (f->type) { 1507 case PACKET_FANOUT_LB: 1508 atomic_set(&f->rr_cur, 0); 1509 break; 1510 case PACKET_FANOUT_CBPF: 1511 case PACKET_FANOUT_EBPF: 1512 RCU_INIT_POINTER(f->bpf_prog, NULL); 1513 break; 1514 } 1515 } 1516 1517 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1518 { 1519 struct bpf_prog *old; 1520 1521 spin_lock(&f->lock); 1522 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1523 rcu_assign_pointer(f->bpf_prog, new); 1524 spin_unlock(&f->lock); 1525 1526 if (old) { 1527 synchronize_net(); 1528 bpf_prog_destroy(old); 1529 } 1530 } 1531 1532 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data, 1533 unsigned int len) 1534 { 1535 struct bpf_prog *new; 1536 struct sock_fprog fprog; 1537 int ret; 1538 1539 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1540 return -EPERM; 1541 if (len != sizeof(fprog)) 1542 return -EINVAL; 1543 if (copy_from_user(&fprog, data, len)) 1544 return -EFAULT; 1545 1546 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1547 if (ret) 1548 return ret; 1549 1550 __fanout_set_data_bpf(po->fanout, new); 1551 return 0; 1552 } 1553 1554 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data, 1555 unsigned int len) 1556 { 1557 struct bpf_prog *new; 1558 u32 fd; 1559 1560 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1561 return -EPERM; 1562 if (len != sizeof(fd)) 1563 return -EINVAL; 1564 if (copy_from_user(&fd, data, len)) 1565 return -EFAULT; 1566 1567 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1568 if (IS_ERR(new)) 1569 return PTR_ERR(new); 1570 1571 __fanout_set_data_bpf(po->fanout, new); 1572 return 0; 1573 } 1574 1575 static int fanout_set_data(struct packet_sock *po, char __user *data, 1576 unsigned int len) 1577 { 1578 switch (po->fanout->type) { 1579 case PACKET_FANOUT_CBPF: 1580 return fanout_set_data_cbpf(po, data, len); 1581 case PACKET_FANOUT_EBPF: 1582 return fanout_set_data_ebpf(po, data, len); 1583 default: 1584 return -EINVAL; 1585 } 1586 } 1587 1588 static void fanout_release_data(struct packet_fanout *f) 1589 { 1590 switch (f->type) { 1591 case PACKET_FANOUT_CBPF: 1592 case PACKET_FANOUT_EBPF: 1593 __fanout_set_data_bpf(f, NULL); 1594 } 1595 } 1596 1597 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) 1598 { 1599 struct packet_fanout *f; 1600 1601 list_for_each_entry(f, &fanout_list, list) { 1602 if (f->id == candidate_id && 1603 read_pnet(&f->net) == sock_net(sk)) { 1604 return false; 1605 } 1606 } 1607 return true; 1608 } 1609 1610 static bool fanout_find_new_id(struct sock *sk, u16 *new_id) 1611 { 1612 u16 id = fanout_next_id; 1613 1614 do { 1615 if (__fanout_id_is_free(sk, id)) { 1616 *new_id = id; 1617 fanout_next_id = id + 1; 1618 return true; 1619 } 1620 1621 id++; 1622 } while (id != fanout_next_id); 1623 1624 return false; 1625 } 1626 1627 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1628 { 1629 struct packet_rollover *rollover = NULL; 1630 struct packet_sock *po = pkt_sk(sk); 1631 struct packet_fanout *f, *match; 1632 u8 type = type_flags & 0xff; 1633 u8 flags = type_flags >> 8; 1634 int err; 1635 1636 switch (type) { 1637 case PACKET_FANOUT_ROLLOVER: 1638 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1639 return -EINVAL; 1640 case PACKET_FANOUT_HASH: 1641 case PACKET_FANOUT_LB: 1642 case PACKET_FANOUT_CPU: 1643 case PACKET_FANOUT_RND: 1644 case PACKET_FANOUT_QM: 1645 case PACKET_FANOUT_CBPF: 1646 case PACKET_FANOUT_EBPF: 1647 break; 1648 default: 1649 return -EINVAL; 1650 } 1651 1652 mutex_lock(&fanout_mutex); 1653 1654 err = -EALREADY; 1655 if (po->fanout) 1656 goto out; 1657 1658 if (type == PACKET_FANOUT_ROLLOVER || 1659 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1660 err = -ENOMEM; 1661 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); 1662 if (!rollover) 1663 goto out; 1664 atomic_long_set(&rollover->num, 0); 1665 atomic_long_set(&rollover->num_huge, 0); 1666 atomic_long_set(&rollover->num_failed, 0); 1667 } 1668 1669 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { 1670 if (id != 0) { 1671 err = -EINVAL; 1672 goto out; 1673 } 1674 if (!fanout_find_new_id(sk, &id)) { 1675 err = -ENOMEM; 1676 goto out; 1677 } 1678 /* ephemeral flag for the first socket in the group: drop it */ 1679 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); 1680 } 1681 1682 match = NULL; 1683 list_for_each_entry(f, &fanout_list, list) { 1684 if (f->id == id && 1685 read_pnet(&f->net) == sock_net(sk)) { 1686 match = f; 1687 break; 1688 } 1689 } 1690 err = -EINVAL; 1691 if (match && match->flags != flags) 1692 goto out; 1693 if (!match) { 1694 err = -ENOMEM; 1695 match = kzalloc(sizeof(*match), GFP_KERNEL); 1696 if (!match) 1697 goto out; 1698 write_pnet(&match->net, sock_net(sk)); 1699 match->id = id; 1700 match->type = type; 1701 match->flags = flags; 1702 INIT_LIST_HEAD(&match->list); 1703 spin_lock_init(&match->lock); 1704 refcount_set(&match->sk_ref, 0); 1705 fanout_init_data(match); 1706 match->prot_hook.type = po->prot_hook.type; 1707 match->prot_hook.dev = po->prot_hook.dev; 1708 match->prot_hook.func = packet_rcv_fanout; 1709 match->prot_hook.af_packet_priv = match; 1710 match->prot_hook.id_match = match_fanout_group; 1711 list_add(&match->list, &fanout_list); 1712 } 1713 err = -EINVAL; 1714 1715 spin_lock(&po->bind_lock); 1716 if (po->running && 1717 match->type == type && 1718 match->prot_hook.type == po->prot_hook.type && 1719 match->prot_hook.dev == po->prot_hook.dev) { 1720 err = -ENOSPC; 1721 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1722 __dev_remove_pack(&po->prot_hook); 1723 po->fanout = match; 1724 po->rollover = rollover; 1725 rollover = NULL; 1726 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); 1727 __fanout_link(sk, po); 1728 err = 0; 1729 } 1730 } 1731 spin_unlock(&po->bind_lock); 1732 1733 if (err && !refcount_read(&match->sk_ref)) { 1734 list_del(&match->list); 1735 kfree(match); 1736 } 1737 1738 out: 1739 kfree(rollover); 1740 mutex_unlock(&fanout_mutex); 1741 return err; 1742 } 1743 1744 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes 1745 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. 1746 * It is the responsibility of the caller to call fanout_release_data() and 1747 * free the returned packet_fanout (after synchronize_net()) 1748 */ 1749 static struct packet_fanout *fanout_release(struct sock *sk) 1750 { 1751 struct packet_sock *po = pkt_sk(sk); 1752 struct packet_fanout *f; 1753 1754 mutex_lock(&fanout_mutex); 1755 f = po->fanout; 1756 if (f) { 1757 po->fanout = NULL; 1758 1759 if (refcount_dec_and_test(&f->sk_ref)) 1760 list_del(&f->list); 1761 else 1762 f = NULL; 1763 } 1764 mutex_unlock(&fanout_mutex); 1765 1766 return f; 1767 } 1768 1769 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1770 struct sk_buff *skb) 1771 { 1772 /* Earlier code assumed this would be a VLAN pkt, double-check 1773 * this now that we have the actual packet in hand. We can only 1774 * do this check on Ethernet devices. 1775 */ 1776 if (unlikely(dev->type != ARPHRD_ETHER)) 1777 return false; 1778 1779 skb_reset_mac_header(skb); 1780 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1781 } 1782 1783 static const struct proto_ops packet_ops; 1784 1785 static const struct proto_ops packet_ops_spkt; 1786 1787 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1788 struct packet_type *pt, struct net_device *orig_dev) 1789 { 1790 struct sock *sk; 1791 struct sockaddr_pkt *spkt; 1792 1793 /* 1794 * When we registered the protocol we saved the socket in the data 1795 * field for just this event. 1796 */ 1797 1798 sk = pt->af_packet_priv; 1799 1800 /* 1801 * Yank back the headers [hope the device set this 1802 * right or kerboom...] 1803 * 1804 * Incoming packets have ll header pulled, 1805 * push it back. 1806 * 1807 * For outgoing ones skb->data == skb_mac_header(skb) 1808 * so that this procedure is noop. 1809 */ 1810 1811 if (skb->pkt_type == PACKET_LOOPBACK) 1812 goto out; 1813 1814 if (!net_eq(dev_net(dev), sock_net(sk))) 1815 goto out; 1816 1817 skb = skb_share_check(skb, GFP_ATOMIC); 1818 if (skb == NULL) 1819 goto oom; 1820 1821 /* drop any routing info */ 1822 skb_dst_drop(skb); 1823 1824 /* drop conntrack reference */ 1825 nf_reset(skb); 1826 1827 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1828 1829 skb_push(skb, skb->data - skb_mac_header(skb)); 1830 1831 /* 1832 * The SOCK_PACKET socket receives _all_ frames. 1833 */ 1834 1835 spkt->spkt_family = dev->type; 1836 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1837 spkt->spkt_protocol = skb->protocol; 1838 1839 /* 1840 * Charge the memory to the socket. This is done specifically 1841 * to prevent sockets using all the memory up. 1842 */ 1843 1844 if (sock_queue_rcv_skb(sk, skb) == 0) 1845 return 0; 1846 1847 out: 1848 kfree_skb(skb); 1849 oom: 1850 return 0; 1851 } 1852 1853 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock) 1854 { 1855 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && 1856 sock->type == SOCK_RAW) { 1857 skb_reset_mac_header(skb); 1858 skb->protocol = dev_parse_header_protocol(skb); 1859 } 1860 1861 skb_probe_transport_header(skb); 1862 } 1863 1864 /* 1865 * Output a raw packet to a device layer. This bypasses all the other 1866 * protocol layers and you must therefore supply it with a complete frame 1867 */ 1868 1869 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1870 size_t len) 1871 { 1872 struct sock *sk = sock->sk; 1873 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1874 struct sk_buff *skb = NULL; 1875 struct net_device *dev; 1876 struct sockcm_cookie sockc; 1877 __be16 proto = 0; 1878 int err; 1879 int extra_len = 0; 1880 1881 /* 1882 * Get and verify the address. 1883 */ 1884 1885 if (saddr) { 1886 if (msg->msg_namelen < sizeof(struct sockaddr)) 1887 return -EINVAL; 1888 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1889 proto = saddr->spkt_protocol; 1890 } else 1891 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1892 1893 /* 1894 * Find the device first to size check it 1895 */ 1896 1897 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1898 retry: 1899 rcu_read_lock(); 1900 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1901 err = -ENODEV; 1902 if (dev == NULL) 1903 goto out_unlock; 1904 1905 err = -ENETDOWN; 1906 if (!(dev->flags & IFF_UP)) 1907 goto out_unlock; 1908 1909 /* 1910 * You may not queue a frame bigger than the mtu. This is the lowest level 1911 * raw protocol and you must do your own fragmentation at this level. 1912 */ 1913 1914 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1915 if (!netif_supports_nofcs(dev)) { 1916 err = -EPROTONOSUPPORT; 1917 goto out_unlock; 1918 } 1919 extra_len = 4; /* We're doing our own CRC */ 1920 } 1921 1922 err = -EMSGSIZE; 1923 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1924 goto out_unlock; 1925 1926 if (!skb) { 1927 size_t reserved = LL_RESERVED_SPACE(dev); 1928 int tlen = dev->needed_tailroom; 1929 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1930 1931 rcu_read_unlock(); 1932 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1933 if (skb == NULL) 1934 return -ENOBUFS; 1935 /* FIXME: Save some space for broken drivers that write a hard 1936 * header at transmission time by themselves. PPP is the notable 1937 * one here. This should really be fixed at the driver level. 1938 */ 1939 skb_reserve(skb, reserved); 1940 skb_reset_network_header(skb); 1941 1942 /* Try to align data part correctly */ 1943 if (hhlen) { 1944 skb->data -= hhlen; 1945 skb->tail -= hhlen; 1946 if (len < hhlen) 1947 skb_reset_network_header(skb); 1948 } 1949 err = memcpy_from_msg(skb_put(skb, len), msg, len); 1950 if (err) 1951 goto out_free; 1952 goto retry; 1953 } 1954 1955 if (!dev_validate_header(dev, skb->data, len)) { 1956 err = -EINVAL; 1957 goto out_unlock; 1958 } 1959 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 1960 !packet_extra_vlan_len_allowed(dev, skb)) { 1961 err = -EMSGSIZE; 1962 goto out_unlock; 1963 } 1964 1965 sockcm_init(&sockc, sk); 1966 if (msg->msg_controllen) { 1967 err = sock_cmsg_send(sk, msg, &sockc); 1968 if (unlikely(err)) 1969 goto out_unlock; 1970 } 1971 1972 skb->protocol = proto; 1973 skb->dev = dev; 1974 skb->priority = sk->sk_priority; 1975 skb->mark = sk->sk_mark; 1976 skb->tstamp = sockc.transmit_time; 1977 1978 skb_setup_tx_timestamp(skb, sockc.tsflags); 1979 1980 if (unlikely(extra_len == 4)) 1981 skb->no_fcs = 1; 1982 1983 packet_parse_headers(skb, sock); 1984 1985 dev_queue_xmit(skb); 1986 rcu_read_unlock(); 1987 return len; 1988 1989 out_unlock: 1990 rcu_read_unlock(); 1991 out_free: 1992 kfree_skb(skb); 1993 return err; 1994 } 1995 1996 static unsigned int run_filter(struct sk_buff *skb, 1997 const struct sock *sk, 1998 unsigned int res) 1999 { 2000 struct sk_filter *filter; 2001 2002 rcu_read_lock(); 2003 filter = rcu_dereference(sk->sk_filter); 2004 if (filter != NULL) 2005 res = bpf_prog_run_clear_cb(filter->prog, skb); 2006 rcu_read_unlock(); 2007 2008 return res; 2009 } 2010 2011 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2012 size_t *len) 2013 { 2014 struct virtio_net_hdr vnet_hdr; 2015 2016 if (*len < sizeof(vnet_hdr)) 2017 return -EINVAL; 2018 *len -= sizeof(vnet_hdr); 2019 2020 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0)) 2021 return -EINVAL; 2022 2023 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 2024 } 2025 2026 /* 2027 * This function makes lazy skb cloning in hope that most of packets 2028 * are discarded by BPF. 2029 * 2030 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2031 * and skb->cb are mangled. It works because (and until) packets 2032 * falling here are owned by current CPU. Output packets are cloned 2033 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2034 * sequencially, so that if we return skb to original state on exit, 2035 * we will not harm anyone. 2036 */ 2037 2038 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2039 struct packet_type *pt, struct net_device *orig_dev) 2040 { 2041 struct sock *sk; 2042 struct sockaddr_ll *sll; 2043 struct packet_sock *po; 2044 u8 *skb_head = skb->data; 2045 int skb_len = skb->len; 2046 unsigned int snaplen, res; 2047 bool is_drop_n_account = false; 2048 2049 if (skb->pkt_type == PACKET_LOOPBACK) 2050 goto drop; 2051 2052 sk = pt->af_packet_priv; 2053 po = pkt_sk(sk); 2054 2055 if (!net_eq(dev_net(dev), sock_net(sk))) 2056 goto drop; 2057 2058 skb->dev = dev; 2059 2060 if (dev->header_ops) { 2061 /* The device has an explicit notion of ll header, 2062 * exported to higher levels. 2063 * 2064 * Otherwise, the device hides details of its frame 2065 * structure, so that corresponding packet head is 2066 * never delivered to user. 2067 */ 2068 if (sk->sk_type != SOCK_DGRAM) 2069 skb_push(skb, skb->data - skb_mac_header(skb)); 2070 else if (skb->pkt_type == PACKET_OUTGOING) { 2071 /* Special case: outgoing packets have ll header at head */ 2072 skb_pull(skb, skb_network_offset(skb)); 2073 } 2074 } 2075 2076 snaplen = skb->len; 2077 2078 res = run_filter(skb, sk, snaplen); 2079 if (!res) 2080 goto drop_n_restore; 2081 if (snaplen > res) 2082 snaplen = res; 2083 2084 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2085 goto drop_n_acct; 2086 2087 if (skb_shared(skb)) { 2088 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2089 if (nskb == NULL) 2090 goto drop_n_acct; 2091 2092 if (skb_head != skb->data) { 2093 skb->data = skb_head; 2094 skb->len = skb_len; 2095 } 2096 consume_skb(skb); 2097 skb = nskb; 2098 } 2099 2100 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2101 2102 sll = &PACKET_SKB_CB(skb)->sa.ll; 2103 sll->sll_hatype = dev->type; 2104 sll->sll_pkttype = skb->pkt_type; 2105 if (unlikely(po->origdev)) 2106 sll->sll_ifindex = orig_dev->ifindex; 2107 else 2108 sll->sll_ifindex = dev->ifindex; 2109 2110 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2111 2112 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2113 * Use their space for storing the original skb length. 2114 */ 2115 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2116 2117 if (pskb_trim(skb, snaplen)) 2118 goto drop_n_acct; 2119 2120 skb_set_owner_r(skb, sk); 2121 skb->dev = NULL; 2122 skb_dst_drop(skb); 2123 2124 /* drop conntrack reference */ 2125 nf_reset(skb); 2126 2127 spin_lock(&sk->sk_receive_queue.lock); 2128 po->stats.stats1.tp_packets++; 2129 sock_skb_set_dropcount(sk, skb); 2130 __skb_queue_tail(&sk->sk_receive_queue, skb); 2131 spin_unlock(&sk->sk_receive_queue.lock); 2132 sk->sk_data_ready(sk); 2133 return 0; 2134 2135 drop_n_acct: 2136 is_drop_n_account = true; 2137 spin_lock(&sk->sk_receive_queue.lock); 2138 po->stats.stats1.tp_drops++; 2139 atomic_inc(&sk->sk_drops); 2140 spin_unlock(&sk->sk_receive_queue.lock); 2141 2142 drop_n_restore: 2143 if (skb_head != skb->data && skb_shared(skb)) { 2144 skb->data = skb_head; 2145 skb->len = skb_len; 2146 } 2147 drop: 2148 if (!is_drop_n_account) 2149 consume_skb(skb); 2150 else 2151 kfree_skb(skb); 2152 return 0; 2153 } 2154 2155 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2156 struct packet_type *pt, struct net_device *orig_dev) 2157 { 2158 struct sock *sk; 2159 struct packet_sock *po; 2160 struct sockaddr_ll *sll; 2161 union tpacket_uhdr h; 2162 u8 *skb_head = skb->data; 2163 int skb_len = skb->len; 2164 unsigned int snaplen, res; 2165 unsigned long status = TP_STATUS_USER; 2166 unsigned short macoff, netoff, hdrlen; 2167 struct sk_buff *copy_skb = NULL; 2168 struct timespec ts; 2169 __u32 ts_status; 2170 bool is_drop_n_account = false; 2171 bool do_vnet = false; 2172 2173 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2174 * We may add members to them until current aligned size without forcing 2175 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2176 */ 2177 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2178 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2179 2180 if (skb->pkt_type == PACKET_LOOPBACK) 2181 goto drop; 2182 2183 sk = pt->af_packet_priv; 2184 po = pkt_sk(sk); 2185 2186 if (!net_eq(dev_net(dev), sock_net(sk))) 2187 goto drop; 2188 2189 if (dev->header_ops) { 2190 if (sk->sk_type != SOCK_DGRAM) 2191 skb_push(skb, skb->data - skb_mac_header(skb)); 2192 else if (skb->pkt_type == PACKET_OUTGOING) { 2193 /* Special case: outgoing packets have ll header at head */ 2194 skb_pull(skb, skb_network_offset(skb)); 2195 } 2196 } 2197 2198 snaplen = skb->len; 2199 2200 res = run_filter(skb, sk, snaplen); 2201 if (!res) 2202 goto drop_n_restore; 2203 2204 if (skb->ip_summed == CHECKSUM_PARTIAL) 2205 status |= TP_STATUS_CSUMNOTREADY; 2206 else if (skb->pkt_type != PACKET_OUTGOING && 2207 (skb->ip_summed == CHECKSUM_COMPLETE || 2208 skb_csum_unnecessary(skb))) 2209 status |= TP_STATUS_CSUM_VALID; 2210 2211 if (snaplen > res) 2212 snaplen = res; 2213 2214 if (sk->sk_type == SOCK_DGRAM) { 2215 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2216 po->tp_reserve; 2217 } else { 2218 unsigned int maclen = skb_network_offset(skb); 2219 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2220 (maclen < 16 ? 16 : maclen)) + 2221 po->tp_reserve; 2222 if (po->has_vnet_hdr) { 2223 netoff += sizeof(struct virtio_net_hdr); 2224 do_vnet = true; 2225 } 2226 macoff = netoff - maclen; 2227 } 2228 if (po->tp_version <= TPACKET_V2) { 2229 if (macoff + snaplen > po->rx_ring.frame_size) { 2230 if (po->copy_thresh && 2231 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2232 if (skb_shared(skb)) { 2233 copy_skb = skb_clone(skb, GFP_ATOMIC); 2234 } else { 2235 copy_skb = skb_get(skb); 2236 skb_head = skb->data; 2237 } 2238 if (copy_skb) 2239 skb_set_owner_r(copy_skb, sk); 2240 } 2241 snaplen = po->rx_ring.frame_size - macoff; 2242 if ((int)snaplen < 0) { 2243 snaplen = 0; 2244 do_vnet = false; 2245 } 2246 } 2247 } else if (unlikely(macoff + snaplen > 2248 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2249 u32 nval; 2250 2251 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2252 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2253 snaplen, nval, macoff); 2254 snaplen = nval; 2255 if (unlikely((int)snaplen < 0)) { 2256 snaplen = 0; 2257 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2258 do_vnet = false; 2259 } 2260 } 2261 spin_lock(&sk->sk_receive_queue.lock); 2262 h.raw = packet_current_rx_frame(po, skb, 2263 TP_STATUS_KERNEL, (macoff+snaplen)); 2264 if (!h.raw) 2265 goto drop_n_account; 2266 if (po->tp_version <= TPACKET_V2) { 2267 packet_increment_rx_head(po, &po->rx_ring); 2268 /* 2269 * LOSING will be reported till you read the stats, 2270 * because it's COR - Clear On Read. 2271 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2272 * at packet level. 2273 */ 2274 if (po->stats.stats1.tp_drops) 2275 status |= TP_STATUS_LOSING; 2276 } 2277 2278 if (do_vnet && 2279 virtio_net_hdr_from_skb(skb, h.raw + macoff - 2280 sizeof(struct virtio_net_hdr), 2281 vio_le(), true, 0)) 2282 goto drop_n_account; 2283 2284 po->stats.stats1.tp_packets++; 2285 if (copy_skb) { 2286 status |= TP_STATUS_COPY; 2287 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2288 } 2289 spin_unlock(&sk->sk_receive_queue.lock); 2290 2291 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2292 2293 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 2294 getnstimeofday(&ts); 2295 2296 status |= ts_status; 2297 2298 switch (po->tp_version) { 2299 case TPACKET_V1: 2300 h.h1->tp_len = skb->len; 2301 h.h1->tp_snaplen = snaplen; 2302 h.h1->tp_mac = macoff; 2303 h.h1->tp_net = netoff; 2304 h.h1->tp_sec = ts.tv_sec; 2305 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2306 hdrlen = sizeof(*h.h1); 2307 break; 2308 case TPACKET_V2: 2309 h.h2->tp_len = skb->len; 2310 h.h2->tp_snaplen = snaplen; 2311 h.h2->tp_mac = macoff; 2312 h.h2->tp_net = netoff; 2313 h.h2->tp_sec = ts.tv_sec; 2314 h.h2->tp_nsec = ts.tv_nsec; 2315 if (skb_vlan_tag_present(skb)) { 2316 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2317 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2318 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2319 } else { 2320 h.h2->tp_vlan_tci = 0; 2321 h.h2->tp_vlan_tpid = 0; 2322 } 2323 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2324 hdrlen = sizeof(*h.h2); 2325 break; 2326 case TPACKET_V3: 2327 /* tp_nxt_offset,vlan are already populated above. 2328 * So DONT clear those fields here 2329 */ 2330 h.h3->tp_status |= status; 2331 h.h3->tp_len = skb->len; 2332 h.h3->tp_snaplen = snaplen; 2333 h.h3->tp_mac = macoff; 2334 h.h3->tp_net = netoff; 2335 h.h3->tp_sec = ts.tv_sec; 2336 h.h3->tp_nsec = ts.tv_nsec; 2337 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2338 hdrlen = sizeof(*h.h3); 2339 break; 2340 default: 2341 BUG(); 2342 } 2343 2344 sll = h.raw + TPACKET_ALIGN(hdrlen); 2345 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2346 sll->sll_family = AF_PACKET; 2347 sll->sll_hatype = dev->type; 2348 sll->sll_protocol = skb->protocol; 2349 sll->sll_pkttype = skb->pkt_type; 2350 if (unlikely(po->origdev)) 2351 sll->sll_ifindex = orig_dev->ifindex; 2352 else 2353 sll->sll_ifindex = dev->ifindex; 2354 2355 smp_mb(); 2356 2357 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2358 if (po->tp_version <= TPACKET_V2) { 2359 u8 *start, *end; 2360 2361 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2362 macoff + snaplen); 2363 2364 for (start = h.raw; start < end; start += PAGE_SIZE) 2365 flush_dcache_page(pgv_to_page(start)); 2366 } 2367 smp_wmb(); 2368 #endif 2369 2370 if (po->tp_version <= TPACKET_V2) { 2371 __packet_set_status(po, h.raw, status); 2372 sk->sk_data_ready(sk); 2373 } else { 2374 prb_clear_blk_fill_status(&po->rx_ring); 2375 } 2376 2377 drop_n_restore: 2378 if (skb_head != skb->data && skb_shared(skb)) { 2379 skb->data = skb_head; 2380 skb->len = skb_len; 2381 } 2382 drop: 2383 if (!is_drop_n_account) 2384 consume_skb(skb); 2385 else 2386 kfree_skb(skb); 2387 return 0; 2388 2389 drop_n_account: 2390 is_drop_n_account = true; 2391 po->stats.stats1.tp_drops++; 2392 spin_unlock(&sk->sk_receive_queue.lock); 2393 2394 sk->sk_data_ready(sk); 2395 kfree_skb(copy_skb); 2396 goto drop_n_restore; 2397 } 2398 2399 static void tpacket_destruct_skb(struct sk_buff *skb) 2400 { 2401 struct packet_sock *po = pkt_sk(skb->sk); 2402 2403 if (likely(po->tx_ring.pg_vec)) { 2404 void *ph; 2405 __u32 ts; 2406 2407 ph = skb_zcopy_get_nouarg(skb); 2408 packet_dec_pending(&po->tx_ring); 2409 2410 ts = __packet_set_timestamp(po, ph, skb); 2411 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2412 } 2413 2414 sock_wfree(skb); 2415 } 2416 2417 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2418 { 2419 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2420 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2421 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2422 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2423 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2424 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2425 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2426 2427 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2428 return -EINVAL; 2429 2430 return 0; 2431 } 2432 2433 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2434 struct virtio_net_hdr *vnet_hdr) 2435 { 2436 if (*len < sizeof(*vnet_hdr)) 2437 return -EINVAL; 2438 *len -= sizeof(*vnet_hdr); 2439 2440 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2441 return -EFAULT; 2442 2443 return __packet_snd_vnet_parse(vnet_hdr, *len); 2444 } 2445 2446 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2447 void *frame, struct net_device *dev, void *data, int tp_len, 2448 __be16 proto, unsigned char *addr, int hlen, int copylen, 2449 const struct sockcm_cookie *sockc) 2450 { 2451 union tpacket_uhdr ph; 2452 int to_write, offset, len, nr_frags, len_max; 2453 struct socket *sock = po->sk.sk_socket; 2454 struct page *page; 2455 int err; 2456 2457 ph.raw = frame; 2458 2459 skb->protocol = proto; 2460 skb->dev = dev; 2461 skb->priority = po->sk.sk_priority; 2462 skb->mark = po->sk.sk_mark; 2463 skb->tstamp = sockc->transmit_time; 2464 skb_setup_tx_timestamp(skb, sockc->tsflags); 2465 skb_zcopy_set_nouarg(skb, ph.raw); 2466 2467 skb_reserve(skb, hlen); 2468 skb_reset_network_header(skb); 2469 2470 to_write = tp_len; 2471 2472 if (sock->type == SOCK_DGRAM) { 2473 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2474 NULL, tp_len); 2475 if (unlikely(err < 0)) 2476 return -EINVAL; 2477 } else if (copylen) { 2478 int hdrlen = min_t(int, copylen, tp_len); 2479 2480 skb_push(skb, dev->hard_header_len); 2481 skb_put(skb, copylen - dev->hard_header_len); 2482 err = skb_store_bits(skb, 0, data, hdrlen); 2483 if (unlikely(err)) 2484 return err; 2485 if (!dev_validate_header(dev, skb->data, hdrlen)) 2486 return -EINVAL; 2487 2488 data += hdrlen; 2489 to_write -= hdrlen; 2490 } 2491 2492 offset = offset_in_page(data); 2493 len_max = PAGE_SIZE - offset; 2494 len = ((to_write > len_max) ? len_max : to_write); 2495 2496 skb->data_len = to_write; 2497 skb->len += to_write; 2498 skb->truesize += to_write; 2499 refcount_add(to_write, &po->sk.sk_wmem_alloc); 2500 2501 while (likely(to_write)) { 2502 nr_frags = skb_shinfo(skb)->nr_frags; 2503 2504 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2505 pr_err("Packet exceed the number of skb frags(%lu)\n", 2506 MAX_SKB_FRAGS); 2507 return -EFAULT; 2508 } 2509 2510 page = pgv_to_page(data); 2511 data += len; 2512 flush_dcache_page(page); 2513 get_page(page); 2514 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2515 to_write -= len; 2516 offset = 0; 2517 len_max = PAGE_SIZE; 2518 len = ((to_write > len_max) ? len_max : to_write); 2519 } 2520 2521 packet_parse_headers(skb, sock); 2522 2523 return tp_len; 2524 } 2525 2526 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2527 int size_max, void **data) 2528 { 2529 union tpacket_uhdr ph; 2530 int tp_len, off; 2531 2532 ph.raw = frame; 2533 2534 switch (po->tp_version) { 2535 case TPACKET_V3: 2536 if (ph.h3->tp_next_offset != 0) { 2537 pr_warn_once("variable sized slot not supported"); 2538 return -EINVAL; 2539 } 2540 tp_len = ph.h3->tp_len; 2541 break; 2542 case TPACKET_V2: 2543 tp_len = ph.h2->tp_len; 2544 break; 2545 default: 2546 tp_len = ph.h1->tp_len; 2547 break; 2548 } 2549 if (unlikely(tp_len > size_max)) { 2550 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2551 return -EMSGSIZE; 2552 } 2553 2554 if (unlikely(po->tp_tx_has_off)) { 2555 int off_min, off_max; 2556 2557 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2558 off_max = po->tx_ring.frame_size - tp_len; 2559 if (po->sk.sk_type == SOCK_DGRAM) { 2560 switch (po->tp_version) { 2561 case TPACKET_V3: 2562 off = ph.h3->tp_net; 2563 break; 2564 case TPACKET_V2: 2565 off = ph.h2->tp_net; 2566 break; 2567 default: 2568 off = ph.h1->tp_net; 2569 break; 2570 } 2571 } else { 2572 switch (po->tp_version) { 2573 case TPACKET_V3: 2574 off = ph.h3->tp_mac; 2575 break; 2576 case TPACKET_V2: 2577 off = ph.h2->tp_mac; 2578 break; 2579 default: 2580 off = ph.h1->tp_mac; 2581 break; 2582 } 2583 } 2584 if (unlikely((off < off_min) || (off_max < off))) 2585 return -EINVAL; 2586 } else { 2587 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2588 } 2589 2590 *data = frame + off; 2591 return tp_len; 2592 } 2593 2594 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2595 { 2596 struct sk_buff *skb; 2597 struct net_device *dev; 2598 struct virtio_net_hdr *vnet_hdr = NULL; 2599 struct sockcm_cookie sockc; 2600 __be16 proto; 2601 int err, reserve = 0; 2602 void *ph; 2603 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2604 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2605 unsigned char *addr = NULL; 2606 int tp_len, size_max; 2607 void *data; 2608 int len_sum = 0; 2609 int status = TP_STATUS_AVAILABLE; 2610 int hlen, tlen, copylen = 0; 2611 2612 mutex_lock(&po->pg_vec_lock); 2613 2614 if (likely(saddr == NULL)) { 2615 dev = packet_cached_dev_get(po); 2616 proto = po->num; 2617 } else { 2618 err = -EINVAL; 2619 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2620 goto out; 2621 if (msg->msg_namelen < (saddr->sll_halen 2622 + offsetof(struct sockaddr_ll, 2623 sll_addr))) 2624 goto out; 2625 proto = saddr->sll_protocol; 2626 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2627 if (po->sk.sk_socket->type == SOCK_DGRAM) { 2628 if (dev && msg->msg_namelen < dev->addr_len + 2629 offsetof(struct sockaddr_ll, sll_addr)) 2630 goto out_put; 2631 addr = saddr->sll_addr; 2632 } 2633 } 2634 2635 err = -ENXIO; 2636 if (unlikely(dev == NULL)) 2637 goto out; 2638 err = -ENETDOWN; 2639 if (unlikely(!(dev->flags & IFF_UP))) 2640 goto out_put; 2641 2642 sockcm_init(&sockc, &po->sk); 2643 if (msg->msg_controllen) { 2644 err = sock_cmsg_send(&po->sk, msg, &sockc); 2645 if (unlikely(err)) 2646 goto out_put; 2647 } 2648 2649 if (po->sk.sk_socket->type == SOCK_RAW) 2650 reserve = dev->hard_header_len; 2651 size_max = po->tx_ring.frame_size 2652 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2653 2654 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) 2655 size_max = dev->mtu + reserve + VLAN_HLEN; 2656 2657 do { 2658 ph = packet_current_frame(po, &po->tx_ring, 2659 TP_STATUS_SEND_REQUEST); 2660 if (unlikely(ph == NULL)) { 2661 if (need_wait && need_resched()) 2662 schedule(); 2663 continue; 2664 } 2665 2666 skb = NULL; 2667 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2668 if (tp_len < 0) 2669 goto tpacket_error; 2670 2671 status = TP_STATUS_SEND_REQUEST; 2672 hlen = LL_RESERVED_SPACE(dev); 2673 tlen = dev->needed_tailroom; 2674 if (po->has_vnet_hdr) { 2675 vnet_hdr = data; 2676 data += sizeof(*vnet_hdr); 2677 tp_len -= sizeof(*vnet_hdr); 2678 if (tp_len < 0 || 2679 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2680 tp_len = -EINVAL; 2681 goto tpacket_error; 2682 } 2683 copylen = __virtio16_to_cpu(vio_le(), 2684 vnet_hdr->hdr_len); 2685 } 2686 copylen = max_t(int, copylen, dev->hard_header_len); 2687 skb = sock_alloc_send_skb(&po->sk, 2688 hlen + tlen + sizeof(struct sockaddr_ll) + 2689 (copylen - dev->hard_header_len), 2690 !need_wait, &err); 2691 2692 if (unlikely(skb == NULL)) { 2693 /* we assume the socket was initially writeable ... */ 2694 if (likely(len_sum > 0)) 2695 err = len_sum; 2696 goto out_status; 2697 } 2698 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2699 addr, hlen, copylen, &sockc); 2700 if (likely(tp_len >= 0) && 2701 tp_len > dev->mtu + reserve && 2702 !po->has_vnet_hdr && 2703 !packet_extra_vlan_len_allowed(dev, skb)) 2704 tp_len = -EMSGSIZE; 2705 2706 if (unlikely(tp_len < 0)) { 2707 tpacket_error: 2708 if (po->tp_loss) { 2709 __packet_set_status(po, ph, 2710 TP_STATUS_AVAILABLE); 2711 packet_increment_head(&po->tx_ring); 2712 kfree_skb(skb); 2713 continue; 2714 } else { 2715 status = TP_STATUS_WRONG_FORMAT; 2716 err = tp_len; 2717 goto out_status; 2718 } 2719 } 2720 2721 if (po->has_vnet_hdr) { 2722 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) { 2723 tp_len = -EINVAL; 2724 goto tpacket_error; 2725 } 2726 virtio_net_hdr_set_proto(skb, vnet_hdr); 2727 } 2728 2729 skb->destructor = tpacket_destruct_skb; 2730 __packet_set_status(po, ph, TP_STATUS_SENDING); 2731 packet_inc_pending(&po->tx_ring); 2732 2733 status = TP_STATUS_SEND_REQUEST; 2734 err = po->xmit(skb); 2735 if (unlikely(err > 0)) { 2736 err = net_xmit_errno(err); 2737 if (err && __packet_get_status(po, ph) == 2738 TP_STATUS_AVAILABLE) { 2739 /* skb was destructed already */ 2740 skb = NULL; 2741 goto out_status; 2742 } 2743 /* 2744 * skb was dropped but not destructed yet; 2745 * let's treat it like congestion or err < 0 2746 */ 2747 err = 0; 2748 } 2749 packet_increment_head(&po->tx_ring); 2750 len_sum += tp_len; 2751 } while (likely((ph != NULL) || 2752 /* Note: packet_read_pending() might be slow if we have 2753 * to call it as it's per_cpu variable, but in fast-path 2754 * we already short-circuit the loop with the first 2755 * condition, and luckily don't have to go that path 2756 * anyway. 2757 */ 2758 (need_wait && packet_read_pending(&po->tx_ring)))); 2759 2760 err = len_sum; 2761 goto out_put; 2762 2763 out_status: 2764 __packet_set_status(po, ph, status); 2765 kfree_skb(skb); 2766 out_put: 2767 dev_put(dev); 2768 out: 2769 mutex_unlock(&po->pg_vec_lock); 2770 return err; 2771 } 2772 2773 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2774 size_t reserve, size_t len, 2775 size_t linear, int noblock, 2776 int *err) 2777 { 2778 struct sk_buff *skb; 2779 2780 /* Under a page? Don't bother with paged skb. */ 2781 if (prepad + len < PAGE_SIZE || !linear) 2782 linear = len; 2783 2784 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2785 err, 0); 2786 if (!skb) 2787 return NULL; 2788 2789 skb_reserve(skb, reserve); 2790 skb_put(skb, linear); 2791 skb->data_len = len - linear; 2792 skb->len += len - linear; 2793 2794 return skb; 2795 } 2796 2797 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2798 { 2799 struct sock *sk = sock->sk; 2800 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2801 struct sk_buff *skb; 2802 struct net_device *dev; 2803 __be16 proto; 2804 unsigned char *addr = NULL; 2805 int err, reserve = 0; 2806 struct sockcm_cookie sockc; 2807 struct virtio_net_hdr vnet_hdr = { 0 }; 2808 int offset = 0; 2809 struct packet_sock *po = pkt_sk(sk); 2810 bool has_vnet_hdr = false; 2811 int hlen, tlen, linear; 2812 int extra_len = 0; 2813 2814 /* 2815 * Get and verify the address. 2816 */ 2817 2818 if (likely(saddr == NULL)) { 2819 dev = packet_cached_dev_get(po); 2820 proto = po->num; 2821 } else { 2822 err = -EINVAL; 2823 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2824 goto out; 2825 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2826 goto out; 2827 proto = saddr->sll_protocol; 2828 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2829 if (sock->type == SOCK_DGRAM) { 2830 if (dev && msg->msg_namelen < dev->addr_len + 2831 offsetof(struct sockaddr_ll, sll_addr)) 2832 goto out_unlock; 2833 addr = saddr->sll_addr; 2834 } 2835 } 2836 2837 err = -ENXIO; 2838 if (unlikely(dev == NULL)) 2839 goto out_unlock; 2840 err = -ENETDOWN; 2841 if (unlikely(!(dev->flags & IFF_UP))) 2842 goto out_unlock; 2843 2844 sockcm_init(&sockc, sk); 2845 sockc.mark = sk->sk_mark; 2846 if (msg->msg_controllen) { 2847 err = sock_cmsg_send(sk, msg, &sockc); 2848 if (unlikely(err)) 2849 goto out_unlock; 2850 } 2851 2852 if (sock->type == SOCK_RAW) 2853 reserve = dev->hard_header_len; 2854 if (po->has_vnet_hdr) { 2855 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2856 if (err) 2857 goto out_unlock; 2858 has_vnet_hdr = true; 2859 } 2860 2861 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2862 if (!netif_supports_nofcs(dev)) { 2863 err = -EPROTONOSUPPORT; 2864 goto out_unlock; 2865 } 2866 extra_len = 4; /* We're doing our own CRC */ 2867 } 2868 2869 err = -EMSGSIZE; 2870 if (!vnet_hdr.gso_type && 2871 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2872 goto out_unlock; 2873 2874 err = -ENOBUFS; 2875 hlen = LL_RESERVED_SPACE(dev); 2876 tlen = dev->needed_tailroom; 2877 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); 2878 linear = max(linear, min_t(int, len, dev->hard_header_len)); 2879 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, 2880 msg->msg_flags & MSG_DONTWAIT, &err); 2881 if (skb == NULL) 2882 goto out_unlock; 2883 2884 skb_reset_network_header(skb); 2885 2886 err = -EINVAL; 2887 if (sock->type == SOCK_DGRAM) { 2888 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 2889 if (unlikely(offset < 0)) 2890 goto out_free; 2891 } else if (reserve) { 2892 skb_reserve(skb, -reserve); 2893 if (len < reserve + sizeof(struct ipv6hdr) && 2894 dev->min_header_len != dev->hard_header_len) 2895 skb_reset_network_header(skb); 2896 } 2897 2898 /* Returns -EFAULT on error */ 2899 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 2900 if (err) 2901 goto out_free; 2902 2903 if (sock->type == SOCK_RAW && 2904 !dev_validate_header(dev, skb->data, len)) { 2905 err = -EINVAL; 2906 goto out_free; 2907 } 2908 2909 skb_setup_tx_timestamp(skb, sockc.tsflags); 2910 2911 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 2912 !packet_extra_vlan_len_allowed(dev, skb)) { 2913 err = -EMSGSIZE; 2914 goto out_free; 2915 } 2916 2917 skb->protocol = proto; 2918 skb->dev = dev; 2919 skb->priority = sk->sk_priority; 2920 skb->mark = sockc.mark; 2921 skb->tstamp = sockc.transmit_time; 2922 2923 if (has_vnet_hdr) { 2924 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 2925 if (err) 2926 goto out_free; 2927 len += sizeof(vnet_hdr); 2928 virtio_net_hdr_set_proto(skb, &vnet_hdr); 2929 } 2930 2931 packet_parse_headers(skb, sock); 2932 2933 if (unlikely(extra_len == 4)) 2934 skb->no_fcs = 1; 2935 2936 err = po->xmit(skb); 2937 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2938 goto out_unlock; 2939 2940 dev_put(dev); 2941 2942 return len; 2943 2944 out_free: 2945 kfree_skb(skb); 2946 out_unlock: 2947 if (dev) 2948 dev_put(dev); 2949 out: 2950 return err; 2951 } 2952 2953 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2954 { 2955 struct sock *sk = sock->sk; 2956 struct packet_sock *po = pkt_sk(sk); 2957 2958 if (po->tx_ring.pg_vec) 2959 return tpacket_snd(po, msg); 2960 else 2961 return packet_snd(sock, msg, len); 2962 } 2963 2964 /* 2965 * Close a PACKET socket. This is fairly simple. We immediately go 2966 * to 'closed' state and remove our protocol entry in the device list. 2967 */ 2968 2969 static int packet_release(struct socket *sock) 2970 { 2971 struct sock *sk = sock->sk; 2972 struct packet_sock *po; 2973 struct packet_fanout *f; 2974 struct net *net; 2975 union tpacket_req_u req_u; 2976 2977 if (!sk) 2978 return 0; 2979 2980 net = sock_net(sk); 2981 po = pkt_sk(sk); 2982 2983 mutex_lock(&net->packet.sklist_lock); 2984 sk_del_node_init_rcu(sk); 2985 mutex_unlock(&net->packet.sklist_lock); 2986 2987 preempt_disable(); 2988 sock_prot_inuse_add(net, sk->sk_prot, -1); 2989 preempt_enable(); 2990 2991 spin_lock(&po->bind_lock); 2992 unregister_prot_hook(sk, false); 2993 packet_cached_dev_reset(po); 2994 2995 if (po->prot_hook.dev) { 2996 dev_put(po->prot_hook.dev); 2997 po->prot_hook.dev = NULL; 2998 } 2999 spin_unlock(&po->bind_lock); 3000 3001 packet_flush_mclist(sk); 3002 3003 lock_sock(sk); 3004 if (po->rx_ring.pg_vec) { 3005 memset(&req_u, 0, sizeof(req_u)); 3006 packet_set_ring(sk, &req_u, 1, 0); 3007 } 3008 3009 if (po->tx_ring.pg_vec) { 3010 memset(&req_u, 0, sizeof(req_u)); 3011 packet_set_ring(sk, &req_u, 1, 1); 3012 } 3013 release_sock(sk); 3014 3015 f = fanout_release(sk); 3016 3017 synchronize_net(); 3018 3019 if (f) { 3020 kfree(po->rollover); 3021 fanout_release_data(f); 3022 kfree(f); 3023 } 3024 /* 3025 * Now the socket is dead. No more input will appear. 3026 */ 3027 sock_orphan(sk); 3028 sock->sk = NULL; 3029 3030 /* Purge queues */ 3031 3032 skb_queue_purge(&sk->sk_receive_queue); 3033 packet_free_pending(po); 3034 sk_refcnt_debug_release(sk); 3035 3036 sock_put(sk); 3037 return 0; 3038 } 3039 3040 /* 3041 * Attach a packet hook. 3042 */ 3043 3044 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3045 __be16 proto) 3046 { 3047 struct packet_sock *po = pkt_sk(sk); 3048 struct net_device *dev_curr; 3049 __be16 proto_curr; 3050 bool need_rehook; 3051 struct net_device *dev = NULL; 3052 int ret = 0; 3053 bool unlisted = false; 3054 3055 lock_sock(sk); 3056 spin_lock(&po->bind_lock); 3057 rcu_read_lock(); 3058 3059 if (po->fanout) { 3060 ret = -EINVAL; 3061 goto out_unlock; 3062 } 3063 3064 if (name) { 3065 dev = dev_get_by_name_rcu(sock_net(sk), name); 3066 if (!dev) { 3067 ret = -ENODEV; 3068 goto out_unlock; 3069 } 3070 } else if (ifindex) { 3071 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3072 if (!dev) { 3073 ret = -ENODEV; 3074 goto out_unlock; 3075 } 3076 } 3077 3078 if (dev) 3079 dev_hold(dev); 3080 3081 proto_curr = po->prot_hook.type; 3082 dev_curr = po->prot_hook.dev; 3083 3084 need_rehook = proto_curr != proto || dev_curr != dev; 3085 3086 if (need_rehook) { 3087 if (po->running) { 3088 rcu_read_unlock(); 3089 /* prevents packet_notifier() from calling 3090 * register_prot_hook() 3091 */ 3092 po->num = 0; 3093 __unregister_prot_hook(sk, true); 3094 rcu_read_lock(); 3095 dev_curr = po->prot_hook.dev; 3096 if (dev) 3097 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3098 dev->ifindex); 3099 } 3100 3101 BUG_ON(po->running); 3102 po->num = proto; 3103 po->prot_hook.type = proto; 3104 3105 if (unlikely(unlisted)) { 3106 dev_put(dev); 3107 po->prot_hook.dev = NULL; 3108 po->ifindex = -1; 3109 packet_cached_dev_reset(po); 3110 } else { 3111 po->prot_hook.dev = dev; 3112 po->ifindex = dev ? dev->ifindex : 0; 3113 packet_cached_dev_assign(po, dev); 3114 } 3115 } 3116 if (dev_curr) 3117 dev_put(dev_curr); 3118 3119 if (proto == 0 || !need_rehook) 3120 goto out_unlock; 3121 3122 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3123 register_prot_hook(sk); 3124 } else { 3125 sk->sk_err = ENETDOWN; 3126 if (!sock_flag(sk, SOCK_DEAD)) 3127 sk->sk_error_report(sk); 3128 } 3129 3130 out_unlock: 3131 rcu_read_unlock(); 3132 spin_unlock(&po->bind_lock); 3133 release_sock(sk); 3134 return ret; 3135 } 3136 3137 /* 3138 * Bind a packet socket to a device 3139 */ 3140 3141 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3142 int addr_len) 3143 { 3144 struct sock *sk = sock->sk; 3145 char name[sizeof(uaddr->sa_data) + 1]; 3146 3147 /* 3148 * Check legality 3149 */ 3150 3151 if (addr_len != sizeof(struct sockaddr)) 3152 return -EINVAL; 3153 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be 3154 * zero-terminated. 3155 */ 3156 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); 3157 name[sizeof(uaddr->sa_data)] = 0; 3158 3159 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3160 } 3161 3162 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3163 { 3164 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3165 struct sock *sk = sock->sk; 3166 3167 /* 3168 * Check legality 3169 */ 3170 3171 if (addr_len < sizeof(struct sockaddr_ll)) 3172 return -EINVAL; 3173 if (sll->sll_family != AF_PACKET) 3174 return -EINVAL; 3175 3176 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3177 sll->sll_protocol ? : pkt_sk(sk)->num); 3178 } 3179 3180 static struct proto packet_proto = { 3181 .name = "PACKET", 3182 .owner = THIS_MODULE, 3183 .obj_size = sizeof(struct packet_sock), 3184 }; 3185 3186 /* 3187 * Create a packet of type SOCK_PACKET. 3188 */ 3189 3190 static int packet_create(struct net *net, struct socket *sock, int protocol, 3191 int kern) 3192 { 3193 struct sock *sk; 3194 struct packet_sock *po; 3195 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3196 int err; 3197 3198 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3199 return -EPERM; 3200 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3201 sock->type != SOCK_PACKET) 3202 return -ESOCKTNOSUPPORT; 3203 3204 sock->state = SS_UNCONNECTED; 3205 3206 err = -ENOBUFS; 3207 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3208 if (sk == NULL) 3209 goto out; 3210 3211 sock->ops = &packet_ops; 3212 if (sock->type == SOCK_PACKET) 3213 sock->ops = &packet_ops_spkt; 3214 3215 sock_init_data(sock, sk); 3216 3217 po = pkt_sk(sk); 3218 sk->sk_family = PF_PACKET; 3219 po->num = proto; 3220 po->xmit = dev_queue_xmit; 3221 3222 err = packet_alloc_pending(po); 3223 if (err) 3224 goto out2; 3225 3226 packet_cached_dev_reset(po); 3227 3228 sk->sk_destruct = packet_sock_destruct; 3229 sk_refcnt_debug_inc(sk); 3230 3231 /* 3232 * Attach a protocol block 3233 */ 3234 3235 spin_lock_init(&po->bind_lock); 3236 mutex_init(&po->pg_vec_lock); 3237 po->rollover = NULL; 3238 po->prot_hook.func = packet_rcv; 3239 3240 if (sock->type == SOCK_PACKET) 3241 po->prot_hook.func = packet_rcv_spkt; 3242 3243 po->prot_hook.af_packet_priv = sk; 3244 3245 if (proto) { 3246 po->prot_hook.type = proto; 3247 __register_prot_hook(sk); 3248 } 3249 3250 mutex_lock(&net->packet.sklist_lock); 3251 sk_add_node_tail_rcu(sk, &net->packet.sklist); 3252 mutex_unlock(&net->packet.sklist_lock); 3253 3254 preempt_disable(); 3255 sock_prot_inuse_add(net, &packet_proto, 1); 3256 preempt_enable(); 3257 3258 return 0; 3259 out2: 3260 sk_free(sk); 3261 out: 3262 return err; 3263 } 3264 3265 /* 3266 * Pull a packet from our receive queue and hand it to the user. 3267 * If necessary we block. 3268 */ 3269 3270 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3271 int flags) 3272 { 3273 struct sock *sk = sock->sk; 3274 struct sk_buff *skb; 3275 int copied, err; 3276 int vnet_hdr_len = 0; 3277 unsigned int origlen = 0; 3278 3279 err = -EINVAL; 3280 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3281 goto out; 3282 3283 #if 0 3284 /* What error should we return now? EUNATTACH? */ 3285 if (pkt_sk(sk)->ifindex < 0) 3286 return -ENODEV; 3287 #endif 3288 3289 if (flags & MSG_ERRQUEUE) { 3290 err = sock_recv_errqueue(sk, msg, len, 3291 SOL_PACKET, PACKET_TX_TIMESTAMP); 3292 goto out; 3293 } 3294 3295 /* 3296 * Call the generic datagram receiver. This handles all sorts 3297 * of horrible races and re-entrancy so we can forget about it 3298 * in the protocol layers. 3299 * 3300 * Now it will return ENETDOWN, if device have just gone down, 3301 * but then it will block. 3302 */ 3303 3304 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 3305 3306 /* 3307 * An error occurred so return it. Because skb_recv_datagram() 3308 * handles the blocking we don't see and worry about blocking 3309 * retries. 3310 */ 3311 3312 if (skb == NULL) 3313 goto out; 3314 3315 if (pkt_sk(sk)->pressure) 3316 packet_rcv_has_room(pkt_sk(sk), NULL); 3317 3318 if (pkt_sk(sk)->has_vnet_hdr) { 3319 err = packet_rcv_vnet(msg, skb, &len); 3320 if (err) 3321 goto out_free; 3322 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3323 } 3324 3325 /* You lose any data beyond the buffer you gave. If it worries 3326 * a user program they can ask the device for its MTU 3327 * anyway. 3328 */ 3329 copied = skb->len; 3330 if (copied > len) { 3331 copied = len; 3332 msg->msg_flags |= MSG_TRUNC; 3333 } 3334 3335 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3336 if (err) 3337 goto out_free; 3338 3339 if (sock->type != SOCK_PACKET) { 3340 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3341 3342 /* Original length was stored in sockaddr_ll fields */ 3343 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3344 sll->sll_family = AF_PACKET; 3345 sll->sll_protocol = skb->protocol; 3346 } 3347 3348 sock_recv_ts_and_drops(msg, sk, skb); 3349 3350 if (msg->msg_name) { 3351 int copy_len; 3352 3353 /* If the address length field is there to be filled 3354 * in, we fill it in now. 3355 */ 3356 if (sock->type == SOCK_PACKET) { 3357 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3358 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3359 copy_len = msg->msg_namelen; 3360 } else { 3361 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3362 3363 msg->msg_namelen = sll->sll_halen + 3364 offsetof(struct sockaddr_ll, sll_addr); 3365 copy_len = msg->msg_namelen; 3366 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { 3367 memset(msg->msg_name + 3368 offsetof(struct sockaddr_ll, sll_addr), 3369 0, sizeof(sll->sll_addr)); 3370 msg->msg_namelen = sizeof(struct sockaddr_ll); 3371 } 3372 } 3373 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); 3374 } 3375 3376 if (pkt_sk(sk)->auxdata) { 3377 struct tpacket_auxdata aux; 3378 3379 aux.tp_status = TP_STATUS_USER; 3380 if (skb->ip_summed == CHECKSUM_PARTIAL) 3381 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3382 else if (skb->pkt_type != PACKET_OUTGOING && 3383 (skb->ip_summed == CHECKSUM_COMPLETE || 3384 skb_csum_unnecessary(skb))) 3385 aux.tp_status |= TP_STATUS_CSUM_VALID; 3386 3387 aux.tp_len = origlen; 3388 aux.tp_snaplen = skb->len; 3389 aux.tp_mac = 0; 3390 aux.tp_net = skb_network_offset(skb); 3391 if (skb_vlan_tag_present(skb)) { 3392 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3393 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3394 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3395 } else { 3396 aux.tp_vlan_tci = 0; 3397 aux.tp_vlan_tpid = 0; 3398 } 3399 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3400 } 3401 3402 /* 3403 * Free or return the buffer as appropriate. Again this 3404 * hides all the races and re-entrancy issues from us. 3405 */ 3406 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3407 3408 out_free: 3409 skb_free_datagram(sk, skb); 3410 out: 3411 return err; 3412 } 3413 3414 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3415 int peer) 3416 { 3417 struct net_device *dev; 3418 struct sock *sk = sock->sk; 3419 3420 if (peer) 3421 return -EOPNOTSUPP; 3422 3423 uaddr->sa_family = AF_PACKET; 3424 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3425 rcu_read_lock(); 3426 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3427 if (dev) 3428 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3429 rcu_read_unlock(); 3430 3431 return sizeof(*uaddr); 3432 } 3433 3434 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3435 int peer) 3436 { 3437 struct net_device *dev; 3438 struct sock *sk = sock->sk; 3439 struct packet_sock *po = pkt_sk(sk); 3440 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3441 3442 if (peer) 3443 return -EOPNOTSUPP; 3444 3445 sll->sll_family = AF_PACKET; 3446 sll->sll_ifindex = po->ifindex; 3447 sll->sll_protocol = po->num; 3448 sll->sll_pkttype = 0; 3449 rcu_read_lock(); 3450 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 3451 if (dev) { 3452 sll->sll_hatype = dev->type; 3453 sll->sll_halen = dev->addr_len; 3454 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3455 } else { 3456 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3457 sll->sll_halen = 0; 3458 } 3459 rcu_read_unlock(); 3460 3461 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3462 } 3463 3464 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3465 int what) 3466 { 3467 switch (i->type) { 3468 case PACKET_MR_MULTICAST: 3469 if (i->alen != dev->addr_len) 3470 return -EINVAL; 3471 if (what > 0) 3472 return dev_mc_add(dev, i->addr); 3473 else 3474 return dev_mc_del(dev, i->addr); 3475 break; 3476 case PACKET_MR_PROMISC: 3477 return dev_set_promiscuity(dev, what); 3478 case PACKET_MR_ALLMULTI: 3479 return dev_set_allmulti(dev, what); 3480 case PACKET_MR_UNICAST: 3481 if (i->alen != dev->addr_len) 3482 return -EINVAL; 3483 if (what > 0) 3484 return dev_uc_add(dev, i->addr); 3485 else 3486 return dev_uc_del(dev, i->addr); 3487 break; 3488 default: 3489 break; 3490 } 3491 return 0; 3492 } 3493 3494 static void packet_dev_mclist_delete(struct net_device *dev, 3495 struct packet_mclist **mlp) 3496 { 3497 struct packet_mclist *ml; 3498 3499 while ((ml = *mlp) != NULL) { 3500 if (ml->ifindex == dev->ifindex) { 3501 packet_dev_mc(dev, ml, -1); 3502 *mlp = ml->next; 3503 kfree(ml); 3504 } else 3505 mlp = &ml->next; 3506 } 3507 } 3508 3509 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3510 { 3511 struct packet_sock *po = pkt_sk(sk); 3512 struct packet_mclist *ml, *i; 3513 struct net_device *dev; 3514 int err; 3515 3516 rtnl_lock(); 3517 3518 err = -ENODEV; 3519 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3520 if (!dev) 3521 goto done; 3522 3523 err = -EINVAL; 3524 if (mreq->mr_alen > dev->addr_len) 3525 goto done; 3526 3527 err = -ENOBUFS; 3528 i = kmalloc(sizeof(*i), GFP_KERNEL); 3529 if (i == NULL) 3530 goto done; 3531 3532 err = 0; 3533 for (ml = po->mclist; ml; ml = ml->next) { 3534 if (ml->ifindex == mreq->mr_ifindex && 3535 ml->type == mreq->mr_type && 3536 ml->alen == mreq->mr_alen && 3537 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3538 ml->count++; 3539 /* Free the new element ... */ 3540 kfree(i); 3541 goto done; 3542 } 3543 } 3544 3545 i->type = mreq->mr_type; 3546 i->ifindex = mreq->mr_ifindex; 3547 i->alen = mreq->mr_alen; 3548 memcpy(i->addr, mreq->mr_address, i->alen); 3549 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3550 i->count = 1; 3551 i->next = po->mclist; 3552 po->mclist = i; 3553 err = packet_dev_mc(dev, i, 1); 3554 if (err) { 3555 po->mclist = i->next; 3556 kfree(i); 3557 } 3558 3559 done: 3560 rtnl_unlock(); 3561 return err; 3562 } 3563 3564 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3565 { 3566 struct packet_mclist *ml, **mlp; 3567 3568 rtnl_lock(); 3569 3570 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3571 if (ml->ifindex == mreq->mr_ifindex && 3572 ml->type == mreq->mr_type && 3573 ml->alen == mreq->mr_alen && 3574 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3575 if (--ml->count == 0) { 3576 struct net_device *dev; 3577 *mlp = ml->next; 3578 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3579 if (dev) 3580 packet_dev_mc(dev, ml, -1); 3581 kfree(ml); 3582 } 3583 break; 3584 } 3585 } 3586 rtnl_unlock(); 3587 return 0; 3588 } 3589 3590 static void packet_flush_mclist(struct sock *sk) 3591 { 3592 struct packet_sock *po = pkt_sk(sk); 3593 struct packet_mclist *ml; 3594 3595 if (!po->mclist) 3596 return; 3597 3598 rtnl_lock(); 3599 while ((ml = po->mclist) != NULL) { 3600 struct net_device *dev; 3601 3602 po->mclist = ml->next; 3603 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3604 if (dev != NULL) 3605 packet_dev_mc(dev, ml, -1); 3606 kfree(ml); 3607 } 3608 rtnl_unlock(); 3609 } 3610 3611 static int 3612 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3613 { 3614 struct sock *sk = sock->sk; 3615 struct packet_sock *po = pkt_sk(sk); 3616 int ret; 3617 3618 if (level != SOL_PACKET) 3619 return -ENOPROTOOPT; 3620 3621 switch (optname) { 3622 case PACKET_ADD_MEMBERSHIP: 3623 case PACKET_DROP_MEMBERSHIP: 3624 { 3625 struct packet_mreq_max mreq; 3626 int len = optlen; 3627 memset(&mreq, 0, sizeof(mreq)); 3628 if (len < sizeof(struct packet_mreq)) 3629 return -EINVAL; 3630 if (len > sizeof(mreq)) 3631 len = sizeof(mreq); 3632 if (copy_from_user(&mreq, optval, len)) 3633 return -EFAULT; 3634 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3635 return -EINVAL; 3636 if (optname == PACKET_ADD_MEMBERSHIP) 3637 ret = packet_mc_add(sk, &mreq); 3638 else 3639 ret = packet_mc_drop(sk, &mreq); 3640 return ret; 3641 } 3642 3643 case PACKET_RX_RING: 3644 case PACKET_TX_RING: 3645 { 3646 union tpacket_req_u req_u; 3647 int len; 3648 3649 lock_sock(sk); 3650 switch (po->tp_version) { 3651 case TPACKET_V1: 3652 case TPACKET_V2: 3653 len = sizeof(req_u.req); 3654 break; 3655 case TPACKET_V3: 3656 default: 3657 len = sizeof(req_u.req3); 3658 break; 3659 } 3660 if (optlen < len) { 3661 ret = -EINVAL; 3662 } else { 3663 if (copy_from_user(&req_u.req, optval, len)) 3664 ret = -EFAULT; 3665 else 3666 ret = packet_set_ring(sk, &req_u, 0, 3667 optname == PACKET_TX_RING); 3668 } 3669 release_sock(sk); 3670 return ret; 3671 } 3672 case PACKET_COPY_THRESH: 3673 { 3674 int val; 3675 3676 if (optlen != sizeof(val)) 3677 return -EINVAL; 3678 if (copy_from_user(&val, optval, sizeof(val))) 3679 return -EFAULT; 3680 3681 pkt_sk(sk)->copy_thresh = val; 3682 return 0; 3683 } 3684 case PACKET_VERSION: 3685 { 3686 int val; 3687 3688 if (optlen != sizeof(val)) 3689 return -EINVAL; 3690 if (copy_from_user(&val, optval, sizeof(val))) 3691 return -EFAULT; 3692 switch (val) { 3693 case TPACKET_V1: 3694 case TPACKET_V2: 3695 case TPACKET_V3: 3696 break; 3697 default: 3698 return -EINVAL; 3699 } 3700 lock_sock(sk); 3701 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3702 ret = -EBUSY; 3703 } else { 3704 po->tp_version = val; 3705 ret = 0; 3706 } 3707 release_sock(sk); 3708 return ret; 3709 } 3710 case PACKET_RESERVE: 3711 { 3712 unsigned int val; 3713 3714 if (optlen != sizeof(val)) 3715 return -EINVAL; 3716 if (copy_from_user(&val, optval, sizeof(val))) 3717 return -EFAULT; 3718 if (val > INT_MAX) 3719 return -EINVAL; 3720 lock_sock(sk); 3721 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3722 ret = -EBUSY; 3723 } else { 3724 po->tp_reserve = val; 3725 ret = 0; 3726 } 3727 release_sock(sk); 3728 return ret; 3729 } 3730 case PACKET_LOSS: 3731 { 3732 unsigned int val; 3733 3734 if (optlen != sizeof(val)) 3735 return -EINVAL; 3736 if (copy_from_user(&val, optval, sizeof(val))) 3737 return -EFAULT; 3738 3739 lock_sock(sk); 3740 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3741 ret = -EBUSY; 3742 } else { 3743 po->tp_loss = !!val; 3744 ret = 0; 3745 } 3746 release_sock(sk); 3747 return ret; 3748 } 3749 case PACKET_AUXDATA: 3750 { 3751 int val; 3752 3753 if (optlen < sizeof(val)) 3754 return -EINVAL; 3755 if (copy_from_user(&val, optval, sizeof(val))) 3756 return -EFAULT; 3757 3758 lock_sock(sk); 3759 po->auxdata = !!val; 3760 release_sock(sk); 3761 return 0; 3762 } 3763 case PACKET_ORIGDEV: 3764 { 3765 int val; 3766 3767 if (optlen < sizeof(val)) 3768 return -EINVAL; 3769 if (copy_from_user(&val, optval, sizeof(val))) 3770 return -EFAULT; 3771 3772 lock_sock(sk); 3773 po->origdev = !!val; 3774 release_sock(sk); 3775 return 0; 3776 } 3777 case PACKET_VNET_HDR: 3778 { 3779 int val; 3780 3781 if (sock->type != SOCK_RAW) 3782 return -EINVAL; 3783 if (optlen < sizeof(val)) 3784 return -EINVAL; 3785 if (copy_from_user(&val, optval, sizeof(val))) 3786 return -EFAULT; 3787 3788 lock_sock(sk); 3789 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3790 ret = -EBUSY; 3791 } else { 3792 po->has_vnet_hdr = !!val; 3793 ret = 0; 3794 } 3795 release_sock(sk); 3796 return ret; 3797 } 3798 case PACKET_TIMESTAMP: 3799 { 3800 int val; 3801 3802 if (optlen != sizeof(val)) 3803 return -EINVAL; 3804 if (copy_from_user(&val, optval, sizeof(val))) 3805 return -EFAULT; 3806 3807 po->tp_tstamp = val; 3808 return 0; 3809 } 3810 case PACKET_FANOUT: 3811 { 3812 int val; 3813 3814 if (optlen != sizeof(val)) 3815 return -EINVAL; 3816 if (copy_from_user(&val, optval, sizeof(val))) 3817 return -EFAULT; 3818 3819 return fanout_add(sk, val & 0xffff, val >> 16); 3820 } 3821 case PACKET_FANOUT_DATA: 3822 { 3823 if (!po->fanout) 3824 return -EINVAL; 3825 3826 return fanout_set_data(po, optval, optlen); 3827 } 3828 case PACKET_IGNORE_OUTGOING: 3829 { 3830 int val; 3831 3832 if (optlen != sizeof(val)) 3833 return -EINVAL; 3834 if (copy_from_user(&val, optval, sizeof(val))) 3835 return -EFAULT; 3836 if (val < 0 || val > 1) 3837 return -EINVAL; 3838 3839 po->prot_hook.ignore_outgoing = !!val; 3840 return 0; 3841 } 3842 case PACKET_TX_HAS_OFF: 3843 { 3844 unsigned int val; 3845 3846 if (optlen != sizeof(val)) 3847 return -EINVAL; 3848 if (copy_from_user(&val, optval, sizeof(val))) 3849 return -EFAULT; 3850 3851 lock_sock(sk); 3852 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3853 ret = -EBUSY; 3854 } else { 3855 po->tp_tx_has_off = !!val; 3856 ret = 0; 3857 } 3858 release_sock(sk); 3859 return 0; 3860 } 3861 case PACKET_QDISC_BYPASS: 3862 { 3863 int val; 3864 3865 if (optlen != sizeof(val)) 3866 return -EINVAL; 3867 if (copy_from_user(&val, optval, sizeof(val))) 3868 return -EFAULT; 3869 3870 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; 3871 return 0; 3872 } 3873 default: 3874 return -ENOPROTOOPT; 3875 } 3876 } 3877 3878 static int packet_getsockopt(struct socket *sock, int level, int optname, 3879 char __user *optval, int __user *optlen) 3880 { 3881 int len; 3882 int val, lv = sizeof(val); 3883 struct sock *sk = sock->sk; 3884 struct packet_sock *po = pkt_sk(sk); 3885 void *data = &val; 3886 union tpacket_stats_u st; 3887 struct tpacket_rollover_stats rstats; 3888 3889 if (level != SOL_PACKET) 3890 return -ENOPROTOOPT; 3891 3892 if (get_user(len, optlen)) 3893 return -EFAULT; 3894 3895 if (len < 0) 3896 return -EINVAL; 3897 3898 switch (optname) { 3899 case PACKET_STATISTICS: 3900 spin_lock_bh(&sk->sk_receive_queue.lock); 3901 memcpy(&st, &po->stats, sizeof(st)); 3902 memset(&po->stats, 0, sizeof(po->stats)); 3903 spin_unlock_bh(&sk->sk_receive_queue.lock); 3904 3905 if (po->tp_version == TPACKET_V3) { 3906 lv = sizeof(struct tpacket_stats_v3); 3907 st.stats3.tp_packets += st.stats3.tp_drops; 3908 data = &st.stats3; 3909 } else { 3910 lv = sizeof(struct tpacket_stats); 3911 st.stats1.tp_packets += st.stats1.tp_drops; 3912 data = &st.stats1; 3913 } 3914 3915 break; 3916 case PACKET_AUXDATA: 3917 val = po->auxdata; 3918 break; 3919 case PACKET_ORIGDEV: 3920 val = po->origdev; 3921 break; 3922 case PACKET_VNET_HDR: 3923 val = po->has_vnet_hdr; 3924 break; 3925 case PACKET_VERSION: 3926 val = po->tp_version; 3927 break; 3928 case PACKET_HDRLEN: 3929 if (len > sizeof(int)) 3930 len = sizeof(int); 3931 if (len < sizeof(int)) 3932 return -EINVAL; 3933 if (copy_from_user(&val, optval, len)) 3934 return -EFAULT; 3935 switch (val) { 3936 case TPACKET_V1: 3937 val = sizeof(struct tpacket_hdr); 3938 break; 3939 case TPACKET_V2: 3940 val = sizeof(struct tpacket2_hdr); 3941 break; 3942 case TPACKET_V3: 3943 val = sizeof(struct tpacket3_hdr); 3944 break; 3945 default: 3946 return -EINVAL; 3947 } 3948 break; 3949 case PACKET_RESERVE: 3950 val = po->tp_reserve; 3951 break; 3952 case PACKET_LOSS: 3953 val = po->tp_loss; 3954 break; 3955 case PACKET_TIMESTAMP: 3956 val = po->tp_tstamp; 3957 break; 3958 case PACKET_FANOUT: 3959 val = (po->fanout ? 3960 ((u32)po->fanout->id | 3961 ((u32)po->fanout->type << 16) | 3962 ((u32)po->fanout->flags << 24)) : 3963 0); 3964 break; 3965 case PACKET_IGNORE_OUTGOING: 3966 val = po->prot_hook.ignore_outgoing; 3967 break; 3968 case PACKET_ROLLOVER_STATS: 3969 if (!po->rollover) 3970 return -EINVAL; 3971 rstats.tp_all = atomic_long_read(&po->rollover->num); 3972 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 3973 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 3974 data = &rstats; 3975 lv = sizeof(rstats); 3976 break; 3977 case PACKET_TX_HAS_OFF: 3978 val = po->tp_tx_has_off; 3979 break; 3980 case PACKET_QDISC_BYPASS: 3981 val = packet_use_direct_xmit(po); 3982 break; 3983 default: 3984 return -ENOPROTOOPT; 3985 } 3986 3987 if (len > lv) 3988 len = lv; 3989 if (put_user(len, optlen)) 3990 return -EFAULT; 3991 if (copy_to_user(optval, data, len)) 3992 return -EFAULT; 3993 return 0; 3994 } 3995 3996 3997 #ifdef CONFIG_COMPAT 3998 static int compat_packet_setsockopt(struct socket *sock, int level, int optname, 3999 char __user *optval, unsigned int optlen) 4000 { 4001 struct packet_sock *po = pkt_sk(sock->sk); 4002 4003 if (level != SOL_PACKET) 4004 return -ENOPROTOOPT; 4005 4006 if (optname == PACKET_FANOUT_DATA && 4007 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) { 4008 optval = (char __user *)get_compat_bpf_fprog(optval); 4009 if (!optval) 4010 return -EFAULT; 4011 optlen = sizeof(struct sock_fprog); 4012 } 4013 4014 return packet_setsockopt(sock, level, optname, optval, optlen); 4015 } 4016 #endif 4017 4018 static int packet_notifier(struct notifier_block *this, 4019 unsigned long msg, void *ptr) 4020 { 4021 struct sock *sk; 4022 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4023 struct net *net = dev_net(dev); 4024 4025 rcu_read_lock(); 4026 sk_for_each_rcu(sk, &net->packet.sklist) { 4027 struct packet_sock *po = pkt_sk(sk); 4028 4029 switch (msg) { 4030 case NETDEV_UNREGISTER: 4031 if (po->mclist) 4032 packet_dev_mclist_delete(dev, &po->mclist); 4033 /* fallthrough */ 4034 4035 case NETDEV_DOWN: 4036 if (dev->ifindex == po->ifindex) { 4037 spin_lock(&po->bind_lock); 4038 if (po->running) { 4039 __unregister_prot_hook(sk, false); 4040 sk->sk_err = ENETDOWN; 4041 if (!sock_flag(sk, SOCK_DEAD)) 4042 sk->sk_error_report(sk); 4043 } 4044 if (msg == NETDEV_UNREGISTER) { 4045 packet_cached_dev_reset(po); 4046 po->ifindex = -1; 4047 if (po->prot_hook.dev) 4048 dev_put(po->prot_hook.dev); 4049 po->prot_hook.dev = NULL; 4050 } 4051 spin_unlock(&po->bind_lock); 4052 } 4053 break; 4054 case NETDEV_UP: 4055 if (dev->ifindex == po->ifindex) { 4056 spin_lock(&po->bind_lock); 4057 if (po->num) 4058 register_prot_hook(sk); 4059 spin_unlock(&po->bind_lock); 4060 } 4061 break; 4062 } 4063 } 4064 rcu_read_unlock(); 4065 return NOTIFY_DONE; 4066 } 4067 4068 4069 static int packet_ioctl(struct socket *sock, unsigned int cmd, 4070 unsigned long arg) 4071 { 4072 struct sock *sk = sock->sk; 4073 4074 switch (cmd) { 4075 case SIOCOUTQ: 4076 { 4077 int amount = sk_wmem_alloc_get(sk); 4078 4079 return put_user(amount, (int __user *)arg); 4080 } 4081 case SIOCINQ: 4082 { 4083 struct sk_buff *skb; 4084 int amount = 0; 4085 4086 spin_lock_bh(&sk->sk_receive_queue.lock); 4087 skb = skb_peek(&sk->sk_receive_queue); 4088 if (skb) 4089 amount = skb->len; 4090 spin_unlock_bh(&sk->sk_receive_queue.lock); 4091 return put_user(amount, (int __user *)arg); 4092 } 4093 case SIOCGSTAMP: 4094 return sock_get_timestamp(sk, (struct timeval __user *)arg); 4095 case SIOCGSTAMPNS: 4096 return sock_get_timestampns(sk, (struct timespec __user *)arg); 4097 4098 #ifdef CONFIG_INET 4099 case SIOCADDRT: 4100 case SIOCDELRT: 4101 case SIOCDARP: 4102 case SIOCGARP: 4103 case SIOCSARP: 4104 case SIOCGIFADDR: 4105 case SIOCSIFADDR: 4106 case SIOCGIFBRDADDR: 4107 case SIOCSIFBRDADDR: 4108 case SIOCGIFNETMASK: 4109 case SIOCSIFNETMASK: 4110 case SIOCGIFDSTADDR: 4111 case SIOCSIFDSTADDR: 4112 case SIOCSIFFLAGS: 4113 return inet_dgram_ops.ioctl(sock, cmd, arg); 4114 #endif 4115 4116 default: 4117 return -ENOIOCTLCMD; 4118 } 4119 return 0; 4120 } 4121 4122 static __poll_t packet_poll(struct file *file, struct socket *sock, 4123 poll_table *wait) 4124 { 4125 struct sock *sk = sock->sk; 4126 struct packet_sock *po = pkt_sk(sk); 4127 __poll_t mask = datagram_poll(file, sock, wait); 4128 4129 spin_lock_bh(&sk->sk_receive_queue.lock); 4130 if (po->rx_ring.pg_vec) { 4131 if (!packet_previous_rx_frame(po, &po->rx_ring, 4132 TP_STATUS_KERNEL)) 4133 mask |= EPOLLIN | EPOLLRDNORM; 4134 } 4135 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 4136 po->pressure = 0; 4137 spin_unlock_bh(&sk->sk_receive_queue.lock); 4138 spin_lock_bh(&sk->sk_write_queue.lock); 4139 if (po->tx_ring.pg_vec) { 4140 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4141 mask |= EPOLLOUT | EPOLLWRNORM; 4142 } 4143 spin_unlock_bh(&sk->sk_write_queue.lock); 4144 return mask; 4145 } 4146 4147 4148 /* Dirty? Well, I still did not learn better way to account 4149 * for user mmaps. 4150 */ 4151 4152 static void packet_mm_open(struct vm_area_struct *vma) 4153 { 4154 struct file *file = vma->vm_file; 4155 struct socket *sock = file->private_data; 4156 struct sock *sk = sock->sk; 4157 4158 if (sk) 4159 atomic_inc(&pkt_sk(sk)->mapped); 4160 } 4161 4162 static void packet_mm_close(struct vm_area_struct *vma) 4163 { 4164 struct file *file = vma->vm_file; 4165 struct socket *sock = file->private_data; 4166 struct sock *sk = sock->sk; 4167 4168 if (sk) 4169 atomic_dec(&pkt_sk(sk)->mapped); 4170 } 4171 4172 static const struct vm_operations_struct packet_mmap_ops = { 4173 .open = packet_mm_open, 4174 .close = packet_mm_close, 4175 }; 4176 4177 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4178 unsigned int len) 4179 { 4180 int i; 4181 4182 for (i = 0; i < len; i++) { 4183 if (likely(pg_vec[i].buffer)) { 4184 if (is_vmalloc_addr(pg_vec[i].buffer)) 4185 vfree(pg_vec[i].buffer); 4186 else 4187 free_pages((unsigned long)pg_vec[i].buffer, 4188 order); 4189 pg_vec[i].buffer = NULL; 4190 } 4191 } 4192 kfree(pg_vec); 4193 } 4194 4195 static char *alloc_one_pg_vec_page(unsigned long order) 4196 { 4197 char *buffer; 4198 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4199 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4200 4201 buffer = (char *) __get_free_pages(gfp_flags, order); 4202 if (buffer) 4203 return buffer; 4204 4205 /* __get_free_pages failed, fall back to vmalloc */ 4206 buffer = vzalloc(array_size((1 << order), PAGE_SIZE)); 4207 if (buffer) 4208 return buffer; 4209 4210 /* vmalloc failed, lets dig into swap here */ 4211 gfp_flags &= ~__GFP_NORETRY; 4212 buffer = (char *) __get_free_pages(gfp_flags, order); 4213 if (buffer) 4214 return buffer; 4215 4216 /* complete and utter failure */ 4217 return NULL; 4218 } 4219 4220 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4221 { 4222 unsigned int block_nr = req->tp_block_nr; 4223 struct pgv *pg_vec; 4224 int i; 4225 4226 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN); 4227 if (unlikely(!pg_vec)) 4228 goto out; 4229 4230 for (i = 0; i < block_nr; i++) { 4231 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4232 if (unlikely(!pg_vec[i].buffer)) 4233 goto out_free_pgvec; 4234 } 4235 4236 out: 4237 return pg_vec; 4238 4239 out_free_pgvec: 4240 free_pg_vec(pg_vec, order, block_nr); 4241 pg_vec = NULL; 4242 goto out; 4243 } 4244 4245 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4246 int closing, int tx_ring) 4247 { 4248 struct pgv *pg_vec = NULL; 4249 struct packet_sock *po = pkt_sk(sk); 4250 int was_running, order = 0; 4251 struct packet_ring_buffer *rb; 4252 struct sk_buff_head *rb_queue; 4253 __be16 num; 4254 int err = -EINVAL; 4255 /* Added to avoid minimal code churn */ 4256 struct tpacket_req *req = &req_u->req; 4257 4258 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4259 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4260 4261 err = -EBUSY; 4262 if (!closing) { 4263 if (atomic_read(&po->mapped)) 4264 goto out; 4265 if (packet_read_pending(rb)) 4266 goto out; 4267 } 4268 4269 if (req->tp_block_nr) { 4270 unsigned int min_frame_size; 4271 4272 /* Sanity tests and some calculations */ 4273 err = -EBUSY; 4274 if (unlikely(rb->pg_vec)) 4275 goto out; 4276 4277 switch (po->tp_version) { 4278 case TPACKET_V1: 4279 po->tp_hdrlen = TPACKET_HDRLEN; 4280 break; 4281 case TPACKET_V2: 4282 po->tp_hdrlen = TPACKET2_HDRLEN; 4283 break; 4284 case TPACKET_V3: 4285 po->tp_hdrlen = TPACKET3_HDRLEN; 4286 break; 4287 } 4288 4289 err = -EINVAL; 4290 if (unlikely((int)req->tp_block_size <= 0)) 4291 goto out; 4292 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4293 goto out; 4294 min_frame_size = po->tp_hdrlen + po->tp_reserve; 4295 if (po->tp_version >= TPACKET_V3 && 4296 req->tp_block_size < 4297 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) 4298 goto out; 4299 if (unlikely(req->tp_frame_size < min_frame_size)) 4300 goto out; 4301 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4302 goto out; 4303 4304 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4305 if (unlikely(rb->frames_per_block == 0)) 4306 goto out; 4307 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) 4308 goto out; 4309 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4310 req->tp_frame_nr)) 4311 goto out; 4312 4313 err = -ENOMEM; 4314 order = get_order(req->tp_block_size); 4315 pg_vec = alloc_pg_vec(req, order); 4316 if (unlikely(!pg_vec)) 4317 goto out; 4318 switch (po->tp_version) { 4319 case TPACKET_V3: 4320 /* Block transmit is not supported yet */ 4321 if (!tx_ring) { 4322 init_prb_bdqc(po, rb, pg_vec, req_u); 4323 } else { 4324 struct tpacket_req3 *req3 = &req_u->req3; 4325 4326 if (req3->tp_retire_blk_tov || 4327 req3->tp_sizeof_priv || 4328 req3->tp_feature_req_word) { 4329 err = -EINVAL; 4330 goto out; 4331 } 4332 } 4333 break; 4334 default: 4335 break; 4336 } 4337 } 4338 /* Done */ 4339 else { 4340 err = -EINVAL; 4341 if (unlikely(req->tp_frame_nr)) 4342 goto out; 4343 } 4344 4345 4346 /* Detach socket from network */ 4347 spin_lock(&po->bind_lock); 4348 was_running = po->running; 4349 num = po->num; 4350 if (was_running) { 4351 po->num = 0; 4352 __unregister_prot_hook(sk, false); 4353 } 4354 spin_unlock(&po->bind_lock); 4355 4356 synchronize_net(); 4357 4358 err = -EBUSY; 4359 mutex_lock(&po->pg_vec_lock); 4360 if (closing || atomic_read(&po->mapped) == 0) { 4361 err = 0; 4362 spin_lock_bh(&rb_queue->lock); 4363 swap(rb->pg_vec, pg_vec); 4364 rb->frame_max = (req->tp_frame_nr - 1); 4365 rb->head = 0; 4366 rb->frame_size = req->tp_frame_size; 4367 spin_unlock_bh(&rb_queue->lock); 4368 4369 swap(rb->pg_vec_order, order); 4370 swap(rb->pg_vec_len, req->tp_block_nr); 4371 4372 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4373 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4374 tpacket_rcv : packet_rcv; 4375 skb_queue_purge(rb_queue); 4376 if (atomic_read(&po->mapped)) 4377 pr_err("packet_mmap: vma is busy: %d\n", 4378 atomic_read(&po->mapped)); 4379 } 4380 mutex_unlock(&po->pg_vec_lock); 4381 4382 spin_lock(&po->bind_lock); 4383 if (was_running) { 4384 po->num = num; 4385 register_prot_hook(sk); 4386 } 4387 spin_unlock(&po->bind_lock); 4388 if (pg_vec && (po->tp_version > TPACKET_V2)) { 4389 /* Because we don't support block-based V3 on tx-ring */ 4390 if (!tx_ring) 4391 prb_shutdown_retire_blk_timer(po, rb_queue); 4392 } 4393 4394 if (pg_vec) 4395 free_pg_vec(pg_vec, order, req->tp_block_nr); 4396 out: 4397 return err; 4398 } 4399 4400 static int packet_mmap(struct file *file, struct socket *sock, 4401 struct vm_area_struct *vma) 4402 { 4403 struct sock *sk = sock->sk; 4404 struct packet_sock *po = pkt_sk(sk); 4405 unsigned long size, expected_size; 4406 struct packet_ring_buffer *rb; 4407 unsigned long start; 4408 int err = -EINVAL; 4409 int i; 4410 4411 if (vma->vm_pgoff) 4412 return -EINVAL; 4413 4414 mutex_lock(&po->pg_vec_lock); 4415 4416 expected_size = 0; 4417 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4418 if (rb->pg_vec) { 4419 expected_size += rb->pg_vec_len 4420 * rb->pg_vec_pages 4421 * PAGE_SIZE; 4422 } 4423 } 4424 4425 if (expected_size == 0) 4426 goto out; 4427 4428 size = vma->vm_end - vma->vm_start; 4429 if (size != expected_size) 4430 goto out; 4431 4432 start = vma->vm_start; 4433 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4434 if (rb->pg_vec == NULL) 4435 continue; 4436 4437 for (i = 0; i < rb->pg_vec_len; i++) { 4438 struct page *page; 4439 void *kaddr = rb->pg_vec[i].buffer; 4440 int pg_num; 4441 4442 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4443 page = pgv_to_page(kaddr); 4444 err = vm_insert_page(vma, start, page); 4445 if (unlikely(err)) 4446 goto out; 4447 start += PAGE_SIZE; 4448 kaddr += PAGE_SIZE; 4449 } 4450 } 4451 } 4452 4453 atomic_inc(&po->mapped); 4454 vma->vm_ops = &packet_mmap_ops; 4455 err = 0; 4456 4457 out: 4458 mutex_unlock(&po->pg_vec_lock); 4459 return err; 4460 } 4461 4462 static const struct proto_ops packet_ops_spkt = { 4463 .family = PF_PACKET, 4464 .owner = THIS_MODULE, 4465 .release = packet_release, 4466 .bind = packet_bind_spkt, 4467 .connect = sock_no_connect, 4468 .socketpair = sock_no_socketpair, 4469 .accept = sock_no_accept, 4470 .getname = packet_getname_spkt, 4471 .poll = datagram_poll, 4472 .ioctl = packet_ioctl, 4473 .listen = sock_no_listen, 4474 .shutdown = sock_no_shutdown, 4475 .setsockopt = sock_no_setsockopt, 4476 .getsockopt = sock_no_getsockopt, 4477 .sendmsg = packet_sendmsg_spkt, 4478 .recvmsg = packet_recvmsg, 4479 .mmap = sock_no_mmap, 4480 .sendpage = sock_no_sendpage, 4481 }; 4482 4483 static const struct proto_ops packet_ops = { 4484 .family = PF_PACKET, 4485 .owner = THIS_MODULE, 4486 .release = packet_release, 4487 .bind = packet_bind, 4488 .connect = sock_no_connect, 4489 .socketpair = sock_no_socketpair, 4490 .accept = sock_no_accept, 4491 .getname = packet_getname, 4492 .poll = packet_poll, 4493 .ioctl = packet_ioctl, 4494 .listen = sock_no_listen, 4495 .shutdown = sock_no_shutdown, 4496 .setsockopt = packet_setsockopt, 4497 .getsockopt = packet_getsockopt, 4498 #ifdef CONFIG_COMPAT 4499 .compat_setsockopt = compat_packet_setsockopt, 4500 #endif 4501 .sendmsg = packet_sendmsg, 4502 .recvmsg = packet_recvmsg, 4503 .mmap = packet_mmap, 4504 .sendpage = sock_no_sendpage, 4505 }; 4506 4507 static const struct net_proto_family packet_family_ops = { 4508 .family = PF_PACKET, 4509 .create = packet_create, 4510 .owner = THIS_MODULE, 4511 }; 4512 4513 static struct notifier_block packet_netdev_notifier = { 4514 .notifier_call = packet_notifier, 4515 }; 4516 4517 #ifdef CONFIG_PROC_FS 4518 4519 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4520 __acquires(RCU) 4521 { 4522 struct net *net = seq_file_net(seq); 4523 4524 rcu_read_lock(); 4525 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4526 } 4527 4528 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4529 { 4530 struct net *net = seq_file_net(seq); 4531 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4532 } 4533 4534 static void packet_seq_stop(struct seq_file *seq, void *v) 4535 __releases(RCU) 4536 { 4537 rcu_read_unlock(); 4538 } 4539 4540 static int packet_seq_show(struct seq_file *seq, void *v) 4541 { 4542 if (v == SEQ_START_TOKEN) 4543 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4544 else { 4545 struct sock *s = sk_entry(v); 4546 const struct packet_sock *po = pkt_sk(s); 4547 4548 seq_printf(seq, 4549 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4550 s, 4551 refcount_read(&s->sk_refcnt), 4552 s->sk_type, 4553 ntohs(po->num), 4554 po->ifindex, 4555 po->running, 4556 atomic_read(&s->sk_rmem_alloc), 4557 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4558 sock_i_ino(s)); 4559 } 4560 4561 return 0; 4562 } 4563 4564 static const struct seq_operations packet_seq_ops = { 4565 .start = packet_seq_start, 4566 .next = packet_seq_next, 4567 .stop = packet_seq_stop, 4568 .show = packet_seq_show, 4569 }; 4570 #endif 4571 4572 static int __net_init packet_net_init(struct net *net) 4573 { 4574 mutex_init(&net->packet.sklist_lock); 4575 INIT_HLIST_HEAD(&net->packet.sklist); 4576 4577 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, 4578 sizeof(struct seq_net_private))) 4579 return -ENOMEM; 4580 4581 return 0; 4582 } 4583 4584 static void __net_exit packet_net_exit(struct net *net) 4585 { 4586 remove_proc_entry("packet", net->proc_net); 4587 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); 4588 } 4589 4590 static struct pernet_operations packet_net_ops = { 4591 .init = packet_net_init, 4592 .exit = packet_net_exit, 4593 }; 4594 4595 4596 static void __exit packet_exit(void) 4597 { 4598 unregister_netdevice_notifier(&packet_netdev_notifier); 4599 unregister_pernet_subsys(&packet_net_ops); 4600 sock_unregister(PF_PACKET); 4601 proto_unregister(&packet_proto); 4602 } 4603 4604 static int __init packet_init(void) 4605 { 4606 int rc = proto_register(&packet_proto, 0); 4607 4608 if (rc != 0) 4609 goto out; 4610 4611 sock_register(&packet_family_ops); 4612 register_pernet_subsys(&packet_net_ops); 4613 register_netdevice_notifier(&packet_netdev_notifier); 4614 out: 4615 return rc; 4616 } 4617 4618 module_init(packet_init); 4619 module_exit(packet_exit); 4620 MODULE_LICENSE("GPL"); 4621 MODULE_ALIAS_NETPROTO(PF_PACKET); 4622