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