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 <asm/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 PGV_FROM_VMALLOC 1 181 182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) 183 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) 184 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) 185 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) 186 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) 187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) 188 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x))) 189 190 struct packet_sock; 191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg); 192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 193 struct packet_type *pt, struct net_device *orig_dev); 194 195 static void *packet_previous_frame(struct packet_sock *po, 196 struct packet_ring_buffer *rb, 197 int status); 198 static void packet_increment_head(struct packet_ring_buffer *buff); 199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *, 200 struct tpacket_block_desc *); 201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 202 struct packet_sock *); 203 static void prb_retire_current_block(struct tpacket_kbdq_core *, 204 struct packet_sock *, unsigned int status); 205 static int prb_queue_frozen(struct tpacket_kbdq_core *); 206 static void prb_open_block(struct tpacket_kbdq_core *, 207 struct tpacket_block_desc *); 208 static void prb_retire_rx_blk_timer_expired(unsigned long); 209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); 210 static void prb_init_blk_timer(struct packet_sock *, 211 struct tpacket_kbdq_core *, 212 void (*func) (unsigned long)); 213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 214 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 215 struct tpacket3_hdr *); 216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 217 struct tpacket3_hdr *); 218 static void packet_flush_mclist(struct sock *sk); 219 220 struct packet_skb_cb { 221 union { 222 struct sockaddr_pkt pkt; 223 union { 224 /* Trick: alias skb original length with 225 * ll.sll_family and ll.protocol in order 226 * to save room. 227 */ 228 unsigned int origlen; 229 struct sockaddr_ll ll; 230 }; 231 } sa; 232 }; 233 234 #define vio_le() virtio_legacy_is_little_endian() 235 236 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 237 238 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 239 #define GET_PBLOCK_DESC(x, bid) \ 240 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 242 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 243 #define GET_NEXT_PRB_BLK_NUM(x) \ 244 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 245 ((x)->kactive_blk_num+1) : 0) 246 247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 248 static void __fanout_link(struct sock *sk, struct packet_sock *po); 249 250 static int packet_direct_xmit(struct sk_buff *skb) 251 { 252 struct net_device *dev = skb->dev; 253 netdev_features_t features; 254 struct netdev_queue *txq; 255 int ret = NETDEV_TX_BUSY; 256 257 if (unlikely(!netif_running(dev) || 258 !netif_carrier_ok(dev))) 259 goto drop; 260 261 features = netif_skb_features(skb); 262 if (skb_needs_linearize(skb, features) && 263 __skb_linearize(skb)) 264 goto drop; 265 266 txq = skb_get_tx_queue(dev, skb); 267 268 local_bh_disable(); 269 270 HARD_TX_LOCK(dev, txq, smp_processor_id()); 271 if (!netif_xmit_frozen_or_drv_stopped(txq)) 272 ret = netdev_start_xmit(skb, dev, txq, false); 273 HARD_TX_UNLOCK(dev, txq); 274 275 local_bh_enable(); 276 277 if (!dev_xmit_complete(ret)) 278 kfree_skb(skb); 279 280 return ret; 281 drop: 282 atomic_long_inc(&dev->tx_dropped); 283 kfree_skb(skb); 284 return NET_XMIT_DROP; 285 } 286 287 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 288 { 289 struct net_device *dev; 290 291 rcu_read_lock(); 292 dev = rcu_dereference(po->cached_dev); 293 if (likely(dev)) 294 dev_hold(dev); 295 rcu_read_unlock(); 296 297 return dev; 298 } 299 300 static void packet_cached_dev_assign(struct packet_sock *po, 301 struct net_device *dev) 302 { 303 rcu_assign_pointer(po->cached_dev, dev); 304 } 305 306 static void packet_cached_dev_reset(struct packet_sock *po) 307 { 308 RCU_INIT_POINTER(po->cached_dev, NULL); 309 } 310 311 static bool packet_use_direct_xmit(const struct packet_sock *po) 312 { 313 return po->xmit == packet_direct_xmit; 314 } 315 316 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 317 { 318 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues; 319 } 320 321 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 322 { 323 const struct net_device_ops *ops = dev->netdev_ops; 324 u16 queue_index; 325 326 if (ops->ndo_select_queue) { 327 queue_index = ops->ndo_select_queue(dev, skb, NULL, 328 __packet_pick_tx_queue); 329 queue_index = netdev_cap_txqueue(dev, queue_index); 330 } else { 331 queue_index = __packet_pick_tx_queue(dev, skb); 332 } 333 334 skb_set_queue_mapping(skb, queue_index); 335 } 336 337 /* register_prot_hook must be invoked with the po->bind_lock held, 338 * or from a context in which asynchronous accesses to the packet 339 * socket is not possible (packet_create()). 340 */ 341 static void register_prot_hook(struct sock *sk) 342 { 343 struct packet_sock *po = pkt_sk(sk); 344 345 if (!po->running) { 346 if (po->fanout) 347 __fanout_link(sk, po); 348 else 349 dev_add_pack(&po->prot_hook); 350 351 sock_hold(sk); 352 po->running = 1; 353 } 354 } 355 356 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock 357 * held. If the sync parameter is true, we will temporarily drop 358 * the po->bind_lock and do a synchronize_net to make sure no 359 * asynchronous packet processing paths still refer to the elements 360 * of po->prot_hook. If the sync parameter is false, it is the 361 * callers responsibility to take care of this. 362 */ 363 static void __unregister_prot_hook(struct sock *sk, bool sync) 364 { 365 struct packet_sock *po = pkt_sk(sk); 366 367 po->running = 0; 368 369 if (po->fanout) 370 __fanout_unlink(sk, po); 371 else 372 __dev_remove_pack(&po->prot_hook); 373 374 __sock_put(sk); 375 376 if (sync) { 377 spin_unlock(&po->bind_lock); 378 synchronize_net(); 379 spin_lock(&po->bind_lock); 380 } 381 } 382 383 static void unregister_prot_hook(struct sock *sk, bool sync) 384 { 385 struct packet_sock *po = pkt_sk(sk); 386 387 if (po->running) 388 __unregister_prot_hook(sk, sync); 389 } 390 391 static inline struct page * __pure pgv_to_page(void *addr) 392 { 393 if (is_vmalloc_addr(addr)) 394 return vmalloc_to_page(addr); 395 return virt_to_page(addr); 396 } 397 398 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 399 { 400 union tpacket_uhdr h; 401 402 h.raw = frame; 403 switch (po->tp_version) { 404 case TPACKET_V1: 405 h.h1->tp_status = status; 406 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 407 break; 408 case TPACKET_V2: 409 h.h2->tp_status = status; 410 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 411 break; 412 case TPACKET_V3: 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 default: 437 WARN(1, "TPACKET version not supported.\n"); 438 BUG(); 439 return 0; 440 } 441 } 442 443 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, 444 unsigned int flags) 445 { 446 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 447 448 if (shhwtstamps && 449 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 450 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) 451 return TP_STATUS_TS_RAW_HARDWARE; 452 453 if (ktime_to_timespec_cond(skb->tstamp, ts)) 454 return TP_STATUS_TS_SOFTWARE; 455 456 return 0; 457 } 458 459 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 460 struct sk_buff *skb) 461 { 462 union tpacket_uhdr h; 463 struct timespec ts; 464 __u32 ts_status; 465 466 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 467 return 0; 468 469 h.raw = frame; 470 switch (po->tp_version) { 471 case TPACKET_V1: 472 h.h1->tp_sec = ts.tv_sec; 473 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 474 break; 475 case TPACKET_V2: 476 h.h2->tp_sec = ts.tv_sec; 477 h.h2->tp_nsec = ts.tv_nsec; 478 break; 479 case TPACKET_V3: 480 default: 481 WARN(1, "TPACKET version not supported.\n"); 482 BUG(); 483 } 484 485 /* one flush is safe, as both fields always lie on the same cacheline */ 486 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 487 smp_wmb(); 488 489 return ts_status; 490 } 491 492 static void *packet_lookup_frame(struct packet_sock *po, 493 struct packet_ring_buffer *rb, 494 unsigned int position, 495 int status) 496 { 497 unsigned int pg_vec_pos, frame_offset; 498 union tpacket_uhdr h; 499 500 pg_vec_pos = position / rb->frames_per_block; 501 frame_offset = position % rb->frames_per_block; 502 503 h.raw = rb->pg_vec[pg_vec_pos].buffer + 504 (frame_offset * rb->frame_size); 505 506 if (status != __packet_get_status(po, h.raw)) 507 return NULL; 508 509 return h.raw; 510 } 511 512 static void *packet_current_frame(struct packet_sock *po, 513 struct packet_ring_buffer *rb, 514 int status) 515 { 516 return packet_lookup_frame(po, rb, rb->head, status); 517 } 518 519 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 520 { 521 del_timer_sync(&pkc->retire_blk_timer); 522 } 523 524 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 525 struct sk_buff_head *rb_queue) 526 { 527 struct tpacket_kbdq_core *pkc; 528 529 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 530 531 spin_lock_bh(&rb_queue->lock); 532 pkc->delete_blk_timer = 1; 533 spin_unlock_bh(&rb_queue->lock); 534 535 prb_del_retire_blk_timer(pkc); 536 } 537 538 static void prb_init_blk_timer(struct packet_sock *po, 539 struct tpacket_kbdq_core *pkc, 540 void (*func) (unsigned long)) 541 { 542 init_timer(&pkc->retire_blk_timer); 543 pkc->retire_blk_timer.data = (long)po; 544 pkc->retire_blk_timer.function = func; 545 pkc->retire_blk_timer.expires = jiffies; 546 } 547 548 static void prb_setup_retire_blk_timer(struct packet_sock *po) 549 { 550 struct tpacket_kbdq_core *pkc; 551 552 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 553 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired); 554 } 555 556 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 557 int blk_size_in_bytes) 558 { 559 struct net_device *dev; 560 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; 561 struct ethtool_link_ksettings ecmd; 562 int err; 563 564 rtnl_lock(); 565 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 566 if (unlikely(!dev)) { 567 rtnl_unlock(); 568 return DEFAULT_PRB_RETIRE_TOV; 569 } 570 err = __ethtool_get_link_ksettings(dev, &ecmd); 571 rtnl_unlock(); 572 if (!err) { 573 /* 574 * If the link speed is so slow you don't really 575 * need to worry about perf anyways 576 */ 577 if (ecmd.base.speed < SPEED_1000 || 578 ecmd.base.speed == SPEED_UNKNOWN) { 579 return DEFAULT_PRB_RETIRE_TOV; 580 } else { 581 msec = 1; 582 div = ecmd.base.speed / 1000; 583 } 584 } 585 586 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 587 588 if (div) 589 mbits /= div; 590 591 tmo = mbits * msec; 592 593 if (div) 594 return tmo+1; 595 return tmo; 596 } 597 598 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 599 union tpacket_req_u *req_u) 600 { 601 p1->feature_req_word = req_u->req3.tp_feature_req_word; 602 } 603 604 static void init_prb_bdqc(struct packet_sock *po, 605 struct packet_ring_buffer *rb, 606 struct pgv *pg_vec, 607 union tpacket_req_u *req_u) 608 { 609 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 610 struct tpacket_block_desc *pbd; 611 612 memset(p1, 0x0, sizeof(*p1)); 613 614 p1->knxt_seq_num = 1; 615 p1->pkbdq = pg_vec; 616 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 617 p1->pkblk_start = pg_vec[0].buffer; 618 p1->kblk_size = req_u->req3.tp_block_size; 619 p1->knum_blocks = req_u->req3.tp_block_nr; 620 p1->hdrlen = po->tp_hdrlen; 621 p1->version = po->tp_version; 622 p1->last_kactive_blk_num = 0; 623 po->stats.stats3.tp_freeze_q_cnt = 0; 624 if (req_u->req3.tp_retire_blk_tov) 625 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 626 else 627 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 628 req_u->req3.tp_block_size); 629 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 630 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 631 632 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 633 prb_init_ft_ops(p1, req_u); 634 prb_setup_retire_blk_timer(po); 635 prb_open_block(p1, pbd); 636 } 637 638 /* Do NOT update the last_blk_num first. 639 * Assumes sk_buff_head lock is held. 640 */ 641 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 642 { 643 mod_timer(&pkc->retire_blk_timer, 644 jiffies + pkc->tov_in_jiffies); 645 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 646 } 647 648 /* 649 * Timer logic: 650 * 1) We refresh the timer only when we open a block. 651 * By doing this we don't waste cycles refreshing the timer 652 * on packet-by-packet basis. 653 * 654 * With a 1MB block-size, on a 1Gbps line, it will take 655 * i) ~8 ms to fill a block + ii) memcpy etc. 656 * In this cut we are not accounting for the memcpy time. 657 * 658 * So, if the user sets the 'tmo' to 10ms then the timer 659 * will never fire while the block is still getting filled 660 * (which is what we want). However, the user could choose 661 * to close a block early and that's fine. 662 * 663 * But when the timer does fire, we check whether or not to refresh it. 664 * Since the tmo granularity is in msecs, it is not too expensive 665 * to refresh the timer, lets say every '8' msecs. 666 * Either the user can set the 'tmo' or we can derive it based on 667 * a) line-speed and b) block-size. 668 * prb_calc_retire_blk_tmo() calculates the tmo. 669 * 670 */ 671 static void prb_retire_rx_blk_timer_expired(unsigned long data) 672 { 673 struct packet_sock *po = (struct packet_sock *)data; 674 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 675 unsigned int frozen; 676 struct tpacket_block_desc *pbd; 677 678 spin_lock(&po->sk.sk_receive_queue.lock); 679 680 frozen = prb_queue_frozen(pkc); 681 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 682 683 if (unlikely(pkc->delete_blk_timer)) 684 goto out; 685 686 /* We only need to plug the race when the block is partially filled. 687 * tpacket_rcv: 688 * lock(); increment BLOCK_NUM_PKTS; unlock() 689 * copy_bits() is in progress ... 690 * timer fires on other cpu: 691 * we can't retire the current block because copy_bits 692 * is in progress. 693 * 694 */ 695 if (BLOCK_NUM_PKTS(pbd)) { 696 while (atomic_read(&pkc->blk_fill_in_prog)) { 697 /* Waiting for skb_copy_bits to finish... */ 698 cpu_relax(); 699 } 700 } 701 702 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 703 if (!frozen) { 704 if (!BLOCK_NUM_PKTS(pbd)) { 705 /* An empty block. Just refresh the timer. */ 706 goto refresh_timer; 707 } 708 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 709 if (!prb_dispatch_next_block(pkc, po)) 710 goto refresh_timer; 711 else 712 goto out; 713 } else { 714 /* Case 1. Queue was frozen because user-space was 715 * lagging behind. 716 */ 717 if (prb_curr_blk_in_use(pkc, pbd)) { 718 /* 719 * Ok, user-space is still behind. 720 * So just refresh the timer. 721 */ 722 goto refresh_timer; 723 } else { 724 /* Case 2. queue was frozen,user-space caught up, 725 * now the link went idle && the timer fired. 726 * We don't have a block to close.So we open this 727 * block and restart the timer. 728 * opening a block thaws the queue,restarts timer 729 * Thawing/timer-refresh is a side effect. 730 */ 731 prb_open_block(pkc, pbd); 732 goto out; 733 } 734 } 735 } 736 737 refresh_timer: 738 _prb_refresh_rx_retire_blk_timer(pkc); 739 740 out: 741 spin_unlock(&po->sk.sk_receive_queue.lock); 742 } 743 744 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 745 struct tpacket_block_desc *pbd1, __u32 status) 746 { 747 /* Flush everything minus the block header */ 748 749 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 750 u8 *start, *end; 751 752 start = (u8 *)pbd1; 753 754 /* Skip the block header(we know header WILL fit in 4K) */ 755 start += PAGE_SIZE; 756 757 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 758 for (; start < end; start += PAGE_SIZE) 759 flush_dcache_page(pgv_to_page(start)); 760 761 smp_wmb(); 762 #endif 763 764 /* Now update the block status. */ 765 766 BLOCK_STATUS(pbd1) = status; 767 768 /* Flush the block header */ 769 770 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 771 start = (u8 *)pbd1; 772 flush_dcache_page(pgv_to_page(start)); 773 774 smp_wmb(); 775 #endif 776 } 777 778 /* 779 * Side effect: 780 * 781 * 1) flush the block 782 * 2) Increment active_blk_num 783 * 784 * Note:We DONT refresh the timer on purpose. 785 * Because almost always the next block will be opened. 786 */ 787 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 788 struct tpacket_block_desc *pbd1, 789 struct packet_sock *po, unsigned int stat) 790 { 791 __u32 status = TP_STATUS_USER | stat; 792 793 struct tpacket3_hdr *last_pkt; 794 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 795 struct sock *sk = &po->sk; 796 797 if (po->stats.stats3.tp_drops) 798 status |= TP_STATUS_LOSING; 799 800 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 801 last_pkt->tp_next_offset = 0; 802 803 /* Get the ts of the last pkt */ 804 if (BLOCK_NUM_PKTS(pbd1)) { 805 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 806 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 807 } else { 808 /* Ok, we tmo'd - so get the current time. 809 * 810 * It shouldn't really happen as we don't close empty 811 * blocks. See prb_retire_rx_blk_timer_expired(). 812 */ 813 struct timespec ts; 814 getnstimeofday(&ts); 815 h1->ts_last_pkt.ts_sec = ts.tv_sec; 816 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 817 } 818 819 smp_wmb(); 820 821 /* Flush the block */ 822 prb_flush_block(pkc1, pbd1, status); 823 824 sk->sk_data_ready(sk); 825 826 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 827 } 828 829 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 830 { 831 pkc->reset_pending_on_curr_blk = 0; 832 } 833 834 /* 835 * Side effect of opening a block: 836 * 837 * 1) prb_queue is thawed. 838 * 2) retire_blk_timer is refreshed. 839 * 840 */ 841 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 842 struct tpacket_block_desc *pbd1) 843 { 844 struct timespec ts; 845 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 846 847 smp_rmb(); 848 849 /* We could have just memset this but we will lose the 850 * flexibility of making the priv area sticky 851 */ 852 853 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 854 BLOCK_NUM_PKTS(pbd1) = 0; 855 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 856 857 getnstimeofday(&ts); 858 859 h1->ts_first_pkt.ts_sec = ts.tv_sec; 860 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 861 862 pkc1->pkblk_start = (char *)pbd1; 863 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 864 865 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 866 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 867 868 pbd1->version = pkc1->version; 869 pkc1->prev = pkc1->nxt_offset; 870 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 871 872 prb_thaw_queue(pkc1); 873 _prb_refresh_rx_retire_blk_timer(pkc1); 874 875 smp_wmb(); 876 } 877 878 /* 879 * Queue freeze logic: 880 * 1) Assume tp_block_nr = 8 blocks. 881 * 2) At time 't0', user opens Rx ring. 882 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 883 * 4) user-space is either sleeping or processing block '0'. 884 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 885 * it will close block-7,loop around and try to fill block '0'. 886 * call-flow: 887 * __packet_lookup_frame_in_block 888 * prb_retire_current_block() 889 * prb_dispatch_next_block() 890 * |->(BLOCK_STATUS == USER) evaluates to true 891 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 892 * 6) Now there are two cases: 893 * 6.1) Link goes idle right after the queue is frozen. 894 * But remember, the last open_block() refreshed the timer. 895 * When this timer expires,it will refresh itself so that we can 896 * re-open block-0 in near future. 897 * 6.2) Link is busy and keeps on receiving packets. This is a simple 898 * case and __packet_lookup_frame_in_block will check if block-0 899 * is free and can now be re-used. 900 */ 901 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 902 struct packet_sock *po) 903 { 904 pkc->reset_pending_on_curr_blk = 1; 905 po->stats.stats3.tp_freeze_q_cnt++; 906 } 907 908 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 909 910 /* 911 * If the next block is free then we will dispatch it 912 * and return a good offset. 913 * Else, we will freeze the queue. 914 * So, caller must check the return value. 915 */ 916 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 917 struct packet_sock *po) 918 { 919 struct tpacket_block_desc *pbd; 920 921 smp_rmb(); 922 923 /* 1. Get current block num */ 924 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 925 926 /* 2. If this block is currently in_use then freeze the queue */ 927 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 928 prb_freeze_queue(pkc, po); 929 return NULL; 930 } 931 932 /* 933 * 3. 934 * open this block and return the offset where the first packet 935 * needs to get stored. 936 */ 937 prb_open_block(pkc, pbd); 938 return (void *)pkc->nxt_offset; 939 } 940 941 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 942 struct packet_sock *po, unsigned int status) 943 { 944 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 945 946 /* retire/close the current block */ 947 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 948 /* 949 * Plug the case where copy_bits() is in progress on 950 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 951 * have space to copy the pkt in the current block and 952 * called prb_retire_current_block() 953 * 954 * We don't need to worry about the TMO case because 955 * the timer-handler already handled this case. 956 */ 957 if (!(status & TP_STATUS_BLK_TMO)) { 958 while (atomic_read(&pkc->blk_fill_in_prog)) { 959 /* Waiting for skb_copy_bits to finish... */ 960 cpu_relax(); 961 } 962 } 963 prb_close_block(pkc, pbd, po, status); 964 return; 965 } 966 } 967 968 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc, 969 struct tpacket_block_desc *pbd) 970 { 971 return TP_STATUS_USER & BLOCK_STATUS(pbd); 972 } 973 974 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 975 { 976 return pkc->reset_pending_on_curr_blk; 977 } 978 979 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 980 { 981 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 982 atomic_dec(&pkc->blk_fill_in_prog); 983 } 984 985 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 986 struct tpacket3_hdr *ppd) 987 { 988 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 989 } 990 991 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 992 struct tpacket3_hdr *ppd) 993 { 994 ppd->hv1.tp_rxhash = 0; 995 } 996 997 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 998 struct tpacket3_hdr *ppd) 999 { 1000 if (skb_vlan_tag_present(pkc->skb)) { 1001 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 1002 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 1003 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 1004 } else { 1005 ppd->hv1.tp_vlan_tci = 0; 1006 ppd->hv1.tp_vlan_tpid = 0; 1007 ppd->tp_status = TP_STATUS_AVAILABLE; 1008 } 1009 } 1010 1011 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 1012 struct tpacket3_hdr *ppd) 1013 { 1014 ppd->hv1.tp_padding = 0; 1015 prb_fill_vlan_info(pkc, ppd); 1016 1017 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 1018 prb_fill_rxhash(pkc, ppd); 1019 else 1020 prb_clear_rxhash(pkc, ppd); 1021 } 1022 1023 static void prb_fill_curr_block(char *curr, 1024 struct tpacket_kbdq_core *pkc, 1025 struct tpacket_block_desc *pbd, 1026 unsigned int len) 1027 { 1028 struct tpacket3_hdr *ppd; 1029 1030 ppd = (struct tpacket3_hdr *)curr; 1031 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 1032 pkc->prev = curr; 1033 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 1034 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1035 BLOCK_NUM_PKTS(pbd) += 1; 1036 atomic_inc(&pkc->blk_fill_in_prog); 1037 prb_run_all_ft_ops(pkc, ppd); 1038 } 1039 1040 /* Assumes caller has the sk->rx_queue.lock */ 1041 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1042 struct sk_buff *skb, 1043 int status, 1044 unsigned int len 1045 ) 1046 { 1047 struct tpacket_kbdq_core *pkc; 1048 struct tpacket_block_desc *pbd; 1049 char *curr, *end; 1050 1051 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1052 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1053 1054 /* Queue is frozen when user space is lagging behind */ 1055 if (prb_queue_frozen(pkc)) { 1056 /* 1057 * Check if that last block which caused the queue to freeze, 1058 * is still in_use by user-space. 1059 */ 1060 if (prb_curr_blk_in_use(pkc, pbd)) { 1061 /* Can't record this packet */ 1062 return NULL; 1063 } else { 1064 /* 1065 * Ok, the block was released by user-space. 1066 * Now let's open that block. 1067 * opening a block also thaws the queue. 1068 * Thawing is a side effect. 1069 */ 1070 prb_open_block(pkc, pbd); 1071 } 1072 } 1073 1074 smp_mb(); 1075 curr = pkc->nxt_offset; 1076 pkc->skb = skb; 1077 end = (char *)pbd + pkc->kblk_size; 1078 1079 /* first try the current block */ 1080 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1081 prb_fill_curr_block(curr, pkc, pbd, len); 1082 return (void *)curr; 1083 } 1084 1085 /* Ok, close the current block */ 1086 prb_retire_current_block(pkc, po, 0); 1087 1088 /* Now, try to dispatch the next block */ 1089 curr = (char *)prb_dispatch_next_block(pkc, po); 1090 if (curr) { 1091 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1092 prb_fill_curr_block(curr, pkc, pbd, len); 1093 return (void *)curr; 1094 } 1095 1096 /* 1097 * No free blocks are available.user_space hasn't caught up yet. 1098 * Queue was just frozen and now this packet will get dropped. 1099 */ 1100 return NULL; 1101 } 1102 1103 static void *packet_current_rx_frame(struct packet_sock *po, 1104 struct sk_buff *skb, 1105 int status, unsigned int len) 1106 { 1107 char *curr = NULL; 1108 switch (po->tp_version) { 1109 case TPACKET_V1: 1110 case TPACKET_V2: 1111 curr = packet_lookup_frame(po, &po->rx_ring, 1112 po->rx_ring.head, status); 1113 return curr; 1114 case TPACKET_V3: 1115 return __packet_lookup_frame_in_block(po, skb, status, len); 1116 default: 1117 WARN(1, "TPACKET version not supported\n"); 1118 BUG(); 1119 return NULL; 1120 } 1121 } 1122 1123 static void *prb_lookup_block(struct packet_sock *po, 1124 struct packet_ring_buffer *rb, 1125 unsigned int idx, 1126 int status) 1127 { 1128 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1129 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1130 1131 if (status != BLOCK_STATUS(pbd)) 1132 return NULL; 1133 return pbd; 1134 } 1135 1136 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1137 { 1138 unsigned int prev; 1139 if (rb->prb_bdqc.kactive_blk_num) 1140 prev = rb->prb_bdqc.kactive_blk_num-1; 1141 else 1142 prev = rb->prb_bdqc.knum_blocks-1; 1143 return prev; 1144 } 1145 1146 /* Assumes caller has held the rx_queue.lock */ 1147 static void *__prb_previous_block(struct packet_sock *po, 1148 struct packet_ring_buffer *rb, 1149 int status) 1150 { 1151 unsigned int previous = prb_previous_blk_num(rb); 1152 return prb_lookup_block(po, rb, previous, status); 1153 } 1154 1155 static void *packet_previous_rx_frame(struct packet_sock *po, 1156 struct packet_ring_buffer *rb, 1157 int status) 1158 { 1159 if (po->tp_version <= TPACKET_V2) 1160 return packet_previous_frame(po, rb, status); 1161 1162 return __prb_previous_block(po, rb, status); 1163 } 1164 1165 static void packet_increment_rx_head(struct packet_sock *po, 1166 struct packet_ring_buffer *rb) 1167 { 1168 switch (po->tp_version) { 1169 case TPACKET_V1: 1170 case TPACKET_V2: 1171 return packet_increment_head(rb); 1172 case TPACKET_V3: 1173 default: 1174 WARN(1, "TPACKET version not supported.\n"); 1175 BUG(); 1176 return; 1177 } 1178 } 1179 1180 static void *packet_previous_frame(struct packet_sock *po, 1181 struct packet_ring_buffer *rb, 1182 int status) 1183 { 1184 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1185 return packet_lookup_frame(po, rb, previous, status); 1186 } 1187 1188 static void packet_increment_head(struct packet_ring_buffer *buff) 1189 { 1190 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1191 } 1192 1193 static void packet_inc_pending(struct packet_ring_buffer *rb) 1194 { 1195 this_cpu_inc(*rb->pending_refcnt); 1196 } 1197 1198 static void packet_dec_pending(struct packet_ring_buffer *rb) 1199 { 1200 this_cpu_dec(*rb->pending_refcnt); 1201 } 1202 1203 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1204 { 1205 unsigned int refcnt = 0; 1206 int cpu; 1207 1208 /* We don't use pending refcount in rx_ring. */ 1209 if (rb->pending_refcnt == NULL) 1210 return 0; 1211 1212 for_each_possible_cpu(cpu) 1213 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1214 1215 return refcnt; 1216 } 1217 1218 static int packet_alloc_pending(struct packet_sock *po) 1219 { 1220 po->rx_ring.pending_refcnt = NULL; 1221 1222 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1223 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1224 return -ENOBUFS; 1225 1226 return 0; 1227 } 1228 1229 static void packet_free_pending(struct packet_sock *po) 1230 { 1231 free_percpu(po->tx_ring.pending_refcnt); 1232 } 1233 1234 #define ROOM_POW_OFF 2 1235 #define ROOM_NONE 0x0 1236 #define ROOM_LOW 0x1 1237 #define ROOM_NORMAL 0x2 1238 1239 static bool __tpacket_has_room(struct packet_sock *po, int pow_off) 1240 { 1241 int idx, len; 1242 1243 len = po->rx_ring.frame_max + 1; 1244 idx = po->rx_ring.head; 1245 if (pow_off) 1246 idx += len >> pow_off; 1247 if (idx >= len) 1248 idx -= len; 1249 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1250 } 1251 1252 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off) 1253 { 1254 int idx, len; 1255 1256 len = po->rx_ring.prb_bdqc.knum_blocks; 1257 idx = po->rx_ring.prb_bdqc.kactive_blk_num; 1258 if (pow_off) 1259 idx += len >> pow_off; 1260 if (idx >= len) 1261 idx -= len; 1262 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1263 } 1264 1265 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1266 { 1267 struct sock *sk = &po->sk; 1268 int ret = ROOM_NONE; 1269 1270 if (po->prot_hook.func != tpacket_rcv) { 1271 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1272 - (skb ? skb->truesize : 0); 1273 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF)) 1274 return ROOM_NORMAL; 1275 else if (avail > 0) 1276 return ROOM_LOW; 1277 else 1278 return ROOM_NONE; 1279 } 1280 1281 if (po->tp_version == TPACKET_V3) { 1282 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1283 ret = ROOM_NORMAL; 1284 else if (__tpacket_v3_has_room(po, 0)) 1285 ret = ROOM_LOW; 1286 } else { 1287 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1288 ret = ROOM_NORMAL; 1289 else if (__tpacket_has_room(po, 0)) 1290 ret = ROOM_LOW; 1291 } 1292 1293 return ret; 1294 } 1295 1296 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1297 { 1298 int ret; 1299 bool has_room; 1300 1301 spin_lock_bh(&po->sk.sk_receive_queue.lock); 1302 ret = __packet_rcv_has_room(po, skb); 1303 has_room = ret == ROOM_NORMAL; 1304 if (po->pressure == has_room) 1305 po->pressure = !has_room; 1306 spin_unlock_bh(&po->sk.sk_receive_queue.lock); 1307 1308 return ret; 1309 } 1310 1311 static void packet_sock_destruct(struct sock *sk) 1312 { 1313 skb_queue_purge(&sk->sk_error_queue); 1314 1315 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1316 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 1317 1318 if (!sock_flag(sk, SOCK_DEAD)) { 1319 pr_err("Attempt to release alive packet socket: %p\n", sk); 1320 return; 1321 } 1322 1323 sk_refcnt_debug_dec(sk); 1324 } 1325 1326 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1327 { 1328 u32 rxhash; 1329 int i, count = 0; 1330 1331 rxhash = skb_get_hash(skb); 1332 for (i = 0; i < ROLLOVER_HLEN; i++) 1333 if (po->rollover->history[i] == rxhash) 1334 count++; 1335 1336 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash; 1337 return count > (ROLLOVER_HLEN >> 1); 1338 } 1339 1340 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1341 struct sk_buff *skb, 1342 unsigned int num) 1343 { 1344 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1345 } 1346 1347 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1348 struct sk_buff *skb, 1349 unsigned int num) 1350 { 1351 unsigned int val = atomic_inc_return(&f->rr_cur); 1352 1353 return val % num; 1354 } 1355 1356 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1357 struct sk_buff *skb, 1358 unsigned int num) 1359 { 1360 return smp_processor_id() % num; 1361 } 1362 1363 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1364 struct sk_buff *skb, 1365 unsigned int num) 1366 { 1367 return prandom_u32_max(num); 1368 } 1369 1370 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1371 struct sk_buff *skb, 1372 unsigned int idx, bool try_self, 1373 unsigned int num) 1374 { 1375 struct packet_sock *po, *po_next, *po_skip = NULL; 1376 unsigned int i, j, room = ROOM_NONE; 1377 1378 po = pkt_sk(f->arr[idx]); 1379 1380 if (try_self) { 1381 room = packet_rcv_has_room(po, skb); 1382 if (room == ROOM_NORMAL || 1383 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1384 return idx; 1385 po_skip = po; 1386 } 1387 1388 i = j = min_t(int, po->rollover->sock, num - 1); 1389 do { 1390 po_next = pkt_sk(f->arr[i]); 1391 if (po_next != po_skip && !po_next->pressure && 1392 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1393 if (i != j) 1394 po->rollover->sock = i; 1395 atomic_long_inc(&po->rollover->num); 1396 if (room == ROOM_LOW) 1397 atomic_long_inc(&po->rollover->num_huge); 1398 return i; 1399 } 1400 1401 if (++i == num) 1402 i = 0; 1403 } while (i != j); 1404 1405 atomic_long_inc(&po->rollover->num_failed); 1406 return idx; 1407 } 1408 1409 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1410 struct sk_buff *skb, 1411 unsigned int num) 1412 { 1413 return skb_get_queue_mapping(skb) % num; 1414 } 1415 1416 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1417 struct sk_buff *skb, 1418 unsigned int num) 1419 { 1420 struct bpf_prog *prog; 1421 unsigned int ret = 0; 1422 1423 rcu_read_lock(); 1424 prog = rcu_dereference(f->bpf_prog); 1425 if (prog) 1426 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1427 rcu_read_unlock(); 1428 1429 return ret; 1430 } 1431 1432 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1433 { 1434 return f->flags & (flag >> 8); 1435 } 1436 1437 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1438 struct packet_type *pt, struct net_device *orig_dev) 1439 { 1440 struct packet_fanout *f = pt->af_packet_priv; 1441 unsigned int num = READ_ONCE(f->num_members); 1442 struct net *net = read_pnet(&f->net); 1443 struct packet_sock *po; 1444 unsigned int idx; 1445 1446 if (!net_eq(dev_net(dev), net) || !num) { 1447 kfree_skb(skb); 1448 return 0; 1449 } 1450 1451 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1452 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1453 if (!skb) 1454 return 0; 1455 } 1456 switch (f->type) { 1457 case PACKET_FANOUT_HASH: 1458 default: 1459 idx = fanout_demux_hash(f, skb, num); 1460 break; 1461 case PACKET_FANOUT_LB: 1462 idx = fanout_demux_lb(f, skb, num); 1463 break; 1464 case PACKET_FANOUT_CPU: 1465 idx = fanout_demux_cpu(f, skb, num); 1466 break; 1467 case PACKET_FANOUT_RND: 1468 idx = fanout_demux_rnd(f, skb, num); 1469 break; 1470 case PACKET_FANOUT_QM: 1471 idx = fanout_demux_qm(f, skb, num); 1472 break; 1473 case PACKET_FANOUT_ROLLOVER: 1474 idx = fanout_demux_rollover(f, skb, 0, false, num); 1475 break; 1476 case PACKET_FANOUT_CBPF: 1477 case PACKET_FANOUT_EBPF: 1478 idx = fanout_demux_bpf(f, skb, num); 1479 break; 1480 } 1481 1482 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1483 idx = fanout_demux_rollover(f, skb, idx, true, num); 1484 1485 po = pkt_sk(f->arr[idx]); 1486 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1487 } 1488 1489 DEFINE_MUTEX(fanout_mutex); 1490 EXPORT_SYMBOL_GPL(fanout_mutex); 1491 static LIST_HEAD(fanout_list); 1492 1493 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1494 { 1495 struct packet_fanout *f = po->fanout; 1496 1497 spin_lock(&f->lock); 1498 f->arr[f->num_members] = sk; 1499 smp_wmb(); 1500 f->num_members++; 1501 spin_unlock(&f->lock); 1502 } 1503 1504 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1505 { 1506 struct packet_fanout *f = po->fanout; 1507 int i; 1508 1509 spin_lock(&f->lock); 1510 for (i = 0; i < f->num_members; i++) { 1511 if (f->arr[i] == sk) 1512 break; 1513 } 1514 BUG_ON(i >= f->num_members); 1515 f->arr[i] = f->arr[f->num_members - 1]; 1516 f->num_members--; 1517 spin_unlock(&f->lock); 1518 } 1519 1520 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1521 { 1522 if (sk->sk_family != PF_PACKET) 1523 return false; 1524 1525 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1526 } 1527 1528 static void fanout_init_data(struct packet_fanout *f) 1529 { 1530 switch (f->type) { 1531 case PACKET_FANOUT_LB: 1532 atomic_set(&f->rr_cur, 0); 1533 break; 1534 case PACKET_FANOUT_CBPF: 1535 case PACKET_FANOUT_EBPF: 1536 RCU_INIT_POINTER(f->bpf_prog, NULL); 1537 break; 1538 } 1539 } 1540 1541 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1542 { 1543 struct bpf_prog *old; 1544 1545 spin_lock(&f->lock); 1546 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1547 rcu_assign_pointer(f->bpf_prog, new); 1548 spin_unlock(&f->lock); 1549 1550 if (old) { 1551 synchronize_net(); 1552 bpf_prog_destroy(old); 1553 } 1554 } 1555 1556 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data, 1557 unsigned int len) 1558 { 1559 struct bpf_prog *new; 1560 struct sock_fprog fprog; 1561 int ret; 1562 1563 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1564 return -EPERM; 1565 if (len != sizeof(fprog)) 1566 return -EINVAL; 1567 if (copy_from_user(&fprog, data, len)) 1568 return -EFAULT; 1569 1570 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1571 if (ret) 1572 return ret; 1573 1574 __fanout_set_data_bpf(po->fanout, new); 1575 return 0; 1576 } 1577 1578 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data, 1579 unsigned int len) 1580 { 1581 struct bpf_prog *new; 1582 u32 fd; 1583 1584 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1585 return -EPERM; 1586 if (len != sizeof(fd)) 1587 return -EINVAL; 1588 if (copy_from_user(&fd, data, len)) 1589 return -EFAULT; 1590 1591 new = bpf_prog_get(fd); 1592 if (IS_ERR(new)) 1593 return PTR_ERR(new); 1594 if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) { 1595 bpf_prog_put(new); 1596 return -EINVAL; 1597 } 1598 1599 __fanout_set_data_bpf(po->fanout, new); 1600 return 0; 1601 } 1602 1603 static int fanout_set_data(struct packet_sock *po, char __user *data, 1604 unsigned int len) 1605 { 1606 switch (po->fanout->type) { 1607 case PACKET_FANOUT_CBPF: 1608 return fanout_set_data_cbpf(po, data, len); 1609 case PACKET_FANOUT_EBPF: 1610 return fanout_set_data_ebpf(po, data, len); 1611 default: 1612 return -EINVAL; 1613 }; 1614 } 1615 1616 static void fanout_release_data(struct packet_fanout *f) 1617 { 1618 switch (f->type) { 1619 case PACKET_FANOUT_CBPF: 1620 case PACKET_FANOUT_EBPF: 1621 __fanout_set_data_bpf(f, NULL); 1622 }; 1623 } 1624 1625 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1626 { 1627 struct packet_sock *po = pkt_sk(sk); 1628 struct packet_fanout *f, *match; 1629 u8 type = type_flags & 0xff; 1630 u8 flags = type_flags >> 8; 1631 int err; 1632 1633 switch (type) { 1634 case PACKET_FANOUT_ROLLOVER: 1635 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1636 return -EINVAL; 1637 case PACKET_FANOUT_HASH: 1638 case PACKET_FANOUT_LB: 1639 case PACKET_FANOUT_CPU: 1640 case PACKET_FANOUT_RND: 1641 case PACKET_FANOUT_QM: 1642 case PACKET_FANOUT_CBPF: 1643 case PACKET_FANOUT_EBPF: 1644 break; 1645 default: 1646 return -EINVAL; 1647 } 1648 1649 if (!po->running) 1650 return -EINVAL; 1651 1652 if (po->fanout) 1653 return -EALREADY; 1654 1655 if (type == PACKET_FANOUT_ROLLOVER || 1656 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1657 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL); 1658 if (!po->rollover) 1659 return -ENOMEM; 1660 atomic_long_set(&po->rollover->num, 0); 1661 atomic_long_set(&po->rollover->num_huge, 0); 1662 atomic_long_set(&po->rollover->num_failed, 0); 1663 } 1664 1665 mutex_lock(&fanout_mutex); 1666 match = NULL; 1667 list_for_each_entry(f, &fanout_list, list) { 1668 if (f->id == id && 1669 read_pnet(&f->net) == sock_net(sk)) { 1670 match = f; 1671 break; 1672 } 1673 } 1674 err = -EINVAL; 1675 if (match && match->flags != flags) 1676 goto out; 1677 if (!match) { 1678 err = -ENOMEM; 1679 match = kzalloc(sizeof(*match), GFP_KERNEL); 1680 if (!match) 1681 goto out; 1682 write_pnet(&match->net, sock_net(sk)); 1683 match->id = id; 1684 match->type = type; 1685 match->flags = flags; 1686 INIT_LIST_HEAD(&match->list); 1687 spin_lock_init(&match->lock); 1688 atomic_set(&match->sk_ref, 0); 1689 fanout_init_data(match); 1690 match->prot_hook.type = po->prot_hook.type; 1691 match->prot_hook.dev = po->prot_hook.dev; 1692 match->prot_hook.func = packet_rcv_fanout; 1693 match->prot_hook.af_packet_priv = match; 1694 match->prot_hook.id_match = match_fanout_group; 1695 dev_add_pack(&match->prot_hook); 1696 list_add(&match->list, &fanout_list); 1697 } 1698 err = -EINVAL; 1699 if (match->type == type && 1700 match->prot_hook.type == po->prot_hook.type && 1701 match->prot_hook.dev == po->prot_hook.dev) { 1702 err = -ENOSPC; 1703 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1704 __dev_remove_pack(&po->prot_hook); 1705 po->fanout = match; 1706 atomic_inc(&match->sk_ref); 1707 __fanout_link(sk, po); 1708 err = 0; 1709 } 1710 } 1711 out: 1712 mutex_unlock(&fanout_mutex); 1713 if (err) { 1714 kfree(po->rollover); 1715 po->rollover = NULL; 1716 } 1717 return err; 1718 } 1719 1720 static void fanout_release(struct sock *sk) 1721 { 1722 struct packet_sock *po = pkt_sk(sk); 1723 struct packet_fanout *f; 1724 1725 f = po->fanout; 1726 if (!f) 1727 return; 1728 1729 mutex_lock(&fanout_mutex); 1730 po->fanout = NULL; 1731 1732 if (atomic_dec_and_test(&f->sk_ref)) { 1733 list_del(&f->list); 1734 dev_remove_pack(&f->prot_hook); 1735 fanout_release_data(f); 1736 kfree(f); 1737 } 1738 mutex_unlock(&fanout_mutex); 1739 1740 if (po->rollover) 1741 kfree_rcu(po->rollover, rcu); 1742 } 1743 1744 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1745 struct sk_buff *skb) 1746 { 1747 /* Earlier code assumed this would be a VLAN pkt, double-check 1748 * this now that we have the actual packet in hand. We can only 1749 * do this check on Ethernet devices. 1750 */ 1751 if (unlikely(dev->type != ARPHRD_ETHER)) 1752 return false; 1753 1754 skb_reset_mac_header(skb); 1755 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1756 } 1757 1758 static const struct proto_ops packet_ops; 1759 1760 static const struct proto_ops packet_ops_spkt; 1761 1762 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1763 struct packet_type *pt, struct net_device *orig_dev) 1764 { 1765 struct sock *sk; 1766 struct sockaddr_pkt *spkt; 1767 1768 /* 1769 * When we registered the protocol we saved the socket in the data 1770 * field for just this event. 1771 */ 1772 1773 sk = pt->af_packet_priv; 1774 1775 /* 1776 * Yank back the headers [hope the device set this 1777 * right or kerboom...] 1778 * 1779 * Incoming packets have ll header pulled, 1780 * push it back. 1781 * 1782 * For outgoing ones skb->data == skb_mac_header(skb) 1783 * so that this procedure is noop. 1784 */ 1785 1786 if (skb->pkt_type == PACKET_LOOPBACK) 1787 goto out; 1788 1789 if (!net_eq(dev_net(dev), sock_net(sk))) 1790 goto out; 1791 1792 skb = skb_share_check(skb, GFP_ATOMIC); 1793 if (skb == NULL) 1794 goto oom; 1795 1796 /* drop any routing info */ 1797 skb_dst_drop(skb); 1798 1799 /* drop conntrack reference */ 1800 nf_reset(skb); 1801 1802 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1803 1804 skb_push(skb, skb->data - skb_mac_header(skb)); 1805 1806 /* 1807 * The SOCK_PACKET socket receives _all_ frames. 1808 */ 1809 1810 spkt->spkt_family = dev->type; 1811 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1812 spkt->spkt_protocol = skb->protocol; 1813 1814 /* 1815 * Charge the memory to the socket. This is done specifically 1816 * to prevent sockets using all the memory up. 1817 */ 1818 1819 if (sock_queue_rcv_skb(sk, skb) == 0) 1820 return 0; 1821 1822 out: 1823 kfree_skb(skb); 1824 oom: 1825 return 0; 1826 } 1827 1828 1829 /* 1830 * Output a raw packet to a device layer. This bypasses all the other 1831 * protocol layers and you must therefore supply it with a complete frame 1832 */ 1833 1834 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1835 size_t len) 1836 { 1837 struct sock *sk = sock->sk; 1838 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1839 struct sk_buff *skb = NULL; 1840 struct net_device *dev; 1841 struct sockcm_cookie sockc; 1842 __be16 proto = 0; 1843 int err; 1844 int extra_len = 0; 1845 1846 /* 1847 * Get and verify the address. 1848 */ 1849 1850 if (saddr) { 1851 if (msg->msg_namelen < sizeof(struct sockaddr)) 1852 return -EINVAL; 1853 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1854 proto = saddr->spkt_protocol; 1855 } else 1856 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1857 1858 /* 1859 * Find the device first to size check it 1860 */ 1861 1862 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1863 retry: 1864 rcu_read_lock(); 1865 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1866 err = -ENODEV; 1867 if (dev == NULL) 1868 goto out_unlock; 1869 1870 err = -ENETDOWN; 1871 if (!(dev->flags & IFF_UP)) 1872 goto out_unlock; 1873 1874 /* 1875 * You may not queue a frame bigger than the mtu. This is the lowest level 1876 * raw protocol and you must do your own fragmentation at this level. 1877 */ 1878 1879 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1880 if (!netif_supports_nofcs(dev)) { 1881 err = -EPROTONOSUPPORT; 1882 goto out_unlock; 1883 } 1884 extra_len = 4; /* We're doing our own CRC */ 1885 } 1886 1887 err = -EMSGSIZE; 1888 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1889 goto out_unlock; 1890 1891 if (!skb) { 1892 size_t reserved = LL_RESERVED_SPACE(dev); 1893 int tlen = dev->needed_tailroom; 1894 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1895 1896 rcu_read_unlock(); 1897 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1898 if (skb == NULL) 1899 return -ENOBUFS; 1900 /* FIXME: Save some space for broken drivers that write a hard 1901 * header at transmission time by themselves. PPP is the notable 1902 * one here. This should really be fixed at the driver level. 1903 */ 1904 skb_reserve(skb, reserved); 1905 skb_reset_network_header(skb); 1906 1907 /* Try to align data part correctly */ 1908 if (hhlen) { 1909 skb->data -= hhlen; 1910 skb->tail -= hhlen; 1911 if (len < hhlen) 1912 skb_reset_network_header(skb); 1913 } 1914 err = memcpy_from_msg(skb_put(skb, len), msg, len); 1915 if (err) 1916 goto out_free; 1917 goto retry; 1918 } 1919 1920 if (!dev_validate_header(dev, skb->data, len)) { 1921 err = -EINVAL; 1922 goto out_unlock; 1923 } 1924 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 1925 !packet_extra_vlan_len_allowed(dev, skb)) { 1926 err = -EMSGSIZE; 1927 goto out_unlock; 1928 } 1929 1930 sockc.tsflags = sk->sk_tsflags; 1931 if (msg->msg_controllen) { 1932 err = sock_cmsg_send(sk, msg, &sockc); 1933 if (unlikely(err)) 1934 goto out_unlock; 1935 } 1936 1937 skb->protocol = proto; 1938 skb->dev = dev; 1939 skb->priority = sk->sk_priority; 1940 skb->mark = sk->sk_mark; 1941 1942 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags); 1943 1944 if (unlikely(extra_len == 4)) 1945 skb->no_fcs = 1; 1946 1947 skb_probe_transport_header(skb, 0); 1948 1949 dev_queue_xmit(skb); 1950 rcu_read_unlock(); 1951 return len; 1952 1953 out_unlock: 1954 rcu_read_unlock(); 1955 out_free: 1956 kfree_skb(skb); 1957 return err; 1958 } 1959 1960 static unsigned int run_filter(struct sk_buff *skb, 1961 const struct sock *sk, 1962 unsigned int res) 1963 { 1964 struct sk_filter *filter; 1965 1966 rcu_read_lock(); 1967 filter = rcu_dereference(sk->sk_filter); 1968 if (filter != NULL) 1969 res = bpf_prog_run_clear_cb(filter->prog, skb); 1970 rcu_read_unlock(); 1971 1972 return res; 1973 } 1974 1975 static int __packet_rcv_vnet(const struct sk_buff *skb, 1976 struct virtio_net_hdr *vnet_hdr) 1977 { 1978 *vnet_hdr = (const struct virtio_net_hdr) { 0 }; 1979 1980 if (skb_is_gso(skb)) { 1981 struct skb_shared_info *sinfo = skb_shinfo(skb); 1982 1983 /* This is a hint as to how much should be linear. */ 1984 vnet_hdr->hdr_len = 1985 __cpu_to_virtio16(vio_le(), skb_headlen(skb)); 1986 vnet_hdr->gso_size = 1987 __cpu_to_virtio16(vio_le(), sinfo->gso_size); 1988 1989 if (sinfo->gso_type & SKB_GSO_TCPV4) 1990 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 1991 else if (sinfo->gso_type & SKB_GSO_TCPV6) 1992 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 1993 else if (sinfo->gso_type & SKB_GSO_UDP) 1994 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; 1995 else if (sinfo->gso_type & SKB_GSO_FCOE) 1996 return -EINVAL; 1997 else 1998 BUG(); 1999 2000 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 2001 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 2002 } else 2003 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; 2004 2005 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2006 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 2007 vnet_hdr->csum_start = __cpu_to_virtio16(vio_le(), 2008 skb_checksum_start_offset(skb)); 2009 vnet_hdr->csum_offset = __cpu_to_virtio16(vio_le(), 2010 skb->csum_offset); 2011 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2012 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; 2013 } /* else everything is zero */ 2014 2015 return 0; 2016 } 2017 2018 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2019 size_t *len) 2020 { 2021 struct virtio_net_hdr vnet_hdr; 2022 2023 if (*len < sizeof(vnet_hdr)) 2024 return -EINVAL; 2025 *len -= sizeof(vnet_hdr); 2026 2027 if (__packet_rcv_vnet(skb, &vnet_hdr)) 2028 return -EINVAL; 2029 2030 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 2031 } 2032 2033 /* 2034 * This function makes lazy skb cloning in hope that most of packets 2035 * are discarded by BPF. 2036 * 2037 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2038 * and skb->cb are mangled. It works because (and until) packets 2039 * falling here are owned by current CPU. Output packets are cloned 2040 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2041 * sequencially, so that if we return skb to original state on exit, 2042 * we will not harm anyone. 2043 */ 2044 2045 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2046 struct packet_type *pt, struct net_device *orig_dev) 2047 { 2048 struct sock *sk; 2049 struct sockaddr_ll *sll; 2050 struct packet_sock *po; 2051 u8 *skb_head = skb->data; 2052 int skb_len = skb->len; 2053 unsigned int snaplen, res; 2054 bool is_drop_n_account = false; 2055 2056 if (skb->pkt_type == PACKET_LOOPBACK) 2057 goto drop; 2058 2059 sk = pt->af_packet_priv; 2060 po = pkt_sk(sk); 2061 2062 if (!net_eq(dev_net(dev), sock_net(sk))) 2063 goto drop; 2064 2065 skb->dev = dev; 2066 2067 if (dev->header_ops) { 2068 /* The device has an explicit notion of ll header, 2069 * exported to higher levels. 2070 * 2071 * Otherwise, the device hides details of its frame 2072 * structure, so that corresponding packet head is 2073 * never delivered to user. 2074 */ 2075 if (sk->sk_type != SOCK_DGRAM) 2076 skb_push(skb, skb->data - skb_mac_header(skb)); 2077 else if (skb->pkt_type == PACKET_OUTGOING) { 2078 /* Special case: outgoing packets have ll header at head */ 2079 skb_pull(skb, skb_network_offset(skb)); 2080 } 2081 } 2082 2083 snaplen = skb->len; 2084 2085 res = run_filter(skb, sk, snaplen); 2086 if (!res) 2087 goto drop_n_restore; 2088 if (snaplen > res) 2089 snaplen = res; 2090 2091 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2092 goto drop_n_acct; 2093 2094 if (skb_shared(skb)) { 2095 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2096 if (nskb == NULL) 2097 goto drop_n_acct; 2098 2099 if (skb_head != skb->data) { 2100 skb->data = skb_head; 2101 skb->len = skb_len; 2102 } 2103 consume_skb(skb); 2104 skb = nskb; 2105 } 2106 2107 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2108 2109 sll = &PACKET_SKB_CB(skb)->sa.ll; 2110 sll->sll_hatype = dev->type; 2111 sll->sll_pkttype = skb->pkt_type; 2112 if (unlikely(po->origdev)) 2113 sll->sll_ifindex = orig_dev->ifindex; 2114 else 2115 sll->sll_ifindex = dev->ifindex; 2116 2117 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2118 2119 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2120 * Use their space for storing the original skb length. 2121 */ 2122 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2123 2124 if (pskb_trim(skb, snaplen)) 2125 goto drop_n_acct; 2126 2127 skb_set_owner_r(skb, sk); 2128 skb->dev = NULL; 2129 skb_dst_drop(skb); 2130 2131 /* drop conntrack reference */ 2132 nf_reset(skb); 2133 2134 spin_lock(&sk->sk_receive_queue.lock); 2135 po->stats.stats1.tp_packets++; 2136 sock_skb_set_dropcount(sk, skb); 2137 __skb_queue_tail(&sk->sk_receive_queue, skb); 2138 spin_unlock(&sk->sk_receive_queue.lock); 2139 sk->sk_data_ready(sk); 2140 return 0; 2141 2142 drop_n_acct: 2143 is_drop_n_account = true; 2144 spin_lock(&sk->sk_receive_queue.lock); 2145 po->stats.stats1.tp_drops++; 2146 atomic_inc(&sk->sk_drops); 2147 spin_unlock(&sk->sk_receive_queue.lock); 2148 2149 drop_n_restore: 2150 if (skb_head != skb->data && skb_shared(skb)) { 2151 skb->data = skb_head; 2152 skb->len = skb_len; 2153 } 2154 drop: 2155 if (!is_drop_n_account) 2156 consume_skb(skb); 2157 else 2158 kfree_skb(skb); 2159 return 0; 2160 } 2161 2162 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2163 struct packet_type *pt, struct net_device *orig_dev) 2164 { 2165 struct sock *sk; 2166 struct packet_sock *po; 2167 struct sockaddr_ll *sll; 2168 union tpacket_uhdr h; 2169 u8 *skb_head = skb->data; 2170 int skb_len = skb->len; 2171 unsigned int snaplen, res; 2172 unsigned long status = TP_STATUS_USER; 2173 unsigned short macoff, netoff, hdrlen; 2174 struct sk_buff *copy_skb = NULL; 2175 struct timespec ts; 2176 __u32 ts_status; 2177 bool is_drop_n_account = false; 2178 2179 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2180 * We may add members to them until current aligned size without forcing 2181 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2182 */ 2183 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2184 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2185 2186 if (skb->pkt_type == PACKET_LOOPBACK) 2187 goto drop; 2188 2189 sk = pt->af_packet_priv; 2190 po = pkt_sk(sk); 2191 2192 if (!net_eq(dev_net(dev), sock_net(sk))) 2193 goto drop; 2194 2195 if (dev->header_ops) { 2196 if (sk->sk_type != SOCK_DGRAM) 2197 skb_push(skb, skb->data - skb_mac_header(skb)); 2198 else if (skb->pkt_type == PACKET_OUTGOING) { 2199 /* Special case: outgoing packets have ll header at head */ 2200 skb_pull(skb, skb_network_offset(skb)); 2201 } 2202 } 2203 2204 snaplen = skb->len; 2205 2206 res = run_filter(skb, sk, snaplen); 2207 if (!res) 2208 goto drop_n_restore; 2209 2210 if (skb->ip_summed == CHECKSUM_PARTIAL) 2211 status |= TP_STATUS_CSUMNOTREADY; 2212 else if (skb->pkt_type != PACKET_OUTGOING && 2213 (skb->ip_summed == CHECKSUM_COMPLETE || 2214 skb_csum_unnecessary(skb))) 2215 status |= TP_STATUS_CSUM_VALID; 2216 2217 if (snaplen > res) 2218 snaplen = res; 2219 2220 if (sk->sk_type == SOCK_DGRAM) { 2221 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2222 po->tp_reserve; 2223 } else { 2224 unsigned int maclen = skb_network_offset(skb); 2225 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2226 (maclen < 16 ? 16 : maclen)) + 2227 po->tp_reserve; 2228 if (po->has_vnet_hdr) 2229 netoff += sizeof(struct virtio_net_hdr); 2230 macoff = netoff - maclen; 2231 } 2232 if (po->tp_version <= TPACKET_V2) { 2233 if (macoff + snaplen > po->rx_ring.frame_size) { 2234 if (po->copy_thresh && 2235 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2236 if (skb_shared(skb)) { 2237 copy_skb = skb_clone(skb, GFP_ATOMIC); 2238 } else { 2239 copy_skb = skb_get(skb); 2240 skb_head = skb->data; 2241 } 2242 if (copy_skb) 2243 skb_set_owner_r(copy_skb, sk); 2244 } 2245 snaplen = po->rx_ring.frame_size - macoff; 2246 if ((int)snaplen < 0) 2247 snaplen = 0; 2248 } 2249 } else if (unlikely(macoff + snaplen > 2250 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2251 u32 nval; 2252 2253 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2254 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2255 snaplen, nval, macoff); 2256 snaplen = nval; 2257 if (unlikely((int)snaplen < 0)) { 2258 snaplen = 0; 2259 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2260 } 2261 } 2262 spin_lock(&sk->sk_receive_queue.lock); 2263 h.raw = packet_current_rx_frame(po, skb, 2264 TP_STATUS_KERNEL, (macoff+snaplen)); 2265 if (!h.raw) 2266 goto drop_n_account; 2267 if (po->tp_version <= TPACKET_V2) { 2268 packet_increment_rx_head(po, &po->rx_ring); 2269 /* 2270 * LOSING will be reported till you read the stats, 2271 * because it's COR - Clear On Read. 2272 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2273 * at packet level. 2274 */ 2275 if (po->stats.stats1.tp_drops) 2276 status |= TP_STATUS_LOSING; 2277 } 2278 po->stats.stats1.tp_packets++; 2279 if (copy_skb) { 2280 status |= TP_STATUS_COPY; 2281 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2282 } 2283 spin_unlock(&sk->sk_receive_queue.lock); 2284 2285 if (po->has_vnet_hdr) { 2286 if (__packet_rcv_vnet(skb, h.raw + macoff - 2287 sizeof(struct virtio_net_hdr))) { 2288 spin_lock(&sk->sk_receive_queue.lock); 2289 goto drop_n_account; 2290 } 2291 } 2292 2293 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2294 2295 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 2296 getnstimeofday(&ts); 2297 2298 status |= ts_status; 2299 2300 switch (po->tp_version) { 2301 case TPACKET_V1: 2302 h.h1->tp_len = skb->len; 2303 h.h1->tp_snaplen = snaplen; 2304 h.h1->tp_mac = macoff; 2305 h.h1->tp_net = netoff; 2306 h.h1->tp_sec = ts.tv_sec; 2307 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2308 hdrlen = sizeof(*h.h1); 2309 break; 2310 case TPACKET_V2: 2311 h.h2->tp_len = skb->len; 2312 h.h2->tp_snaplen = snaplen; 2313 h.h2->tp_mac = macoff; 2314 h.h2->tp_net = netoff; 2315 h.h2->tp_sec = ts.tv_sec; 2316 h.h2->tp_nsec = ts.tv_nsec; 2317 if (skb_vlan_tag_present(skb)) { 2318 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2319 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2320 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2321 } else { 2322 h.h2->tp_vlan_tci = 0; 2323 h.h2->tp_vlan_tpid = 0; 2324 } 2325 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2326 hdrlen = sizeof(*h.h2); 2327 break; 2328 case TPACKET_V3: 2329 /* tp_nxt_offset,vlan are already populated above. 2330 * So DONT clear those fields here 2331 */ 2332 h.h3->tp_status |= status; 2333 h.h3->tp_len = skb->len; 2334 h.h3->tp_snaplen = snaplen; 2335 h.h3->tp_mac = macoff; 2336 h.h3->tp_net = netoff; 2337 h.h3->tp_sec = ts.tv_sec; 2338 h.h3->tp_nsec = ts.tv_nsec; 2339 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2340 hdrlen = sizeof(*h.h3); 2341 break; 2342 default: 2343 BUG(); 2344 } 2345 2346 sll = h.raw + TPACKET_ALIGN(hdrlen); 2347 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2348 sll->sll_family = AF_PACKET; 2349 sll->sll_hatype = dev->type; 2350 sll->sll_protocol = skb->protocol; 2351 sll->sll_pkttype = skb->pkt_type; 2352 if (unlikely(po->origdev)) 2353 sll->sll_ifindex = orig_dev->ifindex; 2354 else 2355 sll->sll_ifindex = dev->ifindex; 2356 2357 smp_mb(); 2358 2359 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2360 if (po->tp_version <= TPACKET_V2) { 2361 u8 *start, *end; 2362 2363 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2364 macoff + snaplen); 2365 2366 for (start = h.raw; start < end; start += PAGE_SIZE) 2367 flush_dcache_page(pgv_to_page(start)); 2368 } 2369 smp_wmb(); 2370 #endif 2371 2372 if (po->tp_version <= TPACKET_V2) { 2373 __packet_set_status(po, h.raw, status); 2374 sk->sk_data_ready(sk); 2375 } else { 2376 prb_clear_blk_fill_status(&po->rx_ring); 2377 } 2378 2379 drop_n_restore: 2380 if (skb_head != skb->data && skb_shared(skb)) { 2381 skb->data = skb_head; 2382 skb->len = skb_len; 2383 } 2384 drop: 2385 if (!is_drop_n_account) 2386 consume_skb(skb); 2387 else 2388 kfree_skb(skb); 2389 return 0; 2390 2391 drop_n_account: 2392 is_drop_n_account = true; 2393 po->stats.stats1.tp_drops++; 2394 spin_unlock(&sk->sk_receive_queue.lock); 2395 2396 sk->sk_data_ready(sk); 2397 kfree_skb(copy_skb); 2398 goto drop_n_restore; 2399 } 2400 2401 static void tpacket_destruct_skb(struct sk_buff *skb) 2402 { 2403 struct packet_sock *po = pkt_sk(skb->sk); 2404 2405 if (likely(po->tx_ring.pg_vec)) { 2406 void *ph; 2407 __u32 ts; 2408 2409 ph = skb_shinfo(skb)->destructor_arg; 2410 packet_dec_pending(&po->tx_ring); 2411 2412 ts = __packet_set_timestamp(po, ph, skb); 2413 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2414 } 2415 2416 sock_wfree(skb); 2417 } 2418 2419 static void tpacket_set_protocol(const struct net_device *dev, 2420 struct sk_buff *skb) 2421 { 2422 if (dev->type == ARPHRD_ETHER) { 2423 skb_reset_mac_header(skb); 2424 skb->protocol = eth_hdr(skb)->h_proto; 2425 } 2426 } 2427 2428 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2429 { 2430 unsigned short gso_type = 0; 2431 2432 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2433 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2434 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2435 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2436 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2437 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2438 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2439 2440 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2441 return -EINVAL; 2442 2443 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 2444 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 2445 case VIRTIO_NET_HDR_GSO_TCPV4: 2446 gso_type = SKB_GSO_TCPV4; 2447 break; 2448 case VIRTIO_NET_HDR_GSO_TCPV6: 2449 gso_type = SKB_GSO_TCPV6; 2450 break; 2451 case VIRTIO_NET_HDR_GSO_UDP: 2452 gso_type = SKB_GSO_UDP; 2453 break; 2454 default: 2455 return -EINVAL; 2456 } 2457 2458 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) 2459 gso_type |= SKB_GSO_TCP_ECN; 2460 2461 if (vnet_hdr->gso_size == 0) 2462 return -EINVAL; 2463 } 2464 2465 vnet_hdr->gso_type = gso_type; /* changes type, temporary storage */ 2466 return 0; 2467 } 2468 2469 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2470 struct virtio_net_hdr *vnet_hdr) 2471 { 2472 int n; 2473 2474 if (*len < sizeof(*vnet_hdr)) 2475 return -EINVAL; 2476 *len -= sizeof(*vnet_hdr); 2477 2478 n = copy_from_iter(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter); 2479 if (n != sizeof(*vnet_hdr)) 2480 return -EFAULT; 2481 2482 return __packet_snd_vnet_parse(vnet_hdr, *len); 2483 } 2484 2485 static int packet_snd_vnet_gso(struct sk_buff *skb, 2486 struct virtio_net_hdr *vnet_hdr) 2487 { 2488 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 2489 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start); 2490 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset); 2491 2492 if (!skb_partial_csum_set(skb, s, o)) 2493 return -EINVAL; 2494 } 2495 2496 skb_shinfo(skb)->gso_size = 2497 __virtio16_to_cpu(vio_le(), vnet_hdr->gso_size); 2498 skb_shinfo(skb)->gso_type = vnet_hdr->gso_type; 2499 2500 /* Header must be checked, and gso_segs computed. */ 2501 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 2502 skb_shinfo(skb)->gso_segs = 0; 2503 return 0; 2504 } 2505 2506 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2507 void *frame, struct net_device *dev, void *data, int tp_len, 2508 __be16 proto, unsigned char *addr, int hlen, int copylen, 2509 const struct sockcm_cookie *sockc) 2510 { 2511 union tpacket_uhdr ph; 2512 int to_write, offset, len, nr_frags, len_max; 2513 struct socket *sock = po->sk.sk_socket; 2514 struct page *page; 2515 int err; 2516 2517 ph.raw = frame; 2518 2519 skb->protocol = proto; 2520 skb->dev = dev; 2521 skb->priority = po->sk.sk_priority; 2522 skb->mark = po->sk.sk_mark; 2523 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags); 2524 skb_shinfo(skb)->destructor_arg = ph.raw; 2525 2526 skb_reserve(skb, hlen); 2527 skb_reset_network_header(skb); 2528 2529 to_write = tp_len; 2530 2531 if (sock->type == SOCK_DGRAM) { 2532 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2533 NULL, tp_len); 2534 if (unlikely(err < 0)) 2535 return -EINVAL; 2536 } else if (copylen) { 2537 int hdrlen = min_t(int, copylen, tp_len); 2538 2539 skb_push(skb, dev->hard_header_len); 2540 skb_put(skb, copylen - dev->hard_header_len); 2541 err = skb_store_bits(skb, 0, data, hdrlen); 2542 if (unlikely(err)) 2543 return err; 2544 if (!dev_validate_header(dev, skb->data, hdrlen)) 2545 return -EINVAL; 2546 if (!skb->protocol) 2547 tpacket_set_protocol(dev, skb); 2548 2549 data += hdrlen; 2550 to_write -= hdrlen; 2551 } 2552 2553 offset = offset_in_page(data); 2554 len_max = PAGE_SIZE - offset; 2555 len = ((to_write > len_max) ? len_max : to_write); 2556 2557 skb->data_len = to_write; 2558 skb->len += to_write; 2559 skb->truesize += to_write; 2560 atomic_add(to_write, &po->sk.sk_wmem_alloc); 2561 2562 while (likely(to_write)) { 2563 nr_frags = skb_shinfo(skb)->nr_frags; 2564 2565 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2566 pr_err("Packet exceed the number of skb frags(%lu)\n", 2567 MAX_SKB_FRAGS); 2568 return -EFAULT; 2569 } 2570 2571 page = pgv_to_page(data); 2572 data += len; 2573 flush_dcache_page(page); 2574 get_page(page); 2575 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2576 to_write -= len; 2577 offset = 0; 2578 len_max = PAGE_SIZE; 2579 len = ((to_write > len_max) ? len_max : to_write); 2580 } 2581 2582 skb_probe_transport_header(skb, 0); 2583 2584 return tp_len; 2585 } 2586 2587 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2588 int size_max, void **data) 2589 { 2590 union tpacket_uhdr ph; 2591 int tp_len, off; 2592 2593 ph.raw = frame; 2594 2595 switch (po->tp_version) { 2596 case TPACKET_V2: 2597 tp_len = ph.h2->tp_len; 2598 break; 2599 default: 2600 tp_len = ph.h1->tp_len; 2601 break; 2602 } 2603 if (unlikely(tp_len > size_max)) { 2604 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2605 return -EMSGSIZE; 2606 } 2607 2608 if (unlikely(po->tp_tx_has_off)) { 2609 int off_min, off_max; 2610 2611 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2612 off_max = po->tx_ring.frame_size - tp_len; 2613 if (po->sk.sk_type == SOCK_DGRAM) { 2614 switch (po->tp_version) { 2615 case TPACKET_V2: 2616 off = ph.h2->tp_net; 2617 break; 2618 default: 2619 off = ph.h1->tp_net; 2620 break; 2621 } 2622 } else { 2623 switch (po->tp_version) { 2624 case TPACKET_V2: 2625 off = ph.h2->tp_mac; 2626 break; 2627 default: 2628 off = ph.h1->tp_mac; 2629 break; 2630 } 2631 } 2632 if (unlikely((off < off_min) || (off_max < off))) 2633 return -EINVAL; 2634 } else { 2635 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2636 } 2637 2638 *data = frame + off; 2639 return tp_len; 2640 } 2641 2642 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2643 { 2644 struct sk_buff *skb; 2645 struct net_device *dev; 2646 struct virtio_net_hdr *vnet_hdr = NULL; 2647 struct sockcm_cookie sockc; 2648 __be16 proto; 2649 int err, reserve = 0; 2650 void *ph; 2651 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2652 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2653 int tp_len, size_max; 2654 unsigned char *addr; 2655 void *data; 2656 int len_sum = 0; 2657 int status = TP_STATUS_AVAILABLE; 2658 int hlen, tlen, copylen = 0; 2659 2660 mutex_lock(&po->pg_vec_lock); 2661 2662 if (likely(saddr == NULL)) { 2663 dev = packet_cached_dev_get(po); 2664 proto = po->num; 2665 addr = NULL; 2666 } else { 2667 err = -EINVAL; 2668 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2669 goto out; 2670 if (msg->msg_namelen < (saddr->sll_halen 2671 + offsetof(struct sockaddr_ll, 2672 sll_addr))) 2673 goto out; 2674 proto = saddr->sll_protocol; 2675 addr = saddr->sll_addr; 2676 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2677 } 2678 2679 sockc.tsflags = po->sk.sk_tsflags; 2680 if (msg->msg_controllen) { 2681 err = sock_cmsg_send(&po->sk, msg, &sockc); 2682 if (unlikely(err)) 2683 goto out; 2684 } 2685 2686 err = -ENXIO; 2687 if (unlikely(dev == NULL)) 2688 goto out; 2689 err = -ENETDOWN; 2690 if (unlikely(!(dev->flags & IFF_UP))) 2691 goto out_put; 2692 2693 if (po->sk.sk_socket->type == SOCK_RAW) 2694 reserve = dev->hard_header_len; 2695 size_max = po->tx_ring.frame_size 2696 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2697 2698 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) 2699 size_max = dev->mtu + reserve + VLAN_HLEN; 2700 2701 do { 2702 ph = packet_current_frame(po, &po->tx_ring, 2703 TP_STATUS_SEND_REQUEST); 2704 if (unlikely(ph == NULL)) { 2705 if (need_wait && need_resched()) 2706 schedule(); 2707 continue; 2708 } 2709 2710 skb = NULL; 2711 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2712 if (tp_len < 0) 2713 goto tpacket_error; 2714 2715 status = TP_STATUS_SEND_REQUEST; 2716 hlen = LL_RESERVED_SPACE(dev); 2717 tlen = dev->needed_tailroom; 2718 if (po->has_vnet_hdr) { 2719 vnet_hdr = data; 2720 data += sizeof(*vnet_hdr); 2721 tp_len -= sizeof(*vnet_hdr); 2722 if (tp_len < 0 || 2723 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2724 tp_len = -EINVAL; 2725 goto tpacket_error; 2726 } 2727 copylen = __virtio16_to_cpu(vio_le(), 2728 vnet_hdr->hdr_len); 2729 } 2730 copylen = max_t(int, copylen, dev->hard_header_len); 2731 skb = sock_alloc_send_skb(&po->sk, 2732 hlen + tlen + sizeof(struct sockaddr_ll) + 2733 (copylen - dev->hard_header_len), 2734 !need_wait, &err); 2735 2736 if (unlikely(skb == NULL)) { 2737 /* we assume the socket was initially writeable ... */ 2738 if (likely(len_sum > 0)) 2739 err = len_sum; 2740 goto out_status; 2741 } 2742 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2743 addr, hlen, copylen, &sockc); 2744 if (likely(tp_len >= 0) && 2745 tp_len > dev->mtu + reserve && 2746 !po->has_vnet_hdr && 2747 !packet_extra_vlan_len_allowed(dev, skb)) 2748 tp_len = -EMSGSIZE; 2749 2750 if (unlikely(tp_len < 0)) { 2751 tpacket_error: 2752 if (po->tp_loss) { 2753 __packet_set_status(po, ph, 2754 TP_STATUS_AVAILABLE); 2755 packet_increment_head(&po->tx_ring); 2756 kfree_skb(skb); 2757 continue; 2758 } else { 2759 status = TP_STATUS_WRONG_FORMAT; 2760 err = tp_len; 2761 goto out_status; 2762 } 2763 } 2764 2765 if (po->has_vnet_hdr && packet_snd_vnet_gso(skb, vnet_hdr)) { 2766 tp_len = -EINVAL; 2767 goto tpacket_error; 2768 } 2769 2770 packet_pick_tx_queue(dev, skb); 2771 2772 skb->destructor = tpacket_destruct_skb; 2773 __packet_set_status(po, ph, TP_STATUS_SENDING); 2774 packet_inc_pending(&po->tx_ring); 2775 2776 status = TP_STATUS_SEND_REQUEST; 2777 err = po->xmit(skb); 2778 if (unlikely(err > 0)) { 2779 err = net_xmit_errno(err); 2780 if (err && __packet_get_status(po, ph) == 2781 TP_STATUS_AVAILABLE) { 2782 /* skb was destructed already */ 2783 skb = NULL; 2784 goto out_status; 2785 } 2786 /* 2787 * skb was dropped but not destructed yet; 2788 * let's treat it like congestion or err < 0 2789 */ 2790 err = 0; 2791 } 2792 packet_increment_head(&po->tx_ring); 2793 len_sum += tp_len; 2794 } while (likely((ph != NULL) || 2795 /* Note: packet_read_pending() might be slow if we have 2796 * to call it as it's per_cpu variable, but in fast-path 2797 * we already short-circuit the loop with the first 2798 * condition, and luckily don't have to go that path 2799 * anyway. 2800 */ 2801 (need_wait && packet_read_pending(&po->tx_ring)))); 2802 2803 err = len_sum; 2804 goto out_put; 2805 2806 out_status: 2807 __packet_set_status(po, ph, status); 2808 kfree_skb(skb); 2809 out_put: 2810 dev_put(dev); 2811 out: 2812 mutex_unlock(&po->pg_vec_lock); 2813 return err; 2814 } 2815 2816 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2817 size_t reserve, size_t len, 2818 size_t linear, int noblock, 2819 int *err) 2820 { 2821 struct sk_buff *skb; 2822 2823 /* Under a page? Don't bother with paged skb. */ 2824 if (prepad + len < PAGE_SIZE || !linear) 2825 linear = len; 2826 2827 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2828 err, 0); 2829 if (!skb) 2830 return NULL; 2831 2832 skb_reserve(skb, reserve); 2833 skb_put(skb, linear); 2834 skb->data_len = len - linear; 2835 skb->len += len - linear; 2836 2837 return skb; 2838 } 2839 2840 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2841 { 2842 struct sock *sk = sock->sk; 2843 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2844 struct sk_buff *skb; 2845 struct net_device *dev; 2846 __be16 proto; 2847 unsigned char *addr; 2848 int err, reserve = 0; 2849 struct sockcm_cookie sockc; 2850 struct virtio_net_hdr vnet_hdr = { 0 }; 2851 int offset = 0; 2852 struct packet_sock *po = pkt_sk(sk); 2853 int hlen, tlen; 2854 int extra_len = 0; 2855 2856 /* 2857 * Get and verify the address. 2858 */ 2859 2860 if (likely(saddr == NULL)) { 2861 dev = packet_cached_dev_get(po); 2862 proto = po->num; 2863 addr = NULL; 2864 } else { 2865 err = -EINVAL; 2866 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2867 goto out; 2868 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2869 goto out; 2870 proto = saddr->sll_protocol; 2871 addr = saddr->sll_addr; 2872 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2873 } 2874 2875 err = -ENXIO; 2876 if (unlikely(dev == NULL)) 2877 goto out_unlock; 2878 err = -ENETDOWN; 2879 if (unlikely(!(dev->flags & IFF_UP))) 2880 goto out_unlock; 2881 2882 sockc.tsflags = sk->sk_tsflags; 2883 sockc.mark = sk->sk_mark; 2884 if (msg->msg_controllen) { 2885 err = sock_cmsg_send(sk, msg, &sockc); 2886 if (unlikely(err)) 2887 goto out_unlock; 2888 } 2889 2890 if (sock->type == SOCK_RAW) 2891 reserve = dev->hard_header_len; 2892 if (po->has_vnet_hdr) { 2893 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2894 if (err) 2895 goto out_unlock; 2896 } 2897 2898 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2899 if (!netif_supports_nofcs(dev)) { 2900 err = -EPROTONOSUPPORT; 2901 goto out_unlock; 2902 } 2903 extra_len = 4; /* We're doing our own CRC */ 2904 } 2905 2906 err = -EMSGSIZE; 2907 if (!vnet_hdr.gso_type && 2908 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2909 goto out_unlock; 2910 2911 err = -ENOBUFS; 2912 hlen = LL_RESERVED_SPACE(dev); 2913 tlen = dev->needed_tailroom; 2914 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, 2915 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len), 2916 msg->msg_flags & MSG_DONTWAIT, &err); 2917 if (skb == NULL) 2918 goto out_unlock; 2919 2920 skb_set_network_header(skb, reserve); 2921 2922 err = -EINVAL; 2923 if (sock->type == SOCK_DGRAM) { 2924 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 2925 if (unlikely(offset < 0)) 2926 goto out_free; 2927 } 2928 2929 /* Returns -EFAULT on error */ 2930 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 2931 if (err) 2932 goto out_free; 2933 2934 if (sock->type == SOCK_RAW && 2935 !dev_validate_header(dev, skb->data, len)) { 2936 err = -EINVAL; 2937 goto out_free; 2938 } 2939 2940 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags); 2941 2942 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 2943 !packet_extra_vlan_len_allowed(dev, skb)) { 2944 err = -EMSGSIZE; 2945 goto out_free; 2946 } 2947 2948 skb->protocol = proto; 2949 skb->dev = dev; 2950 skb->priority = sk->sk_priority; 2951 skb->mark = sockc.mark; 2952 2953 packet_pick_tx_queue(dev, skb); 2954 2955 if (po->has_vnet_hdr) { 2956 err = packet_snd_vnet_gso(skb, &vnet_hdr); 2957 if (err) 2958 goto out_free; 2959 len += sizeof(vnet_hdr); 2960 } 2961 2962 skb_probe_transport_header(skb, reserve); 2963 2964 if (unlikely(extra_len == 4)) 2965 skb->no_fcs = 1; 2966 2967 err = po->xmit(skb); 2968 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2969 goto out_unlock; 2970 2971 dev_put(dev); 2972 2973 return len; 2974 2975 out_free: 2976 kfree_skb(skb); 2977 out_unlock: 2978 if (dev) 2979 dev_put(dev); 2980 out: 2981 return err; 2982 } 2983 2984 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2985 { 2986 struct sock *sk = sock->sk; 2987 struct packet_sock *po = pkt_sk(sk); 2988 2989 if (po->tx_ring.pg_vec) 2990 return tpacket_snd(po, msg); 2991 else 2992 return packet_snd(sock, msg, len); 2993 } 2994 2995 /* 2996 * Close a PACKET socket. This is fairly simple. We immediately go 2997 * to 'closed' state and remove our protocol entry in the device list. 2998 */ 2999 3000 static int packet_release(struct socket *sock) 3001 { 3002 struct sock *sk = sock->sk; 3003 struct packet_sock *po; 3004 struct net *net; 3005 union tpacket_req_u req_u; 3006 3007 if (!sk) 3008 return 0; 3009 3010 net = sock_net(sk); 3011 po = pkt_sk(sk); 3012 3013 mutex_lock(&net->packet.sklist_lock); 3014 sk_del_node_init_rcu(sk); 3015 mutex_unlock(&net->packet.sklist_lock); 3016 3017 preempt_disable(); 3018 sock_prot_inuse_add(net, sk->sk_prot, -1); 3019 preempt_enable(); 3020 3021 spin_lock(&po->bind_lock); 3022 unregister_prot_hook(sk, false); 3023 packet_cached_dev_reset(po); 3024 3025 if (po->prot_hook.dev) { 3026 dev_put(po->prot_hook.dev); 3027 po->prot_hook.dev = NULL; 3028 } 3029 spin_unlock(&po->bind_lock); 3030 3031 packet_flush_mclist(sk); 3032 3033 if (po->rx_ring.pg_vec) { 3034 memset(&req_u, 0, sizeof(req_u)); 3035 packet_set_ring(sk, &req_u, 1, 0); 3036 } 3037 3038 if (po->tx_ring.pg_vec) { 3039 memset(&req_u, 0, sizeof(req_u)); 3040 packet_set_ring(sk, &req_u, 1, 1); 3041 } 3042 3043 fanout_release(sk); 3044 3045 synchronize_net(); 3046 /* 3047 * Now the socket is dead. No more input will appear. 3048 */ 3049 sock_orphan(sk); 3050 sock->sk = NULL; 3051 3052 /* Purge queues */ 3053 3054 skb_queue_purge(&sk->sk_receive_queue); 3055 packet_free_pending(po); 3056 sk_refcnt_debug_release(sk); 3057 3058 sock_put(sk); 3059 return 0; 3060 } 3061 3062 /* 3063 * Attach a packet hook. 3064 */ 3065 3066 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3067 __be16 proto) 3068 { 3069 struct packet_sock *po = pkt_sk(sk); 3070 struct net_device *dev_curr; 3071 __be16 proto_curr; 3072 bool need_rehook; 3073 struct net_device *dev = NULL; 3074 int ret = 0; 3075 bool unlisted = false; 3076 3077 if (po->fanout) 3078 return -EINVAL; 3079 3080 lock_sock(sk); 3081 spin_lock(&po->bind_lock); 3082 rcu_read_lock(); 3083 3084 if (name) { 3085 dev = dev_get_by_name_rcu(sock_net(sk), name); 3086 if (!dev) { 3087 ret = -ENODEV; 3088 goto out_unlock; 3089 } 3090 } else if (ifindex) { 3091 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3092 if (!dev) { 3093 ret = -ENODEV; 3094 goto out_unlock; 3095 } 3096 } 3097 3098 if (dev) 3099 dev_hold(dev); 3100 3101 proto_curr = po->prot_hook.type; 3102 dev_curr = po->prot_hook.dev; 3103 3104 need_rehook = proto_curr != proto || dev_curr != dev; 3105 3106 if (need_rehook) { 3107 if (po->running) { 3108 rcu_read_unlock(); 3109 __unregister_prot_hook(sk, true); 3110 rcu_read_lock(); 3111 dev_curr = po->prot_hook.dev; 3112 if (dev) 3113 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3114 dev->ifindex); 3115 } 3116 3117 po->num = proto; 3118 po->prot_hook.type = proto; 3119 3120 if (unlikely(unlisted)) { 3121 dev_put(dev); 3122 po->prot_hook.dev = NULL; 3123 po->ifindex = -1; 3124 packet_cached_dev_reset(po); 3125 } else { 3126 po->prot_hook.dev = dev; 3127 po->ifindex = dev ? dev->ifindex : 0; 3128 packet_cached_dev_assign(po, dev); 3129 } 3130 } 3131 if (dev_curr) 3132 dev_put(dev_curr); 3133 3134 if (proto == 0 || !need_rehook) 3135 goto out_unlock; 3136 3137 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3138 register_prot_hook(sk); 3139 } else { 3140 sk->sk_err = ENETDOWN; 3141 if (!sock_flag(sk, SOCK_DEAD)) 3142 sk->sk_error_report(sk); 3143 } 3144 3145 out_unlock: 3146 rcu_read_unlock(); 3147 spin_unlock(&po->bind_lock); 3148 release_sock(sk); 3149 return ret; 3150 } 3151 3152 /* 3153 * Bind a packet socket to a device 3154 */ 3155 3156 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3157 int addr_len) 3158 { 3159 struct sock *sk = sock->sk; 3160 char name[15]; 3161 3162 /* 3163 * Check legality 3164 */ 3165 3166 if (addr_len != sizeof(struct sockaddr)) 3167 return -EINVAL; 3168 strlcpy(name, uaddr->sa_data, sizeof(name)); 3169 3170 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3171 } 3172 3173 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3174 { 3175 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3176 struct sock *sk = sock->sk; 3177 3178 /* 3179 * Check legality 3180 */ 3181 3182 if (addr_len < sizeof(struct sockaddr_ll)) 3183 return -EINVAL; 3184 if (sll->sll_family != AF_PACKET) 3185 return -EINVAL; 3186 3187 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3188 sll->sll_protocol ? : pkt_sk(sk)->num); 3189 } 3190 3191 static struct proto packet_proto = { 3192 .name = "PACKET", 3193 .owner = THIS_MODULE, 3194 .obj_size = sizeof(struct packet_sock), 3195 }; 3196 3197 /* 3198 * Create a packet of type SOCK_PACKET. 3199 */ 3200 3201 static int packet_create(struct net *net, struct socket *sock, int protocol, 3202 int kern) 3203 { 3204 struct sock *sk; 3205 struct packet_sock *po; 3206 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3207 int err; 3208 3209 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3210 return -EPERM; 3211 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3212 sock->type != SOCK_PACKET) 3213 return -ESOCKTNOSUPPORT; 3214 3215 sock->state = SS_UNCONNECTED; 3216 3217 err = -ENOBUFS; 3218 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3219 if (sk == NULL) 3220 goto out; 3221 3222 sock->ops = &packet_ops; 3223 if (sock->type == SOCK_PACKET) 3224 sock->ops = &packet_ops_spkt; 3225 3226 sock_init_data(sock, sk); 3227 3228 po = pkt_sk(sk); 3229 sk->sk_family = PF_PACKET; 3230 po->num = proto; 3231 po->xmit = dev_queue_xmit; 3232 3233 err = packet_alloc_pending(po); 3234 if (err) 3235 goto out2; 3236 3237 packet_cached_dev_reset(po); 3238 3239 sk->sk_destruct = packet_sock_destruct; 3240 sk_refcnt_debug_inc(sk); 3241 3242 /* 3243 * Attach a protocol block 3244 */ 3245 3246 spin_lock_init(&po->bind_lock); 3247 mutex_init(&po->pg_vec_lock); 3248 po->rollover = NULL; 3249 po->prot_hook.func = packet_rcv; 3250 3251 if (sock->type == SOCK_PACKET) 3252 po->prot_hook.func = packet_rcv_spkt; 3253 3254 po->prot_hook.af_packet_priv = sk; 3255 3256 if (proto) { 3257 po->prot_hook.type = proto; 3258 register_prot_hook(sk); 3259 } 3260 3261 mutex_lock(&net->packet.sklist_lock); 3262 sk_add_node_rcu(sk, &net->packet.sklist); 3263 mutex_unlock(&net->packet.sklist_lock); 3264 3265 preempt_disable(); 3266 sock_prot_inuse_add(net, &packet_proto, 1); 3267 preempt_enable(); 3268 3269 return 0; 3270 out2: 3271 sk_free(sk); 3272 out: 3273 return err; 3274 } 3275 3276 /* 3277 * Pull a packet from our receive queue and hand it to the user. 3278 * If necessary we block. 3279 */ 3280 3281 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3282 int flags) 3283 { 3284 struct sock *sk = sock->sk; 3285 struct sk_buff *skb; 3286 int copied, err; 3287 int vnet_hdr_len = 0; 3288 unsigned int origlen = 0; 3289 3290 err = -EINVAL; 3291 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3292 goto out; 3293 3294 #if 0 3295 /* What error should we return now? EUNATTACH? */ 3296 if (pkt_sk(sk)->ifindex < 0) 3297 return -ENODEV; 3298 #endif 3299 3300 if (flags & MSG_ERRQUEUE) { 3301 err = sock_recv_errqueue(sk, msg, len, 3302 SOL_PACKET, PACKET_TX_TIMESTAMP); 3303 goto out; 3304 } 3305 3306 /* 3307 * Call the generic datagram receiver. This handles all sorts 3308 * of horrible races and re-entrancy so we can forget about it 3309 * in the protocol layers. 3310 * 3311 * Now it will return ENETDOWN, if device have just gone down, 3312 * but then it will block. 3313 */ 3314 3315 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 3316 3317 /* 3318 * An error occurred so return it. Because skb_recv_datagram() 3319 * handles the blocking we don't see and worry about blocking 3320 * retries. 3321 */ 3322 3323 if (skb == NULL) 3324 goto out; 3325 3326 if (pkt_sk(sk)->pressure) 3327 packet_rcv_has_room(pkt_sk(sk), NULL); 3328 3329 if (pkt_sk(sk)->has_vnet_hdr) { 3330 err = packet_rcv_vnet(msg, skb, &len); 3331 if (err) 3332 goto out_free; 3333 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3334 } 3335 3336 /* You lose any data beyond the buffer you gave. If it worries 3337 * a user program they can ask the device for its MTU 3338 * anyway. 3339 */ 3340 copied = skb->len; 3341 if (copied > len) { 3342 copied = len; 3343 msg->msg_flags |= MSG_TRUNC; 3344 } 3345 3346 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3347 if (err) 3348 goto out_free; 3349 3350 if (sock->type != SOCK_PACKET) { 3351 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3352 3353 /* Original length was stored in sockaddr_ll fields */ 3354 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3355 sll->sll_family = AF_PACKET; 3356 sll->sll_protocol = skb->protocol; 3357 } 3358 3359 sock_recv_ts_and_drops(msg, sk, skb); 3360 3361 if (msg->msg_name) { 3362 /* If the address length field is there to be filled 3363 * in, we fill it in now. 3364 */ 3365 if (sock->type == SOCK_PACKET) { 3366 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3367 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3368 } else { 3369 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3370 3371 msg->msg_namelen = sll->sll_halen + 3372 offsetof(struct sockaddr_ll, sll_addr); 3373 } 3374 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, 3375 msg->msg_namelen); 3376 } 3377 3378 if (pkt_sk(sk)->auxdata) { 3379 struct tpacket_auxdata aux; 3380 3381 aux.tp_status = TP_STATUS_USER; 3382 if (skb->ip_summed == CHECKSUM_PARTIAL) 3383 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3384 else if (skb->pkt_type != PACKET_OUTGOING && 3385 (skb->ip_summed == CHECKSUM_COMPLETE || 3386 skb_csum_unnecessary(skb))) 3387 aux.tp_status |= TP_STATUS_CSUM_VALID; 3388 3389 aux.tp_len = origlen; 3390 aux.tp_snaplen = skb->len; 3391 aux.tp_mac = 0; 3392 aux.tp_net = skb_network_offset(skb); 3393 if (skb_vlan_tag_present(skb)) { 3394 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3395 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3396 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3397 } else { 3398 aux.tp_vlan_tci = 0; 3399 aux.tp_vlan_tpid = 0; 3400 } 3401 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3402 } 3403 3404 /* 3405 * Free or return the buffer as appropriate. Again this 3406 * hides all the races and re-entrancy issues from us. 3407 */ 3408 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3409 3410 out_free: 3411 skb_free_datagram(sk, skb); 3412 out: 3413 return err; 3414 } 3415 3416 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3417 int *uaddr_len, int peer) 3418 { 3419 struct net_device *dev; 3420 struct sock *sk = sock->sk; 3421 3422 if (peer) 3423 return -EOPNOTSUPP; 3424 3425 uaddr->sa_family = AF_PACKET; 3426 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3427 rcu_read_lock(); 3428 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3429 if (dev) 3430 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3431 rcu_read_unlock(); 3432 *uaddr_len = sizeof(*uaddr); 3433 3434 return 0; 3435 } 3436 3437 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3438 int *uaddr_len, int peer) 3439 { 3440 struct net_device *dev; 3441 struct sock *sk = sock->sk; 3442 struct packet_sock *po = pkt_sk(sk); 3443 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3444 3445 if (peer) 3446 return -EOPNOTSUPP; 3447 3448 sll->sll_family = AF_PACKET; 3449 sll->sll_ifindex = po->ifindex; 3450 sll->sll_protocol = po->num; 3451 sll->sll_pkttype = 0; 3452 rcu_read_lock(); 3453 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 3454 if (dev) { 3455 sll->sll_hatype = dev->type; 3456 sll->sll_halen = dev->addr_len; 3457 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3458 } else { 3459 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3460 sll->sll_halen = 0; 3461 } 3462 rcu_read_unlock(); 3463 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3464 3465 return 0; 3466 } 3467 3468 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3469 int what) 3470 { 3471 switch (i->type) { 3472 case PACKET_MR_MULTICAST: 3473 if (i->alen != dev->addr_len) 3474 return -EINVAL; 3475 if (what > 0) 3476 return dev_mc_add(dev, i->addr); 3477 else 3478 return dev_mc_del(dev, i->addr); 3479 break; 3480 case PACKET_MR_PROMISC: 3481 return dev_set_promiscuity(dev, what); 3482 case PACKET_MR_ALLMULTI: 3483 return dev_set_allmulti(dev, what); 3484 case PACKET_MR_UNICAST: 3485 if (i->alen != dev->addr_len) 3486 return -EINVAL; 3487 if (what > 0) 3488 return dev_uc_add(dev, i->addr); 3489 else 3490 return dev_uc_del(dev, i->addr); 3491 break; 3492 default: 3493 break; 3494 } 3495 return 0; 3496 } 3497 3498 static void packet_dev_mclist_delete(struct net_device *dev, 3499 struct packet_mclist **mlp) 3500 { 3501 struct packet_mclist *ml; 3502 3503 while ((ml = *mlp) != NULL) { 3504 if (ml->ifindex == dev->ifindex) { 3505 packet_dev_mc(dev, ml, -1); 3506 *mlp = ml->next; 3507 kfree(ml); 3508 } else 3509 mlp = &ml->next; 3510 } 3511 } 3512 3513 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3514 { 3515 struct packet_sock *po = pkt_sk(sk); 3516 struct packet_mclist *ml, *i; 3517 struct net_device *dev; 3518 int err; 3519 3520 rtnl_lock(); 3521 3522 err = -ENODEV; 3523 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3524 if (!dev) 3525 goto done; 3526 3527 err = -EINVAL; 3528 if (mreq->mr_alen > dev->addr_len) 3529 goto done; 3530 3531 err = -ENOBUFS; 3532 i = kmalloc(sizeof(*i), GFP_KERNEL); 3533 if (i == NULL) 3534 goto done; 3535 3536 err = 0; 3537 for (ml = po->mclist; ml; ml = ml->next) { 3538 if (ml->ifindex == mreq->mr_ifindex && 3539 ml->type == mreq->mr_type && 3540 ml->alen == mreq->mr_alen && 3541 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3542 ml->count++; 3543 /* Free the new element ... */ 3544 kfree(i); 3545 goto done; 3546 } 3547 } 3548 3549 i->type = mreq->mr_type; 3550 i->ifindex = mreq->mr_ifindex; 3551 i->alen = mreq->mr_alen; 3552 memcpy(i->addr, mreq->mr_address, i->alen); 3553 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3554 i->count = 1; 3555 i->next = po->mclist; 3556 po->mclist = i; 3557 err = packet_dev_mc(dev, i, 1); 3558 if (err) { 3559 po->mclist = i->next; 3560 kfree(i); 3561 } 3562 3563 done: 3564 rtnl_unlock(); 3565 return err; 3566 } 3567 3568 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3569 { 3570 struct packet_mclist *ml, **mlp; 3571 3572 rtnl_lock(); 3573 3574 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3575 if (ml->ifindex == mreq->mr_ifindex && 3576 ml->type == mreq->mr_type && 3577 ml->alen == mreq->mr_alen && 3578 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3579 if (--ml->count == 0) { 3580 struct net_device *dev; 3581 *mlp = ml->next; 3582 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3583 if (dev) 3584 packet_dev_mc(dev, ml, -1); 3585 kfree(ml); 3586 } 3587 break; 3588 } 3589 } 3590 rtnl_unlock(); 3591 return 0; 3592 } 3593 3594 static void packet_flush_mclist(struct sock *sk) 3595 { 3596 struct packet_sock *po = pkt_sk(sk); 3597 struct packet_mclist *ml; 3598 3599 if (!po->mclist) 3600 return; 3601 3602 rtnl_lock(); 3603 while ((ml = po->mclist) != NULL) { 3604 struct net_device *dev; 3605 3606 po->mclist = ml->next; 3607 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3608 if (dev != NULL) 3609 packet_dev_mc(dev, ml, -1); 3610 kfree(ml); 3611 } 3612 rtnl_unlock(); 3613 } 3614 3615 static int 3616 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3617 { 3618 struct sock *sk = sock->sk; 3619 struct packet_sock *po = pkt_sk(sk); 3620 int ret; 3621 3622 if (level != SOL_PACKET) 3623 return -ENOPROTOOPT; 3624 3625 switch (optname) { 3626 case PACKET_ADD_MEMBERSHIP: 3627 case PACKET_DROP_MEMBERSHIP: 3628 { 3629 struct packet_mreq_max mreq; 3630 int len = optlen; 3631 memset(&mreq, 0, sizeof(mreq)); 3632 if (len < sizeof(struct packet_mreq)) 3633 return -EINVAL; 3634 if (len > sizeof(mreq)) 3635 len = sizeof(mreq); 3636 if (copy_from_user(&mreq, optval, len)) 3637 return -EFAULT; 3638 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3639 return -EINVAL; 3640 if (optname == PACKET_ADD_MEMBERSHIP) 3641 ret = packet_mc_add(sk, &mreq); 3642 else 3643 ret = packet_mc_drop(sk, &mreq); 3644 return ret; 3645 } 3646 3647 case PACKET_RX_RING: 3648 case PACKET_TX_RING: 3649 { 3650 union tpacket_req_u req_u; 3651 int len; 3652 3653 switch (po->tp_version) { 3654 case TPACKET_V1: 3655 case TPACKET_V2: 3656 len = sizeof(req_u.req); 3657 break; 3658 case TPACKET_V3: 3659 default: 3660 len = sizeof(req_u.req3); 3661 break; 3662 } 3663 if (optlen < len) 3664 return -EINVAL; 3665 if (copy_from_user(&req_u.req, optval, len)) 3666 return -EFAULT; 3667 return packet_set_ring(sk, &req_u, 0, 3668 optname == PACKET_TX_RING); 3669 } 3670 case PACKET_COPY_THRESH: 3671 { 3672 int val; 3673 3674 if (optlen != sizeof(val)) 3675 return -EINVAL; 3676 if (copy_from_user(&val, optval, sizeof(val))) 3677 return -EFAULT; 3678 3679 pkt_sk(sk)->copy_thresh = val; 3680 return 0; 3681 } 3682 case PACKET_VERSION: 3683 { 3684 int val; 3685 3686 if (optlen != sizeof(val)) 3687 return -EINVAL; 3688 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3689 return -EBUSY; 3690 if (copy_from_user(&val, optval, sizeof(val))) 3691 return -EFAULT; 3692 switch (val) { 3693 case TPACKET_V1: 3694 case TPACKET_V2: 3695 case TPACKET_V3: 3696 po->tp_version = val; 3697 return 0; 3698 default: 3699 return -EINVAL; 3700 } 3701 } 3702 case PACKET_RESERVE: 3703 { 3704 unsigned int val; 3705 3706 if (optlen != sizeof(val)) 3707 return -EINVAL; 3708 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3709 return -EBUSY; 3710 if (copy_from_user(&val, optval, sizeof(val))) 3711 return -EFAULT; 3712 po->tp_reserve = val; 3713 return 0; 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 (copy_from_user(&val, optval, len)) 3882 return -EFAULT; 3883 switch (val) { 3884 case TPACKET_V1: 3885 val = sizeof(struct tpacket_hdr); 3886 break; 3887 case TPACKET_V2: 3888 val = sizeof(struct tpacket2_hdr); 3889 break; 3890 case TPACKET_V3: 3891 val = sizeof(struct tpacket3_hdr); 3892 break; 3893 default: 3894 return -EINVAL; 3895 } 3896 break; 3897 case PACKET_RESERVE: 3898 val = po->tp_reserve; 3899 break; 3900 case PACKET_LOSS: 3901 val = po->tp_loss; 3902 break; 3903 case PACKET_TIMESTAMP: 3904 val = po->tp_tstamp; 3905 break; 3906 case PACKET_FANOUT: 3907 val = (po->fanout ? 3908 ((u32)po->fanout->id | 3909 ((u32)po->fanout->type << 16) | 3910 ((u32)po->fanout->flags << 24)) : 3911 0); 3912 break; 3913 case PACKET_ROLLOVER_STATS: 3914 if (!po->rollover) 3915 return -EINVAL; 3916 rstats.tp_all = atomic_long_read(&po->rollover->num); 3917 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 3918 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 3919 data = &rstats; 3920 lv = sizeof(rstats); 3921 break; 3922 case PACKET_TX_HAS_OFF: 3923 val = po->tp_tx_has_off; 3924 break; 3925 case PACKET_QDISC_BYPASS: 3926 val = packet_use_direct_xmit(po); 3927 break; 3928 default: 3929 return -ENOPROTOOPT; 3930 } 3931 3932 if (len > lv) 3933 len = lv; 3934 if (put_user(len, optlen)) 3935 return -EFAULT; 3936 if (copy_to_user(optval, data, len)) 3937 return -EFAULT; 3938 return 0; 3939 } 3940 3941 3942 #ifdef CONFIG_COMPAT 3943 static int compat_packet_setsockopt(struct socket *sock, int level, int optname, 3944 char __user *optval, unsigned int optlen) 3945 { 3946 struct packet_sock *po = pkt_sk(sock->sk); 3947 3948 if (level != SOL_PACKET) 3949 return -ENOPROTOOPT; 3950 3951 if (optname == PACKET_FANOUT_DATA && 3952 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) { 3953 optval = (char __user *)get_compat_bpf_fprog(optval); 3954 if (!optval) 3955 return -EFAULT; 3956 optlen = sizeof(struct sock_fprog); 3957 } 3958 3959 return packet_setsockopt(sock, level, optname, optval, optlen); 3960 } 3961 #endif 3962 3963 static int packet_notifier(struct notifier_block *this, 3964 unsigned long msg, void *ptr) 3965 { 3966 struct sock *sk; 3967 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 3968 struct net *net = dev_net(dev); 3969 3970 rcu_read_lock(); 3971 sk_for_each_rcu(sk, &net->packet.sklist) { 3972 struct packet_sock *po = pkt_sk(sk); 3973 3974 switch (msg) { 3975 case NETDEV_UNREGISTER: 3976 if (po->mclist) 3977 packet_dev_mclist_delete(dev, &po->mclist); 3978 /* fallthrough */ 3979 3980 case NETDEV_DOWN: 3981 if (dev->ifindex == po->ifindex) { 3982 spin_lock(&po->bind_lock); 3983 if (po->running) { 3984 __unregister_prot_hook(sk, false); 3985 sk->sk_err = ENETDOWN; 3986 if (!sock_flag(sk, SOCK_DEAD)) 3987 sk->sk_error_report(sk); 3988 } 3989 if (msg == NETDEV_UNREGISTER) { 3990 packet_cached_dev_reset(po); 3991 po->ifindex = -1; 3992 if (po->prot_hook.dev) 3993 dev_put(po->prot_hook.dev); 3994 po->prot_hook.dev = NULL; 3995 } 3996 spin_unlock(&po->bind_lock); 3997 } 3998 break; 3999 case NETDEV_UP: 4000 if (dev->ifindex == po->ifindex) { 4001 spin_lock(&po->bind_lock); 4002 if (po->num) 4003 register_prot_hook(sk); 4004 spin_unlock(&po->bind_lock); 4005 } 4006 break; 4007 } 4008 } 4009 rcu_read_unlock(); 4010 return NOTIFY_DONE; 4011 } 4012 4013 4014 static int packet_ioctl(struct socket *sock, unsigned int cmd, 4015 unsigned long arg) 4016 { 4017 struct sock *sk = sock->sk; 4018 4019 switch (cmd) { 4020 case SIOCOUTQ: 4021 { 4022 int amount = sk_wmem_alloc_get(sk); 4023 4024 return put_user(amount, (int __user *)arg); 4025 } 4026 case SIOCINQ: 4027 { 4028 struct sk_buff *skb; 4029 int amount = 0; 4030 4031 spin_lock_bh(&sk->sk_receive_queue.lock); 4032 skb = skb_peek(&sk->sk_receive_queue); 4033 if (skb) 4034 amount = skb->len; 4035 spin_unlock_bh(&sk->sk_receive_queue.lock); 4036 return put_user(amount, (int __user *)arg); 4037 } 4038 case SIOCGSTAMP: 4039 return sock_get_timestamp(sk, (struct timeval __user *)arg); 4040 case SIOCGSTAMPNS: 4041 return sock_get_timestampns(sk, (struct timespec __user *)arg); 4042 4043 #ifdef CONFIG_INET 4044 case SIOCADDRT: 4045 case SIOCDELRT: 4046 case SIOCDARP: 4047 case SIOCGARP: 4048 case SIOCSARP: 4049 case SIOCGIFADDR: 4050 case SIOCSIFADDR: 4051 case SIOCGIFBRDADDR: 4052 case SIOCSIFBRDADDR: 4053 case SIOCGIFNETMASK: 4054 case SIOCSIFNETMASK: 4055 case SIOCGIFDSTADDR: 4056 case SIOCSIFDSTADDR: 4057 case SIOCSIFFLAGS: 4058 return inet_dgram_ops.ioctl(sock, cmd, arg); 4059 #endif 4060 4061 default: 4062 return -ENOIOCTLCMD; 4063 } 4064 return 0; 4065 } 4066 4067 static unsigned int packet_poll(struct file *file, struct socket *sock, 4068 poll_table *wait) 4069 { 4070 struct sock *sk = sock->sk; 4071 struct packet_sock *po = pkt_sk(sk); 4072 unsigned int mask = datagram_poll(file, sock, wait); 4073 4074 spin_lock_bh(&sk->sk_receive_queue.lock); 4075 if (po->rx_ring.pg_vec) { 4076 if (!packet_previous_rx_frame(po, &po->rx_ring, 4077 TP_STATUS_KERNEL)) 4078 mask |= POLLIN | POLLRDNORM; 4079 } 4080 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 4081 po->pressure = 0; 4082 spin_unlock_bh(&sk->sk_receive_queue.lock); 4083 spin_lock_bh(&sk->sk_write_queue.lock); 4084 if (po->tx_ring.pg_vec) { 4085 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4086 mask |= POLLOUT | POLLWRNORM; 4087 } 4088 spin_unlock_bh(&sk->sk_write_queue.lock); 4089 return mask; 4090 } 4091 4092 4093 /* Dirty? Well, I still did not learn better way to account 4094 * for user mmaps. 4095 */ 4096 4097 static void packet_mm_open(struct vm_area_struct *vma) 4098 { 4099 struct file *file = vma->vm_file; 4100 struct socket *sock = file->private_data; 4101 struct sock *sk = sock->sk; 4102 4103 if (sk) 4104 atomic_inc(&pkt_sk(sk)->mapped); 4105 } 4106 4107 static void packet_mm_close(struct vm_area_struct *vma) 4108 { 4109 struct file *file = vma->vm_file; 4110 struct socket *sock = file->private_data; 4111 struct sock *sk = sock->sk; 4112 4113 if (sk) 4114 atomic_dec(&pkt_sk(sk)->mapped); 4115 } 4116 4117 static const struct vm_operations_struct packet_mmap_ops = { 4118 .open = packet_mm_open, 4119 .close = packet_mm_close, 4120 }; 4121 4122 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4123 unsigned int len) 4124 { 4125 int i; 4126 4127 for (i = 0; i < len; i++) { 4128 if (likely(pg_vec[i].buffer)) { 4129 if (is_vmalloc_addr(pg_vec[i].buffer)) 4130 vfree(pg_vec[i].buffer); 4131 else 4132 free_pages((unsigned long)pg_vec[i].buffer, 4133 order); 4134 pg_vec[i].buffer = NULL; 4135 } 4136 } 4137 kfree(pg_vec); 4138 } 4139 4140 static char *alloc_one_pg_vec_page(unsigned long order) 4141 { 4142 char *buffer; 4143 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4144 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4145 4146 buffer = (char *) __get_free_pages(gfp_flags, order); 4147 if (buffer) 4148 return buffer; 4149 4150 /* __get_free_pages failed, fall back to vmalloc */ 4151 buffer = vzalloc((1 << order) * PAGE_SIZE); 4152 if (buffer) 4153 return buffer; 4154 4155 /* vmalloc failed, lets dig into swap here */ 4156 gfp_flags &= ~__GFP_NORETRY; 4157 buffer = (char *) __get_free_pages(gfp_flags, order); 4158 if (buffer) 4159 return buffer; 4160 4161 /* complete and utter failure */ 4162 return NULL; 4163 } 4164 4165 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4166 { 4167 unsigned int block_nr = req->tp_block_nr; 4168 struct pgv *pg_vec; 4169 int i; 4170 4171 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL); 4172 if (unlikely(!pg_vec)) 4173 goto out; 4174 4175 for (i = 0; i < block_nr; i++) { 4176 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4177 if (unlikely(!pg_vec[i].buffer)) 4178 goto out_free_pgvec; 4179 } 4180 4181 out: 4182 return pg_vec; 4183 4184 out_free_pgvec: 4185 free_pg_vec(pg_vec, order, block_nr); 4186 pg_vec = NULL; 4187 goto out; 4188 } 4189 4190 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4191 int closing, int tx_ring) 4192 { 4193 struct pgv *pg_vec = NULL; 4194 struct packet_sock *po = pkt_sk(sk); 4195 int was_running, order = 0; 4196 struct packet_ring_buffer *rb; 4197 struct sk_buff_head *rb_queue; 4198 __be16 num; 4199 int err = -EINVAL; 4200 /* Added to avoid minimal code churn */ 4201 struct tpacket_req *req = &req_u->req; 4202 4203 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */ 4204 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) { 4205 net_warn_ratelimited("Tx-ring is not supported.\n"); 4206 goto out; 4207 } 4208 4209 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4210 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4211 4212 err = -EBUSY; 4213 if (!closing) { 4214 if (atomic_read(&po->mapped)) 4215 goto out; 4216 if (packet_read_pending(rb)) 4217 goto out; 4218 } 4219 4220 if (req->tp_block_nr) { 4221 /* Sanity tests and some calculations */ 4222 err = -EBUSY; 4223 if (unlikely(rb->pg_vec)) 4224 goto out; 4225 4226 switch (po->tp_version) { 4227 case TPACKET_V1: 4228 po->tp_hdrlen = TPACKET_HDRLEN; 4229 break; 4230 case TPACKET_V2: 4231 po->tp_hdrlen = TPACKET2_HDRLEN; 4232 break; 4233 case TPACKET_V3: 4234 po->tp_hdrlen = TPACKET3_HDRLEN; 4235 break; 4236 } 4237 4238 err = -EINVAL; 4239 if (unlikely((int)req->tp_block_size <= 0)) 4240 goto out; 4241 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4242 goto out; 4243 if (po->tp_version >= TPACKET_V3 && 4244 (int)(req->tp_block_size - 4245 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0) 4246 goto out; 4247 if (unlikely(req->tp_frame_size < po->tp_hdrlen + 4248 po->tp_reserve)) 4249 goto out; 4250 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4251 goto out; 4252 4253 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4254 if (unlikely(rb->frames_per_block == 0)) 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 /* Transmit path is not supported. We checked 4268 * it above but just being paranoid 4269 */ 4270 if (!tx_ring) 4271 init_prb_bdqc(po, rb, pg_vec, req_u); 4272 break; 4273 default: 4274 break; 4275 } 4276 } 4277 /* Done */ 4278 else { 4279 err = -EINVAL; 4280 if (unlikely(req->tp_frame_nr)) 4281 goto out; 4282 } 4283 4284 lock_sock(sk); 4285 4286 /* Detach socket from network */ 4287 spin_lock(&po->bind_lock); 4288 was_running = po->running; 4289 num = po->num; 4290 if (was_running) { 4291 po->num = 0; 4292 __unregister_prot_hook(sk, false); 4293 } 4294 spin_unlock(&po->bind_lock); 4295 4296 synchronize_net(); 4297 4298 err = -EBUSY; 4299 mutex_lock(&po->pg_vec_lock); 4300 if (closing || atomic_read(&po->mapped) == 0) { 4301 err = 0; 4302 spin_lock_bh(&rb_queue->lock); 4303 swap(rb->pg_vec, pg_vec); 4304 rb->frame_max = (req->tp_frame_nr - 1); 4305 rb->head = 0; 4306 rb->frame_size = req->tp_frame_size; 4307 spin_unlock_bh(&rb_queue->lock); 4308 4309 swap(rb->pg_vec_order, order); 4310 swap(rb->pg_vec_len, req->tp_block_nr); 4311 4312 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4313 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4314 tpacket_rcv : packet_rcv; 4315 skb_queue_purge(rb_queue); 4316 if (atomic_read(&po->mapped)) 4317 pr_err("packet_mmap: vma is busy: %d\n", 4318 atomic_read(&po->mapped)); 4319 } 4320 mutex_unlock(&po->pg_vec_lock); 4321 4322 spin_lock(&po->bind_lock); 4323 if (was_running) { 4324 po->num = num; 4325 register_prot_hook(sk); 4326 } 4327 spin_unlock(&po->bind_lock); 4328 if (closing && (po->tp_version > TPACKET_V2)) { 4329 /* Because we don't support block-based V3 on tx-ring */ 4330 if (!tx_ring) 4331 prb_shutdown_retire_blk_timer(po, rb_queue); 4332 } 4333 release_sock(sk); 4334 4335 if (pg_vec) 4336 free_pg_vec(pg_vec, order, req->tp_block_nr); 4337 out: 4338 return err; 4339 } 4340 4341 static int packet_mmap(struct file *file, struct socket *sock, 4342 struct vm_area_struct *vma) 4343 { 4344 struct sock *sk = sock->sk; 4345 struct packet_sock *po = pkt_sk(sk); 4346 unsigned long size, expected_size; 4347 struct packet_ring_buffer *rb; 4348 unsigned long start; 4349 int err = -EINVAL; 4350 int i; 4351 4352 if (vma->vm_pgoff) 4353 return -EINVAL; 4354 4355 mutex_lock(&po->pg_vec_lock); 4356 4357 expected_size = 0; 4358 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4359 if (rb->pg_vec) { 4360 expected_size += rb->pg_vec_len 4361 * rb->pg_vec_pages 4362 * PAGE_SIZE; 4363 } 4364 } 4365 4366 if (expected_size == 0) 4367 goto out; 4368 4369 size = vma->vm_end - vma->vm_start; 4370 if (size != expected_size) 4371 goto out; 4372 4373 start = vma->vm_start; 4374 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4375 if (rb->pg_vec == NULL) 4376 continue; 4377 4378 for (i = 0; i < rb->pg_vec_len; i++) { 4379 struct page *page; 4380 void *kaddr = rb->pg_vec[i].buffer; 4381 int pg_num; 4382 4383 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4384 page = pgv_to_page(kaddr); 4385 err = vm_insert_page(vma, start, page); 4386 if (unlikely(err)) 4387 goto out; 4388 start += PAGE_SIZE; 4389 kaddr += PAGE_SIZE; 4390 } 4391 } 4392 } 4393 4394 atomic_inc(&po->mapped); 4395 vma->vm_ops = &packet_mmap_ops; 4396 err = 0; 4397 4398 out: 4399 mutex_unlock(&po->pg_vec_lock); 4400 return err; 4401 } 4402 4403 static const struct proto_ops packet_ops_spkt = { 4404 .family = PF_PACKET, 4405 .owner = THIS_MODULE, 4406 .release = packet_release, 4407 .bind = packet_bind_spkt, 4408 .connect = sock_no_connect, 4409 .socketpair = sock_no_socketpair, 4410 .accept = sock_no_accept, 4411 .getname = packet_getname_spkt, 4412 .poll = datagram_poll, 4413 .ioctl = packet_ioctl, 4414 .listen = sock_no_listen, 4415 .shutdown = sock_no_shutdown, 4416 .setsockopt = sock_no_setsockopt, 4417 .getsockopt = sock_no_getsockopt, 4418 .sendmsg = packet_sendmsg_spkt, 4419 .recvmsg = packet_recvmsg, 4420 .mmap = sock_no_mmap, 4421 .sendpage = sock_no_sendpage, 4422 }; 4423 4424 static const struct proto_ops packet_ops = { 4425 .family = PF_PACKET, 4426 .owner = THIS_MODULE, 4427 .release = packet_release, 4428 .bind = packet_bind, 4429 .connect = sock_no_connect, 4430 .socketpair = sock_no_socketpair, 4431 .accept = sock_no_accept, 4432 .getname = packet_getname, 4433 .poll = packet_poll, 4434 .ioctl = packet_ioctl, 4435 .listen = sock_no_listen, 4436 .shutdown = sock_no_shutdown, 4437 .setsockopt = packet_setsockopt, 4438 .getsockopt = packet_getsockopt, 4439 #ifdef CONFIG_COMPAT 4440 .compat_setsockopt = compat_packet_setsockopt, 4441 #endif 4442 .sendmsg = packet_sendmsg, 4443 .recvmsg = packet_recvmsg, 4444 .mmap = packet_mmap, 4445 .sendpage = sock_no_sendpage, 4446 }; 4447 4448 static const struct net_proto_family packet_family_ops = { 4449 .family = PF_PACKET, 4450 .create = packet_create, 4451 .owner = THIS_MODULE, 4452 }; 4453 4454 static struct notifier_block packet_netdev_notifier = { 4455 .notifier_call = packet_notifier, 4456 }; 4457 4458 #ifdef CONFIG_PROC_FS 4459 4460 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4461 __acquires(RCU) 4462 { 4463 struct net *net = seq_file_net(seq); 4464 4465 rcu_read_lock(); 4466 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4467 } 4468 4469 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4470 { 4471 struct net *net = seq_file_net(seq); 4472 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4473 } 4474 4475 static void packet_seq_stop(struct seq_file *seq, void *v) 4476 __releases(RCU) 4477 { 4478 rcu_read_unlock(); 4479 } 4480 4481 static int packet_seq_show(struct seq_file *seq, void *v) 4482 { 4483 if (v == SEQ_START_TOKEN) 4484 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4485 else { 4486 struct sock *s = sk_entry(v); 4487 const struct packet_sock *po = pkt_sk(s); 4488 4489 seq_printf(seq, 4490 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4491 s, 4492 atomic_read(&s->sk_refcnt), 4493 s->sk_type, 4494 ntohs(po->num), 4495 po->ifindex, 4496 po->running, 4497 atomic_read(&s->sk_rmem_alloc), 4498 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4499 sock_i_ino(s)); 4500 } 4501 4502 return 0; 4503 } 4504 4505 static const struct seq_operations packet_seq_ops = { 4506 .start = packet_seq_start, 4507 .next = packet_seq_next, 4508 .stop = packet_seq_stop, 4509 .show = packet_seq_show, 4510 }; 4511 4512 static int packet_seq_open(struct inode *inode, struct file *file) 4513 { 4514 return seq_open_net(inode, file, &packet_seq_ops, 4515 sizeof(struct seq_net_private)); 4516 } 4517 4518 static const struct file_operations packet_seq_fops = { 4519 .owner = THIS_MODULE, 4520 .open = packet_seq_open, 4521 .read = seq_read, 4522 .llseek = seq_lseek, 4523 .release = seq_release_net, 4524 }; 4525 4526 #endif 4527 4528 static int __net_init packet_net_init(struct net *net) 4529 { 4530 mutex_init(&net->packet.sklist_lock); 4531 INIT_HLIST_HEAD(&net->packet.sklist); 4532 4533 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops)) 4534 return -ENOMEM; 4535 4536 return 0; 4537 } 4538 4539 static void __net_exit packet_net_exit(struct net *net) 4540 { 4541 remove_proc_entry("packet", net->proc_net); 4542 } 4543 4544 static struct pernet_operations packet_net_ops = { 4545 .init = packet_net_init, 4546 .exit = packet_net_exit, 4547 }; 4548 4549 4550 static void __exit packet_exit(void) 4551 { 4552 unregister_netdevice_notifier(&packet_netdev_notifier); 4553 unregister_pernet_subsys(&packet_net_ops); 4554 sock_unregister(PF_PACKET); 4555 proto_unregister(&packet_proto); 4556 } 4557 4558 static int __init packet_init(void) 4559 { 4560 int rc = proto_register(&packet_proto, 0); 4561 4562 if (rc != 0) 4563 goto out; 4564 4565 sock_register(&packet_family_ops); 4566 register_pernet_subsys(&packet_net_ops); 4567 register_netdevice_notifier(&packet_netdev_notifier); 4568 out: 4569 return rc; 4570 } 4571 4572 module_init(packet_init); 4573 module_exit(packet_exit); 4574 MODULE_LICENSE("GPL"); 4575 MODULE_ALIAS_NETPROTO(PF_PACKET); 4576