1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * PACKET - implements raw packet sockets. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Alan Cox, <gw4pts@gw4pts.ampr.org> 11 * 12 * Fixes: 13 * Alan Cox : verify_area() now used correctly 14 * Alan Cox : new skbuff lists, look ma no backlogs! 15 * Alan Cox : tidied skbuff lists. 16 * Alan Cox : Now uses generic datagram routines I 17 * added. Also fixed the peek/read crash 18 * from all old Linux datagram code. 19 * Alan Cox : Uses the improved datagram code. 20 * Alan Cox : Added NULL's for socket options. 21 * Alan Cox : Re-commented the code. 22 * Alan Cox : Use new kernel side addressing 23 * Rob Janssen : Correct MTU usage. 24 * Dave Platt : Counter leaks caused by incorrect 25 * interrupt locking and some slightly 26 * dubious gcc output. Can you read 27 * compiler: it said _VOLATILE_ 28 * Richard Kooijman : Timestamp fixes. 29 * Alan Cox : New buffers. Use sk->mac.raw. 30 * Alan Cox : sendmsg/recvmsg support. 31 * Alan Cox : Protocol setting support 32 * Alexey Kuznetsov : Untied from IPv4 stack. 33 * Cyrus Durgin : Fixed kerneld for kmod. 34 * Michal Ostrowski : Module initialization cleanup. 35 * Ulises Alonso : Frame number limit removal and 36 * packet_set_ring memory leak. 37 * Eric Biederman : Allow for > 8 byte hardware addresses. 38 * The convention is that longer addresses 39 * will simply extend the hardware address 40 * byte arrays at the end of sockaddr_ll 41 * and packet_mreq. 42 * Johann Baudy : Added TX RING. 43 * Chetan Loke : Implemented TPACKET_V3 block abstraction 44 * layer. 45 * Copyright (C) 2011, <lokec@ccs.neu.edu> 46 * 47 * 48 * This program is free software; you can redistribute it and/or 49 * modify it under the terms of the GNU General Public License 50 * as published by the Free Software Foundation; either version 51 * 2 of the License, or (at your option) any later version. 52 * 53 */ 54 55 #include <linux/types.h> 56 #include <linux/mm.h> 57 #include <linux/capability.h> 58 #include <linux/fcntl.h> 59 #include <linux/socket.h> 60 #include <linux/in.h> 61 #include <linux/inet.h> 62 #include <linux/netdevice.h> 63 #include <linux/if_packet.h> 64 #include <linux/wireless.h> 65 #include <linux/kernel.h> 66 #include <linux/kmod.h> 67 #include <linux/slab.h> 68 #include <linux/vmalloc.h> 69 #include <net/net_namespace.h> 70 #include <net/ip.h> 71 #include <net/protocol.h> 72 #include <linux/skbuff.h> 73 #include <net/sock.h> 74 #include <linux/errno.h> 75 #include <linux/timer.h> 76 #include <linux/uaccess.h> 77 #include <asm/ioctls.h> 78 #include <asm/page.h> 79 #include <asm/cacheflush.h> 80 #include <asm/io.h> 81 #include <linux/proc_fs.h> 82 #include <linux/seq_file.h> 83 #include <linux/poll.h> 84 #include <linux/module.h> 85 #include <linux/init.h> 86 #include <linux/mutex.h> 87 #include <linux/if_vlan.h> 88 #include <linux/virtio_net.h> 89 #include <linux/errqueue.h> 90 #include <linux/net_tstamp.h> 91 #include <linux/percpu.h> 92 #ifdef CONFIG_INET 93 #include <net/inet_common.h> 94 #endif 95 #include <linux/bpf.h> 96 #include <net/compat.h> 97 98 #include "internal.h" 99 100 /* 101 Assumptions: 102 - if device has no dev->hard_header routine, it adds and removes ll header 103 inside itself. In this case ll header is invisible outside of device, 104 but higher levels still should reserve dev->hard_header_len. 105 Some devices are enough clever to reallocate skb, when header 106 will not fit to reserved space (tunnel), another ones are silly 107 (PPP). 108 - packet socket receives packets with pulled ll header, 109 so that SOCK_RAW should push it back. 110 111 On receive: 112 ----------- 113 114 Incoming, dev->hard_header!=NULL 115 mac_header -> ll header 116 data -> data 117 118 Outgoing, dev->hard_header!=NULL 119 mac_header -> ll header 120 data -> ll header 121 122 Incoming, dev->hard_header==NULL 123 mac_header -> UNKNOWN position. It is very likely, that it points to ll 124 header. PPP makes it, that is wrong, because introduce 125 assymetry between rx and tx paths. 126 data -> data 127 128 Outgoing, dev->hard_header==NULL 129 mac_header -> data. ll header is still not built! 130 data -> data 131 132 Resume 133 If dev->hard_header==NULL we are unlikely to restore sensible ll header. 134 135 136 On transmit: 137 ------------ 138 139 dev->hard_header != NULL 140 mac_header -> ll header 141 data -> ll header 142 143 dev->hard_header == NULL (ll header is added by device, we cannot control it) 144 mac_header -> data 145 data -> data 146 147 We should set nh.raw on output to correct posistion, 148 packet classifier depends on it. 149 */ 150 151 /* Private packet socket structures. */ 152 153 /* identical to struct packet_mreq except it has 154 * a longer address field. 155 */ 156 struct packet_mreq_max { 157 int mr_ifindex; 158 unsigned short mr_type; 159 unsigned short mr_alen; 160 unsigned char mr_address[MAX_ADDR_LEN]; 161 }; 162 163 union tpacket_uhdr { 164 struct tpacket_hdr *h1; 165 struct tpacket2_hdr *h2; 166 struct tpacket3_hdr *h3; 167 void *raw; 168 }; 169 170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 171 int closing, int tx_ring); 172 173 #define V3_ALIGNMENT (8) 174 175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) 176 177 #define BLK_PLUS_PRIV(sz_of_priv) \ 178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) 179 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) 181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) 182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) 183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) 184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) 186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x))) 187 188 struct packet_sock; 189 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 190 struct packet_type *pt, struct net_device *orig_dev); 191 192 static void *packet_previous_frame(struct packet_sock *po, 193 struct packet_ring_buffer *rb, 194 int status); 195 static void packet_increment_head(struct packet_ring_buffer *buff); 196 static int prb_curr_blk_in_use(struct tpacket_block_desc *); 197 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 198 struct packet_sock *); 199 static void prb_retire_current_block(struct tpacket_kbdq_core *, 200 struct packet_sock *, unsigned int status); 201 static int prb_queue_frozen(struct tpacket_kbdq_core *); 202 static void prb_open_block(struct tpacket_kbdq_core *, 203 struct tpacket_block_desc *); 204 static void prb_retire_rx_blk_timer_expired(unsigned long); 205 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); 206 static void prb_init_blk_timer(struct packet_sock *, 207 struct tpacket_kbdq_core *, 208 void (*func) (unsigned long)); 209 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 210 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 211 struct tpacket3_hdr *); 212 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 213 struct tpacket3_hdr *); 214 static void packet_flush_mclist(struct sock *sk); 215 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb); 216 217 struct packet_skb_cb { 218 union { 219 struct sockaddr_pkt pkt; 220 union { 221 /* Trick: alias skb original length with 222 * ll.sll_family and ll.protocol in order 223 * to save room. 224 */ 225 unsigned int origlen; 226 struct sockaddr_ll ll; 227 }; 228 } sa; 229 }; 230 231 #define vio_le() virtio_legacy_is_little_endian() 232 233 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 234 235 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 236 #define GET_PBLOCK_DESC(x, bid) \ 237 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 238 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 239 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 240 #define GET_NEXT_PRB_BLK_NUM(x) \ 241 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 242 ((x)->kactive_blk_num+1) : 0) 243 244 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 245 static void __fanout_link(struct sock *sk, struct packet_sock *po); 246 247 static int packet_direct_xmit(struct sk_buff *skb) 248 { 249 struct net_device *dev = skb->dev; 250 struct sk_buff *orig_skb = skb; 251 struct netdev_queue *txq; 252 int ret = NETDEV_TX_BUSY; 253 254 if (unlikely(!netif_running(dev) || 255 !netif_carrier_ok(dev))) 256 goto drop; 257 258 skb = validate_xmit_skb_list(skb, dev); 259 if (skb != orig_skb) 260 goto drop; 261 262 packet_pick_tx_queue(dev, skb); 263 txq = skb_get_tx_queue(dev, skb); 264 265 local_bh_disable(); 266 267 HARD_TX_LOCK(dev, txq, smp_processor_id()); 268 if (!netif_xmit_frozen_or_drv_stopped(txq)) 269 ret = netdev_start_xmit(skb, dev, txq, false); 270 HARD_TX_UNLOCK(dev, txq); 271 272 local_bh_enable(); 273 274 if (!dev_xmit_complete(ret)) 275 kfree_skb(skb); 276 277 return ret; 278 drop: 279 atomic_long_inc(&dev->tx_dropped); 280 kfree_skb_list(skb); 281 return NET_XMIT_DROP; 282 } 283 284 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 285 { 286 struct net_device *dev; 287 288 rcu_read_lock(); 289 dev = rcu_dereference(po->cached_dev); 290 if (likely(dev)) 291 dev_hold(dev); 292 rcu_read_unlock(); 293 294 return dev; 295 } 296 297 static void packet_cached_dev_assign(struct packet_sock *po, 298 struct net_device *dev) 299 { 300 rcu_assign_pointer(po->cached_dev, dev); 301 } 302 303 static void packet_cached_dev_reset(struct packet_sock *po) 304 { 305 RCU_INIT_POINTER(po->cached_dev, NULL); 306 } 307 308 static bool packet_use_direct_xmit(const struct packet_sock *po) 309 { 310 return po->xmit == packet_direct_xmit; 311 } 312 313 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 314 { 315 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues; 316 } 317 318 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 319 { 320 const struct net_device_ops *ops = dev->netdev_ops; 321 u16 queue_index; 322 323 if (ops->ndo_select_queue) { 324 queue_index = ops->ndo_select_queue(dev, skb, NULL, 325 __packet_pick_tx_queue); 326 queue_index = netdev_cap_txqueue(dev, queue_index); 327 } else { 328 queue_index = __packet_pick_tx_queue(dev, skb); 329 } 330 331 skb_set_queue_mapping(skb, queue_index); 332 } 333 334 /* register_prot_hook must be invoked with the po->bind_lock held, 335 * or from a context in which asynchronous accesses to the packet 336 * socket is not possible (packet_create()). 337 */ 338 static void register_prot_hook(struct sock *sk) 339 { 340 struct packet_sock *po = pkt_sk(sk); 341 342 if (!po->running) { 343 if (po->fanout) 344 __fanout_link(sk, po); 345 else 346 dev_add_pack(&po->prot_hook); 347 348 sock_hold(sk); 349 po->running = 1; 350 } 351 } 352 353 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock 354 * held. If the sync parameter is true, we will temporarily drop 355 * the po->bind_lock and do a synchronize_net to make sure no 356 * asynchronous packet processing paths still refer to the elements 357 * of po->prot_hook. If the sync parameter is false, it is the 358 * callers responsibility to take care of this. 359 */ 360 static void __unregister_prot_hook(struct sock *sk, bool sync) 361 { 362 struct packet_sock *po = pkt_sk(sk); 363 364 po->running = 0; 365 366 if (po->fanout) 367 __fanout_unlink(sk, po); 368 else 369 __dev_remove_pack(&po->prot_hook); 370 371 __sock_put(sk); 372 373 if (sync) { 374 spin_unlock(&po->bind_lock); 375 synchronize_net(); 376 spin_lock(&po->bind_lock); 377 } 378 } 379 380 static void unregister_prot_hook(struct sock *sk, bool sync) 381 { 382 struct packet_sock *po = pkt_sk(sk); 383 384 if (po->running) 385 __unregister_prot_hook(sk, sync); 386 } 387 388 static inline struct page * __pure pgv_to_page(void *addr) 389 { 390 if (is_vmalloc_addr(addr)) 391 return vmalloc_to_page(addr); 392 return virt_to_page(addr); 393 } 394 395 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 396 { 397 union tpacket_uhdr h; 398 399 h.raw = frame; 400 switch (po->tp_version) { 401 case TPACKET_V1: 402 h.h1->tp_status = status; 403 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 404 break; 405 case TPACKET_V2: 406 h.h2->tp_status = status; 407 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 408 break; 409 case TPACKET_V3: 410 h.h3->tp_status = status; 411 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 412 break; 413 default: 414 WARN(1, "TPACKET version not supported.\n"); 415 BUG(); 416 } 417 418 smp_wmb(); 419 } 420 421 static int __packet_get_status(struct packet_sock *po, void *frame) 422 { 423 union tpacket_uhdr h; 424 425 smp_rmb(); 426 427 h.raw = frame; 428 switch (po->tp_version) { 429 case TPACKET_V1: 430 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 431 return h.h1->tp_status; 432 case TPACKET_V2: 433 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 434 return h.h2->tp_status; 435 case TPACKET_V3: 436 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 437 return h.h3->tp_status; 438 default: 439 WARN(1, "TPACKET version not supported.\n"); 440 BUG(); 441 return 0; 442 } 443 } 444 445 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, 446 unsigned int flags) 447 { 448 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 449 450 if (shhwtstamps && 451 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 452 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) 453 return TP_STATUS_TS_RAW_HARDWARE; 454 455 if (ktime_to_timespec_cond(skb->tstamp, ts)) 456 return TP_STATUS_TS_SOFTWARE; 457 458 return 0; 459 } 460 461 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 462 struct sk_buff *skb) 463 { 464 union tpacket_uhdr h; 465 struct timespec ts; 466 __u32 ts_status; 467 468 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 469 return 0; 470 471 h.raw = frame; 472 switch (po->tp_version) { 473 case TPACKET_V1: 474 h.h1->tp_sec = ts.tv_sec; 475 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 476 break; 477 case TPACKET_V2: 478 h.h2->tp_sec = ts.tv_sec; 479 h.h2->tp_nsec = ts.tv_nsec; 480 break; 481 case TPACKET_V3: 482 h.h3->tp_sec = ts.tv_sec; 483 h.h3->tp_nsec = ts.tv_nsec; 484 break; 485 default: 486 WARN(1, "TPACKET version not supported.\n"); 487 BUG(); 488 } 489 490 /* one flush is safe, as both fields always lie on the same cacheline */ 491 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 492 smp_wmb(); 493 494 return ts_status; 495 } 496 497 static void *packet_lookup_frame(struct packet_sock *po, 498 struct packet_ring_buffer *rb, 499 unsigned int position, 500 int status) 501 { 502 unsigned int pg_vec_pos, frame_offset; 503 union tpacket_uhdr h; 504 505 pg_vec_pos = position / rb->frames_per_block; 506 frame_offset = position % rb->frames_per_block; 507 508 h.raw = rb->pg_vec[pg_vec_pos].buffer + 509 (frame_offset * rb->frame_size); 510 511 if (status != __packet_get_status(po, h.raw)) 512 return NULL; 513 514 return h.raw; 515 } 516 517 static void *packet_current_frame(struct packet_sock *po, 518 struct packet_ring_buffer *rb, 519 int status) 520 { 521 return packet_lookup_frame(po, rb, rb->head, status); 522 } 523 524 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 525 { 526 del_timer_sync(&pkc->retire_blk_timer); 527 } 528 529 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 530 struct sk_buff_head *rb_queue) 531 { 532 struct tpacket_kbdq_core *pkc; 533 534 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 535 536 spin_lock_bh(&rb_queue->lock); 537 pkc->delete_blk_timer = 1; 538 spin_unlock_bh(&rb_queue->lock); 539 540 prb_del_retire_blk_timer(pkc); 541 } 542 543 static void prb_init_blk_timer(struct packet_sock *po, 544 struct tpacket_kbdq_core *pkc, 545 void (*func) (unsigned long)) 546 { 547 init_timer(&pkc->retire_blk_timer); 548 pkc->retire_blk_timer.data = (long)po; 549 pkc->retire_blk_timer.function = func; 550 pkc->retire_blk_timer.expires = jiffies; 551 } 552 553 static void prb_setup_retire_blk_timer(struct packet_sock *po) 554 { 555 struct tpacket_kbdq_core *pkc; 556 557 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 558 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired); 559 } 560 561 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 562 int blk_size_in_bytes) 563 { 564 struct net_device *dev; 565 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; 566 struct ethtool_link_ksettings ecmd; 567 int err; 568 569 rtnl_lock(); 570 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 571 if (unlikely(!dev)) { 572 rtnl_unlock(); 573 return DEFAULT_PRB_RETIRE_TOV; 574 } 575 err = __ethtool_get_link_ksettings(dev, &ecmd); 576 rtnl_unlock(); 577 if (!err) { 578 /* 579 * If the link speed is so slow you don't really 580 * need to worry about perf anyways 581 */ 582 if (ecmd.base.speed < SPEED_1000 || 583 ecmd.base.speed == SPEED_UNKNOWN) { 584 return DEFAULT_PRB_RETIRE_TOV; 585 } else { 586 msec = 1; 587 div = ecmd.base.speed / 1000; 588 } 589 } 590 591 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 592 593 if (div) 594 mbits /= div; 595 596 tmo = mbits * msec; 597 598 if (div) 599 return tmo+1; 600 return tmo; 601 } 602 603 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 604 union tpacket_req_u *req_u) 605 { 606 p1->feature_req_word = req_u->req3.tp_feature_req_word; 607 } 608 609 static void init_prb_bdqc(struct packet_sock *po, 610 struct packet_ring_buffer *rb, 611 struct pgv *pg_vec, 612 union tpacket_req_u *req_u) 613 { 614 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 615 struct tpacket_block_desc *pbd; 616 617 memset(p1, 0x0, sizeof(*p1)); 618 619 p1->knxt_seq_num = 1; 620 p1->pkbdq = pg_vec; 621 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 622 p1->pkblk_start = pg_vec[0].buffer; 623 p1->kblk_size = req_u->req3.tp_block_size; 624 p1->knum_blocks = req_u->req3.tp_block_nr; 625 p1->hdrlen = po->tp_hdrlen; 626 p1->version = po->tp_version; 627 p1->last_kactive_blk_num = 0; 628 po->stats.stats3.tp_freeze_q_cnt = 0; 629 if (req_u->req3.tp_retire_blk_tov) 630 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 631 else 632 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 633 req_u->req3.tp_block_size); 634 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 635 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 636 637 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 638 prb_init_ft_ops(p1, req_u); 639 prb_setup_retire_blk_timer(po); 640 prb_open_block(p1, pbd); 641 } 642 643 /* Do NOT update the last_blk_num first. 644 * Assumes sk_buff_head lock is held. 645 */ 646 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 647 { 648 mod_timer(&pkc->retire_blk_timer, 649 jiffies + pkc->tov_in_jiffies); 650 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 651 } 652 653 /* 654 * Timer logic: 655 * 1) We refresh the timer only when we open a block. 656 * By doing this we don't waste cycles refreshing the timer 657 * on packet-by-packet basis. 658 * 659 * With a 1MB block-size, on a 1Gbps line, it will take 660 * i) ~8 ms to fill a block + ii) memcpy etc. 661 * In this cut we are not accounting for the memcpy time. 662 * 663 * So, if the user sets the 'tmo' to 10ms then the timer 664 * will never fire while the block is still getting filled 665 * (which is what we want). However, the user could choose 666 * to close a block early and that's fine. 667 * 668 * But when the timer does fire, we check whether or not to refresh it. 669 * Since the tmo granularity is in msecs, it is not too expensive 670 * to refresh the timer, lets say every '8' msecs. 671 * Either the user can set the 'tmo' or we can derive it based on 672 * a) line-speed and b) block-size. 673 * prb_calc_retire_blk_tmo() calculates the tmo. 674 * 675 */ 676 static void prb_retire_rx_blk_timer_expired(unsigned long data) 677 { 678 struct packet_sock *po = (struct packet_sock *)data; 679 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 680 unsigned int frozen; 681 struct tpacket_block_desc *pbd; 682 683 spin_lock(&po->sk.sk_receive_queue.lock); 684 685 frozen = prb_queue_frozen(pkc); 686 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 687 688 if (unlikely(pkc->delete_blk_timer)) 689 goto out; 690 691 /* We only need to plug the race when the block is partially filled. 692 * tpacket_rcv: 693 * lock(); increment BLOCK_NUM_PKTS; unlock() 694 * copy_bits() is in progress ... 695 * timer fires on other cpu: 696 * we can't retire the current block because copy_bits 697 * is in progress. 698 * 699 */ 700 if (BLOCK_NUM_PKTS(pbd)) { 701 while (atomic_read(&pkc->blk_fill_in_prog)) { 702 /* Waiting for skb_copy_bits to finish... */ 703 cpu_relax(); 704 } 705 } 706 707 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 708 if (!frozen) { 709 if (!BLOCK_NUM_PKTS(pbd)) { 710 /* An empty block. Just refresh the timer. */ 711 goto refresh_timer; 712 } 713 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 714 if (!prb_dispatch_next_block(pkc, po)) 715 goto refresh_timer; 716 else 717 goto out; 718 } else { 719 /* Case 1. Queue was frozen because user-space was 720 * lagging behind. 721 */ 722 if (prb_curr_blk_in_use(pbd)) { 723 /* 724 * Ok, user-space is still behind. 725 * So just refresh the timer. 726 */ 727 goto refresh_timer; 728 } else { 729 /* Case 2. queue was frozen,user-space caught up, 730 * now the link went idle && the timer fired. 731 * We don't have a block to close.So we open this 732 * block and restart the timer. 733 * opening a block thaws the queue,restarts timer 734 * Thawing/timer-refresh is a side effect. 735 */ 736 prb_open_block(pkc, pbd); 737 goto out; 738 } 739 } 740 } 741 742 refresh_timer: 743 _prb_refresh_rx_retire_blk_timer(pkc); 744 745 out: 746 spin_unlock(&po->sk.sk_receive_queue.lock); 747 } 748 749 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 750 struct tpacket_block_desc *pbd1, __u32 status) 751 { 752 /* Flush everything minus the block header */ 753 754 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 755 u8 *start, *end; 756 757 start = (u8 *)pbd1; 758 759 /* Skip the block header(we know header WILL fit in 4K) */ 760 start += PAGE_SIZE; 761 762 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 763 for (; start < end; start += PAGE_SIZE) 764 flush_dcache_page(pgv_to_page(start)); 765 766 smp_wmb(); 767 #endif 768 769 /* Now update the block status. */ 770 771 BLOCK_STATUS(pbd1) = status; 772 773 /* Flush the block header */ 774 775 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 776 start = (u8 *)pbd1; 777 flush_dcache_page(pgv_to_page(start)); 778 779 smp_wmb(); 780 #endif 781 } 782 783 /* 784 * Side effect: 785 * 786 * 1) flush the block 787 * 2) Increment active_blk_num 788 * 789 * Note:We DONT refresh the timer on purpose. 790 * Because almost always the next block will be opened. 791 */ 792 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 793 struct tpacket_block_desc *pbd1, 794 struct packet_sock *po, unsigned int stat) 795 { 796 __u32 status = TP_STATUS_USER | stat; 797 798 struct tpacket3_hdr *last_pkt; 799 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 800 struct sock *sk = &po->sk; 801 802 if (po->stats.stats3.tp_drops) 803 status |= TP_STATUS_LOSING; 804 805 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 806 last_pkt->tp_next_offset = 0; 807 808 /* Get the ts of the last pkt */ 809 if (BLOCK_NUM_PKTS(pbd1)) { 810 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 811 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 812 } else { 813 /* Ok, we tmo'd - so get the current time. 814 * 815 * It shouldn't really happen as we don't close empty 816 * blocks. See prb_retire_rx_blk_timer_expired(). 817 */ 818 struct timespec ts; 819 getnstimeofday(&ts); 820 h1->ts_last_pkt.ts_sec = ts.tv_sec; 821 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 822 } 823 824 smp_wmb(); 825 826 /* Flush the block */ 827 prb_flush_block(pkc1, pbd1, status); 828 829 sk->sk_data_ready(sk); 830 831 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 832 } 833 834 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 835 { 836 pkc->reset_pending_on_curr_blk = 0; 837 } 838 839 /* 840 * Side effect of opening a block: 841 * 842 * 1) prb_queue is thawed. 843 * 2) retire_blk_timer is refreshed. 844 * 845 */ 846 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 847 struct tpacket_block_desc *pbd1) 848 { 849 struct timespec ts; 850 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 851 852 smp_rmb(); 853 854 /* We could have just memset this but we will lose the 855 * flexibility of making the priv area sticky 856 */ 857 858 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 859 BLOCK_NUM_PKTS(pbd1) = 0; 860 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 861 862 getnstimeofday(&ts); 863 864 h1->ts_first_pkt.ts_sec = ts.tv_sec; 865 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 866 867 pkc1->pkblk_start = (char *)pbd1; 868 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 869 870 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 871 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 872 873 pbd1->version = pkc1->version; 874 pkc1->prev = pkc1->nxt_offset; 875 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 876 877 prb_thaw_queue(pkc1); 878 _prb_refresh_rx_retire_blk_timer(pkc1); 879 880 smp_wmb(); 881 } 882 883 /* 884 * Queue freeze logic: 885 * 1) Assume tp_block_nr = 8 blocks. 886 * 2) At time 't0', user opens Rx ring. 887 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 888 * 4) user-space is either sleeping or processing block '0'. 889 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 890 * it will close block-7,loop around and try to fill block '0'. 891 * call-flow: 892 * __packet_lookup_frame_in_block 893 * prb_retire_current_block() 894 * prb_dispatch_next_block() 895 * |->(BLOCK_STATUS == USER) evaluates to true 896 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 897 * 6) Now there are two cases: 898 * 6.1) Link goes idle right after the queue is frozen. 899 * But remember, the last open_block() refreshed the timer. 900 * When this timer expires,it will refresh itself so that we can 901 * re-open block-0 in near future. 902 * 6.2) Link is busy and keeps on receiving packets. This is a simple 903 * case and __packet_lookup_frame_in_block will check if block-0 904 * is free and can now be re-used. 905 */ 906 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 907 struct packet_sock *po) 908 { 909 pkc->reset_pending_on_curr_blk = 1; 910 po->stats.stats3.tp_freeze_q_cnt++; 911 } 912 913 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 914 915 /* 916 * If the next block is free then we will dispatch it 917 * and return a good offset. 918 * Else, we will freeze the queue. 919 * So, caller must check the return value. 920 */ 921 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 922 struct packet_sock *po) 923 { 924 struct tpacket_block_desc *pbd; 925 926 smp_rmb(); 927 928 /* 1. Get current block num */ 929 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 930 931 /* 2. If this block is currently in_use then freeze the queue */ 932 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 933 prb_freeze_queue(pkc, po); 934 return NULL; 935 } 936 937 /* 938 * 3. 939 * open this block and return the offset where the first packet 940 * needs to get stored. 941 */ 942 prb_open_block(pkc, pbd); 943 return (void *)pkc->nxt_offset; 944 } 945 946 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 947 struct packet_sock *po, unsigned int status) 948 { 949 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 950 951 /* retire/close the current block */ 952 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 953 /* 954 * Plug the case where copy_bits() is in progress on 955 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 956 * have space to copy the pkt in the current block and 957 * called prb_retire_current_block() 958 * 959 * We don't need to worry about the TMO case because 960 * the timer-handler already handled this case. 961 */ 962 if (!(status & TP_STATUS_BLK_TMO)) { 963 while (atomic_read(&pkc->blk_fill_in_prog)) { 964 /* Waiting for skb_copy_bits to finish... */ 965 cpu_relax(); 966 } 967 } 968 prb_close_block(pkc, pbd, po, status); 969 return; 970 } 971 } 972 973 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) 974 { 975 return TP_STATUS_USER & BLOCK_STATUS(pbd); 976 } 977 978 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 979 { 980 return pkc->reset_pending_on_curr_blk; 981 } 982 983 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 984 { 985 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 986 atomic_dec(&pkc->blk_fill_in_prog); 987 } 988 989 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 990 struct tpacket3_hdr *ppd) 991 { 992 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 993 } 994 995 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 996 struct tpacket3_hdr *ppd) 997 { 998 ppd->hv1.tp_rxhash = 0; 999 } 1000 1001 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 1002 struct tpacket3_hdr *ppd) 1003 { 1004 if (skb_vlan_tag_present(pkc->skb)) { 1005 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 1006 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 1007 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 1008 } else { 1009 ppd->hv1.tp_vlan_tci = 0; 1010 ppd->hv1.tp_vlan_tpid = 0; 1011 ppd->tp_status = TP_STATUS_AVAILABLE; 1012 } 1013 } 1014 1015 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 1016 struct tpacket3_hdr *ppd) 1017 { 1018 ppd->hv1.tp_padding = 0; 1019 prb_fill_vlan_info(pkc, ppd); 1020 1021 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 1022 prb_fill_rxhash(pkc, ppd); 1023 else 1024 prb_clear_rxhash(pkc, ppd); 1025 } 1026 1027 static void prb_fill_curr_block(char *curr, 1028 struct tpacket_kbdq_core *pkc, 1029 struct tpacket_block_desc *pbd, 1030 unsigned int len) 1031 { 1032 struct tpacket3_hdr *ppd; 1033 1034 ppd = (struct tpacket3_hdr *)curr; 1035 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 1036 pkc->prev = curr; 1037 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 1038 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1039 BLOCK_NUM_PKTS(pbd) += 1; 1040 atomic_inc(&pkc->blk_fill_in_prog); 1041 prb_run_all_ft_ops(pkc, ppd); 1042 } 1043 1044 /* Assumes caller has the sk->rx_queue.lock */ 1045 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1046 struct sk_buff *skb, 1047 int status, 1048 unsigned int len 1049 ) 1050 { 1051 struct tpacket_kbdq_core *pkc; 1052 struct tpacket_block_desc *pbd; 1053 char *curr, *end; 1054 1055 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1056 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1057 1058 /* Queue is frozen when user space is lagging behind */ 1059 if (prb_queue_frozen(pkc)) { 1060 /* 1061 * Check if that last block which caused the queue to freeze, 1062 * is still in_use by user-space. 1063 */ 1064 if (prb_curr_blk_in_use(pbd)) { 1065 /* Can't record this packet */ 1066 return NULL; 1067 } else { 1068 /* 1069 * Ok, the block was released by user-space. 1070 * Now let's open that block. 1071 * opening a block also thaws the queue. 1072 * Thawing is a side effect. 1073 */ 1074 prb_open_block(pkc, pbd); 1075 } 1076 } 1077 1078 smp_mb(); 1079 curr = pkc->nxt_offset; 1080 pkc->skb = skb; 1081 end = (char *)pbd + pkc->kblk_size; 1082 1083 /* first try the current block */ 1084 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1085 prb_fill_curr_block(curr, pkc, pbd, len); 1086 return (void *)curr; 1087 } 1088 1089 /* Ok, close the current block */ 1090 prb_retire_current_block(pkc, po, 0); 1091 1092 /* Now, try to dispatch the next block */ 1093 curr = (char *)prb_dispatch_next_block(pkc, po); 1094 if (curr) { 1095 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1096 prb_fill_curr_block(curr, pkc, pbd, len); 1097 return (void *)curr; 1098 } 1099 1100 /* 1101 * No free blocks are available.user_space hasn't caught up yet. 1102 * Queue was just frozen and now this packet will get dropped. 1103 */ 1104 return NULL; 1105 } 1106 1107 static void *packet_current_rx_frame(struct packet_sock *po, 1108 struct sk_buff *skb, 1109 int status, unsigned int len) 1110 { 1111 char *curr = NULL; 1112 switch (po->tp_version) { 1113 case TPACKET_V1: 1114 case TPACKET_V2: 1115 curr = packet_lookup_frame(po, &po->rx_ring, 1116 po->rx_ring.head, status); 1117 return curr; 1118 case TPACKET_V3: 1119 return __packet_lookup_frame_in_block(po, skb, status, len); 1120 default: 1121 WARN(1, "TPACKET version not supported\n"); 1122 BUG(); 1123 return NULL; 1124 } 1125 } 1126 1127 static void *prb_lookup_block(struct packet_sock *po, 1128 struct packet_ring_buffer *rb, 1129 unsigned int idx, 1130 int status) 1131 { 1132 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1133 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1134 1135 if (status != BLOCK_STATUS(pbd)) 1136 return NULL; 1137 return pbd; 1138 } 1139 1140 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1141 { 1142 unsigned int prev; 1143 if (rb->prb_bdqc.kactive_blk_num) 1144 prev = rb->prb_bdqc.kactive_blk_num-1; 1145 else 1146 prev = rb->prb_bdqc.knum_blocks-1; 1147 return prev; 1148 } 1149 1150 /* Assumes caller has held the rx_queue.lock */ 1151 static void *__prb_previous_block(struct packet_sock *po, 1152 struct packet_ring_buffer *rb, 1153 int status) 1154 { 1155 unsigned int previous = prb_previous_blk_num(rb); 1156 return prb_lookup_block(po, rb, previous, status); 1157 } 1158 1159 static void *packet_previous_rx_frame(struct packet_sock *po, 1160 struct packet_ring_buffer *rb, 1161 int status) 1162 { 1163 if (po->tp_version <= TPACKET_V2) 1164 return packet_previous_frame(po, rb, status); 1165 1166 return __prb_previous_block(po, rb, status); 1167 } 1168 1169 static void packet_increment_rx_head(struct packet_sock *po, 1170 struct packet_ring_buffer *rb) 1171 { 1172 switch (po->tp_version) { 1173 case TPACKET_V1: 1174 case TPACKET_V2: 1175 return packet_increment_head(rb); 1176 case TPACKET_V3: 1177 default: 1178 WARN(1, "TPACKET version not supported.\n"); 1179 BUG(); 1180 return; 1181 } 1182 } 1183 1184 static void *packet_previous_frame(struct packet_sock *po, 1185 struct packet_ring_buffer *rb, 1186 int status) 1187 { 1188 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1189 return packet_lookup_frame(po, rb, previous, status); 1190 } 1191 1192 static void packet_increment_head(struct packet_ring_buffer *buff) 1193 { 1194 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1195 } 1196 1197 static void packet_inc_pending(struct packet_ring_buffer *rb) 1198 { 1199 this_cpu_inc(*rb->pending_refcnt); 1200 } 1201 1202 static void packet_dec_pending(struct packet_ring_buffer *rb) 1203 { 1204 this_cpu_dec(*rb->pending_refcnt); 1205 } 1206 1207 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1208 { 1209 unsigned int refcnt = 0; 1210 int cpu; 1211 1212 /* We don't use pending refcount in rx_ring. */ 1213 if (rb->pending_refcnt == NULL) 1214 return 0; 1215 1216 for_each_possible_cpu(cpu) 1217 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1218 1219 return refcnt; 1220 } 1221 1222 static int packet_alloc_pending(struct packet_sock *po) 1223 { 1224 po->rx_ring.pending_refcnt = NULL; 1225 1226 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1227 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1228 return -ENOBUFS; 1229 1230 return 0; 1231 } 1232 1233 static void packet_free_pending(struct packet_sock *po) 1234 { 1235 free_percpu(po->tx_ring.pending_refcnt); 1236 } 1237 1238 #define ROOM_POW_OFF 2 1239 #define ROOM_NONE 0x0 1240 #define ROOM_LOW 0x1 1241 #define ROOM_NORMAL 0x2 1242 1243 static bool __tpacket_has_room(struct packet_sock *po, int pow_off) 1244 { 1245 int idx, len; 1246 1247 len = po->rx_ring.frame_max + 1; 1248 idx = po->rx_ring.head; 1249 if (pow_off) 1250 idx += len >> pow_off; 1251 if (idx >= len) 1252 idx -= len; 1253 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1254 } 1255 1256 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off) 1257 { 1258 int idx, len; 1259 1260 len = po->rx_ring.prb_bdqc.knum_blocks; 1261 idx = po->rx_ring.prb_bdqc.kactive_blk_num; 1262 if (pow_off) 1263 idx += len >> pow_off; 1264 if (idx >= len) 1265 idx -= len; 1266 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1267 } 1268 1269 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1270 { 1271 struct sock *sk = &po->sk; 1272 int ret = ROOM_NONE; 1273 1274 if (po->prot_hook.func != tpacket_rcv) { 1275 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1276 - (skb ? skb->truesize : 0); 1277 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF)) 1278 return ROOM_NORMAL; 1279 else if (avail > 0) 1280 return ROOM_LOW; 1281 else 1282 return ROOM_NONE; 1283 } 1284 1285 if (po->tp_version == TPACKET_V3) { 1286 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1287 ret = ROOM_NORMAL; 1288 else if (__tpacket_v3_has_room(po, 0)) 1289 ret = ROOM_LOW; 1290 } else { 1291 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1292 ret = ROOM_NORMAL; 1293 else if (__tpacket_has_room(po, 0)) 1294 ret = ROOM_LOW; 1295 } 1296 1297 return ret; 1298 } 1299 1300 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1301 { 1302 int ret; 1303 bool has_room; 1304 1305 spin_lock_bh(&po->sk.sk_receive_queue.lock); 1306 ret = __packet_rcv_has_room(po, skb); 1307 has_room = ret == ROOM_NORMAL; 1308 if (po->pressure == has_room) 1309 po->pressure = !has_room; 1310 spin_unlock_bh(&po->sk.sk_receive_queue.lock); 1311 1312 return ret; 1313 } 1314 1315 static void packet_sock_destruct(struct sock *sk) 1316 { 1317 skb_queue_purge(&sk->sk_error_queue); 1318 1319 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1320 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 1321 1322 if (!sock_flag(sk, SOCK_DEAD)) { 1323 pr_err("Attempt to release alive packet socket: %p\n", sk); 1324 return; 1325 } 1326 1327 sk_refcnt_debug_dec(sk); 1328 } 1329 1330 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1331 { 1332 u32 rxhash; 1333 int i, count = 0; 1334 1335 rxhash = skb_get_hash(skb); 1336 for (i = 0; i < ROLLOVER_HLEN; i++) 1337 if (po->rollover->history[i] == rxhash) 1338 count++; 1339 1340 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash; 1341 return count > (ROLLOVER_HLEN >> 1); 1342 } 1343 1344 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1345 struct sk_buff *skb, 1346 unsigned int num) 1347 { 1348 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1349 } 1350 1351 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1352 struct sk_buff *skb, 1353 unsigned int num) 1354 { 1355 unsigned int val = atomic_inc_return(&f->rr_cur); 1356 1357 return val % num; 1358 } 1359 1360 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1361 struct sk_buff *skb, 1362 unsigned int num) 1363 { 1364 return smp_processor_id() % num; 1365 } 1366 1367 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1368 struct sk_buff *skb, 1369 unsigned int num) 1370 { 1371 return prandom_u32_max(num); 1372 } 1373 1374 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1375 struct sk_buff *skb, 1376 unsigned int idx, bool try_self, 1377 unsigned int num) 1378 { 1379 struct packet_sock *po, *po_next, *po_skip = NULL; 1380 unsigned int i, j, room = ROOM_NONE; 1381 1382 po = pkt_sk(f->arr[idx]); 1383 1384 if (try_self) { 1385 room = packet_rcv_has_room(po, skb); 1386 if (room == ROOM_NORMAL || 1387 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1388 return idx; 1389 po_skip = po; 1390 } 1391 1392 i = j = min_t(int, po->rollover->sock, num - 1); 1393 do { 1394 po_next = pkt_sk(f->arr[i]); 1395 if (po_next != po_skip && !po_next->pressure && 1396 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1397 if (i != j) 1398 po->rollover->sock = i; 1399 atomic_long_inc(&po->rollover->num); 1400 if (room == ROOM_LOW) 1401 atomic_long_inc(&po->rollover->num_huge); 1402 return i; 1403 } 1404 1405 if (++i == num) 1406 i = 0; 1407 } while (i != j); 1408 1409 atomic_long_inc(&po->rollover->num_failed); 1410 return idx; 1411 } 1412 1413 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1414 struct sk_buff *skb, 1415 unsigned int num) 1416 { 1417 return skb_get_queue_mapping(skb) % num; 1418 } 1419 1420 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1421 struct sk_buff *skb, 1422 unsigned int num) 1423 { 1424 struct bpf_prog *prog; 1425 unsigned int ret = 0; 1426 1427 rcu_read_lock(); 1428 prog = rcu_dereference(f->bpf_prog); 1429 if (prog) 1430 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1431 rcu_read_unlock(); 1432 1433 return ret; 1434 } 1435 1436 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1437 { 1438 return f->flags & (flag >> 8); 1439 } 1440 1441 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1442 struct packet_type *pt, struct net_device *orig_dev) 1443 { 1444 struct packet_fanout *f = pt->af_packet_priv; 1445 unsigned int num = READ_ONCE(f->num_members); 1446 struct net *net = read_pnet(&f->net); 1447 struct packet_sock *po; 1448 unsigned int idx; 1449 1450 if (!net_eq(dev_net(dev), net) || !num) { 1451 kfree_skb(skb); 1452 return 0; 1453 } 1454 1455 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1456 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1457 if (!skb) 1458 return 0; 1459 } 1460 switch (f->type) { 1461 case PACKET_FANOUT_HASH: 1462 default: 1463 idx = fanout_demux_hash(f, skb, num); 1464 break; 1465 case PACKET_FANOUT_LB: 1466 idx = fanout_demux_lb(f, skb, num); 1467 break; 1468 case PACKET_FANOUT_CPU: 1469 idx = fanout_demux_cpu(f, skb, num); 1470 break; 1471 case PACKET_FANOUT_RND: 1472 idx = fanout_demux_rnd(f, skb, num); 1473 break; 1474 case PACKET_FANOUT_QM: 1475 idx = fanout_demux_qm(f, skb, num); 1476 break; 1477 case PACKET_FANOUT_ROLLOVER: 1478 idx = fanout_demux_rollover(f, skb, 0, false, num); 1479 break; 1480 case PACKET_FANOUT_CBPF: 1481 case PACKET_FANOUT_EBPF: 1482 idx = fanout_demux_bpf(f, skb, num); 1483 break; 1484 } 1485 1486 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1487 idx = fanout_demux_rollover(f, skb, idx, true, num); 1488 1489 po = pkt_sk(f->arr[idx]); 1490 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1491 } 1492 1493 DEFINE_MUTEX(fanout_mutex); 1494 EXPORT_SYMBOL_GPL(fanout_mutex); 1495 static LIST_HEAD(fanout_list); 1496 static u16 fanout_next_id; 1497 1498 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1499 { 1500 struct packet_fanout *f = po->fanout; 1501 1502 spin_lock(&f->lock); 1503 f->arr[f->num_members] = sk; 1504 smp_wmb(); 1505 f->num_members++; 1506 if (f->num_members == 1) 1507 dev_add_pack(&f->prot_hook); 1508 spin_unlock(&f->lock); 1509 } 1510 1511 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1512 { 1513 struct packet_fanout *f = po->fanout; 1514 int i; 1515 1516 spin_lock(&f->lock); 1517 for (i = 0; i < f->num_members; i++) { 1518 if (f->arr[i] == sk) 1519 break; 1520 } 1521 BUG_ON(i >= f->num_members); 1522 f->arr[i] = f->arr[f->num_members - 1]; 1523 f->num_members--; 1524 if (f->num_members == 0) 1525 __dev_remove_pack(&f->prot_hook); 1526 spin_unlock(&f->lock); 1527 } 1528 1529 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1530 { 1531 if (sk->sk_family != PF_PACKET) 1532 return false; 1533 1534 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1535 } 1536 1537 static void fanout_init_data(struct packet_fanout *f) 1538 { 1539 switch (f->type) { 1540 case PACKET_FANOUT_LB: 1541 atomic_set(&f->rr_cur, 0); 1542 break; 1543 case PACKET_FANOUT_CBPF: 1544 case PACKET_FANOUT_EBPF: 1545 RCU_INIT_POINTER(f->bpf_prog, NULL); 1546 break; 1547 } 1548 } 1549 1550 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1551 { 1552 struct bpf_prog *old; 1553 1554 spin_lock(&f->lock); 1555 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1556 rcu_assign_pointer(f->bpf_prog, new); 1557 spin_unlock(&f->lock); 1558 1559 if (old) { 1560 synchronize_net(); 1561 bpf_prog_destroy(old); 1562 } 1563 } 1564 1565 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data, 1566 unsigned int len) 1567 { 1568 struct bpf_prog *new; 1569 struct sock_fprog fprog; 1570 int ret; 1571 1572 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1573 return -EPERM; 1574 if (len != sizeof(fprog)) 1575 return -EINVAL; 1576 if (copy_from_user(&fprog, data, len)) 1577 return -EFAULT; 1578 1579 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1580 if (ret) 1581 return ret; 1582 1583 __fanout_set_data_bpf(po->fanout, new); 1584 return 0; 1585 } 1586 1587 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data, 1588 unsigned int len) 1589 { 1590 struct bpf_prog *new; 1591 u32 fd; 1592 1593 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1594 return -EPERM; 1595 if (len != sizeof(fd)) 1596 return -EINVAL; 1597 if (copy_from_user(&fd, data, len)) 1598 return -EFAULT; 1599 1600 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1601 if (IS_ERR(new)) 1602 return PTR_ERR(new); 1603 1604 __fanout_set_data_bpf(po->fanout, new); 1605 return 0; 1606 } 1607 1608 static int fanout_set_data(struct packet_sock *po, char __user *data, 1609 unsigned int len) 1610 { 1611 switch (po->fanout->type) { 1612 case PACKET_FANOUT_CBPF: 1613 return fanout_set_data_cbpf(po, data, len); 1614 case PACKET_FANOUT_EBPF: 1615 return fanout_set_data_ebpf(po, data, len); 1616 default: 1617 return -EINVAL; 1618 }; 1619 } 1620 1621 static void fanout_release_data(struct packet_fanout *f) 1622 { 1623 switch (f->type) { 1624 case PACKET_FANOUT_CBPF: 1625 case PACKET_FANOUT_EBPF: 1626 __fanout_set_data_bpf(f, NULL); 1627 }; 1628 } 1629 1630 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) 1631 { 1632 struct packet_fanout *f; 1633 1634 list_for_each_entry(f, &fanout_list, list) { 1635 if (f->id == candidate_id && 1636 read_pnet(&f->net) == sock_net(sk)) { 1637 return false; 1638 } 1639 } 1640 return true; 1641 } 1642 1643 static bool fanout_find_new_id(struct sock *sk, u16 *new_id) 1644 { 1645 u16 id = fanout_next_id; 1646 1647 do { 1648 if (__fanout_id_is_free(sk, id)) { 1649 *new_id = id; 1650 fanout_next_id = id + 1; 1651 return true; 1652 } 1653 1654 id++; 1655 } while (id != fanout_next_id); 1656 1657 return false; 1658 } 1659 1660 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1661 { 1662 struct packet_rollover *rollover = NULL; 1663 struct packet_sock *po = pkt_sk(sk); 1664 struct packet_fanout *f, *match; 1665 u8 type = type_flags & 0xff; 1666 u8 flags = type_flags >> 8; 1667 int err; 1668 1669 switch (type) { 1670 case PACKET_FANOUT_ROLLOVER: 1671 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1672 return -EINVAL; 1673 case PACKET_FANOUT_HASH: 1674 case PACKET_FANOUT_LB: 1675 case PACKET_FANOUT_CPU: 1676 case PACKET_FANOUT_RND: 1677 case PACKET_FANOUT_QM: 1678 case PACKET_FANOUT_CBPF: 1679 case PACKET_FANOUT_EBPF: 1680 break; 1681 default: 1682 return -EINVAL; 1683 } 1684 1685 mutex_lock(&fanout_mutex); 1686 1687 err = -EALREADY; 1688 if (po->fanout) 1689 goto out; 1690 1691 if (type == PACKET_FANOUT_ROLLOVER || 1692 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1693 err = -ENOMEM; 1694 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); 1695 if (!rollover) 1696 goto out; 1697 atomic_long_set(&rollover->num, 0); 1698 atomic_long_set(&rollover->num_huge, 0); 1699 atomic_long_set(&rollover->num_failed, 0); 1700 po->rollover = rollover; 1701 } 1702 1703 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { 1704 if (id != 0) { 1705 err = -EINVAL; 1706 goto out; 1707 } 1708 if (!fanout_find_new_id(sk, &id)) { 1709 err = -ENOMEM; 1710 goto out; 1711 } 1712 /* ephemeral flag for the first socket in the group: drop it */ 1713 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); 1714 } 1715 1716 match = NULL; 1717 list_for_each_entry(f, &fanout_list, list) { 1718 if (f->id == id && 1719 read_pnet(&f->net) == sock_net(sk)) { 1720 match = f; 1721 break; 1722 } 1723 } 1724 err = -EINVAL; 1725 if (match && match->flags != flags) 1726 goto out; 1727 if (!match) { 1728 err = -ENOMEM; 1729 match = kzalloc(sizeof(*match), GFP_KERNEL); 1730 if (!match) 1731 goto out; 1732 write_pnet(&match->net, sock_net(sk)); 1733 match->id = id; 1734 match->type = type; 1735 match->flags = flags; 1736 INIT_LIST_HEAD(&match->list); 1737 spin_lock_init(&match->lock); 1738 refcount_set(&match->sk_ref, 0); 1739 fanout_init_data(match); 1740 match->prot_hook.type = po->prot_hook.type; 1741 match->prot_hook.dev = po->prot_hook.dev; 1742 match->prot_hook.func = packet_rcv_fanout; 1743 match->prot_hook.af_packet_priv = match; 1744 match->prot_hook.id_match = match_fanout_group; 1745 list_add(&match->list, &fanout_list); 1746 } 1747 err = -EINVAL; 1748 1749 spin_lock(&po->bind_lock); 1750 if (po->running && 1751 match->type == type && 1752 match->prot_hook.type == po->prot_hook.type && 1753 match->prot_hook.dev == po->prot_hook.dev) { 1754 err = -ENOSPC; 1755 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1756 __dev_remove_pack(&po->prot_hook); 1757 po->fanout = match; 1758 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); 1759 __fanout_link(sk, po); 1760 err = 0; 1761 } 1762 } 1763 spin_unlock(&po->bind_lock); 1764 1765 if (err && !refcount_read(&match->sk_ref)) { 1766 list_del(&match->list); 1767 kfree(match); 1768 } 1769 1770 out: 1771 if (err && rollover) { 1772 kfree_rcu(rollover, rcu); 1773 po->rollover = NULL; 1774 } 1775 mutex_unlock(&fanout_mutex); 1776 return err; 1777 } 1778 1779 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes 1780 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. 1781 * It is the responsibility of the caller to call fanout_release_data() and 1782 * free the returned packet_fanout (after synchronize_net()) 1783 */ 1784 static struct packet_fanout *fanout_release(struct sock *sk) 1785 { 1786 struct packet_sock *po = pkt_sk(sk); 1787 struct packet_fanout *f; 1788 1789 mutex_lock(&fanout_mutex); 1790 f = po->fanout; 1791 if (f) { 1792 po->fanout = NULL; 1793 1794 if (refcount_dec_and_test(&f->sk_ref)) 1795 list_del(&f->list); 1796 else 1797 f = NULL; 1798 1799 if (po->rollover) { 1800 kfree_rcu(po->rollover, rcu); 1801 po->rollover = NULL; 1802 } 1803 } 1804 mutex_unlock(&fanout_mutex); 1805 1806 return f; 1807 } 1808 1809 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1810 struct sk_buff *skb) 1811 { 1812 /* Earlier code assumed this would be a VLAN pkt, double-check 1813 * this now that we have the actual packet in hand. We can only 1814 * do this check on Ethernet devices. 1815 */ 1816 if (unlikely(dev->type != ARPHRD_ETHER)) 1817 return false; 1818 1819 skb_reset_mac_header(skb); 1820 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1821 } 1822 1823 static const struct proto_ops packet_ops; 1824 1825 static const struct proto_ops packet_ops_spkt; 1826 1827 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1828 struct packet_type *pt, struct net_device *orig_dev) 1829 { 1830 struct sock *sk; 1831 struct sockaddr_pkt *spkt; 1832 1833 /* 1834 * When we registered the protocol we saved the socket in the data 1835 * field for just this event. 1836 */ 1837 1838 sk = pt->af_packet_priv; 1839 1840 /* 1841 * Yank back the headers [hope the device set this 1842 * right or kerboom...] 1843 * 1844 * Incoming packets have ll header pulled, 1845 * push it back. 1846 * 1847 * For outgoing ones skb->data == skb_mac_header(skb) 1848 * so that this procedure is noop. 1849 */ 1850 1851 if (skb->pkt_type == PACKET_LOOPBACK) 1852 goto out; 1853 1854 if (!net_eq(dev_net(dev), sock_net(sk))) 1855 goto out; 1856 1857 skb = skb_share_check(skb, GFP_ATOMIC); 1858 if (skb == NULL) 1859 goto oom; 1860 1861 /* drop any routing info */ 1862 skb_dst_drop(skb); 1863 1864 /* drop conntrack reference */ 1865 nf_reset(skb); 1866 1867 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1868 1869 skb_push(skb, skb->data - skb_mac_header(skb)); 1870 1871 /* 1872 * The SOCK_PACKET socket receives _all_ frames. 1873 */ 1874 1875 spkt->spkt_family = dev->type; 1876 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1877 spkt->spkt_protocol = skb->protocol; 1878 1879 /* 1880 * Charge the memory to the socket. This is done specifically 1881 * to prevent sockets using all the memory up. 1882 */ 1883 1884 if (sock_queue_rcv_skb(sk, skb) == 0) 1885 return 0; 1886 1887 out: 1888 kfree_skb(skb); 1889 oom: 1890 return 0; 1891 } 1892 1893 1894 /* 1895 * Output a raw packet to a device layer. This bypasses all the other 1896 * protocol layers and you must therefore supply it with a complete frame 1897 */ 1898 1899 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1900 size_t len) 1901 { 1902 struct sock *sk = sock->sk; 1903 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1904 struct sk_buff *skb = NULL; 1905 struct net_device *dev; 1906 struct sockcm_cookie sockc; 1907 __be16 proto = 0; 1908 int err; 1909 int extra_len = 0; 1910 1911 /* 1912 * Get and verify the address. 1913 */ 1914 1915 if (saddr) { 1916 if (msg->msg_namelen < sizeof(struct sockaddr)) 1917 return -EINVAL; 1918 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1919 proto = saddr->spkt_protocol; 1920 } else 1921 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1922 1923 /* 1924 * Find the device first to size check it 1925 */ 1926 1927 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1928 retry: 1929 rcu_read_lock(); 1930 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1931 err = -ENODEV; 1932 if (dev == NULL) 1933 goto out_unlock; 1934 1935 err = -ENETDOWN; 1936 if (!(dev->flags & IFF_UP)) 1937 goto out_unlock; 1938 1939 /* 1940 * You may not queue a frame bigger than the mtu. This is the lowest level 1941 * raw protocol and you must do your own fragmentation at this level. 1942 */ 1943 1944 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1945 if (!netif_supports_nofcs(dev)) { 1946 err = -EPROTONOSUPPORT; 1947 goto out_unlock; 1948 } 1949 extra_len = 4; /* We're doing our own CRC */ 1950 } 1951 1952 err = -EMSGSIZE; 1953 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1954 goto out_unlock; 1955 1956 if (!skb) { 1957 size_t reserved = LL_RESERVED_SPACE(dev); 1958 int tlen = dev->needed_tailroom; 1959 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1960 1961 rcu_read_unlock(); 1962 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1963 if (skb == NULL) 1964 return -ENOBUFS; 1965 /* FIXME: Save some space for broken drivers that write a hard 1966 * header at transmission time by themselves. PPP is the notable 1967 * one here. This should really be fixed at the driver level. 1968 */ 1969 skb_reserve(skb, reserved); 1970 skb_reset_network_header(skb); 1971 1972 /* Try to align data part correctly */ 1973 if (hhlen) { 1974 skb->data -= hhlen; 1975 skb->tail -= hhlen; 1976 if (len < hhlen) 1977 skb_reset_network_header(skb); 1978 } 1979 err = memcpy_from_msg(skb_put(skb, len), msg, len); 1980 if (err) 1981 goto out_free; 1982 goto retry; 1983 } 1984 1985 if (!dev_validate_header(dev, skb->data, len)) { 1986 err = -EINVAL; 1987 goto out_unlock; 1988 } 1989 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 1990 !packet_extra_vlan_len_allowed(dev, skb)) { 1991 err = -EMSGSIZE; 1992 goto out_unlock; 1993 } 1994 1995 sockc.tsflags = sk->sk_tsflags; 1996 if (msg->msg_controllen) { 1997 err = sock_cmsg_send(sk, msg, &sockc); 1998 if (unlikely(err)) 1999 goto out_unlock; 2000 } 2001 2002 skb->protocol = proto; 2003 skb->dev = dev; 2004 skb->priority = sk->sk_priority; 2005 skb->mark = sk->sk_mark; 2006 2007 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags); 2008 2009 if (unlikely(extra_len == 4)) 2010 skb->no_fcs = 1; 2011 2012 skb_probe_transport_header(skb, 0); 2013 2014 dev_queue_xmit(skb); 2015 rcu_read_unlock(); 2016 return len; 2017 2018 out_unlock: 2019 rcu_read_unlock(); 2020 out_free: 2021 kfree_skb(skb); 2022 return err; 2023 } 2024 2025 static unsigned int run_filter(struct sk_buff *skb, 2026 const struct sock *sk, 2027 unsigned int res) 2028 { 2029 struct sk_filter *filter; 2030 2031 rcu_read_lock(); 2032 filter = rcu_dereference(sk->sk_filter); 2033 if (filter != NULL) 2034 res = bpf_prog_run_clear_cb(filter->prog, skb); 2035 rcu_read_unlock(); 2036 2037 return res; 2038 } 2039 2040 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2041 size_t *len) 2042 { 2043 struct virtio_net_hdr vnet_hdr; 2044 2045 if (*len < sizeof(vnet_hdr)) 2046 return -EINVAL; 2047 *len -= sizeof(vnet_hdr); 2048 2049 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true)) 2050 return -EINVAL; 2051 2052 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 2053 } 2054 2055 /* 2056 * This function makes lazy skb cloning in hope that most of packets 2057 * are discarded by BPF. 2058 * 2059 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2060 * and skb->cb are mangled. It works because (and until) packets 2061 * falling here are owned by current CPU. Output packets are cloned 2062 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2063 * sequencially, so that if we return skb to original state on exit, 2064 * we will not harm anyone. 2065 */ 2066 2067 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2068 struct packet_type *pt, struct net_device *orig_dev) 2069 { 2070 struct sock *sk; 2071 struct sockaddr_ll *sll; 2072 struct packet_sock *po; 2073 u8 *skb_head = skb->data; 2074 int skb_len = skb->len; 2075 unsigned int snaplen, res; 2076 bool is_drop_n_account = false; 2077 2078 if (skb->pkt_type == PACKET_LOOPBACK) 2079 goto drop; 2080 2081 sk = pt->af_packet_priv; 2082 po = pkt_sk(sk); 2083 2084 if (!net_eq(dev_net(dev), sock_net(sk))) 2085 goto drop; 2086 2087 skb->dev = dev; 2088 2089 if (dev->header_ops) { 2090 /* The device has an explicit notion of ll header, 2091 * exported to higher levels. 2092 * 2093 * Otherwise, the device hides details of its frame 2094 * structure, so that corresponding packet head is 2095 * never delivered to user. 2096 */ 2097 if (sk->sk_type != SOCK_DGRAM) 2098 skb_push(skb, skb->data - skb_mac_header(skb)); 2099 else if (skb->pkt_type == PACKET_OUTGOING) { 2100 /* Special case: outgoing packets have ll header at head */ 2101 skb_pull(skb, skb_network_offset(skb)); 2102 } 2103 } 2104 2105 snaplen = skb->len; 2106 2107 res = run_filter(skb, sk, snaplen); 2108 if (!res) 2109 goto drop_n_restore; 2110 if (snaplen > res) 2111 snaplen = res; 2112 2113 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2114 goto drop_n_acct; 2115 2116 if (skb_shared(skb)) { 2117 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2118 if (nskb == NULL) 2119 goto drop_n_acct; 2120 2121 if (skb_head != skb->data) { 2122 skb->data = skb_head; 2123 skb->len = skb_len; 2124 } 2125 consume_skb(skb); 2126 skb = nskb; 2127 } 2128 2129 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2130 2131 sll = &PACKET_SKB_CB(skb)->sa.ll; 2132 sll->sll_hatype = dev->type; 2133 sll->sll_pkttype = skb->pkt_type; 2134 if (unlikely(po->origdev)) 2135 sll->sll_ifindex = orig_dev->ifindex; 2136 else 2137 sll->sll_ifindex = dev->ifindex; 2138 2139 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2140 2141 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2142 * Use their space for storing the original skb length. 2143 */ 2144 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2145 2146 if (pskb_trim(skb, snaplen)) 2147 goto drop_n_acct; 2148 2149 skb_set_owner_r(skb, sk); 2150 skb->dev = NULL; 2151 skb_dst_drop(skb); 2152 2153 /* drop conntrack reference */ 2154 nf_reset(skb); 2155 2156 spin_lock(&sk->sk_receive_queue.lock); 2157 po->stats.stats1.tp_packets++; 2158 sock_skb_set_dropcount(sk, skb); 2159 __skb_queue_tail(&sk->sk_receive_queue, skb); 2160 spin_unlock(&sk->sk_receive_queue.lock); 2161 sk->sk_data_ready(sk); 2162 return 0; 2163 2164 drop_n_acct: 2165 is_drop_n_account = true; 2166 spin_lock(&sk->sk_receive_queue.lock); 2167 po->stats.stats1.tp_drops++; 2168 atomic_inc(&sk->sk_drops); 2169 spin_unlock(&sk->sk_receive_queue.lock); 2170 2171 drop_n_restore: 2172 if (skb_head != skb->data && skb_shared(skb)) { 2173 skb->data = skb_head; 2174 skb->len = skb_len; 2175 } 2176 drop: 2177 if (!is_drop_n_account) 2178 consume_skb(skb); 2179 else 2180 kfree_skb(skb); 2181 return 0; 2182 } 2183 2184 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2185 struct packet_type *pt, struct net_device *orig_dev) 2186 { 2187 struct sock *sk; 2188 struct packet_sock *po; 2189 struct sockaddr_ll *sll; 2190 union tpacket_uhdr h; 2191 u8 *skb_head = skb->data; 2192 int skb_len = skb->len; 2193 unsigned int snaplen, res; 2194 unsigned long status = TP_STATUS_USER; 2195 unsigned short macoff, netoff, hdrlen; 2196 struct sk_buff *copy_skb = NULL; 2197 struct timespec ts; 2198 __u32 ts_status; 2199 bool is_drop_n_account = false; 2200 bool do_vnet = false; 2201 2202 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2203 * We may add members to them until current aligned size without forcing 2204 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2205 */ 2206 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2207 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2208 2209 if (skb->pkt_type == PACKET_LOOPBACK) 2210 goto drop; 2211 2212 sk = pt->af_packet_priv; 2213 po = pkt_sk(sk); 2214 2215 if (!net_eq(dev_net(dev), sock_net(sk))) 2216 goto drop; 2217 2218 if (dev->header_ops) { 2219 if (sk->sk_type != SOCK_DGRAM) 2220 skb_push(skb, skb->data - skb_mac_header(skb)); 2221 else if (skb->pkt_type == PACKET_OUTGOING) { 2222 /* Special case: outgoing packets have ll header at head */ 2223 skb_pull(skb, skb_network_offset(skb)); 2224 } 2225 } 2226 2227 snaplen = skb->len; 2228 2229 res = run_filter(skb, sk, snaplen); 2230 if (!res) 2231 goto drop_n_restore; 2232 2233 if (skb->ip_summed == CHECKSUM_PARTIAL) 2234 status |= TP_STATUS_CSUMNOTREADY; 2235 else if (skb->pkt_type != PACKET_OUTGOING && 2236 (skb->ip_summed == CHECKSUM_COMPLETE || 2237 skb_csum_unnecessary(skb))) 2238 status |= TP_STATUS_CSUM_VALID; 2239 2240 if (snaplen > res) 2241 snaplen = res; 2242 2243 if (sk->sk_type == SOCK_DGRAM) { 2244 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2245 po->tp_reserve; 2246 } else { 2247 unsigned int maclen = skb_network_offset(skb); 2248 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2249 (maclen < 16 ? 16 : maclen)) + 2250 po->tp_reserve; 2251 if (po->has_vnet_hdr) { 2252 netoff += sizeof(struct virtio_net_hdr); 2253 do_vnet = true; 2254 } 2255 macoff = netoff - maclen; 2256 } 2257 if (po->tp_version <= TPACKET_V2) { 2258 if (macoff + snaplen > po->rx_ring.frame_size) { 2259 if (po->copy_thresh && 2260 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2261 if (skb_shared(skb)) { 2262 copy_skb = skb_clone(skb, GFP_ATOMIC); 2263 } else { 2264 copy_skb = skb_get(skb); 2265 skb_head = skb->data; 2266 } 2267 if (copy_skb) 2268 skb_set_owner_r(copy_skb, sk); 2269 } 2270 snaplen = po->rx_ring.frame_size - macoff; 2271 if ((int)snaplen < 0) { 2272 snaplen = 0; 2273 do_vnet = false; 2274 } 2275 } 2276 } else if (unlikely(macoff + snaplen > 2277 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2278 u32 nval; 2279 2280 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2281 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2282 snaplen, nval, macoff); 2283 snaplen = nval; 2284 if (unlikely((int)snaplen < 0)) { 2285 snaplen = 0; 2286 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2287 do_vnet = false; 2288 } 2289 } 2290 spin_lock(&sk->sk_receive_queue.lock); 2291 h.raw = packet_current_rx_frame(po, skb, 2292 TP_STATUS_KERNEL, (macoff+snaplen)); 2293 if (!h.raw) 2294 goto drop_n_account; 2295 if (po->tp_version <= TPACKET_V2) { 2296 packet_increment_rx_head(po, &po->rx_ring); 2297 /* 2298 * LOSING will be reported till you read the stats, 2299 * because it's COR - Clear On Read. 2300 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2301 * at packet level. 2302 */ 2303 if (po->stats.stats1.tp_drops) 2304 status |= TP_STATUS_LOSING; 2305 } 2306 po->stats.stats1.tp_packets++; 2307 if (copy_skb) { 2308 status |= TP_STATUS_COPY; 2309 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2310 } 2311 spin_unlock(&sk->sk_receive_queue.lock); 2312 2313 if (do_vnet) { 2314 if (virtio_net_hdr_from_skb(skb, h.raw + macoff - 2315 sizeof(struct virtio_net_hdr), 2316 vio_le(), true)) { 2317 spin_lock(&sk->sk_receive_queue.lock); 2318 goto drop_n_account; 2319 } 2320 } 2321 2322 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2323 2324 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 2325 getnstimeofday(&ts); 2326 2327 status |= ts_status; 2328 2329 switch (po->tp_version) { 2330 case TPACKET_V1: 2331 h.h1->tp_len = skb->len; 2332 h.h1->tp_snaplen = snaplen; 2333 h.h1->tp_mac = macoff; 2334 h.h1->tp_net = netoff; 2335 h.h1->tp_sec = ts.tv_sec; 2336 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2337 hdrlen = sizeof(*h.h1); 2338 break; 2339 case TPACKET_V2: 2340 h.h2->tp_len = skb->len; 2341 h.h2->tp_snaplen = snaplen; 2342 h.h2->tp_mac = macoff; 2343 h.h2->tp_net = netoff; 2344 h.h2->tp_sec = ts.tv_sec; 2345 h.h2->tp_nsec = ts.tv_nsec; 2346 if (skb_vlan_tag_present(skb)) { 2347 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2348 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2349 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2350 } else { 2351 h.h2->tp_vlan_tci = 0; 2352 h.h2->tp_vlan_tpid = 0; 2353 } 2354 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2355 hdrlen = sizeof(*h.h2); 2356 break; 2357 case TPACKET_V3: 2358 /* tp_nxt_offset,vlan are already populated above. 2359 * So DONT clear those fields here 2360 */ 2361 h.h3->tp_status |= status; 2362 h.h3->tp_len = skb->len; 2363 h.h3->tp_snaplen = snaplen; 2364 h.h3->tp_mac = macoff; 2365 h.h3->tp_net = netoff; 2366 h.h3->tp_sec = ts.tv_sec; 2367 h.h3->tp_nsec = ts.tv_nsec; 2368 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2369 hdrlen = sizeof(*h.h3); 2370 break; 2371 default: 2372 BUG(); 2373 } 2374 2375 sll = h.raw + TPACKET_ALIGN(hdrlen); 2376 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2377 sll->sll_family = AF_PACKET; 2378 sll->sll_hatype = dev->type; 2379 sll->sll_protocol = skb->protocol; 2380 sll->sll_pkttype = skb->pkt_type; 2381 if (unlikely(po->origdev)) 2382 sll->sll_ifindex = orig_dev->ifindex; 2383 else 2384 sll->sll_ifindex = dev->ifindex; 2385 2386 smp_mb(); 2387 2388 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2389 if (po->tp_version <= TPACKET_V2) { 2390 u8 *start, *end; 2391 2392 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2393 macoff + snaplen); 2394 2395 for (start = h.raw; start < end; start += PAGE_SIZE) 2396 flush_dcache_page(pgv_to_page(start)); 2397 } 2398 smp_wmb(); 2399 #endif 2400 2401 if (po->tp_version <= TPACKET_V2) { 2402 __packet_set_status(po, h.raw, status); 2403 sk->sk_data_ready(sk); 2404 } else { 2405 prb_clear_blk_fill_status(&po->rx_ring); 2406 } 2407 2408 drop_n_restore: 2409 if (skb_head != skb->data && skb_shared(skb)) { 2410 skb->data = skb_head; 2411 skb->len = skb_len; 2412 } 2413 drop: 2414 if (!is_drop_n_account) 2415 consume_skb(skb); 2416 else 2417 kfree_skb(skb); 2418 return 0; 2419 2420 drop_n_account: 2421 is_drop_n_account = true; 2422 po->stats.stats1.tp_drops++; 2423 spin_unlock(&sk->sk_receive_queue.lock); 2424 2425 sk->sk_data_ready(sk); 2426 kfree_skb(copy_skb); 2427 goto drop_n_restore; 2428 } 2429 2430 static void tpacket_destruct_skb(struct sk_buff *skb) 2431 { 2432 struct packet_sock *po = pkt_sk(skb->sk); 2433 2434 if (likely(po->tx_ring.pg_vec)) { 2435 void *ph; 2436 __u32 ts; 2437 2438 ph = skb_shinfo(skb)->destructor_arg; 2439 packet_dec_pending(&po->tx_ring); 2440 2441 ts = __packet_set_timestamp(po, ph, skb); 2442 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2443 } 2444 2445 sock_wfree(skb); 2446 } 2447 2448 static void tpacket_set_protocol(const struct net_device *dev, 2449 struct sk_buff *skb) 2450 { 2451 if (dev->type == ARPHRD_ETHER) { 2452 skb_reset_mac_header(skb); 2453 skb->protocol = eth_hdr(skb)->h_proto; 2454 } 2455 } 2456 2457 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2458 { 2459 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2460 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2461 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2462 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2463 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2464 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2465 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2466 2467 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2468 return -EINVAL; 2469 2470 return 0; 2471 } 2472 2473 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2474 struct virtio_net_hdr *vnet_hdr) 2475 { 2476 if (*len < sizeof(*vnet_hdr)) 2477 return -EINVAL; 2478 *len -= sizeof(*vnet_hdr); 2479 2480 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2481 return -EFAULT; 2482 2483 return __packet_snd_vnet_parse(vnet_hdr, *len); 2484 } 2485 2486 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2487 void *frame, struct net_device *dev, void *data, int tp_len, 2488 __be16 proto, unsigned char *addr, int hlen, int copylen, 2489 const struct sockcm_cookie *sockc) 2490 { 2491 union tpacket_uhdr ph; 2492 int to_write, offset, len, nr_frags, len_max; 2493 struct socket *sock = po->sk.sk_socket; 2494 struct page *page; 2495 int err; 2496 2497 ph.raw = frame; 2498 2499 skb->protocol = proto; 2500 skb->dev = dev; 2501 skb->priority = po->sk.sk_priority; 2502 skb->mark = po->sk.sk_mark; 2503 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags); 2504 skb_shinfo(skb)->destructor_arg = ph.raw; 2505 2506 skb_reserve(skb, hlen); 2507 skb_reset_network_header(skb); 2508 2509 to_write = tp_len; 2510 2511 if (sock->type == SOCK_DGRAM) { 2512 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2513 NULL, tp_len); 2514 if (unlikely(err < 0)) 2515 return -EINVAL; 2516 } else if (copylen) { 2517 int hdrlen = min_t(int, copylen, tp_len); 2518 2519 skb_push(skb, dev->hard_header_len); 2520 skb_put(skb, copylen - dev->hard_header_len); 2521 err = skb_store_bits(skb, 0, data, hdrlen); 2522 if (unlikely(err)) 2523 return err; 2524 if (!dev_validate_header(dev, skb->data, hdrlen)) 2525 return -EINVAL; 2526 if (!skb->protocol) 2527 tpacket_set_protocol(dev, skb); 2528 2529 data += hdrlen; 2530 to_write -= hdrlen; 2531 } 2532 2533 offset = offset_in_page(data); 2534 len_max = PAGE_SIZE - offset; 2535 len = ((to_write > len_max) ? len_max : to_write); 2536 2537 skb->data_len = to_write; 2538 skb->len += to_write; 2539 skb->truesize += to_write; 2540 refcount_add(to_write, &po->sk.sk_wmem_alloc); 2541 2542 while (likely(to_write)) { 2543 nr_frags = skb_shinfo(skb)->nr_frags; 2544 2545 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2546 pr_err("Packet exceed the number of skb frags(%lu)\n", 2547 MAX_SKB_FRAGS); 2548 return -EFAULT; 2549 } 2550 2551 page = pgv_to_page(data); 2552 data += len; 2553 flush_dcache_page(page); 2554 get_page(page); 2555 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2556 to_write -= len; 2557 offset = 0; 2558 len_max = PAGE_SIZE; 2559 len = ((to_write > len_max) ? len_max : to_write); 2560 } 2561 2562 skb_probe_transport_header(skb, 0); 2563 2564 return tp_len; 2565 } 2566 2567 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2568 int size_max, void **data) 2569 { 2570 union tpacket_uhdr ph; 2571 int tp_len, off; 2572 2573 ph.raw = frame; 2574 2575 switch (po->tp_version) { 2576 case TPACKET_V3: 2577 if (ph.h3->tp_next_offset != 0) { 2578 pr_warn_once("variable sized slot not supported"); 2579 return -EINVAL; 2580 } 2581 tp_len = ph.h3->tp_len; 2582 break; 2583 case TPACKET_V2: 2584 tp_len = ph.h2->tp_len; 2585 break; 2586 default: 2587 tp_len = ph.h1->tp_len; 2588 break; 2589 } 2590 if (unlikely(tp_len > size_max)) { 2591 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2592 return -EMSGSIZE; 2593 } 2594 2595 if (unlikely(po->tp_tx_has_off)) { 2596 int off_min, off_max; 2597 2598 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2599 off_max = po->tx_ring.frame_size - tp_len; 2600 if (po->sk.sk_type == SOCK_DGRAM) { 2601 switch (po->tp_version) { 2602 case TPACKET_V3: 2603 off = ph.h3->tp_net; 2604 break; 2605 case TPACKET_V2: 2606 off = ph.h2->tp_net; 2607 break; 2608 default: 2609 off = ph.h1->tp_net; 2610 break; 2611 } 2612 } else { 2613 switch (po->tp_version) { 2614 case TPACKET_V3: 2615 off = ph.h3->tp_mac; 2616 break; 2617 case TPACKET_V2: 2618 off = ph.h2->tp_mac; 2619 break; 2620 default: 2621 off = ph.h1->tp_mac; 2622 break; 2623 } 2624 } 2625 if (unlikely((off < off_min) || (off_max < off))) 2626 return -EINVAL; 2627 } else { 2628 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2629 } 2630 2631 *data = frame + off; 2632 return tp_len; 2633 } 2634 2635 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2636 { 2637 struct sk_buff *skb; 2638 struct net_device *dev; 2639 struct virtio_net_hdr *vnet_hdr = NULL; 2640 struct sockcm_cookie sockc; 2641 __be16 proto; 2642 int err, reserve = 0; 2643 void *ph; 2644 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2645 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2646 int tp_len, size_max; 2647 unsigned char *addr; 2648 void *data; 2649 int len_sum = 0; 2650 int status = TP_STATUS_AVAILABLE; 2651 int hlen, tlen, copylen = 0; 2652 2653 mutex_lock(&po->pg_vec_lock); 2654 2655 if (likely(saddr == NULL)) { 2656 dev = packet_cached_dev_get(po); 2657 proto = po->num; 2658 addr = NULL; 2659 } else { 2660 err = -EINVAL; 2661 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2662 goto out; 2663 if (msg->msg_namelen < (saddr->sll_halen 2664 + offsetof(struct sockaddr_ll, 2665 sll_addr))) 2666 goto out; 2667 proto = saddr->sll_protocol; 2668 addr = saddr->sll_addr; 2669 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2670 } 2671 2672 err = -ENXIO; 2673 if (unlikely(dev == NULL)) 2674 goto out; 2675 err = -ENETDOWN; 2676 if (unlikely(!(dev->flags & IFF_UP))) 2677 goto out_put; 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_put; 2684 } 2685 2686 if (po->sk.sk_socket->type == SOCK_RAW) 2687 reserve = dev->hard_header_len; 2688 size_max = po->tx_ring.frame_size 2689 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2690 2691 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) 2692 size_max = dev->mtu + reserve + VLAN_HLEN; 2693 2694 do { 2695 ph = packet_current_frame(po, &po->tx_ring, 2696 TP_STATUS_SEND_REQUEST); 2697 if (unlikely(ph == NULL)) { 2698 if (need_wait && need_resched()) 2699 schedule(); 2700 continue; 2701 } 2702 2703 skb = NULL; 2704 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2705 if (tp_len < 0) 2706 goto tpacket_error; 2707 2708 status = TP_STATUS_SEND_REQUEST; 2709 hlen = LL_RESERVED_SPACE(dev); 2710 tlen = dev->needed_tailroom; 2711 if (po->has_vnet_hdr) { 2712 vnet_hdr = data; 2713 data += sizeof(*vnet_hdr); 2714 tp_len -= sizeof(*vnet_hdr); 2715 if (tp_len < 0 || 2716 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2717 tp_len = -EINVAL; 2718 goto tpacket_error; 2719 } 2720 copylen = __virtio16_to_cpu(vio_le(), 2721 vnet_hdr->hdr_len); 2722 } 2723 copylen = max_t(int, copylen, dev->hard_header_len); 2724 skb = sock_alloc_send_skb(&po->sk, 2725 hlen + tlen + sizeof(struct sockaddr_ll) + 2726 (copylen - dev->hard_header_len), 2727 !need_wait, &err); 2728 2729 if (unlikely(skb == NULL)) { 2730 /* we assume the socket was initially writeable ... */ 2731 if (likely(len_sum > 0)) 2732 err = len_sum; 2733 goto out_status; 2734 } 2735 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2736 addr, hlen, copylen, &sockc); 2737 if (likely(tp_len >= 0) && 2738 tp_len > dev->mtu + reserve && 2739 !po->has_vnet_hdr && 2740 !packet_extra_vlan_len_allowed(dev, skb)) 2741 tp_len = -EMSGSIZE; 2742 2743 if (unlikely(tp_len < 0)) { 2744 tpacket_error: 2745 if (po->tp_loss) { 2746 __packet_set_status(po, ph, 2747 TP_STATUS_AVAILABLE); 2748 packet_increment_head(&po->tx_ring); 2749 kfree_skb(skb); 2750 continue; 2751 } else { 2752 status = TP_STATUS_WRONG_FORMAT; 2753 err = tp_len; 2754 goto out_status; 2755 } 2756 } 2757 2758 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr, 2759 vio_le())) { 2760 tp_len = -EINVAL; 2761 goto tpacket_error; 2762 } 2763 2764 skb->destructor = tpacket_destruct_skb; 2765 __packet_set_status(po, ph, TP_STATUS_SENDING); 2766 packet_inc_pending(&po->tx_ring); 2767 2768 status = TP_STATUS_SEND_REQUEST; 2769 err = po->xmit(skb); 2770 if (unlikely(err > 0)) { 2771 err = net_xmit_errno(err); 2772 if (err && __packet_get_status(po, ph) == 2773 TP_STATUS_AVAILABLE) { 2774 /* skb was destructed already */ 2775 skb = NULL; 2776 goto out_status; 2777 } 2778 /* 2779 * skb was dropped but not destructed yet; 2780 * let's treat it like congestion or err < 0 2781 */ 2782 err = 0; 2783 } 2784 packet_increment_head(&po->tx_ring); 2785 len_sum += tp_len; 2786 } while (likely((ph != NULL) || 2787 /* Note: packet_read_pending() might be slow if we have 2788 * to call it as it's per_cpu variable, but in fast-path 2789 * we already short-circuit the loop with the first 2790 * condition, and luckily don't have to go that path 2791 * anyway. 2792 */ 2793 (need_wait && packet_read_pending(&po->tx_ring)))); 2794 2795 err = len_sum; 2796 goto out_put; 2797 2798 out_status: 2799 __packet_set_status(po, ph, status); 2800 kfree_skb(skb); 2801 out_put: 2802 dev_put(dev); 2803 out: 2804 mutex_unlock(&po->pg_vec_lock); 2805 return err; 2806 } 2807 2808 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2809 size_t reserve, size_t len, 2810 size_t linear, int noblock, 2811 int *err) 2812 { 2813 struct sk_buff *skb; 2814 2815 /* Under a page? Don't bother with paged skb. */ 2816 if (prepad + len < PAGE_SIZE || !linear) 2817 linear = len; 2818 2819 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2820 err, 0); 2821 if (!skb) 2822 return NULL; 2823 2824 skb_reserve(skb, reserve); 2825 skb_put(skb, linear); 2826 skb->data_len = len - linear; 2827 skb->len += len - linear; 2828 2829 return skb; 2830 } 2831 2832 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2833 { 2834 struct sock *sk = sock->sk; 2835 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2836 struct sk_buff *skb; 2837 struct net_device *dev; 2838 __be16 proto; 2839 unsigned char *addr; 2840 int err, reserve = 0; 2841 struct sockcm_cookie sockc; 2842 struct virtio_net_hdr vnet_hdr = { 0 }; 2843 int offset = 0; 2844 struct packet_sock *po = pkt_sk(sk); 2845 bool has_vnet_hdr = false; 2846 int hlen, tlen, linear; 2847 int extra_len = 0; 2848 2849 /* 2850 * Get and verify the address. 2851 */ 2852 2853 if (likely(saddr == NULL)) { 2854 dev = packet_cached_dev_get(po); 2855 proto = po->num; 2856 addr = NULL; 2857 } else { 2858 err = -EINVAL; 2859 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2860 goto out; 2861 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2862 goto out; 2863 proto = saddr->sll_protocol; 2864 addr = saddr->sll_addr; 2865 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2866 } 2867 2868 err = -ENXIO; 2869 if (unlikely(dev == NULL)) 2870 goto out_unlock; 2871 err = -ENETDOWN; 2872 if (unlikely(!(dev->flags & IFF_UP))) 2873 goto out_unlock; 2874 2875 sockc.tsflags = sk->sk_tsflags; 2876 sockc.mark = sk->sk_mark; 2877 if (msg->msg_controllen) { 2878 err = sock_cmsg_send(sk, msg, &sockc); 2879 if (unlikely(err)) 2880 goto out_unlock; 2881 } 2882 2883 if (sock->type == SOCK_RAW) 2884 reserve = dev->hard_header_len; 2885 if (po->has_vnet_hdr) { 2886 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2887 if (err) 2888 goto out_unlock; 2889 has_vnet_hdr = true; 2890 } 2891 2892 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2893 if (!netif_supports_nofcs(dev)) { 2894 err = -EPROTONOSUPPORT; 2895 goto out_unlock; 2896 } 2897 extra_len = 4; /* We're doing our own CRC */ 2898 } 2899 2900 err = -EMSGSIZE; 2901 if (!vnet_hdr.gso_type && 2902 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2903 goto out_unlock; 2904 2905 err = -ENOBUFS; 2906 hlen = LL_RESERVED_SPACE(dev); 2907 tlen = dev->needed_tailroom; 2908 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); 2909 linear = max(linear, min_t(int, len, dev->hard_header_len)); 2910 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, 2911 msg->msg_flags & MSG_DONTWAIT, &err); 2912 if (skb == NULL) 2913 goto out_unlock; 2914 2915 skb_set_network_header(skb, reserve); 2916 2917 err = -EINVAL; 2918 if (sock->type == SOCK_DGRAM) { 2919 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 2920 if (unlikely(offset < 0)) 2921 goto out_free; 2922 } 2923 2924 /* Returns -EFAULT on error */ 2925 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 2926 if (err) 2927 goto out_free; 2928 2929 if (sock->type == SOCK_RAW && 2930 !dev_validate_header(dev, skb->data, len)) { 2931 err = -EINVAL; 2932 goto out_free; 2933 } 2934 2935 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags); 2936 2937 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 2938 !packet_extra_vlan_len_allowed(dev, skb)) { 2939 err = -EMSGSIZE; 2940 goto out_free; 2941 } 2942 2943 skb->protocol = proto; 2944 skb->dev = dev; 2945 skb->priority = sk->sk_priority; 2946 skb->mark = sockc.mark; 2947 2948 if (has_vnet_hdr) { 2949 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 2950 if (err) 2951 goto out_free; 2952 len += sizeof(vnet_hdr); 2953 } 2954 2955 skb_probe_transport_header(skb, reserve); 2956 2957 if (unlikely(extra_len == 4)) 2958 skb->no_fcs = 1; 2959 2960 err = po->xmit(skb); 2961 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2962 goto out_unlock; 2963 2964 dev_put(dev); 2965 2966 return len; 2967 2968 out_free: 2969 kfree_skb(skb); 2970 out_unlock: 2971 if (dev) 2972 dev_put(dev); 2973 out: 2974 return err; 2975 } 2976 2977 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2978 { 2979 struct sock *sk = sock->sk; 2980 struct packet_sock *po = pkt_sk(sk); 2981 2982 if (po->tx_ring.pg_vec) 2983 return tpacket_snd(po, msg); 2984 else 2985 return packet_snd(sock, msg, len); 2986 } 2987 2988 /* 2989 * Close a PACKET socket. This is fairly simple. We immediately go 2990 * to 'closed' state and remove our protocol entry in the device list. 2991 */ 2992 2993 static int packet_release(struct socket *sock) 2994 { 2995 struct sock *sk = sock->sk; 2996 struct packet_sock *po; 2997 struct packet_fanout *f; 2998 struct net *net; 2999 union tpacket_req_u req_u; 3000 3001 if (!sk) 3002 return 0; 3003 3004 net = sock_net(sk); 3005 po = pkt_sk(sk); 3006 3007 mutex_lock(&net->packet.sklist_lock); 3008 sk_del_node_init_rcu(sk); 3009 mutex_unlock(&net->packet.sklist_lock); 3010 3011 preempt_disable(); 3012 sock_prot_inuse_add(net, sk->sk_prot, -1); 3013 preempt_enable(); 3014 3015 spin_lock(&po->bind_lock); 3016 unregister_prot_hook(sk, false); 3017 packet_cached_dev_reset(po); 3018 3019 if (po->prot_hook.dev) { 3020 dev_put(po->prot_hook.dev); 3021 po->prot_hook.dev = NULL; 3022 } 3023 spin_unlock(&po->bind_lock); 3024 3025 packet_flush_mclist(sk); 3026 3027 if (po->rx_ring.pg_vec) { 3028 memset(&req_u, 0, sizeof(req_u)); 3029 packet_set_ring(sk, &req_u, 1, 0); 3030 } 3031 3032 if (po->tx_ring.pg_vec) { 3033 memset(&req_u, 0, sizeof(req_u)); 3034 packet_set_ring(sk, &req_u, 1, 1); 3035 } 3036 3037 f = fanout_release(sk); 3038 3039 synchronize_net(); 3040 3041 if (f) { 3042 fanout_release_data(f); 3043 kfree(f); 3044 } 3045 /* 3046 * Now the socket is dead. No more input will appear. 3047 */ 3048 sock_orphan(sk); 3049 sock->sk = NULL; 3050 3051 /* Purge queues */ 3052 3053 skb_queue_purge(&sk->sk_receive_queue); 3054 packet_free_pending(po); 3055 sk_refcnt_debug_release(sk); 3056 3057 sock_put(sk); 3058 return 0; 3059 } 3060 3061 /* 3062 * Attach a packet hook. 3063 */ 3064 3065 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3066 __be16 proto) 3067 { 3068 struct packet_sock *po = pkt_sk(sk); 3069 struct net_device *dev_curr; 3070 __be16 proto_curr; 3071 bool need_rehook; 3072 struct net_device *dev = NULL; 3073 int ret = 0; 3074 bool unlisted = false; 3075 3076 lock_sock(sk); 3077 spin_lock(&po->bind_lock); 3078 rcu_read_lock(); 3079 3080 if (po->fanout) { 3081 ret = -EINVAL; 3082 goto out_unlock; 3083 } 3084 3085 if (name) { 3086 dev = dev_get_by_name_rcu(sock_net(sk), name); 3087 if (!dev) { 3088 ret = -ENODEV; 3089 goto out_unlock; 3090 } 3091 } else if (ifindex) { 3092 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3093 if (!dev) { 3094 ret = -ENODEV; 3095 goto out_unlock; 3096 } 3097 } 3098 3099 if (dev) 3100 dev_hold(dev); 3101 3102 proto_curr = po->prot_hook.type; 3103 dev_curr = po->prot_hook.dev; 3104 3105 need_rehook = proto_curr != proto || dev_curr != dev; 3106 3107 if (need_rehook) { 3108 if (po->running) { 3109 rcu_read_unlock(); 3110 __unregister_prot_hook(sk, true); 3111 rcu_read_lock(); 3112 dev_curr = po->prot_hook.dev; 3113 if (dev) 3114 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3115 dev->ifindex); 3116 } 3117 3118 po->num = proto; 3119 po->prot_hook.type = proto; 3120 3121 if (unlikely(unlisted)) { 3122 dev_put(dev); 3123 po->prot_hook.dev = NULL; 3124 po->ifindex = -1; 3125 packet_cached_dev_reset(po); 3126 } else { 3127 po->prot_hook.dev = dev; 3128 po->ifindex = dev ? dev->ifindex : 0; 3129 packet_cached_dev_assign(po, dev); 3130 } 3131 } 3132 if (dev_curr) 3133 dev_put(dev_curr); 3134 3135 if (proto == 0 || !need_rehook) 3136 goto out_unlock; 3137 3138 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3139 register_prot_hook(sk); 3140 } else { 3141 sk->sk_err = ENETDOWN; 3142 if (!sock_flag(sk, SOCK_DEAD)) 3143 sk->sk_error_report(sk); 3144 } 3145 3146 out_unlock: 3147 rcu_read_unlock(); 3148 spin_unlock(&po->bind_lock); 3149 release_sock(sk); 3150 return ret; 3151 } 3152 3153 /* 3154 * Bind a packet socket to a device 3155 */ 3156 3157 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3158 int addr_len) 3159 { 3160 struct sock *sk = sock->sk; 3161 char name[sizeof(uaddr->sa_data) + 1]; 3162 3163 /* 3164 * Check legality 3165 */ 3166 3167 if (addr_len != sizeof(struct sockaddr)) 3168 return -EINVAL; 3169 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be 3170 * zero-terminated. 3171 */ 3172 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); 3173 name[sizeof(uaddr->sa_data)] = 0; 3174 3175 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3176 } 3177 3178 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3179 { 3180 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3181 struct sock *sk = sock->sk; 3182 3183 /* 3184 * Check legality 3185 */ 3186 3187 if (addr_len < sizeof(struct sockaddr_ll)) 3188 return -EINVAL; 3189 if (sll->sll_family != AF_PACKET) 3190 return -EINVAL; 3191 3192 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3193 sll->sll_protocol ? : pkt_sk(sk)->num); 3194 } 3195 3196 static struct proto packet_proto = { 3197 .name = "PACKET", 3198 .owner = THIS_MODULE, 3199 .obj_size = sizeof(struct packet_sock), 3200 }; 3201 3202 /* 3203 * Create a packet of type SOCK_PACKET. 3204 */ 3205 3206 static int packet_create(struct net *net, struct socket *sock, int protocol, 3207 int kern) 3208 { 3209 struct sock *sk; 3210 struct packet_sock *po; 3211 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3212 int err; 3213 3214 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3215 return -EPERM; 3216 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3217 sock->type != SOCK_PACKET) 3218 return -ESOCKTNOSUPPORT; 3219 3220 sock->state = SS_UNCONNECTED; 3221 3222 err = -ENOBUFS; 3223 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3224 if (sk == NULL) 3225 goto out; 3226 3227 sock->ops = &packet_ops; 3228 if (sock->type == SOCK_PACKET) 3229 sock->ops = &packet_ops_spkt; 3230 3231 sock_init_data(sock, sk); 3232 3233 po = pkt_sk(sk); 3234 sk->sk_family = PF_PACKET; 3235 po->num = proto; 3236 po->xmit = dev_queue_xmit; 3237 3238 err = packet_alloc_pending(po); 3239 if (err) 3240 goto out2; 3241 3242 packet_cached_dev_reset(po); 3243 3244 sk->sk_destruct = packet_sock_destruct; 3245 sk_refcnt_debug_inc(sk); 3246 3247 /* 3248 * Attach a protocol block 3249 */ 3250 3251 spin_lock_init(&po->bind_lock); 3252 mutex_init(&po->pg_vec_lock); 3253 po->rollover = NULL; 3254 po->prot_hook.func = packet_rcv; 3255 3256 if (sock->type == SOCK_PACKET) 3257 po->prot_hook.func = packet_rcv_spkt; 3258 3259 po->prot_hook.af_packet_priv = sk; 3260 3261 if (proto) { 3262 po->prot_hook.type = proto; 3263 register_prot_hook(sk); 3264 } 3265 3266 mutex_lock(&net->packet.sklist_lock); 3267 sk_add_node_rcu(sk, &net->packet.sklist); 3268 mutex_unlock(&net->packet.sklist_lock); 3269 3270 preempt_disable(); 3271 sock_prot_inuse_add(net, &packet_proto, 1); 3272 preempt_enable(); 3273 3274 return 0; 3275 out2: 3276 sk_free(sk); 3277 out: 3278 return err; 3279 } 3280 3281 /* 3282 * Pull a packet from our receive queue and hand it to the user. 3283 * If necessary we block. 3284 */ 3285 3286 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3287 int flags) 3288 { 3289 struct sock *sk = sock->sk; 3290 struct sk_buff *skb; 3291 int copied, err; 3292 int vnet_hdr_len = 0; 3293 unsigned int origlen = 0; 3294 3295 err = -EINVAL; 3296 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3297 goto out; 3298 3299 #if 0 3300 /* What error should we return now? EUNATTACH? */ 3301 if (pkt_sk(sk)->ifindex < 0) 3302 return -ENODEV; 3303 #endif 3304 3305 if (flags & MSG_ERRQUEUE) { 3306 err = sock_recv_errqueue(sk, msg, len, 3307 SOL_PACKET, PACKET_TX_TIMESTAMP); 3308 goto out; 3309 } 3310 3311 /* 3312 * Call the generic datagram receiver. This handles all sorts 3313 * of horrible races and re-entrancy so we can forget about it 3314 * in the protocol layers. 3315 * 3316 * Now it will return ENETDOWN, if device have just gone down, 3317 * but then it will block. 3318 */ 3319 3320 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 3321 3322 /* 3323 * An error occurred so return it. Because skb_recv_datagram() 3324 * handles the blocking we don't see and worry about blocking 3325 * retries. 3326 */ 3327 3328 if (skb == NULL) 3329 goto out; 3330 3331 if (pkt_sk(sk)->pressure) 3332 packet_rcv_has_room(pkt_sk(sk), NULL); 3333 3334 if (pkt_sk(sk)->has_vnet_hdr) { 3335 err = packet_rcv_vnet(msg, skb, &len); 3336 if (err) 3337 goto out_free; 3338 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3339 } 3340 3341 /* You lose any data beyond the buffer you gave. If it worries 3342 * a user program they can ask the device for its MTU 3343 * anyway. 3344 */ 3345 copied = skb->len; 3346 if (copied > len) { 3347 copied = len; 3348 msg->msg_flags |= MSG_TRUNC; 3349 } 3350 3351 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3352 if (err) 3353 goto out_free; 3354 3355 if (sock->type != SOCK_PACKET) { 3356 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3357 3358 /* Original length was stored in sockaddr_ll fields */ 3359 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3360 sll->sll_family = AF_PACKET; 3361 sll->sll_protocol = skb->protocol; 3362 } 3363 3364 sock_recv_ts_and_drops(msg, sk, skb); 3365 3366 if (msg->msg_name) { 3367 /* If the address length field is there to be filled 3368 * in, we fill it in now. 3369 */ 3370 if (sock->type == SOCK_PACKET) { 3371 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3372 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3373 } else { 3374 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3375 3376 msg->msg_namelen = sll->sll_halen + 3377 offsetof(struct sockaddr_ll, sll_addr); 3378 } 3379 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, 3380 msg->msg_namelen); 3381 } 3382 3383 if (pkt_sk(sk)->auxdata) { 3384 struct tpacket_auxdata aux; 3385 3386 aux.tp_status = TP_STATUS_USER; 3387 if (skb->ip_summed == CHECKSUM_PARTIAL) 3388 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3389 else if (skb->pkt_type != PACKET_OUTGOING && 3390 (skb->ip_summed == CHECKSUM_COMPLETE || 3391 skb_csum_unnecessary(skb))) 3392 aux.tp_status |= TP_STATUS_CSUM_VALID; 3393 3394 aux.tp_len = origlen; 3395 aux.tp_snaplen = skb->len; 3396 aux.tp_mac = 0; 3397 aux.tp_net = skb_network_offset(skb); 3398 if (skb_vlan_tag_present(skb)) { 3399 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3400 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3401 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3402 } else { 3403 aux.tp_vlan_tci = 0; 3404 aux.tp_vlan_tpid = 0; 3405 } 3406 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3407 } 3408 3409 /* 3410 * Free or return the buffer as appropriate. Again this 3411 * hides all the races and re-entrancy issues from us. 3412 */ 3413 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3414 3415 out_free: 3416 skb_free_datagram(sk, skb); 3417 out: 3418 return err; 3419 } 3420 3421 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3422 int *uaddr_len, int peer) 3423 { 3424 struct net_device *dev; 3425 struct sock *sk = sock->sk; 3426 3427 if (peer) 3428 return -EOPNOTSUPP; 3429 3430 uaddr->sa_family = AF_PACKET; 3431 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3432 rcu_read_lock(); 3433 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3434 if (dev) 3435 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3436 rcu_read_unlock(); 3437 *uaddr_len = sizeof(*uaddr); 3438 3439 return 0; 3440 } 3441 3442 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3443 int *uaddr_len, int peer) 3444 { 3445 struct net_device *dev; 3446 struct sock *sk = sock->sk; 3447 struct packet_sock *po = pkt_sk(sk); 3448 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3449 3450 if (peer) 3451 return -EOPNOTSUPP; 3452 3453 sll->sll_family = AF_PACKET; 3454 sll->sll_ifindex = po->ifindex; 3455 sll->sll_protocol = po->num; 3456 sll->sll_pkttype = 0; 3457 rcu_read_lock(); 3458 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 3459 if (dev) { 3460 sll->sll_hatype = dev->type; 3461 sll->sll_halen = dev->addr_len; 3462 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3463 } else { 3464 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3465 sll->sll_halen = 0; 3466 } 3467 rcu_read_unlock(); 3468 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3469 3470 return 0; 3471 } 3472 3473 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3474 int what) 3475 { 3476 switch (i->type) { 3477 case PACKET_MR_MULTICAST: 3478 if (i->alen != dev->addr_len) 3479 return -EINVAL; 3480 if (what > 0) 3481 return dev_mc_add(dev, i->addr); 3482 else 3483 return dev_mc_del(dev, i->addr); 3484 break; 3485 case PACKET_MR_PROMISC: 3486 return dev_set_promiscuity(dev, what); 3487 case PACKET_MR_ALLMULTI: 3488 return dev_set_allmulti(dev, what); 3489 case PACKET_MR_UNICAST: 3490 if (i->alen != dev->addr_len) 3491 return -EINVAL; 3492 if (what > 0) 3493 return dev_uc_add(dev, i->addr); 3494 else 3495 return dev_uc_del(dev, i->addr); 3496 break; 3497 default: 3498 break; 3499 } 3500 return 0; 3501 } 3502 3503 static void packet_dev_mclist_delete(struct net_device *dev, 3504 struct packet_mclist **mlp) 3505 { 3506 struct packet_mclist *ml; 3507 3508 while ((ml = *mlp) != NULL) { 3509 if (ml->ifindex == dev->ifindex) { 3510 packet_dev_mc(dev, ml, -1); 3511 *mlp = ml->next; 3512 kfree(ml); 3513 } else 3514 mlp = &ml->next; 3515 } 3516 } 3517 3518 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3519 { 3520 struct packet_sock *po = pkt_sk(sk); 3521 struct packet_mclist *ml, *i; 3522 struct net_device *dev; 3523 int err; 3524 3525 rtnl_lock(); 3526 3527 err = -ENODEV; 3528 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3529 if (!dev) 3530 goto done; 3531 3532 err = -EINVAL; 3533 if (mreq->mr_alen > dev->addr_len) 3534 goto done; 3535 3536 err = -ENOBUFS; 3537 i = kmalloc(sizeof(*i), GFP_KERNEL); 3538 if (i == NULL) 3539 goto done; 3540 3541 err = 0; 3542 for (ml = po->mclist; ml; ml = ml->next) { 3543 if (ml->ifindex == mreq->mr_ifindex && 3544 ml->type == mreq->mr_type && 3545 ml->alen == mreq->mr_alen && 3546 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3547 ml->count++; 3548 /* Free the new element ... */ 3549 kfree(i); 3550 goto done; 3551 } 3552 } 3553 3554 i->type = mreq->mr_type; 3555 i->ifindex = mreq->mr_ifindex; 3556 i->alen = mreq->mr_alen; 3557 memcpy(i->addr, mreq->mr_address, i->alen); 3558 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3559 i->count = 1; 3560 i->next = po->mclist; 3561 po->mclist = i; 3562 err = packet_dev_mc(dev, i, 1); 3563 if (err) { 3564 po->mclist = i->next; 3565 kfree(i); 3566 } 3567 3568 done: 3569 rtnl_unlock(); 3570 return err; 3571 } 3572 3573 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3574 { 3575 struct packet_mclist *ml, **mlp; 3576 3577 rtnl_lock(); 3578 3579 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3580 if (ml->ifindex == mreq->mr_ifindex && 3581 ml->type == mreq->mr_type && 3582 ml->alen == mreq->mr_alen && 3583 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3584 if (--ml->count == 0) { 3585 struct net_device *dev; 3586 *mlp = ml->next; 3587 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3588 if (dev) 3589 packet_dev_mc(dev, ml, -1); 3590 kfree(ml); 3591 } 3592 break; 3593 } 3594 } 3595 rtnl_unlock(); 3596 return 0; 3597 } 3598 3599 static void packet_flush_mclist(struct sock *sk) 3600 { 3601 struct packet_sock *po = pkt_sk(sk); 3602 struct packet_mclist *ml; 3603 3604 if (!po->mclist) 3605 return; 3606 3607 rtnl_lock(); 3608 while ((ml = po->mclist) != NULL) { 3609 struct net_device *dev; 3610 3611 po->mclist = ml->next; 3612 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3613 if (dev != NULL) 3614 packet_dev_mc(dev, ml, -1); 3615 kfree(ml); 3616 } 3617 rtnl_unlock(); 3618 } 3619 3620 static int 3621 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3622 { 3623 struct sock *sk = sock->sk; 3624 struct packet_sock *po = pkt_sk(sk); 3625 int ret; 3626 3627 if (level != SOL_PACKET) 3628 return -ENOPROTOOPT; 3629 3630 switch (optname) { 3631 case PACKET_ADD_MEMBERSHIP: 3632 case PACKET_DROP_MEMBERSHIP: 3633 { 3634 struct packet_mreq_max mreq; 3635 int len = optlen; 3636 memset(&mreq, 0, sizeof(mreq)); 3637 if (len < sizeof(struct packet_mreq)) 3638 return -EINVAL; 3639 if (len > sizeof(mreq)) 3640 len = sizeof(mreq); 3641 if (copy_from_user(&mreq, optval, len)) 3642 return -EFAULT; 3643 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3644 return -EINVAL; 3645 if (optname == PACKET_ADD_MEMBERSHIP) 3646 ret = packet_mc_add(sk, &mreq); 3647 else 3648 ret = packet_mc_drop(sk, &mreq); 3649 return ret; 3650 } 3651 3652 case PACKET_RX_RING: 3653 case PACKET_TX_RING: 3654 { 3655 union tpacket_req_u req_u; 3656 int len; 3657 3658 switch (po->tp_version) { 3659 case TPACKET_V1: 3660 case TPACKET_V2: 3661 len = sizeof(req_u.req); 3662 break; 3663 case TPACKET_V3: 3664 default: 3665 len = sizeof(req_u.req3); 3666 break; 3667 } 3668 if (optlen < len) 3669 return -EINVAL; 3670 if (copy_from_user(&req_u.req, optval, len)) 3671 return -EFAULT; 3672 return packet_set_ring(sk, &req_u, 0, 3673 optname == PACKET_TX_RING); 3674 } 3675 case PACKET_COPY_THRESH: 3676 { 3677 int val; 3678 3679 if (optlen != sizeof(val)) 3680 return -EINVAL; 3681 if (copy_from_user(&val, optval, sizeof(val))) 3682 return -EFAULT; 3683 3684 pkt_sk(sk)->copy_thresh = val; 3685 return 0; 3686 } 3687 case PACKET_VERSION: 3688 { 3689 int val; 3690 3691 if (optlen != sizeof(val)) 3692 return -EINVAL; 3693 if (copy_from_user(&val, optval, sizeof(val))) 3694 return -EFAULT; 3695 switch (val) { 3696 case TPACKET_V1: 3697 case TPACKET_V2: 3698 case TPACKET_V3: 3699 break; 3700 default: 3701 return -EINVAL; 3702 } 3703 lock_sock(sk); 3704 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3705 ret = -EBUSY; 3706 } else { 3707 po->tp_version = val; 3708 ret = 0; 3709 } 3710 release_sock(sk); 3711 return ret; 3712 } 3713 case PACKET_RESERVE: 3714 { 3715 unsigned int val; 3716 3717 if (optlen != sizeof(val)) 3718 return -EINVAL; 3719 if (copy_from_user(&val, optval, sizeof(val))) 3720 return -EFAULT; 3721 if (val > INT_MAX) 3722 return -EINVAL; 3723 lock_sock(sk); 3724 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3725 ret = -EBUSY; 3726 } else { 3727 po->tp_reserve = val; 3728 ret = 0; 3729 } 3730 release_sock(sk); 3731 return ret; 3732 } 3733 case PACKET_LOSS: 3734 { 3735 unsigned int val; 3736 3737 if (optlen != sizeof(val)) 3738 return -EINVAL; 3739 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3740 return -EBUSY; 3741 if (copy_from_user(&val, optval, sizeof(val))) 3742 return -EFAULT; 3743 po->tp_loss = !!val; 3744 return 0; 3745 } 3746 case PACKET_AUXDATA: 3747 { 3748 int val; 3749 3750 if (optlen < sizeof(val)) 3751 return -EINVAL; 3752 if (copy_from_user(&val, optval, sizeof(val))) 3753 return -EFAULT; 3754 3755 po->auxdata = !!val; 3756 return 0; 3757 } 3758 case PACKET_ORIGDEV: 3759 { 3760 int val; 3761 3762 if (optlen < sizeof(val)) 3763 return -EINVAL; 3764 if (copy_from_user(&val, optval, sizeof(val))) 3765 return -EFAULT; 3766 3767 po->origdev = !!val; 3768 return 0; 3769 } 3770 case PACKET_VNET_HDR: 3771 { 3772 int val; 3773 3774 if (sock->type != SOCK_RAW) 3775 return -EINVAL; 3776 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3777 return -EBUSY; 3778 if (optlen < sizeof(val)) 3779 return -EINVAL; 3780 if (copy_from_user(&val, optval, sizeof(val))) 3781 return -EFAULT; 3782 3783 po->has_vnet_hdr = !!val; 3784 return 0; 3785 } 3786 case PACKET_TIMESTAMP: 3787 { 3788 int val; 3789 3790 if (optlen != sizeof(val)) 3791 return -EINVAL; 3792 if (copy_from_user(&val, optval, sizeof(val))) 3793 return -EFAULT; 3794 3795 po->tp_tstamp = val; 3796 return 0; 3797 } 3798 case PACKET_FANOUT: 3799 { 3800 int val; 3801 3802 if (optlen != sizeof(val)) 3803 return -EINVAL; 3804 if (copy_from_user(&val, optval, sizeof(val))) 3805 return -EFAULT; 3806 3807 return fanout_add(sk, val & 0xffff, val >> 16); 3808 } 3809 case PACKET_FANOUT_DATA: 3810 { 3811 if (!po->fanout) 3812 return -EINVAL; 3813 3814 return fanout_set_data(po, optval, optlen); 3815 } 3816 case PACKET_TX_HAS_OFF: 3817 { 3818 unsigned int val; 3819 3820 if (optlen != sizeof(val)) 3821 return -EINVAL; 3822 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3823 return -EBUSY; 3824 if (copy_from_user(&val, optval, sizeof(val))) 3825 return -EFAULT; 3826 po->tp_tx_has_off = !!val; 3827 return 0; 3828 } 3829 case PACKET_QDISC_BYPASS: 3830 { 3831 int val; 3832 3833 if (optlen != sizeof(val)) 3834 return -EINVAL; 3835 if (copy_from_user(&val, optval, sizeof(val))) 3836 return -EFAULT; 3837 3838 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; 3839 return 0; 3840 } 3841 default: 3842 return -ENOPROTOOPT; 3843 } 3844 } 3845 3846 static int packet_getsockopt(struct socket *sock, int level, int optname, 3847 char __user *optval, int __user *optlen) 3848 { 3849 int len; 3850 int val, lv = sizeof(val); 3851 struct sock *sk = sock->sk; 3852 struct packet_sock *po = pkt_sk(sk); 3853 void *data = &val; 3854 union tpacket_stats_u st; 3855 struct tpacket_rollover_stats rstats; 3856 struct packet_rollover *rollover; 3857 3858 if (level != SOL_PACKET) 3859 return -ENOPROTOOPT; 3860 3861 if (get_user(len, optlen)) 3862 return -EFAULT; 3863 3864 if (len < 0) 3865 return -EINVAL; 3866 3867 switch (optname) { 3868 case PACKET_STATISTICS: 3869 spin_lock_bh(&sk->sk_receive_queue.lock); 3870 memcpy(&st, &po->stats, sizeof(st)); 3871 memset(&po->stats, 0, sizeof(po->stats)); 3872 spin_unlock_bh(&sk->sk_receive_queue.lock); 3873 3874 if (po->tp_version == TPACKET_V3) { 3875 lv = sizeof(struct tpacket_stats_v3); 3876 st.stats3.tp_packets += st.stats3.tp_drops; 3877 data = &st.stats3; 3878 } else { 3879 lv = sizeof(struct tpacket_stats); 3880 st.stats1.tp_packets += st.stats1.tp_drops; 3881 data = &st.stats1; 3882 } 3883 3884 break; 3885 case PACKET_AUXDATA: 3886 val = po->auxdata; 3887 break; 3888 case PACKET_ORIGDEV: 3889 val = po->origdev; 3890 break; 3891 case PACKET_VNET_HDR: 3892 val = po->has_vnet_hdr; 3893 break; 3894 case PACKET_VERSION: 3895 val = po->tp_version; 3896 break; 3897 case PACKET_HDRLEN: 3898 if (len > sizeof(int)) 3899 len = sizeof(int); 3900 if (len < sizeof(int)) 3901 return -EINVAL; 3902 if (copy_from_user(&val, optval, len)) 3903 return -EFAULT; 3904 switch (val) { 3905 case TPACKET_V1: 3906 val = sizeof(struct tpacket_hdr); 3907 break; 3908 case TPACKET_V2: 3909 val = sizeof(struct tpacket2_hdr); 3910 break; 3911 case TPACKET_V3: 3912 val = sizeof(struct tpacket3_hdr); 3913 break; 3914 default: 3915 return -EINVAL; 3916 } 3917 break; 3918 case PACKET_RESERVE: 3919 val = po->tp_reserve; 3920 break; 3921 case PACKET_LOSS: 3922 val = po->tp_loss; 3923 break; 3924 case PACKET_TIMESTAMP: 3925 val = po->tp_tstamp; 3926 break; 3927 case PACKET_FANOUT: 3928 val = (po->fanout ? 3929 ((u32)po->fanout->id | 3930 ((u32)po->fanout->type << 16) | 3931 ((u32)po->fanout->flags << 24)) : 3932 0); 3933 break; 3934 case PACKET_ROLLOVER_STATS: 3935 rcu_read_lock(); 3936 rollover = rcu_dereference(po->rollover); 3937 if (rollover) { 3938 rstats.tp_all = atomic_long_read(&rollover->num); 3939 rstats.tp_huge = atomic_long_read(&rollover->num_huge); 3940 rstats.tp_failed = atomic_long_read(&rollover->num_failed); 3941 data = &rstats; 3942 lv = sizeof(rstats); 3943 } 3944 rcu_read_unlock(); 3945 if (!rollover) 3946 return -EINVAL; 3947 break; 3948 case PACKET_TX_HAS_OFF: 3949 val = po->tp_tx_has_off; 3950 break; 3951 case PACKET_QDISC_BYPASS: 3952 val = packet_use_direct_xmit(po); 3953 break; 3954 default: 3955 return -ENOPROTOOPT; 3956 } 3957 3958 if (len > lv) 3959 len = lv; 3960 if (put_user(len, optlen)) 3961 return -EFAULT; 3962 if (copy_to_user(optval, data, len)) 3963 return -EFAULT; 3964 return 0; 3965 } 3966 3967 3968 #ifdef CONFIG_COMPAT 3969 static int compat_packet_setsockopt(struct socket *sock, int level, int optname, 3970 char __user *optval, unsigned int optlen) 3971 { 3972 struct packet_sock *po = pkt_sk(sock->sk); 3973 3974 if (level != SOL_PACKET) 3975 return -ENOPROTOOPT; 3976 3977 if (optname == PACKET_FANOUT_DATA && 3978 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) { 3979 optval = (char __user *)get_compat_bpf_fprog(optval); 3980 if (!optval) 3981 return -EFAULT; 3982 optlen = sizeof(struct sock_fprog); 3983 } 3984 3985 return packet_setsockopt(sock, level, optname, optval, optlen); 3986 } 3987 #endif 3988 3989 static int packet_notifier(struct notifier_block *this, 3990 unsigned long msg, void *ptr) 3991 { 3992 struct sock *sk; 3993 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 3994 struct net *net = dev_net(dev); 3995 3996 rcu_read_lock(); 3997 sk_for_each_rcu(sk, &net->packet.sklist) { 3998 struct packet_sock *po = pkt_sk(sk); 3999 4000 switch (msg) { 4001 case NETDEV_UNREGISTER: 4002 if (po->mclist) 4003 packet_dev_mclist_delete(dev, &po->mclist); 4004 /* fallthrough */ 4005 4006 case NETDEV_DOWN: 4007 if (dev->ifindex == po->ifindex) { 4008 spin_lock(&po->bind_lock); 4009 if (po->running) { 4010 __unregister_prot_hook(sk, false); 4011 sk->sk_err = ENETDOWN; 4012 if (!sock_flag(sk, SOCK_DEAD)) 4013 sk->sk_error_report(sk); 4014 } 4015 if (msg == NETDEV_UNREGISTER) { 4016 packet_cached_dev_reset(po); 4017 po->ifindex = -1; 4018 if (po->prot_hook.dev) 4019 dev_put(po->prot_hook.dev); 4020 po->prot_hook.dev = NULL; 4021 } 4022 spin_unlock(&po->bind_lock); 4023 } 4024 break; 4025 case NETDEV_UP: 4026 if (dev->ifindex == po->ifindex) { 4027 spin_lock(&po->bind_lock); 4028 if (po->num) 4029 register_prot_hook(sk); 4030 spin_unlock(&po->bind_lock); 4031 } 4032 break; 4033 } 4034 } 4035 rcu_read_unlock(); 4036 return NOTIFY_DONE; 4037 } 4038 4039 4040 static int packet_ioctl(struct socket *sock, unsigned int cmd, 4041 unsigned long arg) 4042 { 4043 struct sock *sk = sock->sk; 4044 4045 switch (cmd) { 4046 case SIOCOUTQ: 4047 { 4048 int amount = sk_wmem_alloc_get(sk); 4049 4050 return put_user(amount, (int __user *)arg); 4051 } 4052 case SIOCINQ: 4053 { 4054 struct sk_buff *skb; 4055 int amount = 0; 4056 4057 spin_lock_bh(&sk->sk_receive_queue.lock); 4058 skb = skb_peek(&sk->sk_receive_queue); 4059 if (skb) 4060 amount = skb->len; 4061 spin_unlock_bh(&sk->sk_receive_queue.lock); 4062 return put_user(amount, (int __user *)arg); 4063 } 4064 case SIOCGSTAMP: 4065 return sock_get_timestamp(sk, (struct timeval __user *)arg); 4066 case SIOCGSTAMPNS: 4067 return sock_get_timestampns(sk, (struct timespec __user *)arg); 4068 4069 #ifdef CONFIG_INET 4070 case SIOCADDRT: 4071 case SIOCDELRT: 4072 case SIOCDARP: 4073 case SIOCGARP: 4074 case SIOCSARP: 4075 case SIOCGIFADDR: 4076 case SIOCSIFADDR: 4077 case SIOCGIFBRDADDR: 4078 case SIOCSIFBRDADDR: 4079 case SIOCGIFNETMASK: 4080 case SIOCSIFNETMASK: 4081 case SIOCGIFDSTADDR: 4082 case SIOCSIFDSTADDR: 4083 case SIOCSIFFLAGS: 4084 return inet_dgram_ops.ioctl(sock, cmd, arg); 4085 #endif 4086 4087 default: 4088 return -ENOIOCTLCMD; 4089 } 4090 return 0; 4091 } 4092 4093 static unsigned int packet_poll(struct file *file, struct socket *sock, 4094 poll_table *wait) 4095 { 4096 struct sock *sk = sock->sk; 4097 struct packet_sock *po = pkt_sk(sk); 4098 unsigned int mask = datagram_poll(file, sock, wait); 4099 4100 spin_lock_bh(&sk->sk_receive_queue.lock); 4101 if (po->rx_ring.pg_vec) { 4102 if (!packet_previous_rx_frame(po, &po->rx_ring, 4103 TP_STATUS_KERNEL)) 4104 mask |= POLLIN | POLLRDNORM; 4105 } 4106 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 4107 po->pressure = 0; 4108 spin_unlock_bh(&sk->sk_receive_queue.lock); 4109 spin_lock_bh(&sk->sk_write_queue.lock); 4110 if (po->tx_ring.pg_vec) { 4111 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4112 mask |= POLLOUT | POLLWRNORM; 4113 } 4114 spin_unlock_bh(&sk->sk_write_queue.lock); 4115 return mask; 4116 } 4117 4118 4119 /* Dirty? Well, I still did not learn better way to account 4120 * for user mmaps. 4121 */ 4122 4123 static void packet_mm_open(struct vm_area_struct *vma) 4124 { 4125 struct file *file = vma->vm_file; 4126 struct socket *sock = file->private_data; 4127 struct sock *sk = sock->sk; 4128 4129 if (sk) 4130 atomic_inc(&pkt_sk(sk)->mapped); 4131 } 4132 4133 static void packet_mm_close(struct vm_area_struct *vma) 4134 { 4135 struct file *file = vma->vm_file; 4136 struct socket *sock = file->private_data; 4137 struct sock *sk = sock->sk; 4138 4139 if (sk) 4140 atomic_dec(&pkt_sk(sk)->mapped); 4141 } 4142 4143 static const struct vm_operations_struct packet_mmap_ops = { 4144 .open = packet_mm_open, 4145 .close = packet_mm_close, 4146 }; 4147 4148 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4149 unsigned int len) 4150 { 4151 int i; 4152 4153 for (i = 0; i < len; i++) { 4154 if (likely(pg_vec[i].buffer)) { 4155 if (is_vmalloc_addr(pg_vec[i].buffer)) 4156 vfree(pg_vec[i].buffer); 4157 else 4158 free_pages((unsigned long)pg_vec[i].buffer, 4159 order); 4160 pg_vec[i].buffer = NULL; 4161 } 4162 } 4163 kfree(pg_vec); 4164 } 4165 4166 static char *alloc_one_pg_vec_page(unsigned long order) 4167 { 4168 char *buffer; 4169 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4170 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4171 4172 buffer = (char *) __get_free_pages(gfp_flags, order); 4173 if (buffer) 4174 return buffer; 4175 4176 /* __get_free_pages failed, fall back to vmalloc */ 4177 buffer = vzalloc((1 << order) * PAGE_SIZE); 4178 if (buffer) 4179 return buffer; 4180 4181 /* vmalloc failed, lets dig into swap here */ 4182 gfp_flags &= ~__GFP_NORETRY; 4183 buffer = (char *) __get_free_pages(gfp_flags, order); 4184 if (buffer) 4185 return buffer; 4186 4187 /* complete and utter failure */ 4188 return NULL; 4189 } 4190 4191 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4192 { 4193 unsigned int block_nr = req->tp_block_nr; 4194 struct pgv *pg_vec; 4195 int i; 4196 4197 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL); 4198 if (unlikely(!pg_vec)) 4199 goto out; 4200 4201 for (i = 0; i < block_nr; i++) { 4202 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4203 if (unlikely(!pg_vec[i].buffer)) 4204 goto out_free_pgvec; 4205 } 4206 4207 out: 4208 return pg_vec; 4209 4210 out_free_pgvec: 4211 free_pg_vec(pg_vec, order, block_nr); 4212 pg_vec = NULL; 4213 goto out; 4214 } 4215 4216 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4217 int closing, int tx_ring) 4218 { 4219 struct pgv *pg_vec = NULL; 4220 struct packet_sock *po = pkt_sk(sk); 4221 int was_running, order = 0; 4222 struct packet_ring_buffer *rb; 4223 struct sk_buff_head *rb_queue; 4224 __be16 num; 4225 int err = -EINVAL; 4226 /* Added to avoid minimal code churn */ 4227 struct tpacket_req *req = &req_u->req; 4228 4229 lock_sock(sk); 4230 4231 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4232 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4233 4234 err = -EBUSY; 4235 if (!closing) { 4236 if (atomic_read(&po->mapped)) 4237 goto out; 4238 if (packet_read_pending(rb)) 4239 goto out; 4240 } 4241 4242 if (req->tp_block_nr) { 4243 /* Sanity tests and some calculations */ 4244 err = -EBUSY; 4245 if (unlikely(rb->pg_vec)) 4246 goto out; 4247 4248 switch (po->tp_version) { 4249 case TPACKET_V1: 4250 po->tp_hdrlen = TPACKET_HDRLEN; 4251 break; 4252 case TPACKET_V2: 4253 po->tp_hdrlen = TPACKET2_HDRLEN; 4254 break; 4255 case TPACKET_V3: 4256 po->tp_hdrlen = TPACKET3_HDRLEN; 4257 break; 4258 } 4259 4260 err = -EINVAL; 4261 if (unlikely((int)req->tp_block_size <= 0)) 4262 goto out; 4263 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4264 goto out; 4265 if (po->tp_version >= TPACKET_V3 && 4266 req->tp_block_size <= 4267 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv)) 4268 goto out; 4269 if (unlikely(req->tp_frame_size < po->tp_hdrlen + 4270 po->tp_reserve)) 4271 goto out; 4272 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4273 goto out; 4274 4275 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4276 if (unlikely(rb->frames_per_block == 0)) 4277 goto out; 4278 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr)) 4279 goto out; 4280 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4281 req->tp_frame_nr)) 4282 goto out; 4283 4284 err = -ENOMEM; 4285 order = get_order(req->tp_block_size); 4286 pg_vec = alloc_pg_vec(req, order); 4287 if (unlikely(!pg_vec)) 4288 goto out; 4289 switch (po->tp_version) { 4290 case TPACKET_V3: 4291 /* Block transmit is not supported yet */ 4292 if (!tx_ring) { 4293 init_prb_bdqc(po, rb, pg_vec, req_u); 4294 } else { 4295 struct tpacket_req3 *req3 = &req_u->req3; 4296 4297 if (req3->tp_retire_blk_tov || 4298 req3->tp_sizeof_priv || 4299 req3->tp_feature_req_word) { 4300 err = -EINVAL; 4301 goto out; 4302 } 4303 } 4304 break; 4305 default: 4306 break; 4307 } 4308 } 4309 /* Done */ 4310 else { 4311 err = -EINVAL; 4312 if (unlikely(req->tp_frame_nr)) 4313 goto out; 4314 } 4315 4316 4317 /* Detach socket from network */ 4318 spin_lock(&po->bind_lock); 4319 was_running = po->running; 4320 num = po->num; 4321 if (was_running) { 4322 po->num = 0; 4323 __unregister_prot_hook(sk, false); 4324 } 4325 spin_unlock(&po->bind_lock); 4326 4327 synchronize_net(); 4328 4329 err = -EBUSY; 4330 mutex_lock(&po->pg_vec_lock); 4331 if (closing || atomic_read(&po->mapped) == 0) { 4332 err = 0; 4333 spin_lock_bh(&rb_queue->lock); 4334 swap(rb->pg_vec, pg_vec); 4335 rb->frame_max = (req->tp_frame_nr - 1); 4336 rb->head = 0; 4337 rb->frame_size = req->tp_frame_size; 4338 spin_unlock_bh(&rb_queue->lock); 4339 4340 swap(rb->pg_vec_order, order); 4341 swap(rb->pg_vec_len, req->tp_block_nr); 4342 4343 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4344 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4345 tpacket_rcv : packet_rcv; 4346 skb_queue_purge(rb_queue); 4347 if (atomic_read(&po->mapped)) 4348 pr_err("packet_mmap: vma is busy: %d\n", 4349 atomic_read(&po->mapped)); 4350 } 4351 mutex_unlock(&po->pg_vec_lock); 4352 4353 spin_lock(&po->bind_lock); 4354 if (was_running) { 4355 po->num = num; 4356 register_prot_hook(sk); 4357 } 4358 spin_unlock(&po->bind_lock); 4359 if (pg_vec && (po->tp_version > TPACKET_V2)) { 4360 /* Because we don't support block-based V3 on tx-ring */ 4361 if (!tx_ring) 4362 prb_shutdown_retire_blk_timer(po, rb_queue); 4363 } 4364 4365 if (pg_vec) 4366 free_pg_vec(pg_vec, order, req->tp_block_nr); 4367 out: 4368 release_sock(sk); 4369 return err; 4370 } 4371 4372 static int packet_mmap(struct file *file, struct socket *sock, 4373 struct vm_area_struct *vma) 4374 { 4375 struct sock *sk = sock->sk; 4376 struct packet_sock *po = pkt_sk(sk); 4377 unsigned long size, expected_size; 4378 struct packet_ring_buffer *rb; 4379 unsigned long start; 4380 int err = -EINVAL; 4381 int i; 4382 4383 if (vma->vm_pgoff) 4384 return -EINVAL; 4385 4386 mutex_lock(&po->pg_vec_lock); 4387 4388 expected_size = 0; 4389 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4390 if (rb->pg_vec) { 4391 expected_size += rb->pg_vec_len 4392 * rb->pg_vec_pages 4393 * PAGE_SIZE; 4394 } 4395 } 4396 4397 if (expected_size == 0) 4398 goto out; 4399 4400 size = vma->vm_end - vma->vm_start; 4401 if (size != expected_size) 4402 goto out; 4403 4404 start = vma->vm_start; 4405 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4406 if (rb->pg_vec == NULL) 4407 continue; 4408 4409 for (i = 0; i < rb->pg_vec_len; i++) { 4410 struct page *page; 4411 void *kaddr = rb->pg_vec[i].buffer; 4412 int pg_num; 4413 4414 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4415 page = pgv_to_page(kaddr); 4416 err = vm_insert_page(vma, start, page); 4417 if (unlikely(err)) 4418 goto out; 4419 start += PAGE_SIZE; 4420 kaddr += PAGE_SIZE; 4421 } 4422 } 4423 } 4424 4425 atomic_inc(&po->mapped); 4426 vma->vm_ops = &packet_mmap_ops; 4427 err = 0; 4428 4429 out: 4430 mutex_unlock(&po->pg_vec_lock); 4431 return err; 4432 } 4433 4434 static const struct proto_ops packet_ops_spkt = { 4435 .family = PF_PACKET, 4436 .owner = THIS_MODULE, 4437 .release = packet_release, 4438 .bind = packet_bind_spkt, 4439 .connect = sock_no_connect, 4440 .socketpair = sock_no_socketpair, 4441 .accept = sock_no_accept, 4442 .getname = packet_getname_spkt, 4443 .poll = datagram_poll, 4444 .ioctl = packet_ioctl, 4445 .listen = sock_no_listen, 4446 .shutdown = sock_no_shutdown, 4447 .setsockopt = sock_no_setsockopt, 4448 .getsockopt = sock_no_getsockopt, 4449 .sendmsg = packet_sendmsg_spkt, 4450 .recvmsg = packet_recvmsg, 4451 .mmap = sock_no_mmap, 4452 .sendpage = sock_no_sendpage, 4453 }; 4454 4455 static const struct proto_ops packet_ops = { 4456 .family = PF_PACKET, 4457 .owner = THIS_MODULE, 4458 .release = packet_release, 4459 .bind = packet_bind, 4460 .connect = sock_no_connect, 4461 .socketpair = sock_no_socketpair, 4462 .accept = sock_no_accept, 4463 .getname = packet_getname, 4464 .poll = packet_poll, 4465 .ioctl = packet_ioctl, 4466 .listen = sock_no_listen, 4467 .shutdown = sock_no_shutdown, 4468 .setsockopt = packet_setsockopt, 4469 .getsockopt = packet_getsockopt, 4470 #ifdef CONFIG_COMPAT 4471 .compat_setsockopt = compat_packet_setsockopt, 4472 #endif 4473 .sendmsg = packet_sendmsg, 4474 .recvmsg = packet_recvmsg, 4475 .mmap = packet_mmap, 4476 .sendpage = sock_no_sendpage, 4477 }; 4478 4479 static const struct net_proto_family packet_family_ops = { 4480 .family = PF_PACKET, 4481 .create = packet_create, 4482 .owner = THIS_MODULE, 4483 }; 4484 4485 static struct notifier_block packet_netdev_notifier = { 4486 .notifier_call = packet_notifier, 4487 }; 4488 4489 #ifdef CONFIG_PROC_FS 4490 4491 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4492 __acquires(RCU) 4493 { 4494 struct net *net = seq_file_net(seq); 4495 4496 rcu_read_lock(); 4497 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4498 } 4499 4500 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4501 { 4502 struct net *net = seq_file_net(seq); 4503 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4504 } 4505 4506 static void packet_seq_stop(struct seq_file *seq, void *v) 4507 __releases(RCU) 4508 { 4509 rcu_read_unlock(); 4510 } 4511 4512 static int packet_seq_show(struct seq_file *seq, void *v) 4513 { 4514 if (v == SEQ_START_TOKEN) 4515 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4516 else { 4517 struct sock *s = sk_entry(v); 4518 const struct packet_sock *po = pkt_sk(s); 4519 4520 seq_printf(seq, 4521 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4522 s, 4523 refcount_read(&s->sk_refcnt), 4524 s->sk_type, 4525 ntohs(po->num), 4526 po->ifindex, 4527 po->running, 4528 atomic_read(&s->sk_rmem_alloc), 4529 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4530 sock_i_ino(s)); 4531 } 4532 4533 return 0; 4534 } 4535 4536 static const struct seq_operations packet_seq_ops = { 4537 .start = packet_seq_start, 4538 .next = packet_seq_next, 4539 .stop = packet_seq_stop, 4540 .show = packet_seq_show, 4541 }; 4542 4543 static int packet_seq_open(struct inode *inode, struct file *file) 4544 { 4545 return seq_open_net(inode, file, &packet_seq_ops, 4546 sizeof(struct seq_net_private)); 4547 } 4548 4549 static const struct file_operations packet_seq_fops = { 4550 .owner = THIS_MODULE, 4551 .open = packet_seq_open, 4552 .read = seq_read, 4553 .llseek = seq_lseek, 4554 .release = seq_release_net, 4555 }; 4556 4557 #endif 4558 4559 static int __net_init packet_net_init(struct net *net) 4560 { 4561 mutex_init(&net->packet.sklist_lock); 4562 INIT_HLIST_HEAD(&net->packet.sklist); 4563 4564 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops)) 4565 return -ENOMEM; 4566 4567 return 0; 4568 } 4569 4570 static void __net_exit packet_net_exit(struct net *net) 4571 { 4572 remove_proc_entry("packet", net->proc_net); 4573 } 4574 4575 static struct pernet_operations packet_net_ops = { 4576 .init = packet_net_init, 4577 .exit = packet_net_exit, 4578 }; 4579 4580 4581 static void __exit packet_exit(void) 4582 { 4583 unregister_netdevice_notifier(&packet_netdev_notifier); 4584 unregister_pernet_subsys(&packet_net_ops); 4585 sock_unregister(PF_PACKET); 4586 proto_unregister(&packet_proto); 4587 } 4588 4589 static int __init packet_init(void) 4590 { 4591 int rc = proto_register(&packet_proto, 0); 4592 4593 if (rc != 0) 4594 goto out; 4595 4596 sock_register(&packet_family_ops); 4597 register_pernet_subsys(&packet_net_ops); 4598 register_netdevice_notifier(&packet_netdev_notifier); 4599 out: 4600 return rc; 4601 } 4602 4603 module_init(packet_init); 4604 module_exit(packet_exit); 4605 MODULE_LICENSE("GPL"); 4606 MODULE_ALIAS_NETPROTO(PF_PACKET); 4607