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