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 PGV_FROM_VMALLOC 1 181 182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) 183 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) 184 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) 185 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) 186 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) 187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) 188 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x))) 189 190 struct packet_sock; 191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg); 192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 193 struct packet_type *pt, struct net_device *orig_dev); 194 195 static void *packet_previous_frame(struct packet_sock *po, 196 struct packet_ring_buffer *rb, 197 int status); 198 static void packet_increment_head(struct packet_ring_buffer *buff); 199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *, 200 struct tpacket_block_desc *); 201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 202 struct packet_sock *); 203 static void prb_retire_current_block(struct tpacket_kbdq_core *, 204 struct packet_sock *, unsigned int status); 205 static int prb_queue_frozen(struct tpacket_kbdq_core *); 206 static void prb_open_block(struct tpacket_kbdq_core *, 207 struct tpacket_block_desc *); 208 static void prb_retire_rx_blk_timer_expired(unsigned long); 209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); 210 static void prb_init_blk_timer(struct packet_sock *, 211 struct tpacket_kbdq_core *, 212 void (*func) (unsigned long)); 213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 214 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 215 struct tpacket3_hdr *); 216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 217 struct tpacket3_hdr *); 218 static void packet_flush_mclist(struct sock *sk); 219 220 struct packet_skb_cb { 221 union { 222 struct sockaddr_pkt pkt; 223 union { 224 /* Trick: alias skb original length with 225 * ll.sll_family and ll.protocol in order 226 * to save room. 227 */ 228 unsigned int origlen; 229 struct sockaddr_ll ll; 230 }; 231 } sa; 232 }; 233 234 #define vio_le() virtio_legacy_is_little_endian() 235 236 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 237 238 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 239 #define GET_PBLOCK_DESC(x, bid) \ 240 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 242 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 243 #define GET_NEXT_PRB_BLK_NUM(x) \ 244 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 245 ((x)->kactive_blk_num+1) : 0) 246 247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 248 static void __fanout_link(struct sock *sk, struct packet_sock *po); 249 250 static int packet_direct_xmit(struct sk_buff *skb) 251 { 252 struct net_device *dev = skb->dev; 253 struct sk_buff *orig_skb = skb; 254 struct netdev_queue *txq; 255 int ret = NETDEV_TX_BUSY; 256 257 if (unlikely(!netif_running(dev) || 258 !netif_carrier_ok(dev))) 259 goto drop; 260 261 skb = validate_xmit_skb_list(skb, dev); 262 if (skb != orig_skb) 263 goto drop; 264 265 txq = skb_get_tx_queue(dev, skb); 266 267 local_bh_disable(); 268 269 HARD_TX_LOCK(dev, txq, smp_processor_id()); 270 if (!netif_xmit_frozen_or_drv_stopped(txq)) 271 ret = netdev_start_xmit(skb, dev, txq, false); 272 HARD_TX_UNLOCK(dev, txq); 273 274 local_bh_enable(); 275 276 if (!dev_xmit_complete(ret)) 277 kfree_skb(skb); 278 279 return ret; 280 drop: 281 atomic_long_inc(&dev->tx_dropped); 282 kfree_skb_list(skb); 283 return NET_XMIT_DROP; 284 } 285 286 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 287 { 288 struct net_device *dev; 289 290 rcu_read_lock(); 291 dev = rcu_dereference(po->cached_dev); 292 if (likely(dev)) 293 dev_hold(dev); 294 rcu_read_unlock(); 295 296 return dev; 297 } 298 299 static void packet_cached_dev_assign(struct packet_sock *po, 300 struct net_device *dev) 301 { 302 rcu_assign_pointer(po->cached_dev, dev); 303 } 304 305 static void packet_cached_dev_reset(struct packet_sock *po) 306 { 307 RCU_INIT_POINTER(po->cached_dev, NULL); 308 } 309 310 static bool packet_use_direct_xmit(const struct packet_sock *po) 311 { 312 return po->xmit == packet_direct_xmit; 313 } 314 315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 316 { 317 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues; 318 } 319 320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 321 { 322 const struct net_device_ops *ops = dev->netdev_ops; 323 u16 queue_index; 324 325 if (ops->ndo_select_queue) { 326 queue_index = ops->ndo_select_queue(dev, skb, NULL, 327 __packet_pick_tx_queue); 328 queue_index = netdev_cap_txqueue(dev, queue_index); 329 } else { 330 queue_index = __packet_pick_tx_queue(dev, skb); 331 } 332 333 skb_set_queue_mapping(skb, queue_index); 334 } 335 336 /* register_prot_hook must be invoked with the po->bind_lock held, 337 * or from a context in which asynchronous accesses to the packet 338 * socket is not possible (packet_create()). 339 */ 340 static void register_prot_hook(struct sock *sk) 341 { 342 struct packet_sock *po = pkt_sk(sk); 343 344 if (!po->running) { 345 if (po->fanout) 346 __fanout_link(sk, po); 347 else 348 dev_add_pack(&po->prot_hook); 349 350 sock_hold(sk); 351 po->running = 1; 352 } 353 } 354 355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock 356 * held. If the sync parameter is true, we will temporarily drop 357 * the po->bind_lock and do a synchronize_net to make sure no 358 * asynchronous packet processing paths still refer to the elements 359 * of po->prot_hook. If the sync parameter is false, it is the 360 * callers responsibility to take care of this. 361 */ 362 static void __unregister_prot_hook(struct sock *sk, bool sync) 363 { 364 struct packet_sock *po = pkt_sk(sk); 365 366 po->running = 0; 367 368 if (po->fanout) 369 __fanout_unlink(sk, po); 370 else 371 __dev_remove_pack(&po->prot_hook); 372 373 __sock_put(sk); 374 375 if (sync) { 376 spin_unlock(&po->bind_lock); 377 synchronize_net(); 378 spin_lock(&po->bind_lock); 379 } 380 } 381 382 static void unregister_prot_hook(struct sock *sk, bool sync) 383 { 384 struct packet_sock *po = pkt_sk(sk); 385 386 if (po->running) 387 __unregister_prot_hook(sk, sync); 388 } 389 390 static inline struct page * __pure pgv_to_page(void *addr) 391 { 392 if (is_vmalloc_addr(addr)) 393 return vmalloc_to_page(addr); 394 return virt_to_page(addr); 395 } 396 397 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 398 { 399 union tpacket_uhdr h; 400 401 h.raw = frame; 402 switch (po->tp_version) { 403 case TPACKET_V1: 404 h.h1->tp_status = status; 405 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 406 break; 407 case TPACKET_V2: 408 h.h2->tp_status = status; 409 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 410 break; 411 case TPACKET_V3: 412 h.h3->tp_status = status; 413 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 414 break; 415 default: 416 WARN(1, "TPACKET version not supported.\n"); 417 BUG(); 418 } 419 420 smp_wmb(); 421 } 422 423 static int __packet_get_status(struct packet_sock *po, void *frame) 424 { 425 union tpacket_uhdr h; 426 427 smp_rmb(); 428 429 h.raw = frame; 430 switch (po->tp_version) { 431 case TPACKET_V1: 432 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 433 return h.h1->tp_status; 434 case TPACKET_V2: 435 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 436 return h.h2->tp_status; 437 case TPACKET_V3: 438 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 439 return h.h3->tp_status; 440 default: 441 WARN(1, "TPACKET version not supported.\n"); 442 BUG(); 443 return 0; 444 } 445 } 446 447 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, 448 unsigned int flags) 449 { 450 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 451 452 if (shhwtstamps && 453 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 454 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) 455 return TP_STATUS_TS_RAW_HARDWARE; 456 457 if (ktime_to_timespec_cond(skb->tstamp, ts)) 458 return TP_STATUS_TS_SOFTWARE; 459 460 return 0; 461 } 462 463 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 464 struct sk_buff *skb) 465 { 466 union tpacket_uhdr h; 467 struct timespec ts; 468 __u32 ts_status; 469 470 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 471 return 0; 472 473 h.raw = frame; 474 switch (po->tp_version) { 475 case TPACKET_V1: 476 h.h1->tp_sec = ts.tv_sec; 477 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 478 break; 479 case TPACKET_V2: 480 h.h2->tp_sec = ts.tv_sec; 481 h.h2->tp_nsec = ts.tv_nsec; 482 break; 483 case TPACKET_V3: 484 h.h3->tp_sec = ts.tv_sec; 485 h.h3->tp_nsec = ts.tv_nsec; 486 break; 487 default: 488 WARN(1, "TPACKET version not supported.\n"); 489 BUG(); 490 } 491 492 /* one flush is safe, as both fields always lie on the same cacheline */ 493 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 494 smp_wmb(); 495 496 return ts_status; 497 } 498 499 static void *packet_lookup_frame(struct packet_sock *po, 500 struct packet_ring_buffer *rb, 501 unsigned int position, 502 int status) 503 { 504 unsigned int pg_vec_pos, frame_offset; 505 union tpacket_uhdr h; 506 507 pg_vec_pos = position / rb->frames_per_block; 508 frame_offset = position % rb->frames_per_block; 509 510 h.raw = rb->pg_vec[pg_vec_pos].buffer + 511 (frame_offset * rb->frame_size); 512 513 if (status != __packet_get_status(po, h.raw)) 514 return NULL; 515 516 return h.raw; 517 } 518 519 static void *packet_current_frame(struct packet_sock *po, 520 struct packet_ring_buffer *rb, 521 int status) 522 { 523 return packet_lookup_frame(po, rb, rb->head, status); 524 } 525 526 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 527 { 528 del_timer_sync(&pkc->retire_blk_timer); 529 } 530 531 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 532 struct sk_buff_head *rb_queue) 533 { 534 struct tpacket_kbdq_core *pkc; 535 536 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 537 538 spin_lock_bh(&rb_queue->lock); 539 pkc->delete_blk_timer = 1; 540 spin_unlock_bh(&rb_queue->lock); 541 542 prb_del_retire_blk_timer(pkc); 543 } 544 545 static void prb_init_blk_timer(struct packet_sock *po, 546 struct tpacket_kbdq_core *pkc, 547 void (*func) (unsigned long)) 548 { 549 init_timer(&pkc->retire_blk_timer); 550 pkc->retire_blk_timer.data = (long)po; 551 pkc->retire_blk_timer.function = func; 552 pkc->retire_blk_timer.expires = jiffies; 553 } 554 555 static void prb_setup_retire_blk_timer(struct packet_sock *po) 556 { 557 struct tpacket_kbdq_core *pkc; 558 559 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 560 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired); 561 } 562 563 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 564 int blk_size_in_bytes) 565 { 566 struct net_device *dev; 567 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; 568 struct ethtool_link_ksettings ecmd; 569 int err; 570 571 rtnl_lock(); 572 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 573 if (unlikely(!dev)) { 574 rtnl_unlock(); 575 return DEFAULT_PRB_RETIRE_TOV; 576 } 577 err = __ethtool_get_link_ksettings(dev, &ecmd); 578 rtnl_unlock(); 579 if (!err) { 580 /* 581 * If the link speed is so slow you don't really 582 * need to worry about perf anyways 583 */ 584 if (ecmd.base.speed < SPEED_1000 || 585 ecmd.base.speed == SPEED_UNKNOWN) { 586 return DEFAULT_PRB_RETIRE_TOV; 587 } else { 588 msec = 1; 589 div = ecmd.base.speed / 1000; 590 } 591 } 592 593 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 594 595 if (div) 596 mbits /= div; 597 598 tmo = mbits * msec; 599 600 if (div) 601 return tmo+1; 602 return tmo; 603 } 604 605 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 606 union tpacket_req_u *req_u) 607 { 608 p1->feature_req_word = req_u->req3.tp_feature_req_word; 609 } 610 611 static void init_prb_bdqc(struct packet_sock *po, 612 struct packet_ring_buffer *rb, 613 struct pgv *pg_vec, 614 union tpacket_req_u *req_u) 615 { 616 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 617 struct tpacket_block_desc *pbd; 618 619 memset(p1, 0x0, sizeof(*p1)); 620 621 p1->knxt_seq_num = 1; 622 p1->pkbdq = pg_vec; 623 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 624 p1->pkblk_start = pg_vec[0].buffer; 625 p1->kblk_size = req_u->req3.tp_block_size; 626 p1->knum_blocks = req_u->req3.tp_block_nr; 627 p1->hdrlen = po->tp_hdrlen; 628 p1->version = po->tp_version; 629 p1->last_kactive_blk_num = 0; 630 po->stats.stats3.tp_freeze_q_cnt = 0; 631 if (req_u->req3.tp_retire_blk_tov) 632 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 633 else 634 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 635 req_u->req3.tp_block_size); 636 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 637 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 638 639 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 640 prb_init_ft_ops(p1, req_u); 641 prb_setup_retire_blk_timer(po); 642 prb_open_block(p1, pbd); 643 } 644 645 /* Do NOT update the last_blk_num first. 646 * Assumes sk_buff_head lock is held. 647 */ 648 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 649 { 650 mod_timer(&pkc->retire_blk_timer, 651 jiffies + pkc->tov_in_jiffies); 652 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 653 } 654 655 /* 656 * Timer logic: 657 * 1) We refresh the timer only when we open a block. 658 * By doing this we don't waste cycles refreshing the timer 659 * on packet-by-packet basis. 660 * 661 * With a 1MB block-size, on a 1Gbps line, it will take 662 * i) ~8 ms to fill a block + ii) memcpy etc. 663 * In this cut we are not accounting for the memcpy time. 664 * 665 * So, if the user sets the 'tmo' to 10ms then the timer 666 * will never fire while the block is still getting filled 667 * (which is what we want). However, the user could choose 668 * to close a block early and that's fine. 669 * 670 * But when the timer does fire, we check whether or not to refresh it. 671 * Since the tmo granularity is in msecs, it is not too expensive 672 * to refresh the timer, lets say every '8' msecs. 673 * Either the user can set the 'tmo' or we can derive it based on 674 * a) line-speed and b) block-size. 675 * prb_calc_retire_blk_tmo() calculates the tmo. 676 * 677 */ 678 static void prb_retire_rx_blk_timer_expired(unsigned long data) 679 { 680 struct packet_sock *po = (struct packet_sock *)data; 681 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 682 unsigned int frozen; 683 struct tpacket_block_desc *pbd; 684 685 spin_lock(&po->sk.sk_receive_queue.lock); 686 687 frozen = prb_queue_frozen(pkc); 688 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 689 690 if (unlikely(pkc->delete_blk_timer)) 691 goto out; 692 693 /* We only need to plug the race when the block is partially filled. 694 * tpacket_rcv: 695 * lock(); increment BLOCK_NUM_PKTS; unlock() 696 * copy_bits() is in progress ... 697 * timer fires on other cpu: 698 * we can't retire the current block because copy_bits 699 * is in progress. 700 * 701 */ 702 if (BLOCK_NUM_PKTS(pbd)) { 703 while (atomic_read(&pkc->blk_fill_in_prog)) { 704 /* Waiting for skb_copy_bits to finish... */ 705 cpu_relax(); 706 } 707 } 708 709 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 710 if (!frozen) { 711 if (!BLOCK_NUM_PKTS(pbd)) { 712 /* An empty block. Just refresh the timer. */ 713 goto refresh_timer; 714 } 715 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 716 if (!prb_dispatch_next_block(pkc, po)) 717 goto refresh_timer; 718 else 719 goto out; 720 } else { 721 /* Case 1. Queue was frozen because user-space was 722 * lagging behind. 723 */ 724 if (prb_curr_blk_in_use(pkc, pbd)) { 725 /* 726 * Ok, user-space is still behind. 727 * So just refresh the timer. 728 */ 729 goto refresh_timer; 730 } else { 731 /* Case 2. queue was frozen,user-space caught up, 732 * now the link went idle && the timer fired. 733 * We don't have a block to close.So we open this 734 * block and restart the timer. 735 * opening a block thaws the queue,restarts timer 736 * Thawing/timer-refresh is a side effect. 737 */ 738 prb_open_block(pkc, pbd); 739 goto out; 740 } 741 } 742 } 743 744 refresh_timer: 745 _prb_refresh_rx_retire_blk_timer(pkc); 746 747 out: 748 spin_unlock(&po->sk.sk_receive_queue.lock); 749 } 750 751 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 752 struct tpacket_block_desc *pbd1, __u32 status) 753 { 754 /* Flush everything minus the block header */ 755 756 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 757 u8 *start, *end; 758 759 start = (u8 *)pbd1; 760 761 /* Skip the block header(we know header WILL fit in 4K) */ 762 start += PAGE_SIZE; 763 764 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 765 for (; start < end; start += PAGE_SIZE) 766 flush_dcache_page(pgv_to_page(start)); 767 768 smp_wmb(); 769 #endif 770 771 /* Now update the block status. */ 772 773 BLOCK_STATUS(pbd1) = status; 774 775 /* Flush the block header */ 776 777 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 778 start = (u8 *)pbd1; 779 flush_dcache_page(pgv_to_page(start)); 780 781 smp_wmb(); 782 #endif 783 } 784 785 /* 786 * Side effect: 787 * 788 * 1) flush the block 789 * 2) Increment active_blk_num 790 * 791 * Note:We DONT refresh the timer on purpose. 792 * Because almost always the next block will be opened. 793 */ 794 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 795 struct tpacket_block_desc *pbd1, 796 struct packet_sock *po, unsigned int stat) 797 { 798 __u32 status = TP_STATUS_USER | stat; 799 800 struct tpacket3_hdr *last_pkt; 801 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 802 struct sock *sk = &po->sk; 803 804 if (po->stats.stats3.tp_drops) 805 status |= TP_STATUS_LOSING; 806 807 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 808 last_pkt->tp_next_offset = 0; 809 810 /* Get the ts of the last pkt */ 811 if (BLOCK_NUM_PKTS(pbd1)) { 812 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 813 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 814 } else { 815 /* Ok, we tmo'd - so get the current time. 816 * 817 * It shouldn't really happen as we don't close empty 818 * blocks. See prb_retire_rx_blk_timer_expired(). 819 */ 820 struct timespec ts; 821 getnstimeofday(&ts); 822 h1->ts_last_pkt.ts_sec = ts.tv_sec; 823 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 824 } 825 826 smp_wmb(); 827 828 /* Flush the block */ 829 prb_flush_block(pkc1, pbd1, status); 830 831 sk->sk_data_ready(sk); 832 833 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 834 } 835 836 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 837 { 838 pkc->reset_pending_on_curr_blk = 0; 839 } 840 841 /* 842 * Side effect of opening a block: 843 * 844 * 1) prb_queue is thawed. 845 * 2) retire_blk_timer is refreshed. 846 * 847 */ 848 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 849 struct tpacket_block_desc *pbd1) 850 { 851 struct timespec ts; 852 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 853 854 smp_rmb(); 855 856 /* We could have just memset this but we will lose the 857 * flexibility of making the priv area sticky 858 */ 859 860 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 861 BLOCK_NUM_PKTS(pbd1) = 0; 862 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 863 864 getnstimeofday(&ts); 865 866 h1->ts_first_pkt.ts_sec = ts.tv_sec; 867 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 868 869 pkc1->pkblk_start = (char *)pbd1; 870 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 871 872 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 873 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 874 875 pbd1->version = pkc1->version; 876 pkc1->prev = pkc1->nxt_offset; 877 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 878 879 prb_thaw_queue(pkc1); 880 _prb_refresh_rx_retire_blk_timer(pkc1); 881 882 smp_wmb(); 883 } 884 885 /* 886 * Queue freeze logic: 887 * 1) Assume tp_block_nr = 8 blocks. 888 * 2) At time 't0', user opens Rx ring. 889 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 890 * 4) user-space is either sleeping or processing block '0'. 891 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 892 * it will close block-7,loop around and try to fill block '0'. 893 * call-flow: 894 * __packet_lookup_frame_in_block 895 * prb_retire_current_block() 896 * prb_dispatch_next_block() 897 * |->(BLOCK_STATUS == USER) evaluates to true 898 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 899 * 6) Now there are two cases: 900 * 6.1) Link goes idle right after the queue is frozen. 901 * But remember, the last open_block() refreshed the timer. 902 * When this timer expires,it will refresh itself so that we can 903 * re-open block-0 in near future. 904 * 6.2) Link is busy and keeps on receiving packets. This is a simple 905 * case and __packet_lookup_frame_in_block will check if block-0 906 * is free and can now be re-used. 907 */ 908 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 909 struct packet_sock *po) 910 { 911 pkc->reset_pending_on_curr_blk = 1; 912 po->stats.stats3.tp_freeze_q_cnt++; 913 } 914 915 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 916 917 /* 918 * If the next block is free then we will dispatch it 919 * and return a good offset. 920 * Else, we will freeze the queue. 921 * So, caller must check the return value. 922 */ 923 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 924 struct packet_sock *po) 925 { 926 struct tpacket_block_desc *pbd; 927 928 smp_rmb(); 929 930 /* 1. Get current block num */ 931 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 932 933 /* 2. If this block is currently in_use then freeze the queue */ 934 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 935 prb_freeze_queue(pkc, po); 936 return NULL; 937 } 938 939 /* 940 * 3. 941 * open this block and return the offset where the first packet 942 * needs to get stored. 943 */ 944 prb_open_block(pkc, pbd); 945 return (void *)pkc->nxt_offset; 946 } 947 948 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 949 struct packet_sock *po, unsigned int status) 950 { 951 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 952 953 /* retire/close the current block */ 954 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 955 /* 956 * Plug the case where copy_bits() is in progress on 957 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 958 * have space to copy the pkt in the current block and 959 * called prb_retire_current_block() 960 * 961 * We don't need to worry about the TMO case because 962 * the timer-handler already handled this case. 963 */ 964 if (!(status & TP_STATUS_BLK_TMO)) { 965 while (atomic_read(&pkc->blk_fill_in_prog)) { 966 /* Waiting for skb_copy_bits to finish... */ 967 cpu_relax(); 968 } 969 } 970 prb_close_block(pkc, pbd, po, status); 971 return; 972 } 973 } 974 975 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc, 976 struct tpacket_block_desc *pbd) 977 { 978 return TP_STATUS_USER & BLOCK_STATUS(pbd); 979 } 980 981 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 982 { 983 return pkc->reset_pending_on_curr_blk; 984 } 985 986 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 987 { 988 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 989 atomic_dec(&pkc->blk_fill_in_prog); 990 } 991 992 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 993 struct tpacket3_hdr *ppd) 994 { 995 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 996 } 997 998 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 999 struct tpacket3_hdr *ppd) 1000 { 1001 ppd->hv1.tp_rxhash = 0; 1002 } 1003 1004 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 1005 struct tpacket3_hdr *ppd) 1006 { 1007 if (skb_vlan_tag_present(pkc->skb)) { 1008 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 1009 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 1010 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 1011 } else { 1012 ppd->hv1.tp_vlan_tci = 0; 1013 ppd->hv1.tp_vlan_tpid = 0; 1014 ppd->tp_status = TP_STATUS_AVAILABLE; 1015 } 1016 } 1017 1018 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 1019 struct tpacket3_hdr *ppd) 1020 { 1021 ppd->hv1.tp_padding = 0; 1022 prb_fill_vlan_info(pkc, ppd); 1023 1024 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 1025 prb_fill_rxhash(pkc, ppd); 1026 else 1027 prb_clear_rxhash(pkc, ppd); 1028 } 1029 1030 static void prb_fill_curr_block(char *curr, 1031 struct tpacket_kbdq_core *pkc, 1032 struct tpacket_block_desc *pbd, 1033 unsigned int len) 1034 { 1035 struct tpacket3_hdr *ppd; 1036 1037 ppd = (struct tpacket3_hdr *)curr; 1038 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 1039 pkc->prev = curr; 1040 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 1041 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1042 BLOCK_NUM_PKTS(pbd) += 1; 1043 atomic_inc(&pkc->blk_fill_in_prog); 1044 prb_run_all_ft_ops(pkc, ppd); 1045 } 1046 1047 /* Assumes caller has the sk->rx_queue.lock */ 1048 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1049 struct sk_buff *skb, 1050 int status, 1051 unsigned int len 1052 ) 1053 { 1054 struct tpacket_kbdq_core *pkc; 1055 struct tpacket_block_desc *pbd; 1056 char *curr, *end; 1057 1058 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1059 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1060 1061 /* Queue is frozen when user space is lagging behind */ 1062 if (prb_queue_frozen(pkc)) { 1063 /* 1064 * Check if that last block which caused the queue to freeze, 1065 * is still in_use by user-space. 1066 */ 1067 if (prb_curr_blk_in_use(pkc, pbd)) { 1068 /* Can't record this packet */ 1069 return NULL; 1070 } else { 1071 /* 1072 * Ok, the block was released by user-space. 1073 * Now let's open that block. 1074 * opening a block also thaws the queue. 1075 * Thawing is a side effect. 1076 */ 1077 prb_open_block(pkc, pbd); 1078 } 1079 } 1080 1081 smp_mb(); 1082 curr = pkc->nxt_offset; 1083 pkc->skb = skb; 1084 end = (char *)pbd + pkc->kblk_size; 1085 1086 /* first try the current block */ 1087 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1088 prb_fill_curr_block(curr, pkc, pbd, len); 1089 return (void *)curr; 1090 } 1091 1092 /* Ok, close the current block */ 1093 prb_retire_current_block(pkc, po, 0); 1094 1095 /* Now, try to dispatch the next block */ 1096 curr = (char *)prb_dispatch_next_block(pkc, po); 1097 if (curr) { 1098 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1099 prb_fill_curr_block(curr, pkc, pbd, len); 1100 return (void *)curr; 1101 } 1102 1103 /* 1104 * No free blocks are available.user_space hasn't caught up yet. 1105 * Queue was just frozen and now this packet will get dropped. 1106 */ 1107 return NULL; 1108 } 1109 1110 static void *packet_current_rx_frame(struct packet_sock *po, 1111 struct sk_buff *skb, 1112 int status, unsigned int len) 1113 { 1114 char *curr = NULL; 1115 switch (po->tp_version) { 1116 case TPACKET_V1: 1117 case TPACKET_V2: 1118 curr = packet_lookup_frame(po, &po->rx_ring, 1119 po->rx_ring.head, status); 1120 return curr; 1121 case TPACKET_V3: 1122 return __packet_lookup_frame_in_block(po, skb, status, len); 1123 default: 1124 WARN(1, "TPACKET version not supported\n"); 1125 BUG(); 1126 return NULL; 1127 } 1128 } 1129 1130 static void *prb_lookup_block(struct packet_sock *po, 1131 struct packet_ring_buffer *rb, 1132 unsigned int idx, 1133 int status) 1134 { 1135 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1136 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1137 1138 if (status != BLOCK_STATUS(pbd)) 1139 return NULL; 1140 return pbd; 1141 } 1142 1143 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1144 { 1145 unsigned int prev; 1146 if (rb->prb_bdqc.kactive_blk_num) 1147 prev = rb->prb_bdqc.kactive_blk_num-1; 1148 else 1149 prev = rb->prb_bdqc.knum_blocks-1; 1150 return prev; 1151 } 1152 1153 /* Assumes caller has held the rx_queue.lock */ 1154 static void *__prb_previous_block(struct packet_sock *po, 1155 struct packet_ring_buffer *rb, 1156 int status) 1157 { 1158 unsigned int previous = prb_previous_blk_num(rb); 1159 return prb_lookup_block(po, rb, previous, status); 1160 } 1161 1162 static void *packet_previous_rx_frame(struct packet_sock *po, 1163 struct packet_ring_buffer *rb, 1164 int status) 1165 { 1166 if (po->tp_version <= TPACKET_V2) 1167 return packet_previous_frame(po, rb, status); 1168 1169 return __prb_previous_block(po, rb, status); 1170 } 1171 1172 static void packet_increment_rx_head(struct packet_sock *po, 1173 struct packet_ring_buffer *rb) 1174 { 1175 switch (po->tp_version) { 1176 case TPACKET_V1: 1177 case TPACKET_V2: 1178 return packet_increment_head(rb); 1179 case TPACKET_V3: 1180 default: 1181 WARN(1, "TPACKET version not supported.\n"); 1182 BUG(); 1183 return; 1184 } 1185 } 1186 1187 static void *packet_previous_frame(struct packet_sock *po, 1188 struct packet_ring_buffer *rb, 1189 int status) 1190 { 1191 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1192 return packet_lookup_frame(po, rb, previous, status); 1193 } 1194 1195 static void packet_increment_head(struct packet_ring_buffer *buff) 1196 { 1197 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1198 } 1199 1200 static void packet_inc_pending(struct packet_ring_buffer *rb) 1201 { 1202 this_cpu_inc(*rb->pending_refcnt); 1203 } 1204 1205 static void packet_dec_pending(struct packet_ring_buffer *rb) 1206 { 1207 this_cpu_dec(*rb->pending_refcnt); 1208 } 1209 1210 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1211 { 1212 unsigned int refcnt = 0; 1213 int cpu; 1214 1215 /* We don't use pending refcount in rx_ring. */ 1216 if (rb->pending_refcnt == NULL) 1217 return 0; 1218 1219 for_each_possible_cpu(cpu) 1220 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1221 1222 return refcnt; 1223 } 1224 1225 static int packet_alloc_pending(struct packet_sock *po) 1226 { 1227 po->rx_ring.pending_refcnt = NULL; 1228 1229 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1230 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1231 return -ENOBUFS; 1232 1233 return 0; 1234 } 1235 1236 static void packet_free_pending(struct packet_sock *po) 1237 { 1238 free_percpu(po->tx_ring.pending_refcnt); 1239 } 1240 1241 #define ROOM_POW_OFF 2 1242 #define ROOM_NONE 0x0 1243 #define ROOM_LOW 0x1 1244 #define ROOM_NORMAL 0x2 1245 1246 static bool __tpacket_has_room(struct packet_sock *po, int pow_off) 1247 { 1248 int idx, len; 1249 1250 len = po->rx_ring.frame_max + 1; 1251 idx = po->rx_ring.head; 1252 if (pow_off) 1253 idx += len >> pow_off; 1254 if (idx >= len) 1255 idx -= len; 1256 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1257 } 1258 1259 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off) 1260 { 1261 int idx, len; 1262 1263 len = po->rx_ring.prb_bdqc.knum_blocks; 1264 idx = po->rx_ring.prb_bdqc.kactive_blk_num; 1265 if (pow_off) 1266 idx += len >> pow_off; 1267 if (idx >= len) 1268 idx -= len; 1269 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1270 } 1271 1272 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1273 { 1274 struct sock *sk = &po->sk; 1275 int ret = ROOM_NONE; 1276 1277 if (po->prot_hook.func != tpacket_rcv) { 1278 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1279 - (skb ? skb->truesize : 0); 1280 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF)) 1281 return ROOM_NORMAL; 1282 else if (avail > 0) 1283 return ROOM_LOW; 1284 else 1285 return ROOM_NONE; 1286 } 1287 1288 if (po->tp_version == TPACKET_V3) { 1289 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1290 ret = ROOM_NORMAL; 1291 else if (__tpacket_v3_has_room(po, 0)) 1292 ret = ROOM_LOW; 1293 } else { 1294 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1295 ret = ROOM_NORMAL; 1296 else if (__tpacket_has_room(po, 0)) 1297 ret = ROOM_LOW; 1298 } 1299 1300 return ret; 1301 } 1302 1303 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1304 { 1305 int ret; 1306 bool has_room; 1307 1308 spin_lock_bh(&po->sk.sk_receive_queue.lock); 1309 ret = __packet_rcv_has_room(po, skb); 1310 has_room = ret == ROOM_NORMAL; 1311 if (po->pressure == has_room) 1312 po->pressure = !has_room; 1313 spin_unlock_bh(&po->sk.sk_receive_queue.lock); 1314 1315 return ret; 1316 } 1317 1318 static void packet_sock_destruct(struct sock *sk) 1319 { 1320 skb_queue_purge(&sk->sk_error_queue); 1321 1322 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1323 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 1324 1325 if (!sock_flag(sk, SOCK_DEAD)) { 1326 pr_err("Attempt to release alive packet socket: %p\n", sk); 1327 return; 1328 } 1329 1330 sk_refcnt_debug_dec(sk); 1331 } 1332 1333 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1334 { 1335 u32 rxhash; 1336 int i, count = 0; 1337 1338 rxhash = skb_get_hash(skb); 1339 for (i = 0; i < ROLLOVER_HLEN; i++) 1340 if (po->rollover->history[i] == rxhash) 1341 count++; 1342 1343 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash; 1344 return count > (ROLLOVER_HLEN >> 1); 1345 } 1346 1347 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1348 struct sk_buff *skb, 1349 unsigned int num) 1350 { 1351 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1352 } 1353 1354 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1355 struct sk_buff *skb, 1356 unsigned int num) 1357 { 1358 unsigned int val = atomic_inc_return(&f->rr_cur); 1359 1360 return val % num; 1361 } 1362 1363 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1364 struct sk_buff *skb, 1365 unsigned int num) 1366 { 1367 return smp_processor_id() % num; 1368 } 1369 1370 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1371 struct sk_buff *skb, 1372 unsigned int num) 1373 { 1374 return prandom_u32_max(num); 1375 } 1376 1377 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1378 struct sk_buff *skb, 1379 unsigned int idx, bool try_self, 1380 unsigned int num) 1381 { 1382 struct packet_sock *po, *po_next, *po_skip = NULL; 1383 unsigned int i, j, room = ROOM_NONE; 1384 1385 po = pkt_sk(f->arr[idx]); 1386 1387 if (try_self) { 1388 room = packet_rcv_has_room(po, skb); 1389 if (room == ROOM_NORMAL || 1390 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1391 return idx; 1392 po_skip = po; 1393 } 1394 1395 i = j = min_t(int, po->rollover->sock, num - 1); 1396 do { 1397 po_next = pkt_sk(f->arr[i]); 1398 if (po_next != po_skip && !po_next->pressure && 1399 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1400 if (i != j) 1401 po->rollover->sock = i; 1402 atomic_long_inc(&po->rollover->num); 1403 if (room == ROOM_LOW) 1404 atomic_long_inc(&po->rollover->num_huge); 1405 return i; 1406 } 1407 1408 if (++i == num) 1409 i = 0; 1410 } while (i != j); 1411 1412 atomic_long_inc(&po->rollover->num_failed); 1413 return idx; 1414 } 1415 1416 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1417 struct sk_buff *skb, 1418 unsigned int num) 1419 { 1420 return skb_get_queue_mapping(skb) % num; 1421 } 1422 1423 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1424 struct sk_buff *skb, 1425 unsigned int num) 1426 { 1427 struct bpf_prog *prog; 1428 unsigned int ret = 0; 1429 1430 rcu_read_lock(); 1431 prog = rcu_dereference(f->bpf_prog); 1432 if (prog) 1433 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1434 rcu_read_unlock(); 1435 1436 return ret; 1437 } 1438 1439 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1440 { 1441 return f->flags & (flag >> 8); 1442 } 1443 1444 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1445 struct packet_type *pt, struct net_device *orig_dev) 1446 { 1447 struct packet_fanout *f = pt->af_packet_priv; 1448 unsigned int num = READ_ONCE(f->num_members); 1449 struct net *net = read_pnet(&f->net); 1450 struct packet_sock *po; 1451 unsigned int idx; 1452 1453 if (!net_eq(dev_net(dev), net) || !num) { 1454 kfree_skb(skb); 1455 return 0; 1456 } 1457 1458 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1459 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1460 if (!skb) 1461 return 0; 1462 } 1463 switch (f->type) { 1464 case PACKET_FANOUT_HASH: 1465 default: 1466 idx = fanout_demux_hash(f, skb, num); 1467 break; 1468 case PACKET_FANOUT_LB: 1469 idx = fanout_demux_lb(f, skb, num); 1470 break; 1471 case PACKET_FANOUT_CPU: 1472 idx = fanout_demux_cpu(f, skb, num); 1473 break; 1474 case PACKET_FANOUT_RND: 1475 idx = fanout_demux_rnd(f, skb, num); 1476 break; 1477 case PACKET_FANOUT_QM: 1478 idx = fanout_demux_qm(f, skb, num); 1479 break; 1480 case PACKET_FANOUT_ROLLOVER: 1481 idx = fanout_demux_rollover(f, skb, 0, false, num); 1482 break; 1483 case PACKET_FANOUT_CBPF: 1484 case PACKET_FANOUT_EBPF: 1485 idx = fanout_demux_bpf(f, skb, num); 1486 break; 1487 } 1488 1489 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1490 idx = fanout_demux_rollover(f, skb, idx, true, num); 1491 1492 po = pkt_sk(f->arr[idx]); 1493 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1494 } 1495 1496 DEFINE_MUTEX(fanout_mutex); 1497 EXPORT_SYMBOL_GPL(fanout_mutex); 1498 static LIST_HEAD(fanout_list); 1499 1500 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1501 { 1502 struct packet_fanout *f = po->fanout; 1503 1504 spin_lock(&f->lock); 1505 f->arr[f->num_members] = sk; 1506 smp_wmb(); 1507 f->num_members++; 1508 spin_unlock(&f->lock); 1509 } 1510 1511 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1512 { 1513 struct packet_fanout *f = po->fanout; 1514 int i; 1515 1516 spin_lock(&f->lock); 1517 for (i = 0; i < f->num_members; i++) { 1518 if (f->arr[i] == sk) 1519 break; 1520 } 1521 BUG_ON(i >= f->num_members); 1522 f->arr[i] = f->arr[f->num_members - 1]; 1523 f->num_members--; 1524 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 int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1629 { 1630 struct packet_sock *po = pkt_sk(sk); 1631 struct packet_fanout *f, *match; 1632 u8 type = type_flags & 0xff; 1633 u8 flags = type_flags >> 8; 1634 int err; 1635 1636 switch (type) { 1637 case PACKET_FANOUT_ROLLOVER: 1638 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1639 return -EINVAL; 1640 case PACKET_FANOUT_HASH: 1641 case PACKET_FANOUT_LB: 1642 case PACKET_FANOUT_CPU: 1643 case PACKET_FANOUT_RND: 1644 case PACKET_FANOUT_QM: 1645 case PACKET_FANOUT_CBPF: 1646 case PACKET_FANOUT_EBPF: 1647 break; 1648 default: 1649 return -EINVAL; 1650 } 1651 1652 if (!po->running) 1653 return -EINVAL; 1654 1655 if (po->fanout) 1656 return -EALREADY; 1657 1658 if (type == PACKET_FANOUT_ROLLOVER || 1659 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1660 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL); 1661 if (!po->rollover) 1662 return -ENOMEM; 1663 atomic_long_set(&po->rollover->num, 0); 1664 atomic_long_set(&po->rollover->num_huge, 0); 1665 atomic_long_set(&po->rollover->num_failed, 0); 1666 } 1667 1668 mutex_lock(&fanout_mutex); 1669 match = NULL; 1670 list_for_each_entry(f, &fanout_list, list) { 1671 if (f->id == id && 1672 read_pnet(&f->net) == sock_net(sk)) { 1673 match = f; 1674 break; 1675 } 1676 } 1677 err = -EINVAL; 1678 if (match && match->flags != flags) 1679 goto out; 1680 if (!match) { 1681 err = -ENOMEM; 1682 match = kzalloc(sizeof(*match), GFP_KERNEL); 1683 if (!match) 1684 goto out; 1685 write_pnet(&match->net, sock_net(sk)); 1686 match->id = id; 1687 match->type = type; 1688 match->flags = flags; 1689 INIT_LIST_HEAD(&match->list); 1690 spin_lock_init(&match->lock); 1691 atomic_set(&match->sk_ref, 0); 1692 fanout_init_data(match); 1693 match->prot_hook.type = po->prot_hook.type; 1694 match->prot_hook.dev = po->prot_hook.dev; 1695 match->prot_hook.func = packet_rcv_fanout; 1696 match->prot_hook.af_packet_priv = match; 1697 match->prot_hook.id_match = match_fanout_group; 1698 dev_add_pack(&match->prot_hook); 1699 list_add(&match->list, &fanout_list); 1700 } 1701 err = -EINVAL; 1702 if (match->type == type && 1703 match->prot_hook.type == po->prot_hook.type && 1704 match->prot_hook.dev == po->prot_hook.dev) { 1705 err = -ENOSPC; 1706 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1707 __dev_remove_pack(&po->prot_hook); 1708 po->fanout = match; 1709 atomic_inc(&match->sk_ref); 1710 __fanout_link(sk, po); 1711 err = 0; 1712 } 1713 } 1714 out: 1715 mutex_unlock(&fanout_mutex); 1716 if (err) { 1717 kfree(po->rollover); 1718 po->rollover = NULL; 1719 } 1720 return err; 1721 } 1722 1723 static void fanout_release(struct sock *sk) 1724 { 1725 struct packet_sock *po = pkt_sk(sk); 1726 struct packet_fanout *f; 1727 1728 f = po->fanout; 1729 if (!f) 1730 return; 1731 1732 mutex_lock(&fanout_mutex); 1733 po->fanout = NULL; 1734 1735 if (atomic_dec_and_test(&f->sk_ref)) { 1736 list_del(&f->list); 1737 dev_remove_pack(&f->prot_hook); 1738 fanout_release_data(f); 1739 kfree(f); 1740 } 1741 mutex_unlock(&fanout_mutex); 1742 1743 if (po->rollover) 1744 kfree_rcu(po->rollover, rcu); 1745 } 1746 1747 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1748 struct sk_buff *skb) 1749 { 1750 /* Earlier code assumed this would be a VLAN pkt, double-check 1751 * this now that we have the actual packet in hand. We can only 1752 * do this check on Ethernet devices. 1753 */ 1754 if (unlikely(dev->type != ARPHRD_ETHER)) 1755 return false; 1756 1757 skb_reset_mac_header(skb); 1758 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1759 } 1760 1761 static const struct proto_ops packet_ops; 1762 1763 static const struct proto_ops packet_ops_spkt; 1764 1765 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1766 struct packet_type *pt, struct net_device *orig_dev) 1767 { 1768 struct sock *sk; 1769 struct sockaddr_pkt *spkt; 1770 1771 /* 1772 * When we registered the protocol we saved the socket in the data 1773 * field for just this event. 1774 */ 1775 1776 sk = pt->af_packet_priv; 1777 1778 /* 1779 * Yank back the headers [hope the device set this 1780 * right or kerboom...] 1781 * 1782 * Incoming packets have ll header pulled, 1783 * push it back. 1784 * 1785 * For outgoing ones skb->data == skb_mac_header(skb) 1786 * so that this procedure is noop. 1787 */ 1788 1789 if (skb->pkt_type == PACKET_LOOPBACK) 1790 goto out; 1791 1792 if (!net_eq(dev_net(dev), sock_net(sk))) 1793 goto out; 1794 1795 skb = skb_share_check(skb, GFP_ATOMIC); 1796 if (skb == NULL) 1797 goto oom; 1798 1799 /* drop any routing info */ 1800 skb_dst_drop(skb); 1801 1802 /* drop conntrack reference */ 1803 nf_reset(skb); 1804 1805 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1806 1807 skb_push(skb, skb->data - skb_mac_header(skb)); 1808 1809 /* 1810 * The SOCK_PACKET socket receives _all_ frames. 1811 */ 1812 1813 spkt->spkt_family = dev->type; 1814 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1815 spkt->spkt_protocol = skb->protocol; 1816 1817 /* 1818 * Charge the memory to the socket. This is done specifically 1819 * to prevent sockets using all the memory up. 1820 */ 1821 1822 if (sock_queue_rcv_skb(sk, skb) == 0) 1823 return 0; 1824 1825 out: 1826 kfree_skb(skb); 1827 oom: 1828 return 0; 1829 } 1830 1831 1832 /* 1833 * Output a raw packet to a device layer. This bypasses all the other 1834 * protocol layers and you must therefore supply it with a complete frame 1835 */ 1836 1837 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1838 size_t len) 1839 { 1840 struct sock *sk = sock->sk; 1841 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1842 struct sk_buff *skb = NULL; 1843 struct net_device *dev; 1844 struct sockcm_cookie sockc; 1845 __be16 proto = 0; 1846 int err; 1847 int extra_len = 0; 1848 1849 /* 1850 * Get and verify the address. 1851 */ 1852 1853 if (saddr) { 1854 if (msg->msg_namelen < sizeof(struct sockaddr)) 1855 return -EINVAL; 1856 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1857 proto = saddr->spkt_protocol; 1858 } else 1859 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1860 1861 /* 1862 * Find the device first to size check it 1863 */ 1864 1865 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1866 retry: 1867 rcu_read_lock(); 1868 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1869 err = -ENODEV; 1870 if (dev == NULL) 1871 goto out_unlock; 1872 1873 err = -ENETDOWN; 1874 if (!(dev->flags & IFF_UP)) 1875 goto out_unlock; 1876 1877 /* 1878 * You may not queue a frame bigger than the mtu. This is the lowest level 1879 * raw protocol and you must do your own fragmentation at this level. 1880 */ 1881 1882 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1883 if (!netif_supports_nofcs(dev)) { 1884 err = -EPROTONOSUPPORT; 1885 goto out_unlock; 1886 } 1887 extra_len = 4; /* We're doing our own CRC */ 1888 } 1889 1890 err = -EMSGSIZE; 1891 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1892 goto out_unlock; 1893 1894 if (!skb) { 1895 size_t reserved = LL_RESERVED_SPACE(dev); 1896 int tlen = dev->needed_tailroom; 1897 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1898 1899 rcu_read_unlock(); 1900 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1901 if (skb == NULL) 1902 return -ENOBUFS; 1903 /* FIXME: Save some space for broken drivers that write a hard 1904 * header at transmission time by themselves. PPP is the notable 1905 * one here. This should really be fixed at the driver level. 1906 */ 1907 skb_reserve(skb, reserved); 1908 skb_reset_network_header(skb); 1909 1910 /* Try to align data part correctly */ 1911 if (hhlen) { 1912 skb->data -= hhlen; 1913 skb->tail -= hhlen; 1914 if (len < hhlen) 1915 skb_reset_network_header(skb); 1916 } 1917 err = memcpy_from_msg(skb_put(skb, len), msg, len); 1918 if (err) 1919 goto out_free; 1920 goto retry; 1921 } 1922 1923 if (!dev_validate_header(dev, skb->data, len)) { 1924 err = -EINVAL; 1925 goto out_unlock; 1926 } 1927 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 1928 !packet_extra_vlan_len_allowed(dev, skb)) { 1929 err = -EMSGSIZE; 1930 goto out_unlock; 1931 } 1932 1933 sockc.tsflags = sk->sk_tsflags; 1934 if (msg->msg_controllen) { 1935 err = sock_cmsg_send(sk, msg, &sockc); 1936 if (unlikely(err)) 1937 goto out_unlock; 1938 } 1939 1940 skb->protocol = proto; 1941 skb->dev = dev; 1942 skb->priority = sk->sk_priority; 1943 skb->mark = sk->sk_mark; 1944 1945 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags); 1946 1947 if (unlikely(extra_len == 4)) 1948 skb->no_fcs = 1; 1949 1950 skb_probe_transport_header(skb, 0); 1951 1952 dev_queue_xmit(skb); 1953 rcu_read_unlock(); 1954 return len; 1955 1956 out_unlock: 1957 rcu_read_unlock(); 1958 out_free: 1959 kfree_skb(skb); 1960 return err; 1961 } 1962 1963 static unsigned int run_filter(struct sk_buff *skb, 1964 const struct sock *sk, 1965 unsigned int res) 1966 { 1967 struct sk_filter *filter; 1968 1969 rcu_read_lock(); 1970 filter = rcu_dereference(sk->sk_filter); 1971 if (filter != NULL) 1972 res = bpf_prog_run_clear_cb(filter->prog, skb); 1973 rcu_read_unlock(); 1974 1975 return res; 1976 } 1977 1978 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 1979 size_t *len) 1980 { 1981 struct virtio_net_hdr vnet_hdr; 1982 1983 if (*len < sizeof(vnet_hdr)) 1984 return -EINVAL; 1985 *len -= sizeof(vnet_hdr); 1986 1987 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le())) 1988 return -EINVAL; 1989 1990 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 1991 } 1992 1993 /* 1994 * This function makes lazy skb cloning in hope that most of packets 1995 * are discarded by BPF. 1996 * 1997 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 1998 * and skb->cb are mangled. It works because (and until) packets 1999 * falling here are owned by current CPU. Output packets are cloned 2000 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2001 * sequencially, so that if we return skb to original state on exit, 2002 * we will not harm anyone. 2003 */ 2004 2005 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2006 struct packet_type *pt, struct net_device *orig_dev) 2007 { 2008 struct sock *sk; 2009 struct sockaddr_ll *sll; 2010 struct packet_sock *po; 2011 u8 *skb_head = skb->data; 2012 int skb_len = skb->len; 2013 unsigned int snaplen, res; 2014 bool is_drop_n_account = false; 2015 2016 if (skb->pkt_type == PACKET_LOOPBACK) 2017 goto drop; 2018 2019 sk = pt->af_packet_priv; 2020 po = pkt_sk(sk); 2021 2022 if (!net_eq(dev_net(dev), sock_net(sk))) 2023 goto drop; 2024 2025 skb->dev = dev; 2026 2027 if (dev->header_ops) { 2028 /* The device has an explicit notion of ll header, 2029 * exported to higher levels. 2030 * 2031 * Otherwise, the device hides details of its frame 2032 * structure, so that corresponding packet head is 2033 * never delivered to user. 2034 */ 2035 if (sk->sk_type != SOCK_DGRAM) 2036 skb_push(skb, skb->data - skb_mac_header(skb)); 2037 else if (skb->pkt_type == PACKET_OUTGOING) { 2038 /* Special case: outgoing packets have ll header at head */ 2039 skb_pull(skb, skb_network_offset(skb)); 2040 } 2041 } 2042 2043 snaplen = skb->len; 2044 2045 res = run_filter(skb, sk, snaplen); 2046 if (!res) 2047 goto drop_n_restore; 2048 if (snaplen > res) 2049 snaplen = res; 2050 2051 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2052 goto drop_n_acct; 2053 2054 if (skb_shared(skb)) { 2055 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2056 if (nskb == NULL) 2057 goto drop_n_acct; 2058 2059 if (skb_head != skb->data) { 2060 skb->data = skb_head; 2061 skb->len = skb_len; 2062 } 2063 consume_skb(skb); 2064 skb = nskb; 2065 } 2066 2067 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2068 2069 sll = &PACKET_SKB_CB(skb)->sa.ll; 2070 sll->sll_hatype = dev->type; 2071 sll->sll_pkttype = skb->pkt_type; 2072 if (unlikely(po->origdev)) 2073 sll->sll_ifindex = orig_dev->ifindex; 2074 else 2075 sll->sll_ifindex = dev->ifindex; 2076 2077 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2078 2079 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2080 * Use their space for storing the original skb length. 2081 */ 2082 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2083 2084 if (pskb_trim(skb, snaplen)) 2085 goto drop_n_acct; 2086 2087 skb_set_owner_r(skb, sk); 2088 skb->dev = NULL; 2089 skb_dst_drop(skb); 2090 2091 /* drop conntrack reference */ 2092 nf_reset(skb); 2093 2094 spin_lock(&sk->sk_receive_queue.lock); 2095 po->stats.stats1.tp_packets++; 2096 sock_skb_set_dropcount(sk, skb); 2097 __skb_queue_tail(&sk->sk_receive_queue, skb); 2098 spin_unlock(&sk->sk_receive_queue.lock); 2099 sk->sk_data_ready(sk); 2100 return 0; 2101 2102 drop_n_acct: 2103 is_drop_n_account = true; 2104 spin_lock(&sk->sk_receive_queue.lock); 2105 po->stats.stats1.tp_drops++; 2106 atomic_inc(&sk->sk_drops); 2107 spin_unlock(&sk->sk_receive_queue.lock); 2108 2109 drop_n_restore: 2110 if (skb_head != skb->data && skb_shared(skb)) { 2111 skb->data = skb_head; 2112 skb->len = skb_len; 2113 } 2114 drop: 2115 if (!is_drop_n_account) 2116 consume_skb(skb); 2117 else 2118 kfree_skb(skb); 2119 return 0; 2120 } 2121 2122 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2123 struct packet_type *pt, struct net_device *orig_dev) 2124 { 2125 struct sock *sk; 2126 struct packet_sock *po; 2127 struct sockaddr_ll *sll; 2128 union tpacket_uhdr h; 2129 u8 *skb_head = skb->data; 2130 int skb_len = skb->len; 2131 unsigned int snaplen, res; 2132 unsigned long status = TP_STATUS_USER; 2133 unsigned short macoff, netoff, hdrlen; 2134 struct sk_buff *copy_skb = NULL; 2135 struct timespec ts; 2136 __u32 ts_status; 2137 bool is_drop_n_account = false; 2138 2139 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2140 * We may add members to them until current aligned size without forcing 2141 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2142 */ 2143 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2144 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2145 2146 if (skb->pkt_type == PACKET_LOOPBACK) 2147 goto drop; 2148 2149 sk = pt->af_packet_priv; 2150 po = pkt_sk(sk); 2151 2152 if (!net_eq(dev_net(dev), sock_net(sk))) 2153 goto drop; 2154 2155 if (dev->header_ops) { 2156 if (sk->sk_type != SOCK_DGRAM) 2157 skb_push(skb, skb->data - skb_mac_header(skb)); 2158 else if (skb->pkt_type == PACKET_OUTGOING) { 2159 /* Special case: outgoing packets have ll header at head */ 2160 skb_pull(skb, skb_network_offset(skb)); 2161 } 2162 } 2163 2164 snaplen = skb->len; 2165 2166 res = run_filter(skb, sk, snaplen); 2167 if (!res) 2168 goto drop_n_restore; 2169 2170 if (skb->ip_summed == CHECKSUM_PARTIAL) 2171 status |= TP_STATUS_CSUMNOTREADY; 2172 else if (skb->pkt_type != PACKET_OUTGOING && 2173 (skb->ip_summed == CHECKSUM_COMPLETE || 2174 skb_csum_unnecessary(skb))) 2175 status |= TP_STATUS_CSUM_VALID; 2176 2177 if (snaplen > res) 2178 snaplen = res; 2179 2180 if (sk->sk_type == SOCK_DGRAM) { 2181 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2182 po->tp_reserve; 2183 } else { 2184 unsigned int maclen = skb_network_offset(skb); 2185 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2186 (maclen < 16 ? 16 : maclen)) + 2187 po->tp_reserve; 2188 if (po->has_vnet_hdr) 2189 netoff += sizeof(struct virtio_net_hdr); 2190 macoff = netoff - maclen; 2191 } 2192 if (po->tp_version <= TPACKET_V2) { 2193 if (macoff + snaplen > po->rx_ring.frame_size) { 2194 if (po->copy_thresh && 2195 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2196 if (skb_shared(skb)) { 2197 copy_skb = skb_clone(skb, GFP_ATOMIC); 2198 } else { 2199 copy_skb = skb_get(skb); 2200 skb_head = skb->data; 2201 } 2202 if (copy_skb) 2203 skb_set_owner_r(copy_skb, sk); 2204 } 2205 snaplen = po->rx_ring.frame_size - macoff; 2206 if ((int)snaplen < 0) 2207 snaplen = 0; 2208 } 2209 } else if (unlikely(macoff + snaplen > 2210 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2211 u32 nval; 2212 2213 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2214 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2215 snaplen, nval, macoff); 2216 snaplen = nval; 2217 if (unlikely((int)snaplen < 0)) { 2218 snaplen = 0; 2219 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2220 } 2221 } 2222 spin_lock(&sk->sk_receive_queue.lock); 2223 h.raw = packet_current_rx_frame(po, skb, 2224 TP_STATUS_KERNEL, (macoff+snaplen)); 2225 if (!h.raw) 2226 goto drop_n_account; 2227 if (po->tp_version <= TPACKET_V2) { 2228 packet_increment_rx_head(po, &po->rx_ring); 2229 /* 2230 * LOSING will be reported till you read the stats, 2231 * because it's COR - Clear On Read. 2232 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2233 * at packet level. 2234 */ 2235 if (po->stats.stats1.tp_drops) 2236 status |= TP_STATUS_LOSING; 2237 } 2238 po->stats.stats1.tp_packets++; 2239 if (copy_skb) { 2240 status |= TP_STATUS_COPY; 2241 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2242 } 2243 spin_unlock(&sk->sk_receive_queue.lock); 2244 2245 if (po->has_vnet_hdr) { 2246 if (virtio_net_hdr_from_skb(skb, h.raw + macoff - 2247 sizeof(struct virtio_net_hdr), 2248 vio_le())) { 2249 spin_lock(&sk->sk_receive_queue.lock); 2250 goto drop_n_account; 2251 } 2252 } 2253 2254 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2255 2256 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 2257 getnstimeofday(&ts); 2258 2259 status |= ts_status; 2260 2261 switch (po->tp_version) { 2262 case TPACKET_V1: 2263 h.h1->tp_len = skb->len; 2264 h.h1->tp_snaplen = snaplen; 2265 h.h1->tp_mac = macoff; 2266 h.h1->tp_net = netoff; 2267 h.h1->tp_sec = ts.tv_sec; 2268 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2269 hdrlen = sizeof(*h.h1); 2270 break; 2271 case TPACKET_V2: 2272 h.h2->tp_len = skb->len; 2273 h.h2->tp_snaplen = snaplen; 2274 h.h2->tp_mac = macoff; 2275 h.h2->tp_net = netoff; 2276 h.h2->tp_sec = ts.tv_sec; 2277 h.h2->tp_nsec = ts.tv_nsec; 2278 if (skb_vlan_tag_present(skb)) { 2279 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2280 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2281 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2282 } else { 2283 h.h2->tp_vlan_tci = 0; 2284 h.h2->tp_vlan_tpid = 0; 2285 } 2286 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2287 hdrlen = sizeof(*h.h2); 2288 break; 2289 case TPACKET_V3: 2290 /* tp_nxt_offset,vlan are already populated above. 2291 * So DONT clear those fields here 2292 */ 2293 h.h3->tp_status |= status; 2294 h.h3->tp_len = skb->len; 2295 h.h3->tp_snaplen = snaplen; 2296 h.h3->tp_mac = macoff; 2297 h.h3->tp_net = netoff; 2298 h.h3->tp_sec = ts.tv_sec; 2299 h.h3->tp_nsec = ts.tv_nsec; 2300 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2301 hdrlen = sizeof(*h.h3); 2302 break; 2303 default: 2304 BUG(); 2305 } 2306 2307 sll = h.raw + TPACKET_ALIGN(hdrlen); 2308 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2309 sll->sll_family = AF_PACKET; 2310 sll->sll_hatype = dev->type; 2311 sll->sll_protocol = skb->protocol; 2312 sll->sll_pkttype = skb->pkt_type; 2313 if (unlikely(po->origdev)) 2314 sll->sll_ifindex = orig_dev->ifindex; 2315 else 2316 sll->sll_ifindex = dev->ifindex; 2317 2318 smp_mb(); 2319 2320 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2321 if (po->tp_version <= TPACKET_V2) { 2322 u8 *start, *end; 2323 2324 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2325 macoff + snaplen); 2326 2327 for (start = h.raw; start < end; start += PAGE_SIZE) 2328 flush_dcache_page(pgv_to_page(start)); 2329 } 2330 smp_wmb(); 2331 #endif 2332 2333 if (po->tp_version <= TPACKET_V2) { 2334 __packet_set_status(po, h.raw, status); 2335 sk->sk_data_ready(sk); 2336 } else { 2337 prb_clear_blk_fill_status(&po->rx_ring); 2338 } 2339 2340 drop_n_restore: 2341 if (skb_head != skb->data && skb_shared(skb)) { 2342 skb->data = skb_head; 2343 skb->len = skb_len; 2344 } 2345 drop: 2346 if (!is_drop_n_account) 2347 consume_skb(skb); 2348 else 2349 kfree_skb(skb); 2350 return 0; 2351 2352 drop_n_account: 2353 is_drop_n_account = true; 2354 po->stats.stats1.tp_drops++; 2355 spin_unlock(&sk->sk_receive_queue.lock); 2356 2357 sk->sk_data_ready(sk); 2358 kfree_skb(copy_skb); 2359 goto drop_n_restore; 2360 } 2361 2362 static void tpacket_destruct_skb(struct sk_buff *skb) 2363 { 2364 struct packet_sock *po = pkt_sk(skb->sk); 2365 2366 if (likely(po->tx_ring.pg_vec)) { 2367 void *ph; 2368 __u32 ts; 2369 2370 ph = skb_shinfo(skb)->destructor_arg; 2371 packet_dec_pending(&po->tx_ring); 2372 2373 ts = __packet_set_timestamp(po, ph, skb); 2374 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2375 } 2376 2377 sock_wfree(skb); 2378 } 2379 2380 static void tpacket_set_protocol(const struct net_device *dev, 2381 struct sk_buff *skb) 2382 { 2383 if (dev->type == ARPHRD_ETHER) { 2384 skb_reset_mac_header(skb); 2385 skb->protocol = eth_hdr(skb)->h_proto; 2386 } 2387 } 2388 2389 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2390 { 2391 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2392 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2393 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2394 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2395 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2396 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2397 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2398 2399 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2400 return -EINVAL; 2401 2402 return 0; 2403 } 2404 2405 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2406 struct virtio_net_hdr *vnet_hdr) 2407 { 2408 if (*len < sizeof(*vnet_hdr)) 2409 return -EINVAL; 2410 *len -= sizeof(*vnet_hdr); 2411 2412 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2413 return -EFAULT; 2414 2415 return __packet_snd_vnet_parse(vnet_hdr, *len); 2416 } 2417 2418 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2419 void *frame, struct net_device *dev, void *data, int tp_len, 2420 __be16 proto, unsigned char *addr, int hlen, int copylen, 2421 const struct sockcm_cookie *sockc) 2422 { 2423 union tpacket_uhdr ph; 2424 int to_write, offset, len, nr_frags, len_max; 2425 struct socket *sock = po->sk.sk_socket; 2426 struct page *page; 2427 int err; 2428 2429 ph.raw = frame; 2430 2431 skb->protocol = proto; 2432 skb->dev = dev; 2433 skb->priority = po->sk.sk_priority; 2434 skb->mark = po->sk.sk_mark; 2435 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags); 2436 skb_shinfo(skb)->destructor_arg = ph.raw; 2437 2438 skb_reserve(skb, hlen); 2439 skb_reset_network_header(skb); 2440 2441 to_write = tp_len; 2442 2443 if (sock->type == SOCK_DGRAM) { 2444 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2445 NULL, tp_len); 2446 if (unlikely(err < 0)) 2447 return -EINVAL; 2448 } else if (copylen) { 2449 int hdrlen = min_t(int, copylen, tp_len); 2450 2451 skb_push(skb, dev->hard_header_len); 2452 skb_put(skb, copylen - dev->hard_header_len); 2453 err = skb_store_bits(skb, 0, data, hdrlen); 2454 if (unlikely(err)) 2455 return err; 2456 if (!dev_validate_header(dev, skb->data, hdrlen)) 2457 return -EINVAL; 2458 if (!skb->protocol) 2459 tpacket_set_protocol(dev, skb); 2460 2461 data += hdrlen; 2462 to_write -= hdrlen; 2463 } 2464 2465 offset = offset_in_page(data); 2466 len_max = PAGE_SIZE - offset; 2467 len = ((to_write > len_max) ? len_max : to_write); 2468 2469 skb->data_len = to_write; 2470 skb->len += to_write; 2471 skb->truesize += to_write; 2472 atomic_add(to_write, &po->sk.sk_wmem_alloc); 2473 2474 while (likely(to_write)) { 2475 nr_frags = skb_shinfo(skb)->nr_frags; 2476 2477 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2478 pr_err("Packet exceed the number of skb frags(%lu)\n", 2479 MAX_SKB_FRAGS); 2480 return -EFAULT; 2481 } 2482 2483 page = pgv_to_page(data); 2484 data += len; 2485 flush_dcache_page(page); 2486 get_page(page); 2487 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2488 to_write -= len; 2489 offset = 0; 2490 len_max = PAGE_SIZE; 2491 len = ((to_write > len_max) ? len_max : to_write); 2492 } 2493 2494 skb_probe_transport_header(skb, 0); 2495 2496 return tp_len; 2497 } 2498 2499 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2500 int size_max, void **data) 2501 { 2502 union tpacket_uhdr ph; 2503 int tp_len, off; 2504 2505 ph.raw = frame; 2506 2507 switch (po->tp_version) { 2508 case TPACKET_V3: 2509 if (ph.h3->tp_next_offset != 0) { 2510 pr_warn_once("variable sized slot not supported"); 2511 return -EINVAL; 2512 } 2513 tp_len = ph.h3->tp_len; 2514 break; 2515 case TPACKET_V2: 2516 tp_len = ph.h2->tp_len; 2517 break; 2518 default: 2519 tp_len = ph.h1->tp_len; 2520 break; 2521 } 2522 if (unlikely(tp_len > size_max)) { 2523 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2524 return -EMSGSIZE; 2525 } 2526 2527 if (unlikely(po->tp_tx_has_off)) { 2528 int off_min, off_max; 2529 2530 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2531 off_max = po->tx_ring.frame_size - tp_len; 2532 if (po->sk.sk_type == SOCK_DGRAM) { 2533 switch (po->tp_version) { 2534 case TPACKET_V3: 2535 off = ph.h3->tp_net; 2536 break; 2537 case TPACKET_V2: 2538 off = ph.h2->tp_net; 2539 break; 2540 default: 2541 off = ph.h1->tp_net; 2542 break; 2543 } 2544 } else { 2545 switch (po->tp_version) { 2546 case TPACKET_V3: 2547 off = ph.h3->tp_mac; 2548 break; 2549 case TPACKET_V2: 2550 off = ph.h2->tp_mac; 2551 break; 2552 default: 2553 off = ph.h1->tp_mac; 2554 break; 2555 } 2556 } 2557 if (unlikely((off < off_min) || (off_max < off))) 2558 return -EINVAL; 2559 } else { 2560 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2561 } 2562 2563 *data = frame + off; 2564 return tp_len; 2565 } 2566 2567 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2568 { 2569 struct sk_buff *skb; 2570 struct net_device *dev; 2571 struct virtio_net_hdr *vnet_hdr = NULL; 2572 struct sockcm_cookie sockc; 2573 __be16 proto; 2574 int err, reserve = 0; 2575 void *ph; 2576 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2577 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2578 int tp_len, size_max; 2579 unsigned char *addr; 2580 void *data; 2581 int len_sum = 0; 2582 int status = TP_STATUS_AVAILABLE; 2583 int hlen, tlen, copylen = 0; 2584 2585 mutex_lock(&po->pg_vec_lock); 2586 2587 if (likely(saddr == NULL)) { 2588 dev = packet_cached_dev_get(po); 2589 proto = po->num; 2590 addr = NULL; 2591 } else { 2592 err = -EINVAL; 2593 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2594 goto out; 2595 if (msg->msg_namelen < (saddr->sll_halen 2596 + offsetof(struct sockaddr_ll, 2597 sll_addr))) 2598 goto out; 2599 proto = saddr->sll_protocol; 2600 addr = saddr->sll_addr; 2601 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2602 } 2603 2604 sockc.tsflags = po->sk.sk_tsflags; 2605 if (msg->msg_controllen) { 2606 err = sock_cmsg_send(&po->sk, msg, &sockc); 2607 if (unlikely(err)) 2608 goto out; 2609 } 2610 2611 err = -ENXIO; 2612 if (unlikely(dev == NULL)) 2613 goto out; 2614 err = -ENETDOWN; 2615 if (unlikely(!(dev->flags & IFF_UP))) 2616 goto out_put; 2617 2618 if (po->sk.sk_socket->type == SOCK_RAW) 2619 reserve = dev->hard_header_len; 2620 size_max = po->tx_ring.frame_size 2621 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2622 2623 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) 2624 size_max = dev->mtu + reserve + VLAN_HLEN; 2625 2626 do { 2627 ph = packet_current_frame(po, &po->tx_ring, 2628 TP_STATUS_SEND_REQUEST); 2629 if (unlikely(ph == NULL)) { 2630 if (need_wait && need_resched()) 2631 schedule(); 2632 continue; 2633 } 2634 2635 skb = NULL; 2636 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2637 if (tp_len < 0) 2638 goto tpacket_error; 2639 2640 status = TP_STATUS_SEND_REQUEST; 2641 hlen = LL_RESERVED_SPACE(dev); 2642 tlen = dev->needed_tailroom; 2643 if (po->has_vnet_hdr) { 2644 vnet_hdr = data; 2645 data += sizeof(*vnet_hdr); 2646 tp_len -= sizeof(*vnet_hdr); 2647 if (tp_len < 0 || 2648 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2649 tp_len = -EINVAL; 2650 goto tpacket_error; 2651 } 2652 copylen = __virtio16_to_cpu(vio_le(), 2653 vnet_hdr->hdr_len); 2654 } 2655 copylen = max_t(int, copylen, dev->hard_header_len); 2656 skb = sock_alloc_send_skb(&po->sk, 2657 hlen + tlen + sizeof(struct sockaddr_ll) + 2658 (copylen - dev->hard_header_len), 2659 !need_wait, &err); 2660 2661 if (unlikely(skb == NULL)) { 2662 /* we assume the socket was initially writeable ... */ 2663 if (likely(len_sum > 0)) 2664 err = len_sum; 2665 goto out_status; 2666 } 2667 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2668 addr, hlen, copylen, &sockc); 2669 if (likely(tp_len >= 0) && 2670 tp_len > dev->mtu + reserve && 2671 !po->has_vnet_hdr && 2672 !packet_extra_vlan_len_allowed(dev, skb)) 2673 tp_len = -EMSGSIZE; 2674 2675 if (unlikely(tp_len < 0)) { 2676 tpacket_error: 2677 if (po->tp_loss) { 2678 __packet_set_status(po, ph, 2679 TP_STATUS_AVAILABLE); 2680 packet_increment_head(&po->tx_ring); 2681 kfree_skb(skb); 2682 continue; 2683 } else { 2684 status = TP_STATUS_WRONG_FORMAT; 2685 err = tp_len; 2686 goto out_status; 2687 } 2688 } 2689 2690 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr, 2691 vio_le())) { 2692 tp_len = -EINVAL; 2693 goto tpacket_error; 2694 } 2695 2696 packet_pick_tx_queue(dev, skb); 2697 2698 skb->destructor = tpacket_destruct_skb; 2699 __packet_set_status(po, ph, TP_STATUS_SENDING); 2700 packet_inc_pending(&po->tx_ring); 2701 2702 status = TP_STATUS_SEND_REQUEST; 2703 err = po->xmit(skb); 2704 if (unlikely(err > 0)) { 2705 err = net_xmit_errno(err); 2706 if (err && __packet_get_status(po, ph) == 2707 TP_STATUS_AVAILABLE) { 2708 /* skb was destructed already */ 2709 skb = NULL; 2710 goto out_status; 2711 } 2712 /* 2713 * skb was dropped but not destructed yet; 2714 * let's treat it like congestion or err < 0 2715 */ 2716 err = 0; 2717 } 2718 packet_increment_head(&po->tx_ring); 2719 len_sum += tp_len; 2720 } while (likely((ph != NULL) || 2721 /* Note: packet_read_pending() might be slow if we have 2722 * to call it as it's per_cpu variable, but in fast-path 2723 * we already short-circuit the loop with the first 2724 * condition, and luckily don't have to go that path 2725 * anyway. 2726 */ 2727 (need_wait && packet_read_pending(&po->tx_ring)))); 2728 2729 err = len_sum; 2730 goto out_put; 2731 2732 out_status: 2733 __packet_set_status(po, ph, status); 2734 kfree_skb(skb); 2735 out_put: 2736 dev_put(dev); 2737 out: 2738 mutex_unlock(&po->pg_vec_lock); 2739 return err; 2740 } 2741 2742 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2743 size_t reserve, size_t len, 2744 size_t linear, int noblock, 2745 int *err) 2746 { 2747 struct sk_buff *skb; 2748 2749 /* Under a page? Don't bother with paged skb. */ 2750 if (prepad + len < PAGE_SIZE || !linear) 2751 linear = len; 2752 2753 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2754 err, 0); 2755 if (!skb) 2756 return NULL; 2757 2758 skb_reserve(skb, reserve); 2759 skb_put(skb, linear); 2760 skb->data_len = len - linear; 2761 skb->len += len - linear; 2762 2763 return skb; 2764 } 2765 2766 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2767 { 2768 struct sock *sk = sock->sk; 2769 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2770 struct sk_buff *skb; 2771 struct net_device *dev; 2772 __be16 proto; 2773 unsigned char *addr; 2774 int err, reserve = 0; 2775 struct sockcm_cookie sockc; 2776 struct virtio_net_hdr vnet_hdr = { 0 }; 2777 int offset = 0; 2778 struct packet_sock *po = pkt_sk(sk); 2779 int hlen, tlen; 2780 int extra_len = 0; 2781 2782 /* 2783 * Get and verify the address. 2784 */ 2785 2786 if (likely(saddr == NULL)) { 2787 dev = packet_cached_dev_get(po); 2788 proto = po->num; 2789 addr = NULL; 2790 } else { 2791 err = -EINVAL; 2792 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2793 goto out; 2794 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2795 goto out; 2796 proto = saddr->sll_protocol; 2797 addr = saddr->sll_addr; 2798 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2799 } 2800 2801 err = -ENXIO; 2802 if (unlikely(dev == NULL)) 2803 goto out_unlock; 2804 err = -ENETDOWN; 2805 if (unlikely(!(dev->flags & IFF_UP))) 2806 goto out_unlock; 2807 2808 sockc.tsflags = sk->sk_tsflags; 2809 sockc.mark = sk->sk_mark; 2810 if (msg->msg_controllen) { 2811 err = sock_cmsg_send(sk, msg, &sockc); 2812 if (unlikely(err)) 2813 goto out_unlock; 2814 } 2815 2816 if (sock->type == SOCK_RAW) 2817 reserve = dev->hard_header_len; 2818 if (po->has_vnet_hdr) { 2819 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2820 if (err) 2821 goto out_unlock; 2822 } 2823 2824 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2825 if (!netif_supports_nofcs(dev)) { 2826 err = -EPROTONOSUPPORT; 2827 goto out_unlock; 2828 } 2829 extra_len = 4; /* We're doing our own CRC */ 2830 } 2831 2832 err = -EMSGSIZE; 2833 if (!vnet_hdr.gso_type && 2834 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2835 goto out_unlock; 2836 2837 err = -ENOBUFS; 2838 hlen = LL_RESERVED_SPACE(dev); 2839 tlen = dev->needed_tailroom; 2840 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, 2841 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len), 2842 msg->msg_flags & MSG_DONTWAIT, &err); 2843 if (skb == NULL) 2844 goto out_unlock; 2845 2846 skb_set_network_header(skb, reserve); 2847 2848 err = -EINVAL; 2849 if (sock->type == SOCK_DGRAM) { 2850 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 2851 if (unlikely(offset < 0)) 2852 goto out_free; 2853 } 2854 2855 /* Returns -EFAULT on error */ 2856 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 2857 if (err) 2858 goto out_free; 2859 2860 if (sock->type == SOCK_RAW && 2861 !dev_validate_header(dev, skb->data, len)) { 2862 err = -EINVAL; 2863 goto out_free; 2864 } 2865 2866 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags); 2867 2868 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 2869 !packet_extra_vlan_len_allowed(dev, skb)) { 2870 err = -EMSGSIZE; 2871 goto out_free; 2872 } 2873 2874 skb->protocol = proto; 2875 skb->dev = dev; 2876 skb->priority = sk->sk_priority; 2877 skb->mark = sockc.mark; 2878 2879 packet_pick_tx_queue(dev, skb); 2880 2881 if (po->has_vnet_hdr) { 2882 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 2883 if (err) 2884 goto out_free; 2885 len += sizeof(vnet_hdr); 2886 } 2887 2888 skb_probe_transport_header(skb, reserve); 2889 2890 if (unlikely(extra_len == 4)) 2891 skb->no_fcs = 1; 2892 2893 err = po->xmit(skb); 2894 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2895 goto out_unlock; 2896 2897 dev_put(dev); 2898 2899 return len; 2900 2901 out_free: 2902 kfree_skb(skb); 2903 out_unlock: 2904 if (dev) 2905 dev_put(dev); 2906 out: 2907 return err; 2908 } 2909 2910 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2911 { 2912 struct sock *sk = sock->sk; 2913 struct packet_sock *po = pkt_sk(sk); 2914 2915 if (po->tx_ring.pg_vec) 2916 return tpacket_snd(po, msg); 2917 else 2918 return packet_snd(sock, msg, len); 2919 } 2920 2921 /* 2922 * Close a PACKET socket. This is fairly simple. We immediately go 2923 * to 'closed' state and remove our protocol entry in the device list. 2924 */ 2925 2926 static int packet_release(struct socket *sock) 2927 { 2928 struct sock *sk = sock->sk; 2929 struct packet_sock *po; 2930 struct net *net; 2931 union tpacket_req_u req_u; 2932 2933 if (!sk) 2934 return 0; 2935 2936 net = sock_net(sk); 2937 po = pkt_sk(sk); 2938 2939 mutex_lock(&net->packet.sklist_lock); 2940 sk_del_node_init_rcu(sk); 2941 mutex_unlock(&net->packet.sklist_lock); 2942 2943 preempt_disable(); 2944 sock_prot_inuse_add(net, sk->sk_prot, -1); 2945 preempt_enable(); 2946 2947 spin_lock(&po->bind_lock); 2948 unregister_prot_hook(sk, false); 2949 packet_cached_dev_reset(po); 2950 2951 if (po->prot_hook.dev) { 2952 dev_put(po->prot_hook.dev); 2953 po->prot_hook.dev = NULL; 2954 } 2955 spin_unlock(&po->bind_lock); 2956 2957 packet_flush_mclist(sk); 2958 2959 if (po->rx_ring.pg_vec) { 2960 memset(&req_u, 0, sizeof(req_u)); 2961 packet_set_ring(sk, &req_u, 1, 0); 2962 } 2963 2964 if (po->tx_ring.pg_vec) { 2965 memset(&req_u, 0, sizeof(req_u)); 2966 packet_set_ring(sk, &req_u, 1, 1); 2967 } 2968 2969 fanout_release(sk); 2970 2971 synchronize_net(); 2972 /* 2973 * Now the socket is dead. No more input will appear. 2974 */ 2975 sock_orphan(sk); 2976 sock->sk = NULL; 2977 2978 /* Purge queues */ 2979 2980 skb_queue_purge(&sk->sk_receive_queue); 2981 packet_free_pending(po); 2982 sk_refcnt_debug_release(sk); 2983 2984 sock_put(sk); 2985 return 0; 2986 } 2987 2988 /* 2989 * Attach a packet hook. 2990 */ 2991 2992 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 2993 __be16 proto) 2994 { 2995 struct packet_sock *po = pkt_sk(sk); 2996 struct net_device *dev_curr; 2997 __be16 proto_curr; 2998 bool need_rehook; 2999 struct net_device *dev = NULL; 3000 int ret = 0; 3001 bool unlisted = false; 3002 3003 if (po->fanout) 3004 return -EINVAL; 3005 3006 lock_sock(sk); 3007 spin_lock(&po->bind_lock); 3008 rcu_read_lock(); 3009 3010 if (name) { 3011 dev = dev_get_by_name_rcu(sock_net(sk), name); 3012 if (!dev) { 3013 ret = -ENODEV; 3014 goto out_unlock; 3015 } 3016 } else if (ifindex) { 3017 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3018 if (!dev) { 3019 ret = -ENODEV; 3020 goto out_unlock; 3021 } 3022 } 3023 3024 if (dev) 3025 dev_hold(dev); 3026 3027 proto_curr = po->prot_hook.type; 3028 dev_curr = po->prot_hook.dev; 3029 3030 need_rehook = proto_curr != proto || dev_curr != dev; 3031 3032 if (need_rehook) { 3033 if (po->running) { 3034 rcu_read_unlock(); 3035 __unregister_prot_hook(sk, true); 3036 rcu_read_lock(); 3037 dev_curr = po->prot_hook.dev; 3038 if (dev) 3039 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3040 dev->ifindex); 3041 } 3042 3043 po->num = proto; 3044 po->prot_hook.type = proto; 3045 3046 if (unlikely(unlisted)) { 3047 dev_put(dev); 3048 po->prot_hook.dev = NULL; 3049 po->ifindex = -1; 3050 packet_cached_dev_reset(po); 3051 } else { 3052 po->prot_hook.dev = dev; 3053 po->ifindex = dev ? dev->ifindex : 0; 3054 packet_cached_dev_assign(po, dev); 3055 } 3056 } 3057 if (dev_curr) 3058 dev_put(dev_curr); 3059 3060 if (proto == 0 || !need_rehook) 3061 goto out_unlock; 3062 3063 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3064 register_prot_hook(sk); 3065 } else { 3066 sk->sk_err = ENETDOWN; 3067 if (!sock_flag(sk, SOCK_DEAD)) 3068 sk->sk_error_report(sk); 3069 } 3070 3071 out_unlock: 3072 rcu_read_unlock(); 3073 spin_unlock(&po->bind_lock); 3074 release_sock(sk); 3075 return ret; 3076 } 3077 3078 /* 3079 * Bind a packet socket to a device 3080 */ 3081 3082 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3083 int addr_len) 3084 { 3085 struct sock *sk = sock->sk; 3086 char name[15]; 3087 3088 /* 3089 * Check legality 3090 */ 3091 3092 if (addr_len != sizeof(struct sockaddr)) 3093 return -EINVAL; 3094 strlcpy(name, uaddr->sa_data, sizeof(name)); 3095 3096 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3097 } 3098 3099 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3100 { 3101 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3102 struct sock *sk = sock->sk; 3103 3104 /* 3105 * Check legality 3106 */ 3107 3108 if (addr_len < sizeof(struct sockaddr_ll)) 3109 return -EINVAL; 3110 if (sll->sll_family != AF_PACKET) 3111 return -EINVAL; 3112 3113 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3114 sll->sll_protocol ? : pkt_sk(sk)->num); 3115 } 3116 3117 static struct proto packet_proto = { 3118 .name = "PACKET", 3119 .owner = THIS_MODULE, 3120 .obj_size = sizeof(struct packet_sock), 3121 }; 3122 3123 /* 3124 * Create a packet of type SOCK_PACKET. 3125 */ 3126 3127 static int packet_create(struct net *net, struct socket *sock, int protocol, 3128 int kern) 3129 { 3130 struct sock *sk; 3131 struct packet_sock *po; 3132 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3133 int err; 3134 3135 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3136 return -EPERM; 3137 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3138 sock->type != SOCK_PACKET) 3139 return -ESOCKTNOSUPPORT; 3140 3141 sock->state = SS_UNCONNECTED; 3142 3143 err = -ENOBUFS; 3144 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3145 if (sk == NULL) 3146 goto out; 3147 3148 sock->ops = &packet_ops; 3149 if (sock->type == SOCK_PACKET) 3150 sock->ops = &packet_ops_spkt; 3151 3152 sock_init_data(sock, sk); 3153 3154 po = pkt_sk(sk); 3155 sk->sk_family = PF_PACKET; 3156 po->num = proto; 3157 po->xmit = dev_queue_xmit; 3158 3159 err = packet_alloc_pending(po); 3160 if (err) 3161 goto out2; 3162 3163 packet_cached_dev_reset(po); 3164 3165 sk->sk_destruct = packet_sock_destruct; 3166 sk_refcnt_debug_inc(sk); 3167 3168 /* 3169 * Attach a protocol block 3170 */ 3171 3172 spin_lock_init(&po->bind_lock); 3173 mutex_init(&po->pg_vec_lock); 3174 po->rollover = NULL; 3175 po->prot_hook.func = packet_rcv; 3176 3177 if (sock->type == SOCK_PACKET) 3178 po->prot_hook.func = packet_rcv_spkt; 3179 3180 po->prot_hook.af_packet_priv = sk; 3181 3182 if (proto) { 3183 po->prot_hook.type = proto; 3184 register_prot_hook(sk); 3185 } 3186 3187 mutex_lock(&net->packet.sklist_lock); 3188 sk_add_node_rcu(sk, &net->packet.sklist); 3189 mutex_unlock(&net->packet.sklist_lock); 3190 3191 preempt_disable(); 3192 sock_prot_inuse_add(net, &packet_proto, 1); 3193 preempt_enable(); 3194 3195 return 0; 3196 out2: 3197 sk_free(sk); 3198 out: 3199 return err; 3200 } 3201 3202 /* 3203 * Pull a packet from our receive queue and hand it to the user. 3204 * If necessary we block. 3205 */ 3206 3207 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3208 int flags) 3209 { 3210 struct sock *sk = sock->sk; 3211 struct sk_buff *skb; 3212 int copied, err; 3213 int vnet_hdr_len = 0; 3214 unsigned int origlen = 0; 3215 3216 err = -EINVAL; 3217 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3218 goto out; 3219 3220 #if 0 3221 /* What error should we return now? EUNATTACH? */ 3222 if (pkt_sk(sk)->ifindex < 0) 3223 return -ENODEV; 3224 #endif 3225 3226 if (flags & MSG_ERRQUEUE) { 3227 err = sock_recv_errqueue(sk, msg, len, 3228 SOL_PACKET, PACKET_TX_TIMESTAMP); 3229 goto out; 3230 } 3231 3232 /* 3233 * Call the generic datagram receiver. This handles all sorts 3234 * of horrible races and re-entrancy so we can forget about it 3235 * in the protocol layers. 3236 * 3237 * Now it will return ENETDOWN, if device have just gone down, 3238 * but then it will block. 3239 */ 3240 3241 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 3242 3243 /* 3244 * An error occurred so return it. Because skb_recv_datagram() 3245 * handles the blocking we don't see and worry about blocking 3246 * retries. 3247 */ 3248 3249 if (skb == NULL) 3250 goto out; 3251 3252 if (pkt_sk(sk)->pressure) 3253 packet_rcv_has_room(pkt_sk(sk), NULL); 3254 3255 if (pkt_sk(sk)->has_vnet_hdr) { 3256 err = packet_rcv_vnet(msg, skb, &len); 3257 if (err) 3258 goto out_free; 3259 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3260 } 3261 3262 /* You lose any data beyond the buffer you gave. If it worries 3263 * a user program they can ask the device for its MTU 3264 * anyway. 3265 */ 3266 copied = skb->len; 3267 if (copied > len) { 3268 copied = len; 3269 msg->msg_flags |= MSG_TRUNC; 3270 } 3271 3272 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3273 if (err) 3274 goto out_free; 3275 3276 if (sock->type != SOCK_PACKET) { 3277 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3278 3279 /* Original length was stored in sockaddr_ll fields */ 3280 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3281 sll->sll_family = AF_PACKET; 3282 sll->sll_protocol = skb->protocol; 3283 } 3284 3285 sock_recv_ts_and_drops(msg, sk, skb); 3286 3287 if (msg->msg_name) { 3288 /* If the address length field is there to be filled 3289 * in, we fill it in now. 3290 */ 3291 if (sock->type == SOCK_PACKET) { 3292 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3293 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3294 } else { 3295 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3296 3297 msg->msg_namelen = sll->sll_halen + 3298 offsetof(struct sockaddr_ll, sll_addr); 3299 } 3300 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, 3301 msg->msg_namelen); 3302 } 3303 3304 if (pkt_sk(sk)->auxdata) { 3305 struct tpacket_auxdata aux; 3306 3307 aux.tp_status = TP_STATUS_USER; 3308 if (skb->ip_summed == CHECKSUM_PARTIAL) 3309 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3310 else if (skb->pkt_type != PACKET_OUTGOING && 3311 (skb->ip_summed == CHECKSUM_COMPLETE || 3312 skb_csum_unnecessary(skb))) 3313 aux.tp_status |= TP_STATUS_CSUM_VALID; 3314 3315 aux.tp_len = origlen; 3316 aux.tp_snaplen = skb->len; 3317 aux.tp_mac = 0; 3318 aux.tp_net = skb_network_offset(skb); 3319 if (skb_vlan_tag_present(skb)) { 3320 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3321 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3322 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3323 } else { 3324 aux.tp_vlan_tci = 0; 3325 aux.tp_vlan_tpid = 0; 3326 } 3327 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3328 } 3329 3330 /* 3331 * Free or return the buffer as appropriate. Again this 3332 * hides all the races and re-entrancy issues from us. 3333 */ 3334 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3335 3336 out_free: 3337 skb_free_datagram(sk, skb); 3338 out: 3339 return err; 3340 } 3341 3342 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3343 int *uaddr_len, int peer) 3344 { 3345 struct net_device *dev; 3346 struct sock *sk = sock->sk; 3347 3348 if (peer) 3349 return -EOPNOTSUPP; 3350 3351 uaddr->sa_family = AF_PACKET; 3352 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3353 rcu_read_lock(); 3354 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3355 if (dev) 3356 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3357 rcu_read_unlock(); 3358 *uaddr_len = sizeof(*uaddr); 3359 3360 return 0; 3361 } 3362 3363 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3364 int *uaddr_len, int peer) 3365 { 3366 struct net_device *dev; 3367 struct sock *sk = sock->sk; 3368 struct packet_sock *po = pkt_sk(sk); 3369 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3370 3371 if (peer) 3372 return -EOPNOTSUPP; 3373 3374 sll->sll_family = AF_PACKET; 3375 sll->sll_ifindex = po->ifindex; 3376 sll->sll_protocol = po->num; 3377 sll->sll_pkttype = 0; 3378 rcu_read_lock(); 3379 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 3380 if (dev) { 3381 sll->sll_hatype = dev->type; 3382 sll->sll_halen = dev->addr_len; 3383 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3384 } else { 3385 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3386 sll->sll_halen = 0; 3387 } 3388 rcu_read_unlock(); 3389 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3390 3391 return 0; 3392 } 3393 3394 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3395 int what) 3396 { 3397 switch (i->type) { 3398 case PACKET_MR_MULTICAST: 3399 if (i->alen != dev->addr_len) 3400 return -EINVAL; 3401 if (what > 0) 3402 return dev_mc_add(dev, i->addr); 3403 else 3404 return dev_mc_del(dev, i->addr); 3405 break; 3406 case PACKET_MR_PROMISC: 3407 return dev_set_promiscuity(dev, what); 3408 case PACKET_MR_ALLMULTI: 3409 return dev_set_allmulti(dev, what); 3410 case PACKET_MR_UNICAST: 3411 if (i->alen != dev->addr_len) 3412 return -EINVAL; 3413 if (what > 0) 3414 return dev_uc_add(dev, i->addr); 3415 else 3416 return dev_uc_del(dev, i->addr); 3417 break; 3418 default: 3419 break; 3420 } 3421 return 0; 3422 } 3423 3424 static void packet_dev_mclist_delete(struct net_device *dev, 3425 struct packet_mclist **mlp) 3426 { 3427 struct packet_mclist *ml; 3428 3429 while ((ml = *mlp) != NULL) { 3430 if (ml->ifindex == dev->ifindex) { 3431 packet_dev_mc(dev, ml, -1); 3432 *mlp = ml->next; 3433 kfree(ml); 3434 } else 3435 mlp = &ml->next; 3436 } 3437 } 3438 3439 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3440 { 3441 struct packet_sock *po = pkt_sk(sk); 3442 struct packet_mclist *ml, *i; 3443 struct net_device *dev; 3444 int err; 3445 3446 rtnl_lock(); 3447 3448 err = -ENODEV; 3449 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3450 if (!dev) 3451 goto done; 3452 3453 err = -EINVAL; 3454 if (mreq->mr_alen > dev->addr_len) 3455 goto done; 3456 3457 err = -ENOBUFS; 3458 i = kmalloc(sizeof(*i), GFP_KERNEL); 3459 if (i == NULL) 3460 goto done; 3461 3462 err = 0; 3463 for (ml = po->mclist; ml; ml = ml->next) { 3464 if (ml->ifindex == mreq->mr_ifindex && 3465 ml->type == mreq->mr_type && 3466 ml->alen == mreq->mr_alen && 3467 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3468 ml->count++; 3469 /* Free the new element ... */ 3470 kfree(i); 3471 goto done; 3472 } 3473 } 3474 3475 i->type = mreq->mr_type; 3476 i->ifindex = mreq->mr_ifindex; 3477 i->alen = mreq->mr_alen; 3478 memcpy(i->addr, mreq->mr_address, i->alen); 3479 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3480 i->count = 1; 3481 i->next = po->mclist; 3482 po->mclist = i; 3483 err = packet_dev_mc(dev, i, 1); 3484 if (err) { 3485 po->mclist = i->next; 3486 kfree(i); 3487 } 3488 3489 done: 3490 rtnl_unlock(); 3491 return err; 3492 } 3493 3494 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3495 { 3496 struct packet_mclist *ml, **mlp; 3497 3498 rtnl_lock(); 3499 3500 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3501 if (ml->ifindex == mreq->mr_ifindex && 3502 ml->type == mreq->mr_type && 3503 ml->alen == mreq->mr_alen && 3504 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3505 if (--ml->count == 0) { 3506 struct net_device *dev; 3507 *mlp = ml->next; 3508 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3509 if (dev) 3510 packet_dev_mc(dev, ml, -1); 3511 kfree(ml); 3512 } 3513 break; 3514 } 3515 } 3516 rtnl_unlock(); 3517 return 0; 3518 } 3519 3520 static void packet_flush_mclist(struct sock *sk) 3521 { 3522 struct packet_sock *po = pkt_sk(sk); 3523 struct packet_mclist *ml; 3524 3525 if (!po->mclist) 3526 return; 3527 3528 rtnl_lock(); 3529 while ((ml = po->mclist) != NULL) { 3530 struct net_device *dev; 3531 3532 po->mclist = ml->next; 3533 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3534 if (dev != NULL) 3535 packet_dev_mc(dev, ml, -1); 3536 kfree(ml); 3537 } 3538 rtnl_unlock(); 3539 } 3540 3541 static int 3542 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3543 { 3544 struct sock *sk = sock->sk; 3545 struct packet_sock *po = pkt_sk(sk); 3546 int ret; 3547 3548 if (level != SOL_PACKET) 3549 return -ENOPROTOOPT; 3550 3551 switch (optname) { 3552 case PACKET_ADD_MEMBERSHIP: 3553 case PACKET_DROP_MEMBERSHIP: 3554 { 3555 struct packet_mreq_max mreq; 3556 int len = optlen; 3557 memset(&mreq, 0, sizeof(mreq)); 3558 if (len < sizeof(struct packet_mreq)) 3559 return -EINVAL; 3560 if (len > sizeof(mreq)) 3561 len = sizeof(mreq); 3562 if (copy_from_user(&mreq, optval, len)) 3563 return -EFAULT; 3564 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3565 return -EINVAL; 3566 if (optname == PACKET_ADD_MEMBERSHIP) 3567 ret = packet_mc_add(sk, &mreq); 3568 else 3569 ret = packet_mc_drop(sk, &mreq); 3570 return ret; 3571 } 3572 3573 case PACKET_RX_RING: 3574 case PACKET_TX_RING: 3575 { 3576 union tpacket_req_u req_u; 3577 int len; 3578 3579 switch (po->tp_version) { 3580 case TPACKET_V1: 3581 case TPACKET_V2: 3582 len = sizeof(req_u.req); 3583 break; 3584 case TPACKET_V3: 3585 default: 3586 len = sizeof(req_u.req3); 3587 break; 3588 } 3589 if (optlen < len) 3590 return -EINVAL; 3591 if (copy_from_user(&req_u.req, optval, len)) 3592 return -EFAULT; 3593 return packet_set_ring(sk, &req_u, 0, 3594 optname == PACKET_TX_RING); 3595 } 3596 case PACKET_COPY_THRESH: 3597 { 3598 int val; 3599 3600 if (optlen != sizeof(val)) 3601 return -EINVAL; 3602 if (copy_from_user(&val, optval, sizeof(val))) 3603 return -EFAULT; 3604 3605 pkt_sk(sk)->copy_thresh = val; 3606 return 0; 3607 } 3608 case PACKET_VERSION: 3609 { 3610 int val; 3611 3612 if (optlen != sizeof(val)) 3613 return -EINVAL; 3614 if (copy_from_user(&val, optval, sizeof(val))) 3615 return -EFAULT; 3616 switch (val) { 3617 case TPACKET_V1: 3618 case TPACKET_V2: 3619 case TPACKET_V3: 3620 break; 3621 default: 3622 return -EINVAL; 3623 } 3624 lock_sock(sk); 3625 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3626 ret = -EBUSY; 3627 } else { 3628 po->tp_version = val; 3629 ret = 0; 3630 } 3631 release_sock(sk); 3632 return ret; 3633 } 3634 case PACKET_RESERVE: 3635 { 3636 unsigned int val; 3637 3638 if (optlen != sizeof(val)) 3639 return -EINVAL; 3640 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3641 return -EBUSY; 3642 if (copy_from_user(&val, optval, sizeof(val))) 3643 return -EFAULT; 3644 po->tp_reserve = val; 3645 return 0; 3646 } 3647 case PACKET_LOSS: 3648 { 3649 unsigned int val; 3650 3651 if (optlen != sizeof(val)) 3652 return -EINVAL; 3653 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3654 return -EBUSY; 3655 if (copy_from_user(&val, optval, sizeof(val))) 3656 return -EFAULT; 3657 po->tp_loss = !!val; 3658 return 0; 3659 } 3660 case PACKET_AUXDATA: 3661 { 3662 int val; 3663 3664 if (optlen < sizeof(val)) 3665 return -EINVAL; 3666 if (copy_from_user(&val, optval, sizeof(val))) 3667 return -EFAULT; 3668 3669 po->auxdata = !!val; 3670 return 0; 3671 } 3672 case PACKET_ORIGDEV: 3673 { 3674 int val; 3675 3676 if (optlen < sizeof(val)) 3677 return -EINVAL; 3678 if (copy_from_user(&val, optval, sizeof(val))) 3679 return -EFAULT; 3680 3681 po->origdev = !!val; 3682 return 0; 3683 } 3684 case PACKET_VNET_HDR: 3685 { 3686 int val; 3687 3688 if (sock->type != SOCK_RAW) 3689 return -EINVAL; 3690 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3691 return -EBUSY; 3692 if (optlen < sizeof(val)) 3693 return -EINVAL; 3694 if (copy_from_user(&val, optval, sizeof(val))) 3695 return -EFAULT; 3696 3697 po->has_vnet_hdr = !!val; 3698 return 0; 3699 } 3700 case PACKET_TIMESTAMP: 3701 { 3702 int val; 3703 3704 if (optlen != sizeof(val)) 3705 return -EINVAL; 3706 if (copy_from_user(&val, optval, sizeof(val))) 3707 return -EFAULT; 3708 3709 po->tp_tstamp = val; 3710 return 0; 3711 } 3712 case PACKET_FANOUT: 3713 { 3714 int val; 3715 3716 if (optlen != sizeof(val)) 3717 return -EINVAL; 3718 if (copy_from_user(&val, optval, sizeof(val))) 3719 return -EFAULT; 3720 3721 return fanout_add(sk, val & 0xffff, val >> 16); 3722 } 3723 case PACKET_FANOUT_DATA: 3724 { 3725 if (!po->fanout) 3726 return -EINVAL; 3727 3728 return fanout_set_data(po, optval, optlen); 3729 } 3730 case PACKET_TX_HAS_OFF: 3731 { 3732 unsigned int val; 3733 3734 if (optlen != sizeof(val)) 3735 return -EINVAL; 3736 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3737 return -EBUSY; 3738 if (copy_from_user(&val, optval, sizeof(val))) 3739 return -EFAULT; 3740 po->tp_tx_has_off = !!val; 3741 return 0; 3742 } 3743 case PACKET_QDISC_BYPASS: 3744 { 3745 int val; 3746 3747 if (optlen != sizeof(val)) 3748 return -EINVAL; 3749 if (copy_from_user(&val, optval, sizeof(val))) 3750 return -EFAULT; 3751 3752 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; 3753 return 0; 3754 } 3755 default: 3756 return -ENOPROTOOPT; 3757 } 3758 } 3759 3760 static int packet_getsockopt(struct socket *sock, int level, int optname, 3761 char __user *optval, int __user *optlen) 3762 { 3763 int len; 3764 int val, lv = sizeof(val); 3765 struct sock *sk = sock->sk; 3766 struct packet_sock *po = pkt_sk(sk); 3767 void *data = &val; 3768 union tpacket_stats_u st; 3769 struct tpacket_rollover_stats rstats; 3770 3771 if (level != SOL_PACKET) 3772 return -ENOPROTOOPT; 3773 3774 if (get_user(len, optlen)) 3775 return -EFAULT; 3776 3777 if (len < 0) 3778 return -EINVAL; 3779 3780 switch (optname) { 3781 case PACKET_STATISTICS: 3782 spin_lock_bh(&sk->sk_receive_queue.lock); 3783 memcpy(&st, &po->stats, sizeof(st)); 3784 memset(&po->stats, 0, sizeof(po->stats)); 3785 spin_unlock_bh(&sk->sk_receive_queue.lock); 3786 3787 if (po->tp_version == TPACKET_V3) { 3788 lv = sizeof(struct tpacket_stats_v3); 3789 st.stats3.tp_packets += st.stats3.tp_drops; 3790 data = &st.stats3; 3791 } else { 3792 lv = sizeof(struct tpacket_stats); 3793 st.stats1.tp_packets += st.stats1.tp_drops; 3794 data = &st.stats1; 3795 } 3796 3797 break; 3798 case PACKET_AUXDATA: 3799 val = po->auxdata; 3800 break; 3801 case PACKET_ORIGDEV: 3802 val = po->origdev; 3803 break; 3804 case PACKET_VNET_HDR: 3805 val = po->has_vnet_hdr; 3806 break; 3807 case PACKET_VERSION: 3808 val = po->tp_version; 3809 break; 3810 case PACKET_HDRLEN: 3811 if (len > sizeof(int)) 3812 len = sizeof(int); 3813 if (copy_from_user(&val, optval, len)) 3814 return -EFAULT; 3815 switch (val) { 3816 case TPACKET_V1: 3817 val = sizeof(struct tpacket_hdr); 3818 break; 3819 case TPACKET_V2: 3820 val = sizeof(struct tpacket2_hdr); 3821 break; 3822 case TPACKET_V3: 3823 val = sizeof(struct tpacket3_hdr); 3824 break; 3825 default: 3826 return -EINVAL; 3827 } 3828 break; 3829 case PACKET_RESERVE: 3830 val = po->tp_reserve; 3831 break; 3832 case PACKET_LOSS: 3833 val = po->tp_loss; 3834 break; 3835 case PACKET_TIMESTAMP: 3836 val = po->tp_tstamp; 3837 break; 3838 case PACKET_FANOUT: 3839 val = (po->fanout ? 3840 ((u32)po->fanout->id | 3841 ((u32)po->fanout->type << 16) | 3842 ((u32)po->fanout->flags << 24)) : 3843 0); 3844 break; 3845 case PACKET_ROLLOVER_STATS: 3846 if (!po->rollover) 3847 return -EINVAL; 3848 rstats.tp_all = atomic_long_read(&po->rollover->num); 3849 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 3850 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 3851 data = &rstats; 3852 lv = sizeof(rstats); 3853 break; 3854 case PACKET_TX_HAS_OFF: 3855 val = po->tp_tx_has_off; 3856 break; 3857 case PACKET_QDISC_BYPASS: 3858 val = packet_use_direct_xmit(po); 3859 break; 3860 default: 3861 return -ENOPROTOOPT; 3862 } 3863 3864 if (len > lv) 3865 len = lv; 3866 if (put_user(len, optlen)) 3867 return -EFAULT; 3868 if (copy_to_user(optval, data, len)) 3869 return -EFAULT; 3870 return 0; 3871 } 3872 3873 3874 #ifdef CONFIG_COMPAT 3875 static int compat_packet_setsockopt(struct socket *sock, int level, int optname, 3876 char __user *optval, unsigned int optlen) 3877 { 3878 struct packet_sock *po = pkt_sk(sock->sk); 3879 3880 if (level != SOL_PACKET) 3881 return -ENOPROTOOPT; 3882 3883 if (optname == PACKET_FANOUT_DATA && 3884 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) { 3885 optval = (char __user *)get_compat_bpf_fprog(optval); 3886 if (!optval) 3887 return -EFAULT; 3888 optlen = sizeof(struct sock_fprog); 3889 } 3890 3891 return packet_setsockopt(sock, level, optname, optval, optlen); 3892 } 3893 #endif 3894 3895 static int packet_notifier(struct notifier_block *this, 3896 unsigned long msg, void *ptr) 3897 { 3898 struct sock *sk; 3899 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 3900 struct net *net = dev_net(dev); 3901 3902 rcu_read_lock(); 3903 sk_for_each_rcu(sk, &net->packet.sklist) { 3904 struct packet_sock *po = pkt_sk(sk); 3905 3906 switch (msg) { 3907 case NETDEV_UNREGISTER: 3908 if (po->mclist) 3909 packet_dev_mclist_delete(dev, &po->mclist); 3910 /* fallthrough */ 3911 3912 case NETDEV_DOWN: 3913 if (dev->ifindex == po->ifindex) { 3914 spin_lock(&po->bind_lock); 3915 if (po->running) { 3916 __unregister_prot_hook(sk, false); 3917 sk->sk_err = ENETDOWN; 3918 if (!sock_flag(sk, SOCK_DEAD)) 3919 sk->sk_error_report(sk); 3920 } 3921 if (msg == NETDEV_UNREGISTER) { 3922 packet_cached_dev_reset(po); 3923 fanout_release(sk); 3924 po->ifindex = -1; 3925 if (po->prot_hook.dev) 3926 dev_put(po->prot_hook.dev); 3927 po->prot_hook.dev = NULL; 3928 } 3929 spin_unlock(&po->bind_lock); 3930 } 3931 break; 3932 case NETDEV_UP: 3933 if (dev->ifindex == po->ifindex) { 3934 spin_lock(&po->bind_lock); 3935 if (po->num) 3936 register_prot_hook(sk); 3937 spin_unlock(&po->bind_lock); 3938 } 3939 break; 3940 } 3941 } 3942 rcu_read_unlock(); 3943 return NOTIFY_DONE; 3944 } 3945 3946 3947 static int packet_ioctl(struct socket *sock, unsigned int cmd, 3948 unsigned long arg) 3949 { 3950 struct sock *sk = sock->sk; 3951 3952 switch (cmd) { 3953 case SIOCOUTQ: 3954 { 3955 int amount = sk_wmem_alloc_get(sk); 3956 3957 return put_user(amount, (int __user *)arg); 3958 } 3959 case SIOCINQ: 3960 { 3961 struct sk_buff *skb; 3962 int amount = 0; 3963 3964 spin_lock_bh(&sk->sk_receive_queue.lock); 3965 skb = skb_peek(&sk->sk_receive_queue); 3966 if (skb) 3967 amount = skb->len; 3968 spin_unlock_bh(&sk->sk_receive_queue.lock); 3969 return put_user(amount, (int __user *)arg); 3970 } 3971 case SIOCGSTAMP: 3972 return sock_get_timestamp(sk, (struct timeval __user *)arg); 3973 case SIOCGSTAMPNS: 3974 return sock_get_timestampns(sk, (struct timespec __user *)arg); 3975 3976 #ifdef CONFIG_INET 3977 case SIOCADDRT: 3978 case SIOCDELRT: 3979 case SIOCDARP: 3980 case SIOCGARP: 3981 case SIOCSARP: 3982 case SIOCGIFADDR: 3983 case SIOCSIFADDR: 3984 case SIOCGIFBRDADDR: 3985 case SIOCSIFBRDADDR: 3986 case SIOCGIFNETMASK: 3987 case SIOCSIFNETMASK: 3988 case SIOCGIFDSTADDR: 3989 case SIOCSIFDSTADDR: 3990 case SIOCSIFFLAGS: 3991 return inet_dgram_ops.ioctl(sock, cmd, arg); 3992 #endif 3993 3994 default: 3995 return -ENOIOCTLCMD; 3996 } 3997 return 0; 3998 } 3999 4000 static unsigned int packet_poll(struct file *file, struct socket *sock, 4001 poll_table *wait) 4002 { 4003 struct sock *sk = sock->sk; 4004 struct packet_sock *po = pkt_sk(sk); 4005 unsigned int mask = datagram_poll(file, sock, wait); 4006 4007 spin_lock_bh(&sk->sk_receive_queue.lock); 4008 if (po->rx_ring.pg_vec) { 4009 if (!packet_previous_rx_frame(po, &po->rx_ring, 4010 TP_STATUS_KERNEL)) 4011 mask |= POLLIN | POLLRDNORM; 4012 } 4013 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 4014 po->pressure = 0; 4015 spin_unlock_bh(&sk->sk_receive_queue.lock); 4016 spin_lock_bh(&sk->sk_write_queue.lock); 4017 if (po->tx_ring.pg_vec) { 4018 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4019 mask |= POLLOUT | POLLWRNORM; 4020 } 4021 spin_unlock_bh(&sk->sk_write_queue.lock); 4022 return mask; 4023 } 4024 4025 4026 /* Dirty? Well, I still did not learn better way to account 4027 * for user mmaps. 4028 */ 4029 4030 static void packet_mm_open(struct vm_area_struct *vma) 4031 { 4032 struct file *file = vma->vm_file; 4033 struct socket *sock = file->private_data; 4034 struct sock *sk = sock->sk; 4035 4036 if (sk) 4037 atomic_inc(&pkt_sk(sk)->mapped); 4038 } 4039 4040 static void packet_mm_close(struct vm_area_struct *vma) 4041 { 4042 struct file *file = vma->vm_file; 4043 struct socket *sock = file->private_data; 4044 struct sock *sk = sock->sk; 4045 4046 if (sk) 4047 atomic_dec(&pkt_sk(sk)->mapped); 4048 } 4049 4050 static const struct vm_operations_struct packet_mmap_ops = { 4051 .open = packet_mm_open, 4052 .close = packet_mm_close, 4053 }; 4054 4055 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4056 unsigned int len) 4057 { 4058 int i; 4059 4060 for (i = 0; i < len; i++) { 4061 if (likely(pg_vec[i].buffer)) { 4062 if (is_vmalloc_addr(pg_vec[i].buffer)) 4063 vfree(pg_vec[i].buffer); 4064 else 4065 free_pages((unsigned long)pg_vec[i].buffer, 4066 order); 4067 pg_vec[i].buffer = NULL; 4068 } 4069 } 4070 kfree(pg_vec); 4071 } 4072 4073 static char *alloc_one_pg_vec_page(unsigned long order) 4074 { 4075 char *buffer; 4076 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4077 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4078 4079 buffer = (char *) __get_free_pages(gfp_flags, order); 4080 if (buffer) 4081 return buffer; 4082 4083 /* __get_free_pages failed, fall back to vmalloc */ 4084 buffer = vzalloc((1 << order) * PAGE_SIZE); 4085 if (buffer) 4086 return buffer; 4087 4088 /* vmalloc failed, lets dig into swap here */ 4089 gfp_flags &= ~__GFP_NORETRY; 4090 buffer = (char *) __get_free_pages(gfp_flags, order); 4091 if (buffer) 4092 return buffer; 4093 4094 /* complete and utter failure */ 4095 return NULL; 4096 } 4097 4098 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4099 { 4100 unsigned int block_nr = req->tp_block_nr; 4101 struct pgv *pg_vec; 4102 int i; 4103 4104 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL); 4105 if (unlikely(!pg_vec)) 4106 goto out; 4107 4108 for (i = 0; i < block_nr; i++) { 4109 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4110 if (unlikely(!pg_vec[i].buffer)) 4111 goto out_free_pgvec; 4112 } 4113 4114 out: 4115 return pg_vec; 4116 4117 out_free_pgvec: 4118 free_pg_vec(pg_vec, order, block_nr); 4119 pg_vec = NULL; 4120 goto out; 4121 } 4122 4123 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4124 int closing, int tx_ring) 4125 { 4126 struct pgv *pg_vec = NULL; 4127 struct packet_sock *po = pkt_sk(sk); 4128 int was_running, order = 0; 4129 struct packet_ring_buffer *rb; 4130 struct sk_buff_head *rb_queue; 4131 __be16 num; 4132 int err = -EINVAL; 4133 /* Added to avoid minimal code churn */ 4134 struct tpacket_req *req = &req_u->req; 4135 4136 lock_sock(sk); 4137 4138 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4139 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4140 4141 err = -EBUSY; 4142 if (!closing) { 4143 if (atomic_read(&po->mapped)) 4144 goto out; 4145 if (packet_read_pending(rb)) 4146 goto out; 4147 } 4148 4149 if (req->tp_block_nr) { 4150 /* Sanity tests and some calculations */ 4151 err = -EBUSY; 4152 if (unlikely(rb->pg_vec)) 4153 goto out; 4154 4155 switch (po->tp_version) { 4156 case TPACKET_V1: 4157 po->tp_hdrlen = TPACKET_HDRLEN; 4158 break; 4159 case TPACKET_V2: 4160 po->tp_hdrlen = TPACKET2_HDRLEN; 4161 break; 4162 case TPACKET_V3: 4163 po->tp_hdrlen = TPACKET3_HDRLEN; 4164 break; 4165 } 4166 4167 err = -EINVAL; 4168 if (unlikely((int)req->tp_block_size <= 0)) 4169 goto out; 4170 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4171 goto out; 4172 if (po->tp_version >= TPACKET_V3 && 4173 (int)(req->tp_block_size - 4174 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0) 4175 goto out; 4176 if (unlikely(req->tp_frame_size < po->tp_hdrlen + 4177 po->tp_reserve)) 4178 goto out; 4179 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4180 goto out; 4181 4182 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4183 if (unlikely(rb->frames_per_block == 0)) 4184 goto out; 4185 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4186 req->tp_frame_nr)) 4187 goto out; 4188 4189 err = -ENOMEM; 4190 order = get_order(req->tp_block_size); 4191 pg_vec = alloc_pg_vec(req, order); 4192 if (unlikely(!pg_vec)) 4193 goto out; 4194 switch (po->tp_version) { 4195 case TPACKET_V3: 4196 /* Block transmit is not supported yet */ 4197 if (!tx_ring) { 4198 init_prb_bdqc(po, rb, pg_vec, req_u); 4199 } else { 4200 struct tpacket_req3 *req3 = &req_u->req3; 4201 4202 if (req3->tp_retire_blk_tov || 4203 req3->tp_sizeof_priv || 4204 req3->tp_feature_req_word) { 4205 err = -EINVAL; 4206 goto out; 4207 } 4208 } 4209 break; 4210 default: 4211 break; 4212 } 4213 } 4214 /* Done */ 4215 else { 4216 err = -EINVAL; 4217 if (unlikely(req->tp_frame_nr)) 4218 goto out; 4219 } 4220 4221 4222 /* Detach socket from network */ 4223 spin_lock(&po->bind_lock); 4224 was_running = po->running; 4225 num = po->num; 4226 if (was_running) { 4227 po->num = 0; 4228 __unregister_prot_hook(sk, false); 4229 } 4230 spin_unlock(&po->bind_lock); 4231 4232 synchronize_net(); 4233 4234 err = -EBUSY; 4235 mutex_lock(&po->pg_vec_lock); 4236 if (closing || atomic_read(&po->mapped) == 0) { 4237 err = 0; 4238 spin_lock_bh(&rb_queue->lock); 4239 swap(rb->pg_vec, pg_vec); 4240 rb->frame_max = (req->tp_frame_nr - 1); 4241 rb->head = 0; 4242 rb->frame_size = req->tp_frame_size; 4243 spin_unlock_bh(&rb_queue->lock); 4244 4245 swap(rb->pg_vec_order, order); 4246 swap(rb->pg_vec_len, req->tp_block_nr); 4247 4248 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4249 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4250 tpacket_rcv : packet_rcv; 4251 skb_queue_purge(rb_queue); 4252 if (atomic_read(&po->mapped)) 4253 pr_err("packet_mmap: vma is busy: %d\n", 4254 atomic_read(&po->mapped)); 4255 } 4256 mutex_unlock(&po->pg_vec_lock); 4257 4258 spin_lock(&po->bind_lock); 4259 if (was_running) { 4260 po->num = num; 4261 register_prot_hook(sk); 4262 } 4263 spin_unlock(&po->bind_lock); 4264 if (closing && (po->tp_version > TPACKET_V2)) { 4265 /* Because we don't support block-based V3 on tx-ring */ 4266 if (!tx_ring) 4267 prb_shutdown_retire_blk_timer(po, rb_queue); 4268 } 4269 4270 if (pg_vec) 4271 free_pg_vec(pg_vec, order, req->tp_block_nr); 4272 out: 4273 release_sock(sk); 4274 return err; 4275 } 4276 4277 static int packet_mmap(struct file *file, struct socket *sock, 4278 struct vm_area_struct *vma) 4279 { 4280 struct sock *sk = sock->sk; 4281 struct packet_sock *po = pkt_sk(sk); 4282 unsigned long size, expected_size; 4283 struct packet_ring_buffer *rb; 4284 unsigned long start; 4285 int err = -EINVAL; 4286 int i; 4287 4288 if (vma->vm_pgoff) 4289 return -EINVAL; 4290 4291 mutex_lock(&po->pg_vec_lock); 4292 4293 expected_size = 0; 4294 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4295 if (rb->pg_vec) { 4296 expected_size += rb->pg_vec_len 4297 * rb->pg_vec_pages 4298 * PAGE_SIZE; 4299 } 4300 } 4301 4302 if (expected_size == 0) 4303 goto out; 4304 4305 size = vma->vm_end - vma->vm_start; 4306 if (size != expected_size) 4307 goto out; 4308 4309 start = vma->vm_start; 4310 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4311 if (rb->pg_vec == NULL) 4312 continue; 4313 4314 for (i = 0; i < rb->pg_vec_len; i++) { 4315 struct page *page; 4316 void *kaddr = rb->pg_vec[i].buffer; 4317 int pg_num; 4318 4319 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4320 page = pgv_to_page(kaddr); 4321 err = vm_insert_page(vma, start, page); 4322 if (unlikely(err)) 4323 goto out; 4324 start += PAGE_SIZE; 4325 kaddr += PAGE_SIZE; 4326 } 4327 } 4328 } 4329 4330 atomic_inc(&po->mapped); 4331 vma->vm_ops = &packet_mmap_ops; 4332 err = 0; 4333 4334 out: 4335 mutex_unlock(&po->pg_vec_lock); 4336 return err; 4337 } 4338 4339 static const struct proto_ops packet_ops_spkt = { 4340 .family = PF_PACKET, 4341 .owner = THIS_MODULE, 4342 .release = packet_release, 4343 .bind = packet_bind_spkt, 4344 .connect = sock_no_connect, 4345 .socketpair = sock_no_socketpair, 4346 .accept = sock_no_accept, 4347 .getname = packet_getname_spkt, 4348 .poll = datagram_poll, 4349 .ioctl = packet_ioctl, 4350 .listen = sock_no_listen, 4351 .shutdown = sock_no_shutdown, 4352 .setsockopt = sock_no_setsockopt, 4353 .getsockopt = sock_no_getsockopt, 4354 .sendmsg = packet_sendmsg_spkt, 4355 .recvmsg = packet_recvmsg, 4356 .mmap = sock_no_mmap, 4357 .sendpage = sock_no_sendpage, 4358 }; 4359 4360 static const struct proto_ops packet_ops = { 4361 .family = PF_PACKET, 4362 .owner = THIS_MODULE, 4363 .release = packet_release, 4364 .bind = packet_bind, 4365 .connect = sock_no_connect, 4366 .socketpair = sock_no_socketpair, 4367 .accept = sock_no_accept, 4368 .getname = packet_getname, 4369 .poll = packet_poll, 4370 .ioctl = packet_ioctl, 4371 .listen = sock_no_listen, 4372 .shutdown = sock_no_shutdown, 4373 .setsockopt = packet_setsockopt, 4374 .getsockopt = packet_getsockopt, 4375 #ifdef CONFIG_COMPAT 4376 .compat_setsockopt = compat_packet_setsockopt, 4377 #endif 4378 .sendmsg = packet_sendmsg, 4379 .recvmsg = packet_recvmsg, 4380 .mmap = packet_mmap, 4381 .sendpage = sock_no_sendpage, 4382 }; 4383 4384 static const struct net_proto_family packet_family_ops = { 4385 .family = PF_PACKET, 4386 .create = packet_create, 4387 .owner = THIS_MODULE, 4388 }; 4389 4390 static struct notifier_block packet_netdev_notifier = { 4391 .notifier_call = packet_notifier, 4392 }; 4393 4394 #ifdef CONFIG_PROC_FS 4395 4396 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4397 __acquires(RCU) 4398 { 4399 struct net *net = seq_file_net(seq); 4400 4401 rcu_read_lock(); 4402 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4403 } 4404 4405 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4406 { 4407 struct net *net = seq_file_net(seq); 4408 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4409 } 4410 4411 static void packet_seq_stop(struct seq_file *seq, void *v) 4412 __releases(RCU) 4413 { 4414 rcu_read_unlock(); 4415 } 4416 4417 static int packet_seq_show(struct seq_file *seq, void *v) 4418 { 4419 if (v == SEQ_START_TOKEN) 4420 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4421 else { 4422 struct sock *s = sk_entry(v); 4423 const struct packet_sock *po = pkt_sk(s); 4424 4425 seq_printf(seq, 4426 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4427 s, 4428 atomic_read(&s->sk_refcnt), 4429 s->sk_type, 4430 ntohs(po->num), 4431 po->ifindex, 4432 po->running, 4433 atomic_read(&s->sk_rmem_alloc), 4434 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4435 sock_i_ino(s)); 4436 } 4437 4438 return 0; 4439 } 4440 4441 static const struct seq_operations packet_seq_ops = { 4442 .start = packet_seq_start, 4443 .next = packet_seq_next, 4444 .stop = packet_seq_stop, 4445 .show = packet_seq_show, 4446 }; 4447 4448 static int packet_seq_open(struct inode *inode, struct file *file) 4449 { 4450 return seq_open_net(inode, file, &packet_seq_ops, 4451 sizeof(struct seq_net_private)); 4452 } 4453 4454 static const struct file_operations packet_seq_fops = { 4455 .owner = THIS_MODULE, 4456 .open = packet_seq_open, 4457 .read = seq_read, 4458 .llseek = seq_lseek, 4459 .release = seq_release_net, 4460 }; 4461 4462 #endif 4463 4464 static int __net_init packet_net_init(struct net *net) 4465 { 4466 mutex_init(&net->packet.sklist_lock); 4467 INIT_HLIST_HEAD(&net->packet.sklist); 4468 4469 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops)) 4470 return -ENOMEM; 4471 4472 return 0; 4473 } 4474 4475 static void __net_exit packet_net_exit(struct net *net) 4476 { 4477 remove_proc_entry("packet", net->proc_net); 4478 } 4479 4480 static struct pernet_operations packet_net_ops = { 4481 .init = packet_net_init, 4482 .exit = packet_net_exit, 4483 }; 4484 4485 4486 static void __exit packet_exit(void) 4487 { 4488 unregister_netdevice_notifier(&packet_netdev_notifier); 4489 unregister_pernet_subsys(&packet_net_ops); 4490 sock_unregister(PF_PACKET); 4491 proto_unregister(&packet_proto); 4492 } 4493 4494 static int __init packet_init(void) 4495 { 4496 int rc = proto_register(&packet_proto, 0); 4497 4498 if (rc != 0) 4499 goto out; 4500 4501 sock_register(&packet_family_ops); 4502 register_pernet_subsys(&packet_net_ops); 4503 register_netdevice_notifier(&packet_netdev_notifier); 4504 out: 4505 return rc; 4506 } 4507 4508 module_init(packet_init); 4509 module_exit(packet_exit); 4510 MODULE_LICENSE("GPL"); 4511 MODULE_ALIAS_NETPROTO(PF_PACKET); 4512