1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * PACKET - implements raw packet sockets. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Alan Cox, <gw4pts@gw4pts.ampr.org> 11 * 12 * Fixes: 13 * Alan Cox : verify_area() now used correctly 14 * Alan Cox : new skbuff lists, look ma no backlogs! 15 * Alan Cox : tidied skbuff lists. 16 * Alan Cox : Now uses generic datagram routines I 17 * added. Also fixed the peek/read crash 18 * from all old Linux datagram code. 19 * Alan Cox : Uses the improved datagram code. 20 * Alan Cox : Added NULL's for socket options. 21 * Alan Cox : Re-commented the code. 22 * Alan Cox : Use new kernel side addressing 23 * Rob Janssen : Correct MTU usage. 24 * Dave Platt : Counter leaks caused by incorrect 25 * interrupt locking and some slightly 26 * dubious gcc output. Can you read 27 * compiler: it said _VOLATILE_ 28 * Richard Kooijman : Timestamp fixes. 29 * Alan Cox : New buffers. Use sk->mac.raw. 30 * Alan Cox : sendmsg/recvmsg support. 31 * Alan Cox : Protocol setting support 32 * Alexey Kuznetsov : Untied from IPv4 stack. 33 * Cyrus Durgin : Fixed kerneld for kmod. 34 * Michal Ostrowski : Module initialization cleanup. 35 * Ulises Alonso : Frame number limit removal and 36 * packet_set_ring memory leak. 37 * Eric Biederman : Allow for > 8 byte hardware addresses. 38 * The convention is that longer addresses 39 * will simply extend the hardware address 40 * byte arrays at the end of sockaddr_ll 41 * and packet_mreq. 42 * Johann Baudy : Added TX RING. 43 * Chetan Loke : Implemented TPACKET_V3 block abstraction 44 * layer. 45 * Copyright (C) 2011, <lokec@ccs.neu.edu> 46 * 47 * 48 * This program is free software; you can redistribute it and/or 49 * modify it under the terms of the GNU General Public License 50 * as published by the Free Software Foundation; either version 51 * 2 of the License, or (at your option) any later version. 52 * 53 */ 54 55 #include <linux/types.h> 56 #include <linux/mm.h> 57 #include <linux/capability.h> 58 #include <linux/fcntl.h> 59 #include <linux/socket.h> 60 #include <linux/in.h> 61 #include <linux/inet.h> 62 #include <linux/netdevice.h> 63 #include <linux/if_packet.h> 64 #include <linux/wireless.h> 65 #include <linux/kernel.h> 66 #include <linux/kmod.h> 67 #include <linux/slab.h> 68 #include <linux/vmalloc.h> 69 #include <net/net_namespace.h> 70 #include <net/ip.h> 71 #include <net/protocol.h> 72 #include <linux/skbuff.h> 73 #include <net/sock.h> 74 #include <linux/errno.h> 75 #include <linux/timer.h> 76 #include <asm/uaccess.h> 77 #include <asm/ioctls.h> 78 #include <asm/page.h> 79 #include <asm/cacheflush.h> 80 #include <asm/io.h> 81 #include <linux/proc_fs.h> 82 #include <linux/seq_file.h> 83 #include <linux/poll.h> 84 #include <linux/module.h> 85 #include <linux/init.h> 86 #include <linux/mutex.h> 87 #include <linux/if_vlan.h> 88 #include <linux/virtio_net.h> 89 #include <linux/errqueue.h> 90 #include <linux/net_tstamp.h> 91 #include <linux/percpu.h> 92 #ifdef CONFIG_INET 93 #include <net/inet_common.h> 94 #endif 95 #include <linux/bpf.h> 96 #include <net/compat.h> 97 98 #include "internal.h" 99 100 /* 101 Assumptions: 102 - if device has no dev->hard_header routine, it adds and removes ll header 103 inside itself. In this case ll header is invisible outside of device, 104 but higher levels still should reserve dev->hard_header_len. 105 Some devices are enough clever to reallocate skb, when header 106 will not fit to reserved space (tunnel), another ones are silly 107 (PPP). 108 - packet socket receives packets with pulled ll header, 109 so that SOCK_RAW should push it back. 110 111 On receive: 112 ----------- 113 114 Incoming, dev->hard_header!=NULL 115 mac_header -> ll header 116 data -> data 117 118 Outgoing, dev->hard_header!=NULL 119 mac_header -> ll header 120 data -> ll header 121 122 Incoming, dev->hard_header==NULL 123 mac_header -> UNKNOWN position. It is very likely, that it points to ll 124 header. PPP makes it, that is wrong, because introduce 125 assymetry between rx and tx paths. 126 data -> data 127 128 Outgoing, dev->hard_header==NULL 129 mac_header -> data. ll header is still not built! 130 data -> data 131 132 Resume 133 If dev->hard_header==NULL we are unlikely to restore sensible ll header. 134 135 136 On transmit: 137 ------------ 138 139 dev->hard_header != NULL 140 mac_header -> ll header 141 data -> ll header 142 143 dev->hard_header == NULL (ll header is added by device, we cannot control it) 144 mac_header -> data 145 data -> data 146 147 We should set nh.raw on output to correct posistion, 148 packet classifier depends on it. 149 */ 150 151 /* Private packet socket structures. */ 152 153 /* identical to struct packet_mreq except it has 154 * a longer address field. 155 */ 156 struct packet_mreq_max { 157 int mr_ifindex; 158 unsigned short mr_type; 159 unsigned short mr_alen; 160 unsigned char mr_address[MAX_ADDR_LEN]; 161 }; 162 163 union tpacket_uhdr { 164 struct tpacket_hdr *h1; 165 struct tpacket2_hdr *h2; 166 struct tpacket3_hdr *h3; 167 void *raw; 168 }; 169 170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 171 int closing, int tx_ring); 172 173 #define V3_ALIGNMENT (8) 174 175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) 176 177 #define BLK_PLUS_PRIV(sz_of_priv) \ 178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) 179 180 #define PGV_FROM_VMALLOC 1 181 182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) 183 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) 184 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) 185 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) 186 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) 187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) 188 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x))) 189 190 struct packet_sock; 191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg); 192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 193 struct packet_type *pt, struct net_device *orig_dev); 194 195 static void *packet_previous_frame(struct packet_sock *po, 196 struct packet_ring_buffer *rb, 197 int status); 198 static void packet_increment_head(struct packet_ring_buffer *buff); 199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *, 200 struct tpacket_block_desc *); 201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 202 struct packet_sock *); 203 static void prb_retire_current_block(struct tpacket_kbdq_core *, 204 struct packet_sock *, unsigned int status); 205 static int prb_queue_frozen(struct tpacket_kbdq_core *); 206 static void prb_open_block(struct tpacket_kbdq_core *, 207 struct tpacket_block_desc *); 208 static void prb_retire_rx_blk_timer_expired(unsigned long); 209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); 210 static void prb_init_blk_timer(struct packet_sock *, 211 struct tpacket_kbdq_core *, 212 void (*func) (unsigned long)); 213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 214 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 215 struct tpacket3_hdr *); 216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 217 struct tpacket3_hdr *); 218 static void packet_flush_mclist(struct sock *sk); 219 220 struct packet_skb_cb { 221 union { 222 struct sockaddr_pkt pkt; 223 union { 224 /* Trick: alias skb original length with 225 * ll.sll_family and ll.protocol in order 226 * to save room. 227 */ 228 unsigned int origlen; 229 struct sockaddr_ll ll; 230 }; 231 } sa; 232 }; 233 234 #define vio_le() virtio_legacy_is_little_endian() 235 236 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 237 238 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 239 #define GET_PBLOCK_DESC(x, bid) \ 240 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 242 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 243 #define GET_NEXT_PRB_BLK_NUM(x) \ 244 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 245 ((x)->kactive_blk_num+1) : 0) 246 247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 248 static void __fanout_link(struct sock *sk, struct packet_sock *po); 249 250 static int packet_direct_xmit(struct sk_buff *skb) 251 { 252 struct net_device *dev = skb->dev; 253 netdev_features_t features; 254 struct netdev_queue *txq; 255 int ret = NETDEV_TX_BUSY; 256 257 if (unlikely(!netif_running(dev) || 258 !netif_carrier_ok(dev))) 259 goto drop; 260 261 features = netif_skb_features(skb); 262 if (skb_needs_linearize(skb, features) && 263 __skb_linearize(skb)) 264 goto drop; 265 266 txq = skb_get_tx_queue(dev, skb); 267 268 local_bh_disable(); 269 270 HARD_TX_LOCK(dev, txq, smp_processor_id()); 271 if (!netif_xmit_frozen_or_drv_stopped(txq)) 272 ret = netdev_start_xmit(skb, dev, txq, false); 273 HARD_TX_UNLOCK(dev, txq); 274 275 local_bh_enable(); 276 277 if (!dev_xmit_complete(ret)) 278 kfree_skb(skb); 279 280 return ret; 281 drop: 282 atomic_long_inc(&dev->tx_dropped); 283 kfree_skb(skb); 284 return NET_XMIT_DROP; 285 } 286 287 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 288 { 289 struct net_device *dev; 290 291 rcu_read_lock(); 292 dev = rcu_dereference(po->cached_dev); 293 if (likely(dev)) 294 dev_hold(dev); 295 rcu_read_unlock(); 296 297 return dev; 298 } 299 300 static void packet_cached_dev_assign(struct packet_sock *po, 301 struct net_device *dev) 302 { 303 rcu_assign_pointer(po->cached_dev, dev); 304 } 305 306 static void packet_cached_dev_reset(struct packet_sock *po) 307 { 308 RCU_INIT_POINTER(po->cached_dev, NULL); 309 } 310 311 static bool packet_use_direct_xmit(const struct packet_sock *po) 312 { 313 return po->xmit == packet_direct_xmit; 314 } 315 316 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 317 { 318 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues; 319 } 320 321 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb) 322 { 323 const struct net_device_ops *ops = dev->netdev_ops; 324 u16 queue_index; 325 326 if (ops->ndo_select_queue) { 327 queue_index = ops->ndo_select_queue(dev, skb, NULL, 328 __packet_pick_tx_queue); 329 queue_index = netdev_cap_txqueue(dev, queue_index); 330 } else { 331 queue_index = __packet_pick_tx_queue(dev, skb); 332 } 333 334 skb_set_queue_mapping(skb, queue_index); 335 } 336 337 /* register_prot_hook must be invoked with the po->bind_lock held, 338 * or from a context in which asynchronous accesses to the packet 339 * socket is not possible (packet_create()). 340 */ 341 static void register_prot_hook(struct sock *sk) 342 { 343 struct packet_sock *po = pkt_sk(sk); 344 345 if (!po->running) { 346 if (po->fanout) 347 __fanout_link(sk, po); 348 else 349 dev_add_pack(&po->prot_hook); 350 351 sock_hold(sk); 352 po->running = 1; 353 } 354 } 355 356 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock 357 * held. If the sync parameter is true, we will temporarily drop 358 * the po->bind_lock and do a synchronize_net to make sure no 359 * asynchronous packet processing paths still refer to the elements 360 * of po->prot_hook. If the sync parameter is false, it is the 361 * callers responsibility to take care of this. 362 */ 363 static void __unregister_prot_hook(struct sock *sk, bool sync) 364 { 365 struct packet_sock *po = pkt_sk(sk); 366 367 po->running = 0; 368 369 if (po->fanout) 370 __fanout_unlink(sk, po); 371 else 372 __dev_remove_pack(&po->prot_hook); 373 374 __sock_put(sk); 375 376 if (sync) { 377 spin_unlock(&po->bind_lock); 378 synchronize_net(); 379 spin_lock(&po->bind_lock); 380 } 381 } 382 383 static void unregister_prot_hook(struct sock *sk, bool sync) 384 { 385 struct packet_sock *po = pkt_sk(sk); 386 387 if (po->running) 388 __unregister_prot_hook(sk, sync); 389 } 390 391 static inline struct page * __pure pgv_to_page(void *addr) 392 { 393 if (is_vmalloc_addr(addr)) 394 return vmalloc_to_page(addr); 395 return virt_to_page(addr); 396 } 397 398 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 399 { 400 union tpacket_uhdr h; 401 402 h.raw = frame; 403 switch (po->tp_version) { 404 case TPACKET_V1: 405 h.h1->tp_status = status; 406 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 407 break; 408 case TPACKET_V2: 409 h.h2->tp_status = status; 410 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 411 break; 412 case TPACKET_V3: 413 default: 414 WARN(1, "TPACKET version not supported.\n"); 415 BUG(); 416 } 417 418 smp_wmb(); 419 } 420 421 static int __packet_get_status(struct packet_sock *po, void *frame) 422 { 423 union tpacket_uhdr h; 424 425 smp_rmb(); 426 427 h.raw = frame; 428 switch (po->tp_version) { 429 case TPACKET_V1: 430 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 431 return h.h1->tp_status; 432 case TPACKET_V2: 433 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 434 return h.h2->tp_status; 435 case TPACKET_V3: 436 default: 437 WARN(1, "TPACKET version not supported.\n"); 438 BUG(); 439 return 0; 440 } 441 } 442 443 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, 444 unsigned int flags) 445 { 446 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 447 448 if (shhwtstamps && 449 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 450 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) 451 return TP_STATUS_TS_RAW_HARDWARE; 452 453 if (ktime_to_timespec_cond(skb->tstamp, ts)) 454 return TP_STATUS_TS_SOFTWARE; 455 456 return 0; 457 } 458 459 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 460 struct sk_buff *skb) 461 { 462 union tpacket_uhdr h; 463 struct timespec ts; 464 __u32 ts_status; 465 466 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 467 return 0; 468 469 h.raw = frame; 470 switch (po->tp_version) { 471 case TPACKET_V1: 472 h.h1->tp_sec = ts.tv_sec; 473 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 474 break; 475 case TPACKET_V2: 476 h.h2->tp_sec = ts.tv_sec; 477 h.h2->tp_nsec = ts.tv_nsec; 478 break; 479 case TPACKET_V3: 480 default: 481 WARN(1, "TPACKET version not supported.\n"); 482 BUG(); 483 } 484 485 /* one flush is safe, as both fields always lie on the same cacheline */ 486 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 487 smp_wmb(); 488 489 return ts_status; 490 } 491 492 static void *packet_lookup_frame(struct packet_sock *po, 493 struct packet_ring_buffer *rb, 494 unsigned int position, 495 int status) 496 { 497 unsigned int pg_vec_pos, frame_offset; 498 union tpacket_uhdr h; 499 500 pg_vec_pos = position / rb->frames_per_block; 501 frame_offset = position % rb->frames_per_block; 502 503 h.raw = rb->pg_vec[pg_vec_pos].buffer + 504 (frame_offset * rb->frame_size); 505 506 if (status != __packet_get_status(po, h.raw)) 507 return NULL; 508 509 return h.raw; 510 } 511 512 static void *packet_current_frame(struct packet_sock *po, 513 struct packet_ring_buffer *rb, 514 int status) 515 { 516 return packet_lookup_frame(po, rb, rb->head, status); 517 } 518 519 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 520 { 521 del_timer_sync(&pkc->retire_blk_timer); 522 } 523 524 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 525 struct sk_buff_head *rb_queue) 526 { 527 struct tpacket_kbdq_core *pkc; 528 529 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 530 531 spin_lock_bh(&rb_queue->lock); 532 pkc->delete_blk_timer = 1; 533 spin_unlock_bh(&rb_queue->lock); 534 535 prb_del_retire_blk_timer(pkc); 536 } 537 538 static void prb_init_blk_timer(struct packet_sock *po, 539 struct tpacket_kbdq_core *pkc, 540 void (*func) (unsigned long)) 541 { 542 init_timer(&pkc->retire_blk_timer); 543 pkc->retire_blk_timer.data = (long)po; 544 pkc->retire_blk_timer.function = func; 545 pkc->retire_blk_timer.expires = jiffies; 546 } 547 548 static void prb_setup_retire_blk_timer(struct packet_sock *po) 549 { 550 struct tpacket_kbdq_core *pkc; 551 552 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 553 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired); 554 } 555 556 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 557 int blk_size_in_bytes) 558 { 559 struct net_device *dev; 560 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; 561 struct ethtool_link_ksettings ecmd; 562 int err; 563 564 rtnl_lock(); 565 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 566 if (unlikely(!dev)) { 567 rtnl_unlock(); 568 return DEFAULT_PRB_RETIRE_TOV; 569 } 570 err = __ethtool_get_link_ksettings(dev, &ecmd); 571 rtnl_unlock(); 572 if (!err) { 573 /* 574 * If the link speed is so slow you don't really 575 * need to worry about perf anyways 576 */ 577 if (ecmd.base.speed < SPEED_1000 || 578 ecmd.base.speed == SPEED_UNKNOWN) { 579 return DEFAULT_PRB_RETIRE_TOV; 580 } else { 581 msec = 1; 582 div = ecmd.base.speed / 1000; 583 } 584 } 585 586 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 587 588 if (div) 589 mbits /= div; 590 591 tmo = mbits * msec; 592 593 if (div) 594 return tmo+1; 595 return tmo; 596 } 597 598 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 599 union tpacket_req_u *req_u) 600 { 601 p1->feature_req_word = req_u->req3.tp_feature_req_word; 602 } 603 604 static void init_prb_bdqc(struct packet_sock *po, 605 struct packet_ring_buffer *rb, 606 struct pgv *pg_vec, 607 union tpacket_req_u *req_u) 608 { 609 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 610 struct tpacket_block_desc *pbd; 611 612 memset(p1, 0x0, sizeof(*p1)); 613 614 p1->knxt_seq_num = 1; 615 p1->pkbdq = pg_vec; 616 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 617 p1->pkblk_start = pg_vec[0].buffer; 618 p1->kblk_size = req_u->req3.tp_block_size; 619 p1->knum_blocks = req_u->req3.tp_block_nr; 620 p1->hdrlen = po->tp_hdrlen; 621 p1->version = po->tp_version; 622 p1->last_kactive_blk_num = 0; 623 po->stats.stats3.tp_freeze_q_cnt = 0; 624 if (req_u->req3.tp_retire_blk_tov) 625 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 626 else 627 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 628 req_u->req3.tp_block_size); 629 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 630 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 631 632 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 633 prb_init_ft_ops(p1, req_u); 634 prb_setup_retire_blk_timer(po); 635 prb_open_block(p1, pbd); 636 } 637 638 /* Do NOT update the last_blk_num first. 639 * Assumes sk_buff_head lock is held. 640 */ 641 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 642 { 643 mod_timer(&pkc->retire_blk_timer, 644 jiffies + pkc->tov_in_jiffies); 645 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 646 } 647 648 /* 649 * Timer logic: 650 * 1) We refresh the timer only when we open a block. 651 * By doing this we don't waste cycles refreshing the timer 652 * on packet-by-packet basis. 653 * 654 * With a 1MB block-size, on a 1Gbps line, it will take 655 * i) ~8 ms to fill a block + ii) memcpy etc. 656 * In this cut we are not accounting for the memcpy time. 657 * 658 * So, if the user sets the 'tmo' to 10ms then the timer 659 * will never fire while the block is still getting filled 660 * (which is what we want). However, the user could choose 661 * to close a block early and that's fine. 662 * 663 * But when the timer does fire, we check whether or not to refresh it. 664 * Since the tmo granularity is in msecs, it is not too expensive 665 * to refresh the timer, lets say every '8' msecs. 666 * Either the user can set the 'tmo' or we can derive it based on 667 * a) line-speed and b) block-size. 668 * prb_calc_retire_blk_tmo() calculates the tmo. 669 * 670 */ 671 static void prb_retire_rx_blk_timer_expired(unsigned long data) 672 { 673 struct packet_sock *po = (struct packet_sock *)data; 674 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 675 unsigned int frozen; 676 struct tpacket_block_desc *pbd; 677 678 spin_lock(&po->sk.sk_receive_queue.lock); 679 680 frozen = prb_queue_frozen(pkc); 681 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 682 683 if (unlikely(pkc->delete_blk_timer)) 684 goto out; 685 686 /* We only need to plug the race when the block is partially filled. 687 * tpacket_rcv: 688 * lock(); increment BLOCK_NUM_PKTS; unlock() 689 * copy_bits() is in progress ... 690 * timer fires on other cpu: 691 * we can't retire the current block because copy_bits 692 * is in progress. 693 * 694 */ 695 if (BLOCK_NUM_PKTS(pbd)) { 696 while (atomic_read(&pkc->blk_fill_in_prog)) { 697 /* Waiting for skb_copy_bits to finish... */ 698 cpu_relax(); 699 } 700 } 701 702 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 703 if (!frozen) { 704 if (!BLOCK_NUM_PKTS(pbd)) { 705 /* An empty block. Just refresh the timer. */ 706 goto refresh_timer; 707 } 708 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 709 if (!prb_dispatch_next_block(pkc, po)) 710 goto refresh_timer; 711 else 712 goto out; 713 } else { 714 /* Case 1. Queue was frozen because user-space was 715 * lagging behind. 716 */ 717 if (prb_curr_blk_in_use(pkc, pbd)) { 718 /* 719 * Ok, user-space is still behind. 720 * So just refresh the timer. 721 */ 722 goto refresh_timer; 723 } else { 724 /* Case 2. queue was frozen,user-space caught up, 725 * now the link went idle && the timer fired. 726 * We don't have a block to close.So we open this 727 * block and restart the timer. 728 * opening a block thaws the queue,restarts timer 729 * Thawing/timer-refresh is a side effect. 730 */ 731 prb_open_block(pkc, pbd); 732 goto out; 733 } 734 } 735 } 736 737 refresh_timer: 738 _prb_refresh_rx_retire_blk_timer(pkc); 739 740 out: 741 spin_unlock(&po->sk.sk_receive_queue.lock); 742 } 743 744 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 745 struct tpacket_block_desc *pbd1, __u32 status) 746 { 747 /* Flush everything minus the block header */ 748 749 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 750 u8 *start, *end; 751 752 start = (u8 *)pbd1; 753 754 /* Skip the block header(we know header WILL fit in 4K) */ 755 start += PAGE_SIZE; 756 757 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 758 for (; start < end; start += PAGE_SIZE) 759 flush_dcache_page(pgv_to_page(start)); 760 761 smp_wmb(); 762 #endif 763 764 /* Now update the block status. */ 765 766 BLOCK_STATUS(pbd1) = status; 767 768 /* Flush the block header */ 769 770 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 771 start = (u8 *)pbd1; 772 flush_dcache_page(pgv_to_page(start)); 773 774 smp_wmb(); 775 #endif 776 } 777 778 /* 779 * Side effect: 780 * 781 * 1) flush the block 782 * 2) Increment active_blk_num 783 * 784 * Note:We DONT refresh the timer on purpose. 785 * Because almost always the next block will be opened. 786 */ 787 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 788 struct tpacket_block_desc *pbd1, 789 struct packet_sock *po, unsigned int stat) 790 { 791 __u32 status = TP_STATUS_USER | stat; 792 793 struct tpacket3_hdr *last_pkt; 794 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 795 struct sock *sk = &po->sk; 796 797 if (po->stats.stats3.tp_drops) 798 status |= TP_STATUS_LOSING; 799 800 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 801 last_pkt->tp_next_offset = 0; 802 803 /* Get the ts of the last pkt */ 804 if (BLOCK_NUM_PKTS(pbd1)) { 805 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 806 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 807 } else { 808 /* Ok, we tmo'd - so get the current time. 809 * 810 * It shouldn't really happen as we don't close empty 811 * blocks. See prb_retire_rx_blk_timer_expired(). 812 */ 813 struct timespec ts; 814 getnstimeofday(&ts); 815 h1->ts_last_pkt.ts_sec = ts.tv_sec; 816 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 817 } 818 819 smp_wmb(); 820 821 /* Flush the block */ 822 prb_flush_block(pkc1, pbd1, status); 823 824 sk->sk_data_ready(sk); 825 826 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 827 } 828 829 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 830 { 831 pkc->reset_pending_on_curr_blk = 0; 832 } 833 834 /* 835 * Side effect of opening a block: 836 * 837 * 1) prb_queue is thawed. 838 * 2) retire_blk_timer is refreshed. 839 * 840 */ 841 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 842 struct tpacket_block_desc *pbd1) 843 { 844 struct timespec ts; 845 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 846 847 smp_rmb(); 848 849 /* We could have just memset this but we will lose the 850 * flexibility of making the priv area sticky 851 */ 852 853 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 854 BLOCK_NUM_PKTS(pbd1) = 0; 855 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 856 857 getnstimeofday(&ts); 858 859 h1->ts_first_pkt.ts_sec = ts.tv_sec; 860 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 861 862 pkc1->pkblk_start = (char *)pbd1; 863 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 864 865 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 866 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 867 868 pbd1->version = pkc1->version; 869 pkc1->prev = pkc1->nxt_offset; 870 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 871 872 prb_thaw_queue(pkc1); 873 _prb_refresh_rx_retire_blk_timer(pkc1); 874 875 smp_wmb(); 876 } 877 878 /* 879 * Queue freeze logic: 880 * 1) Assume tp_block_nr = 8 blocks. 881 * 2) At time 't0', user opens Rx ring. 882 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 883 * 4) user-space is either sleeping or processing block '0'. 884 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 885 * it will close block-7,loop around and try to fill block '0'. 886 * call-flow: 887 * __packet_lookup_frame_in_block 888 * prb_retire_current_block() 889 * prb_dispatch_next_block() 890 * |->(BLOCK_STATUS == USER) evaluates to true 891 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 892 * 6) Now there are two cases: 893 * 6.1) Link goes idle right after the queue is frozen. 894 * But remember, the last open_block() refreshed the timer. 895 * When this timer expires,it will refresh itself so that we can 896 * re-open block-0 in near future. 897 * 6.2) Link is busy and keeps on receiving packets. This is a simple 898 * case and __packet_lookup_frame_in_block will check if block-0 899 * is free and can now be re-used. 900 */ 901 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 902 struct packet_sock *po) 903 { 904 pkc->reset_pending_on_curr_blk = 1; 905 po->stats.stats3.tp_freeze_q_cnt++; 906 } 907 908 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 909 910 /* 911 * If the next block is free then we will dispatch it 912 * and return a good offset. 913 * Else, we will freeze the queue. 914 * So, caller must check the return value. 915 */ 916 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 917 struct packet_sock *po) 918 { 919 struct tpacket_block_desc *pbd; 920 921 smp_rmb(); 922 923 /* 1. Get current block num */ 924 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 925 926 /* 2. If this block is currently in_use then freeze the queue */ 927 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 928 prb_freeze_queue(pkc, po); 929 return NULL; 930 } 931 932 /* 933 * 3. 934 * open this block and return the offset where the first packet 935 * needs to get stored. 936 */ 937 prb_open_block(pkc, pbd); 938 return (void *)pkc->nxt_offset; 939 } 940 941 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 942 struct packet_sock *po, unsigned int status) 943 { 944 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 945 946 /* retire/close the current block */ 947 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 948 /* 949 * Plug the case where copy_bits() is in progress on 950 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 951 * have space to copy the pkt in the current block and 952 * called prb_retire_current_block() 953 * 954 * We don't need to worry about the TMO case because 955 * the timer-handler already handled this case. 956 */ 957 if (!(status & TP_STATUS_BLK_TMO)) { 958 while (atomic_read(&pkc->blk_fill_in_prog)) { 959 /* Waiting for skb_copy_bits to finish... */ 960 cpu_relax(); 961 } 962 } 963 prb_close_block(pkc, pbd, po, status); 964 return; 965 } 966 } 967 968 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc, 969 struct tpacket_block_desc *pbd) 970 { 971 return TP_STATUS_USER & BLOCK_STATUS(pbd); 972 } 973 974 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 975 { 976 return pkc->reset_pending_on_curr_blk; 977 } 978 979 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 980 { 981 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 982 atomic_dec(&pkc->blk_fill_in_prog); 983 } 984 985 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 986 struct tpacket3_hdr *ppd) 987 { 988 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 989 } 990 991 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 992 struct tpacket3_hdr *ppd) 993 { 994 ppd->hv1.tp_rxhash = 0; 995 } 996 997 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 998 struct tpacket3_hdr *ppd) 999 { 1000 if (skb_vlan_tag_present(pkc->skb)) { 1001 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 1002 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 1003 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 1004 } else { 1005 ppd->hv1.tp_vlan_tci = 0; 1006 ppd->hv1.tp_vlan_tpid = 0; 1007 ppd->tp_status = TP_STATUS_AVAILABLE; 1008 } 1009 } 1010 1011 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 1012 struct tpacket3_hdr *ppd) 1013 { 1014 ppd->hv1.tp_padding = 0; 1015 prb_fill_vlan_info(pkc, ppd); 1016 1017 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 1018 prb_fill_rxhash(pkc, ppd); 1019 else 1020 prb_clear_rxhash(pkc, ppd); 1021 } 1022 1023 static void prb_fill_curr_block(char *curr, 1024 struct tpacket_kbdq_core *pkc, 1025 struct tpacket_block_desc *pbd, 1026 unsigned int len) 1027 { 1028 struct tpacket3_hdr *ppd; 1029 1030 ppd = (struct tpacket3_hdr *)curr; 1031 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 1032 pkc->prev = curr; 1033 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 1034 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1035 BLOCK_NUM_PKTS(pbd) += 1; 1036 atomic_inc(&pkc->blk_fill_in_prog); 1037 prb_run_all_ft_ops(pkc, ppd); 1038 } 1039 1040 /* Assumes caller has the sk->rx_queue.lock */ 1041 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1042 struct sk_buff *skb, 1043 int status, 1044 unsigned int len 1045 ) 1046 { 1047 struct tpacket_kbdq_core *pkc; 1048 struct tpacket_block_desc *pbd; 1049 char *curr, *end; 1050 1051 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1052 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1053 1054 /* Queue is frozen when user space is lagging behind */ 1055 if (prb_queue_frozen(pkc)) { 1056 /* 1057 * Check if that last block which caused the queue to freeze, 1058 * is still in_use by user-space. 1059 */ 1060 if (prb_curr_blk_in_use(pkc, pbd)) { 1061 /* Can't record this packet */ 1062 return NULL; 1063 } else { 1064 /* 1065 * Ok, the block was released by user-space. 1066 * Now let's open that block. 1067 * opening a block also thaws the queue. 1068 * Thawing is a side effect. 1069 */ 1070 prb_open_block(pkc, pbd); 1071 } 1072 } 1073 1074 smp_mb(); 1075 curr = pkc->nxt_offset; 1076 pkc->skb = skb; 1077 end = (char *)pbd + pkc->kblk_size; 1078 1079 /* first try the current block */ 1080 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1081 prb_fill_curr_block(curr, pkc, pbd, len); 1082 return (void *)curr; 1083 } 1084 1085 /* Ok, close the current block */ 1086 prb_retire_current_block(pkc, po, 0); 1087 1088 /* Now, try to dispatch the next block */ 1089 curr = (char *)prb_dispatch_next_block(pkc, po); 1090 if (curr) { 1091 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1092 prb_fill_curr_block(curr, pkc, pbd, len); 1093 return (void *)curr; 1094 } 1095 1096 /* 1097 * No free blocks are available.user_space hasn't caught up yet. 1098 * Queue was just frozen and now this packet will get dropped. 1099 */ 1100 return NULL; 1101 } 1102 1103 static void *packet_current_rx_frame(struct packet_sock *po, 1104 struct sk_buff *skb, 1105 int status, unsigned int len) 1106 { 1107 char *curr = NULL; 1108 switch (po->tp_version) { 1109 case TPACKET_V1: 1110 case TPACKET_V2: 1111 curr = packet_lookup_frame(po, &po->rx_ring, 1112 po->rx_ring.head, status); 1113 return curr; 1114 case TPACKET_V3: 1115 return __packet_lookup_frame_in_block(po, skb, status, len); 1116 default: 1117 WARN(1, "TPACKET version not supported\n"); 1118 BUG(); 1119 return NULL; 1120 } 1121 } 1122 1123 static void *prb_lookup_block(struct packet_sock *po, 1124 struct packet_ring_buffer *rb, 1125 unsigned int idx, 1126 int status) 1127 { 1128 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1129 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1130 1131 if (status != BLOCK_STATUS(pbd)) 1132 return NULL; 1133 return pbd; 1134 } 1135 1136 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1137 { 1138 unsigned int prev; 1139 if (rb->prb_bdqc.kactive_blk_num) 1140 prev = rb->prb_bdqc.kactive_blk_num-1; 1141 else 1142 prev = rb->prb_bdqc.knum_blocks-1; 1143 return prev; 1144 } 1145 1146 /* Assumes caller has held the rx_queue.lock */ 1147 static void *__prb_previous_block(struct packet_sock *po, 1148 struct packet_ring_buffer *rb, 1149 int status) 1150 { 1151 unsigned int previous = prb_previous_blk_num(rb); 1152 return prb_lookup_block(po, rb, previous, status); 1153 } 1154 1155 static void *packet_previous_rx_frame(struct packet_sock *po, 1156 struct packet_ring_buffer *rb, 1157 int status) 1158 { 1159 if (po->tp_version <= TPACKET_V2) 1160 return packet_previous_frame(po, rb, status); 1161 1162 return __prb_previous_block(po, rb, status); 1163 } 1164 1165 static void packet_increment_rx_head(struct packet_sock *po, 1166 struct packet_ring_buffer *rb) 1167 { 1168 switch (po->tp_version) { 1169 case TPACKET_V1: 1170 case TPACKET_V2: 1171 return packet_increment_head(rb); 1172 case TPACKET_V3: 1173 default: 1174 WARN(1, "TPACKET version not supported.\n"); 1175 BUG(); 1176 return; 1177 } 1178 } 1179 1180 static void *packet_previous_frame(struct packet_sock *po, 1181 struct packet_ring_buffer *rb, 1182 int status) 1183 { 1184 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1185 return packet_lookup_frame(po, rb, previous, status); 1186 } 1187 1188 static void packet_increment_head(struct packet_ring_buffer *buff) 1189 { 1190 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1191 } 1192 1193 static void packet_inc_pending(struct packet_ring_buffer *rb) 1194 { 1195 this_cpu_inc(*rb->pending_refcnt); 1196 } 1197 1198 static void packet_dec_pending(struct packet_ring_buffer *rb) 1199 { 1200 this_cpu_dec(*rb->pending_refcnt); 1201 } 1202 1203 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1204 { 1205 unsigned int refcnt = 0; 1206 int cpu; 1207 1208 /* We don't use pending refcount in rx_ring. */ 1209 if (rb->pending_refcnt == NULL) 1210 return 0; 1211 1212 for_each_possible_cpu(cpu) 1213 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1214 1215 return refcnt; 1216 } 1217 1218 static int packet_alloc_pending(struct packet_sock *po) 1219 { 1220 po->rx_ring.pending_refcnt = NULL; 1221 1222 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1223 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1224 return -ENOBUFS; 1225 1226 return 0; 1227 } 1228 1229 static void packet_free_pending(struct packet_sock *po) 1230 { 1231 free_percpu(po->tx_ring.pending_refcnt); 1232 } 1233 1234 #define ROOM_POW_OFF 2 1235 #define ROOM_NONE 0x0 1236 #define ROOM_LOW 0x1 1237 #define ROOM_NORMAL 0x2 1238 1239 static bool __tpacket_has_room(struct packet_sock *po, int pow_off) 1240 { 1241 int idx, len; 1242 1243 len = po->rx_ring.frame_max + 1; 1244 idx = po->rx_ring.head; 1245 if (pow_off) 1246 idx += len >> pow_off; 1247 if (idx >= len) 1248 idx -= len; 1249 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1250 } 1251 1252 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off) 1253 { 1254 int idx, len; 1255 1256 len = po->rx_ring.prb_bdqc.knum_blocks; 1257 idx = po->rx_ring.prb_bdqc.kactive_blk_num; 1258 if (pow_off) 1259 idx += len >> pow_off; 1260 if (idx >= len) 1261 idx -= len; 1262 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1263 } 1264 1265 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1266 { 1267 struct sock *sk = &po->sk; 1268 int ret = ROOM_NONE; 1269 1270 if (po->prot_hook.func != tpacket_rcv) { 1271 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1272 - (skb ? skb->truesize : 0); 1273 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF)) 1274 return ROOM_NORMAL; 1275 else if (avail > 0) 1276 return ROOM_LOW; 1277 else 1278 return ROOM_NONE; 1279 } 1280 1281 if (po->tp_version == TPACKET_V3) { 1282 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1283 ret = ROOM_NORMAL; 1284 else if (__tpacket_v3_has_room(po, 0)) 1285 ret = ROOM_LOW; 1286 } else { 1287 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1288 ret = ROOM_NORMAL; 1289 else if (__tpacket_has_room(po, 0)) 1290 ret = ROOM_LOW; 1291 } 1292 1293 return ret; 1294 } 1295 1296 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1297 { 1298 int ret; 1299 bool has_room; 1300 1301 spin_lock_bh(&po->sk.sk_receive_queue.lock); 1302 ret = __packet_rcv_has_room(po, skb); 1303 has_room = ret == ROOM_NORMAL; 1304 if (po->pressure == has_room) 1305 po->pressure = !has_room; 1306 spin_unlock_bh(&po->sk.sk_receive_queue.lock); 1307 1308 return ret; 1309 } 1310 1311 static void packet_sock_destruct(struct sock *sk) 1312 { 1313 skb_queue_purge(&sk->sk_error_queue); 1314 1315 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1316 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 1317 1318 if (!sock_flag(sk, SOCK_DEAD)) { 1319 pr_err("Attempt to release alive packet socket: %p\n", sk); 1320 return; 1321 } 1322 1323 sk_refcnt_debug_dec(sk); 1324 } 1325 1326 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1327 { 1328 u32 rxhash; 1329 int i, count = 0; 1330 1331 rxhash = skb_get_hash(skb); 1332 for (i = 0; i < ROLLOVER_HLEN; i++) 1333 if (po->rollover->history[i] == rxhash) 1334 count++; 1335 1336 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash; 1337 return count > (ROLLOVER_HLEN >> 1); 1338 } 1339 1340 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1341 struct sk_buff *skb, 1342 unsigned int num) 1343 { 1344 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1345 } 1346 1347 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1348 struct sk_buff *skb, 1349 unsigned int num) 1350 { 1351 unsigned int val = atomic_inc_return(&f->rr_cur); 1352 1353 return val % num; 1354 } 1355 1356 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1357 struct sk_buff *skb, 1358 unsigned int num) 1359 { 1360 return smp_processor_id() % num; 1361 } 1362 1363 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1364 struct sk_buff *skb, 1365 unsigned int num) 1366 { 1367 return prandom_u32_max(num); 1368 } 1369 1370 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1371 struct sk_buff *skb, 1372 unsigned int idx, bool try_self, 1373 unsigned int num) 1374 { 1375 struct packet_sock *po, *po_next, *po_skip = NULL; 1376 unsigned int i, j, room = ROOM_NONE; 1377 1378 po = pkt_sk(f->arr[idx]); 1379 1380 if (try_self) { 1381 room = packet_rcv_has_room(po, skb); 1382 if (room == ROOM_NORMAL || 1383 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1384 return idx; 1385 po_skip = po; 1386 } 1387 1388 i = j = min_t(int, po->rollover->sock, num - 1); 1389 do { 1390 po_next = pkt_sk(f->arr[i]); 1391 if (po_next != po_skip && !po_next->pressure && 1392 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1393 if (i != j) 1394 po->rollover->sock = i; 1395 atomic_long_inc(&po->rollover->num); 1396 if (room == ROOM_LOW) 1397 atomic_long_inc(&po->rollover->num_huge); 1398 return i; 1399 } 1400 1401 if (++i == num) 1402 i = 0; 1403 } while (i != j); 1404 1405 atomic_long_inc(&po->rollover->num_failed); 1406 return idx; 1407 } 1408 1409 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1410 struct sk_buff *skb, 1411 unsigned int num) 1412 { 1413 return skb_get_queue_mapping(skb) % num; 1414 } 1415 1416 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1417 struct sk_buff *skb, 1418 unsigned int num) 1419 { 1420 struct bpf_prog *prog; 1421 unsigned int ret = 0; 1422 1423 rcu_read_lock(); 1424 prog = rcu_dereference(f->bpf_prog); 1425 if (prog) 1426 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1427 rcu_read_unlock(); 1428 1429 return ret; 1430 } 1431 1432 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1433 { 1434 return f->flags & (flag >> 8); 1435 } 1436 1437 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1438 struct packet_type *pt, struct net_device *orig_dev) 1439 { 1440 struct packet_fanout *f = pt->af_packet_priv; 1441 unsigned int num = READ_ONCE(f->num_members); 1442 struct net *net = read_pnet(&f->net); 1443 struct packet_sock *po; 1444 unsigned int idx; 1445 1446 if (!net_eq(dev_net(dev), net) || !num) { 1447 kfree_skb(skb); 1448 return 0; 1449 } 1450 1451 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1452 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1453 if (!skb) 1454 return 0; 1455 } 1456 switch (f->type) { 1457 case PACKET_FANOUT_HASH: 1458 default: 1459 idx = fanout_demux_hash(f, skb, num); 1460 break; 1461 case PACKET_FANOUT_LB: 1462 idx = fanout_demux_lb(f, skb, num); 1463 break; 1464 case PACKET_FANOUT_CPU: 1465 idx = fanout_demux_cpu(f, skb, num); 1466 break; 1467 case PACKET_FANOUT_RND: 1468 idx = fanout_demux_rnd(f, skb, num); 1469 break; 1470 case PACKET_FANOUT_QM: 1471 idx = fanout_demux_qm(f, skb, num); 1472 break; 1473 case PACKET_FANOUT_ROLLOVER: 1474 idx = fanout_demux_rollover(f, skb, 0, false, num); 1475 break; 1476 case PACKET_FANOUT_CBPF: 1477 case PACKET_FANOUT_EBPF: 1478 idx = fanout_demux_bpf(f, skb, num); 1479 break; 1480 } 1481 1482 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1483 idx = fanout_demux_rollover(f, skb, idx, true, num); 1484 1485 po = pkt_sk(f->arr[idx]); 1486 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1487 } 1488 1489 DEFINE_MUTEX(fanout_mutex); 1490 EXPORT_SYMBOL_GPL(fanout_mutex); 1491 static LIST_HEAD(fanout_list); 1492 1493 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1494 { 1495 struct packet_fanout *f = po->fanout; 1496 1497 spin_lock(&f->lock); 1498 f->arr[f->num_members] = sk; 1499 smp_wmb(); 1500 f->num_members++; 1501 spin_unlock(&f->lock); 1502 } 1503 1504 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1505 { 1506 struct packet_fanout *f = po->fanout; 1507 int i; 1508 1509 spin_lock(&f->lock); 1510 for (i = 0; i < f->num_members; i++) { 1511 if (f->arr[i] == sk) 1512 break; 1513 } 1514 BUG_ON(i >= f->num_members); 1515 f->arr[i] = f->arr[f->num_members - 1]; 1516 f->num_members--; 1517 spin_unlock(&f->lock); 1518 } 1519 1520 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1521 { 1522 if (sk->sk_family != PF_PACKET) 1523 return false; 1524 1525 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1526 } 1527 1528 static void fanout_init_data(struct packet_fanout *f) 1529 { 1530 switch (f->type) { 1531 case PACKET_FANOUT_LB: 1532 atomic_set(&f->rr_cur, 0); 1533 break; 1534 case PACKET_FANOUT_CBPF: 1535 case PACKET_FANOUT_EBPF: 1536 RCU_INIT_POINTER(f->bpf_prog, NULL); 1537 break; 1538 } 1539 } 1540 1541 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1542 { 1543 struct bpf_prog *old; 1544 1545 spin_lock(&f->lock); 1546 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1547 rcu_assign_pointer(f->bpf_prog, new); 1548 spin_unlock(&f->lock); 1549 1550 if (old) { 1551 synchronize_net(); 1552 bpf_prog_destroy(old); 1553 } 1554 } 1555 1556 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data, 1557 unsigned int len) 1558 { 1559 struct bpf_prog *new; 1560 struct sock_fprog fprog; 1561 int ret; 1562 1563 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1564 return -EPERM; 1565 if (len != sizeof(fprog)) 1566 return -EINVAL; 1567 if (copy_from_user(&fprog, data, len)) 1568 return -EFAULT; 1569 1570 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1571 if (ret) 1572 return ret; 1573 1574 __fanout_set_data_bpf(po->fanout, new); 1575 return 0; 1576 } 1577 1578 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data, 1579 unsigned int len) 1580 { 1581 struct bpf_prog *new; 1582 u32 fd; 1583 1584 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1585 return -EPERM; 1586 if (len != sizeof(fd)) 1587 return -EINVAL; 1588 if (copy_from_user(&fd, data, len)) 1589 return -EFAULT; 1590 1591 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1592 if (IS_ERR(new)) 1593 return PTR_ERR(new); 1594 1595 __fanout_set_data_bpf(po->fanout, new); 1596 return 0; 1597 } 1598 1599 static int fanout_set_data(struct packet_sock *po, char __user *data, 1600 unsigned int len) 1601 { 1602 switch (po->fanout->type) { 1603 case PACKET_FANOUT_CBPF: 1604 return fanout_set_data_cbpf(po, data, len); 1605 case PACKET_FANOUT_EBPF: 1606 return fanout_set_data_ebpf(po, data, len); 1607 default: 1608 return -EINVAL; 1609 }; 1610 } 1611 1612 static void fanout_release_data(struct packet_fanout *f) 1613 { 1614 switch (f->type) { 1615 case PACKET_FANOUT_CBPF: 1616 case PACKET_FANOUT_EBPF: 1617 __fanout_set_data_bpf(f, NULL); 1618 }; 1619 } 1620 1621 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1622 { 1623 struct packet_sock *po = pkt_sk(sk); 1624 struct packet_fanout *f, *match; 1625 u8 type = type_flags & 0xff; 1626 u8 flags = type_flags >> 8; 1627 int err; 1628 1629 switch (type) { 1630 case PACKET_FANOUT_ROLLOVER: 1631 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1632 return -EINVAL; 1633 case PACKET_FANOUT_HASH: 1634 case PACKET_FANOUT_LB: 1635 case PACKET_FANOUT_CPU: 1636 case PACKET_FANOUT_RND: 1637 case PACKET_FANOUT_QM: 1638 case PACKET_FANOUT_CBPF: 1639 case PACKET_FANOUT_EBPF: 1640 break; 1641 default: 1642 return -EINVAL; 1643 } 1644 1645 if (!po->running) 1646 return -EINVAL; 1647 1648 if (po->fanout) 1649 return -EALREADY; 1650 1651 if (type == PACKET_FANOUT_ROLLOVER || 1652 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1653 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL); 1654 if (!po->rollover) 1655 return -ENOMEM; 1656 atomic_long_set(&po->rollover->num, 0); 1657 atomic_long_set(&po->rollover->num_huge, 0); 1658 atomic_long_set(&po->rollover->num_failed, 0); 1659 } 1660 1661 mutex_lock(&fanout_mutex); 1662 match = NULL; 1663 list_for_each_entry(f, &fanout_list, list) { 1664 if (f->id == id && 1665 read_pnet(&f->net) == sock_net(sk)) { 1666 match = f; 1667 break; 1668 } 1669 } 1670 err = -EINVAL; 1671 if (match && match->flags != flags) 1672 goto out; 1673 if (!match) { 1674 err = -ENOMEM; 1675 match = kzalloc(sizeof(*match), GFP_KERNEL); 1676 if (!match) 1677 goto out; 1678 write_pnet(&match->net, sock_net(sk)); 1679 match->id = id; 1680 match->type = type; 1681 match->flags = flags; 1682 INIT_LIST_HEAD(&match->list); 1683 spin_lock_init(&match->lock); 1684 atomic_set(&match->sk_ref, 0); 1685 fanout_init_data(match); 1686 match->prot_hook.type = po->prot_hook.type; 1687 match->prot_hook.dev = po->prot_hook.dev; 1688 match->prot_hook.func = packet_rcv_fanout; 1689 match->prot_hook.af_packet_priv = match; 1690 match->prot_hook.id_match = match_fanout_group; 1691 dev_add_pack(&match->prot_hook); 1692 list_add(&match->list, &fanout_list); 1693 } 1694 err = -EINVAL; 1695 if (match->type == type && 1696 match->prot_hook.type == po->prot_hook.type && 1697 match->prot_hook.dev == po->prot_hook.dev) { 1698 err = -ENOSPC; 1699 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1700 __dev_remove_pack(&po->prot_hook); 1701 po->fanout = match; 1702 atomic_inc(&match->sk_ref); 1703 __fanout_link(sk, po); 1704 err = 0; 1705 } 1706 } 1707 out: 1708 mutex_unlock(&fanout_mutex); 1709 if (err) { 1710 kfree(po->rollover); 1711 po->rollover = NULL; 1712 } 1713 return err; 1714 } 1715 1716 static void fanout_release(struct sock *sk) 1717 { 1718 struct packet_sock *po = pkt_sk(sk); 1719 struct packet_fanout *f; 1720 1721 f = po->fanout; 1722 if (!f) 1723 return; 1724 1725 mutex_lock(&fanout_mutex); 1726 po->fanout = NULL; 1727 1728 if (atomic_dec_and_test(&f->sk_ref)) { 1729 list_del(&f->list); 1730 dev_remove_pack(&f->prot_hook); 1731 fanout_release_data(f); 1732 kfree(f); 1733 } 1734 mutex_unlock(&fanout_mutex); 1735 1736 if (po->rollover) 1737 kfree_rcu(po->rollover, rcu); 1738 } 1739 1740 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1741 struct sk_buff *skb) 1742 { 1743 /* Earlier code assumed this would be a VLAN pkt, double-check 1744 * this now that we have the actual packet in hand. We can only 1745 * do this check on Ethernet devices. 1746 */ 1747 if (unlikely(dev->type != ARPHRD_ETHER)) 1748 return false; 1749 1750 skb_reset_mac_header(skb); 1751 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1752 } 1753 1754 static const struct proto_ops packet_ops; 1755 1756 static const struct proto_ops packet_ops_spkt; 1757 1758 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1759 struct packet_type *pt, struct net_device *orig_dev) 1760 { 1761 struct sock *sk; 1762 struct sockaddr_pkt *spkt; 1763 1764 /* 1765 * When we registered the protocol we saved the socket in the data 1766 * field for just this event. 1767 */ 1768 1769 sk = pt->af_packet_priv; 1770 1771 /* 1772 * Yank back the headers [hope the device set this 1773 * right or kerboom...] 1774 * 1775 * Incoming packets have ll header pulled, 1776 * push it back. 1777 * 1778 * For outgoing ones skb->data == skb_mac_header(skb) 1779 * so that this procedure is noop. 1780 */ 1781 1782 if (skb->pkt_type == PACKET_LOOPBACK) 1783 goto out; 1784 1785 if (!net_eq(dev_net(dev), sock_net(sk))) 1786 goto out; 1787 1788 skb = skb_share_check(skb, GFP_ATOMIC); 1789 if (skb == NULL) 1790 goto oom; 1791 1792 /* drop any routing info */ 1793 skb_dst_drop(skb); 1794 1795 /* drop conntrack reference */ 1796 nf_reset(skb); 1797 1798 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1799 1800 skb_push(skb, skb->data - skb_mac_header(skb)); 1801 1802 /* 1803 * The SOCK_PACKET socket receives _all_ frames. 1804 */ 1805 1806 spkt->spkt_family = dev->type; 1807 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1808 spkt->spkt_protocol = skb->protocol; 1809 1810 /* 1811 * Charge the memory to the socket. This is done specifically 1812 * to prevent sockets using all the memory up. 1813 */ 1814 1815 if (sock_queue_rcv_skb(sk, skb) == 0) 1816 return 0; 1817 1818 out: 1819 kfree_skb(skb); 1820 oom: 1821 return 0; 1822 } 1823 1824 1825 /* 1826 * Output a raw packet to a device layer. This bypasses all the other 1827 * protocol layers and you must therefore supply it with a complete frame 1828 */ 1829 1830 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1831 size_t len) 1832 { 1833 struct sock *sk = sock->sk; 1834 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1835 struct sk_buff *skb = NULL; 1836 struct net_device *dev; 1837 struct sockcm_cookie sockc; 1838 __be16 proto = 0; 1839 int err; 1840 int extra_len = 0; 1841 1842 /* 1843 * Get and verify the address. 1844 */ 1845 1846 if (saddr) { 1847 if (msg->msg_namelen < sizeof(struct sockaddr)) 1848 return -EINVAL; 1849 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1850 proto = saddr->spkt_protocol; 1851 } else 1852 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1853 1854 /* 1855 * Find the device first to size check it 1856 */ 1857 1858 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1859 retry: 1860 rcu_read_lock(); 1861 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1862 err = -ENODEV; 1863 if (dev == NULL) 1864 goto out_unlock; 1865 1866 err = -ENETDOWN; 1867 if (!(dev->flags & IFF_UP)) 1868 goto out_unlock; 1869 1870 /* 1871 * You may not queue a frame bigger than the mtu. This is the lowest level 1872 * raw protocol and you must do your own fragmentation at this level. 1873 */ 1874 1875 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1876 if (!netif_supports_nofcs(dev)) { 1877 err = -EPROTONOSUPPORT; 1878 goto out_unlock; 1879 } 1880 extra_len = 4; /* We're doing our own CRC */ 1881 } 1882 1883 err = -EMSGSIZE; 1884 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1885 goto out_unlock; 1886 1887 if (!skb) { 1888 size_t reserved = LL_RESERVED_SPACE(dev); 1889 int tlen = dev->needed_tailroom; 1890 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1891 1892 rcu_read_unlock(); 1893 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1894 if (skb == NULL) 1895 return -ENOBUFS; 1896 /* FIXME: Save some space for broken drivers that write a hard 1897 * header at transmission time by themselves. PPP is the notable 1898 * one here. This should really be fixed at the driver level. 1899 */ 1900 skb_reserve(skb, reserved); 1901 skb_reset_network_header(skb); 1902 1903 /* Try to align data part correctly */ 1904 if (hhlen) { 1905 skb->data -= hhlen; 1906 skb->tail -= hhlen; 1907 if (len < hhlen) 1908 skb_reset_network_header(skb); 1909 } 1910 err = memcpy_from_msg(skb_put(skb, len), msg, len); 1911 if (err) 1912 goto out_free; 1913 goto retry; 1914 } 1915 1916 if (!dev_validate_header(dev, skb->data, len)) { 1917 err = -EINVAL; 1918 goto out_unlock; 1919 } 1920 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 1921 !packet_extra_vlan_len_allowed(dev, skb)) { 1922 err = -EMSGSIZE; 1923 goto out_unlock; 1924 } 1925 1926 sockc.tsflags = sk->sk_tsflags; 1927 if (msg->msg_controllen) { 1928 err = sock_cmsg_send(sk, msg, &sockc); 1929 if (unlikely(err)) 1930 goto out_unlock; 1931 } 1932 1933 skb->protocol = proto; 1934 skb->dev = dev; 1935 skb->priority = sk->sk_priority; 1936 skb->mark = sk->sk_mark; 1937 1938 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags); 1939 1940 if (unlikely(extra_len == 4)) 1941 skb->no_fcs = 1; 1942 1943 skb_probe_transport_header(skb, 0); 1944 1945 dev_queue_xmit(skb); 1946 rcu_read_unlock(); 1947 return len; 1948 1949 out_unlock: 1950 rcu_read_unlock(); 1951 out_free: 1952 kfree_skb(skb); 1953 return err; 1954 } 1955 1956 static unsigned int run_filter(struct sk_buff *skb, 1957 const struct sock *sk, 1958 unsigned int res) 1959 { 1960 struct sk_filter *filter; 1961 1962 rcu_read_lock(); 1963 filter = rcu_dereference(sk->sk_filter); 1964 if (filter != NULL) 1965 res = bpf_prog_run_clear_cb(filter->prog, skb); 1966 rcu_read_unlock(); 1967 1968 return res; 1969 } 1970 1971 static int __packet_rcv_vnet(const struct sk_buff *skb, 1972 struct virtio_net_hdr *vnet_hdr) 1973 { 1974 *vnet_hdr = (const struct virtio_net_hdr) { 0 }; 1975 1976 if (virtio_net_hdr_from_skb(skb, vnet_hdr, vio_le())) 1977 BUG(); 1978 1979 return 0; 1980 } 1981 1982 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 1983 size_t *len) 1984 { 1985 struct virtio_net_hdr vnet_hdr; 1986 1987 if (*len < sizeof(vnet_hdr)) 1988 return -EINVAL; 1989 *len -= sizeof(vnet_hdr); 1990 1991 if (__packet_rcv_vnet(skb, &vnet_hdr)) 1992 return -EINVAL; 1993 1994 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 1995 } 1996 1997 /* 1998 * This function makes lazy skb cloning in hope that most of packets 1999 * are discarded by BPF. 2000 * 2001 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2002 * and skb->cb are mangled. It works because (and until) packets 2003 * falling here are owned by current CPU. Output packets are cloned 2004 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2005 * sequencially, so that if we return skb to original state on exit, 2006 * we will not harm anyone. 2007 */ 2008 2009 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2010 struct packet_type *pt, struct net_device *orig_dev) 2011 { 2012 struct sock *sk; 2013 struct sockaddr_ll *sll; 2014 struct packet_sock *po; 2015 u8 *skb_head = skb->data; 2016 int skb_len = skb->len; 2017 unsigned int snaplen, res; 2018 bool is_drop_n_account = false; 2019 2020 if (skb->pkt_type == PACKET_LOOPBACK) 2021 goto drop; 2022 2023 sk = pt->af_packet_priv; 2024 po = pkt_sk(sk); 2025 2026 if (!net_eq(dev_net(dev), sock_net(sk))) 2027 goto drop; 2028 2029 skb->dev = dev; 2030 2031 if (dev->header_ops) { 2032 /* The device has an explicit notion of ll header, 2033 * exported to higher levels. 2034 * 2035 * Otherwise, the device hides details of its frame 2036 * structure, so that corresponding packet head is 2037 * never delivered to user. 2038 */ 2039 if (sk->sk_type != SOCK_DGRAM) 2040 skb_push(skb, skb->data - skb_mac_header(skb)); 2041 else if (skb->pkt_type == PACKET_OUTGOING) { 2042 /* Special case: outgoing packets have ll header at head */ 2043 skb_pull(skb, skb_network_offset(skb)); 2044 } 2045 } 2046 2047 snaplen = skb->len; 2048 2049 res = run_filter(skb, sk, snaplen); 2050 if (!res) 2051 goto drop_n_restore; 2052 if (snaplen > res) 2053 snaplen = res; 2054 2055 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2056 goto drop_n_acct; 2057 2058 if (skb_shared(skb)) { 2059 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2060 if (nskb == NULL) 2061 goto drop_n_acct; 2062 2063 if (skb_head != skb->data) { 2064 skb->data = skb_head; 2065 skb->len = skb_len; 2066 } 2067 consume_skb(skb); 2068 skb = nskb; 2069 } 2070 2071 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2072 2073 sll = &PACKET_SKB_CB(skb)->sa.ll; 2074 sll->sll_hatype = dev->type; 2075 sll->sll_pkttype = skb->pkt_type; 2076 if (unlikely(po->origdev)) 2077 sll->sll_ifindex = orig_dev->ifindex; 2078 else 2079 sll->sll_ifindex = dev->ifindex; 2080 2081 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2082 2083 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2084 * Use their space for storing the original skb length. 2085 */ 2086 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2087 2088 if (pskb_trim(skb, snaplen)) 2089 goto drop_n_acct; 2090 2091 skb_set_owner_r(skb, sk); 2092 skb->dev = NULL; 2093 skb_dst_drop(skb); 2094 2095 /* drop conntrack reference */ 2096 nf_reset(skb); 2097 2098 spin_lock(&sk->sk_receive_queue.lock); 2099 po->stats.stats1.tp_packets++; 2100 sock_skb_set_dropcount(sk, skb); 2101 __skb_queue_tail(&sk->sk_receive_queue, skb); 2102 spin_unlock(&sk->sk_receive_queue.lock); 2103 sk->sk_data_ready(sk); 2104 return 0; 2105 2106 drop_n_acct: 2107 is_drop_n_account = true; 2108 spin_lock(&sk->sk_receive_queue.lock); 2109 po->stats.stats1.tp_drops++; 2110 atomic_inc(&sk->sk_drops); 2111 spin_unlock(&sk->sk_receive_queue.lock); 2112 2113 drop_n_restore: 2114 if (skb_head != skb->data && skb_shared(skb)) { 2115 skb->data = skb_head; 2116 skb->len = skb_len; 2117 } 2118 drop: 2119 if (!is_drop_n_account) 2120 consume_skb(skb); 2121 else 2122 kfree_skb(skb); 2123 return 0; 2124 } 2125 2126 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2127 struct packet_type *pt, struct net_device *orig_dev) 2128 { 2129 struct sock *sk; 2130 struct packet_sock *po; 2131 struct sockaddr_ll *sll; 2132 union tpacket_uhdr h; 2133 u8 *skb_head = skb->data; 2134 int skb_len = skb->len; 2135 unsigned int snaplen, res; 2136 unsigned long status = TP_STATUS_USER; 2137 unsigned short macoff, netoff, hdrlen; 2138 struct sk_buff *copy_skb = NULL; 2139 struct timespec ts; 2140 __u32 ts_status; 2141 bool is_drop_n_account = false; 2142 2143 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2144 * We may add members to them until current aligned size without forcing 2145 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2146 */ 2147 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2148 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2149 2150 if (skb->pkt_type == PACKET_LOOPBACK) 2151 goto drop; 2152 2153 sk = pt->af_packet_priv; 2154 po = pkt_sk(sk); 2155 2156 if (!net_eq(dev_net(dev), sock_net(sk))) 2157 goto drop; 2158 2159 if (dev->header_ops) { 2160 if (sk->sk_type != SOCK_DGRAM) 2161 skb_push(skb, skb->data - skb_mac_header(skb)); 2162 else if (skb->pkt_type == PACKET_OUTGOING) { 2163 /* Special case: outgoing packets have ll header at head */ 2164 skb_pull(skb, skb_network_offset(skb)); 2165 } 2166 } 2167 2168 snaplen = skb->len; 2169 2170 res = run_filter(skb, sk, snaplen); 2171 if (!res) 2172 goto drop_n_restore; 2173 2174 if (skb->ip_summed == CHECKSUM_PARTIAL) 2175 status |= TP_STATUS_CSUMNOTREADY; 2176 else if (skb->pkt_type != PACKET_OUTGOING && 2177 (skb->ip_summed == CHECKSUM_COMPLETE || 2178 skb_csum_unnecessary(skb))) 2179 status |= TP_STATUS_CSUM_VALID; 2180 2181 if (snaplen > res) 2182 snaplen = res; 2183 2184 if (sk->sk_type == SOCK_DGRAM) { 2185 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2186 po->tp_reserve; 2187 } else { 2188 unsigned int maclen = skb_network_offset(skb); 2189 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2190 (maclen < 16 ? 16 : maclen)) + 2191 po->tp_reserve; 2192 if (po->has_vnet_hdr) 2193 netoff += sizeof(struct virtio_net_hdr); 2194 macoff = netoff - maclen; 2195 } 2196 if (po->tp_version <= TPACKET_V2) { 2197 if (macoff + snaplen > po->rx_ring.frame_size) { 2198 if (po->copy_thresh && 2199 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2200 if (skb_shared(skb)) { 2201 copy_skb = skb_clone(skb, GFP_ATOMIC); 2202 } else { 2203 copy_skb = skb_get(skb); 2204 skb_head = skb->data; 2205 } 2206 if (copy_skb) 2207 skb_set_owner_r(copy_skb, sk); 2208 } 2209 snaplen = po->rx_ring.frame_size - macoff; 2210 if ((int)snaplen < 0) 2211 snaplen = 0; 2212 } 2213 } else if (unlikely(macoff + snaplen > 2214 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2215 u32 nval; 2216 2217 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2218 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2219 snaplen, nval, macoff); 2220 snaplen = nval; 2221 if (unlikely((int)snaplen < 0)) { 2222 snaplen = 0; 2223 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2224 } 2225 } 2226 spin_lock(&sk->sk_receive_queue.lock); 2227 h.raw = packet_current_rx_frame(po, skb, 2228 TP_STATUS_KERNEL, (macoff+snaplen)); 2229 if (!h.raw) 2230 goto drop_n_account; 2231 if (po->tp_version <= TPACKET_V2) { 2232 packet_increment_rx_head(po, &po->rx_ring); 2233 /* 2234 * LOSING will be reported till you read the stats, 2235 * because it's COR - Clear On Read. 2236 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2237 * at packet level. 2238 */ 2239 if (po->stats.stats1.tp_drops) 2240 status |= TP_STATUS_LOSING; 2241 } 2242 po->stats.stats1.tp_packets++; 2243 if (copy_skb) { 2244 status |= TP_STATUS_COPY; 2245 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2246 } 2247 spin_unlock(&sk->sk_receive_queue.lock); 2248 2249 if (po->has_vnet_hdr) { 2250 if (__packet_rcv_vnet(skb, h.raw + macoff - 2251 sizeof(struct virtio_net_hdr))) { 2252 spin_lock(&sk->sk_receive_queue.lock); 2253 goto drop_n_account; 2254 } 2255 } 2256 2257 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2258 2259 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 2260 getnstimeofday(&ts); 2261 2262 status |= ts_status; 2263 2264 switch (po->tp_version) { 2265 case TPACKET_V1: 2266 h.h1->tp_len = skb->len; 2267 h.h1->tp_snaplen = snaplen; 2268 h.h1->tp_mac = macoff; 2269 h.h1->tp_net = netoff; 2270 h.h1->tp_sec = ts.tv_sec; 2271 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2272 hdrlen = sizeof(*h.h1); 2273 break; 2274 case TPACKET_V2: 2275 h.h2->tp_len = skb->len; 2276 h.h2->tp_snaplen = snaplen; 2277 h.h2->tp_mac = macoff; 2278 h.h2->tp_net = netoff; 2279 h.h2->tp_sec = ts.tv_sec; 2280 h.h2->tp_nsec = ts.tv_nsec; 2281 if (skb_vlan_tag_present(skb)) { 2282 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2283 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2284 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2285 } else { 2286 h.h2->tp_vlan_tci = 0; 2287 h.h2->tp_vlan_tpid = 0; 2288 } 2289 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2290 hdrlen = sizeof(*h.h2); 2291 break; 2292 case TPACKET_V3: 2293 /* tp_nxt_offset,vlan are already populated above. 2294 * So DONT clear those fields here 2295 */ 2296 h.h3->tp_status |= status; 2297 h.h3->tp_len = skb->len; 2298 h.h3->tp_snaplen = snaplen; 2299 h.h3->tp_mac = macoff; 2300 h.h3->tp_net = netoff; 2301 h.h3->tp_sec = ts.tv_sec; 2302 h.h3->tp_nsec = ts.tv_nsec; 2303 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2304 hdrlen = sizeof(*h.h3); 2305 break; 2306 default: 2307 BUG(); 2308 } 2309 2310 sll = h.raw + TPACKET_ALIGN(hdrlen); 2311 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2312 sll->sll_family = AF_PACKET; 2313 sll->sll_hatype = dev->type; 2314 sll->sll_protocol = skb->protocol; 2315 sll->sll_pkttype = skb->pkt_type; 2316 if (unlikely(po->origdev)) 2317 sll->sll_ifindex = orig_dev->ifindex; 2318 else 2319 sll->sll_ifindex = dev->ifindex; 2320 2321 smp_mb(); 2322 2323 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2324 if (po->tp_version <= TPACKET_V2) { 2325 u8 *start, *end; 2326 2327 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2328 macoff + snaplen); 2329 2330 for (start = h.raw; start < end; start += PAGE_SIZE) 2331 flush_dcache_page(pgv_to_page(start)); 2332 } 2333 smp_wmb(); 2334 #endif 2335 2336 if (po->tp_version <= TPACKET_V2) { 2337 __packet_set_status(po, h.raw, status); 2338 sk->sk_data_ready(sk); 2339 } else { 2340 prb_clear_blk_fill_status(&po->rx_ring); 2341 } 2342 2343 drop_n_restore: 2344 if (skb_head != skb->data && skb_shared(skb)) { 2345 skb->data = skb_head; 2346 skb->len = skb_len; 2347 } 2348 drop: 2349 if (!is_drop_n_account) 2350 consume_skb(skb); 2351 else 2352 kfree_skb(skb); 2353 return 0; 2354 2355 drop_n_account: 2356 is_drop_n_account = true; 2357 po->stats.stats1.tp_drops++; 2358 spin_unlock(&sk->sk_receive_queue.lock); 2359 2360 sk->sk_data_ready(sk); 2361 kfree_skb(copy_skb); 2362 goto drop_n_restore; 2363 } 2364 2365 static void tpacket_destruct_skb(struct sk_buff *skb) 2366 { 2367 struct packet_sock *po = pkt_sk(skb->sk); 2368 2369 if (likely(po->tx_ring.pg_vec)) { 2370 void *ph; 2371 __u32 ts; 2372 2373 ph = skb_shinfo(skb)->destructor_arg; 2374 packet_dec_pending(&po->tx_ring); 2375 2376 ts = __packet_set_timestamp(po, ph, skb); 2377 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2378 } 2379 2380 sock_wfree(skb); 2381 } 2382 2383 static void tpacket_set_protocol(const struct net_device *dev, 2384 struct sk_buff *skb) 2385 { 2386 if (dev->type == ARPHRD_ETHER) { 2387 skb_reset_mac_header(skb); 2388 skb->protocol = eth_hdr(skb)->h_proto; 2389 } 2390 } 2391 2392 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2393 { 2394 unsigned short gso_type = 0; 2395 2396 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2397 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2398 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2399 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2400 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2401 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2402 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2403 2404 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2405 return -EINVAL; 2406 2407 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 2408 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 2409 case VIRTIO_NET_HDR_GSO_TCPV4: 2410 gso_type = SKB_GSO_TCPV4; 2411 break; 2412 case VIRTIO_NET_HDR_GSO_TCPV6: 2413 gso_type = SKB_GSO_TCPV6; 2414 break; 2415 case VIRTIO_NET_HDR_GSO_UDP: 2416 gso_type = SKB_GSO_UDP; 2417 break; 2418 default: 2419 return -EINVAL; 2420 } 2421 2422 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) 2423 gso_type |= SKB_GSO_TCP_ECN; 2424 2425 if (vnet_hdr->gso_size == 0) 2426 return -EINVAL; 2427 } 2428 2429 vnet_hdr->gso_type = gso_type; /* changes type, temporary storage */ 2430 return 0; 2431 } 2432 2433 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2434 struct virtio_net_hdr *vnet_hdr) 2435 { 2436 int n; 2437 2438 if (*len < sizeof(*vnet_hdr)) 2439 return -EINVAL; 2440 *len -= sizeof(*vnet_hdr); 2441 2442 n = copy_from_iter(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter); 2443 if (n != sizeof(*vnet_hdr)) 2444 return -EFAULT; 2445 2446 return __packet_snd_vnet_parse(vnet_hdr, *len); 2447 } 2448 2449 static int packet_snd_vnet_gso(struct sk_buff *skb, 2450 struct virtio_net_hdr *vnet_hdr) 2451 { 2452 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 2453 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start); 2454 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset); 2455 2456 if (!skb_partial_csum_set(skb, s, o)) 2457 return -EINVAL; 2458 } 2459 2460 skb_shinfo(skb)->gso_size = 2461 __virtio16_to_cpu(vio_le(), vnet_hdr->gso_size); 2462 skb_shinfo(skb)->gso_type = vnet_hdr->gso_type; 2463 2464 /* Header must be checked, and gso_segs computed. */ 2465 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 2466 skb_shinfo(skb)->gso_segs = 0; 2467 return 0; 2468 } 2469 2470 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2471 void *frame, struct net_device *dev, void *data, int tp_len, 2472 __be16 proto, unsigned char *addr, int hlen, int copylen, 2473 const struct sockcm_cookie *sockc) 2474 { 2475 union tpacket_uhdr ph; 2476 int to_write, offset, len, nr_frags, len_max; 2477 struct socket *sock = po->sk.sk_socket; 2478 struct page *page; 2479 int err; 2480 2481 ph.raw = frame; 2482 2483 skb->protocol = proto; 2484 skb->dev = dev; 2485 skb->priority = po->sk.sk_priority; 2486 skb->mark = po->sk.sk_mark; 2487 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags); 2488 skb_shinfo(skb)->destructor_arg = ph.raw; 2489 2490 skb_reserve(skb, hlen); 2491 skb_reset_network_header(skb); 2492 2493 to_write = tp_len; 2494 2495 if (sock->type == SOCK_DGRAM) { 2496 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2497 NULL, tp_len); 2498 if (unlikely(err < 0)) 2499 return -EINVAL; 2500 } else if (copylen) { 2501 int hdrlen = min_t(int, copylen, tp_len); 2502 2503 skb_push(skb, dev->hard_header_len); 2504 skb_put(skb, copylen - dev->hard_header_len); 2505 err = skb_store_bits(skb, 0, data, hdrlen); 2506 if (unlikely(err)) 2507 return err; 2508 if (!dev_validate_header(dev, skb->data, hdrlen)) 2509 return -EINVAL; 2510 if (!skb->protocol) 2511 tpacket_set_protocol(dev, skb); 2512 2513 data += hdrlen; 2514 to_write -= hdrlen; 2515 } 2516 2517 offset = offset_in_page(data); 2518 len_max = PAGE_SIZE - offset; 2519 len = ((to_write > len_max) ? len_max : to_write); 2520 2521 skb->data_len = to_write; 2522 skb->len += to_write; 2523 skb->truesize += to_write; 2524 atomic_add(to_write, &po->sk.sk_wmem_alloc); 2525 2526 while (likely(to_write)) { 2527 nr_frags = skb_shinfo(skb)->nr_frags; 2528 2529 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2530 pr_err("Packet exceed the number of skb frags(%lu)\n", 2531 MAX_SKB_FRAGS); 2532 return -EFAULT; 2533 } 2534 2535 page = pgv_to_page(data); 2536 data += len; 2537 flush_dcache_page(page); 2538 get_page(page); 2539 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2540 to_write -= len; 2541 offset = 0; 2542 len_max = PAGE_SIZE; 2543 len = ((to_write > len_max) ? len_max : to_write); 2544 } 2545 2546 skb_probe_transport_header(skb, 0); 2547 2548 return tp_len; 2549 } 2550 2551 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2552 int size_max, void **data) 2553 { 2554 union tpacket_uhdr ph; 2555 int tp_len, off; 2556 2557 ph.raw = frame; 2558 2559 switch (po->tp_version) { 2560 case TPACKET_V2: 2561 tp_len = ph.h2->tp_len; 2562 break; 2563 default: 2564 tp_len = ph.h1->tp_len; 2565 break; 2566 } 2567 if (unlikely(tp_len > size_max)) { 2568 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2569 return -EMSGSIZE; 2570 } 2571 2572 if (unlikely(po->tp_tx_has_off)) { 2573 int off_min, off_max; 2574 2575 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2576 off_max = po->tx_ring.frame_size - tp_len; 2577 if (po->sk.sk_type == SOCK_DGRAM) { 2578 switch (po->tp_version) { 2579 case TPACKET_V2: 2580 off = ph.h2->tp_net; 2581 break; 2582 default: 2583 off = ph.h1->tp_net; 2584 break; 2585 } 2586 } else { 2587 switch (po->tp_version) { 2588 case TPACKET_V2: 2589 off = ph.h2->tp_mac; 2590 break; 2591 default: 2592 off = ph.h1->tp_mac; 2593 break; 2594 } 2595 } 2596 if (unlikely((off < off_min) || (off_max < off))) 2597 return -EINVAL; 2598 } else { 2599 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2600 } 2601 2602 *data = frame + off; 2603 return tp_len; 2604 } 2605 2606 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2607 { 2608 struct sk_buff *skb; 2609 struct net_device *dev; 2610 struct virtio_net_hdr *vnet_hdr = NULL; 2611 struct sockcm_cookie sockc; 2612 __be16 proto; 2613 int err, reserve = 0; 2614 void *ph; 2615 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2616 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2617 int tp_len, size_max; 2618 unsigned char *addr; 2619 void *data; 2620 int len_sum = 0; 2621 int status = TP_STATUS_AVAILABLE; 2622 int hlen, tlen, copylen = 0; 2623 2624 mutex_lock(&po->pg_vec_lock); 2625 2626 if (likely(saddr == NULL)) { 2627 dev = packet_cached_dev_get(po); 2628 proto = po->num; 2629 addr = NULL; 2630 } else { 2631 err = -EINVAL; 2632 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2633 goto out; 2634 if (msg->msg_namelen < (saddr->sll_halen 2635 + offsetof(struct sockaddr_ll, 2636 sll_addr))) 2637 goto out; 2638 proto = saddr->sll_protocol; 2639 addr = saddr->sll_addr; 2640 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2641 } 2642 2643 sockc.tsflags = po->sk.sk_tsflags; 2644 if (msg->msg_controllen) { 2645 err = sock_cmsg_send(&po->sk, msg, &sockc); 2646 if (unlikely(err)) 2647 goto out; 2648 } 2649 2650 err = -ENXIO; 2651 if (unlikely(dev == NULL)) 2652 goto out; 2653 err = -ENETDOWN; 2654 if (unlikely(!(dev->flags & IFF_UP))) 2655 goto out_put; 2656 2657 if (po->sk.sk_socket->type == SOCK_RAW) 2658 reserve = dev->hard_header_len; 2659 size_max = po->tx_ring.frame_size 2660 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2661 2662 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) 2663 size_max = dev->mtu + reserve + VLAN_HLEN; 2664 2665 do { 2666 ph = packet_current_frame(po, &po->tx_ring, 2667 TP_STATUS_SEND_REQUEST); 2668 if (unlikely(ph == NULL)) { 2669 if (need_wait && need_resched()) 2670 schedule(); 2671 continue; 2672 } 2673 2674 skb = NULL; 2675 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2676 if (tp_len < 0) 2677 goto tpacket_error; 2678 2679 status = TP_STATUS_SEND_REQUEST; 2680 hlen = LL_RESERVED_SPACE(dev); 2681 tlen = dev->needed_tailroom; 2682 if (po->has_vnet_hdr) { 2683 vnet_hdr = data; 2684 data += sizeof(*vnet_hdr); 2685 tp_len -= sizeof(*vnet_hdr); 2686 if (tp_len < 0 || 2687 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2688 tp_len = -EINVAL; 2689 goto tpacket_error; 2690 } 2691 copylen = __virtio16_to_cpu(vio_le(), 2692 vnet_hdr->hdr_len); 2693 } 2694 copylen = max_t(int, copylen, dev->hard_header_len); 2695 skb = sock_alloc_send_skb(&po->sk, 2696 hlen + tlen + sizeof(struct sockaddr_ll) + 2697 (copylen - dev->hard_header_len), 2698 !need_wait, &err); 2699 2700 if (unlikely(skb == NULL)) { 2701 /* we assume the socket was initially writeable ... */ 2702 if (likely(len_sum > 0)) 2703 err = len_sum; 2704 goto out_status; 2705 } 2706 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2707 addr, hlen, copylen, &sockc); 2708 if (likely(tp_len >= 0) && 2709 tp_len > dev->mtu + reserve && 2710 !po->has_vnet_hdr && 2711 !packet_extra_vlan_len_allowed(dev, skb)) 2712 tp_len = -EMSGSIZE; 2713 2714 if (unlikely(tp_len < 0)) { 2715 tpacket_error: 2716 if (po->tp_loss) { 2717 __packet_set_status(po, ph, 2718 TP_STATUS_AVAILABLE); 2719 packet_increment_head(&po->tx_ring); 2720 kfree_skb(skb); 2721 continue; 2722 } else { 2723 status = TP_STATUS_WRONG_FORMAT; 2724 err = tp_len; 2725 goto out_status; 2726 } 2727 } 2728 2729 if (po->has_vnet_hdr && packet_snd_vnet_gso(skb, vnet_hdr)) { 2730 tp_len = -EINVAL; 2731 goto tpacket_error; 2732 } 2733 2734 packet_pick_tx_queue(dev, skb); 2735 2736 skb->destructor = tpacket_destruct_skb; 2737 __packet_set_status(po, ph, TP_STATUS_SENDING); 2738 packet_inc_pending(&po->tx_ring); 2739 2740 status = TP_STATUS_SEND_REQUEST; 2741 err = po->xmit(skb); 2742 if (unlikely(err > 0)) { 2743 err = net_xmit_errno(err); 2744 if (err && __packet_get_status(po, ph) == 2745 TP_STATUS_AVAILABLE) { 2746 /* skb was destructed already */ 2747 skb = NULL; 2748 goto out_status; 2749 } 2750 /* 2751 * skb was dropped but not destructed yet; 2752 * let's treat it like congestion or err < 0 2753 */ 2754 err = 0; 2755 } 2756 packet_increment_head(&po->tx_ring); 2757 len_sum += tp_len; 2758 } while (likely((ph != NULL) || 2759 /* Note: packet_read_pending() might be slow if we have 2760 * to call it as it's per_cpu variable, but in fast-path 2761 * we already short-circuit the loop with the first 2762 * condition, and luckily don't have to go that path 2763 * anyway. 2764 */ 2765 (need_wait && packet_read_pending(&po->tx_ring)))); 2766 2767 err = len_sum; 2768 goto out_put; 2769 2770 out_status: 2771 __packet_set_status(po, ph, status); 2772 kfree_skb(skb); 2773 out_put: 2774 dev_put(dev); 2775 out: 2776 mutex_unlock(&po->pg_vec_lock); 2777 return err; 2778 } 2779 2780 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2781 size_t reserve, size_t len, 2782 size_t linear, int noblock, 2783 int *err) 2784 { 2785 struct sk_buff *skb; 2786 2787 /* Under a page? Don't bother with paged skb. */ 2788 if (prepad + len < PAGE_SIZE || !linear) 2789 linear = len; 2790 2791 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2792 err, 0); 2793 if (!skb) 2794 return NULL; 2795 2796 skb_reserve(skb, reserve); 2797 skb_put(skb, linear); 2798 skb->data_len = len - linear; 2799 skb->len += len - linear; 2800 2801 return skb; 2802 } 2803 2804 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2805 { 2806 struct sock *sk = sock->sk; 2807 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2808 struct sk_buff *skb; 2809 struct net_device *dev; 2810 __be16 proto; 2811 unsigned char *addr; 2812 int err, reserve = 0; 2813 struct sockcm_cookie sockc; 2814 struct virtio_net_hdr vnet_hdr = { 0 }; 2815 int offset = 0; 2816 struct packet_sock *po = pkt_sk(sk); 2817 int hlen, tlen; 2818 int extra_len = 0; 2819 2820 /* 2821 * Get and verify the address. 2822 */ 2823 2824 if (likely(saddr == NULL)) { 2825 dev = packet_cached_dev_get(po); 2826 proto = po->num; 2827 addr = NULL; 2828 } else { 2829 err = -EINVAL; 2830 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2831 goto out; 2832 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2833 goto out; 2834 proto = saddr->sll_protocol; 2835 addr = saddr->sll_addr; 2836 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2837 } 2838 2839 err = -ENXIO; 2840 if (unlikely(dev == NULL)) 2841 goto out_unlock; 2842 err = -ENETDOWN; 2843 if (unlikely(!(dev->flags & IFF_UP))) 2844 goto out_unlock; 2845 2846 sockc.tsflags = sk->sk_tsflags; 2847 sockc.mark = sk->sk_mark; 2848 if (msg->msg_controllen) { 2849 err = sock_cmsg_send(sk, msg, &sockc); 2850 if (unlikely(err)) 2851 goto out_unlock; 2852 } 2853 2854 if (sock->type == SOCK_RAW) 2855 reserve = dev->hard_header_len; 2856 if (po->has_vnet_hdr) { 2857 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2858 if (err) 2859 goto out_unlock; 2860 } 2861 2862 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2863 if (!netif_supports_nofcs(dev)) { 2864 err = -EPROTONOSUPPORT; 2865 goto out_unlock; 2866 } 2867 extra_len = 4; /* We're doing our own CRC */ 2868 } 2869 2870 err = -EMSGSIZE; 2871 if (!vnet_hdr.gso_type && 2872 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2873 goto out_unlock; 2874 2875 err = -ENOBUFS; 2876 hlen = LL_RESERVED_SPACE(dev); 2877 tlen = dev->needed_tailroom; 2878 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, 2879 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len), 2880 msg->msg_flags & MSG_DONTWAIT, &err); 2881 if (skb == NULL) 2882 goto out_unlock; 2883 2884 skb_set_network_header(skb, reserve); 2885 2886 err = -EINVAL; 2887 if (sock->type == SOCK_DGRAM) { 2888 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 2889 if (unlikely(offset < 0)) 2890 goto out_free; 2891 } 2892 2893 /* Returns -EFAULT on error */ 2894 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 2895 if (err) 2896 goto out_free; 2897 2898 if (sock->type == SOCK_RAW && 2899 !dev_validate_header(dev, skb->data, len)) { 2900 err = -EINVAL; 2901 goto out_free; 2902 } 2903 2904 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags); 2905 2906 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 2907 !packet_extra_vlan_len_allowed(dev, skb)) { 2908 err = -EMSGSIZE; 2909 goto out_free; 2910 } 2911 2912 skb->protocol = proto; 2913 skb->dev = dev; 2914 skb->priority = sk->sk_priority; 2915 skb->mark = sockc.mark; 2916 2917 packet_pick_tx_queue(dev, skb); 2918 2919 if (po->has_vnet_hdr) { 2920 err = packet_snd_vnet_gso(skb, &vnet_hdr); 2921 if (err) 2922 goto out_free; 2923 len += sizeof(vnet_hdr); 2924 } 2925 2926 skb_probe_transport_header(skb, reserve); 2927 2928 if (unlikely(extra_len == 4)) 2929 skb->no_fcs = 1; 2930 2931 err = po->xmit(skb); 2932 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2933 goto out_unlock; 2934 2935 dev_put(dev); 2936 2937 return len; 2938 2939 out_free: 2940 kfree_skb(skb); 2941 out_unlock: 2942 if (dev) 2943 dev_put(dev); 2944 out: 2945 return err; 2946 } 2947 2948 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2949 { 2950 struct sock *sk = sock->sk; 2951 struct packet_sock *po = pkt_sk(sk); 2952 2953 if (po->tx_ring.pg_vec) 2954 return tpacket_snd(po, msg); 2955 else 2956 return packet_snd(sock, msg, len); 2957 } 2958 2959 /* 2960 * Close a PACKET socket. This is fairly simple. We immediately go 2961 * to 'closed' state and remove our protocol entry in the device list. 2962 */ 2963 2964 static int packet_release(struct socket *sock) 2965 { 2966 struct sock *sk = sock->sk; 2967 struct packet_sock *po; 2968 struct net *net; 2969 union tpacket_req_u req_u; 2970 2971 if (!sk) 2972 return 0; 2973 2974 net = sock_net(sk); 2975 po = pkt_sk(sk); 2976 2977 mutex_lock(&net->packet.sklist_lock); 2978 sk_del_node_init_rcu(sk); 2979 mutex_unlock(&net->packet.sklist_lock); 2980 2981 preempt_disable(); 2982 sock_prot_inuse_add(net, sk->sk_prot, -1); 2983 preempt_enable(); 2984 2985 spin_lock(&po->bind_lock); 2986 unregister_prot_hook(sk, false); 2987 packet_cached_dev_reset(po); 2988 2989 if (po->prot_hook.dev) { 2990 dev_put(po->prot_hook.dev); 2991 po->prot_hook.dev = NULL; 2992 } 2993 spin_unlock(&po->bind_lock); 2994 2995 packet_flush_mclist(sk); 2996 2997 if (po->rx_ring.pg_vec) { 2998 memset(&req_u, 0, sizeof(req_u)); 2999 packet_set_ring(sk, &req_u, 1, 0); 3000 } 3001 3002 if (po->tx_ring.pg_vec) { 3003 memset(&req_u, 0, sizeof(req_u)); 3004 packet_set_ring(sk, &req_u, 1, 1); 3005 } 3006 3007 fanout_release(sk); 3008 3009 synchronize_net(); 3010 /* 3011 * Now the socket is dead. No more input will appear. 3012 */ 3013 sock_orphan(sk); 3014 sock->sk = NULL; 3015 3016 /* Purge queues */ 3017 3018 skb_queue_purge(&sk->sk_receive_queue); 3019 packet_free_pending(po); 3020 sk_refcnt_debug_release(sk); 3021 3022 sock_put(sk); 3023 return 0; 3024 } 3025 3026 /* 3027 * Attach a packet hook. 3028 */ 3029 3030 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3031 __be16 proto) 3032 { 3033 struct packet_sock *po = pkt_sk(sk); 3034 struct net_device *dev_curr; 3035 __be16 proto_curr; 3036 bool need_rehook; 3037 struct net_device *dev = NULL; 3038 int ret = 0; 3039 bool unlisted = false; 3040 3041 if (po->fanout) 3042 return -EINVAL; 3043 3044 lock_sock(sk); 3045 spin_lock(&po->bind_lock); 3046 rcu_read_lock(); 3047 3048 if (name) { 3049 dev = dev_get_by_name_rcu(sock_net(sk), name); 3050 if (!dev) { 3051 ret = -ENODEV; 3052 goto out_unlock; 3053 } 3054 } else if (ifindex) { 3055 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3056 if (!dev) { 3057 ret = -ENODEV; 3058 goto out_unlock; 3059 } 3060 } 3061 3062 if (dev) 3063 dev_hold(dev); 3064 3065 proto_curr = po->prot_hook.type; 3066 dev_curr = po->prot_hook.dev; 3067 3068 need_rehook = proto_curr != proto || dev_curr != dev; 3069 3070 if (need_rehook) { 3071 if (po->running) { 3072 rcu_read_unlock(); 3073 __unregister_prot_hook(sk, true); 3074 rcu_read_lock(); 3075 dev_curr = po->prot_hook.dev; 3076 if (dev) 3077 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3078 dev->ifindex); 3079 } 3080 3081 po->num = proto; 3082 po->prot_hook.type = proto; 3083 3084 if (unlikely(unlisted)) { 3085 dev_put(dev); 3086 po->prot_hook.dev = NULL; 3087 po->ifindex = -1; 3088 packet_cached_dev_reset(po); 3089 } else { 3090 po->prot_hook.dev = dev; 3091 po->ifindex = dev ? dev->ifindex : 0; 3092 packet_cached_dev_assign(po, dev); 3093 } 3094 } 3095 if (dev_curr) 3096 dev_put(dev_curr); 3097 3098 if (proto == 0 || !need_rehook) 3099 goto out_unlock; 3100 3101 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3102 register_prot_hook(sk); 3103 } else { 3104 sk->sk_err = ENETDOWN; 3105 if (!sock_flag(sk, SOCK_DEAD)) 3106 sk->sk_error_report(sk); 3107 } 3108 3109 out_unlock: 3110 rcu_read_unlock(); 3111 spin_unlock(&po->bind_lock); 3112 release_sock(sk); 3113 return ret; 3114 } 3115 3116 /* 3117 * Bind a packet socket to a device 3118 */ 3119 3120 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3121 int addr_len) 3122 { 3123 struct sock *sk = sock->sk; 3124 char name[15]; 3125 3126 /* 3127 * Check legality 3128 */ 3129 3130 if (addr_len != sizeof(struct sockaddr)) 3131 return -EINVAL; 3132 strlcpy(name, uaddr->sa_data, sizeof(name)); 3133 3134 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3135 } 3136 3137 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3138 { 3139 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3140 struct sock *sk = sock->sk; 3141 3142 /* 3143 * Check legality 3144 */ 3145 3146 if (addr_len < sizeof(struct sockaddr_ll)) 3147 return -EINVAL; 3148 if (sll->sll_family != AF_PACKET) 3149 return -EINVAL; 3150 3151 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3152 sll->sll_protocol ? : pkt_sk(sk)->num); 3153 } 3154 3155 static struct proto packet_proto = { 3156 .name = "PACKET", 3157 .owner = THIS_MODULE, 3158 .obj_size = sizeof(struct packet_sock), 3159 }; 3160 3161 /* 3162 * Create a packet of type SOCK_PACKET. 3163 */ 3164 3165 static int packet_create(struct net *net, struct socket *sock, int protocol, 3166 int kern) 3167 { 3168 struct sock *sk; 3169 struct packet_sock *po; 3170 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3171 int err; 3172 3173 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3174 return -EPERM; 3175 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3176 sock->type != SOCK_PACKET) 3177 return -ESOCKTNOSUPPORT; 3178 3179 sock->state = SS_UNCONNECTED; 3180 3181 err = -ENOBUFS; 3182 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3183 if (sk == NULL) 3184 goto out; 3185 3186 sock->ops = &packet_ops; 3187 if (sock->type == SOCK_PACKET) 3188 sock->ops = &packet_ops_spkt; 3189 3190 sock_init_data(sock, sk); 3191 3192 po = pkt_sk(sk); 3193 sk->sk_family = PF_PACKET; 3194 po->num = proto; 3195 po->xmit = dev_queue_xmit; 3196 3197 err = packet_alloc_pending(po); 3198 if (err) 3199 goto out2; 3200 3201 packet_cached_dev_reset(po); 3202 3203 sk->sk_destruct = packet_sock_destruct; 3204 sk_refcnt_debug_inc(sk); 3205 3206 /* 3207 * Attach a protocol block 3208 */ 3209 3210 spin_lock_init(&po->bind_lock); 3211 mutex_init(&po->pg_vec_lock); 3212 po->rollover = NULL; 3213 po->prot_hook.func = packet_rcv; 3214 3215 if (sock->type == SOCK_PACKET) 3216 po->prot_hook.func = packet_rcv_spkt; 3217 3218 po->prot_hook.af_packet_priv = sk; 3219 3220 if (proto) { 3221 po->prot_hook.type = proto; 3222 register_prot_hook(sk); 3223 } 3224 3225 mutex_lock(&net->packet.sklist_lock); 3226 sk_add_node_rcu(sk, &net->packet.sklist); 3227 mutex_unlock(&net->packet.sklist_lock); 3228 3229 preempt_disable(); 3230 sock_prot_inuse_add(net, &packet_proto, 1); 3231 preempt_enable(); 3232 3233 return 0; 3234 out2: 3235 sk_free(sk); 3236 out: 3237 return err; 3238 } 3239 3240 /* 3241 * Pull a packet from our receive queue and hand it to the user. 3242 * If necessary we block. 3243 */ 3244 3245 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3246 int flags) 3247 { 3248 struct sock *sk = sock->sk; 3249 struct sk_buff *skb; 3250 int copied, err; 3251 int vnet_hdr_len = 0; 3252 unsigned int origlen = 0; 3253 3254 err = -EINVAL; 3255 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3256 goto out; 3257 3258 #if 0 3259 /* What error should we return now? EUNATTACH? */ 3260 if (pkt_sk(sk)->ifindex < 0) 3261 return -ENODEV; 3262 #endif 3263 3264 if (flags & MSG_ERRQUEUE) { 3265 err = sock_recv_errqueue(sk, msg, len, 3266 SOL_PACKET, PACKET_TX_TIMESTAMP); 3267 goto out; 3268 } 3269 3270 /* 3271 * Call the generic datagram receiver. This handles all sorts 3272 * of horrible races and re-entrancy so we can forget about it 3273 * in the protocol layers. 3274 * 3275 * Now it will return ENETDOWN, if device have just gone down, 3276 * but then it will block. 3277 */ 3278 3279 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 3280 3281 /* 3282 * An error occurred so return it. Because skb_recv_datagram() 3283 * handles the blocking we don't see and worry about blocking 3284 * retries. 3285 */ 3286 3287 if (skb == NULL) 3288 goto out; 3289 3290 if (pkt_sk(sk)->pressure) 3291 packet_rcv_has_room(pkt_sk(sk), NULL); 3292 3293 if (pkt_sk(sk)->has_vnet_hdr) { 3294 err = packet_rcv_vnet(msg, skb, &len); 3295 if (err) 3296 goto out_free; 3297 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3298 } 3299 3300 /* You lose any data beyond the buffer you gave. If it worries 3301 * a user program they can ask the device for its MTU 3302 * anyway. 3303 */ 3304 copied = skb->len; 3305 if (copied > len) { 3306 copied = len; 3307 msg->msg_flags |= MSG_TRUNC; 3308 } 3309 3310 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3311 if (err) 3312 goto out_free; 3313 3314 if (sock->type != SOCK_PACKET) { 3315 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3316 3317 /* Original length was stored in sockaddr_ll fields */ 3318 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3319 sll->sll_family = AF_PACKET; 3320 sll->sll_protocol = skb->protocol; 3321 } 3322 3323 sock_recv_ts_and_drops(msg, sk, skb); 3324 3325 if (msg->msg_name) { 3326 /* If the address length field is there to be filled 3327 * in, we fill it in now. 3328 */ 3329 if (sock->type == SOCK_PACKET) { 3330 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3331 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3332 } else { 3333 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3334 3335 msg->msg_namelen = sll->sll_halen + 3336 offsetof(struct sockaddr_ll, sll_addr); 3337 } 3338 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, 3339 msg->msg_namelen); 3340 } 3341 3342 if (pkt_sk(sk)->auxdata) { 3343 struct tpacket_auxdata aux; 3344 3345 aux.tp_status = TP_STATUS_USER; 3346 if (skb->ip_summed == CHECKSUM_PARTIAL) 3347 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3348 else if (skb->pkt_type != PACKET_OUTGOING && 3349 (skb->ip_summed == CHECKSUM_COMPLETE || 3350 skb_csum_unnecessary(skb))) 3351 aux.tp_status |= TP_STATUS_CSUM_VALID; 3352 3353 aux.tp_len = origlen; 3354 aux.tp_snaplen = skb->len; 3355 aux.tp_mac = 0; 3356 aux.tp_net = skb_network_offset(skb); 3357 if (skb_vlan_tag_present(skb)) { 3358 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3359 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3360 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3361 } else { 3362 aux.tp_vlan_tci = 0; 3363 aux.tp_vlan_tpid = 0; 3364 } 3365 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3366 } 3367 3368 /* 3369 * Free or return the buffer as appropriate. Again this 3370 * hides all the races and re-entrancy issues from us. 3371 */ 3372 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3373 3374 out_free: 3375 skb_free_datagram(sk, skb); 3376 out: 3377 return err; 3378 } 3379 3380 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3381 int *uaddr_len, int peer) 3382 { 3383 struct net_device *dev; 3384 struct sock *sk = sock->sk; 3385 3386 if (peer) 3387 return -EOPNOTSUPP; 3388 3389 uaddr->sa_family = AF_PACKET; 3390 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3391 rcu_read_lock(); 3392 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3393 if (dev) 3394 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3395 rcu_read_unlock(); 3396 *uaddr_len = sizeof(*uaddr); 3397 3398 return 0; 3399 } 3400 3401 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3402 int *uaddr_len, int peer) 3403 { 3404 struct net_device *dev; 3405 struct sock *sk = sock->sk; 3406 struct packet_sock *po = pkt_sk(sk); 3407 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3408 3409 if (peer) 3410 return -EOPNOTSUPP; 3411 3412 sll->sll_family = AF_PACKET; 3413 sll->sll_ifindex = po->ifindex; 3414 sll->sll_protocol = po->num; 3415 sll->sll_pkttype = 0; 3416 rcu_read_lock(); 3417 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 3418 if (dev) { 3419 sll->sll_hatype = dev->type; 3420 sll->sll_halen = dev->addr_len; 3421 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3422 } else { 3423 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3424 sll->sll_halen = 0; 3425 } 3426 rcu_read_unlock(); 3427 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3428 3429 return 0; 3430 } 3431 3432 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3433 int what) 3434 { 3435 switch (i->type) { 3436 case PACKET_MR_MULTICAST: 3437 if (i->alen != dev->addr_len) 3438 return -EINVAL; 3439 if (what > 0) 3440 return dev_mc_add(dev, i->addr); 3441 else 3442 return dev_mc_del(dev, i->addr); 3443 break; 3444 case PACKET_MR_PROMISC: 3445 return dev_set_promiscuity(dev, what); 3446 case PACKET_MR_ALLMULTI: 3447 return dev_set_allmulti(dev, what); 3448 case PACKET_MR_UNICAST: 3449 if (i->alen != dev->addr_len) 3450 return -EINVAL; 3451 if (what > 0) 3452 return dev_uc_add(dev, i->addr); 3453 else 3454 return dev_uc_del(dev, i->addr); 3455 break; 3456 default: 3457 break; 3458 } 3459 return 0; 3460 } 3461 3462 static void packet_dev_mclist_delete(struct net_device *dev, 3463 struct packet_mclist **mlp) 3464 { 3465 struct packet_mclist *ml; 3466 3467 while ((ml = *mlp) != NULL) { 3468 if (ml->ifindex == dev->ifindex) { 3469 packet_dev_mc(dev, ml, -1); 3470 *mlp = ml->next; 3471 kfree(ml); 3472 } else 3473 mlp = &ml->next; 3474 } 3475 } 3476 3477 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3478 { 3479 struct packet_sock *po = pkt_sk(sk); 3480 struct packet_mclist *ml, *i; 3481 struct net_device *dev; 3482 int err; 3483 3484 rtnl_lock(); 3485 3486 err = -ENODEV; 3487 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3488 if (!dev) 3489 goto done; 3490 3491 err = -EINVAL; 3492 if (mreq->mr_alen > dev->addr_len) 3493 goto done; 3494 3495 err = -ENOBUFS; 3496 i = kmalloc(sizeof(*i), GFP_KERNEL); 3497 if (i == NULL) 3498 goto done; 3499 3500 err = 0; 3501 for (ml = po->mclist; ml; ml = ml->next) { 3502 if (ml->ifindex == mreq->mr_ifindex && 3503 ml->type == mreq->mr_type && 3504 ml->alen == mreq->mr_alen && 3505 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3506 ml->count++; 3507 /* Free the new element ... */ 3508 kfree(i); 3509 goto done; 3510 } 3511 } 3512 3513 i->type = mreq->mr_type; 3514 i->ifindex = mreq->mr_ifindex; 3515 i->alen = mreq->mr_alen; 3516 memcpy(i->addr, mreq->mr_address, i->alen); 3517 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3518 i->count = 1; 3519 i->next = po->mclist; 3520 po->mclist = i; 3521 err = packet_dev_mc(dev, i, 1); 3522 if (err) { 3523 po->mclist = i->next; 3524 kfree(i); 3525 } 3526 3527 done: 3528 rtnl_unlock(); 3529 return err; 3530 } 3531 3532 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3533 { 3534 struct packet_mclist *ml, **mlp; 3535 3536 rtnl_lock(); 3537 3538 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3539 if (ml->ifindex == mreq->mr_ifindex && 3540 ml->type == mreq->mr_type && 3541 ml->alen == mreq->mr_alen && 3542 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3543 if (--ml->count == 0) { 3544 struct net_device *dev; 3545 *mlp = ml->next; 3546 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3547 if (dev) 3548 packet_dev_mc(dev, ml, -1); 3549 kfree(ml); 3550 } 3551 break; 3552 } 3553 } 3554 rtnl_unlock(); 3555 return 0; 3556 } 3557 3558 static void packet_flush_mclist(struct sock *sk) 3559 { 3560 struct packet_sock *po = pkt_sk(sk); 3561 struct packet_mclist *ml; 3562 3563 if (!po->mclist) 3564 return; 3565 3566 rtnl_lock(); 3567 while ((ml = po->mclist) != NULL) { 3568 struct net_device *dev; 3569 3570 po->mclist = ml->next; 3571 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3572 if (dev != NULL) 3573 packet_dev_mc(dev, ml, -1); 3574 kfree(ml); 3575 } 3576 rtnl_unlock(); 3577 } 3578 3579 static int 3580 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3581 { 3582 struct sock *sk = sock->sk; 3583 struct packet_sock *po = pkt_sk(sk); 3584 int ret; 3585 3586 if (level != SOL_PACKET) 3587 return -ENOPROTOOPT; 3588 3589 switch (optname) { 3590 case PACKET_ADD_MEMBERSHIP: 3591 case PACKET_DROP_MEMBERSHIP: 3592 { 3593 struct packet_mreq_max mreq; 3594 int len = optlen; 3595 memset(&mreq, 0, sizeof(mreq)); 3596 if (len < sizeof(struct packet_mreq)) 3597 return -EINVAL; 3598 if (len > sizeof(mreq)) 3599 len = sizeof(mreq); 3600 if (copy_from_user(&mreq, optval, len)) 3601 return -EFAULT; 3602 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3603 return -EINVAL; 3604 if (optname == PACKET_ADD_MEMBERSHIP) 3605 ret = packet_mc_add(sk, &mreq); 3606 else 3607 ret = packet_mc_drop(sk, &mreq); 3608 return ret; 3609 } 3610 3611 case PACKET_RX_RING: 3612 case PACKET_TX_RING: 3613 { 3614 union tpacket_req_u req_u; 3615 int len; 3616 3617 switch (po->tp_version) { 3618 case TPACKET_V1: 3619 case TPACKET_V2: 3620 len = sizeof(req_u.req); 3621 break; 3622 case TPACKET_V3: 3623 default: 3624 len = sizeof(req_u.req3); 3625 break; 3626 } 3627 if (optlen < len) 3628 return -EINVAL; 3629 if (copy_from_user(&req_u.req, optval, len)) 3630 return -EFAULT; 3631 return packet_set_ring(sk, &req_u, 0, 3632 optname == PACKET_TX_RING); 3633 } 3634 case PACKET_COPY_THRESH: 3635 { 3636 int val; 3637 3638 if (optlen != sizeof(val)) 3639 return -EINVAL; 3640 if (copy_from_user(&val, optval, sizeof(val))) 3641 return -EFAULT; 3642 3643 pkt_sk(sk)->copy_thresh = val; 3644 return 0; 3645 } 3646 case PACKET_VERSION: 3647 { 3648 int val; 3649 3650 if (optlen != sizeof(val)) 3651 return -EINVAL; 3652 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3653 return -EBUSY; 3654 if (copy_from_user(&val, optval, sizeof(val))) 3655 return -EFAULT; 3656 switch (val) { 3657 case TPACKET_V1: 3658 case TPACKET_V2: 3659 case TPACKET_V3: 3660 po->tp_version = val; 3661 return 0; 3662 default: 3663 return -EINVAL; 3664 } 3665 } 3666 case PACKET_RESERVE: 3667 { 3668 unsigned int val; 3669 3670 if (optlen != sizeof(val)) 3671 return -EINVAL; 3672 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3673 return -EBUSY; 3674 if (copy_from_user(&val, optval, sizeof(val))) 3675 return -EFAULT; 3676 po->tp_reserve = val; 3677 return 0; 3678 } 3679 case PACKET_LOSS: 3680 { 3681 unsigned int val; 3682 3683 if (optlen != sizeof(val)) 3684 return -EINVAL; 3685 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3686 return -EBUSY; 3687 if (copy_from_user(&val, optval, sizeof(val))) 3688 return -EFAULT; 3689 po->tp_loss = !!val; 3690 return 0; 3691 } 3692 case PACKET_AUXDATA: 3693 { 3694 int val; 3695 3696 if (optlen < sizeof(val)) 3697 return -EINVAL; 3698 if (copy_from_user(&val, optval, sizeof(val))) 3699 return -EFAULT; 3700 3701 po->auxdata = !!val; 3702 return 0; 3703 } 3704 case PACKET_ORIGDEV: 3705 { 3706 int val; 3707 3708 if (optlen < sizeof(val)) 3709 return -EINVAL; 3710 if (copy_from_user(&val, optval, sizeof(val))) 3711 return -EFAULT; 3712 3713 po->origdev = !!val; 3714 return 0; 3715 } 3716 case PACKET_VNET_HDR: 3717 { 3718 int val; 3719 3720 if (sock->type != SOCK_RAW) 3721 return -EINVAL; 3722 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3723 return -EBUSY; 3724 if (optlen < sizeof(val)) 3725 return -EINVAL; 3726 if (copy_from_user(&val, optval, sizeof(val))) 3727 return -EFAULT; 3728 3729 po->has_vnet_hdr = !!val; 3730 return 0; 3731 } 3732 case PACKET_TIMESTAMP: 3733 { 3734 int val; 3735 3736 if (optlen != sizeof(val)) 3737 return -EINVAL; 3738 if (copy_from_user(&val, optval, sizeof(val))) 3739 return -EFAULT; 3740 3741 po->tp_tstamp = val; 3742 return 0; 3743 } 3744 case PACKET_FANOUT: 3745 { 3746 int val; 3747 3748 if (optlen != sizeof(val)) 3749 return -EINVAL; 3750 if (copy_from_user(&val, optval, sizeof(val))) 3751 return -EFAULT; 3752 3753 return fanout_add(sk, val & 0xffff, val >> 16); 3754 } 3755 case PACKET_FANOUT_DATA: 3756 { 3757 if (!po->fanout) 3758 return -EINVAL; 3759 3760 return fanout_set_data(po, optval, optlen); 3761 } 3762 case PACKET_TX_HAS_OFF: 3763 { 3764 unsigned int val; 3765 3766 if (optlen != sizeof(val)) 3767 return -EINVAL; 3768 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) 3769 return -EBUSY; 3770 if (copy_from_user(&val, optval, sizeof(val))) 3771 return -EFAULT; 3772 po->tp_tx_has_off = !!val; 3773 return 0; 3774 } 3775 case PACKET_QDISC_BYPASS: 3776 { 3777 int val; 3778 3779 if (optlen != sizeof(val)) 3780 return -EINVAL; 3781 if (copy_from_user(&val, optval, sizeof(val))) 3782 return -EFAULT; 3783 3784 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; 3785 return 0; 3786 } 3787 default: 3788 return -ENOPROTOOPT; 3789 } 3790 } 3791 3792 static int packet_getsockopt(struct socket *sock, int level, int optname, 3793 char __user *optval, int __user *optlen) 3794 { 3795 int len; 3796 int val, lv = sizeof(val); 3797 struct sock *sk = sock->sk; 3798 struct packet_sock *po = pkt_sk(sk); 3799 void *data = &val; 3800 union tpacket_stats_u st; 3801 struct tpacket_rollover_stats rstats; 3802 3803 if (level != SOL_PACKET) 3804 return -ENOPROTOOPT; 3805 3806 if (get_user(len, optlen)) 3807 return -EFAULT; 3808 3809 if (len < 0) 3810 return -EINVAL; 3811 3812 switch (optname) { 3813 case PACKET_STATISTICS: 3814 spin_lock_bh(&sk->sk_receive_queue.lock); 3815 memcpy(&st, &po->stats, sizeof(st)); 3816 memset(&po->stats, 0, sizeof(po->stats)); 3817 spin_unlock_bh(&sk->sk_receive_queue.lock); 3818 3819 if (po->tp_version == TPACKET_V3) { 3820 lv = sizeof(struct tpacket_stats_v3); 3821 st.stats3.tp_packets += st.stats3.tp_drops; 3822 data = &st.stats3; 3823 } else { 3824 lv = sizeof(struct tpacket_stats); 3825 st.stats1.tp_packets += st.stats1.tp_drops; 3826 data = &st.stats1; 3827 } 3828 3829 break; 3830 case PACKET_AUXDATA: 3831 val = po->auxdata; 3832 break; 3833 case PACKET_ORIGDEV: 3834 val = po->origdev; 3835 break; 3836 case PACKET_VNET_HDR: 3837 val = po->has_vnet_hdr; 3838 break; 3839 case PACKET_VERSION: 3840 val = po->tp_version; 3841 break; 3842 case PACKET_HDRLEN: 3843 if (len > sizeof(int)) 3844 len = sizeof(int); 3845 if (copy_from_user(&val, optval, len)) 3846 return -EFAULT; 3847 switch (val) { 3848 case TPACKET_V1: 3849 val = sizeof(struct tpacket_hdr); 3850 break; 3851 case TPACKET_V2: 3852 val = sizeof(struct tpacket2_hdr); 3853 break; 3854 case TPACKET_V3: 3855 val = sizeof(struct tpacket3_hdr); 3856 break; 3857 default: 3858 return -EINVAL; 3859 } 3860 break; 3861 case PACKET_RESERVE: 3862 val = po->tp_reserve; 3863 break; 3864 case PACKET_LOSS: 3865 val = po->tp_loss; 3866 break; 3867 case PACKET_TIMESTAMP: 3868 val = po->tp_tstamp; 3869 break; 3870 case PACKET_FANOUT: 3871 val = (po->fanout ? 3872 ((u32)po->fanout->id | 3873 ((u32)po->fanout->type << 16) | 3874 ((u32)po->fanout->flags << 24)) : 3875 0); 3876 break; 3877 case PACKET_ROLLOVER_STATS: 3878 if (!po->rollover) 3879 return -EINVAL; 3880 rstats.tp_all = atomic_long_read(&po->rollover->num); 3881 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 3882 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 3883 data = &rstats; 3884 lv = sizeof(rstats); 3885 break; 3886 case PACKET_TX_HAS_OFF: 3887 val = po->tp_tx_has_off; 3888 break; 3889 case PACKET_QDISC_BYPASS: 3890 val = packet_use_direct_xmit(po); 3891 break; 3892 default: 3893 return -ENOPROTOOPT; 3894 } 3895 3896 if (len > lv) 3897 len = lv; 3898 if (put_user(len, optlen)) 3899 return -EFAULT; 3900 if (copy_to_user(optval, data, len)) 3901 return -EFAULT; 3902 return 0; 3903 } 3904 3905 3906 #ifdef CONFIG_COMPAT 3907 static int compat_packet_setsockopt(struct socket *sock, int level, int optname, 3908 char __user *optval, unsigned int optlen) 3909 { 3910 struct packet_sock *po = pkt_sk(sock->sk); 3911 3912 if (level != SOL_PACKET) 3913 return -ENOPROTOOPT; 3914 3915 if (optname == PACKET_FANOUT_DATA && 3916 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) { 3917 optval = (char __user *)get_compat_bpf_fprog(optval); 3918 if (!optval) 3919 return -EFAULT; 3920 optlen = sizeof(struct sock_fprog); 3921 } 3922 3923 return packet_setsockopt(sock, level, optname, optval, optlen); 3924 } 3925 #endif 3926 3927 static int packet_notifier(struct notifier_block *this, 3928 unsigned long msg, void *ptr) 3929 { 3930 struct sock *sk; 3931 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 3932 struct net *net = dev_net(dev); 3933 3934 rcu_read_lock(); 3935 sk_for_each_rcu(sk, &net->packet.sklist) { 3936 struct packet_sock *po = pkt_sk(sk); 3937 3938 switch (msg) { 3939 case NETDEV_UNREGISTER: 3940 if (po->mclist) 3941 packet_dev_mclist_delete(dev, &po->mclist); 3942 /* fallthrough */ 3943 3944 case NETDEV_DOWN: 3945 if (dev->ifindex == po->ifindex) { 3946 spin_lock(&po->bind_lock); 3947 if (po->running) { 3948 __unregister_prot_hook(sk, false); 3949 sk->sk_err = ENETDOWN; 3950 if (!sock_flag(sk, SOCK_DEAD)) 3951 sk->sk_error_report(sk); 3952 } 3953 if (msg == NETDEV_UNREGISTER) { 3954 packet_cached_dev_reset(po); 3955 po->ifindex = -1; 3956 if (po->prot_hook.dev) 3957 dev_put(po->prot_hook.dev); 3958 po->prot_hook.dev = NULL; 3959 } 3960 spin_unlock(&po->bind_lock); 3961 } 3962 break; 3963 case NETDEV_UP: 3964 if (dev->ifindex == po->ifindex) { 3965 spin_lock(&po->bind_lock); 3966 if (po->num) 3967 register_prot_hook(sk); 3968 spin_unlock(&po->bind_lock); 3969 } 3970 break; 3971 } 3972 } 3973 rcu_read_unlock(); 3974 return NOTIFY_DONE; 3975 } 3976 3977 3978 static int packet_ioctl(struct socket *sock, unsigned int cmd, 3979 unsigned long arg) 3980 { 3981 struct sock *sk = sock->sk; 3982 3983 switch (cmd) { 3984 case SIOCOUTQ: 3985 { 3986 int amount = sk_wmem_alloc_get(sk); 3987 3988 return put_user(amount, (int __user *)arg); 3989 } 3990 case SIOCINQ: 3991 { 3992 struct sk_buff *skb; 3993 int amount = 0; 3994 3995 spin_lock_bh(&sk->sk_receive_queue.lock); 3996 skb = skb_peek(&sk->sk_receive_queue); 3997 if (skb) 3998 amount = skb->len; 3999 spin_unlock_bh(&sk->sk_receive_queue.lock); 4000 return put_user(amount, (int __user *)arg); 4001 } 4002 case SIOCGSTAMP: 4003 return sock_get_timestamp(sk, (struct timeval __user *)arg); 4004 case SIOCGSTAMPNS: 4005 return sock_get_timestampns(sk, (struct timespec __user *)arg); 4006 4007 #ifdef CONFIG_INET 4008 case SIOCADDRT: 4009 case SIOCDELRT: 4010 case SIOCDARP: 4011 case SIOCGARP: 4012 case SIOCSARP: 4013 case SIOCGIFADDR: 4014 case SIOCSIFADDR: 4015 case SIOCGIFBRDADDR: 4016 case SIOCSIFBRDADDR: 4017 case SIOCGIFNETMASK: 4018 case SIOCSIFNETMASK: 4019 case SIOCGIFDSTADDR: 4020 case SIOCSIFDSTADDR: 4021 case SIOCSIFFLAGS: 4022 return inet_dgram_ops.ioctl(sock, cmd, arg); 4023 #endif 4024 4025 default: 4026 return -ENOIOCTLCMD; 4027 } 4028 return 0; 4029 } 4030 4031 static unsigned int packet_poll(struct file *file, struct socket *sock, 4032 poll_table *wait) 4033 { 4034 struct sock *sk = sock->sk; 4035 struct packet_sock *po = pkt_sk(sk); 4036 unsigned int mask = datagram_poll(file, sock, wait); 4037 4038 spin_lock_bh(&sk->sk_receive_queue.lock); 4039 if (po->rx_ring.pg_vec) { 4040 if (!packet_previous_rx_frame(po, &po->rx_ring, 4041 TP_STATUS_KERNEL)) 4042 mask |= POLLIN | POLLRDNORM; 4043 } 4044 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 4045 po->pressure = 0; 4046 spin_unlock_bh(&sk->sk_receive_queue.lock); 4047 spin_lock_bh(&sk->sk_write_queue.lock); 4048 if (po->tx_ring.pg_vec) { 4049 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4050 mask |= POLLOUT | POLLWRNORM; 4051 } 4052 spin_unlock_bh(&sk->sk_write_queue.lock); 4053 return mask; 4054 } 4055 4056 4057 /* Dirty? Well, I still did not learn better way to account 4058 * for user mmaps. 4059 */ 4060 4061 static void packet_mm_open(struct vm_area_struct *vma) 4062 { 4063 struct file *file = vma->vm_file; 4064 struct socket *sock = file->private_data; 4065 struct sock *sk = sock->sk; 4066 4067 if (sk) 4068 atomic_inc(&pkt_sk(sk)->mapped); 4069 } 4070 4071 static void packet_mm_close(struct vm_area_struct *vma) 4072 { 4073 struct file *file = vma->vm_file; 4074 struct socket *sock = file->private_data; 4075 struct sock *sk = sock->sk; 4076 4077 if (sk) 4078 atomic_dec(&pkt_sk(sk)->mapped); 4079 } 4080 4081 static const struct vm_operations_struct packet_mmap_ops = { 4082 .open = packet_mm_open, 4083 .close = packet_mm_close, 4084 }; 4085 4086 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4087 unsigned int len) 4088 { 4089 int i; 4090 4091 for (i = 0; i < len; i++) { 4092 if (likely(pg_vec[i].buffer)) { 4093 if (is_vmalloc_addr(pg_vec[i].buffer)) 4094 vfree(pg_vec[i].buffer); 4095 else 4096 free_pages((unsigned long)pg_vec[i].buffer, 4097 order); 4098 pg_vec[i].buffer = NULL; 4099 } 4100 } 4101 kfree(pg_vec); 4102 } 4103 4104 static char *alloc_one_pg_vec_page(unsigned long order) 4105 { 4106 char *buffer; 4107 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4108 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4109 4110 buffer = (char *) __get_free_pages(gfp_flags, order); 4111 if (buffer) 4112 return buffer; 4113 4114 /* __get_free_pages failed, fall back to vmalloc */ 4115 buffer = vzalloc((1 << order) * PAGE_SIZE); 4116 if (buffer) 4117 return buffer; 4118 4119 /* vmalloc failed, lets dig into swap here */ 4120 gfp_flags &= ~__GFP_NORETRY; 4121 buffer = (char *) __get_free_pages(gfp_flags, order); 4122 if (buffer) 4123 return buffer; 4124 4125 /* complete and utter failure */ 4126 return NULL; 4127 } 4128 4129 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4130 { 4131 unsigned int block_nr = req->tp_block_nr; 4132 struct pgv *pg_vec; 4133 int i; 4134 4135 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL); 4136 if (unlikely(!pg_vec)) 4137 goto out; 4138 4139 for (i = 0; i < block_nr; i++) { 4140 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4141 if (unlikely(!pg_vec[i].buffer)) 4142 goto out_free_pgvec; 4143 } 4144 4145 out: 4146 return pg_vec; 4147 4148 out_free_pgvec: 4149 free_pg_vec(pg_vec, order, block_nr); 4150 pg_vec = NULL; 4151 goto out; 4152 } 4153 4154 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4155 int closing, int tx_ring) 4156 { 4157 struct pgv *pg_vec = NULL; 4158 struct packet_sock *po = pkt_sk(sk); 4159 int was_running, order = 0; 4160 struct packet_ring_buffer *rb; 4161 struct sk_buff_head *rb_queue; 4162 __be16 num; 4163 int err = -EINVAL; 4164 /* Added to avoid minimal code churn */ 4165 struct tpacket_req *req = &req_u->req; 4166 4167 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */ 4168 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) { 4169 net_warn_ratelimited("Tx-ring is not supported.\n"); 4170 goto out; 4171 } 4172 4173 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4174 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4175 4176 err = -EBUSY; 4177 if (!closing) { 4178 if (atomic_read(&po->mapped)) 4179 goto out; 4180 if (packet_read_pending(rb)) 4181 goto out; 4182 } 4183 4184 if (req->tp_block_nr) { 4185 /* Sanity tests and some calculations */ 4186 err = -EBUSY; 4187 if (unlikely(rb->pg_vec)) 4188 goto out; 4189 4190 switch (po->tp_version) { 4191 case TPACKET_V1: 4192 po->tp_hdrlen = TPACKET_HDRLEN; 4193 break; 4194 case TPACKET_V2: 4195 po->tp_hdrlen = TPACKET2_HDRLEN; 4196 break; 4197 case TPACKET_V3: 4198 po->tp_hdrlen = TPACKET3_HDRLEN; 4199 break; 4200 } 4201 4202 err = -EINVAL; 4203 if (unlikely((int)req->tp_block_size <= 0)) 4204 goto out; 4205 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4206 goto out; 4207 if (po->tp_version >= TPACKET_V3 && 4208 (int)(req->tp_block_size - 4209 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0) 4210 goto out; 4211 if (unlikely(req->tp_frame_size < po->tp_hdrlen + 4212 po->tp_reserve)) 4213 goto out; 4214 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4215 goto out; 4216 4217 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4218 if (unlikely(rb->frames_per_block == 0)) 4219 goto out; 4220 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4221 req->tp_frame_nr)) 4222 goto out; 4223 4224 err = -ENOMEM; 4225 order = get_order(req->tp_block_size); 4226 pg_vec = alloc_pg_vec(req, order); 4227 if (unlikely(!pg_vec)) 4228 goto out; 4229 switch (po->tp_version) { 4230 case TPACKET_V3: 4231 /* Transmit path is not supported. We checked 4232 * it above but just being paranoid 4233 */ 4234 if (!tx_ring) 4235 init_prb_bdqc(po, rb, pg_vec, req_u); 4236 break; 4237 default: 4238 break; 4239 } 4240 } 4241 /* Done */ 4242 else { 4243 err = -EINVAL; 4244 if (unlikely(req->tp_frame_nr)) 4245 goto out; 4246 } 4247 4248 lock_sock(sk); 4249 4250 /* Detach socket from network */ 4251 spin_lock(&po->bind_lock); 4252 was_running = po->running; 4253 num = po->num; 4254 if (was_running) { 4255 po->num = 0; 4256 __unregister_prot_hook(sk, false); 4257 } 4258 spin_unlock(&po->bind_lock); 4259 4260 synchronize_net(); 4261 4262 err = -EBUSY; 4263 mutex_lock(&po->pg_vec_lock); 4264 if (closing || atomic_read(&po->mapped) == 0) { 4265 err = 0; 4266 spin_lock_bh(&rb_queue->lock); 4267 swap(rb->pg_vec, pg_vec); 4268 rb->frame_max = (req->tp_frame_nr - 1); 4269 rb->head = 0; 4270 rb->frame_size = req->tp_frame_size; 4271 spin_unlock_bh(&rb_queue->lock); 4272 4273 swap(rb->pg_vec_order, order); 4274 swap(rb->pg_vec_len, req->tp_block_nr); 4275 4276 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4277 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4278 tpacket_rcv : packet_rcv; 4279 skb_queue_purge(rb_queue); 4280 if (atomic_read(&po->mapped)) 4281 pr_err("packet_mmap: vma is busy: %d\n", 4282 atomic_read(&po->mapped)); 4283 } 4284 mutex_unlock(&po->pg_vec_lock); 4285 4286 spin_lock(&po->bind_lock); 4287 if (was_running) { 4288 po->num = num; 4289 register_prot_hook(sk); 4290 } 4291 spin_unlock(&po->bind_lock); 4292 if (closing && (po->tp_version > TPACKET_V2)) { 4293 /* Because we don't support block-based V3 on tx-ring */ 4294 if (!tx_ring) 4295 prb_shutdown_retire_blk_timer(po, rb_queue); 4296 } 4297 release_sock(sk); 4298 4299 if (pg_vec) 4300 free_pg_vec(pg_vec, order, req->tp_block_nr); 4301 out: 4302 return err; 4303 } 4304 4305 static int packet_mmap(struct file *file, struct socket *sock, 4306 struct vm_area_struct *vma) 4307 { 4308 struct sock *sk = sock->sk; 4309 struct packet_sock *po = pkt_sk(sk); 4310 unsigned long size, expected_size; 4311 struct packet_ring_buffer *rb; 4312 unsigned long start; 4313 int err = -EINVAL; 4314 int i; 4315 4316 if (vma->vm_pgoff) 4317 return -EINVAL; 4318 4319 mutex_lock(&po->pg_vec_lock); 4320 4321 expected_size = 0; 4322 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4323 if (rb->pg_vec) { 4324 expected_size += rb->pg_vec_len 4325 * rb->pg_vec_pages 4326 * PAGE_SIZE; 4327 } 4328 } 4329 4330 if (expected_size == 0) 4331 goto out; 4332 4333 size = vma->vm_end - vma->vm_start; 4334 if (size != expected_size) 4335 goto out; 4336 4337 start = vma->vm_start; 4338 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4339 if (rb->pg_vec == NULL) 4340 continue; 4341 4342 for (i = 0; i < rb->pg_vec_len; i++) { 4343 struct page *page; 4344 void *kaddr = rb->pg_vec[i].buffer; 4345 int pg_num; 4346 4347 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4348 page = pgv_to_page(kaddr); 4349 err = vm_insert_page(vma, start, page); 4350 if (unlikely(err)) 4351 goto out; 4352 start += PAGE_SIZE; 4353 kaddr += PAGE_SIZE; 4354 } 4355 } 4356 } 4357 4358 atomic_inc(&po->mapped); 4359 vma->vm_ops = &packet_mmap_ops; 4360 err = 0; 4361 4362 out: 4363 mutex_unlock(&po->pg_vec_lock); 4364 return err; 4365 } 4366 4367 static const struct proto_ops packet_ops_spkt = { 4368 .family = PF_PACKET, 4369 .owner = THIS_MODULE, 4370 .release = packet_release, 4371 .bind = packet_bind_spkt, 4372 .connect = sock_no_connect, 4373 .socketpair = sock_no_socketpair, 4374 .accept = sock_no_accept, 4375 .getname = packet_getname_spkt, 4376 .poll = datagram_poll, 4377 .ioctl = packet_ioctl, 4378 .listen = sock_no_listen, 4379 .shutdown = sock_no_shutdown, 4380 .setsockopt = sock_no_setsockopt, 4381 .getsockopt = sock_no_getsockopt, 4382 .sendmsg = packet_sendmsg_spkt, 4383 .recvmsg = packet_recvmsg, 4384 .mmap = sock_no_mmap, 4385 .sendpage = sock_no_sendpage, 4386 }; 4387 4388 static const struct proto_ops packet_ops = { 4389 .family = PF_PACKET, 4390 .owner = THIS_MODULE, 4391 .release = packet_release, 4392 .bind = packet_bind, 4393 .connect = sock_no_connect, 4394 .socketpair = sock_no_socketpair, 4395 .accept = sock_no_accept, 4396 .getname = packet_getname, 4397 .poll = packet_poll, 4398 .ioctl = packet_ioctl, 4399 .listen = sock_no_listen, 4400 .shutdown = sock_no_shutdown, 4401 .setsockopt = packet_setsockopt, 4402 .getsockopt = packet_getsockopt, 4403 #ifdef CONFIG_COMPAT 4404 .compat_setsockopt = compat_packet_setsockopt, 4405 #endif 4406 .sendmsg = packet_sendmsg, 4407 .recvmsg = packet_recvmsg, 4408 .mmap = packet_mmap, 4409 .sendpage = sock_no_sendpage, 4410 }; 4411 4412 static const struct net_proto_family packet_family_ops = { 4413 .family = PF_PACKET, 4414 .create = packet_create, 4415 .owner = THIS_MODULE, 4416 }; 4417 4418 static struct notifier_block packet_netdev_notifier = { 4419 .notifier_call = packet_notifier, 4420 }; 4421 4422 #ifdef CONFIG_PROC_FS 4423 4424 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4425 __acquires(RCU) 4426 { 4427 struct net *net = seq_file_net(seq); 4428 4429 rcu_read_lock(); 4430 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4431 } 4432 4433 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4434 { 4435 struct net *net = seq_file_net(seq); 4436 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4437 } 4438 4439 static void packet_seq_stop(struct seq_file *seq, void *v) 4440 __releases(RCU) 4441 { 4442 rcu_read_unlock(); 4443 } 4444 4445 static int packet_seq_show(struct seq_file *seq, void *v) 4446 { 4447 if (v == SEQ_START_TOKEN) 4448 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4449 else { 4450 struct sock *s = sk_entry(v); 4451 const struct packet_sock *po = pkt_sk(s); 4452 4453 seq_printf(seq, 4454 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4455 s, 4456 atomic_read(&s->sk_refcnt), 4457 s->sk_type, 4458 ntohs(po->num), 4459 po->ifindex, 4460 po->running, 4461 atomic_read(&s->sk_rmem_alloc), 4462 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4463 sock_i_ino(s)); 4464 } 4465 4466 return 0; 4467 } 4468 4469 static const struct seq_operations packet_seq_ops = { 4470 .start = packet_seq_start, 4471 .next = packet_seq_next, 4472 .stop = packet_seq_stop, 4473 .show = packet_seq_show, 4474 }; 4475 4476 static int packet_seq_open(struct inode *inode, struct file *file) 4477 { 4478 return seq_open_net(inode, file, &packet_seq_ops, 4479 sizeof(struct seq_net_private)); 4480 } 4481 4482 static const struct file_operations packet_seq_fops = { 4483 .owner = THIS_MODULE, 4484 .open = packet_seq_open, 4485 .read = seq_read, 4486 .llseek = seq_lseek, 4487 .release = seq_release_net, 4488 }; 4489 4490 #endif 4491 4492 static int __net_init packet_net_init(struct net *net) 4493 { 4494 mutex_init(&net->packet.sklist_lock); 4495 INIT_HLIST_HEAD(&net->packet.sklist); 4496 4497 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops)) 4498 return -ENOMEM; 4499 4500 return 0; 4501 } 4502 4503 static void __net_exit packet_net_exit(struct net *net) 4504 { 4505 remove_proc_entry("packet", net->proc_net); 4506 } 4507 4508 static struct pernet_operations packet_net_ops = { 4509 .init = packet_net_init, 4510 .exit = packet_net_exit, 4511 }; 4512 4513 4514 static void __exit packet_exit(void) 4515 { 4516 unregister_netdevice_notifier(&packet_netdev_notifier); 4517 unregister_pernet_subsys(&packet_net_ops); 4518 sock_unregister(PF_PACKET); 4519 proto_unregister(&packet_proto); 4520 } 4521 4522 static int __init packet_init(void) 4523 { 4524 int rc = proto_register(&packet_proto, 0); 4525 4526 if (rc != 0) 4527 goto out; 4528 4529 sock_register(&packet_family_ops); 4530 register_pernet_subsys(&packet_net_ops); 4531 register_netdevice_notifier(&packet_netdev_notifier); 4532 out: 4533 return rc; 4534 } 4535 4536 module_init(packet_init); 4537 module_exit(packet_exit); 4538 MODULE_LICENSE("GPL"); 4539 MODULE_ALIAS_NETPROTO(PF_PACKET); 4540