1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * PACKET - implements raw packet sockets. 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Alan Cox, <gw4pts@gw4pts.ampr.org> 12 * 13 * Fixes: 14 * Alan Cox : verify_area() now used correctly 15 * Alan Cox : new skbuff lists, look ma no backlogs! 16 * Alan Cox : tidied skbuff lists. 17 * Alan Cox : Now uses generic datagram routines I 18 * added. Also fixed the peek/read crash 19 * from all old Linux datagram code. 20 * Alan Cox : Uses the improved datagram code. 21 * Alan Cox : Added NULL's for socket options. 22 * Alan Cox : Re-commented the code. 23 * Alan Cox : Use new kernel side addressing 24 * Rob Janssen : Correct MTU usage. 25 * Dave Platt : Counter leaks caused by incorrect 26 * interrupt locking and some slightly 27 * dubious gcc output. Can you read 28 * compiler: it said _VOLATILE_ 29 * Richard Kooijman : Timestamp fixes. 30 * Alan Cox : New buffers. Use sk->mac.raw. 31 * Alan Cox : sendmsg/recvmsg support. 32 * Alan Cox : Protocol setting support 33 * Alexey Kuznetsov : Untied from IPv4 stack. 34 * Cyrus Durgin : Fixed kerneld for kmod. 35 * Michal Ostrowski : Module initialization cleanup. 36 * Ulises Alonso : Frame number limit removal and 37 * packet_set_ring memory leak. 38 * Eric Biederman : Allow for > 8 byte hardware addresses. 39 * The convention is that longer addresses 40 * will simply extend the hardware address 41 * byte arrays at the end of sockaddr_ll 42 * and packet_mreq. 43 * Johann Baudy : Added TX RING. 44 * Chetan Loke : Implemented TPACKET_V3 block abstraction 45 * layer. 46 * Copyright (C) 2011, <lokec@ccs.neu.edu> 47 */ 48 49 #include <linux/types.h> 50 #include <linux/mm.h> 51 #include <linux/capability.h> 52 #include <linux/fcntl.h> 53 #include <linux/socket.h> 54 #include <linux/in.h> 55 #include <linux/inet.h> 56 #include <linux/netdevice.h> 57 #include <linux/if_packet.h> 58 #include <linux/wireless.h> 59 #include <linux/kernel.h> 60 #include <linux/kmod.h> 61 #include <linux/slab.h> 62 #include <linux/vmalloc.h> 63 #include <net/net_namespace.h> 64 #include <net/ip.h> 65 #include <net/protocol.h> 66 #include <linux/skbuff.h> 67 #include <net/sock.h> 68 #include <linux/errno.h> 69 #include <linux/timer.h> 70 #include <linux/uaccess.h> 71 #include <asm/ioctls.h> 72 #include <asm/page.h> 73 #include <asm/cacheflush.h> 74 #include <asm/io.h> 75 #include <linux/proc_fs.h> 76 #include <linux/seq_file.h> 77 #include <linux/poll.h> 78 #include <linux/module.h> 79 #include <linux/init.h> 80 #include <linux/mutex.h> 81 #include <linux/if_vlan.h> 82 #include <linux/virtio_net.h> 83 #include <linux/errqueue.h> 84 #include <linux/net_tstamp.h> 85 #include <linux/percpu.h> 86 #ifdef CONFIG_INET 87 #include <net/inet_common.h> 88 #endif 89 #include <linux/bpf.h> 90 #include <net/compat.h> 91 92 #include "internal.h" 93 94 /* 95 Assumptions: 96 - if device has no dev->hard_header routine, it adds and removes ll header 97 inside itself. In this case ll header is invisible outside of device, 98 but higher levels still should reserve dev->hard_header_len. 99 Some devices are enough clever to reallocate skb, when header 100 will not fit to reserved space (tunnel), another ones are silly 101 (PPP). 102 - packet socket receives packets with pulled ll header, 103 so that SOCK_RAW should push it back. 104 105 On receive: 106 ----------- 107 108 Incoming, dev->hard_header!=NULL 109 mac_header -> ll header 110 data -> data 111 112 Outgoing, dev->hard_header!=NULL 113 mac_header -> ll header 114 data -> ll header 115 116 Incoming, dev->hard_header==NULL 117 mac_header -> UNKNOWN position. It is very likely, that it points to ll 118 header. PPP makes it, that is wrong, because introduce 119 assymetry between rx and tx paths. 120 data -> data 121 122 Outgoing, dev->hard_header==NULL 123 mac_header -> data. ll header is still not built! 124 data -> data 125 126 Resume 127 If dev->hard_header==NULL we are unlikely to restore sensible ll header. 128 129 130 On transmit: 131 ------------ 132 133 dev->hard_header != NULL 134 mac_header -> ll header 135 data -> ll header 136 137 dev->hard_header == NULL (ll header is added by device, we cannot control it) 138 mac_header -> data 139 data -> data 140 141 We should set nh.raw on output to correct posistion, 142 packet classifier depends on it. 143 */ 144 145 /* Private packet socket structures. */ 146 147 /* identical to struct packet_mreq except it has 148 * a longer address field. 149 */ 150 struct packet_mreq_max { 151 int mr_ifindex; 152 unsigned short mr_type; 153 unsigned short mr_alen; 154 unsigned char mr_address[MAX_ADDR_LEN]; 155 }; 156 157 union tpacket_uhdr { 158 struct tpacket_hdr *h1; 159 struct tpacket2_hdr *h2; 160 struct tpacket3_hdr *h3; 161 void *raw; 162 }; 163 164 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 165 int closing, int tx_ring); 166 167 #define V3_ALIGNMENT (8) 168 169 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) 170 171 #define BLK_PLUS_PRIV(sz_of_priv) \ 172 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) 173 174 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) 175 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) 176 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) 177 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) 178 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) 179 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) 180 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x))) 181 182 struct packet_sock; 183 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 184 struct packet_type *pt, struct net_device *orig_dev); 185 186 static void *packet_previous_frame(struct packet_sock *po, 187 struct packet_ring_buffer *rb, 188 int status); 189 static void packet_increment_head(struct packet_ring_buffer *buff); 190 static int prb_curr_blk_in_use(struct tpacket_block_desc *); 191 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 192 struct packet_sock *); 193 static void prb_retire_current_block(struct tpacket_kbdq_core *, 194 struct packet_sock *, unsigned int status); 195 static int prb_queue_frozen(struct tpacket_kbdq_core *); 196 static void prb_open_block(struct tpacket_kbdq_core *, 197 struct tpacket_block_desc *); 198 static void prb_retire_rx_blk_timer_expired(struct timer_list *); 199 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); 200 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 201 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 202 struct tpacket3_hdr *); 203 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 204 struct tpacket3_hdr *); 205 static void packet_flush_mclist(struct sock *sk); 206 static u16 packet_pick_tx_queue(struct sk_buff *skb); 207 208 struct packet_skb_cb { 209 union { 210 struct sockaddr_pkt pkt; 211 union { 212 /* Trick: alias skb original length with 213 * ll.sll_family and ll.protocol in order 214 * to save room. 215 */ 216 unsigned int origlen; 217 struct sockaddr_ll ll; 218 }; 219 } sa; 220 }; 221 222 #define vio_le() virtio_legacy_is_little_endian() 223 224 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 225 226 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 227 #define GET_PBLOCK_DESC(x, bid) \ 228 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 229 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 230 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 231 #define GET_NEXT_PRB_BLK_NUM(x) \ 232 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 233 ((x)->kactive_blk_num+1) : 0) 234 235 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 236 static void __fanout_link(struct sock *sk, struct packet_sock *po); 237 238 static int packet_direct_xmit(struct sk_buff *skb) 239 { 240 return dev_direct_xmit(skb, packet_pick_tx_queue(skb)); 241 } 242 243 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 244 { 245 struct net_device *dev; 246 247 rcu_read_lock(); 248 dev = rcu_dereference(po->cached_dev); 249 if (likely(dev)) 250 dev_hold(dev); 251 rcu_read_unlock(); 252 253 return dev; 254 } 255 256 static void packet_cached_dev_assign(struct packet_sock *po, 257 struct net_device *dev) 258 { 259 rcu_assign_pointer(po->cached_dev, dev); 260 } 261 262 static void packet_cached_dev_reset(struct packet_sock *po) 263 { 264 RCU_INIT_POINTER(po->cached_dev, NULL); 265 } 266 267 static bool packet_use_direct_xmit(const struct packet_sock *po) 268 { 269 return po->xmit == packet_direct_xmit; 270 } 271 272 static u16 packet_pick_tx_queue(struct sk_buff *skb) 273 { 274 struct net_device *dev = skb->dev; 275 const struct net_device_ops *ops = dev->netdev_ops; 276 int cpu = raw_smp_processor_id(); 277 u16 queue_index; 278 279 #ifdef CONFIG_XPS 280 skb->sender_cpu = cpu + 1; 281 #endif 282 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues); 283 if (ops->ndo_select_queue) { 284 queue_index = ops->ndo_select_queue(dev, skb, NULL); 285 queue_index = netdev_cap_txqueue(dev, queue_index); 286 } else { 287 queue_index = netdev_pick_tx(dev, skb, NULL); 288 } 289 290 return queue_index; 291 } 292 293 /* __register_prot_hook must be invoked through register_prot_hook 294 * or from a context in which asynchronous accesses to the packet 295 * socket is not possible (packet_create()). 296 */ 297 static void __register_prot_hook(struct sock *sk) 298 { 299 struct packet_sock *po = pkt_sk(sk); 300 301 if (!po->running) { 302 if (po->fanout) 303 __fanout_link(sk, po); 304 else 305 dev_add_pack(&po->prot_hook); 306 307 sock_hold(sk); 308 po->running = 1; 309 } 310 } 311 312 static void register_prot_hook(struct sock *sk) 313 { 314 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); 315 __register_prot_hook(sk); 316 } 317 318 /* If the sync parameter is true, we will temporarily drop 319 * the po->bind_lock and do a synchronize_net to make sure no 320 * asynchronous packet processing paths still refer to the elements 321 * of po->prot_hook. If the sync parameter is false, it is the 322 * callers responsibility to take care of this. 323 */ 324 static void __unregister_prot_hook(struct sock *sk, bool sync) 325 { 326 struct packet_sock *po = pkt_sk(sk); 327 328 lockdep_assert_held_once(&po->bind_lock); 329 330 po->running = 0; 331 332 if (po->fanout) 333 __fanout_unlink(sk, po); 334 else 335 __dev_remove_pack(&po->prot_hook); 336 337 __sock_put(sk); 338 339 if (sync) { 340 spin_unlock(&po->bind_lock); 341 synchronize_net(); 342 spin_lock(&po->bind_lock); 343 } 344 } 345 346 static void unregister_prot_hook(struct sock *sk, bool sync) 347 { 348 struct packet_sock *po = pkt_sk(sk); 349 350 if (po->running) 351 __unregister_prot_hook(sk, sync); 352 } 353 354 static inline struct page * __pure pgv_to_page(void *addr) 355 { 356 if (is_vmalloc_addr(addr)) 357 return vmalloc_to_page(addr); 358 return virt_to_page(addr); 359 } 360 361 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 362 { 363 union tpacket_uhdr h; 364 365 h.raw = frame; 366 switch (po->tp_version) { 367 case TPACKET_V1: 368 h.h1->tp_status = status; 369 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 370 break; 371 case TPACKET_V2: 372 h.h2->tp_status = status; 373 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 374 break; 375 case TPACKET_V3: 376 h.h3->tp_status = status; 377 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 378 break; 379 default: 380 WARN(1, "TPACKET version not supported.\n"); 381 BUG(); 382 } 383 384 smp_wmb(); 385 } 386 387 static int __packet_get_status(struct packet_sock *po, void *frame) 388 { 389 union tpacket_uhdr h; 390 391 smp_rmb(); 392 393 h.raw = frame; 394 switch (po->tp_version) { 395 case TPACKET_V1: 396 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 397 return h.h1->tp_status; 398 case TPACKET_V2: 399 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 400 return h.h2->tp_status; 401 case TPACKET_V3: 402 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 403 return h.h3->tp_status; 404 default: 405 WARN(1, "TPACKET version not supported.\n"); 406 BUG(); 407 return 0; 408 } 409 } 410 411 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, 412 unsigned int flags) 413 { 414 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 415 416 if (shhwtstamps && 417 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 418 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) 419 return TP_STATUS_TS_RAW_HARDWARE; 420 421 if (ktime_to_timespec_cond(skb->tstamp, ts)) 422 return TP_STATUS_TS_SOFTWARE; 423 424 return 0; 425 } 426 427 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 428 struct sk_buff *skb) 429 { 430 union tpacket_uhdr h; 431 struct timespec ts; 432 __u32 ts_status; 433 434 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 435 return 0; 436 437 h.raw = frame; 438 switch (po->tp_version) { 439 case TPACKET_V1: 440 h.h1->tp_sec = ts.tv_sec; 441 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 442 break; 443 case TPACKET_V2: 444 h.h2->tp_sec = ts.tv_sec; 445 h.h2->tp_nsec = ts.tv_nsec; 446 break; 447 case TPACKET_V3: 448 h.h3->tp_sec = ts.tv_sec; 449 h.h3->tp_nsec = ts.tv_nsec; 450 break; 451 default: 452 WARN(1, "TPACKET version not supported.\n"); 453 BUG(); 454 } 455 456 /* one flush is safe, as both fields always lie on the same cacheline */ 457 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 458 smp_wmb(); 459 460 return ts_status; 461 } 462 463 static void *packet_lookup_frame(struct packet_sock *po, 464 struct packet_ring_buffer *rb, 465 unsigned int position, 466 int status) 467 { 468 unsigned int pg_vec_pos, frame_offset; 469 union tpacket_uhdr h; 470 471 pg_vec_pos = position / rb->frames_per_block; 472 frame_offset = position % rb->frames_per_block; 473 474 h.raw = rb->pg_vec[pg_vec_pos].buffer + 475 (frame_offset * rb->frame_size); 476 477 if (status != __packet_get_status(po, h.raw)) 478 return NULL; 479 480 return h.raw; 481 } 482 483 static void *packet_current_frame(struct packet_sock *po, 484 struct packet_ring_buffer *rb, 485 int status) 486 { 487 return packet_lookup_frame(po, rb, rb->head, status); 488 } 489 490 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 491 { 492 del_timer_sync(&pkc->retire_blk_timer); 493 } 494 495 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 496 struct sk_buff_head *rb_queue) 497 { 498 struct tpacket_kbdq_core *pkc; 499 500 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 501 502 spin_lock_bh(&rb_queue->lock); 503 pkc->delete_blk_timer = 1; 504 spin_unlock_bh(&rb_queue->lock); 505 506 prb_del_retire_blk_timer(pkc); 507 } 508 509 static void prb_setup_retire_blk_timer(struct packet_sock *po) 510 { 511 struct tpacket_kbdq_core *pkc; 512 513 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 514 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, 515 0); 516 pkc->retire_blk_timer.expires = jiffies; 517 } 518 519 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 520 int blk_size_in_bytes) 521 { 522 struct net_device *dev; 523 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; 524 struct ethtool_link_ksettings ecmd; 525 int err; 526 527 rtnl_lock(); 528 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 529 if (unlikely(!dev)) { 530 rtnl_unlock(); 531 return DEFAULT_PRB_RETIRE_TOV; 532 } 533 err = __ethtool_get_link_ksettings(dev, &ecmd); 534 rtnl_unlock(); 535 if (!err) { 536 /* 537 * If the link speed is so slow you don't really 538 * need to worry about perf anyways 539 */ 540 if (ecmd.base.speed < SPEED_1000 || 541 ecmd.base.speed == SPEED_UNKNOWN) { 542 return DEFAULT_PRB_RETIRE_TOV; 543 } else { 544 msec = 1; 545 div = ecmd.base.speed / 1000; 546 } 547 } 548 549 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 550 551 if (div) 552 mbits /= div; 553 554 tmo = mbits * msec; 555 556 if (div) 557 return tmo+1; 558 return tmo; 559 } 560 561 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 562 union tpacket_req_u *req_u) 563 { 564 p1->feature_req_word = req_u->req3.tp_feature_req_word; 565 } 566 567 static void init_prb_bdqc(struct packet_sock *po, 568 struct packet_ring_buffer *rb, 569 struct pgv *pg_vec, 570 union tpacket_req_u *req_u) 571 { 572 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 573 struct tpacket_block_desc *pbd; 574 575 memset(p1, 0x0, sizeof(*p1)); 576 577 p1->knxt_seq_num = 1; 578 p1->pkbdq = pg_vec; 579 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 580 p1->pkblk_start = pg_vec[0].buffer; 581 p1->kblk_size = req_u->req3.tp_block_size; 582 p1->knum_blocks = req_u->req3.tp_block_nr; 583 p1->hdrlen = po->tp_hdrlen; 584 p1->version = po->tp_version; 585 p1->last_kactive_blk_num = 0; 586 po->stats.stats3.tp_freeze_q_cnt = 0; 587 if (req_u->req3.tp_retire_blk_tov) 588 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 589 else 590 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 591 req_u->req3.tp_block_size); 592 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 593 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 594 595 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 596 prb_init_ft_ops(p1, req_u); 597 prb_setup_retire_blk_timer(po); 598 prb_open_block(p1, pbd); 599 } 600 601 /* Do NOT update the last_blk_num first. 602 * Assumes sk_buff_head lock is held. 603 */ 604 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 605 { 606 mod_timer(&pkc->retire_blk_timer, 607 jiffies + pkc->tov_in_jiffies); 608 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 609 } 610 611 /* 612 * Timer logic: 613 * 1) We refresh the timer only when we open a block. 614 * By doing this we don't waste cycles refreshing the timer 615 * on packet-by-packet basis. 616 * 617 * With a 1MB block-size, on a 1Gbps line, it will take 618 * i) ~8 ms to fill a block + ii) memcpy etc. 619 * In this cut we are not accounting for the memcpy time. 620 * 621 * So, if the user sets the 'tmo' to 10ms then the timer 622 * will never fire while the block is still getting filled 623 * (which is what we want). However, the user could choose 624 * to close a block early and that's fine. 625 * 626 * But when the timer does fire, we check whether or not to refresh it. 627 * Since the tmo granularity is in msecs, it is not too expensive 628 * to refresh the timer, lets say every '8' msecs. 629 * Either the user can set the 'tmo' or we can derive it based on 630 * a) line-speed and b) block-size. 631 * prb_calc_retire_blk_tmo() calculates the tmo. 632 * 633 */ 634 static void prb_retire_rx_blk_timer_expired(struct timer_list *t) 635 { 636 struct packet_sock *po = 637 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer); 638 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 639 unsigned int frozen; 640 struct tpacket_block_desc *pbd; 641 642 spin_lock(&po->sk.sk_receive_queue.lock); 643 644 frozen = prb_queue_frozen(pkc); 645 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 646 647 if (unlikely(pkc->delete_blk_timer)) 648 goto out; 649 650 /* We only need to plug the race when the block is partially filled. 651 * tpacket_rcv: 652 * lock(); increment BLOCK_NUM_PKTS; unlock() 653 * copy_bits() is in progress ... 654 * timer fires on other cpu: 655 * we can't retire the current block because copy_bits 656 * is in progress. 657 * 658 */ 659 if (BLOCK_NUM_PKTS(pbd)) { 660 while (atomic_read(&pkc->blk_fill_in_prog)) { 661 /* Waiting for skb_copy_bits to finish... */ 662 cpu_relax(); 663 } 664 } 665 666 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 667 if (!frozen) { 668 if (!BLOCK_NUM_PKTS(pbd)) { 669 /* An empty block. Just refresh the timer. */ 670 goto refresh_timer; 671 } 672 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 673 if (!prb_dispatch_next_block(pkc, po)) 674 goto refresh_timer; 675 else 676 goto out; 677 } else { 678 /* Case 1. Queue was frozen because user-space was 679 * lagging behind. 680 */ 681 if (prb_curr_blk_in_use(pbd)) { 682 /* 683 * Ok, user-space is still behind. 684 * So just refresh the timer. 685 */ 686 goto refresh_timer; 687 } else { 688 /* Case 2. queue was frozen,user-space caught up, 689 * now the link went idle && the timer fired. 690 * We don't have a block to close.So we open this 691 * block and restart the timer. 692 * opening a block thaws the queue,restarts timer 693 * Thawing/timer-refresh is a side effect. 694 */ 695 prb_open_block(pkc, pbd); 696 goto out; 697 } 698 } 699 } 700 701 refresh_timer: 702 _prb_refresh_rx_retire_blk_timer(pkc); 703 704 out: 705 spin_unlock(&po->sk.sk_receive_queue.lock); 706 } 707 708 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 709 struct tpacket_block_desc *pbd1, __u32 status) 710 { 711 /* Flush everything minus the block header */ 712 713 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 714 u8 *start, *end; 715 716 start = (u8 *)pbd1; 717 718 /* Skip the block header(we know header WILL fit in 4K) */ 719 start += PAGE_SIZE; 720 721 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 722 for (; start < end; start += PAGE_SIZE) 723 flush_dcache_page(pgv_to_page(start)); 724 725 smp_wmb(); 726 #endif 727 728 /* Now update the block status. */ 729 730 BLOCK_STATUS(pbd1) = status; 731 732 /* Flush the block header */ 733 734 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 735 start = (u8 *)pbd1; 736 flush_dcache_page(pgv_to_page(start)); 737 738 smp_wmb(); 739 #endif 740 } 741 742 /* 743 * Side effect: 744 * 745 * 1) flush the block 746 * 2) Increment active_blk_num 747 * 748 * Note:We DONT refresh the timer on purpose. 749 * Because almost always the next block will be opened. 750 */ 751 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 752 struct tpacket_block_desc *pbd1, 753 struct packet_sock *po, unsigned int stat) 754 { 755 __u32 status = TP_STATUS_USER | stat; 756 757 struct tpacket3_hdr *last_pkt; 758 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 759 struct sock *sk = &po->sk; 760 761 if (po->stats.stats3.tp_drops) 762 status |= TP_STATUS_LOSING; 763 764 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 765 last_pkt->tp_next_offset = 0; 766 767 /* Get the ts of the last pkt */ 768 if (BLOCK_NUM_PKTS(pbd1)) { 769 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 770 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 771 } else { 772 /* Ok, we tmo'd - so get the current time. 773 * 774 * It shouldn't really happen as we don't close empty 775 * blocks. See prb_retire_rx_blk_timer_expired(). 776 */ 777 struct timespec ts; 778 getnstimeofday(&ts); 779 h1->ts_last_pkt.ts_sec = ts.tv_sec; 780 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 781 } 782 783 smp_wmb(); 784 785 /* Flush the block */ 786 prb_flush_block(pkc1, pbd1, status); 787 788 sk->sk_data_ready(sk); 789 790 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 791 } 792 793 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 794 { 795 pkc->reset_pending_on_curr_blk = 0; 796 } 797 798 /* 799 * Side effect of opening a block: 800 * 801 * 1) prb_queue is thawed. 802 * 2) retire_blk_timer is refreshed. 803 * 804 */ 805 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 806 struct tpacket_block_desc *pbd1) 807 { 808 struct timespec ts; 809 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 810 811 smp_rmb(); 812 813 /* We could have just memset this but we will lose the 814 * flexibility of making the priv area sticky 815 */ 816 817 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 818 BLOCK_NUM_PKTS(pbd1) = 0; 819 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 820 821 getnstimeofday(&ts); 822 823 h1->ts_first_pkt.ts_sec = ts.tv_sec; 824 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 825 826 pkc1->pkblk_start = (char *)pbd1; 827 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 828 829 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 830 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 831 832 pbd1->version = pkc1->version; 833 pkc1->prev = pkc1->nxt_offset; 834 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 835 836 prb_thaw_queue(pkc1); 837 _prb_refresh_rx_retire_blk_timer(pkc1); 838 839 smp_wmb(); 840 } 841 842 /* 843 * Queue freeze logic: 844 * 1) Assume tp_block_nr = 8 blocks. 845 * 2) At time 't0', user opens Rx ring. 846 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 847 * 4) user-space is either sleeping or processing block '0'. 848 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 849 * it will close block-7,loop around and try to fill block '0'. 850 * call-flow: 851 * __packet_lookup_frame_in_block 852 * prb_retire_current_block() 853 * prb_dispatch_next_block() 854 * |->(BLOCK_STATUS == USER) evaluates to true 855 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 856 * 6) Now there are two cases: 857 * 6.1) Link goes idle right after the queue is frozen. 858 * But remember, the last open_block() refreshed the timer. 859 * When this timer expires,it will refresh itself so that we can 860 * re-open block-0 in near future. 861 * 6.2) Link is busy and keeps on receiving packets. This is a simple 862 * case and __packet_lookup_frame_in_block will check if block-0 863 * is free and can now be re-used. 864 */ 865 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 866 struct packet_sock *po) 867 { 868 pkc->reset_pending_on_curr_blk = 1; 869 po->stats.stats3.tp_freeze_q_cnt++; 870 } 871 872 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 873 874 /* 875 * If the next block is free then we will dispatch it 876 * and return a good offset. 877 * Else, we will freeze the queue. 878 * So, caller must check the return value. 879 */ 880 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 881 struct packet_sock *po) 882 { 883 struct tpacket_block_desc *pbd; 884 885 smp_rmb(); 886 887 /* 1. Get current block num */ 888 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 889 890 /* 2. If this block is currently in_use then freeze the queue */ 891 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 892 prb_freeze_queue(pkc, po); 893 return NULL; 894 } 895 896 /* 897 * 3. 898 * open this block and return the offset where the first packet 899 * needs to get stored. 900 */ 901 prb_open_block(pkc, pbd); 902 return (void *)pkc->nxt_offset; 903 } 904 905 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 906 struct packet_sock *po, unsigned int status) 907 { 908 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 909 910 /* retire/close the current block */ 911 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 912 /* 913 * Plug the case where copy_bits() is in progress on 914 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 915 * have space to copy the pkt in the current block and 916 * called prb_retire_current_block() 917 * 918 * We don't need to worry about the TMO case because 919 * the timer-handler already handled this case. 920 */ 921 if (!(status & TP_STATUS_BLK_TMO)) { 922 while (atomic_read(&pkc->blk_fill_in_prog)) { 923 /* Waiting for skb_copy_bits to finish... */ 924 cpu_relax(); 925 } 926 } 927 prb_close_block(pkc, pbd, po, status); 928 return; 929 } 930 } 931 932 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) 933 { 934 return TP_STATUS_USER & BLOCK_STATUS(pbd); 935 } 936 937 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 938 { 939 return pkc->reset_pending_on_curr_blk; 940 } 941 942 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 943 { 944 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 945 atomic_dec(&pkc->blk_fill_in_prog); 946 } 947 948 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 949 struct tpacket3_hdr *ppd) 950 { 951 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 952 } 953 954 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 955 struct tpacket3_hdr *ppd) 956 { 957 ppd->hv1.tp_rxhash = 0; 958 } 959 960 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 961 struct tpacket3_hdr *ppd) 962 { 963 if (skb_vlan_tag_present(pkc->skb)) { 964 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 965 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 966 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 967 } else { 968 ppd->hv1.tp_vlan_tci = 0; 969 ppd->hv1.tp_vlan_tpid = 0; 970 ppd->tp_status = TP_STATUS_AVAILABLE; 971 } 972 } 973 974 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 975 struct tpacket3_hdr *ppd) 976 { 977 ppd->hv1.tp_padding = 0; 978 prb_fill_vlan_info(pkc, ppd); 979 980 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 981 prb_fill_rxhash(pkc, ppd); 982 else 983 prb_clear_rxhash(pkc, ppd); 984 } 985 986 static void prb_fill_curr_block(char *curr, 987 struct tpacket_kbdq_core *pkc, 988 struct tpacket_block_desc *pbd, 989 unsigned int len) 990 { 991 struct tpacket3_hdr *ppd; 992 993 ppd = (struct tpacket3_hdr *)curr; 994 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 995 pkc->prev = curr; 996 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 997 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 998 BLOCK_NUM_PKTS(pbd) += 1; 999 atomic_inc(&pkc->blk_fill_in_prog); 1000 prb_run_all_ft_ops(pkc, ppd); 1001 } 1002 1003 /* Assumes caller has the sk->rx_queue.lock */ 1004 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1005 struct sk_buff *skb, 1006 int status, 1007 unsigned int len 1008 ) 1009 { 1010 struct tpacket_kbdq_core *pkc; 1011 struct tpacket_block_desc *pbd; 1012 char *curr, *end; 1013 1014 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1015 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1016 1017 /* Queue is frozen when user space is lagging behind */ 1018 if (prb_queue_frozen(pkc)) { 1019 /* 1020 * Check if that last block which caused the queue to freeze, 1021 * is still in_use by user-space. 1022 */ 1023 if (prb_curr_blk_in_use(pbd)) { 1024 /* Can't record this packet */ 1025 return NULL; 1026 } else { 1027 /* 1028 * Ok, the block was released by user-space. 1029 * Now let's open that block. 1030 * opening a block also thaws the queue. 1031 * Thawing is a side effect. 1032 */ 1033 prb_open_block(pkc, pbd); 1034 } 1035 } 1036 1037 smp_mb(); 1038 curr = pkc->nxt_offset; 1039 pkc->skb = skb; 1040 end = (char *)pbd + pkc->kblk_size; 1041 1042 /* first try the current block */ 1043 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1044 prb_fill_curr_block(curr, pkc, pbd, len); 1045 return (void *)curr; 1046 } 1047 1048 /* Ok, close the current block */ 1049 prb_retire_current_block(pkc, po, 0); 1050 1051 /* Now, try to dispatch the next block */ 1052 curr = (char *)prb_dispatch_next_block(pkc, po); 1053 if (curr) { 1054 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1055 prb_fill_curr_block(curr, pkc, pbd, len); 1056 return (void *)curr; 1057 } 1058 1059 /* 1060 * No free blocks are available.user_space hasn't caught up yet. 1061 * Queue was just frozen and now this packet will get dropped. 1062 */ 1063 return NULL; 1064 } 1065 1066 static void *packet_current_rx_frame(struct packet_sock *po, 1067 struct sk_buff *skb, 1068 int status, unsigned int len) 1069 { 1070 char *curr = NULL; 1071 switch (po->tp_version) { 1072 case TPACKET_V1: 1073 case TPACKET_V2: 1074 curr = packet_lookup_frame(po, &po->rx_ring, 1075 po->rx_ring.head, status); 1076 return curr; 1077 case TPACKET_V3: 1078 return __packet_lookup_frame_in_block(po, skb, status, len); 1079 default: 1080 WARN(1, "TPACKET version not supported\n"); 1081 BUG(); 1082 return NULL; 1083 } 1084 } 1085 1086 static void *prb_lookup_block(struct packet_sock *po, 1087 struct packet_ring_buffer *rb, 1088 unsigned int idx, 1089 int status) 1090 { 1091 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1092 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1093 1094 if (status != BLOCK_STATUS(pbd)) 1095 return NULL; 1096 return pbd; 1097 } 1098 1099 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1100 { 1101 unsigned int prev; 1102 if (rb->prb_bdqc.kactive_blk_num) 1103 prev = rb->prb_bdqc.kactive_blk_num-1; 1104 else 1105 prev = rb->prb_bdqc.knum_blocks-1; 1106 return prev; 1107 } 1108 1109 /* Assumes caller has held the rx_queue.lock */ 1110 static void *__prb_previous_block(struct packet_sock *po, 1111 struct packet_ring_buffer *rb, 1112 int status) 1113 { 1114 unsigned int previous = prb_previous_blk_num(rb); 1115 return prb_lookup_block(po, rb, previous, status); 1116 } 1117 1118 static void *packet_previous_rx_frame(struct packet_sock *po, 1119 struct packet_ring_buffer *rb, 1120 int status) 1121 { 1122 if (po->tp_version <= TPACKET_V2) 1123 return packet_previous_frame(po, rb, status); 1124 1125 return __prb_previous_block(po, rb, status); 1126 } 1127 1128 static void packet_increment_rx_head(struct packet_sock *po, 1129 struct packet_ring_buffer *rb) 1130 { 1131 switch (po->tp_version) { 1132 case TPACKET_V1: 1133 case TPACKET_V2: 1134 return packet_increment_head(rb); 1135 case TPACKET_V3: 1136 default: 1137 WARN(1, "TPACKET version not supported.\n"); 1138 BUG(); 1139 return; 1140 } 1141 } 1142 1143 static void *packet_previous_frame(struct packet_sock *po, 1144 struct packet_ring_buffer *rb, 1145 int status) 1146 { 1147 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1148 return packet_lookup_frame(po, rb, previous, status); 1149 } 1150 1151 static void packet_increment_head(struct packet_ring_buffer *buff) 1152 { 1153 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1154 } 1155 1156 static void packet_inc_pending(struct packet_ring_buffer *rb) 1157 { 1158 this_cpu_inc(*rb->pending_refcnt); 1159 } 1160 1161 static void packet_dec_pending(struct packet_ring_buffer *rb) 1162 { 1163 this_cpu_dec(*rb->pending_refcnt); 1164 } 1165 1166 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1167 { 1168 unsigned int refcnt = 0; 1169 int cpu; 1170 1171 /* We don't use pending refcount in rx_ring. */ 1172 if (rb->pending_refcnt == NULL) 1173 return 0; 1174 1175 for_each_possible_cpu(cpu) 1176 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1177 1178 return refcnt; 1179 } 1180 1181 static int packet_alloc_pending(struct packet_sock *po) 1182 { 1183 po->rx_ring.pending_refcnt = NULL; 1184 1185 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1186 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1187 return -ENOBUFS; 1188 1189 return 0; 1190 } 1191 1192 static void packet_free_pending(struct packet_sock *po) 1193 { 1194 free_percpu(po->tx_ring.pending_refcnt); 1195 } 1196 1197 #define ROOM_POW_OFF 2 1198 #define ROOM_NONE 0x0 1199 #define ROOM_LOW 0x1 1200 #define ROOM_NORMAL 0x2 1201 1202 static bool __tpacket_has_room(struct packet_sock *po, int pow_off) 1203 { 1204 int idx, len; 1205 1206 len = po->rx_ring.frame_max + 1; 1207 idx = po->rx_ring.head; 1208 if (pow_off) 1209 idx += len >> pow_off; 1210 if (idx >= len) 1211 idx -= len; 1212 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1213 } 1214 1215 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off) 1216 { 1217 int idx, len; 1218 1219 len = po->rx_ring.prb_bdqc.knum_blocks; 1220 idx = po->rx_ring.prb_bdqc.kactive_blk_num; 1221 if (pow_off) 1222 idx += len >> pow_off; 1223 if (idx >= len) 1224 idx -= len; 1225 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1226 } 1227 1228 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1229 { 1230 struct sock *sk = &po->sk; 1231 int ret = ROOM_NONE; 1232 1233 if (po->prot_hook.func != tpacket_rcv) { 1234 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1235 - (skb ? skb->truesize : 0); 1236 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF)) 1237 return ROOM_NORMAL; 1238 else if (avail > 0) 1239 return ROOM_LOW; 1240 else 1241 return ROOM_NONE; 1242 } 1243 1244 if (po->tp_version == TPACKET_V3) { 1245 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1246 ret = ROOM_NORMAL; 1247 else if (__tpacket_v3_has_room(po, 0)) 1248 ret = ROOM_LOW; 1249 } else { 1250 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1251 ret = ROOM_NORMAL; 1252 else if (__tpacket_has_room(po, 0)) 1253 ret = ROOM_LOW; 1254 } 1255 1256 return ret; 1257 } 1258 1259 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1260 { 1261 int ret; 1262 bool has_room; 1263 1264 spin_lock_bh(&po->sk.sk_receive_queue.lock); 1265 ret = __packet_rcv_has_room(po, skb); 1266 has_room = ret == ROOM_NORMAL; 1267 if (po->pressure == has_room) 1268 po->pressure = !has_room; 1269 spin_unlock_bh(&po->sk.sk_receive_queue.lock); 1270 1271 return ret; 1272 } 1273 1274 static void packet_sock_destruct(struct sock *sk) 1275 { 1276 skb_queue_purge(&sk->sk_error_queue); 1277 1278 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1279 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 1280 1281 if (!sock_flag(sk, SOCK_DEAD)) { 1282 pr_err("Attempt to release alive packet socket: %p\n", sk); 1283 return; 1284 } 1285 1286 sk_refcnt_debug_dec(sk); 1287 } 1288 1289 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1290 { 1291 u32 rxhash; 1292 int i, count = 0; 1293 1294 rxhash = skb_get_hash(skb); 1295 for (i = 0; i < ROLLOVER_HLEN; i++) 1296 if (po->rollover->history[i] == rxhash) 1297 count++; 1298 1299 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash; 1300 return count > (ROLLOVER_HLEN >> 1); 1301 } 1302 1303 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1304 struct sk_buff *skb, 1305 unsigned int num) 1306 { 1307 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1308 } 1309 1310 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1311 struct sk_buff *skb, 1312 unsigned int num) 1313 { 1314 unsigned int val = atomic_inc_return(&f->rr_cur); 1315 1316 return val % num; 1317 } 1318 1319 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1320 struct sk_buff *skb, 1321 unsigned int num) 1322 { 1323 return smp_processor_id() % num; 1324 } 1325 1326 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1327 struct sk_buff *skb, 1328 unsigned int num) 1329 { 1330 return prandom_u32_max(num); 1331 } 1332 1333 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1334 struct sk_buff *skb, 1335 unsigned int idx, bool try_self, 1336 unsigned int num) 1337 { 1338 struct packet_sock *po, *po_next, *po_skip = NULL; 1339 unsigned int i, j, room = ROOM_NONE; 1340 1341 po = pkt_sk(f->arr[idx]); 1342 1343 if (try_self) { 1344 room = packet_rcv_has_room(po, skb); 1345 if (room == ROOM_NORMAL || 1346 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1347 return idx; 1348 po_skip = po; 1349 } 1350 1351 i = j = min_t(int, po->rollover->sock, num - 1); 1352 do { 1353 po_next = pkt_sk(f->arr[i]); 1354 if (po_next != po_skip && !po_next->pressure && 1355 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1356 if (i != j) 1357 po->rollover->sock = i; 1358 atomic_long_inc(&po->rollover->num); 1359 if (room == ROOM_LOW) 1360 atomic_long_inc(&po->rollover->num_huge); 1361 return i; 1362 } 1363 1364 if (++i == num) 1365 i = 0; 1366 } while (i != j); 1367 1368 atomic_long_inc(&po->rollover->num_failed); 1369 return idx; 1370 } 1371 1372 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1373 struct sk_buff *skb, 1374 unsigned int num) 1375 { 1376 return skb_get_queue_mapping(skb) % num; 1377 } 1378 1379 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1380 struct sk_buff *skb, 1381 unsigned int num) 1382 { 1383 struct bpf_prog *prog; 1384 unsigned int ret = 0; 1385 1386 rcu_read_lock(); 1387 prog = rcu_dereference(f->bpf_prog); 1388 if (prog) 1389 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1390 rcu_read_unlock(); 1391 1392 return ret; 1393 } 1394 1395 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1396 { 1397 return f->flags & (flag >> 8); 1398 } 1399 1400 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1401 struct packet_type *pt, struct net_device *orig_dev) 1402 { 1403 struct packet_fanout *f = pt->af_packet_priv; 1404 unsigned int num = READ_ONCE(f->num_members); 1405 struct net *net = read_pnet(&f->net); 1406 struct packet_sock *po; 1407 unsigned int idx; 1408 1409 if (!net_eq(dev_net(dev), net) || !num) { 1410 kfree_skb(skb); 1411 return 0; 1412 } 1413 1414 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1415 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1416 if (!skb) 1417 return 0; 1418 } 1419 switch (f->type) { 1420 case PACKET_FANOUT_HASH: 1421 default: 1422 idx = fanout_demux_hash(f, skb, num); 1423 break; 1424 case PACKET_FANOUT_LB: 1425 idx = fanout_demux_lb(f, skb, num); 1426 break; 1427 case PACKET_FANOUT_CPU: 1428 idx = fanout_demux_cpu(f, skb, num); 1429 break; 1430 case PACKET_FANOUT_RND: 1431 idx = fanout_demux_rnd(f, skb, num); 1432 break; 1433 case PACKET_FANOUT_QM: 1434 idx = fanout_demux_qm(f, skb, num); 1435 break; 1436 case PACKET_FANOUT_ROLLOVER: 1437 idx = fanout_demux_rollover(f, skb, 0, false, num); 1438 break; 1439 case PACKET_FANOUT_CBPF: 1440 case PACKET_FANOUT_EBPF: 1441 idx = fanout_demux_bpf(f, skb, num); 1442 break; 1443 } 1444 1445 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1446 idx = fanout_demux_rollover(f, skb, idx, true, num); 1447 1448 po = pkt_sk(f->arr[idx]); 1449 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1450 } 1451 1452 DEFINE_MUTEX(fanout_mutex); 1453 EXPORT_SYMBOL_GPL(fanout_mutex); 1454 static LIST_HEAD(fanout_list); 1455 static u16 fanout_next_id; 1456 1457 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1458 { 1459 struct packet_fanout *f = po->fanout; 1460 1461 spin_lock(&f->lock); 1462 f->arr[f->num_members] = sk; 1463 smp_wmb(); 1464 f->num_members++; 1465 if (f->num_members == 1) 1466 dev_add_pack(&f->prot_hook); 1467 spin_unlock(&f->lock); 1468 } 1469 1470 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1471 { 1472 struct packet_fanout *f = po->fanout; 1473 int i; 1474 1475 spin_lock(&f->lock); 1476 for (i = 0; i < f->num_members; i++) { 1477 if (f->arr[i] == sk) 1478 break; 1479 } 1480 BUG_ON(i >= f->num_members); 1481 f->arr[i] = f->arr[f->num_members - 1]; 1482 f->num_members--; 1483 if (f->num_members == 0) 1484 __dev_remove_pack(&f->prot_hook); 1485 spin_unlock(&f->lock); 1486 } 1487 1488 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1489 { 1490 if (sk->sk_family != PF_PACKET) 1491 return false; 1492 1493 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1494 } 1495 1496 static void fanout_init_data(struct packet_fanout *f) 1497 { 1498 switch (f->type) { 1499 case PACKET_FANOUT_LB: 1500 atomic_set(&f->rr_cur, 0); 1501 break; 1502 case PACKET_FANOUT_CBPF: 1503 case PACKET_FANOUT_EBPF: 1504 RCU_INIT_POINTER(f->bpf_prog, NULL); 1505 break; 1506 } 1507 } 1508 1509 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1510 { 1511 struct bpf_prog *old; 1512 1513 spin_lock(&f->lock); 1514 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1515 rcu_assign_pointer(f->bpf_prog, new); 1516 spin_unlock(&f->lock); 1517 1518 if (old) { 1519 synchronize_net(); 1520 bpf_prog_destroy(old); 1521 } 1522 } 1523 1524 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data, 1525 unsigned int len) 1526 { 1527 struct bpf_prog *new; 1528 struct sock_fprog fprog; 1529 int ret; 1530 1531 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1532 return -EPERM; 1533 if (len != sizeof(fprog)) 1534 return -EINVAL; 1535 if (copy_from_user(&fprog, data, len)) 1536 return -EFAULT; 1537 1538 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1539 if (ret) 1540 return ret; 1541 1542 __fanout_set_data_bpf(po->fanout, new); 1543 return 0; 1544 } 1545 1546 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data, 1547 unsigned int len) 1548 { 1549 struct bpf_prog *new; 1550 u32 fd; 1551 1552 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1553 return -EPERM; 1554 if (len != sizeof(fd)) 1555 return -EINVAL; 1556 if (copy_from_user(&fd, data, len)) 1557 return -EFAULT; 1558 1559 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1560 if (IS_ERR(new)) 1561 return PTR_ERR(new); 1562 1563 __fanout_set_data_bpf(po->fanout, new); 1564 return 0; 1565 } 1566 1567 static int fanout_set_data(struct packet_sock *po, char __user *data, 1568 unsigned int len) 1569 { 1570 switch (po->fanout->type) { 1571 case PACKET_FANOUT_CBPF: 1572 return fanout_set_data_cbpf(po, data, len); 1573 case PACKET_FANOUT_EBPF: 1574 return fanout_set_data_ebpf(po, data, len); 1575 default: 1576 return -EINVAL; 1577 } 1578 } 1579 1580 static void fanout_release_data(struct packet_fanout *f) 1581 { 1582 switch (f->type) { 1583 case PACKET_FANOUT_CBPF: 1584 case PACKET_FANOUT_EBPF: 1585 __fanout_set_data_bpf(f, NULL); 1586 } 1587 } 1588 1589 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) 1590 { 1591 struct packet_fanout *f; 1592 1593 list_for_each_entry(f, &fanout_list, list) { 1594 if (f->id == candidate_id && 1595 read_pnet(&f->net) == sock_net(sk)) { 1596 return false; 1597 } 1598 } 1599 return true; 1600 } 1601 1602 static bool fanout_find_new_id(struct sock *sk, u16 *new_id) 1603 { 1604 u16 id = fanout_next_id; 1605 1606 do { 1607 if (__fanout_id_is_free(sk, id)) { 1608 *new_id = id; 1609 fanout_next_id = id + 1; 1610 return true; 1611 } 1612 1613 id++; 1614 } while (id != fanout_next_id); 1615 1616 return false; 1617 } 1618 1619 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1620 { 1621 struct packet_rollover *rollover = NULL; 1622 struct packet_sock *po = pkt_sk(sk); 1623 struct packet_fanout *f, *match; 1624 u8 type = type_flags & 0xff; 1625 u8 flags = type_flags >> 8; 1626 int err; 1627 1628 switch (type) { 1629 case PACKET_FANOUT_ROLLOVER: 1630 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1631 return -EINVAL; 1632 case PACKET_FANOUT_HASH: 1633 case PACKET_FANOUT_LB: 1634 case PACKET_FANOUT_CPU: 1635 case PACKET_FANOUT_RND: 1636 case PACKET_FANOUT_QM: 1637 case PACKET_FANOUT_CBPF: 1638 case PACKET_FANOUT_EBPF: 1639 break; 1640 default: 1641 return -EINVAL; 1642 } 1643 1644 mutex_lock(&fanout_mutex); 1645 1646 err = -EALREADY; 1647 if (po->fanout) 1648 goto out; 1649 1650 if (type == PACKET_FANOUT_ROLLOVER || 1651 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1652 err = -ENOMEM; 1653 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); 1654 if (!rollover) 1655 goto out; 1656 atomic_long_set(&rollover->num, 0); 1657 atomic_long_set(&rollover->num_huge, 0); 1658 atomic_long_set(&rollover->num_failed, 0); 1659 } 1660 1661 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { 1662 if (id != 0) { 1663 err = -EINVAL; 1664 goto out; 1665 } 1666 if (!fanout_find_new_id(sk, &id)) { 1667 err = -ENOMEM; 1668 goto out; 1669 } 1670 /* ephemeral flag for the first socket in the group: drop it */ 1671 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); 1672 } 1673 1674 match = NULL; 1675 list_for_each_entry(f, &fanout_list, list) { 1676 if (f->id == id && 1677 read_pnet(&f->net) == sock_net(sk)) { 1678 match = f; 1679 break; 1680 } 1681 } 1682 err = -EINVAL; 1683 if (match && match->flags != flags) 1684 goto out; 1685 if (!match) { 1686 err = -ENOMEM; 1687 match = kzalloc(sizeof(*match), GFP_KERNEL); 1688 if (!match) 1689 goto out; 1690 write_pnet(&match->net, sock_net(sk)); 1691 match->id = id; 1692 match->type = type; 1693 match->flags = flags; 1694 INIT_LIST_HEAD(&match->list); 1695 spin_lock_init(&match->lock); 1696 refcount_set(&match->sk_ref, 0); 1697 fanout_init_data(match); 1698 match->prot_hook.type = po->prot_hook.type; 1699 match->prot_hook.dev = po->prot_hook.dev; 1700 match->prot_hook.func = packet_rcv_fanout; 1701 match->prot_hook.af_packet_priv = match; 1702 match->prot_hook.id_match = match_fanout_group; 1703 list_add(&match->list, &fanout_list); 1704 } 1705 err = -EINVAL; 1706 1707 spin_lock(&po->bind_lock); 1708 if (po->running && 1709 match->type == type && 1710 match->prot_hook.type == po->prot_hook.type && 1711 match->prot_hook.dev == po->prot_hook.dev) { 1712 err = -ENOSPC; 1713 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1714 __dev_remove_pack(&po->prot_hook); 1715 po->fanout = match; 1716 po->rollover = rollover; 1717 rollover = NULL; 1718 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); 1719 __fanout_link(sk, po); 1720 err = 0; 1721 } 1722 } 1723 spin_unlock(&po->bind_lock); 1724 1725 if (err && !refcount_read(&match->sk_ref)) { 1726 list_del(&match->list); 1727 kfree(match); 1728 } 1729 1730 out: 1731 kfree(rollover); 1732 mutex_unlock(&fanout_mutex); 1733 return err; 1734 } 1735 1736 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes 1737 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. 1738 * It is the responsibility of the caller to call fanout_release_data() and 1739 * free the returned packet_fanout (after synchronize_net()) 1740 */ 1741 static struct packet_fanout *fanout_release(struct sock *sk) 1742 { 1743 struct packet_sock *po = pkt_sk(sk); 1744 struct packet_fanout *f; 1745 1746 mutex_lock(&fanout_mutex); 1747 f = po->fanout; 1748 if (f) { 1749 po->fanout = NULL; 1750 1751 if (refcount_dec_and_test(&f->sk_ref)) 1752 list_del(&f->list); 1753 else 1754 f = NULL; 1755 } 1756 mutex_unlock(&fanout_mutex); 1757 1758 return f; 1759 } 1760 1761 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1762 struct sk_buff *skb) 1763 { 1764 /* Earlier code assumed this would be a VLAN pkt, double-check 1765 * this now that we have the actual packet in hand. We can only 1766 * do this check on Ethernet devices. 1767 */ 1768 if (unlikely(dev->type != ARPHRD_ETHER)) 1769 return false; 1770 1771 skb_reset_mac_header(skb); 1772 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1773 } 1774 1775 static const struct proto_ops packet_ops; 1776 1777 static const struct proto_ops packet_ops_spkt; 1778 1779 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1780 struct packet_type *pt, struct net_device *orig_dev) 1781 { 1782 struct sock *sk; 1783 struct sockaddr_pkt *spkt; 1784 1785 /* 1786 * When we registered the protocol we saved the socket in the data 1787 * field for just this event. 1788 */ 1789 1790 sk = pt->af_packet_priv; 1791 1792 /* 1793 * Yank back the headers [hope the device set this 1794 * right or kerboom...] 1795 * 1796 * Incoming packets have ll header pulled, 1797 * push it back. 1798 * 1799 * For outgoing ones skb->data == skb_mac_header(skb) 1800 * so that this procedure is noop. 1801 */ 1802 1803 if (skb->pkt_type == PACKET_LOOPBACK) 1804 goto out; 1805 1806 if (!net_eq(dev_net(dev), sock_net(sk))) 1807 goto out; 1808 1809 skb = skb_share_check(skb, GFP_ATOMIC); 1810 if (skb == NULL) 1811 goto oom; 1812 1813 /* drop any routing info */ 1814 skb_dst_drop(skb); 1815 1816 /* drop conntrack reference */ 1817 nf_reset(skb); 1818 1819 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1820 1821 skb_push(skb, skb->data - skb_mac_header(skb)); 1822 1823 /* 1824 * The SOCK_PACKET socket receives _all_ frames. 1825 */ 1826 1827 spkt->spkt_family = dev->type; 1828 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1829 spkt->spkt_protocol = skb->protocol; 1830 1831 /* 1832 * Charge the memory to the socket. This is done specifically 1833 * to prevent sockets using all the memory up. 1834 */ 1835 1836 if (sock_queue_rcv_skb(sk, skb) == 0) 1837 return 0; 1838 1839 out: 1840 kfree_skb(skb); 1841 oom: 1842 return 0; 1843 } 1844 1845 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock) 1846 { 1847 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && 1848 sock->type == SOCK_RAW) { 1849 skb_reset_mac_header(skb); 1850 skb->protocol = dev_parse_header_protocol(skb); 1851 } 1852 1853 skb_probe_transport_header(skb); 1854 } 1855 1856 /* 1857 * Output a raw packet to a device layer. This bypasses all the other 1858 * protocol layers and you must therefore supply it with a complete frame 1859 */ 1860 1861 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1862 size_t len) 1863 { 1864 struct sock *sk = sock->sk; 1865 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1866 struct sk_buff *skb = NULL; 1867 struct net_device *dev; 1868 struct sockcm_cookie sockc; 1869 __be16 proto = 0; 1870 int err; 1871 int extra_len = 0; 1872 1873 /* 1874 * Get and verify the address. 1875 */ 1876 1877 if (saddr) { 1878 if (msg->msg_namelen < sizeof(struct sockaddr)) 1879 return -EINVAL; 1880 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1881 proto = saddr->spkt_protocol; 1882 } else 1883 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1884 1885 /* 1886 * Find the device first to size check it 1887 */ 1888 1889 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1890 retry: 1891 rcu_read_lock(); 1892 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1893 err = -ENODEV; 1894 if (dev == NULL) 1895 goto out_unlock; 1896 1897 err = -ENETDOWN; 1898 if (!(dev->flags & IFF_UP)) 1899 goto out_unlock; 1900 1901 /* 1902 * You may not queue a frame bigger than the mtu. This is the lowest level 1903 * raw protocol and you must do your own fragmentation at this level. 1904 */ 1905 1906 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1907 if (!netif_supports_nofcs(dev)) { 1908 err = -EPROTONOSUPPORT; 1909 goto out_unlock; 1910 } 1911 extra_len = 4; /* We're doing our own CRC */ 1912 } 1913 1914 err = -EMSGSIZE; 1915 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1916 goto out_unlock; 1917 1918 if (!skb) { 1919 size_t reserved = LL_RESERVED_SPACE(dev); 1920 int tlen = dev->needed_tailroom; 1921 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1922 1923 rcu_read_unlock(); 1924 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1925 if (skb == NULL) 1926 return -ENOBUFS; 1927 /* FIXME: Save some space for broken drivers that write a hard 1928 * header at transmission time by themselves. PPP is the notable 1929 * one here. This should really be fixed at the driver level. 1930 */ 1931 skb_reserve(skb, reserved); 1932 skb_reset_network_header(skb); 1933 1934 /* Try to align data part correctly */ 1935 if (hhlen) { 1936 skb->data -= hhlen; 1937 skb->tail -= hhlen; 1938 if (len < hhlen) 1939 skb_reset_network_header(skb); 1940 } 1941 err = memcpy_from_msg(skb_put(skb, len), msg, len); 1942 if (err) 1943 goto out_free; 1944 goto retry; 1945 } 1946 1947 if (!dev_validate_header(dev, skb->data, len)) { 1948 err = -EINVAL; 1949 goto out_unlock; 1950 } 1951 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 1952 !packet_extra_vlan_len_allowed(dev, skb)) { 1953 err = -EMSGSIZE; 1954 goto out_unlock; 1955 } 1956 1957 sockcm_init(&sockc, sk); 1958 if (msg->msg_controllen) { 1959 err = sock_cmsg_send(sk, msg, &sockc); 1960 if (unlikely(err)) 1961 goto out_unlock; 1962 } 1963 1964 skb->protocol = proto; 1965 skb->dev = dev; 1966 skb->priority = sk->sk_priority; 1967 skb->mark = sk->sk_mark; 1968 skb->tstamp = sockc.transmit_time; 1969 1970 skb_setup_tx_timestamp(skb, sockc.tsflags); 1971 1972 if (unlikely(extra_len == 4)) 1973 skb->no_fcs = 1; 1974 1975 packet_parse_headers(skb, sock); 1976 1977 dev_queue_xmit(skb); 1978 rcu_read_unlock(); 1979 return len; 1980 1981 out_unlock: 1982 rcu_read_unlock(); 1983 out_free: 1984 kfree_skb(skb); 1985 return err; 1986 } 1987 1988 static unsigned int run_filter(struct sk_buff *skb, 1989 const struct sock *sk, 1990 unsigned int res) 1991 { 1992 struct sk_filter *filter; 1993 1994 rcu_read_lock(); 1995 filter = rcu_dereference(sk->sk_filter); 1996 if (filter != NULL) 1997 res = bpf_prog_run_clear_cb(filter->prog, skb); 1998 rcu_read_unlock(); 1999 2000 return res; 2001 } 2002 2003 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2004 size_t *len) 2005 { 2006 struct virtio_net_hdr vnet_hdr; 2007 2008 if (*len < sizeof(vnet_hdr)) 2009 return -EINVAL; 2010 *len -= sizeof(vnet_hdr); 2011 2012 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0)) 2013 return -EINVAL; 2014 2015 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 2016 } 2017 2018 /* 2019 * This function makes lazy skb cloning in hope that most of packets 2020 * are discarded by BPF. 2021 * 2022 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2023 * and skb->cb are mangled. It works because (and until) packets 2024 * falling here are owned by current CPU. Output packets are cloned 2025 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2026 * sequencially, so that if we return skb to original state on exit, 2027 * we will not harm anyone. 2028 */ 2029 2030 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2031 struct packet_type *pt, struct net_device *orig_dev) 2032 { 2033 struct sock *sk; 2034 struct sockaddr_ll *sll; 2035 struct packet_sock *po; 2036 u8 *skb_head = skb->data; 2037 int skb_len = skb->len; 2038 unsigned int snaplen, res; 2039 bool is_drop_n_account = false; 2040 2041 if (skb->pkt_type == PACKET_LOOPBACK) 2042 goto drop; 2043 2044 sk = pt->af_packet_priv; 2045 po = pkt_sk(sk); 2046 2047 if (!net_eq(dev_net(dev), sock_net(sk))) 2048 goto drop; 2049 2050 skb->dev = dev; 2051 2052 if (dev->header_ops) { 2053 /* The device has an explicit notion of ll header, 2054 * exported to higher levels. 2055 * 2056 * Otherwise, the device hides details of its frame 2057 * structure, so that corresponding packet head is 2058 * never delivered to user. 2059 */ 2060 if (sk->sk_type != SOCK_DGRAM) 2061 skb_push(skb, skb->data - skb_mac_header(skb)); 2062 else if (skb->pkt_type == PACKET_OUTGOING) { 2063 /* Special case: outgoing packets have ll header at head */ 2064 skb_pull(skb, skb_network_offset(skb)); 2065 } 2066 } 2067 2068 snaplen = skb->len; 2069 2070 res = run_filter(skb, sk, snaplen); 2071 if (!res) 2072 goto drop_n_restore; 2073 if (snaplen > res) 2074 snaplen = res; 2075 2076 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2077 goto drop_n_acct; 2078 2079 if (skb_shared(skb)) { 2080 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2081 if (nskb == NULL) 2082 goto drop_n_acct; 2083 2084 if (skb_head != skb->data) { 2085 skb->data = skb_head; 2086 skb->len = skb_len; 2087 } 2088 consume_skb(skb); 2089 skb = nskb; 2090 } 2091 2092 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2093 2094 sll = &PACKET_SKB_CB(skb)->sa.ll; 2095 sll->sll_hatype = dev->type; 2096 sll->sll_pkttype = skb->pkt_type; 2097 if (unlikely(po->origdev)) 2098 sll->sll_ifindex = orig_dev->ifindex; 2099 else 2100 sll->sll_ifindex = dev->ifindex; 2101 2102 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2103 2104 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2105 * Use their space for storing the original skb length. 2106 */ 2107 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2108 2109 if (pskb_trim(skb, snaplen)) 2110 goto drop_n_acct; 2111 2112 skb_set_owner_r(skb, sk); 2113 skb->dev = NULL; 2114 skb_dst_drop(skb); 2115 2116 /* drop conntrack reference */ 2117 nf_reset(skb); 2118 2119 spin_lock(&sk->sk_receive_queue.lock); 2120 po->stats.stats1.tp_packets++; 2121 sock_skb_set_dropcount(sk, skb); 2122 __skb_queue_tail(&sk->sk_receive_queue, skb); 2123 spin_unlock(&sk->sk_receive_queue.lock); 2124 sk->sk_data_ready(sk); 2125 return 0; 2126 2127 drop_n_acct: 2128 is_drop_n_account = true; 2129 spin_lock(&sk->sk_receive_queue.lock); 2130 po->stats.stats1.tp_drops++; 2131 atomic_inc(&sk->sk_drops); 2132 spin_unlock(&sk->sk_receive_queue.lock); 2133 2134 drop_n_restore: 2135 if (skb_head != skb->data && skb_shared(skb)) { 2136 skb->data = skb_head; 2137 skb->len = skb_len; 2138 } 2139 drop: 2140 if (!is_drop_n_account) 2141 consume_skb(skb); 2142 else 2143 kfree_skb(skb); 2144 return 0; 2145 } 2146 2147 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2148 struct packet_type *pt, struct net_device *orig_dev) 2149 { 2150 struct sock *sk; 2151 struct packet_sock *po; 2152 struct sockaddr_ll *sll; 2153 union tpacket_uhdr h; 2154 u8 *skb_head = skb->data; 2155 int skb_len = skb->len; 2156 unsigned int snaplen, res; 2157 unsigned long status = TP_STATUS_USER; 2158 unsigned short macoff, netoff, hdrlen; 2159 struct sk_buff *copy_skb = NULL; 2160 struct timespec ts; 2161 __u32 ts_status; 2162 bool is_drop_n_account = false; 2163 bool do_vnet = false; 2164 2165 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2166 * We may add members to them until current aligned size without forcing 2167 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2168 */ 2169 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2170 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2171 2172 if (skb->pkt_type == PACKET_LOOPBACK) 2173 goto drop; 2174 2175 sk = pt->af_packet_priv; 2176 po = pkt_sk(sk); 2177 2178 if (!net_eq(dev_net(dev), sock_net(sk))) 2179 goto drop; 2180 2181 if (dev->header_ops) { 2182 if (sk->sk_type != SOCK_DGRAM) 2183 skb_push(skb, skb->data - skb_mac_header(skb)); 2184 else if (skb->pkt_type == PACKET_OUTGOING) { 2185 /* Special case: outgoing packets have ll header at head */ 2186 skb_pull(skb, skb_network_offset(skb)); 2187 } 2188 } 2189 2190 snaplen = skb->len; 2191 2192 res = run_filter(skb, sk, snaplen); 2193 if (!res) 2194 goto drop_n_restore; 2195 2196 if (skb->ip_summed == CHECKSUM_PARTIAL) 2197 status |= TP_STATUS_CSUMNOTREADY; 2198 else if (skb->pkt_type != PACKET_OUTGOING && 2199 (skb->ip_summed == CHECKSUM_COMPLETE || 2200 skb_csum_unnecessary(skb))) 2201 status |= TP_STATUS_CSUM_VALID; 2202 2203 if (snaplen > res) 2204 snaplen = res; 2205 2206 if (sk->sk_type == SOCK_DGRAM) { 2207 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2208 po->tp_reserve; 2209 } else { 2210 unsigned int maclen = skb_network_offset(skb); 2211 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2212 (maclen < 16 ? 16 : maclen)) + 2213 po->tp_reserve; 2214 if (po->has_vnet_hdr) { 2215 netoff += sizeof(struct virtio_net_hdr); 2216 do_vnet = true; 2217 } 2218 macoff = netoff - maclen; 2219 } 2220 if (po->tp_version <= TPACKET_V2) { 2221 if (macoff + snaplen > po->rx_ring.frame_size) { 2222 if (po->copy_thresh && 2223 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2224 if (skb_shared(skb)) { 2225 copy_skb = skb_clone(skb, GFP_ATOMIC); 2226 } else { 2227 copy_skb = skb_get(skb); 2228 skb_head = skb->data; 2229 } 2230 if (copy_skb) 2231 skb_set_owner_r(copy_skb, sk); 2232 } 2233 snaplen = po->rx_ring.frame_size - macoff; 2234 if ((int)snaplen < 0) { 2235 snaplen = 0; 2236 do_vnet = false; 2237 } 2238 } 2239 } else if (unlikely(macoff + snaplen > 2240 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2241 u32 nval; 2242 2243 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2244 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2245 snaplen, nval, macoff); 2246 snaplen = nval; 2247 if (unlikely((int)snaplen < 0)) { 2248 snaplen = 0; 2249 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2250 do_vnet = false; 2251 } 2252 } 2253 spin_lock(&sk->sk_receive_queue.lock); 2254 h.raw = packet_current_rx_frame(po, skb, 2255 TP_STATUS_KERNEL, (macoff+snaplen)); 2256 if (!h.raw) 2257 goto drop_n_account; 2258 if (po->tp_version <= TPACKET_V2) { 2259 packet_increment_rx_head(po, &po->rx_ring); 2260 /* 2261 * LOSING will be reported till you read the stats, 2262 * because it's COR - Clear On Read. 2263 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2264 * at packet level. 2265 */ 2266 if (po->stats.stats1.tp_drops) 2267 status |= TP_STATUS_LOSING; 2268 } 2269 2270 if (do_vnet && 2271 virtio_net_hdr_from_skb(skb, h.raw + macoff - 2272 sizeof(struct virtio_net_hdr), 2273 vio_le(), true, 0)) 2274 goto drop_n_account; 2275 2276 po->stats.stats1.tp_packets++; 2277 if (copy_skb) { 2278 status |= TP_STATUS_COPY; 2279 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2280 } 2281 spin_unlock(&sk->sk_receive_queue.lock); 2282 2283 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2284 2285 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 2286 getnstimeofday(&ts); 2287 2288 status |= ts_status; 2289 2290 switch (po->tp_version) { 2291 case TPACKET_V1: 2292 h.h1->tp_len = skb->len; 2293 h.h1->tp_snaplen = snaplen; 2294 h.h1->tp_mac = macoff; 2295 h.h1->tp_net = netoff; 2296 h.h1->tp_sec = ts.tv_sec; 2297 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2298 hdrlen = sizeof(*h.h1); 2299 break; 2300 case TPACKET_V2: 2301 h.h2->tp_len = skb->len; 2302 h.h2->tp_snaplen = snaplen; 2303 h.h2->tp_mac = macoff; 2304 h.h2->tp_net = netoff; 2305 h.h2->tp_sec = ts.tv_sec; 2306 h.h2->tp_nsec = ts.tv_nsec; 2307 if (skb_vlan_tag_present(skb)) { 2308 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2309 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2310 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2311 } else { 2312 h.h2->tp_vlan_tci = 0; 2313 h.h2->tp_vlan_tpid = 0; 2314 } 2315 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2316 hdrlen = sizeof(*h.h2); 2317 break; 2318 case TPACKET_V3: 2319 /* tp_nxt_offset,vlan are already populated above. 2320 * So DONT clear those fields here 2321 */ 2322 h.h3->tp_status |= status; 2323 h.h3->tp_len = skb->len; 2324 h.h3->tp_snaplen = snaplen; 2325 h.h3->tp_mac = macoff; 2326 h.h3->tp_net = netoff; 2327 h.h3->tp_sec = ts.tv_sec; 2328 h.h3->tp_nsec = ts.tv_nsec; 2329 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2330 hdrlen = sizeof(*h.h3); 2331 break; 2332 default: 2333 BUG(); 2334 } 2335 2336 sll = h.raw + TPACKET_ALIGN(hdrlen); 2337 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2338 sll->sll_family = AF_PACKET; 2339 sll->sll_hatype = dev->type; 2340 sll->sll_protocol = skb->protocol; 2341 sll->sll_pkttype = skb->pkt_type; 2342 if (unlikely(po->origdev)) 2343 sll->sll_ifindex = orig_dev->ifindex; 2344 else 2345 sll->sll_ifindex = dev->ifindex; 2346 2347 smp_mb(); 2348 2349 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2350 if (po->tp_version <= TPACKET_V2) { 2351 u8 *start, *end; 2352 2353 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2354 macoff + snaplen); 2355 2356 for (start = h.raw; start < end; start += PAGE_SIZE) 2357 flush_dcache_page(pgv_to_page(start)); 2358 } 2359 smp_wmb(); 2360 #endif 2361 2362 if (po->tp_version <= TPACKET_V2) { 2363 __packet_set_status(po, h.raw, status); 2364 sk->sk_data_ready(sk); 2365 } else { 2366 prb_clear_blk_fill_status(&po->rx_ring); 2367 } 2368 2369 drop_n_restore: 2370 if (skb_head != skb->data && skb_shared(skb)) { 2371 skb->data = skb_head; 2372 skb->len = skb_len; 2373 } 2374 drop: 2375 if (!is_drop_n_account) 2376 consume_skb(skb); 2377 else 2378 kfree_skb(skb); 2379 return 0; 2380 2381 drop_n_account: 2382 is_drop_n_account = true; 2383 po->stats.stats1.tp_drops++; 2384 spin_unlock(&sk->sk_receive_queue.lock); 2385 2386 sk->sk_data_ready(sk); 2387 kfree_skb(copy_skb); 2388 goto drop_n_restore; 2389 } 2390 2391 static void tpacket_destruct_skb(struct sk_buff *skb) 2392 { 2393 struct packet_sock *po = pkt_sk(skb->sk); 2394 2395 if (likely(po->tx_ring.pg_vec)) { 2396 void *ph; 2397 __u32 ts; 2398 2399 ph = skb_zcopy_get_nouarg(skb); 2400 packet_dec_pending(&po->tx_ring); 2401 2402 ts = __packet_set_timestamp(po, ph, skb); 2403 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2404 2405 if (!packet_read_pending(&po->tx_ring)) 2406 complete(&po->skb_completion); 2407 } 2408 2409 sock_wfree(skb); 2410 } 2411 2412 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2413 { 2414 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2415 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2416 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2417 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2418 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2419 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2420 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2421 2422 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2423 return -EINVAL; 2424 2425 return 0; 2426 } 2427 2428 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2429 struct virtio_net_hdr *vnet_hdr) 2430 { 2431 if (*len < sizeof(*vnet_hdr)) 2432 return -EINVAL; 2433 *len -= sizeof(*vnet_hdr); 2434 2435 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2436 return -EFAULT; 2437 2438 return __packet_snd_vnet_parse(vnet_hdr, *len); 2439 } 2440 2441 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2442 void *frame, struct net_device *dev, void *data, int tp_len, 2443 __be16 proto, unsigned char *addr, int hlen, int copylen, 2444 const struct sockcm_cookie *sockc) 2445 { 2446 union tpacket_uhdr ph; 2447 int to_write, offset, len, nr_frags, len_max; 2448 struct socket *sock = po->sk.sk_socket; 2449 struct page *page; 2450 int err; 2451 2452 ph.raw = frame; 2453 2454 skb->protocol = proto; 2455 skb->dev = dev; 2456 skb->priority = po->sk.sk_priority; 2457 skb->mark = po->sk.sk_mark; 2458 skb->tstamp = sockc->transmit_time; 2459 skb_setup_tx_timestamp(skb, sockc->tsflags); 2460 skb_zcopy_set_nouarg(skb, ph.raw); 2461 2462 skb_reserve(skb, hlen); 2463 skb_reset_network_header(skb); 2464 2465 to_write = tp_len; 2466 2467 if (sock->type == SOCK_DGRAM) { 2468 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2469 NULL, tp_len); 2470 if (unlikely(err < 0)) 2471 return -EINVAL; 2472 } else if (copylen) { 2473 int hdrlen = min_t(int, copylen, tp_len); 2474 2475 skb_push(skb, dev->hard_header_len); 2476 skb_put(skb, copylen - dev->hard_header_len); 2477 err = skb_store_bits(skb, 0, data, hdrlen); 2478 if (unlikely(err)) 2479 return err; 2480 if (!dev_validate_header(dev, skb->data, hdrlen)) 2481 return -EINVAL; 2482 2483 data += hdrlen; 2484 to_write -= hdrlen; 2485 } 2486 2487 offset = offset_in_page(data); 2488 len_max = PAGE_SIZE - offset; 2489 len = ((to_write > len_max) ? len_max : to_write); 2490 2491 skb->data_len = to_write; 2492 skb->len += to_write; 2493 skb->truesize += to_write; 2494 refcount_add(to_write, &po->sk.sk_wmem_alloc); 2495 2496 while (likely(to_write)) { 2497 nr_frags = skb_shinfo(skb)->nr_frags; 2498 2499 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2500 pr_err("Packet exceed the number of skb frags(%lu)\n", 2501 MAX_SKB_FRAGS); 2502 return -EFAULT; 2503 } 2504 2505 page = pgv_to_page(data); 2506 data += len; 2507 flush_dcache_page(page); 2508 get_page(page); 2509 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2510 to_write -= len; 2511 offset = 0; 2512 len_max = PAGE_SIZE; 2513 len = ((to_write > len_max) ? len_max : to_write); 2514 } 2515 2516 packet_parse_headers(skb, sock); 2517 2518 return tp_len; 2519 } 2520 2521 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2522 int size_max, void **data) 2523 { 2524 union tpacket_uhdr ph; 2525 int tp_len, off; 2526 2527 ph.raw = frame; 2528 2529 switch (po->tp_version) { 2530 case TPACKET_V3: 2531 if (ph.h3->tp_next_offset != 0) { 2532 pr_warn_once("variable sized slot not supported"); 2533 return -EINVAL; 2534 } 2535 tp_len = ph.h3->tp_len; 2536 break; 2537 case TPACKET_V2: 2538 tp_len = ph.h2->tp_len; 2539 break; 2540 default: 2541 tp_len = ph.h1->tp_len; 2542 break; 2543 } 2544 if (unlikely(tp_len > size_max)) { 2545 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2546 return -EMSGSIZE; 2547 } 2548 2549 if (unlikely(po->tp_tx_has_off)) { 2550 int off_min, off_max; 2551 2552 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2553 off_max = po->tx_ring.frame_size - tp_len; 2554 if (po->sk.sk_type == SOCK_DGRAM) { 2555 switch (po->tp_version) { 2556 case TPACKET_V3: 2557 off = ph.h3->tp_net; 2558 break; 2559 case TPACKET_V2: 2560 off = ph.h2->tp_net; 2561 break; 2562 default: 2563 off = ph.h1->tp_net; 2564 break; 2565 } 2566 } else { 2567 switch (po->tp_version) { 2568 case TPACKET_V3: 2569 off = ph.h3->tp_mac; 2570 break; 2571 case TPACKET_V2: 2572 off = ph.h2->tp_mac; 2573 break; 2574 default: 2575 off = ph.h1->tp_mac; 2576 break; 2577 } 2578 } 2579 if (unlikely((off < off_min) || (off_max < off))) 2580 return -EINVAL; 2581 } else { 2582 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2583 } 2584 2585 *data = frame + off; 2586 return tp_len; 2587 } 2588 2589 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2590 { 2591 struct sk_buff *skb = NULL; 2592 struct net_device *dev; 2593 struct virtio_net_hdr *vnet_hdr = NULL; 2594 struct sockcm_cookie sockc; 2595 __be16 proto; 2596 int err, reserve = 0; 2597 void *ph; 2598 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2599 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2600 unsigned char *addr = NULL; 2601 int tp_len, size_max; 2602 void *data; 2603 int len_sum = 0; 2604 int status = TP_STATUS_AVAILABLE; 2605 int hlen, tlen, copylen = 0; 2606 long timeo = 0; 2607 2608 mutex_lock(&po->pg_vec_lock); 2609 2610 if (likely(saddr == NULL)) { 2611 dev = packet_cached_dev_get(po); 2612 proto = po->num; 2613 } else { 2614 err = -EINVAL; 2615 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2616 goto out; 2617 if (msg->msg_namelen < (saddr->sll_halen 2618 + offsetof(struct sockaddr_ll, 2619 sll_addr))) 2620 goto out; 2621 proto = saddr->sll_protocol; 2622 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2623 if (po->sk.sk_socket->type == SOCK_DGRAM) { 2624 if (dev && msg->msg_namelen < dev->addr_len + 2625 offsetof(struct sockaddr_ll, sll_addr)) 2626 goto out_put; 2627 addr = saddr->sll_addr; 2628 } 2629 } 2630 2631 err = -ENXIO; 2632 if (unlikely(dev == NULL)) 2633 goto out; 2634 err = -ENETDOWN; 2635 if (unlikely(!(dev->flags & IFF_UP))) 2636 goto out_put; 2637 2638 sockcm_init(&sockc, &po->sk); 2639 if (msg->msg_controllen) { 2640 err = sock_cmsg_send(&po->sk, msg, &sockc); 2641 if (unlikely(err)) 2642 goto out_put; 2643 } 2644 2645 if (po->sk.sk_socket->type == SOCK_RAW) 2646 reserve = dev->hard_header_len; 2647 size_max = po->tx_ring.frame_size 2648 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2649 2650 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) 2651 size_max = dev->mtu + reserve + VLAN_HLEN; 2652 2653 reinit_completion(&po->skb_completion); 2654 2655 do { 2656 ph = packet_current_frame(po, &po->tx_ring, 2657 TP_STATUS_SEND_REQUEST); 2658 if (unlikely(ph == NULL)) { 2659 if (need_wait && skb) { 2660 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT); 2661 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo); 2662 if (timeo <= 0) { 2663 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS; 2664 goto out_put; 2665 } 2666 } 2667 /* check for additional frames */ 2668 continue; 2669 } 2670 2671 skb = NULL; 2672 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2673 if (tp_len < 0) 2674 goto tpacket_error; 2675 2676 status = TP_STATUS_SEND_REQUEST; 2677 hlen = LL_RESERVED_SPACE(dev); 2678 tlen = dev->needed_tailroom; 2679 if (po->has_vnet_hdr) { 2680 vnet_hdr = data; 2681 data += sizeof(*vnet_hdr); 2682 tp_len -= sizeof(*vnet_hdr); 2683 if (tp_len < 0 || 2684 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2685 tp_len = -EINVAL; 2686 goto tpacket_error; 2687 } 2688 copylen = __virtio16_to_cpu(vio_le(), 2689 vnet_hdr->hdr_len); 2690 } 2691 copylen = max_t(int, copylen, dev->hard_header_len); 2692 skb = sock_alloc_send_skb(&po->sk, 2693 hlen + tlen + sizeof(struct sockaddr_ll) + 2694 (copylen - dev->hard_header_len), 2695 !need_wait, &err); 2696 2697 if (unlikely(skb == NULL)) { 2698 /* we assume the socket was initially writeable ... */ 2699 if (likely(len_sum > 0)) 2700 err = len_sum; 2701 goto out_status; 2702 } 2703 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2704 addr, hlen, copylen, &sockc); 2705 if (likely(tp_len >= 0) && 2706 tp_len > dev->mtu + reserve && 2707 !po->has_vnet_hdr && 2708 !packet_extra_vlan_len_allowed(dev, skb)) 2709 tp_len = -EMSGSIZE; 2710 2711 if (unlikely(tp_len < 0)) { 2712 tpacket_error: 2713 if (po->tp_loss) { 2714 __packet_set_status(po, ph, 2715 TP_STATUS_AVAILABLE); 2716 packet_increment_head(&po->tx_ring); 2717 kfree_skb(skb); 2718 continue; 2719 } else { 2720 status = TP_STATUS_WRONG_FORMAT; 2721 err = tp_len; 2722 goto out_status; 2723 } 2724 } 2725 2726 if (po->has_vnet_hdr) { 2727 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) { 2728 tp_len = -EINVAL; 2729 goto tpacket_error; 2730 } 2731 virtio_net_hdr_set_proto(skb, vnet_hdr); 2732 } 2733 2734 skb->destructor = tpacket_destruct_skb; 2735 __packet_set_status(po, ph, TP_STATUS_SENDING); 2736 packet_inc_pending(&po->tx_ring); 2737 2738 status = TP_STATUS_SEND_REQUEST; 2739 err = po->xmit(skb); 2740 if (unlikely(err > 0)) { 2741 err = net_xmit_errno(err); 2742 if (err && __packet_get_status(po, ph) == 2743 TP_STATUS_AVAILABLE) { 2744 /* skb was destructed already */ 2745 skb = NULL; 2746 goto out_status; 2747 } 2748 /* 2749 * skb was dropped but not destructed yet; 2750 * let's treat it like congestion or err < 0 2751 */ 2752 err = 0; 2753 } 2754 packet_increment_head(&po->tx_ring); 2755 len_sum += tp_len; 2756 } while (likely((ph != NULL) || 2757 /* Note: packet_read_pending() might be slow if we have 2758 * to call it as it's per_cpu variable, but in fast-path 2759 * we already short-circuit the loop with the first 2760 * condition, and luckily don't have to go that path 2761 * anyway. 2762 */ 2763 (need_wait && packet_read_pending(&po->tx_ring)))); 2764 2765 err = len_sum; 2766 goto out_put; 2767 2768 out_status: 2769 __packet_set_status(po, ph, status); 2770 kfree_skb(skb); 2771 out_put: 2772 dev_put(dev); 2773 out: 2774 mutex_unlock(&po->pg_vec_lock); 2775 return err; 2776 } 2777 2778 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2779 size_t reserve, size_t len, 2780 size_t linear, int noblock, 2781 int *err) 2782 { 2783 struct sk_buff *skb; 2784 2785 /* Under a page? Don't bother with paged skb. */ 2786 if (prepad + len < PAGE_SIZE || !linear) 2787 linear = len; 2788 2789 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2790 err, 0); 2791 if (!skb) 2792 return NULL; 2793 2794 skb_reserve(skb, reserve); 2795 skb_put(skb, linear); 2796 skb->data_len = len - linear; 2797 skb->len += len - linear; 2798 2799 return skb; 2800 } 2801 2802 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2803 { 2804 struct sock *sk = sock->sk; 2805 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2806 struct sk_buff *skb; 2807 struct net_device *dev; 2808 __be16 proto; 2809 unsigned char *addr = NULL; 2810 int err, reserve = 0; 2811 struct sockcm_cookie sockc; 2812 struct virtio_net_hdr vnet_hdr = { 0 }; 2813 int offset = 0; 2814 struct packet_sock *po = pkt_sk(sk); 2815 bool has_vnet_hdr = false; 2816 int hlen, tlen, linear; 2817 int extra_len = 0; 2818 2819 /* 2820 * Get and verify the address. 2821 */ 2822 2823 if (likely(saddr == NULL)) { 2824 dev = packet_cached_dev_get(po); 2825 proto = po->num; 2826 } else { 2827 err = -EINVAL; 2828 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2829 goto out; 2830 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2831 goto out; 2832 proto = saddr->sll_protocol; 2833 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2834 if (sock->type == SOCK_DGRAM) { 2835 if (dev && msg->msg_namelen < dev->addr_len + 2836 offsetof(struct sockaddr_ll, sll_addr)) 2837 goto out_unlock; 2838 addr = saddr->sll_addr; 2839 } 2840 } 2841 2842 err = -ENXIO; 2843 if (unlikely(dev == NULL)) 2844 goto out_unlock; 2845 err = -ENETDOWN; 2846 if (unlikely(!(dev->flags & IFF_UP))) 2847 goto out_unlock; 2848 2849 sockcm_init(&sockc, sk); 2850 sockc.mark = sk->sk_mark; 2851 if (msg->msg_controllen) { 2852 err = sock_cmsg_send(sk, msg, &sockc); 2853 if (unlikely(err)) 2854 goto out_unlock; 2855 } 2856 2857 if (sock->type == SOCK_RAW) 2858 reserve = dev->hard_header_len; 2859 if (po->has_vnet_hdr) { 2860 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2861 if (err) 2862 goto out_unlock; 2863 has_vnet_hdr = true; 2864 } 2865 2866 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2867 if (!netif_supports_nofcs(dev)) { 2868 err = -EPROTONOSUPPORT; 2869 goto out_unlock; 2870 } 2871 extra_len = 4; /* We're doing our own CRC */ 2872 } 2873 2874 err = -EMSGSIZE; 2875 if (!vnet_hdr.gso_type && 2876 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2877 goto out_unlock; 2878 2879 err = -ENOBUFS; 2880 hlen = LL_RESERVED_SPACE(dev); 2881 tlen = dev->needed_tailroom; 2882 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); 2883 linear = max(linear, min_t(int, len, dev->hard_header_len)); 2884 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, 2885 msg->msg_flags & MSG_DONTWAIT, &err); 2886 if (skb == NULL) 2887 goto out_unlock; 2888 2889 skb_reset_network_header(skb); 2890 2891 err = -EINVAL; 2892 if (sock->type == SOCK_DGRAM) { 2893 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 2894 if (unlikely(offset < 0)) 2895 goto out_free; 2896 } else if (reserve) { 2897 skb_reserve(skb, -reserve); 2898 if (len < reserve + sizeof(struct ipv6hdr) && 2899 dev->min_header_len != dev->hard_header_len) 2900 skb_reset_network_header(skb); 2901 } 2902 2903 /* Returns -EFAULT on error */ 2904 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 2905 if (err) 2906 goto out_free; 2907 2908 if (sock->type == SOCK_RAW && 2909 !dev_validate_header(dev, skb->data, len)) { 2910 err = -EINVAL; 2911 goto out_free; 2912 } 2913 2914 skb_setup_tx_timestamp(skb, sockc.tsflags); 2915 2916 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 2917 !packet_extra_vlan_len_allowed(dev, skb)) { 2918 err = -EMSGSIZE; 2919 goto out_free; 2920 } 2921 2922 skb->protocol = proto; 2923 skb->dev = dev; 2924 skb->priority = sk->sk_priority; 2925 skb->mark = sockc.mark; 2926 skb->tstamp = sockc.transmit_time; 2927 2928 if (has_vnet_hdr) { 2929 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 2930 if (err) 2931 goto out_free; 2932 len += sizeof(vnet_hdr); 2933 virtio_net_hdr_set_proto(skb, &vnet_hdr); 2934 } 2935 2936 packet_parse_headers(skb, sock); 2937 2938 if (unlikely(extra_len == 4)) 2939 skb->no_fcs = 1; 2940 2941 err = po->xmit(skb); 2942 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2943 goto out_unlock; 2944 2945 dev_put(dev); 2946 2947 return len; 2948 2949 out_free: 2950 kfree_skb(skb); 2951 out_unlock: 2952 if (dev) 2953 dev_put(dev); 2954 out: 2955 return err; 2956 } 2957 2958 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2959 { 2960 struct sock *sk = sock->sk; 2961 struct packet_sock *po = pkt_sk(sk); 2962 2963 if (po->tx_ring.pg_vec) 2964 return tpacket_snd(po, msg); 2965 else 2966 return packet_snd(sock, msg, len); 2967 } 2968 2969 /* 2970 * Close a PACKET socket. This is fairly simple. We immediately go 2971 * to 'closed' state and remove our protocol entry in the device list. 2972 */ 2973 2974 static int packet_release(struct socket *sock) 2975 { 2976 struct sock *sk = sock->sk; 2977 struct packet_sock *po; 2978 struct packet_fanout *f; 2979 struct net *net; 2980 union tpacket_req_u req_u; 2981 2982 if (!sk) 2983 return 0; 2984 2985 net = sock_net(sk); 2986 po = pkt_sk(sk); 2987 2988 mutex_lock(&net->packet.sklist_lock); 2989 sk_del_node_init_rcu(sk); 2990 mutex_unlock(&net->packet.sklist_lock); 2991 2992 preempt_disable(); 2993 sock_prot_inuse_add(net, sk->sk_prot, -1); 2994 preempt_enable(); 2995 2996 spin_lock(&po->bind_lock); 2997 unregister_prot_hook(sk, false); 2998 packet_cached_dev_reset(po); 2999 3000 if (po->prot_hook.dev) { 3001 dev_put(po->prot_hook.dev); 3002 po->prot_hook.dev = NULL; 3003 } 3004 spin_unlock(&po->bind_lock); 3005 3006 packet_flush_mclist(sk); 3007 3008 lock_sock(sk); 3009 if (po->rx_ring.pg_vec) { 3010 memset(&req_u, 0, sizeof(req_u)); 3011 packet_set_ring(sk, &req_u, 1, 0); 3012 } 3013 3014 if (po->tx_ring.pg_vec) { 3015 memset(&req_u, 0, sizeof(req_u)); 3016 packet_set_ring(sk, &req_u, 1, 1); 3017 } 3018 release_sock(sk); 3019 3020 f = fanout_release(sk); 3021 3022 synchronize_net(); 3023 3024 kfree(po->rollover); 3025 if (f) { 3026 fanout_release_data(f); 3027 kfree(f); 3028 } 3029 /* 3030 * Now the socket is dead. No more input will appear. 3031 */ 3032 sock_orphan(sk); 3033 sock->sk = NULL; 3034 3035 /* Purge queues */ 3036 3037 skb_queue_purge(&sk->sk_receive_queue); 3038 packet_free_pending(po); 3039 sk_refcnt_debug_release(sk); 3040 3041 sock_put(sk); 3042 return 0; 3043 } 3044 3045 /* 3046 * Attach a packet hook. 3047 */ 3048 3049 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3050 __be16 proto) 3051 { 3052 struct packet_sock *po = pkt_sk(sk); 3053 struct net_device *dev_curr; 3054 __be16 proto_curr; 3055 bool need_rehook; 3056 struct net_device *dev = NULL; 3057 int ret = 0; 3058 bool unlisted = false; 3059 3060 lock_sock(sk); 3061 spin_lock(&po->bind_lock); 3062 rcu_read_lock(); 3063 3064 if (po->fanout) { 3065 ret = -EINVAL; 3066 goto out_unlock; 3067 } 3068 3069 if (name) { 3070 dev = dev_get_by_name_rcu(sock_net(sk), name); 3071 if (!dev) { 3072 ret = -ENODEV; 3073 goto out_unlock; 3074 } 3075 } else if (ifindex) { 3076 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3077 if (!dev) { 3078 ret = -ENODEV; 3079 goto out_unlock; 3080 } 3081 } 3082 3083 if (dev) 3084 dev_hold(dev); 3085 3086 proto_curr = po->prot_hook.type; 3087 dev_curr = po->prot_hook.dev; 3088 3089 need_rehook = proto_curr != proto || dev_curr != dev; 3090 3091 if (need_rehook) { 3092 if (po->running) { 3093 rcu_read_unlock(); 3094 /* prevents packet_notifier() from calling 3095 * register_prot_hook() 3096 */ 3097 po->num = 0; 3098 __unregister_prot_hook(sk, true); 3099 rcu_read_lock(); 3100 dev_curr = po->prot_hook.dev; 3101 if (dev) 3102 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3103 dev->ifindex); 3104 } 3105 3106 BUG_ON(po->running); 3107 po->num = proto; 3108 po->prot_hook.type = proto; 3109 3110 if (unlikely(unlisted)) { 3111 dev_put(dev); 3112 po->prot_hook.dev = NULL; 3113 po->ifindex = -1; 3114 packet_cached_dev_reset(po); 3115 } else { 3116 po->prot_hook.dev = dev; 3117 po->ifindex = dev ? dev->ifindex : 0; 3118 packet_cached_dev_assign(po, dev); 3119 } 3120 } 3121 if (dev_curr) 3122 dev_put(dev_curr); 3123 3124 if (proto == 0 || !need_rehook) 3125 goto out_unlock; 3126 3127 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3128 register_prot_hook(sk); 3129 } else { 3130 sk->sk_err = ENETDOWN; 3131 if (!sock_flag(sk, SOCK_DEAD)) 3132 sk->sk_error_report(sk); 3133 } 3134 3135 out_unlock: 3136 rcu_read_unlock(); 3137 spin_unlock(&po->bind_lock); 3138 release_sock(sk); 3139 return ret; 3140 } 3141 3142 /* 3143 * Bind a packet socket to a device 3144 */ 3145 3146 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3147 int addr_len) 3148 { 3149 struct sock *sk = sock->sk; 3150 char name[sizeof(uaddr->sa_data) + 1]; 3151 3152 /* 3153 * Check legality 3154 */ 3155 3156 if (addr_len != sizeof(struct sockaddr)) 3157 return -EINVAL; 3158 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be 3159 * zero-terminated. 3160 */ 3161 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); 3162 name[sizeof(uaddr->sa_data)] = 0; 3163 3164 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3165 } 3166 3167 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3168 { 3169 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3170 struct sock *sk = sock->sk; 3171 3172 /* 3173 * Check legality 3174 */ 3175 3176 if (addr_len < sizeof(struct sockaddr_ll)) 3177 return -EINVAL; 3178 if (sll->sll_family != AF_PACKET) 3179 return -EINVAL; 3180 3181 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3182 sll->sll_protocol ? : pkt_sk(sk)->num); 3183 } 3184 3185 static struct proto packet_proto = { 3186 .name = "PACKET", 3187 .owner = THIS_MODULE, 3188 .obj_size = sizeof(struct packet_sock), 3189 }; 3190 3191 /* 3192 * Create a packet of type SOCK_PACKET. 3193 */ 3194 3195 static int packet_create(struct net *net, struct socket *sock, int protocol, 3196 int kern) 3197 { 3198 struct sock *sk; 3199 struct packet_sock *po; 3200 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3201 int err; 3202 3203 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3204 return -EPERM; 3205 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3206 sock->type != SOCK_PACKET) 3207 return -ESOCKTNOSUPPORT; 3208 3209 sock->state = SS_UNCONNECTED; 3210 3211 err = -ENOBUFS; 3212 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3213 if (sk == NULL) 3214 goto out; 3215 3216 sock->ops = &packet_ops; 3217 if (sock->type == SOCK_PACKET) 3218 sock->ops = &packet_ops_spkt; 3219 3220 sock_init_data(sock, sk); 3221 3222 po = pkt_sk(sk); 3223 init_completion(&po->skb_completion); 3224 sk->sk_family = PF_PACKET; 3225 po->num = proto; 3226 po->xmit = dev_queue_xmit; 3227 3228 err = packet_alloc_pending(po); 3229 if (err) 3230 goto out2; 3231 3232 packet_cached_dev_reset(po); 3233 3234 sk->sk_destruct = packet_sock_destruct; 3235 sk_refcnt_debug_inc(sk); 3236 3237 /* 3238 * Attach a protocol block 3239 */ 3240 3241 spin_lock_init(&po->bind_lock); 3242 mutex_init(&po->pg_vec_lock); 3243 po->rollover = NULL; 3244 po->prot_hook.func = packet_rcv; 3245 3246 if (sock->type == SOCK_PACKET) 3247 po->prot_hook.func = packet_rcv_spkt; 3248 3249 po->prot_hook.af_packet_priv = sk; 3250 3251 if (proto) { 3252 po->prot_hook.type = proto; 3253 __register_prot_hook(sk); 3254 } 3255 3256 mutex_lock(&net->packet.sklist_lock); 3257 sk_add_node_tail_rcu(sk, &net->packet.sklist); 3258 mutex_unlock(&net->packet.sklist_lock); 3259 3260 preempt_disable(); 3261 sock_prot_inuse_add(net, &packet_proto, 1); 3262 preempt_enable(); 3263 3264 return 0; 3265 out2: 3266 sk_free(sk); 3267 out: 3268 return err; 3269 } 3270 3271 /* 3272 * Pull a packet from our receive queue and hand it to the user. 3273 * If necessary we block. 3274 */ 3275 3276 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3277 int flags) 3278 { 3279 struct sock *sk = sock->sk; 3280 struct sk_buff *skb; 3281 int copied, err; 3282 int vnet_hdr_len = 0; 3283 unsigned int origlen = 0; 3284 3285 err = -EINVAL; 3286 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3287 goto out; 3288 3289 #if 0 3290 /* What error should we return now? EUNATTACH? */ 3291 if (pkt_sk(sk)->ifindex < 0) 3292 return -ENODEV; 3293 #endif 3294 3295 if (flags & MSG_ERRQUEUE) { 3296 err = sock_recv_errqueue(sk, msg, len, 3297 SOL_PACKET, PACKET_TX_TIMESTAMP); 3298 goto out; 3299 } 3300 3301 /* 3302 * Call the generic datagram receiver. This handles all sorts 3303 * of horrible races and re-entrancy so we can forget about it 3304 * in the protocol layers. 3305 * 3306 * Now it will return ENETDOWN, if device have just gone down, 3307 * but then it will block. 3308 */ 3309 3310 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 3311 3312 /* 3313 * An error occurred so return it. Because skb_recv_datagram() 3314 * handles the blocking we don't see and worry about blocking 3315 * retries. 3316 */ 3317 3318 if (skb == NULL) 3319 goto out; 3320 3321 if (pkt_sk(sk)->pressure) 3322 packet_rcv_has_room(pkt_sk(sk), NULL); 3323 3324 if (pkt_sk(sk)->has_vnet_hdr) { 3325 err = packet_rcv_vnet(msg, skb, &len); 3326 if (err) 3327 goto out_free; 3328 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3329 } 3330 3331 /* You lose any data beyond the buffer you gave. If it worries 3332 * a user program they can ask the device for its MTU 3333 * anyway. 3334 */ 3335 copied = skb->len; 3336 if (copied > len) { 3337 copied = len; 3338 msg->msg_flags |= MSG_TRUNC; 3339 } 3340 3341 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3342 if (err) 3343 goto out_free; 3344 3345 if (sock->type != SOCK_PACKET) { 3346 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3347 3348 /* Original length was stored in sockaddr_ll fields */ 3349 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3350 sll->sll_family = AF_PACKET; 3351 sll->sll_protocol = skb->protocol; 3352 } 3353 3354 sock_recv_ts_and_drops(msg, sk, skb); 3355 3356 if (msg->msg_name) { 3357 int copy_len; 3358 3359 /* If the address length field is there to be filled 3360 * in, we fill it in now. 3361 */ 3362 if (sock->type == SOCK_PACKET) { 3363 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3364 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3365 copy_len = msg->msg_namelen; 3366 } else { 3367 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3368 3369 msg->msg_namelen = sll->sll_halen + 3370 offsetof(struct sockaddr_ll, sll_addr); 3371 copy_len = msg->msg_namelen; 3372 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { 3373 memset(msg->msg_name + 3374 offsetof(struct sockaddr_ll, sll_addr), 3375 0, sizeof(sll->sll_addr)); 3376 msg->msg_namelen = sizeof(struct sockaddr_ll); 3377 } 3378 } 3379 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); 3380 } 3381 3382 if (pkt_sk(sk)->auxdata) { 3383 struct tpacket_auxdata aux; 3384 3385 aux.tp_status = TP_STATUS_USER; 3386 if (skb->ip_summed == CHECKSUM_PARTIAL) 3387 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3388 else if (skb->pkt_type != PACKET_OUTGOING && 3389 (skb->ip_summed == CHECKSUM_COMPLETE || 3390 skb_csum_unnecessary(skb))) 3391 aux.tp_status |= TP_STATUS_CSUM_VALID; 3392 3393 aux.tp_len = origlen; 3394 aux.tp_snaplen = skb->len; 3395 aux.tp_mac = 0; 3396 aux.tp_net = skb_network_offset(skb); 3397 if (skb_vlan_tag_present(skb)) { 3398 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3399 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3400 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3401 } else { 3402 aux.tp_vlan_tci = 0; 3403 aux.tp_vlan_tpid = 0; 3404 } 3405 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3406 } 3407 3408 /* 3409 * Free or return the buffer as appropriate. Again this 3410 * hides all the races and re-entrancy issues from us. 3411 */ 3412 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3413 3414 out_free: 3415 skb_free_datagram(sk, skb); 3416 out: 3417 return err; 3418 } 3419 3420 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3421 int peer) 3422 { 3423 struct net_device *dev; 3424 struct sock *sk = sock->sk; 3425 3426 if (peer) 3427 return -EOPNOTSUPP; 3428 3429 uaddr->sa_family = AF_PACKET; 3430 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3431 rcu_read_lock(); 3432 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3433 if (dev) 3434 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3435 rcu_read_unlock(); 3436 3437 return sizeof(*uaddr); 3438 } 3439 3440 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3441 int peer) 3442 { 3443 struct net_device *dev; 3444 struct sock *sk = sock->sk; 3445 struct packet_sock *po = pkt_sk(sk); 3446 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3447 3448 if (peer) 3449 return -EOPNOTSUPP; 3450 3451 sll->sll_family = AF_PACKET; 3452 sll->sll_ifindex = po->ifindex; 3453 sll->sll_protocol = po->num; 3454 sll->sll_pkttype = 0; 3455 rcu_read_lock(); 3456 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 3457 if (dev) { 3458 sll->sll_hatype = dev->type; 3459 sll->sll_halen = dev->addr_len; 3460 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3461 } else { 3462 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3463 sll->sll_halen = 0; 3464 } 3465 rcu_read_unlock(); 3466 3467 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3468 } 3469 3470 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3471 int what) 3472 { 3473 switch (i->type) { 3474 case PACKET_MR_MULTICAST: 3475 if (i->alen != dev->addr_len) 3476 return -EINVAL; 3477 if (what > 0) 3478 return dev_mc_add(dev, i->addr); 3479 else 3480 return dev_mc_del(dev, i->addr); 3481 break; 3482 case PACKET_MR_PROMISC: 3483 return dev_set_promiscuity(dev, what); 3484 case PACKET_MR_ALLMULTI: 3485 return dev_set_allmulti(dev, what); 3486 case PACKET_MR_UNICAST: 3487 if (i->alen != dev->addr_len) 3488 return -EINVAL; 3489 if (what > 0) 3490 return dev_uc_add(dev, i->addr); 3491 else 3492 return dev_uc_del(dev, i->addr); 3493 break; 3494 default: 3495 break; 3496 } 3497 return 0; 3498 } 3499 3500 static void packet_dev_mclist_delete(struct net_device *dev, 3501 struct packet_mclist **mlp) 3502 { 3503 struct packet_mclist *ml; 3504 3505 while ((ml = *mlp) != NULL) { 3506 if (ml->ifindex == dev->ifindex) { 3507 packet_dev_mc(dev, ml, -1); 3508 *mlp = ml->next; 3509 kfree(ml); 3510 } else 3511 mlp = &ml->next; 3512 } 3513 } 3514 3515 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3516 { 3517 struct packet_sock *po = pkt_sk(sk); 3518 struct packet_mclist *ml, *i; 3519 struct net_device *dev; 3520 int err; 3521 3522 rtnl_lock(); 3523 3524 err = -ENODEV; 3525 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3526 if (!dev) 3527 goto done; 3528 3529 err = -EINVAL; 3530 if (mreq->mr_alen > dev->addr_len) 3531 goto done; 3532 3533 err = -ENOBUFS; 3534 i = kmalloc(sizeof(*i), GFP_KERNEL); 3535 if (i == NULL) 3536 goto done; 3537 3538 err = 0; 3539 for (ml = po->mclist; ml; ml = ml->next) { 3540 if (ml->ifindex == mreq->mr_ifindex && 3541 ml->type == mreq->mr_type && 3542 ml->alen == mreq->mr_alen && 3543 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3544 ml->count++; 3545 /* Free the new element ... */ 3546 kfree(i); 3547 goto done; 3548 } 3549 } 3550 3551 i->type = mreq->mr_type; 3552 i->ifindex = mreq->mr_ifindex; 3553 i->alen = mreq->mr_alen; 3554 memcpy(i->addr, mreq->mr_address, i->alen); 3555 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3556 i->count = 1; 3557 i->next = po->mclist; 3558 po->mclist = i; 3559 err = packet_dev_mc(dev, i, 1); 3560 if (err) { 3561 po->mclist = i->next; 3562 kfree(i); 3563 } 3564 3565 done: 3566 rtnl_unlock(); 3567 return err; 3568 } 3569 3570 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3571 { 3572 struct packet_mclist *ml, **mlp; 3573 3574 rtnl_lock(); 3575 3576 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3577 if (ml->ifindex == mreq->mr_ifindex && 3578 ml->type == mreq->mr_type && 3579 ml->alen == mreq->mr_alen && 3580 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3581 if (--ml->count == 0) { 3582 struct net_device *dev; 3583 *mlp = ml->next; 3584 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3585 if (dev) 3586 packet_dev_mc(dev, ml, -1); 3587 kfree(ml); 3588 } 3589 break; 3590 } 3591 } 3592 rtnl_unlock(); 3593 return 0; 3594 } 3595 3596 static void packet_flush_mclist(struct sock *sk) 3597 { 3598 struct packet_sock *po = pkt_sk(sk); 3599 struct packet_mclist *ml; 3600 3601 if (!po->mclist) 3602 return; 3603 3604 rtnl_lock(); 3605 while ((ml = po->mclist) != NULL) { 3606 struct net_device *dev; 3607 3608 po->mclist = ml->next; 3609 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3610 if (dev != NULL) 3611 packet_dev_mc(dev, ml, -1); 3612 kfree(ml); 3613 } 3614 rtnl_unlock(); 3615 } 3616 3617 static int 3618 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3619 { 3620 struct sock *sk = sock->sk; 3621 struct packet_sock *po = pkt_sk(sk); 3622 int ret; 3623 3624 if (level != SOL_PACKET) 3625 return -ENOPROTOOPT; 3626 3627 switch (optname) { 3628 case PACKET_ADD_MEMBERSHIP: 3629 case PACKET_DROP_MEMBERSHIP: 3630 { 3631 struct packet_mreq_max mreq; 3632 int len = optlen; 3633 memset(&mreq, 0, sizeof(mreq)); 3634 if (len < sizeof(struct packet_mreq)) 3635 return -EINVAL; 3636 if (len > sizeof(mreq)) 3637 len = sizeof(mreq); 3638 if (copy_from_user(&mreq, optval, len)) 3639 return -EFAULT; 3640 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3641 return -EINVAL; 3642 if (optname == PACKET_ADD_MEMBERSHIP) 3643 ret = packet_mc_add(sk, &mreq); 3644 else 3645 ret = packet_mc_drop(sk, &mreq); 3646 return ret; 3647 } 3648 3649 case PACKET_RX_RING: 3650 case PACKET_TX_RING: 3651 { 3652 union tpacket_req_u req_u; 3653 int len; 3654 3655 lock_sock(sk); 3656 switch (po->tp_version) { 3657 case TPACKET_V1: 3658 case TPACKET_V2: 3659 len = sizeof(req_u.req); 3660 break; 3661 case TPACKET_V3: 3662 default: 3663 len = sizeof(req_u.req3); 3664 break; 3665 } 3666 if (optlen < len) { 3667 ret = -EINVAL; 3668 } else { 3669 if (copy_from_user(&req_u.req, optval, len)) 3670 ret = -EFAULT; 3671 else 3672 ret = packet_set_ring(sk, &req_u, 0, 3673 optname == PACKET_TX_RING); 3674 } 3675 release_sock(sk); 3676 return ret; 3677 } 3678 case PACKET_COPY_THRESH: 3679 { 3680 int val; 3681 3682 if (optlen != sizeof(val)) 3683 return -EINVAL; 3684 if (copy_from_user(&val, optval, sizeof(val))) 3685 return -EFAULT; 3686 3687 pkt_sk(sk)->copy_thresh = val; 3688 return 0; 3689 } 3690 case PACKET_VERSION: 3691 { 3692 int val; 3693 3694 if (optlen != sizeof(val)) 3695 return -EINVAL; 3696 if (copy_from_user(&val, optval, sizeof(val))) 3697 return -EFAULT; 3698 switch (val) { 3699 case TPACKET_V1: 3700 case TPACKET_V2: 3701 case TPACKET_V3: 3702 break; 3703 default: 3704 return -EINVAL; 3705 } 3706 lock_sock(sk); 3707 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3708 ret = -EBUSY; 3709 } else { 3710 po->tp_version = val; 3711 ret = 0; 3712 } 3713 release_sock(sk); 3714 return ret; 3715 } 3716 case PACKET_RESERVE: 3717 { 3718 unsigned int val; 3719 3720 if (optlen != sizeof(val)) 3721 return -EINVAL; 3722 if (copy_from_user(&val, optval, sizeof(val))) 3723 return -EFAULT; 3724 if (val > INT_MAX) 3725 return -EINVAL; 3726 lock_sock(sk); 3727 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3728 ret = -EBUSY; 3729 } else { 3730 po->tp_reserve = val; 3731 ret = 0; 3732 } 3733 release_sock(sk); 3734 return ret; 3735 } 3736 case PACKET_LOSS: 3737 { 3738 unsigned int val; 3739 3740 if (optlen != sizeof(val)) 3741 return -EINVAL; 3742 if (copy_from_user(&val, optval, sizeof(val))) 3743 return -EFAULT; 3744 3745 lock_sock(sk); 3746 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3747 ret = -EBUSY; 3748 } else { 3749 po->tp_loss = !!val; 3750 ret = 0; 3751 } 3752 release_sock(sk); 3753 return ret; 3754 } 3755 case PACKET_AUXDATA: 3756 { 3757 int val; 3758 3759 if (optlen < sizeof(val)) 3760 return -EINVAL; 3761 if (copy_from_user(&val, optval, sizeof(val))) 3762 return -EFAULT; 3763 3764 lock_sock(sk); 3765 po->auxdata = !!val; 3766 release_sock(sk); 3767 return 0; 3768 } 3769 case PACKET_ORIGDEV: 3770 { 3771 int val; 3772 3773 if (optlen < sizeof(val)) 3774 return -EINVAL; 3775 if (copy_from_user(&val, optval, sizeof(val))) 3776 return -EFAULT; 3777 3778 lock_sock(sk); 3779 po->origdev = !!val; 3780 release_sock(sk); 3781 return 0; 3782 } 3783 case PACKET_VNET_HDR: 3784 { 3785 int val; 3786 3787 if (sock->type != SOCK_RAW) 3788 return -EINVAL; 3789 if (optlen < sizeof(val)) 3790 return -EINVAL; 3791 if (copy_from_user(&val, optval, sizeof(val))) 3792 return -EFAULT; 3793 3794 lock_sock(sk); 3795 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3796 ret = -EBUSY; 3797 } else { 3798 po->has_vnet_hdr = !!val; 3799 ret = 0; 3800 } 3801 release_sock(sk); 3802 return ret; 3803 } 3804 case PACKET_TIMESTAMP: 3805 { 3806 int val; 3807 3808 if (optlen != sizeof(val)) 3809 return -EINVAL; 3810 if (copy_from_user(&val, optval, sizeof(val))) 3811 return -EFAULT; 3812 3813 po->tp_tstamp = val; 3814 return 0; 3815 } 3816 case PACKET_FANOUT: 3817 { 3818 int val; 3819 3820 if (optlen != sizeof(val)) 3821 return -EINVAL; 3822 if (copy_from_user(&val, optval, sizeof(val))) 3823 return -EFAULT; 3824 3825 return fanout_add(sk, val & 0xffff, val >> 16); 3826 } 3827 case PACKET_FANOUT_DATA: 3828 { 3829 if (!po->fanout) 3830 return -EINVAL; 3831 3832 return fanout_set_data(po, optval, optlen); 3833 } 3834 case PACKET_IGNORE_OUTGOING: 3835 { 3836 int val; 3837 3838 if (optlen != sizeof(val)) 3839 return -EINVAL; 3840 if (copy_from_user(&val, optval, sizeof(val))) 3841 return -EFAULT; 3842 if (val < 0 || val > 1) 3843 return -EINVAL; 3844 3845 po->prot_hook.ignore_outgoing = !!val; 3846 return 0; 3847 } 3848 case PACKET_TX_HAS_OFF: 3849 { 3850 unsigned int val; 3851 3852 if (optlen != sizeof(val)) 3853 return -EINVAL; 3854 if (copy_from_user(&val, optval, sizeof(val))) 3855 return -EFAULT; 3856 3857 lock_sock(sk); 3858 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3859 ret = -EBUSY; 3860 } else { 3861 po->tp_tx_has_off = !!val; 3862 ret = 0; 3863 } 3864 release_sock(sk); 3865 return 0; 3866 } 3867 case PACKET_QDISC_BYPASS: 3868 { 3869 int val; 3870 3871 if (optlen != sizeof(val)) 3872 return -EINVAL; 3873 if (copy_from_user(&val, optval, sizeof(val))) 3874 return -EFAULT; 3875 3876 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; 3877 return 0; 3878 } 3879 default: 3880 return -ENOPROTOOPT; 3881 } 3882 } 3883 3884 static int packet_getsockopt(struct socket *sock, int level, int optname, 3885 char __user *optval, int __user *optlen) 3886 { 3887 int len; 3888 int val, lv = sizeof(val); 3889 struct sock *sk = sock->sk; 3890 struct packet_sock *po = pkt_sk(sk); 3891 void *data = &val; 3892 union tpacket_stats_u st; 3893 struct tpacket_rollover_stats rstats; 3894 3895 if (level != SOL_PACKET) 3896 return -ENOPROTOOPT; 3897 3898 if (get_user(len, optlen)) 3899 return -EFAULT; 3900 3901 if (len < 0) 3902 return -EINVAL; 3903 3904 switch (optname) { 3905 case PACKET_STATISTICS: 3906 spin_lock_bh(&sk->sk_receive_queue.lock); 3907 memcpy(&st, &po->stats, sizeof(st)); 3908 memset(&po->stats, 0, sizeof(po->stats)); 3909 spin_unlock_bh(&sk->sk_receive_queue.lock); 3910 3911 if (po->tp_version == TPACKET_V3) { 3912 lv = sizeof(struct tpacket_stats_v3); 3913 st.stats3.tp_packets += st.stats3.tp_drops; 3914 data = &st.stats3; 3915 } else { 3916 lv = sizeof(struct tpacket_stats); 3917 st.stats1.tp_packets += st.stats1.tp_drops; 3918 data = &st.stats1; 3919 } 3920 3921 break; 3922 case PACKET_AUXDATA: 3923 val = po->auxdata; 3924 break; 3925 case PACKET_ORIGDEV: 3926 val = po->origdev; 3927 break; 3928 case PACKET_VNET_HDR: 3929 val = po->has_vnet_hdr; 3930 break; 3931 case PACKET_VERSION: 3932 val = po->tp_version; 3933 break; 3934 case PACKET_HDRLEN: 3935 if (len > sizeof(int)) 3936 len = sizeof(int); 3937 if (len < sizeof(int)) 3938 return -EINVAL; 3939 if (copy_from_user(&val, optval, len)) 3940 return -EFAULT; 3941 switch (val) { 3942 case TPACKET_V1: 3943 val = sizeof(struct tpacket_hdr); 3944 break; 3945 case TPACKET_V2: 3946 val = sizeof(struct tpacket2_hdr); 3947 break; 3948 case TPACKET_V3: 3949 val = sizeof(struct tpacket3_hdr); 3950 break; 3951 default: 3952 return -EINVAL; 3953 } 3954 break; 3955 case PACKET_RESERVE: 3956 val = po->tp_reserve; 3957 break; 3958 case PACKET_LOSS: 3959 val = po->tp_loss; 3960 break; 3961 case PACKET_TIMESTAMP: 3962 val = po->tp_tstamp; 3963 break; 3964 case PACKET_FANOUT: 3965 val = (po->fanout ? 3966 ((u32)po->fanout->id | 3967 ((u32)po->fanout->type << 16) | 3968 ((u32)po->fanout->flags << 24)) : 3969 0); 3970 break; 3971 case PACKET_IGNORE_OUTGOING: 3972 val = po->prot_hook.ignore_outgoing; 3973 break; 3974 case PACKET_ROLLOVER_STATS: 3975 if (!po->rollover) 3976 return -EINVAL; 3977 rstats.tp_all = atomic_long_read(&po->rollover->num); 3978 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 3979 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 3980 data = &rstats; 3981 lv = sizeof(rstats); 3982 break; 3983 case PACKET_TX_HAS_OFF: 3984 val = po->tp_tx_has_off; 3985 break; 3986 case PACKET_QDISC_BYPASS: 3987 val = packet_use_direct_xmit(po); 3988 break; 3989 default: 3990 return -ENOPROTOOPT; 3991 } 3992 3993 if (len > lv) 3994 len = lv; 3995 if (put_user(len, optlen)) 3996 return -EFAULT; 3997 if (copy_to_user(optval, data, len)) 3998 return -EFAULT; 3999 return 0; 4000 } 4001 4002 4003 #ifdef CONFIG_COMPAT 4004 static int compat_packet_setsockopt(struct socket *sock, int level, int optname, 4005 char __user *optval, unsigned int optlen) 4006 { 4007 struct packet_sock *po = pkt_sk(sock->sk); 4008 4009 if (level != SOL_PACKET) 4010 return -ENOPROTOOPT; 4011 4012 if (optname == PACKET_FANOUT_DATA && 4013 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) { 4014 optval = (char __user *)get_compat_bpf_fprog(optval); 4015 if (!optval) 4016 return -EFAULT; 4017 optlen = sizeof(struct sock_fprog); 4018 } 4019 4020 return packet_setsockopt(sock, level, optname, optval, optlen); 4021 } 4022 #endif 4023 4024 static int packet_notifier(struct notifier_block *this, 4025 unsigned long msg, void *ptr) 4026 { 4027 struct sock *sk; 4028 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4029 struct net *net = dev_net(dev); 4030 4031 rcu_read_lock(); 4032 sk_for_each_rcu(sk, &net->packet.sklist) { 4033 struct packet_sock *po = pkt_sk(sk); 4034 4035 switch (msg) { 4036 case NETDEV_UNREGISTER: 4037 if (po->mclist) 4038 packet_dev_mclist_delete(dev, &po->mclist); 4039 /* fallthrough */ 4040 4041 case NETDEV_DOWN: 4042 if (dev->ifindex == po->ifindex) { 4043 spin_lock(&po->bind_lock); 4044 if (po->running) { 4045 __unregister_prot_hook(sk, false); 4046 sk->sk_err = ENETDOWN; 4047 if (!sock_flag(sk, SOCK_DEAD)) 4048 sk->sk_error_report(sk); 4049 } 4050 if (msg == NETDEV_UNREGISTER) { 4051 packet_cached_dev_reset(po); 4052 po->ifindex = -1; 4053 if (po->prot_hook.dev) 4054 dev_put(po->prot_hook.dev); 4055 po->prot_hook.dev = NULL; 4056 } 4057 spin_unlock(&po->bind_lock); 4058 } 4059 break; 4060 case NETDEV_UP: 4061 if (dev->ifindex == po->ifindex) { 4062 spin_lock(&po->bind_lock); 4063 if (po->num) 4064 register_prot_hook(sk); 4065 spin_unlock(&po->bind_lock); 4066 } 4067 break; 4068 } 4069 } 4070 rcu_read_unlock(); 4071 return NOTIFY_DONE; 4072 } 4073 4074 4075 static int packet_ioctl(struct socket *sock, unsigned int cmd, 4076 unsigned long arg) 4077 { 4078 struct sock *sk = sock->sk; 4079 4080 switch (cmd) { 4081 case SIOCOUTQ: 4082 { 4083 int amount = sk_wmem_alloc_get(sk); 4084 4085 return put_user(amount, (int __user *)arg); 4086 } 4087 case SIOCINQ: 4088 { 4089 struct sk_buff *skb; 4090 int amount = 0; 4091 4092 spin_lock_bh(&sk->sk_receive_queue.lock); 4093 skb = skb_peek(&sk->sk_receive_queue); 4094 if (skb) 4095 amount = skb->len; 4096 spin_unlock_bh(&sk->sk_receive_queue.lock); 4097 return put_user(amount, (int __user *)arg); 4098 } 4099 #ifdef CONFIG_INET 4100 case SIOCADDRT: 4101 case SIOCDELRT: 4102 case SIOCDARP: 4103 case SIOCGARP: 4104 case SIOCSARP: 4105 case SIOCGIFADDR: 4106 case SIOCSIFADDR: 4107 case SIOCGIFBRDADDR: 4108 case SIOCSIFBRDADDR: 4109 case SIOCGIFNETMASK: 4110 case SIOCSIFNETMASK: 4111 case SIOCGIFDSTADDR: 4112 case SIOCSIFDSTADDR: 4113 case SIOCSIFFLAGS: 4114 return inet_dgram_ops.ioctl(sock, cmd, arg); 4115 #endif 4116 4117 default: 4118 return -ENOIOCTLCMD; 4119 } 4120 return 0; 4121 } 4122 4123 static __poll_t packet_poll(struct file *file, struct socket *sock, 4124 poll_table *wait) 4125 { 4126 struct sock *sk = sock->sk; 4127 struct packet_sock *po = pkt_sk(sk); 4128 __poll_t mask = datagram_poll(file, sock, wait); 4129 4130 spin_lock_bh(&sk->sk_receive_queue.lock); 4131 if (po->rx_ring.pg_vec) { 4132 if (!packet_previous_rx_frame(po, &po->rx_ring, 4133 TP_STATUS_KERNEL)) 4134 mask |= EPOLLIN | EPOLLRDNORM; 4135 } 4136 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 4137 po->pressure = 0; 4138 spin_unlock_bh(&sk->sk_receive_queue.lock); 4139 spin_lock_bh(&sk->sk_write_queue.lock); 4140 if (po->tx_ring.pg_vec) { 4141 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4142 mask |= EPOLLOUT | EPOLLWRNORM; 4143 } 4144 spin_unlock_bh(&sk->sk_write_queue.lock); 4145 return mask; 4146 } 4147 4148 4149 /* Dirty? Well, I still did not learn better way to account 4150 * for user mmaps. 4151 */ 4152 4153 static void packet_mm_open(struct vm_area_struct *vma) 4154 { 4155 struct file *file = vma->vm_file; 4156 struct socket *sock = file->private_data; 4157 struct sock *sk = sock->sk; 4158 4159 if (sk) 4160 atomic_inc(&pkt_sk(sk)->mapped); 4161 } 4162 4163 static void packet_mm_close(struct vm_area_struct *vma) 4164 { 4165 struct file *file = vma->vm_file; 4166 struct socket *sock = file->private_data; 4167 struct sock *sk = sock->sk; 4168 4169 if (sk) 4170 atomic_dec(&pkt_sk(sk)->mapped); 4171 } 4172 4173 static const struct vm_operations_struct packet_mmap_ops = { 4174 .open = packet_mm_open, 4175 .close = packet_mm_close, 4176 }; 4177 4178 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4179 unsigned int len) 4180 { 4181 int i; 4182 4183 for (i = 0; i < len; i++) { 4184 if (likely(pg_vec[i].buffer)) { 4185 if (is_vmalloc_addr(pg_vec[i].buffer)) 4186 vfree(pg_vec[i].buffer); 4187 else 4188 free_pages((unsigned long)pg_vec[i].buffer, 4189 order); 4190 pg_vec[i].buffer = NULL; 4191 } 4192 } 4193 kfree(pg_vec); 4194 } 4195 4196 static char *alloc_one_pg_vec_page(unsigned long order) 4197 { 4198 char *buffer; 4199 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4200 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4201 4202 buffer = (char *) __get_free_pages(gfp_flags, order); 4203 if (buffer) 4204 return buffer; 4205 4206 /* __get_free_pages failed, fall back to vmalloc */ 4207 buffer = vzalloc(array_size((1 << order), PAGE_SIZE)); 4208 if (buffer) 4209 return buffer; 4210 4211 /* vmalloc failed, lets dig into swap here */ 4212 gfp_flags &= ~__GFP_NORETRY; 4213 buffer = (char *) __get_free_pages(gfp_flags, order); 4214 if (buffer) 4215 return buffer; 4216 4217 /* complete and utter failure */ 4218 return NULL; 4219 } 4220 4221 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4222 { 4223 unsigned int block_nr = req->tp_block_nr; 4224 struct pgv *pg_vec; 4225 int i; 4226 4227 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN); 4228 if (unlikely(!pg_vec)) 4229 goto out; 4230 4231 for (i = 0; i < block_nr; i++) { 4232 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4233 if (unlikely(!pg_vec[i].buffer)) 4234 goto out_free_pgvec; 4235 } 4236 4237 out: 4238 return pg_vec; 4239 4240 out_free_pgvec: 4241 free_pg_vec(pg_vec, order, block_nr); 4242 pg_vec = NULL; 4243 goto out; 4244 } 4245 4246 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4247 int closing, int tx_ring) 4248 { 4249 struct pgv *pg_vec = NULL; 4250 struct packet_sock *po = pkt_sk(sk); 4251 int was_running, order = 0; 4252 struct packet_ring_buffer *rb; 4253 struct sk_buff_head *rb_queue; 4254 __be16 num; 4255 int err = -EINVAL; 4256 /* Added to avoid minimal code churn */ 4257 struct tpacket_req *req = &req_u->req; 4258 4259 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4260 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4261 4262 err = -EBUSY; 4263 if (!closing) { 4264 if (atomic_read(&po->mapped)) 4265 goto out; 4266 if (packet_read_pending(rb)) 4267 goto out; 4268 } 4269 4270 if (req->tp_block_nr) { 4271 unsigned int min_frame_size; 4272 4273 /* Sanity tests and some calculations */ 4274 err = -EBUSY; 4275 if (unlikely(rb->pg_vec)) 4276 goto out; 4277 4278 switch (po->tp_version) { 4279 case TPACKET_V1: 4280 po->tp_hdrlen = TPACKET_HDRLEN; 4281 break; 4282 case TPACKET_V2: 4283 po->tp_hdrlen = TPACKET2_HDRLEN; 4284 break; 4285 case TPACKET_V3: 4286 po->tp_hdrlen = TPACKET3_HDRLEN; 4287 break; 4288 } 4289 4290 err = -EINVAL; 4291 if (unlikely((int)req->tp_block_size <= 0)) 4292 goto out; 4293 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4294 goto out; 4295 min_frame_size = po->tp_hdrlen + po->tp_reserve; 4296 if (po->tp_version >= TPACKET_V3 && 4297 req->tp_block_size < 4298 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) 4299 goto out; 4300 if (unlikely(req->tp_frame_size < min_frame_size)) 4301 goto out; 4302 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4303 goto out; 4304 4305 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4306 if (unlikely(rb->frames_per_block == 0)) 4307 goto out; 4308 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) 4309 goto out; 4310 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4311 req->tp_frame_nr)) 4312 goto out; 4313 4314 err = -ENOMEM; 4315 order = get_order(req->tp_block_size); 4316 pg_vec = alloc_pg_vec(req, order); 4317 if (unlikely(!pg_vec)) 4318 goto out; 4319 switch (po->tp_version) { 4320 case TPACKET_V3: 4321 /* Block transmit is not supported yet */ 4322 if (!tx_ring) { 4323 init_prb_bdqc(po, rb, pg_vec, req_u); 4324 } else { 4325 struct tpacket_req3 *req3 = &req_u->req3; 4326 4327 if (req3->tp_retire_blk_tov || 4328 req3->tp_sizeof_priv || 4329 req3->tp_feature_req_word) { 4330 err = -EINVAL; 4331 goto out_free_pg_vec; 4332 } 4333 } 4334 break; 4335 default: 4336 break; 4337 } 4338 } 4339 /* Done */ 4340 else { 4341 err = -EINVAL; 4342 if (unlikely(req->tp_frame_nr)) 4343 goto out; 4344 } 4345 4346 4347 /* Detach socket from network */ 4348 spin_lock(&po->bind_lock); 4349 was_running = po->running; 4350 num = po->num; 4351 if (was_running) { 4352 po->num = 0; 4353 __unregister_prot_hook(sk, false); 4354 } 4355 spin_unlock(&po->bind_lock); 4356 4357 synchronize_net(); 4358 4359 err = -EBUSY; 4360 mutex_lock(&po->pg_vec_lock); 4361 if (closing || atomic_read(&po->mapped) == 0) { 4362 err = 0; 4363 spin_lock_bh(&rb_queue->lock); 4364 swap(rb->pg_vec, pg_vec); 4365 rb->frame_max = (req->tp_frame_nr - 1); 4366 rb->head = 0; 4367 rb->frame_size = req->tp_frame_size; 4368 spin_unlock_bh(&rb_queue->lock); 4369 4370 swap(rb->pg_vec_order, order); 4371 swap(rb->pg_vec_len, req->tp_block_nr); 4372 4373 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4374 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4375 tpacket_rcv : packet_rcv; 4376 skb_queue_purge(rb_queue); 4377 if (atomic_read(&po->mapped)) 4378 pr_err("packet_mmap: vma is busy: %d\n", 4379 atomic_read(&po->mapped)); 4380 } 4381 mutex_unlock(&po->pg_vec_lock); 4382 4383 spin_lock(&po->bind_lock); 4384 if (was_running) { 4385 po->num = num; 4386 register_prot_hook(sk); 4387 } 4388 spin_unlock(&po->bind_lock); 4389 if (pg_vec && (po->tp_version > TPACKET_V2)) { 4390 /* Because we don't support block-based V3 on tx-ring */ 4391 if (!tx_ring) 4392 prb_shutdown_retire_blk_timer(po, rb_queue); 4393 } 4394 4395 out_free_pg_vec: 4396 if (pg_vec) 4397 free_pg_vec(pg_vec, order, req->tp_block_nr); 4398 out: 4399 return err; 4400 } 4401 4402 static int packet_mmap(struct file *file, struct socket *sock, 4403 struct vm_area_struct *vma) 4404 { 4405 struct sock *sk = sock->sk; 4406 struct packet_sock *po = pkt_sk(sk); 4407 unsigned long size, expected_size; 4408 struct packet_ring_buffer *rb; 4409 unsigned long start; 4410 int err = -EINVAL; 4411 int i; 4412 4413 if (vma->vm_pgoff) 4414 return -EINVAL; 4415 4416 mutex_lock(&po->pg_vec_lock); 4417 4418 expected_size = 0; 4419 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4420 if (rb->pg_vec) { 4421 expected_size += rb->pg_vec_len 4422 * rb->pg_vec_pages 4423 * PAGE_SIZE; 4424 } 4425 } 4426 4427 if (expected_size == 0) 4428 goto out; 4429 4430 size = vma->vm_end - vma->vm_start; 4431 if (size != expected_size) 4432 goto out; 4433 4434 start = vma->vm_start; 4435 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4436 if (rb->pg_vec == NULL) 4437 continue; 4438 4439 for (i = 0; i < rb->pg_vec_len; i++) { 4440 struct page *page; 4441 void *kaddr = rb->pg_vec[i].buffer; 4442 int pg_num; 4443 4444 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4445 page = pgv_to_page(kaddr); 4446 err = vm_insert_page(vma, start, page); 4447 if (unlikely(err)) 4448 goto out; 4449 start += PAGE_SIZE; 4450 kaddr += PAGE_SIZE; 4451 } 4452 } 4453 } 4454 4455 atomic_inc(&po->mapped); 4456 vma->vm_ops = &packet_mmap_ops; 4457 err = 0; 4458 4459 out: 4460 mutex_unlock(&po->pg_vec_lock); 4461 return err; 4462 } 4463 4464 static const struct proto_ops packet_ops_spkt = { 4465 .family = PF_PACKET, 4466 .owner = THIS_MODULE, 4467 .release = packet_release, 4468 .bind = packet_bind_spkt, 4469 .connect = sock_no_connect, 4470 .socketpair = sock_no_socketpair, 4471 .accept = sock_no_accept, 4472 .getname = packet_getname_spkt, 4473 .poll = datagram_poll, 4474 .ioctl = packet_ioctl, 4475 .gettstamp = sock_gettstamp, 4476 .listen = sock_no_listen, 4477 .shutdown = sock_no_shutdown, 4478 .setsockopt = sock_no_setsockopt, 4479 .getsockopt = sock_no_getsockopt, 4480 .sendmsg = packet_sendmsg_spkt, 4481 .recvmsg = packet_recvmsg, 4482 .mmap = sock_no_mmap, 4483 .sendpage = sock_no_sendpage, 4484 }; 4485 4486 static const struct proto_ops packet_ops = { 4487 .family = PF_PACKET, 4488 .owner = THIS_MODULE, 4489 .release = packet_release, 4490 .bind = packet_bind, 4491 .connect = sock_no_connect, 4492 .socketpair = sock_no_socketpair, 4493 .accept = sock_no_accept, 4494 .getname = packet_getname, 4495 .poll = packet_poll, 4496 .ioctl = packet_ioctl, 4497 .gettstamp = sock_gettstamp, 4498 .listen = sock_no_listen, 4499 .shutdown = sock_no_shutdown, 4500 .setsockopt = packet_setsockopt, 4501 .getsockopt = packet_getsockopt, 4502 #ifdef CONFIG_COMPAT 4503 .compat_setsockopt = compat_packet_setsockopt, 4504 #endif 4505 .sendmsg = packet_sendmsg, 4506 .recvmsg = packet_recvmsg, 4507 .mmap = packet_mmap, 4508 .sendpage = sock_no_sendpage, 4509 }; 4510 4511 static const struct net_proto_family packet_family_ops = { 4512 .family = PF_PACKET, 4513 .create = packet_create, 4514 .owner = THIS_MODULE, 4515 }; 4516 4517 static struct notifier_block packet_netdev_notifier = { 4518 .notifier_call = packet_notifier, 4519 }; 4520 4521 #ifdef CONFIG_PROC_FS 4522 4523 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4524 __acquires(RCU) 4525 { 4526 struct net *net = seq_file_net(seq); 4527 4528 rcu_read_lock(); 4529 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4530 } 4531 4532 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4533 { 4534 struct net *net = seq_file_net(seq); 4535 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4536 } 4537 4538 static void packet_seq_stop(struct seq_file *seq, void *v) 4539 __releases(RCU) 4540 { 4541 rcu_read_unlock(); 4542 } 4543 4544 static int packet_seq_show(struct seq_file *seq, void *v) 4545 { 4546 if (v == SEQ_START_TOKEN) 4547 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4548 else { 4549 struct sock *s = sk_entry(v); 4550 const struct packet_sock *po = pkt_sk(s); 4551 4552 seq_printf(seq, 4553 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4554 s, 4555 refcount_read(&s->sk_refcnt), 4556 s->sk_type, 4557 ntohs(po->num), 4558 po->ifindex, 4559 po->running, 4560 atomic_read(&s->sk_rmem_alloc), 4561 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4562 sock_i_ino(s)); 4563 } 4564 4565 return 0; 4566 } 4567 4568 static const struct seq_operations packet_seq_ops = { 4569 .start = packet_seq_start, 4570 .next = packet_seq_next, 4571 .stop = packet_seq_stop, 4572 .show = packet_seq_show, 4573 }; 4574 #endif 4575 4576 static int __net_init packet_net_init(struct net *net) 4577 { 4578 mutex_init(&net->packet.sklist_lock); 4579 INIT_HLIST_HEAD(&net->packet.sklist); 4580 4581 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, 4582 sizeof(struct seq_net_private))) 4583 return -ENOMEM; 4584 4585 return 0; 4586 } 4587 4588 static void __net_exit packet_net_exit(struct net *net) 4589 { 4590 remove_proc_entry("packet", net->proc_net); 4591 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); 4592 } 4593 4594 static struct pernet_operations packet_net_ops = { 4595 .init = packet_net_init, 4596 .exit = packet_net_exit, 4597 }; 4598 4599 4600 static void __exit packet_exit(void) 4601 { 4602 unregister_netdevice_notifier(&packet_netdev_notifier); 4603 unregister_pernet_subsys(&packet_net_ops); 4604 sock_unregister(PF_PACKET); 4605 proto_unregister(&packet_proto); 4606 } 4607 4608 static int __init packet_init(void) 4609 { 4610 int rc; 4611 4612 rc = proto_register(&packet_proto, 0); 4613 if (rc) 4614 goto out; 4615 rc = sock_register(&packet_family_ops); 4616 if (rc) 4617 goto out_proto; 4618 rc = register_pernet_subsys(&packet_net_ops); 4619 if (rc) 4620 goto out_sock; 4621 rc = register_netdevice_notifier(&packet_netdev_notifier); 4622 if (rc) 4623 goto out_pernet; 4624 4625 return 0; 4626 4627 out_pernet: 4628 unregister_pernet_subsys(&packet_net_ops); 4629 out_sock: 4630 sock_unregister(PF_PACKET); 4631 out_proto: 4632 proto_unregister(&packet_proto); 4633 out: 4634 return rc; 4635 } 4636 4637 module_init(packet_init); 4638 module_exit(packet_exit); 4639 MODULE_LICENSE("GPL"); 4640 MODULE_ALIAS_NETPROTO(PF_PACKET); 4641