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(const 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(const struct packet_sock *po, 464 const 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 (atomic_read(&po->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 unsigned int len 1007 ) 1008 { 1009 struct tpacket_kbdq_core *pkc; 1010 struct tpacket_block_desc *pbd; 1011 char *curr, *end; 1012 1013 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1014 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1015 1016 /* Queue is frozen when user space is lagging behind */ 1017 if (prb_queue_frozen(pkc)) { 1018 /* 1019 * Check if that last block which caused the queue to freeze, 1020 * is still in_use by user-space. 1021 */ 1022 if (prb_curr_blk_in_use(pbd)) { 1023 /* Can't record this packet */ 1024 return NULL; 1025 } else { 1026 /* 1027 * Ok, the block was released by user-space. 1028 * Now let's open that block. 1029 * opening a block also thaws the queue. 1030 * Thawing is a side effect. 1031 */ 1032 prb_open_block(pkc, pbd); 1033 } 1034 } 1035 1036 smp_mb(); 1037 curr = pkc->nxt_offset; 1038 pkc->skb = skb; 1039 end = (char *)pbd + pkc->kblk_size; 1040 1041 /* first try the current block */ 1042 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1043 prb_fill_curr_block(curr, pkc, pbd, len); 1044 return (void *)curr; 1045 } 1046 1047 /* Ok, close the current block */ 1048 prb_retire_current_block(pkc, po, 0); 1049 1050 /* Now, try to dispatch the next block */ 1051 curr = (char *)prb_dispatch_next_block(pkc, po); 1052 if (curr) { 1053 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1054 prb_fill_curr_block(curr, pkc, pbd, len); 1055 return (void *)curr; 1056 } 1057 1058 /* 1059 * No free blocks are available.user_space hasn't caught up yet. 1060 * Queue was just frozen and now this packet will get dropped. 1061 */ 1062 return NULL; 1063 } 1064 1065 static void *packet_current_rx_frame(struct packet_sock *po, 1066 struct sk_buff *skb, 1067 int status, unsigned int len) 1068 { 1069 char *curr = NULL; 1070 switch (po->tp_version) { 1071 case TPACKET_V1: 1072 case TPACKET_V2: 1073 curr = packet_lookup_frame(po, &po->rx_ring, 1074 po->rx_ring.head, status); 1075 return curr; 1076 case TPACKET_V3: 1077 return __packet_lookup_frame_in_block(po, skb, len); 1078 default: 1079 WARN(1, "TPACKET version not supported\n"); 1080 BUG(); 1081 return NULL; 1082 } 1083 } 1084 1085 static void *prb_lookup_block(const struct packet_sock *po, 1086 const struct packet_ring_buffer *rb, 1087 unsigned int idx, 1088 int status) 1089 { 1090 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1091 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1092 1093 if (status != BLOCK_STATUS(pbd)) 1094 return NULL; 1095 return pbd; 1096 } 1097 1098 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1099 { 1100 unsigned int prev; 1101 if (rb->prb_bdqc.kactive_blk_num) 1102 prev = rb->prb_bdqc.kactive_blk_num-1; 1103 else 1104 prev = rb->prb_bdqc.knum_blocks-1; 1105 return prev; 1106 } 1107 1108 /* Assumes caller has held the rx_queue.lock */ 1109 static void *__prb_previous_block(struct packet_sock *po, 1110 struct packet_ring_buffer *rb, 1111 int status) 1112 { 1113 unsigned int previous = prb_previous_blk_num(rb); 1114 return prb_lookup_block(po, rb, previous, status); 1115 } 1116 1117 static void *packet_previous_rx_frame(struct packet_sock *po, 1118 struct packet_ring_buffer *rb, 1119 int status) 1120 { 1121 if (po->tp_version <= TPACKET_V2) 1122 return packet_previous_frame(po, rb, status); 1123 1124 return __prb_previous_block(po, rb, status); 1125 } 1126 1127 static void packet_increment_rx_head(struct packet_sock *po, 1128 struct packet_ring_buffer *rb) 1129 { 1130 switch (po->tp_version) { 1131 case TPACKET_V1: 1132 case TPACKET_V2: 1133 return packet_increment_head(rb); 1134 case TPACKET_V3: 1135 default: 1136 WARN(1, "TPACKET version not supported.\n"); 1137 BUG(); 1138 return; 1139 } 1140 } 1141 1142 static void *packet_previous_frame(struct packet_sock *po, 1143 struct packet_ring_buffer *rb, 1144 int status) 1145 { 1146 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1147 return packet_lookup_frame(po, rb, previous, status); 1148 } 1149 1150 static void packet_increment_head(struct packet_ring_buffer *buff) 1151 { 1152 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1153 } 1154 1155 static void packet_inc_pending(struct packet_ring_buffer *rb) 1156 { 1157 this_cpu_inc(*rb->pending_refcnt); 1158 } 1159 1160 static void packet_dec_pending(struct packet_ring_buffer *rb) 1161 { 1162 this_cpu_dec(*rb->pending_refcnt); 1163 } 1164 1165 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1166 { 1167 unsigned int refcnt = 0; 1168 int cpu; 1169 1170 /* We don't use pending refcount in rx_ring. */ 1171 if (rb->pending_refcnt == NULL) 1172 return 0; 1173 1174 for_each_possible_cpu(cpu) 1175 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1176 1177 return refcnt; 1178 } 1179 1180 static int packet_alloc_pending(struct packet_sock *po) 1181 { 1182 po->rx_ring.pending_refcnt = NULL; 1183 1184 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1185 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1186 return -ENOBUFS; 1187 1188 return 0; 1189 } 1190 1191 static void packet_free_pending(struct packet_sock *po) 1192 { 1193 free_percpu(po->tx_ring.pending_refcnt); 1194 } 1195 1196 #define ROOM_POW_OFF 2 1197 #define ROOM_NONE 0x0 1198 #define ROOM_LOW 0x1 1199 #define ROOM_NORMAL 0x2 1200 1201 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off) 1202 { 1203 int idx, len; 1204 1205 len = READ_ONCE(po->rx_ring.frame_max) + 1; 1206 idx = READ_ONCE(po->rx_ring.head); 1207 if (pow_off) 1208 idx += len >> pow_off; 1209 if (idx >= len) 1210 idx -= len; 1211 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1212 } 1213 1214 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off) 1215 { 1216 int idx, len; 1217 1218 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks); 1219 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num); 1220 if (pow_off) 1221 idx += len >> pow_off; 1222 if (idx >= len) 1223 idx -= len; 1224 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1225 } 1226 1227 static int __packet_rcv_has_room(const struct packet_sock *po, 1228 const struct sk_buff *skb) 1229 { 1230 const struct sock *sk = &po->sk; 1231 int ret = ROOM_NONE; 1232 1233 if (po->prot_hook.func != tpacket_rcv) { 1234 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1235 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1236 - (skb ? skb->truesize : 0); 1237 1238 if (avail > (rcvbuf >> ROOM_POW_OFF)) 1239 return ROOM_NORMAL; 1240 else if (avail > 0) 1241 return ROOM_LOW; 1242 else 1243 return ROOM_NONE; 1244 } 1245 1246 if (po->tp_version == TPACKET_V3) { 1247 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1248 ret = ROOM_NORMAL; 1249 else if (__tpacket_v3_has_room(po, 0)) 1250 ret = ROOM_LOW; 1251 } else { 1252 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1253 ret = ROOM_NORMAL; 1254 else if (__tpacket_has_room(po, 0)) 1255 ret = ROOM_LOW; 1256 } 1257 1258 return ret; 1259 } 1260 1261 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1262 { 1263 int pressure, ret; 1264 1265 ret = __packet_rcv_has_room(po, skb); 1266 pressure = ret != ROOM_NORMAL; 1267 1268 if (READ_ONCE(po->pressure) != pressure) 1269 WRITE_ONCE(po->pressure, pressure); 1270 1271 return ret; 1272 } 1273 1274 static void packet_rcv_try_clear_pressure(struct packet_sock *po) 1275 { 1276 if (READ_ONCE(po->pressure) && 1277 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 1278 WRITE_ONCE(po->pressure, 0); 1279 } 1280 1281 static void packet_sock_destruct(struct sock *sk) 1282 { 1283 skb_queue_purge(&sk->sk_error_queue); 1284 1285 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1286 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 1287 1288 if (!sock_flag(sk, SOCK_DEAD)) { 1289 pr_err("Attempt to release alive packet socket: %p\n", sk); 1290 return; 1291 } 1292 1293 sk_refcnt_debug_dec(sk); 1294 } 1295 1296 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1297 { 1298 u32 rxhash; 1299 int i, count = 0; 1300 1301 rxhash = skb_get_hash(skb); 1302 for (i = 0; i < ROLLOVER_HLEN; i++) 1303 if (po->rollover->history[i] == rxhash) 1304 count++; 1305 1306 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash; 1307 return count > (ROLLOVER_HLEN >> 1); 1308 } 1309 1310 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1311 struct sk_buff *skb, 1312 unsigned int num) 1313 { 1314 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1315 } 1316 1317 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1318 struct sk_buff *skb, 1319 unsigned int num) 1320 { 1321 unsigned int val = atomic_inc_return(&f->rr_cur); 1322 1323 return val % num; 1324 } 1325 1326 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1327 struct sk_buff *skb, 1328 unsigned int num) 1329 { 1330 return smp_processor_id() % num; 1331 } 1332 1333 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1334 struct sk_buff *skb, 1335 unsigned int num) 1336 { 1337 return prandom_u32_max(num); 1338 } 1339 1340 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1341 struct sk_buff *skb, 1342 unsigned int idx, bool try_self, 1343 unsigned int num) 1344 { 1345 struct packet_sock *po, *po_next, *po_skip = NULL; 1346 unsigned int i, j, room = ROOM_NONE; 1347 1348 po = pkt_sk(f->arr[idx]); 1349 1350 if (try_self) { 1351 room = packet_rcv_has_room(po, skb); 1352 if (room == ROOM_NORMAL || 1353 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1354 return idx; 1355 po_skip = po; 1356 } 1357 1358 i = j = min_t(int, po->rollover->sock, num - 1); 1359 do { 1360 po_next = pkt_sk(f->arr[i]); 1361 if (po_next != po_skip && !READ_ONCE(po_next->pressure) && 1362 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1363 if (i != j) 1364 po->rollover->sock = i; 1365 atomic_long_inc(&po->rollover->num); 1366 if (room == ROOM_LOW) 1367 atomic_long_inc(&po->rollover->num_huge); 1368 return i; 1369 } 1370 1371 if (++i == num) 1372 i = 0; 1373 } while (i != j); 1374 1375 atomic_long_inc(&po->rollover->num_failed); 1376 return idx; 1377 } 1378 1379 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1380 struct sk_buff *skb, 1381 unsigned int num) 1382 { 1383 return skb_get_queue_mapping(skb) % num; 1384 } 1385 1386 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1387 struct sk_buff *skb, 1388 unsigned int num) 1389 { 1390 struct bpf_prog *prog; 1391 unsigned int ret = 0; 1392 1393 rcu_read_lock(); 1394 prog = rcu_dereference(f->bpf_prog); 1395 if (prog) 1396 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1397 rcu_read_unlock(); 1398 1399 return ret; 1400 } 1401 1402 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1403 { 1404 return f->flags & (flag >> 8); 1405 } 1406 1407 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1408 struct packet_type *pt, struct net_device *orig_dev) 1409 { 1410 struct packet_fanout *f = pt->af_packet_priv; 1411 unsigned int num = READ_ONCE(f->num_members); 1412 struct net *net = read_pnet(&f->net); 1413 struct packet_sock *po; 1414 unsigned int idx; 1415 1416 if (!net_eq(dev_net(dev), net) || !num) { 1417 kfree_skb(skb); 1418 return 0; 1419 } 1420 1421 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1422 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1423 if (!skb) 1424 return 0; 1425 } 1426 switch (f->type) { 1427 case PACKET_FANOUT_HASH: 1428 default: 1429 idx = fanout_demux_hash(f, skb, num); 1430 break; 1431 case PACKET_FANOUT_LB: 1432 idx = fanout_demux_lb(f, skb, num); 1433 break; 1434 case PACKET_FANOUT_CPU: 1435 idx = fanout_demux_cpu(f, skb, num); 1436 break; 1437 case PACKET_FANOUT_RND: 1438 idx = fanout_demux_rnd(f, skb, num); 1439 break; 1440 case PACKET_FANOUT_QM: 1441 idx = fanout_demux_qm(f, skb, num); 1442 break; 1443 case PACKET_FANOUT_ROLLOVER: 1444 idx = fanout_demux_rollover(f, skb, 0, false, num); 1445 break; 1446 case PACKET_FANOUT_CBPF: 1447 case PACKET_FANOUT_EBPF: 1448 idx = fanout_demux_bpf(f, skb, num); 1449 break; 1450 } 1451 1452 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1453 idx = fanout_demux_rollover(f, skb, idx, true, num); 1454 1455 po = pkt_sk(f->arr[idx]); 1456 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1457 } 1458 1459 DEFINE_MUTEX(fanout_mutex); 1460 EXPORT_SYMBOL_GPL(fanout_mutex); 1461 static LIST_HEAD(fanout_list); 1462 static u16 fanout_next_id; 1463 1464 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1465 { 1466 struct packet_fanout *f = po->fanout; 1467 1468 spin_lock(&f->lock); 1469 f->arr[f->num_members] = sk; 1470 smp_wmb(); 1471 f->num_members++; 1472 if (f->num_members == 1) 1473 dev_add_pack(&f->prot_hook); 1474 spin_unlock(&f->lock); 1475 } 1476 1477 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1478 { 1479 struct packet_fanout *f = po->fanout; 1480 int i; 1481 1482 spin_lock(&f->lock); 1483 for (i = 0; i < f->num_members; i++) { 1484 if (f->arr[i] == sk) 1485 break; 1486 } 1487 BUG_ON(i >= f->num_members); 1488 f->arr[i] = f->arr[f->num_members - 1]; 1489 f->num_members--; 1490 if (f->num_members == 0) 1491 __dev_remove_pack(&f->prot_hook); 1492 spin_unlock(&f->lock); 1493 } 1494 1495 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1496 { 1497 if (sk->sk_family != PF_PACKET) 1498 return false; 1499 1500 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1501 } 1502 1503 static void fanout_init_data(struct packet_fanout *f) 1504 { 1505 switch (f->type) { 1506 case PACKET_FANOUT_LB: 1507 atomic_set(&f->rr_cur, 0); 1508 break; 1509 case PACKET_FANOUT_CBPF: 1510 case PACKET_FANOUT_EBPF: 1511 RCU_INIT_POINTER(f->bpf_prog, NULL); 1512 break; 1513 } 1514 } 1515 1516 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1517 { 1518 struct bpf_prog *old; 1519 1520 spin_lock(&f->lock); 1521 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1522 rcu_assign_pointer(f->bpf_prog, new); 1523 spin_unlock(&f->lock); 1524 1525 if (old) { 1526 synchronize_net(); 1527 bpf_prog_destroy(old); 1528 } 1529 } 1530 1531 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data, 1532 unsigned int len) 1533 { 1534 struct bpf_prog *new; 1535 struct sock_fprog fprog; 1536 int ret; 1537 1538 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1539 return -EPERM; 1540 if (len != sizeof(fprog)) 1541 return -EINVAL; 1542 if (copy_from_user(&fprog, data, len)) 1543 return -EFAULT; 1544 1545 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1546 if (ret) 1547 return ret; 1548 1549 __fanout_set_data_bpf(po->fanout, new); 1550 return 0; 1551 } 1552 1553 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data, 1554 unsigned int len) 1555 { 1556 struct bpf_prog *new; 1557 u32 fd; 1558 1559 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1560 return -EPERM; 1561 if (len != sizeof(fd)) 1562 return -EINVAL; 1563 if (copy_from_user(&fd, data, len)) 1564 return -EFAULT; 1565 1566 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1567 if (IS_ERR(new)) 1568 return PTR_ERR(new); 1569 1570 __fanout_set_data_bpf(po->fanout, new); 1571 return 0; 1572 } 1573 1574 static int fanout_set_data(struct packet_sock *po, char __user *data, 1575 unsigned int len) 1576 { 1577 switch (po->fanout->type) { 1578 case PACKET_FANOUT_CBPF: 1579 return fanout_set_data_cbpf(po, data, len); 1580 case PACKET_FANOUT_EBPF: 1581 return fanout_set_data_ebpf(po, data, len); 1582 default: 1583 return -EINVAL; 1584 } 1585 } 1586 1587 static void fanout_release_data(struct packet_fanout *f) 1588 { 1589 switch (f->type) { 1590 case PACKET_FANOUT_CBPF: 1591 case PACKET_FANOUT_EBPF: 1592 __fanout_set_data_bpf(f, NULL); 1593 } 1594 } 1595 1596 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) 1597 { 1598 struct packet_fanout *f; 1599 1600 list_for_each_entry(f, &fanout_list, list) { 1601 if (f->id == candidate_id && 1602 read_pnet(&f->net) == sock_net(sk)) { 1603 return false; 1604 } 1605 } 1606 return true; 1607 } 1608 1609 static bool fanout_find_new_id(struct sock *sk, u16 *new_id) 1610 { 1611 u16 id = fanout_next_id; 1612 1613 do { 1614 if (__fanout_id_is_free(sk, id)) { 1615 *new_id = id; 1616 fanout_next_id = id + 1; 1617 return true; 1618 } 1619 1620 id++; 1621 } while (id != fanout_next_id); 1622 1623 return false; 1624 } 1625 1626 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1627 { 1628 struct packet_rollover *rollover = NULL; 1629 struct packet_sock *po = pkt_sk(sk); 1630 struct packet_fanout *f, *match; 1631 u8 type = type_flags & 0xff; 1632 u8 flags = type_flags >> 8; 1633 int err; 1634 1635 switch (type) { 1636 case PACKET_FANOUT_ROLLOVER: 1637 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1638 return -EINVAL; 1639 case PACKET_FANOUT_HASH: 1640 case PACKET_FANOUT_LB: 1641 case PACKET_FANOUT_CPU: 1642 case PACKET_FANOUT_RND: 1643 case PACKET_FANOUT_QM: 1644 case PACKET_FANOUT_CBPF: 1645 case PACKET_FANOUT_EBPF: 1646 break; 1647 default: 1648 return -EINVAL; 1649 } 1650 1651 mutex_lock(&fanout_mutex); 1652 1653 err = -EALREADY; 1654 if (po->fanout) 1655 goto out; 1656 1657 if (type == PACKET_FANOUT_ROLLOVER || 1658 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1659 err = -ENOMEM; 1660 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); 1661 if (!rollover) 1662 goto out; 1663 atomic_long_set(&rollover->num, 0); 1664 atomic_long_set(&rollover->num_huge, 0); 1665 atomic_long_set(&rollover->num_failed, 0); 1666 } 1667 1668 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { 1669 if (id != 0) { 1670 err = -EINVAL; 1671 goto out; 1672 } 1673 if (!fanout_find_new_id(sk, &id)) { 1674 err = -ENOMEM; 1675 goto out; 1676 } 1677 /* ephemeral flag for the first socket in the group: drop it */ 1678 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); 1679 } 1680 1681 match = NULL; 1682 list_for_each_entry(f, &fanout_list, list) { 1683 if (f->id == id && 1684 read_pnet(&f->net) == sock_net(sk)) { 1685 match = f; 1686 break; 1687 } 1688 } 1689 err = -EINVAL; 1690 if (match && match->flags != flags) 1691 goto out; 1692 if (!match) { 1693 err = -ENOMEM; 1694 match = kzalloc(sizeof(*match), GFP_KERNEL); 1695 if (!match) 1696 goto out; 1697 write_pnet(&match->net, sock_net(sk)); 1698 match->id = id; 1699 match->type = type; 1700 match->flags = flags; 1701 INIT_LIST_HEAD(&match->list); 1702 spin_lock_init(&match->lock); 1703 refcount_set(&match->sk_ref, 0); 1704 fanout_init_data(match); 1705 match->prot_hook.type = po->prot_hook.type; 1706 match->prot_hook.dev = po->prot_hook.dev; 1707 match->prot_hook.func = packet_rcv_fanout; 1708 match->prot_hook.af_packet_priv = match; 1709 match->prot_hook.id_match = match_fanout_group; 1710 list_add(&match->list, &fanout_list); 1711 } 1712 err = -EINVAL; 1713 1714 spin_lock(&po->bind_lock); 1715 if (po->running && 1716 match->type == type && 1717 match->prot_hook.type == po->prot_hook.type && 1718 match->prot_hook.dev == po->prot_hook.dev) { 1719 err = -ENOSPC; 1720 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1721 __dev_remove_pack(&po->prot_hook); 1722 po->fanout = match; 1723 po->rollover = rollover; 1724 rollover = NULL; 1725 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); 1726 __fanout_link(sk, po); 1727 err = 0; 1728 } 1729 } 1730 spin_unlock(&po->bind_lock); 1731 1732 if (err && !refcount_read(&match->sk_ref)) { 1733 list_del(&match->list); 1734 kfree(match); 1735 } 1736 1737 out: 1738 kfree(rollover); 1739 mutex_unlock(&fanout_mutex); 1740 return err; 1741 } 1742 1743 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes 1744 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. 1745 * It is the responsibility of the caller to call fanout_release_data() and 1746 * free the returned packet_fanout (after synchronize_net()) 1747 */ 1748 static struct packet_fanout *fanout_release(struct sock *sk) 1749 { 1750 struct packet_sock *po = pkt_sk(sk); 1751 struct packet_fanout *f; 1752 1753 mutex_lock(&fanout_mutex); 1754 f = po->fanout; 1755 if (f) { 1756 po->fanout = NULL; 1757 1758 if (refcount_dec_and_test(&f->sk_ref)) 1759 list_del(&f->list); 1760 else 1761 f = NULL; 1762 } 1763 mutex_unlock(&fanout_mutex); 1764 1765 return f; 1766 } 1767 1768 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1769 struct sk_buff *skb) 1770 { 1771 /* Earlier code assumed this would be a VLAN pkt, double-check 1772 * this now that we have the actual packet in hand. We can only 1773 * do this check on Ethernet devices. 1774 */ 1775 if (unlikely(dev->type != ARPHRD_ETHER)) 1776 return false; 1777 1778 skb_reset_mac_header(skb); 1779 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1780 } 1781 1782 static const struct proto_ops packet_ops; 1783 1784 static const struct proto_ops packet_ops_spkt; 1785 1786 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1787 struct packet_type *pt, struct net_device *orig_dev) 1788 { 1789 struct sock *sk; 1790 struct sockaddr_pkt *spkt; 1791 1792 /* 1793 * When we registered the protocol we saved the socket in the data 1794 * field for just this event. 1795 */ 1796 1797 sk = pt->af_packet_priv; 1798 1799 /* 1800 * Yank back the headers [hope the device set this 1801 * right or kerboom...] 1802 * 1803 * Incoming packets have ll header pulled, 1804 * push it back. 1805 * 1806 * For outgoing ones skb->data == skb_mac_header(skb) 1807 * so that this procedure is noop. 1808 */ 1809 1810 if (skb->pkt_type == PACKET_LOOPBACK) 1811 goto out; 1812 1813 if (!net_eq(dev_net(dev), sock_net(sk))) 1814 goto out; 1815 1816 skb = skb_share_check(skb, GFP_ATOMIC); 1817 if (skb == NULL) 1818 goto oom; 1819 1820 /* drop any routing info */ 1821 skb_dst_drop(skb); 1822 1823 /* drop conntrack reference */ 1824 nf_reset(skb); 1825 1826 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1827 1828 skb_push(skb, skb->data - skb_mac_header(skb)); 1829 1830 /* 1831 * The SOCK_PACKET socket receives _all_ frames. 1832 */ 1833 1834 spkt->spkt_family = dev->type; 1835 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1836 spkt->spkt_protocol = skb->protocol; 1837 1838 /* 1839 * Charge the memory to the socket. This is done specifically 1840 * to prevent sockets using all the memory up. 1841 */ 1842 1843 if (sock_queue_rcv_skb(sk, skb) == 0) 1844 return 0; 1845 1846 out: 1847 kfree_skb(skb); 1848 oom: 1849 return 0; 1850 } 1851 1852 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock) 1853 { 1854 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && 1855 sock->type == SOCK_RAW) { 1856 skb_reset_mac_header(skb); 1857 skb->protocol = dev_parse_header_protocol(skb); 1858 } 1859 1860 skb_probe_transport_header(skb); 1861 } 1862 1863 /* 1864 * Output a raw packet to a device layer. This bypasses all the other 1865 * protocol layers and you must therefore supply it with a complete frame 1866 */ 1867 1868 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1869 size_t len) 1870 { 1871 struct sock *sk = sock->sk; 1872 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1873 struct sk_buff *skb = NULL; 1874 struct net_device *dev; 1875 struct sockcm_cookie sockc; 1876 __be16 proto = 0; 1877 int err; 1878 int extra_len = 0; 1879 1880 /* 1881 * Get and verify the address. 1882 */ 1883 1884 if (saddr) { 1885 if (msg->msg_namelen < sizeof(struct sockaddr)) 1886 return -EINVAL; 1887 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1888 proto = saddr->spkt_protocol; 1889 } else 1890 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1891 1892 /* 1893 * Find the device first to size check it 1894 */ 1895 1896 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1897 retry: 1898 rcu_read_lock(); 1899 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1900 err = -ENODEV; 1901 if (dev == NULL) 1902 goto out_unlock; 1903 1904 err = -ENETDOWN; 1905 if (!(dev->flags & IFF_UP)) 1906 goto out_unlock; 1907 1908 /* 1909 * You may not queue a frame bigger than the mtu. This is the lowest level 1910 * raw protocol and you must do your own fragmentation at this level. 1911 */ 1912 1913 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1914 if (!netif_supports_nofcs(dev)) { 1915 err = -EPROTONOSUPPORT; 1916 goto out_unlock; 1917 } 1918 extra_len = 4; /* We're doing our own CRC */ 1919 } 1920 1921 err = -EMSGSIZE; 1922 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1923 goto out_unlock; 1924 1925 if (!skb) { 1926 size_t reserved = LL_RESERVED_SPACE(dev); 1927 int tlen = dev->needed_tailroom; 1928 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1929 1930 rcu_read_unlock(); 1931 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1932 if (skb == NULL) 1933 return -ENOBUFS; 1934 /* FIXME: Save some space for broken drivers that write a hard 1935 * header at transmission time by themselves. PPP is the notable 1936 * one here. This should really be fixed at the driver level. 1937 */ 1938 skb_reserve(skb, reserved); 1939 skb_reset_network_header(skb); 1940 1941 /* Try to align data part correctly */ 1942 if (hhlen) { 1943 skb->data -= hhlen; 1944 skb->tail -= hhlen; 1945 if (len < hhlen) 1946 skb_reset_network_header(skb); 1947 } 1948 err = memcpy_from_msg(skb_put(skb, len), msg, len); 1949 if (err) 1950 goto out_free; 1951 goto retry; 1952 } 1953 1954 if (!dev_validate_header(dev, skb->data, len)) { 1955 err = -EINVAL; 1956 goto out_unlock; 1957 } 1958 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 1959 !packet_extra_vlan_len_allowed(dev, skb)) { 1960 err = -EMSGSIZE; 1961 goto out_unlock; 1962 } 1963 1964 sockcm_init(&sockc, sk); 1965 if (msg->msg_controllen) { 1966 err = sock_cmsg_send(sk, msg, &sockc); 1967 if (unlikely(err)) 1968 goto out_unlock; 1969 } 1970 1971 skb->protocol = proto; 1972 skb->dev = dev; 1973 skb->priority = sk->sk_priority; 1974 skb->mark = sk->sk_mark; 1975 skb->tstamp = sockc.transmit_time; 1976 1977 skb_setup_tx_timestamp(skb, sockc.tsflags); 1978 1979 if (unlikely(extra_len == 4)) 1980 skb->no_fcs = 1; 1981 1982 packet_parse_headers(skb, sock); 1983 1984 dev_queue_xmit(skb); 1985 rcu_read_unlock(); 1986 return len; 1987 1988 out_unlock: 1989 rcu_read_unlock(); 1990 out_free: 1991 kfree_skb(skb); 1992 return err; 1993 } 1994 1995 static unsigned int run_filter(struct sk_buff *skb, 1996 const struct sock *sk, 1997 unsigned int res) 1998 { 1999 struct sk_filter *filter; 2000 2001 rcu_read_lock(); 2002 filter = rcu_dereference(sk->sk_filter); 2003 if (filter != NULL) 2004 res = bpf_prog_run_clear_cb(filter->prog, skb); 2005 rcu_read_unlock(); 2006 2007 return res; 2008 } 2009 2010 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2011 size_t *len) 2012 { 2013 struct virtio_net_hdr vnet_hdr; 2014 2015 if (*len < sizeof(vnet_hdr)) 2016 return -EINVAL; 2017 *len -= sizeof(vnet_hdr); 2018 2019 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0)) 2020 return -EINVAL; 2021 2022 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 2023 } 2024 2025 /* 2026 * This function makes lazy skb cloning in hope that most of packets 2027 * are discarded by BPF. 2028 * 2029 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2030 * and skb->cb are mangled. It works because (and until) packets 2031 * falling here are owned by current CPU. Output packets are cloned 2032 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2033 * sequencially, so that if we return skb to original state on exit, 2034 * we will not harm anyone. 2035 */ 2036 2037 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2038 struct packet_type *pt, struct net_device *orig_dev) 2039 { 2040 struct sock *sk; 2041 struct sockaddr_ll *sll; 2042 struct packet_sock *po; 2043 u8 *skb_head = skb->data; 2044 int skb_len = skb->len; 2045 unsigned int snaplen, res; 2046 bool is_drop_n_account = false; 2047 2048 if (skb->pkt_type == PACKET_LOOPBACK) 2049 goto drop; 2050 2051 sk = pt->af_packet_priv; 2052 po = pkt_sk(sk); 2053 2054 if (!net_eq(dev_net(dev), sock_net(sk))) 2055 goto drop; 2056 2057 skb->dev = dev; 2058 2059 if (dev->header_ops) { 2060 /* The device has an explicit notion of ll header, 2061 * exported to higher levels. 2062 * 2063 * Otherwise, the device hides details of its frame 2064 * structure, so that corresponding packet head is 2065 * never delivered to user. 2066 */ 2067 if (sk->sk_type != SOCK_DGRAM) 2068 skb_push(skb, skb->data - skb_mac_header(skb)); 2069 else if (skb->pkt_type == PACKET_OUTGOING) { 2070 /* Special case: outgoing packets have ll header at head */ 2071 skb_pull(skb, skb_network_offset(skb)); 2072 } 2073 } 2074 2075 snaplen = skb->len; 2076 2077 res = run_filter(skb, sk, snaplen); 2078 if (!res) 2079 goto drop_n_restore; 2080 if (snaplen > res) 2081 snaplen = res; 2082 2083 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2084 goto drop_n_acct; 2085 2086 if (skb_shared(skb)) { 2087 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2088 if (nskb == NULL) 2089 goto drop_n_acct; 2090 2091 if (skb_head != skb->data) { 2092 skb->data = skb_head; 2093 skb->len = skb_len; 2094 } 2095 consume_skb(skb); 2096 skb = nskb; 2097 } 2098 2099 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2100 2101 sll = &PACKET_SKB_CB(skb)->sa.ll; 2102 sll->sll_hatype = dev->type; 2103 sll->sll_pkttype = skb->pkt_type; 2104 if (unlikely(po->origdev)) 2105 sll->sll_ifindex = orig_dev->ifindex; 2106 else 2107 sll->sll_ifindex = dev->ifindex; 2108 2109 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2110 2111 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2112 * Use their space for storing the original skb length. 2113 */ 2114 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2115 2116 if (pskb_trim(skb, snaplen)) 2117 goto drop_n_acct; 2118 2119 skb_set_owner_r(skb, sk); 2120 skb->dev = NULL; 2121 skb_dst_drop(skb); 2122 2123 /* drop conntrack reference */ 2124 nf_reset(skb); 2125 2126 spin_lock(&sk->sk_receive_queue.lock); 2127 po->stats.stats1.tp_packets++; 2128 sock_skb_set_dropcount(sk, skb); 2129 __skb_queue_tail(&sk->sk_receive_queue, skb); 2130 spin_unlock(&sk->sk_receive_queue.lock); 2131 sk->sk_data_ready(sk); 2132 return 0; 2133 2134 drop_n_acct: 2135 is_drop_n_account = true; 2136 atomic_inc(&po->tp_drops); 2137 atomic_inc(&sk->sk_drops); 2138 2139 drop_n_restore: 2140 if (skb_head != skb->data && skb_shared(skb)) { 2141 skb->data = skb_head; 2142 skb->len = skb_len; 2143 } 2144 drop: 2145 if (!is_drop_n_account) 2146 consume_skb(skb); 2147 else 2148 kfree_skb(skb); 2149 return 0; 2150 } 2151 2152 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2153 struct packet_type *pt, struct net_device *orig_dev) 2154 { 2155 struct sock *sk; 2156 struct packet_sock *po; 2157 struct sockaddr_ll *sll; 2158 union tpacket_uhdr h; 2159 u8 *skb_head = skb->data; 2160 int skb_len = skb->len; 2161 unsigned int snaplen, res; 2162 unsigned long status = TP_STATUS_USER; 2163 unsigned short macoff, netoff, hdrlen; 2164 struct sk_buff *copy_skb = NULL; 2165 struct timespec ts; 2166 __u32 ts_status; 2167 bool is_drop_n_account = false; 2168 bool do_vnet = false; 2169 2170 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2171 * We may add members to them until current aligned size without forcing 2172 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2173 */ 2174 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2175 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2176 2177 if (skb->pkt_type == PACKET_LOOPBACK) 2178 goto drop; 2179 2180 sk = pt->af_packet_priv; 2181 po = pkt_sk(sk); 2182 2183 if (!net_eq(dev_net(dev), sock_net(sk))) 2184 goto drop; 2185 2186 if (dev->header_ops) { 2187 if (sk->sk_type != SOCK_DGRAM) 2188 skb_push(skb, skb->data - skb_mac_header(skb)); 2189 else if (skb->pkt_type == PACKET_OUTGOING) { 2190 /* Special case: outgoing packets have ll header at head */ 2191 skb_pull(skb, skb_network_offset(skb)); 2192 } 2193 } 2194 2195 snaplen = skb->len; 2196 2197 res = run_filter(skb, sk, snaplen); 2198 if (!res) 2199 goto drop_n_restore; 2200 2201 /* If we are flooded, just give up */ 2202 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) { 2203 atomic_inc(&po->tp_drops); 2204 goto drop_n_restore; 2205 } 2206 2207 if (skb->ip_summed == CHECKSUM_PARTIAL) 2208 status |= TP_STATUS_CSUMNOTREADY; 2209 else if (skb->pkt_type != PACKET_OUTGOING && 2210 (skb->ip_summed == CHECKSUM_COMPLETE || 2211 skb_csum_unnecessary(skb))) 2212 status |= TP_STATUS_CSUM_VALID; 2213 2214 if (snaplen > res) 2215 snaplen = res; 2216 2217 if (sk->sk_type == SOCK_DGRAM) { 2218 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2219 po->tp_reserve; 2220 } else { 2221 unsigned int maclen = skb_network_offset(skb); 2222 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2223 (maclen < 16 ? 16 : maclen)) + 2224 po->tp_reserve; 2225 if (po->has_vnet_hdr) { 2226 netoff += sizeof(struct virtio_net_hdr); 2227 do_vnet = true; 2228 } 2229 macoff = netoff - maclen; 2230 } 2231 if (po->tp_version <= TPACKET_V2) { 2232 if (macoff + snaplen > po->rx_ring.frame_size) { 2233 if (po->copy_thresh && 2234 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2235 if (skb_shared(skb)) { 2236 copy_skb = skb_clone(skb, GFP_ATOMIC); 2237 } else { 2238 copy_skb = skb_get(skb); 2239 skb_head = skb->data; 2240 } 2241 if (copy_skb) 2242 skb_set_owner_r(copy_skb, sk); 2243 } 2244 snaplen = po->rx_ring.frame_size - macoff; 2245 if ((int)snaplen < 0) { 2246 snaplen = 0; 2247 do_vnet = false; 2248 } 2249 } 2250 } else if (unlikely(macoff + snaplen > 2251 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2252 u32 nval; 2253 2254 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2255 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2256 snaplen, nval, macoff); 2257 snaplen = nval; 2258 if (unlikely((int)snaplen < 0)) { 2259 snaplen = 0; 2260 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2261 do_vnet = false; 2262 } 2263 } 2264 spin_lock(&sk->sk_receive_queue.lock); 2265 h.raw = packet_current_rx_frame(po, skb, 2266 TP_STATUS_KERNEL, (macoff+snaplen)); 2267 if (!h.raw) 2268 goto drop_n_account; 2269 if (po->tp_version <= TPACKET_V2) { 2270 packet_increment_rx_head(po, &po->rx_ring); 2271 /* 2272 * LOSING will be reported till you read the stats, 2273 * because it's COR - Clear On Read. 2274 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2275 * at packet level. 2276 */ 2277 if (atomic_read(&po->tp_drops)) 2278 status |= TP_STATUS_LOSING; 2279 } 2280 2281 if (do_vnet && 2282 virtio_net_hdr_from_skb(skb, h.raw + macoff - 2283 sizeof(struct virtio_net_hdr), 2284 vio_le(), true, 0)) 2285 goto drop_n_account; 2286 2287 po->stats.stats1.tp_packets++; 2288 if (copy_skb) { 2289 status |= TP_STATUS_COPY; 2290 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2291 } 2292 spin_unlock(&sk->sk_receive_queue.lock); 2293 2294 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2295 2296 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 2297 getnstimeofday(&ts); 2298 2299 status |= ts_status; 2300 2301 switch (po->tp_version) { 2302 case TPACKET_V1: 2303 h.h1->tp_len = skb->len; 2304 h.h1->tp_snaplen = snaplen; 2305 h.h1->tp_mac = macoff; 2306 h.h1->tp_net = netoff; 2307 h.h1->tp_sec = ts.tv_sec; 2308 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2309 hdrlen = sizeof(*h.h1); 2310 break; 2311 case TPACKET_V2: 2312 h.h2->tp_len = skb->len; 2313 h.h2->tp_snaplen = snaplen; 2314 h.h2->tp_mac = macoff; 2315 h.h2->tp_net = netoff; 2316 h.h2->tp_sec = ts.tv_sec; 2317 h.h2->tp_nsec = ts.tv_nsec; 2318 if (skb_vlan_tag_present(skb)) { 2319 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2320 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2321 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2322 } else { 2323 h.h2->tp_vlan_tci = 0; 2324 h.h2->tp_vlan_tpid = 0; 2325 } 2326 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2327 hdrlen = sizeof(*h.h2); 2328 break; 2329 case TPACKET_V3: 2330 /* tp_nxt_offset,vlan are already populated above. 2331 * So DONT clear those fields here 2332 */ 2333 h.h3->tp_status |= status; 2334 h.h3->tp_len = skb->len; 2335 h.h3->tp_snaplen = snaplen; 2336 h.h3->tp_mac = macoff; 2337 h.h3->tp_net = netoff; 2338 h.h3->tp_sec = ts.tv_sec; 2339 h.h3->tp_nsec = ts.tv_nsec; 2340 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2341 hdrlen = sizeof(*h.h3); 2342 break; 2343 default: 2344 BUG(); 2345 } 2346 2347 sll = h.raw + TPACKET_ALIGN(hdrlen); 2348 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2349 sll->sll_family = AF_PACKET; 2350 sll->sll_hatype = dev->type; 2351 sll->sll_protocol = skb->protocol; 2352 sll->sll_pkttype = skb->pkt_type; 2353 if (unlikely(po->origdev)) 2354 sll->sll_ifindex = orig_dev->ifindex; 2355 else 2356 sll->sll_ifindex = dev->ifindex; 2357 2358 smp_mb(); 2359 2360 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2361 if (po->tp_version <= TPACKET_V2) { 2362 u8 *start, *end; 2363 2364 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2365 macoff + snaplen); 2366 2367 for (start = h.raw; start < end; start += PAGE_SIZE) 2368 flush_dcache_page(pgv_to_page(start)); 2369 } 2370 smp_wmb(); 2371 #endif 2372 2373 if (po->tp_version <= TPACKET_V2) { 2374 __packet_set_status(po, h.raw, status); 2375 sk->sk_data_ready(sk); 2376 } else { 2377 prb_clear_blk_fill_status(&po->rx_ring); 2378 } 2379 2380 drop_n_restore: 2381 if (skb_head != skb->data && skb_shared(skb)) { 2382 skb->data = skb_head; 2383 skb->len = skb_len; 2384 } 2385 drop: 2386 if (!is_drop_n_account) 2387 consume_skb(skb); 2388 else 2389 kfree_skb(skb); 2390 return 0; 2391 2392 drop_n_account: 2393 spin_unlock(&sk->sk_receive_queue.lock); 2394 atomic_inc(&po->tp_drops); 2395 is_drop_n_account = true; 2396 2397 sk->sk_data_ready(sk); 2398 kfree_skb(copy_skb); 2399 goto drop_n_restore; 2400 } 2401 2402 static void tpacket_destruct_skb(struct sk_buff *skb) 2403 { 2404 struct packet_sock *po = pkt_sk(skb->sk); 2405 2406 if (likely(po->tx_ring.pg_vec)) { 2407 void *ph; 2408 __u32 ts; 2409 2410 ph = skb_zcopy_get_nouarg(skb); 2411 packet_dec_pending(&po->tx_ring); 2412 2413 ts = __packet_set_timestamp(po, ph, skb); 2414 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2415 } 2416 2417 sock_wfree(skb); 2418 } 2419 2420 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2421 { 2422 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2423 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2424 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2425 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2426 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2427 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2428 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2429 2430 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2431 return -EINVAL; 2432 2433 return 0; 2434 } 2435 2436 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2437 struct virtio_net_hdr *vnet_hdr) 2438 { 2439 if (*len < sizeof(*vnet_hdr)) 2440 return -EINVAL; 2441 *len -= sizeof(*vnet_hdr); 2442 2443 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2444 return -EFAULT; 2445 2446 return __packet_snd_vnet_parse(vnet_hdr, *len); 2447 } 2448 2449 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2450 void *frame, struct net_device *dev, void *data, int tp_len, 2451 __be16 proto, unsigned char *addr, int hlen, int copylen, 2452 const struct sockcm_cookie *sockc) 2453 { 2454 union tpacket_uhdr ph; 2455 int to_write, offset, len, nr_frags, len_max; 2456 struct socket *sock = po->sk.sk_socket; 2457 struct page *page; 2458 int err; 2459 2460 ph.raw = frame; 2461 2462 skb->protocol = proto; 2463 skb->dev = dev; 2464 skb->priority = po->sk.sk_priority; 2465 skb->mark = po->sk.sk_mark; 2466 skb->tstamp = sockc->transmit_time; 2467 skb_setup_tx_timestamp(skb, sockc->tsflags); 2468 skb_zcopy_set_nouarg(skb, ph.raw); 2469 2470 skb_reserve(skb, hlen); 2471 skb_reset_network_header(skb); 2472 2473 to_write = tp_len; 2474 2475 if (sock->type == SOCK_DGRAM) { 2476 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2477 NULL, tp_len); 2478 if (unlikely(err < 0)) 2479 return -EINVAL; 2480 } else if (copylen) { 2481 int hdrlen = min_t(int, copylen, tp_len); 2482 2483 skb_push(skb, dev->hard_header_len); 2484 skb_put(skb, copylen - dev->hard_header_len); 2485 err = skb_store_bits(skb, 0, data, hdrlen); 2486 if (unlikely(err)) 2487 return err; 2488 if (!dev_validate_header(dev, skb->data, hdrlen)) 2489 return -EINVAL; 2490 2491 data += hdrlen; 2492 to_write -= hdrlen; 2493 } 2494 2495 offset = offset_in_page(data); 2496 len_max = PAGE_SIZE - offset; 2497 len = ((to_write > len_max) ? len_max : to_write); 2498 2499 skb->data_len = to_write; 2500 skb->len += to_write; 2501 skb->truesize += to_write; 2502 refcount_add(to_write, &po->sk.sk_wmem_alloc); 2503 2504 while (likely(to_write)) { 2505 nr_frags = skb_shinfo(skb)->nr_frags; 2506 2507 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2508 pr_err("Packet exceed the number of skb frags(%lu)\n", 2509 MAX_SKB_FRAGS); 2510 return -EFAULT; 2511 } 2512 2513 page = pgv_to_page(data); 2514 data += len; 2515 flush_dcache_page(page); 2516 get_page(page); 2517 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2518 to_write -= len; 2519 offset = 0; 2520 len_max = PAGE_SIZE; 2521 len = ((to_write > len_max) ? len_max : to_write); 2522 } 2523 2524 packet_parse_headers(skb, sock); 2525 2526 return tp_len; 2527 } 2528 2529 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2530 int size_max, void **data) 2531 { 2532 union tpacket_uhdr ph; 2533 int tp_len, off; 2534 2535 ph.raw = frame; 2536 2537 switch (po->tp_version) { 2538 case TPACKET_V3: 2539 if (ph.h3->tp_next_offset != 0) { 2540 pr_warn_once("variable sized slot not supported"); 2541 return -EINVAL; 2542 } 2543 tp_len = ph.h3->tp_len; 2544 break; 2545 case TPACKET_V2: 2546 tp_len = ph.h2->tp_len; 2547 break; 2548 default: 2549 tp_len = ph.h1->tp_len; 2550 break; 2551 } 2552 if (unlikely(tp_len > size_max)) { 2553 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2554 return -EMSGSIZE; 2555 } 2556 2557 if (unlikely(po->tp_tx_has_off)) { 2558 int off_min, off_max; 2559 2560 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2561 off_max = po->tx_ring.frame_size - tp_len; 2562 if (po->sk.sk_type == SOCK_DGRAM) { 2563 switch (po->tp_version) { 2564 case TPACKET_V3: 2565 off = ph.h3->tp_net; 2566 break; 2567 case TPACKET_V2: 2568 off = ph.h2->tp_net; 2569 break; 2570 default: 2571 off = ph.h1->tp_net; 2572 break; 2573 } 2574 } else { 2575 switch (po->tp_version) { 2576 case TPACKET_V3: 2577 off = ph.h3->tp_mac; 2578 break; 2579 case TPACKET_V2: 2580 off = ph.h2->tp_mac; 2581 break; 2582 default: 2583 off = ph.h1->tp_mac; 2584 break; 2585 } 2586 } 2587 if (unlikely((off < off_min) || (off_max < off))) 2588 return -EINVAL; 2589 } else { 2590 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2591 } 2592 2593 *data = frame + off; 2594 return tp_len; 2595 } 2596 2597 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2598 { 2599 struct sk_buff *skb; 2600 struct net_device *dev; 2601 struct virtio_net_hdr *vnet_hdr = NULL; 2602 struct sockcm_cookie sockc; 2603 __be16 proto; 2604 int err, reserve = 0; 2605 void *ph; 2606 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2607 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2608 unsigned char *addr = NULL; 2609 int tp_len, size_max; 2610 void *data; 2611 int len_sum = 0; 2612 int status = TP_STATUS_AVAILABLE; 2613 int hlen, tlen, copylen = 0; 2614 2615 mutex_lock(&po->pg_vec_lock); 2616 2617 if (likely(saddr == NULL)) { 2618 dev = packet_cached_dev_get(po); 2619 proto = po->num; 2620 } else { 2621 err = -EINVAL; 2622 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2623 goto out; 2624 if (msg->msg_namelen < (saddr->sll_halen 2625 + offsetof(struct sockaddr_ll, 2626 sll_addr))) 2627 goto out; 2628 proto = saddr->sll_protocol; 2629 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2630 if (po->sk.sk_socket->type == SOCK_DGRAM) { 2631 if (dev && msg->msg_namelen < dev->addr_len + 2632 offsetof(struct sockaddr_ll, sll_addr)) 2633 goto out_put; 2634 addr = saddr->sll_addr; 2635 } 2636 } 2637 2638 err = -ENXIO; 2639 if (unlikely(dev == NULL)) 2640 goto out; 2641 err = -ENETDOWN; 2642 if (unlikely(!(dev->flags & IFF_UP))) 2643 goto out_put; 2644 2645 sockcm_init(&sockc, &po->sk); 2646 if (msg->msg_controllen) { 2647 err = sock_cmsg_send(&po->sk, msg, &sockc); 2648 if (unlikely(err)) 2649 goto out_put; 2650 } 2651 2652 if (po->sk.sk_socket->type == SOCK_RAW) 2653 reserve = dev->hard_header_len; 2654 size_max = po->tx_ring.frame_size 2655 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2656 2657 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) 2658 size_max = dev->mtu + reserve + VLAN_HLEN; 2659 2660 do { 2661 ph = packet_current_frame(po, &po->tx_ring, 2662 TP_STATUS_SEND_REQUEST); 2663 if (unlikely(ph == NULL)) { 2664 if (need_wait && need_resched()) 2665 schedule(); 2666 continue; 2667 } 2668 2669 skb = NULL; 2670 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2671 if (tp_len < 0) 2672 goto tpacket_error; 2673 2674 status = TP_STATUS_SEND_REQUEST; 2675 hlen = LL_RESERVED_SPACE(dev); 2676 tlen = dev->needed_tailroom; 2677 if (po->has_vnet_hdr) { 2678 vnet_hdr = data; 2679 data += sizeof(*vnet_hdr); 2680 tp_len -= sizeof(*vnet_hdr); 2681 if (tp_len < 0 || 2682 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2683 tp_len = -EINVAL; 2684 goto tpacket_error; 2685 } 2686 copylen = __virtio16_to_cpu(vio_le(), 2687 vnet_hdr->hdr_len); 2688 } 2689 copylen = max_t(int, copylen, dev->hard_header_len); 2690 skb = sock_alloc_send_skb(&po->sk, 2691 hlen + tlen + sizeof(struct sockaddr_ll) + 2692 (copylen - dev->hard_header_len), 2693 !need_wait, &err); 2694 2695 if (unlikely(skb == NULL)) { 2696 /* we assume the socket was initially writeable ... */ 2697 if (likely(len_sum > 0)) 2698 err = len_sum; 2699 goto out_status; 2700 } 2701 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2702 addr, hlen, copylen, &sockc); 2703 if (likely(tp_len >= 0) && 2704 tp_len > dev->mtu + reserve && 2705 !po->has_vnet_hdr && 2706 !packet_extra_vlan_len_allowed(dev, skb)) 2707 tp_len = -EMSGSIZE; 2708 2709 if (unlikely(tp_len < 0)) { 2710 tpacket_error: 2711 if (po->tp_loss) { 2712 __packet_set_status(po, ph, 2713 TP_STATUS_AVAILABLE); 2714 packet_increment_head(&po->tx_ring); 2715 kfree_skb(skb); 2716 continue; 2717 } else { 2718 status = TP_STATUS_WRONG_FORMAT; 2719 err = tp_len; 2720 goto out_status; 2721 } 2722 } 2723 2724 if (po->has_vnet_hdr) { 2725 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) { 2726 tp_len = -EINVAL; 2727 goto tpacket_error; 2728 } 2729 virtio_net_hdr_set_proto(skb, vnet_hdr); 2730 } 2731 2732 skb->destructor = tpacket_destruct_skb; 2733 __packet_set_status(po, ph, TP_STATUS_SENDING); 2734 packet_inc_pending(&po->tx_ring); 2735 2736 status = TP_STATUS_SEND_REQUEST; 2737 err = po->xmit(skb); 2738 if (unlikely(err > 0)) { 2739 err = net_xmit_errno(err); 2740 if (err && __packet_get_status(po, ph) == 2741 TP_STATUS_AVAILABLE) { 2742 /* skb was destructed already */ 2743 skb = NULL; 2744 goto out_status; 2745 } 2746 /* 2747 * skb was dropped but not destructed yet; 2748 * let's treat it like congestion or err < 0 2749 */ 2750 err = 0; 2751 } 2752 packet_increment_head(&po->tx_ring); 2753 len_sum += tp_len; 2754 } while (likely((ph != NULL) || 2755 /* Note: packet_read_pending() might be slow if we have 2756 * to call it as it's per_cpu variable, but in fast-path 2757 * we already short-circuit the loop with the first 2758 * condition, and luckily don't have to go that path 2759 * anyway. 2760 */ 2761 (need_wait && packet_read_pending(&po->tx_ring)))); 2762 2763 err = len_sum; 2764 goto out_put; 2765 2766 out_status: 2767 __packet_set_status(po, ph, status); 2768 kfree_skb(skb); 2769 out_put: 2770 dev_put(dev); 2771 out: 2772 mutex_unlock(&po->pg_vec_lock); 2773 return err; 2774 } 2775 2776 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2777 size_t reserve, size_t len, 2778 size_t linear, int noblock, 2779 int *err) 2780 { 2781 struct sk_buff *skb; 2782 2783 /* Under a page? Don't bother with paged skb. */ 2784 if (prepad + len < PAGE_SIZE || !linear) 2785 linear = len; 2786 2787 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2788 err, 0); 2789 if (!skb) 2790 return NULL; 2791 2792 skb_reserve(skb, reserve); 2793 skb_put(skb, linear); 2794 skb->data_len = len - linear; 2795 skb->len += len - linear; 2796 2797 return skb; 2798 } 2799 2800 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2801 { 2802 struct sock *sk = sock->sk; 2803 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2804 struct sk_buff *skb; 2805 struct net_device *dev; 2806 __be16 proto; 2807 unsigned char *addr = NULL; 2808 int err, reserve = 0; 2809 struct sockcm_cookie sockc; 2810 struct virtio_net_hdr vnet_hdr = { 0 }; 2811 int offset = 0; 2812 struct packet_sock *po = pkt_sk(sk); 2813 bool has_vnet_hdr = false; 2814 int hlen, tlen, linear; 2815 int extra_len = 0; 2816 2817 /* 2818 * Get and verify the address. 2819 */ 2820 2821 if (likely(saddr == NULL)) { 2822 dev = packet_cached_dev_get(po); 2823 proto = po->num; 2824 } else { 2825 err = -EINVAL; 2826 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2827 goto out; 2828 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2829 goto out; 2830 proto = saddr->sll_protocol; 2831 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2832 if (sock->type == SOCK_DGRAM) { 2833 if (dev && msg->msg_namelen < dev->addr_len + 2834 offsetof(struct sockaddr_ll, sll_addr)) 2835 goto out_unlock; 2836 addr = saddr->sll_addr; 2837 } 2838 } 2839 2840 err = -ENXIO; 2841 if (unlikely(dev == NULL)) 2842 goto out_unlock; 2843 err = -ENETDOWN; 2844 if (unlikely(!(dev->flags & IFF_UP))) 2845 goto out_unlock; 2846 2847 sockcm_init(&sockc, sk); 2848 sockc.mark = sk->sk_mark; 2849 if (msg->msg_controllen) { 2850 err = sock_cmsg_send(sk, msg, &sockc); 2851 if (unlikely(err)) 2852 goto out_unlock; 2853 } 2854 2855 if (sock->type == SOCK_RAW) 2856 reserve = dev->hard_header_len; 2857 if (po->has_vnet_hdr) { 2858 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2859 if (err) 2860 goto out_unlock; 2861 has_vnet_hdr = true; 2862 } 2863 2864 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2865 if (!netif_supports_nofcs(dev)) { 2866 err = -EPROTONOSUPPORT; 2867 goto out_unlock; 2868 } 2869 extra_len = 4; /* We're doing our own CRC */ 2870 } 2871 2872 err = -EMSGSIZE; 2873 if (!vnet_hdr.gso_type && 2874 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2875 goto out_unlock; 2876 2877 err = -ENOBUFS; 2878 hlen = LL_RESERVED_SPACE(dev); 2879 tlen = dev->needed_tailroom; 2880 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); 2881 linear = max(linear, min_t(int, len, dev->hard_header_len)); 2882 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, 2883 msg->msg_flags & MSG_DONTWAIT, &err); 2884 if (skb == NULL) 2885 goto out_unlock; 2886 2887 skb_reset_network_header(skb); 2888 2889 err = -EINVAL; 2890 if (sock->type == SOCK_DGRAM) { 2891 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 2892 if (unlikely(offset < 0)) 2893 goto out_free; 2894 } else if (reserve) { 2895 skb_reserve(skb, -reserve); 2896 if (len < reserve + sizeof(struct ipv6hdr) && 2897 dev->min_header_len != dev->hard_header_len) 2898 skb_reset_network_header(skb); 2899 } 2900 2901 /* Returns -EFAULT on error */ 2902 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 2903 if (err) 2904 goto out_free; 2905 2906 if (sock->type == SOCK_RAW && 2907 !dev_validate_header(dev, skb->data, len)) { 2908 err = -EINVAL; 2909 goto out_free; 2910 } 2911 2912 skb_setup_tx_timestamp(skb, sockc.tsflags); 2913 2914 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 2915 !packet_extra_vlan_len_allowed(dev, skb)) { 2916 err = -EMSGSIZE; 2917 goto out_free; 2918 } 2919 2920 skb->protocol = proto; 2921 skb->dev = dev; 2922 skb->priority = sk->sk_priority; 2923 skb->mark = sockc.mark; 2924 skb->tstamp = sockc.transmit_time; 2925 2926 if (has_vnet_hdr) { 2927 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 2928 if (err) 2929 goto out_free; 2930 len += sizeof(vnet_hdr); 2931 virtio_net_hdr_set_proto(skb, &vnet_hdr); 2932 } 2933 2934 packet_parse_headers(skb, sock); 2935 2936 if (unlikely(extra_len == 4)) 2937 skb->no_fcs = 1; 2938 2939 err = po->xmit(skb); 2940 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2941 goto out_unlock; 2942 2943 dev_put(dev); 2944 2945 return len; 2946 2947 out_free: 2948 kfree_skb(skb); 2949 out_unlock: 2950 if (dev) 2951 dev_put(dev); 2952 out: 2953 return err; 2954 } 2955 2956 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2957 { 2958 struct sock *sk = sock->sk; 2959 struct packet_sock *po = pkt_sk(sk); 2960 2961 if (po->tx_ring.pg_vec) 2962 return tpacket_snd(po, msg); 2963 else 2964 return packet_snd(sock, msg, len); 2965 } 2966 2967 /* 2968 * Close a PACKET socket. This is fairly simple. We immediately go 2969 * to 'closed' state and remove our protocol entry in the device list. 2970 */ 2971 2972 static int packet_release(struct socket *sock) 2973 { 2974 struct sock *sk = sock->sk; 2975 struct packet_sock *po; 2976 struct packet_fanout *f; 2977 struct net *net; 2978 union tpacket_req_u req_u; 2979 2980 if (!sk) 2981 return 0; 2982 2983 net = sock_net(sk); 2984 po = pkt_sk(sk); 2985 2986 mutex_lock(&net->packet.sklist_lock); 2987 sk_del_node_init_rcu(sk); 2988 mutex_unlock(&net->packet.sklist_lock); 2989 2990 preempt_disable(); 2991 sock_prot_inuse_add(net, sk->sk_prot, -1); 2992 preempt_enable(); 2993 2994 spin_lock(&po->bind_lock); 2995 unregister_prot_hook(sk, false); 2996 packet_cached_dev_reset(po); 2997 2998 if (po->prot_hook.dev) { 2999 dev_put(po->prot_hook.dev); 3000 po->prot_hook.dev = NULL; 3001 } 3002 spin_unlock(&po->bind_lock); 3003 3004 packet_flush_mclist(sk); 3005 3006 lock_sock(sk); 3007 if (po->rx_ring.pg_vec) { 3008 memset(&req_u, 0, sizeof(req_u)); 3009 packet_set_ring(sk, &req_u, 1, 0); 3010 } 3011 3012 if (po->tx_ring.pg_vec) { 3013 memset(&req_u, 0, sizeof(req_u)); 3014 packet_set_ring(sk, &req_u, 1, 1); 3015 } 3016 release_sock(sk); 3017 3018 f = fanout_release(sk); 3019 3020 synchronize_net(); 3021 3022 kfree(po->rollover); 3023 if (f) { 3024 fanout_release_data(f); 3025 kfree(f); 3026 } 3027 /* 3028 * Now the socket is dead. No more input will appear. 3029 */ 3030 sock_orphan(sk); 3031 sock->sk = NULL; 3032 3033 /* Purge queues */ 3034 3035 skb_queue_purge(&sk->sk_receive_queue); 3036 packet_free_pending(po); 3037 sk_refcnt_debug_release(sk); 3038 3039 sock_put(sk); 3040 return 0; 3041 } 3042 3043 /* 3044 * Attach a packet hook. 3045 */ 3046 3047 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3048 __be16 proto) 3049 { 3050 struct packet_sock *po = pkt_sk(sk); 3051 struct net_device *dev_curr; 3052 __be16 proto_curr; 3053 bool need_rehook; 3054 struct net_device *dev = NULL; 3055 int ret = 0; 3056 bool unlisted = false; 3057 3058 lock_sock(sk); 3059 spin_lock(&po->bind_lock); 3060 rcu_read_lock(); 3061 3062 if (po->fanout) { 3063 ret = -EINVAL; 3064 goto out_unlock; 3065 } 3066 3067 if (name) { 3068 dev = dev_get_by_name_rcu(sock_net(sk), name); 3069 if (!dev) { 3070 ret = -ENODEV; 3071 goto out_unlock; 3072 } 3073 } else if (ifindex) { 3074 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3075 if (!dev) { 3076 ret = -ENODEV; 3077 goto out_unlock; 3078 } 3079 } 3080 3081 if (dev) 3082 dev_hold(dev); 3083 3084 proto_curr = po->prot_hook.type; 3085 dev_curr = po->prot_hook.dev; 3086 3087 need_rehook = proto_curr != proto || dev_curr != dev; 3088 3089 if (need_rehook) { 3090 if (po->running) { 3091 rcu_read_unlock(); 3092 /* prevents packet_notifier() from calling 3093 * register_prot_hook() 3094 */ 3095 po->num = 0; 3096 __unregister_prot_hook(sk, true); 3097 rcu_read_lock(); 3098 dev_curr = po->prot_hook.dev; 3099 if (dev) 3100 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3101 dev->ifindex); 3102 } 3103 3104 BUG_ON(po->running); 3105 po->num = proto; 3106 po->prot_hook.type = proto; 3107 3108 if (unlikely(unlisted)) { 3109 dev_put(dev); 3110 po->prot_hook.dev = NULL; 3111 po->ifindex = -1; 3112 packet_cached_dev_reset(po); 3113 } else { 3114 po->prot_hook.dev = dev; 3115 po->ifindex = dev ? dev->ifindex : 0; 3116 packet_cached_dev_assign(po, dev); 3117 } 3118 } 3119 if (dev_curr) 3120 dev_put(dev_curr); 3121 3122 if (proto == 0 || !need_rehook) 3123 goto out_unlock; 3124 3125 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3126 register_prot_hook(sk); 3127 } else { 3128 sk->sk_err = ENETDOWN; 3129 if (!sock_flag(sk, SOCK_DEAD)) 3130 sk->sk_error_report(sk); 3131 } 3132 3133 out_unlock: 3134 rcu_read_unlock(); 3135 spin_unlock(&po->bind_lock); 3136 release_sock(sk); 3137 return ret; 3138 } 3139 3140 /* 3141 * Bind a packet socket to a device 3142 */ 3143 3144 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3145 int addr_len) 3146 { 3147 struct sock *sk = sock->sk; 3148 char name[sizeof(uaddr->sa_data) + 1]; 3149 3150 /* 3151 * Check legality 3152 */ 3153 3154 if (addr_len != sizeof(struct sockaddr)) 3155 return -EINVAL; 3156 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be 3157 * zero-terminated. 3158 */ 3159 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); 3160 name[sizeof(uaddr->sa_data)] = 0; 3161 3162 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3163 } 3164 3165 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3166 { 3167 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3168 struct sock *sk = sock->sk; 3169 3170 /* 3171 * Check legality 3172 */ 3173 3174 if (addr_len < sizeof(struct sockaddr_ll)) 3175 return -EINVAL; 3176 if (sll->sll_family != AF_PACKET) 3177 return -EINVAL; 3178 3179 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3180 sll->sll_protocol ? : pkt_sk(sk)->num); 3181 } 3182 3183 static struct proto packet_proto = { 3184 .name = "PACKET", 3185 .owner = THIS_MODULE, 3186 .obj_size = sizeof(struct packet_sock), 3187 }; 3188 3189 /* 3190 * Create a packet of type SOCK_PACKET. 3191 */ 3192 3193 static int packet_create(struct net *net, struct socket *sock, int protocol, 3194 int kern) 3195 { 3196 struct sock *sk; 3197 struct packet_sock *po; 3198 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3199 int err; 3200 3201 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3202 return -EPERM; 3203 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3204 sock->type != SOCK_PACKET) 3205 return -ESOCKTNOSUPPORT; 3206 3207 sock->state = SS_UNCONNECTED; 3208 3209 err = -ENOBUFS; 3210 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3211 if (sk == NULL) 3212 goto out; 3213 3214 sock->ops = &packet_ops; 3215 if (sock->type == SOCK_PACKET) 3216 sock->ops = &packet_ops_spkt; 3217 3218 sock_init_data(sock, sk); 3219 3220 po = pkt_sk(sk); 3221 sk->sk_family = PF_PACKET; 3222 po->num = proto; 3223 po->xmit = dev_queue_xmit; 3224 3225 err = packet_alloc_pending(po); 3226 if (err) 3227 goto out2; 3228 3229 packet_cached_dev_reset(po); 3230 3231 sk->sk_destruct = packet_sock_destruct; 3232 sk_refcnt_debug_inc(sk); 3233 3234 /* 3235 * Attach a protocol block 3236 */ 3237 3238 spin_lock_init(&po->bind_lock); 3239 mutex_init(&po->pg_vec_lock); 3240 po->rollover = NULL; 3241 po->prot_hook.func = packet_rcv; 3242 3243 if (sock->type == SOCK_PACKET) 3244 po->prot_hook.func = packet_rcv_spkt; 3245 3246 po->prot_hook.af_packet_priv = sk; 3247 3248 if (proto) { 3249 po->prot_hook.type = proto; 3250 __register_prot_hook(sk); 3251 } 3252 3253 mutex_lock(&net->packet.sklist_lock); 3254 sk_add_node_tail_rcu(sk, &net->packet.sklist); 3255 mutex_unlock(&net->packet.sklist_lock); 3256 3257 preempt_disable(); 3258 sock_prot_inuse_add(net, &packet_proto, 1); 3259 preempt_enable(); 3260 3261 return 0; 3262 out2: 3263 sk_free(sk); 3264 out: 3265 return err; 3266 } 3267 3268 /* 3269 * Pull a packet from our receive queue and hand it to the user. 3270 * If necessary we block. 3271 */ 3272 3273 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3274 int flags) 3275 { 3276 struct sock *sk = sock->sk; 3277 struct sk_buff *skb; 3278 int copied, err; 3279 int vnet_hdr_len = 0; 3280 unsigned int origlen = 0; 3281 3282 err = -EINVAL; 3283 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3284 goto out; 3285 3286 #if 0 3287 /* What error should we return now? EUNATTACH? */ 3288 if (pkt_sk(sk)->ifindex < 0) 3289 return -ENODEV; 3290 #endif 3291 3292 if (flags & MSG_ERRQUEUE) { 3293 err = sock_recv_errqueue(sk, msg, len, 3294 SOL_PACKET, PACKET_TX_TIMESTAMP); 3295 goto out; 3296 } 3297 3298 /* 3299 * Call the generic datagram receiver. This handles all sorts 3300 * of horrible races and re-entrancy so we can forget about it 3301 * in the protocol layers. 3302 * 3303 * Now it will return ENETDOWN, if device have just gone down, 3304 * but then it will block. 3305 */ 3306 3307 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 3308 3309 /* 3310 * An error occurred so return it. Because skb_recv_datagram() 3311 * handles the blocking we don't see and worry about blocking 3312 * retries. 3313 */ 3314 3315 if (skb == NULL) 3316 goto out; 3317 3318 packet_rcv_try_clear_pressure(pkt_sk(sk)); 3319 3320 if (pkt_sk(sk)->has_vnet_hdr) { 3321 err = packet_rcv_vnet(msg, skb, &len); 3322 if (err) 3323 goto out_free; 3324 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3325 } 3326 3327 /* You lose any data beyond the buffer you gave. If it worries 3328 * a user program they can ask the device for its MTU 3329 * anyway. 3330 */ 3331 copied = skb->len; 3332 if (copied > len) { 3333 copied = len; 3334 msg->msg_flags |= MSG_TRUNC; 3335 } 3336 3337 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3338 if (err) 3339 goto out_free; 3340 3341 if (sock->type != SOCK_PACKET) { 3342 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3343 3344 /* Original length was stored in sockaddr_ll fields */ 3345 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3346 sll->sll_family = AF_PACKET; 3347 sll->sll_protocol = skb->protocol; 3348 } 3349 3350 sock_recv_ts_and_drops(msg, sk, skb); 3351 3352 if (msg->msg_name) { 3353 int copy_len; 3354 3355 /* If the address length field is there to be filled 3356 * in, we fill it in now. 3357 */ 3358 if (sock->type == SOCK_PACKET) { 3359 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3360 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3361 copy_len = msg->msg_namelen; 3362 } else { 3363 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3364 3365 msg->msg_namelen = sll->sll_halen + 3366 offsetof(struct sockaddr_ll, sll_addr); 3367 copy_len = msg->msg_namelen; 3368 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { 3369 memset(msg->msg_name + 3370 offsetof(struct sockaddr_ll, sll_addr), 3371 0, sizeof(sll->sll_addr)); 3372 msg->msg_namelen = sizeof(struct sockaddr_ll); 3373 } 3374 } 3375 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); 3376 } 3377 3378 if (pkt_sk(sk)->auxdata) { 3379 struct tpacket_auxdata aux; 3380 3381 aux.tp_status = TP_STATUS_USER; 3382 if (skb->ip_summed == CHECKSUM_PARTIAL) 3383 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3384 else if (skb->pkt_type != PACKET_OUTGOING && 3385 (skb->ip_summed == CHECKSUM_COMPLETE || 3386 skb_csum_unnecessary(skb))) 3387 aux.tp_status |= TP_STATUS_CSUM_VALID; 3388 3389 aux.tp_len = origlen; 3390 aux.tp_snaplen = skb->len; 3391 aux.tp_mac = 0; 3392 aux.tp_net = skb_network_offset(skb); 3393 if (skb_vlan_tag_present(skb)) { 3394 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3395 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3396 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3397 } else { 3398 aux.tp_vlan_tci = 0; 3399 aux.tp_vlan_tpid = 0; 3400 } 3401 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3402 } 3403 3404 /* 3405 * Free or return the buffer as appropriate. Again this 3406 * hides all the races and re-entrancy issues from us. 3407 */ 3408 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3409 3410 out_free: 3411 skb_free_datagram(sk, skb); 3412 out: 3413 return err; 3414 } 3415 3416 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3417 int peer) 3418 { 3419 struct net_device *dev; 3420 struct sock *sk = sock->sk; 3421 3422 if (peer) 3423 return -EOPNOTSUPP; 3424 3425 uaddr->sa_family = AF_PACKET; 3426 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3427 rcu_read_lock(); 3428 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3429 if (dev) 3430 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3431 rcu_read_unlock(); 3432 3433 return sizeof(*uaddr); 3434 } 3435 3436 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3437 int peer) 3438 { 3439 struct net_device *dev; 3440 struct sock *sk = sock->sk; 3441 struct packet_sock *po = pkt_sk(sk); 3442 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3443 3444 if (peer) 3445 return -EOPNOTSUPP; 3446 3447 sll->sll_family = AF_PACKET; 3448 sll->sll_ifindex = po->ifindex; 3449 sll->sll_protocol = po->num; 3450 sll->sll_pkttype = 0; 3451 rcu_read_lock(); 3452 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 3453 if (dev) { 3454 sll->sll_hatype = dev->type; 3455 sll->sll_halen = dev->addr_len; 3456 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3457 } else { 3458 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3459 sll->sll_halen = 0; 3460 } 3461 rcu_read_unlock(); 3462 3463 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3464 } 3465 3466 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3467 int what) 3468 { 3469 switch (i->type) { 3470 case PACKET_MR_MULTICAST: 3471 if (i->alen != dev->addr_len) 3472 return -EINVAL; 3473 if (what > 0) 3474 return dev_mc_add(dev, i->addr); 3475 else 3476 return dev_mc_del(dev, i->addr); 3477 break; 3478 case PACKET_MR_PROMISC: 3479 return dev_set_promiscuity(dev, what); 3480 case PACKET_MR_ALLMULTI: 3481 return dev_set_allmulti(dev, what); 3482 case PACKET_MR_UNICAST: 3483 if (i->alen != dev->addr_len) 3484 return -EINVAL; 3485 if (what > 0) 3486 return dev_uc_add(dev, i->addr); 3487 else 3488 return dev_uc_del(dev, i->addr); 3489 break; 3490 default: 3491 break; 3492 } 3493 return 0; 3494 } 3495 3496 static void packet_dev_mclist_delete(struct net_device *dev, 3497 struct packet_mclist **mlp) 3498 { 3499 struct packet_mclist *ml; 3500 3501 while ((ml = *mlp) != NULL) { 3502 if (ml->ifindex == dev->ifindex) { 3503 packet_dev_mc(dev, ml, -1); 3504 *mlp = ml->next; 3505 kfree(ml); 3506 } else 3507 mlp = &ml->next; 3508 } 3509 } 3510 3511 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3512 { 3513 struct packet_sock *po = pkt_sk(sk); 3514 struct packet_mclist *ml, *i; 3515 struct net_device *dev; 3516 int err; 3517 3518 rtnl_lock(); 3519 3520 err = -ENODEV; 3521 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3522 if (!dev) 3523 goto done; 3524 3525 err = -EINVAL; 3526 if (mreq->mr_alen > dev->addr_len) 3527 goto done; 3528 3529 err = -ENOBUFS; 3530 i = kmalloc(sizeof(*i), GFP_KERNEL); 3531 if (i == NULL) 3532 goto done; 3533 3534 err = 0; 3535 for (ml = po->mclist; ml; ml = ml->next) { 3536 if (ml->ifindex == mreq->mr_ifindex && 3537 ml->type == mreq->mr_type && 3538 ml->alen == mreq->mr_alen && 3539 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3540 ml->count++; 3541 /* Free the new element ... */ 3542 kfree(i); 3543 goto done; 3544 } 3545 } 3546 3547 i->type = mreq->mr_type; 3548 i->ifindex = mreq->mr_ifindex; 3549 i->alen = mreq->mr_alen; 3550 memcpy(i->addr, mreq->mr_address, i->alen); 3551 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3552 i->count = 1; 3553 i->next = po->mclist; 3554 po->mclist = i; 3555 err = packet_dev_mc(dev, i, 1); 3556 if (err) { 3557 po->mclist = i->next; 3558 kfree(i); 3559 } 3560 3561 done: 3562 rtnl_unlock(); 3563 return err; 3564 } 3565 3566 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3567 { 3568 struct packet_mclist *ml, **mlp; 3569 3570 rtnl_lock(); 3571 3572 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3573 if (ml->ifindex == mreq->mr_ifindex && 3574 ml->type == mreq->mr_type && 3575 ml->alen == mreq->mr_alen && 3576 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3577 if (--ml->count == 0) { 3578 struct net_device *dev; 3579 *mlp = ml->next; 3580 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3581 if (dev) 3582 packet_dev_mc(dev, ml, -1); 3583 kfree(ml); 3584 } 3585 break; 3586 } 3587 } 3588 rtnl_unlock(); 3589 return 0; 3590 } 3591 3592 static void packet_flush_mclist(struct sock *sk) 3593 { 3594 struct packet_sock *po = pkt_sk(sk); 3595 struct packet_mclist *ml; 3596 3597 if (!po->mclist) 3598 return; 3599 3600 rtnl_lock(); 3601 while ((ml = po->mclist) != NULL) { 3602 struct net_device *dev; 3603 3604 po->mclist = ml->next; 3605 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3606 if (dev != NULL) 3607 packet_dev_mc(dev, ml, -1); 3608 kfree(ml); 3609 } 3610 rtnl_unlock(); 3611 } 3612 3613 static int 3614 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3615 { 3616 struct sock *sk = sock->sk; 3617 struct packet_sock *po = pkt_sk(sk); 3618 int ret; 3619 3620 if (level != SOL_PACKET) 3621 return -ENOPROTOOPT; 3622 3623 switch (optname) { 3624 case PACKET_ADD_MEMBERSHIP: 3625 case PACKET_DROP_MEMBERSHIP: 3626 { 3627 struct packet_mreq_max mreq; 3628 int len = optlen; 3629 memset(&mreq, 0, sizeof(mreq)); 3630 if (len < sizeof(struct packet_mreq)) 3631 return -EINVAL; 3632 if (len > sizeof(mreq)) 3633 len = sizeof(mreq); 3634 if (copy_from_user(&mreq, optval, len)) 3635 return -EFAULT; 3636 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3637 return -EINVAL; 3638 if (optname == PACKET_ADD_MEMBERSHIP) 3639 ret = packet_mc_add(sk, &mreq); 3640 else 3641 ret = packet_mc_drop(sk, &mreq); 3642 return ret; 3643 } 3644 3645 case PACKET_RX_RING: 3646 case PACKET_TX_RING: 3647 { 3648 union tpacket_req_u req_u; 3649 int len; 3650 3651 lock_sock(sk); 3652 switch (po->tp_version) { 3653 case TPACKET_V1: 3654 case TPACKET_V2: 3655 len = sizeof(req_u.req); 3656 break; 3657 case TPACKET_V3: 3658 default: 3659 len = sizeof(req_u.req3); 3660 break; 3661 } 3662 if (optlen < len) { 3663 ret = -EINVAL; 3664 } else { 3665 if (copy_from_user(&req_u.req, optval, len)) 3666 ret = -EFAULT; 3667 else 3668 ret = packet_set_ring(sk, &req_u, 0, 3669 optname == PACKET_TX_RING); 3670 } 3671 release_sock(sk); 3672 return ret; 3673 } 3674 case PACKET_COPY_THRESH: 3675 { 3676 int val; 3677 3678 if (optlen != sizeof(val)) 3679 return -EINVAL; 3680 if (copy_from_user(&val, optval, sizeof(val))) 3681 return -EFAULT; 3682 3683 pkt_sk(sk)->copy_thresh = val; 3684 return 0; 3685 } 3686 case PACKET_VERSION: 3687 { 3688 int val; 3689 3690 if (optlen != sizeof(val)) 3691 return -EINVAL; 3692 if (copy_from_user(&val, optval, sizeof(val))) 3693 return -EFAULT; 3694 switch (val) { 3695 case TPACKET_V1: 3696 case TPACKET_V2: 3697 case TPACKET_V3: 3698 break; 3699 default: 3700 return -EINVAL; 3701 } 3702 lock_sock(sk); 3703 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3704 ret = -EBUSY; 3705 } else { 3706 po->tp_version = val; 3707 ret = 0; 3708 } 3709 release_sock(sk); 3710 return ret; 3711 } 3712 case PACKET_RESERVE: 3713 { 3714 unsigned int val; 3715 3716 if (optlen != sizeof(val)) 3717 return -EINVAL; 3718 if (copy_from_user(&val, optval, sizeof(val))) 3719 return -EFAULT; 3720 if (val > INT_MAX) 3721 return -EINVAL; 3722 lock_sock(sk); 3723 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3724 ret = -EBUSY; 3725 } else { 3726 po->tp_reserve = val; 3727 ret = 0; 3728 } 3729 release_sock(sk); 3730 return ret; 3731 } 3732 case PACKET_LOSS: 3733 { 3734 unsigned int val; 3735 3736 if (optlen != sizeof(val)) 3737 return -EINVAL; 3738 if (copy_from_user(&val, optval, sizeof(val))) 3739 return -EFAULT; 3740 3741 lock_sock(sk); 3742 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3743 ret = -EBUSY; 3744 } else { 3745 po->tp_loss = !!val; 3746 ret = 0; 3747 } 3748 release_sock(sk); 3749 return ret; 3750 } 3751 case PACKET_AUXDATA: 3752 { 3753 int val; 3754 3755 if (optlen < sizeof(val)) 3756 return -EINVAL; 3757 if (copy_from_user(&val, optval, sizeof(val))) 3758 return -EFAULT; 3759 3760 lock_sock(sk); 3761 po->auxdata = !!val; 3762 release_sock(sk); 3763 return 0; 3764 } 3765 case PACKET_ORIGDEV: 3766 { 3767 int val; 3768 3769 if (optlen < sizeof(val)) 3770 return -EINVAL; 3771 if (copy_from_user(&val, optval, sizeof(val))) 3772 return -EFAULT; 3773 3774 lock_sock(sk); 3775 po->origdev = !!val; 3776 release_sock(sk); 3777 return 0; 3778 } 3779 case PACKET_VNET_HDR: 3780 { 3781 int val; 3782 3783 if (sock->type != SOCK_RAW) 3784 return -EINVAL; 3785 if (optlen < sizeof(val)) 3786 return -EINVAL; 3787 if (copy_from_user(&val, optval, sizeof(val))) 3788 return -EFAULT; 3789 3790 lock_sock(sk); 3791 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3792 ret = -EBUSY; 3793 } else { 3794 po->has_vnet_hdr = !!val; 3795 ret = 0; 3796 } 3797 release_sock(sk); 3798 return ret; 3799 } 3800 case PACKET_TIMESTAMP: 3801 { 3802 int val; 3803 3804 if (optlen != sizeof(val)) 3805 return -EINVAL; 3806 if (copy_from_user(&val, optval, sizeof(val))) 3807 return -EFAULT; 3808 3809 po->tp_tstamp = val; 3810 return 0; 3811 } 3812 case PACKET_FANOUT: 3813 { 3814 int val; 3815 3816 if (optlen != sizeof(val)) 3817 return -EINVAL; 3818 if (copy_from_user(&val, optval, sizeof(val))) 3819 return -EFAULT; 3820 3821 return fanout_add(sk, val & 0xffff, val >> 16); 3822 } 3823 case PACKET_FANOUT_DATA: 3824 { 3825 if (!po->fanout) 3826 return -EINVAL; 3827 3828 return fanout_set_data(po, optval, optlen); 3829 } 3830 case PACKET_IGNORE_OUTGOING: 3831 { 3832 int val; 3833 3834 if (optlen != sizeof(val)) 3835 return -EINVAL; 3836 if (copy_from_user(&val, optval, sizeof(val))) 3837 return -EFAULT; 3838 if (val < 0 || val > 1) 3839 return -EINVAL; 3840 3841 po->prot_hook.ignore_outgoing = !!val; 3842 return 0; 3843 } 3844 case PACKET_TX_HAS_OFF: 3845 { 3846 unsigned int val; 3847 3848 if (optlen != sizeof(val)) 3849 return -EINVAL; 3850 if (copy_from_user(&val, optval, sizeof(val))) 3851 return -EFAULT; 3852 3853 lock_sock(sk); 3854 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3855 ret = -EBUSY; 3856 } else { 3857 po->tp_tx_has_off = !!val; 3858 ret = 0; 3859 } 3860 release_sock(sk); 3861 return 0; 3862 } 3863 case PACKET_QDISC_BYPASS: 3864 { 3865 int val; 3866 3867 if (optlen != sizeof(val)) 3868 return -EINVAL; 3869 if (copy_from_user(&val, optval, sizeof(val))) 3870 return -EFAULT; 3871 3872 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; 3873 return 0; 3874 } 3875 default: 3876 return -ENOPROTOOPT; 3877 } 3878 } 3879 3880 static int packet_getsockopt(struct socket *sock, int level, int optname, 3881 char __user *optval, int __user *optlen) 3882 { 3883 int len; 3884 int val, lv = sizeof(val); 3885 struct sock *sk = sock->sk; 3886 struct packet_sock *po = pkt_sk(sk); 3887 void *data = &val; 3888 union tpacket_stats_u st; 3889 struct tpacket_rollover_stats rstats; 3890 int drops; 3891 3892 if (level != SOL_PACKET) 3893 return -ENOPROTOOPT; 3894 3895 if (get_user(len, optlen)) 3896 return -EFAULT; 3897 3898 if (len < 0) 3899 return -EINVAL; 3900 3901 switch (optname) { 3902 case PACKET_STATISTICS: 3903 spin_lock_bh(&sk->sk_receive_queue.lock); 3904 memcpy(&st, &po->stats, sizeof(st)); 3905 memset(&po->stats, 0, sizeof(po->stats)); 3906 spin_unlock_bh(&sk->sk_receive_queue.lock); 3907 drops = atomic_xchg(&po->tp_drops, 0); 3908 3909 if (po->tp_version == TPACKET_V3) { 3910 lv = sizeof(struct tpacket_stats_v3); 3911 st.stats3.tp_drops = drops; 3912 st.stats3.tp_packets += drops; 3913 data = &st.stats3; 3914 } else { 3915 lv = sizeof(struct tpacket_stats); 3916 st.stats1.tp_drops = drops; 3917 st.stats1.tp_packets += 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 packet_rcv_try_clear_pressure(po); 4137 spin_unlock_bh(&sk->sk_receive_queue.lock); 4138 spin_lock_bh(&sk->sk_write_queue.lock); 4139 if (po->tx_ring.pg_vec) { 4140 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4141 mask |= EPOLLOUT | EPOLLWRNORM; 4142 } 4143 spin_unlock_bh(&sk->sk_write_queue.lock); 4144 return mask; 4145 } 4146 4147 4148 /* Dirty? Well, I still did not learn better way to account 4149 * for user mmaps. 4150 */ 4151 4152 static void packet_mm_open(struct vm_area_struct *vma) 4153 { 4154 struct file *file = vma->vm_file; 4155 struct socket *sock = file->private_data; 4156 struct sock *sk = sock->sk; 4157 4158 if (sk) 4159 atomic_inc(&pkt_sk(sk)->mapped); 4160 } 4161 4162 static void packet_mm_close(struct vm_area_struct *vma) 4163 { 4164 struct file *file = vma->vm_file; 4165 struct socket *sock = file->private_data; 4166 struct sock *sk = sock->sk; 4167 4168 if (sk) 4169 atomic_dec(&pkt_sk(sk)->mapped); 4170 } 4171 4172 static const struct vm_operations_struct packet_mmap_ops = { 4173 .open = packet_mm_open, 4174 .close = packet_mm_close, 4175 }; 4176 4177 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4178 unsigned int len) 4179 { 4180 int i; 4181 4182 for (i = 0; i < len; i++) { 4183 if (likely(pg_vec[i].buffer)) { 4184 if (is_vmalloc_addr(pg_vec[i].buffer)) 4185 vfree(pg_vec[i].buffer); 4186 else 4187 free_pages((unsigned long)pg_vec[i].buffer, 4188 order); 4189 pg_vec[i].buffer = NULL; 4190 } 4191 } 4192 kfree(pg_vec); 4193 } 4194 4195 static char *alloc_one_pg_vec_page(unsigned long order) 4196 { 4197 char *buffer; 4198 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4199 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4200 4201 buffer = (char *) __get_free_pages(gfp_flags, order); 4202 if (buffer) 4203 return buffer; 4204 4205 /* __get_free_pages failed, fall back to vmalloc */ 4206 buffer = vzalloc(array_size((1 << order), PAGE_SIZE)); 4207 if (buffer) 4208 return buffer; 4209 4210 /* vmalloc failed, lets dig into swap here */ 4211 gfp_flags &= ~__GFP_NORETRY; 4212 buffer = (char *) __get_free_pages(gfp_flags, order); 4213 if (buffer) 4214 return buffer; 4215 4216 /* complete and utter failure */ 4217 return NULL; 4218 } 4219 4220 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4221 { 4222 unsigned int block_nr = req->tp_block_nr; 4223 struct pgv *pg_vec; 4224 int i; 4225 4226 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN); 4227 if (unlikely(!pg_vec)) 4228 goto out; 4229 4230 for (i = 0; i < block_nr; i++) { 4231 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4232 if (unlikely(!pg_vec[i].buffer)) 4233 goto out_free_pgvec; 4234 } 4235 4236 out: 4237 return pg_vec; 4238 4239 out_free_pgvec: 4240 free_pg_vec(pg_vec, order, block_nr); 4241 pg_vec = NULL; 4242 goto out; 4243 } 4244 4245 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4246 int closing, int tx_ring) 4247 { 4248 struct pgv *pg_vec = NULL; 4249 struct packet_sock *po = pkt_sk(sk); 4250 int was_running, order = 0; 4251 struct packet_ring_buffer *rb; 4252 struct sk_buff_head *rb_queue; 4253 __be16 num; 4254 int err = -EINVAL; 4255 /* Added to avoid minimal code churn */ 4256 struct tpacket_req *req = &req_u->req; 4257 4258 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4259 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4260 4261 err = -EBUSY; 4262 if (!closing) { 4263 if (atomic_read(&po->mapped)) 4264 goto out; 4265 if (packet_read_pending(rb)) 4266 goto out; 4267 } 4268 4269 if (req->tp_block_nr) { 4270 unsigned int min_frame_size; 4271 4272 /* Sanity tests and some calculations */ 4273 err = -EBUSY; 4274 if (unlikely(rb->pg_vec)) 4275 goto out; 4276 4277 switch (po->tp_version) { 4278 case TPACKET_V1: 4279 po->tp_hdrlen = TPACKET_HDRLEN; 4280 break; 4281 case TPACKET_V2: 4282 po->tp_hdrlen = TPACKET2_HDRLEN; 4283 break; 4284 case TPACKET_V3: 4285 po->tp_hdrlen = TPACKET3_HDRLEN; 4286 break; 4287 } 4288 4289 err = -EINVAL; 4290 if (unlikely((int)req->tp_block_size <= 0)) 4291 goto out; 4292 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4293 goto out; 4294 min_frame_size = po->tp_hdrlen + po->tp_reserve; 4295 if (po->tp_version >= TPACKET_V3 && 4296 req->tp_block_size < 4297 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) 4298 goto out; 4299 if (unlikely(req->tp_frame_size < min_frame_size)) 4300 goto out; 4301 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4302 goto out; 4303 4304 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4305 if (unlikely(rb->frames_per_block == 0)) 4306 goto out; 4307 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) 4308 goto out; 4309 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4310 req->tp_frame_nr)) 4311 goto out; 4312 4313 err = -ENOMEM; 4314 order = get_order(req->tp_block_size); 4315 pg_vec = alloc_pg_vec(req, order); 4316 if (unlikely(!pg_vec)) 4317 goto out; 4318 switch (po->tp_version) { 4319 case TPACKET_V3: 4320 /* Block transmit is not supported yet */ 4321 if (!tx_ring) { 4322 init_prb_bdqc(po, rb, pg_vec, req_u); 4323 } else { 4324 struct tpacket_req3 *req3 = &req_u->req3; 4325 4326 if (req3->tp_retire_blk_tov || 4327 req3->tp_sizeof_priv || 4328 req3->tp_feature_req_word) { 4329 err = -EINVAL; 4330 goto out; 4331 } 4332 } 4333 break; 4334 default: 4335 break; 4336 } 4337 } 4338 /* Done */ 4339 else { 4340 err = -EINVAL; 4341 if (unlikely(req->tp_frame_nr)) 4342 goto out; 4343 } 4344 4345 4346 /* Detach socket from network */ 4347 spin_lock(&po->bind_lock); 4348 was_running = po->running; 4349 num = po->num; 4350 if (was_running) { 4351 po->num = 0; 4352 __unregister_prot_hook(sk, false); 4353 } 4354 spin_unlock(&po->bind_lock); 4355 4356 synchronize_net(); 4357 4358 err = -EBUSY; 4359 mutex_lock(&po->pg_vec_lock); 4360 if (closing || atomic_read(&po->mapped) == 0) { 4361 err = 0; 4362 spin_lock_bh(&rb_queue->lock); 4363 swap(rb->pg_vec, pg_vec); 4364 rb->frame_max = (req->tp_frame_nr - 1); 4365 rb->head = 0; 4366 rb->frame_size = req->tp_frame_size; 4367 spin_unlock_bh(&rb_queue->lock); 4368 4369 swap(rb->pg_vec_order, order); 4370 swap(rb->pg_vec_len, req->tp_block_nr); 4371 4372 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4373 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4374 tpacket_rcv : packet_rcv; 4375 skb_queue_purge(rb_queue); 4376 if (atomic_read(&po->mapped)) 4377 pr_err("packet_mmap: vma is busy: %d\n", 4378 atomic_read(&po->mapped)); 4379 } 4380 mutex_unlock(&po->pg_vec_lock); 4381 4382 spin_lock(&po->bind_lock); 4383 if (was_running) { 4384 po->num = num; 4385 register_prot_hook(sk); 4386 } 4387 spin_unlock(&po->bind_lock); 4388 if (pg_vec && (po->tp_version > TPACKET_V2)) { 4389 /* Because we don't support block-based V3 on tx-ring */ 4390 if (!tx_ring) 4391 prb_shutdown_retire_blk_timer(po, rb_queue); 4392 } 4393 4394 if (pg_vec) 4395 free_pg_vec(pg_vec, order, req->tp_block_nr); 4396 out: 4397 return err; 4398 } 4399 4400 static int packet_mmap(struct file *file, struct socket *sock, 4401 struct vm_area_struct *vma) 4402 { 4403 struct sock *sk = sock->sk; 4404 struct packet_sock *po = pkt_sk(sk); 4405 unsigned long size, expected_size; 4406 struct packet_ring_buffer *rb; 4407 unsigned long start; 4408 int err = -EINVAL; 4409 int i; 4410 4411 if (vma->vm_pgoff) 4412 return -EINVAL; 4413 4414 mutex_lock(&po->pg_vec_lock); 4415 4416 expected_size = 0; 4417 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4418 if (rb->pg_vec) { 4419 expected_size += rb->pg_vec_len 4420 * rb->pg_vec_pages 4421 * PAGE_SIZE; 4422 } 4423 } 4424 4425 if (expected_size == 0) 4426 goto out; 4427 4428 size = vma->vm_end - vma->vm_start; 4429 if (size != expected_size) 4430 goto out; 4431 4432 start = vma->vm_start; 4433 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4434 if (rb->pg_vec == NULL) 4435 continue; 4436 4437 for (i = 0; i < rb->pg_vec_len; i++) { 4438 struct page *page; 4439 void *kaddr = rb->pg_vec[i].buffer; 4440 int pg_num; 4441 4442 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4443 page = pgv_to_page(kaddr); 4444 err = vm_insert_page(vma, start, page); 4445 if (unlikely(err)) 4446 goto out; 4447 start += PAGE_SIZE; 4448 kaddr += PAGE_SIZE; 4449 } 4450 } 4451 } 4452 4453 atomic_inc(&po->mapped); 4454 vma->vm_ops = &packet_mmap_ops; 4455 err = 0; 4456 4457 out: 4458 mutex_unlock(&po->pg_vec_lock); 4459 return err; 4460 } 4461 4462 static const struct proto_ops packet_ops_spkt = { 4463 .family = PF_PACKET, 4464 .owner = THIS_MODULE, 4465 .release = packet_release, 4466 .bind = packet_bind_spkt, 4467 .connect = sock_no_connect, 4468 .socketpair = sock_no_socketpair, 4469 .accept = sock_no_accept, 4470 .getname = packet_getname_spkt, 4471 .poll = datagram_poll, 4472 .ioctl = packet_ioctl, 4473 .gettstamp = sock_gettstamp, 4474 .listen = sock_no_listen, 4475 .shutdown = sock_no_shutdown, 4476 .setsockopt = sock_no_setsockopt, 4477 .getsockopt = sock_no_getsockopt, 4478 .sendmsg = packet_sendmsg_spkt, 4479 .recvmsg = packet_recvmsg, 4480 .mmap = sock_no_mmap, 4481 .sendpage = sock_no_sendpage, 4482 }; 4483 4484 static const struct proto_ops packet_ops = { 4485 .family = PF_PACKET, 4486 .owner = THIS_MODULE, 4487 .release = packet_release, 4488 .bind = packet_bind, 4489 .connect = sock_no_connect, 4490 .socketpair = sock_no_socketpair, 4491 .accept = sock_no_accept, 4492 .getname = packet_getname, 4493 .poll = packet_poll, 4494 .ioctl = packet_ioctl, 4495 .gettstamp = sock_gettstamp, 4496 .listen = sock_no_listen, 4497 .shutdown = sock_no_shutdown, 4498 .setsockopt = packet_setsockopt, 4499 .getsockopt = packet_getsockopt, 4500 #ifdef CONFIG_COMPAT 4501 .compat_setsockopt = compat_packet_setsockopt, 4502 #endif 4503 .sendmsg = packet_sendmsg, 4504 .recvmsg = packet_recvmsg, 4505 .mmap = packet_mmap, 4506 .sendpage = sock_no_sendpage, 4507 }; 4508 4509 static const struct net_proto_family packet_family_ops = { 4510 .family = PF_PACKET, 4511 .create = packet_create, 4512 .owner = THIS_MODULE, 4513 }; 4514 4515 static struct notifier_block packet_netdev_notifier = { 4516 .notifier_call = packet_notifier, 4517 }; 4518 4519 #ifdef CONFIG_PROC_FS 4520 4521 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4522 __acquires(RCU) 4523 { 4524 struct net *net = seq_file_net(seq); 4525 4526 rcu_read_lock(); 4527 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4528 } 4529 4530 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4531 { 4532 struct net *net = seq_file_net(seq); 4533 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4534 } 4535 4536 static void packet_seq_stop(struct seq_file *seq, void *v) 4537 __releases(RCU) 4538 { 4539 rcu_read_unlock(); 4540 } 4541 4542 static int packet_seq_show(struct seq_file *seq, void *v) 4543 { 4544 if (v == SEQ_START_TOKEN) 4545 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4546 else { 4547 struct sock *s = sk_entry(v); 4548 const struct packet_sock *po = pkt_sk(s); 4549 4550 seq_printf(seq, 4551 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4552 s, 4553 refcount_read(&s->sk_refcnt), 4554 s->sk_type, 4555 ntohs(po->num), 4556 po->ifindex, 4557 po->running, 4558 atomic_read(&s->sk_rmem_alloc), 4559 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4560 sock_i_ino(s)); 4561 } 4562 4563 return 0; 4564 } 4565 4566 static const struct seq_operations packet_seq_ops = { 4567 .start = packet_seq_start, 4568 .next = packet_seq_next, 4569 .stop = packet_seq_stop, 4570 .show = packet_seq_show, 4571 }; 4572 #endif 4573 4574 static int __net_init packet_net_init(struct net *net) 4575 { 4576 mutex_init(&net->packet.sklist_lock); 4577 INIT_HLIST_HEAD(&net->packet.sklist); 4578 4579 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, 4580 sizeof(struct seq_net_private))) 4581 return -ENOMEM; 4582 4583 return 0; 4584 } 4585 4586 static void __net_exit packet_net_exit(struct net *net) 4587 { 4588 remove_proc_entry("packet", net->proc_net); 4589 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); 4590 } 4591 4592 static struct pernet_operations packet_net_ops = { 4593 .init = packet_net_init, 4594 .exit = packet_net_exit, 4595 }; 4596 4597 4598 static void __exit packet_exit(void) 4599 { 4600 unregister_netdevice_notifier(&packet_netdev_notifier); 4601 unregister_pernet_subsys(&packet_net_ops); 4602 sock_unregister(PF_PACKET); 4603 proto_unregister(&packet_proto); 4604 } 4605 4606 static int __init packet_init(void) 4607 { 4608 int rc; 4609 4610 rc = proto_register(&packet_proto, 0); 4611 if (rc) 4612 goto out; 4613 rc = sock_register(&packet_family_ops); 4614 if (rc) 4615 goto out_proto; 4616 rc = register_pernet_subsys(&packet_net_ops); 4617 if (rc) 4618 goto out_sock; 4619 rc = register_netdevice_notifier(&packet_netdev_notifier); 4620 if (rc) 4621 goto out_pernet; 4622 4623 return 0; 4624 4625 out_pernet: 4626 unregister_pernet_subsys(&packet_net_ops); 4627 out_sock: 4628 sock_unregister(PF_PACKET); 4629 out_proto: 4630 proto_unregister(&packet_proto); 4631 out: 4632 return rc; 4633 } 4634 4635 module_init(packet_init); 4636 module_exit(packet_exit); 4637 MODULE_LICENSE("GPL"); 4638 MODULE_ALIAS_NETPROTO(PF_PACKET); 4639