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 timespec64 *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_timespec64_cond(shhwtstamps->hwtstamp, ts)) 419 return TP_STATUS_TS_RAW_HARDWARE; 420 421 if (ktime_to_timespec64_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 timespec64 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 /* 439 * versions 1 through 3 overflow the timestamps in y2106, since they 440 * all store the seconds in a 32-bit unsigned integer. 441 * If we create a version 4, that should have a 64-bit timestamp, 442 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit 443 * nanoseconds. 444 */ 445 switch (po->tp_version) { 446 case TPACKET_V1: 447 h.h1->tp_sec = ts.tv_sec; 448 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 449 break; 450 case TPACKET_V2: 451 h.h2->tp_sec = ts.tv_sec; 452 h.h2->tp_nsec = ts.tv_nsec; 453 break; 454 case TPACKET_V3: 455 h.h3->tp_sec = ts.tv_sec; 456 h.h3->tp_nsec = ts.tv_nsec; 457 break; 458 default: 459 WARN(1, "TPACKET version not supported.\n"); 460 BUG(); 461 } 462 463 /* one flush is safe, as both fields always lie on the same cacheline */ 464 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 465 smp_wmb(); 466 467 return ts_status; 468 } 469 470 static void *packet_lookup_frame(const struct packet_sock *po, 471 const struct packet_ring_buffer *rb, 472 unsigned int position, 473 int status) 474 { 475 unsigned int pg_vec_pos, frame_offset; 476 union tpacket_uhdr h; 477 478 pg_vec_pos = position / rb->frames_per_block; 479 frame_offset = position % rb->frames_per_block; 480 481 h.raw = rb->pg_vec[pg_vec_pos].buffer + 482 (frame_offset * rb->frame_size); 483 484 if (status != __packet_get_status(po, h.raw)) 485 return NULL; 486 487 return h.raw; 488 } 489 490 static void *packet_current_frame(struct packet_sock *po, 491 struct packet_ring_buffer *rb, 492 int status) 493 { 494 return packet_lookup_frame(po, rb, rb->head, status); 495 } 496 497 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 498 { 499 del_timer_sync(&pkc->retire_blk_timer); 500 } 501 502 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 503 struct sk_buff_head *rb_queue) 504 { 505 struct tpacket_kbdq_core *pkc; 506 507 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 508 509 spin_lock_bh(&rb_queue->lock); 510 pkc->delete_blk_timer = 1; 511 spin_unlock_bh(&rb_queue->lock); 512 513 prb_del_retire_blk_timer(pkc); 514 } 515 516 static void prb_setup_retire_blk_timer(struct packet_sock *po) 517 { 518 struct tpacket_kbdq_core *pkc; 519 520 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 521 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, 522 0); 523 pkc->retire_blk_timer.expires = jiffies; 524 } 525 526 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 527 int blk_size_in_bytes) 528 { 529 struct net_device *dev; 530 unsigned int mbits, div; 531 struct ethtool_link_ksettings ecmd; 532 int err; 533 534 rtnl_lock(); 535 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 536 if (unlikely(!dev)) { 537 rtnl_unlock(); 538 return DEFAULT_PRB_RETIRE_TOV; 539 } 540 err = __ethtool_get_link_ksettings(dev, &ecmd); 541 rtnl_unlock(); 542 if (err) 543 return DEFAULT_PRB_RETIRE_TOV; 544 545 /* If the link speed is so slow you don't really 546 * need to worry about perf anyways 547 */ 548 if (ecmd.base.speed < SPEED_1000 || 549 ecmd.base.speed == SPEED_UNKNOWN) 550 return DEFAULT_PRB_RETIRE_TOV; 551 552 div = ecmd.base.speed / 1000; 553 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 554 555 if (div) 556 mbits /= div; 557 558 if (div) 559 return mbits + 1; 560 return mbits; 561 } 562 563 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 564 union tpacket_req_u *req_u) 565 { 566 p1->feature_req_word = req_u->req3.tp_feature_req_word; 567 } 568 569 static void init_prb_bdqc(struct packet_sock *po, 570 struct packet_ring_buffer *rb, 571 struct pgv *pg_vec, 572 union tpacket_req_u *req_u) 573 { 574 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 575 struct tpacket_block_desc *pbd; 576 577 memset(p1, 0x0, sizeof(*p1)); 578 579 p1->knxt_seq_num = 1; 580 p1->pkbdq = pg_vec; 581 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 582 p1->pkblk_start = pg_vec[0].buffer; 583 p1->kblk_size = req_u->req3.tp_block_size; 584 p1->knum_blocks = req_u->req3.tp_block_nr; 585 p1->hdrlen = po->tp_hdrlen; 586 p1->version = po->tp_version; 587 p1->last_kactive_blk_num = 0; 588 po->stats.stats3.tp_freeze_q_cnt = 0; 589 if (req_u->req3.tp_retire_blk_tov) 590 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 591 else 592 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 593 req_u->req3.tp_block_size); 594 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 595 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 596 597 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 598 prb_init_ft_ops(p1, req_u); 599 prb_setup_retire_blk_timer(po); 600 prb_open_block(p1, pbd); 601 } 602 603 /* Do NOT update the last_blk_num first. 604 * Assumes sk_buff_head lock is held. 605 */ 606 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 607 { 608 mod_timer(&pkc->retire_blk_timer, 609 jiffies + pkc->tov_in_jiffies); 610 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 611 } 612 613 /* 614 * Timer logic: 615 * 1) We refresh the timer only when we open a block. 616 * By doing this we don't waste cycles refreshing the timer 617 * on packet-by-packet basis. 618 * 619 * With a 1MB block-size, on a 1Gbps line, it will take 620 * i) ~8 ms to fill a block + ii) memcpy etc. 621 * In this cut we are not accounting for the memcpy time. 622 * 623 * So, if the user sets the 'tmo' to 10ms then the timer 624 * will never fire while the block is still getting filled 625 * (which is what we want). However, the user could choose 626 * to close a block early and that's fine. 627 * 628 * But when the timer does fire, we check whether or not to refresh it. 629 * Since the tmo granularity is in msecs, it is not too expensive 630 * to refresh the timer, lets say every '8' msecs. 631 * Either the user can set the 'tmo' or we can derive it based on 632 * a) line-speed and b) block-size. 633 * prb_calc_retire_blk_tmo() calculates the tmo. 634 * 635 */ 636 static void prb_retire_rx_blk_timer_expired(struct timer_list *t) 637 { 638 struct packet_sock *po = 639 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer); 640 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 641 unsigned int frozen; 642 struct tpacket_block_desc *pbd; 643 644 spin_lock(&po->sk.sk_receive_queue.lock); 645 646 frozen = prb_queue_frozen(pkc); 647 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 648 649 if (unlikely(pkc->delete_blk_timer)) 650 goto out; 651 652 /* We only need to plug the race when the block is partially filled. 653 * tpacket_rcv: 654 * lock(); increment BLOCK_NUM_PKTS; unlock() 655 * copy_bits() is in progress ... 656 * timer fires on other cpu: 657 * we can't retire the current block because copy_bits 658 * is in progress. 659 * 660 */ 661 if (BLOCK_NUM_PKTS(pbd)) { 662 while (atomic_read(&pkc->blk_fill_in_prog)) { 663 /* Waiting for skb_copy_bits to finish... */ 664 cpu_relax(); 665 } 666 } 667 668 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 669 if (!frozen) { 670 if (!BLOCK_NUM_PKTS(pbd)) { 671 /* An empty block. Just refresh the timer. */ 672 goto refresh_timer; 673 } 674 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 675 if (!prb_dispatch_next_block(pkc, po)) 676 goto refresh_timer; 677 else 678 goto out; 679 } else { 680 /* Case 1. Queue was frozen because user-space was 681 * lagging behind. 682 */ 683 if (prb_curr_blk_in_use(pbd)) { 684 /* 685 * Ok, user-space is still behind. 686 * So just refresh the timer. 687 */ 688 goto refresh_timer; 689 } else { 690 /* Case 2. queue was frozen,user-space caught up, 691 * now the link went idle && the timer fired. 692 * We don't have a block to close.So we open this 693 * block and restart the timer. 694 * opening a block thaws the queue,restarts timer 695 * Thawing/timer-refresh is a side effect. 696 */ 697 prb_open_block(pkc, pbd); 698 goto out; 699 } 700 } 701 } 702 703 refresh_timer: 704 _prb_refresh_rx_retire_blk_timer(pkc); 705 706 out: 707 spin_unlock(&po->sk.sk_receive_queue.lock); 708 } 709 710 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 711 struct tpacket_block_desc *pbd1, __u32 status) 712 { 713 /* Flush everything minus the block header */ 714 715 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 716 u8 *start, *end; 717 718 start = (u8 *)pbd1; 719 720 /* Skip the block header(we know header WILL fit in 4K) */ 721 start += PAGE_SIZE; 722 723 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 724 for (; start < end; start += PAGE_SIZE) 725 flush_dcache_page(pgv_to_page(start)); 726 727 smp_wmb(); 728 #endif 729 730 /* Now update the block status. */ 731 732 BLOCK_STATUS(pbd1) = status; 733 734 /* Flush the block header */ 735 736 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 737 start = (u8 *)pbd1; 738 flush_dcache_page(pgv_to_page(start)); 739 740 smp_wmb(); 741 #endif 742 } 743 744 /* 745 * Side effect: 746 * 747 * 1) flush the block 748 * 2) Increment active_blk_num 749 * 750 * Note:We DONT refresh the timer on purpose. 751 * Because almost always the next block will be opened. 752 */ 753 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 754 struct tpacket_block_desc *pbd1, 755 struct packet_sock *po, unsigned int stat) 756 { 757 __u32 status = TP_STATUS_USER | stat; 758 759 struct tpacket3_hdr *last_pkt; 760 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 761 struct sock *sk = &po->sk; 762 763 if (atomic_read(&po->tp_drops)) 764 status |= TP_STATUS_LOSING; 765 766 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 767 last_pkt->tp_next_offset = 0; 768 769 /* Get the ts of the last pkt */ 770 if (BLOCK_NUM_PKTS(pbd1)) { 771 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 772 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 773 } else { 774 /* Ok, we tmo'd - so get the current time. 775 * 776 * It shouldn't really happen as we don't close empty 777 * blocks. See prb_retire_rx_blk_timer_expired(). 778 */ 779 struct timespec64 ts; 780 ktime_get_real_ts64(&ts); 781 h1->ts_last_pkt.ts_sec = ts.tv_sec; 782 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 783 } 784 785 smp_wmb(); 786 787 /* Flush the block */ 788 prb_flush_block(pkc1, pbd1, status); 789 790 sk->sk_data_ready(sk); 791 792 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 793 } 794 795 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 796 { 797 pkc->reset_pending_on_curr_blk = 0; 798 } 799 800 /* 801 * Side effect of opening a block: 802 * 803 * 1) prb_queue is thawed. 804 * 2) retire_blk_timer is refreshed. 805 * 806 */ 807 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 808 struct tpacket_block_desc *pbd1) 809 { 810 struct timespec64 ts; 811 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 812 813 smp_rmb(); 814 815 /* We could have just memset this but we will lose the 816 * flexibility of making the priv area sticky 817 */ 818 819 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 820 BLOCK_NUM_PKTS(pbd1) = 0; 821 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 822 823 ktime_get_real_ts64(&ts); 824 825 h1->ts_first_pkt.ts_sec = ts.tv_sec; 826 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 827 828 pkc1->pkblk_start = (char *)pbd1; 829 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 830 831 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 832 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 833 834 pbd1->version = pkc1->version; 835 pkc1->prev = pkc1->nxt_offset; 836 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 837 838 prb_thaw_queue(pkc1); 839 _prb_refresh_rx_retire_blk_timer(pkc1); 840 841 smp_wmb(); 842 } 843 844 /* 845 * Queue freeze logic: 846 * 1) Assume tp_block_nr = 8 blocks. 847 * 2) At time 't0', user opens Rx ring. 848 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 849 * 4) user-space is either sleeping or processing block '0'. 850 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 851 * it will close block-7,loop around and try to fill block '0'. 852 * call-flow: 853 * __packet_lookup_frame_in_block 854 * prb_retire_current_block() 855 * prb_dispatch_next_block() 856 * |->(BLOCK_STATUS == USER) evaluates to true 857 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 858 * 6) Now there are two cases: 859 * 6.1) Link goes idle right after the queue is frozen. 860 * But remember, the last open_block() refreshed the timer. 861 * When this timer expires,it will refresh itself so that we can 862 * re-open block-0 in near future. 863 * 6.2) Link is busy and keeps on receiving packets. This is a simple 864 * case and __packet_lookup_frame_in_block will check if block-0 865 * is free and can now be re-used. 866 */ 867 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 868 struct packet_sock *po) 869 { 870 pkc->reset_pending_on_curr_blk = 1; 871 po->stats.stats3.tp_freeze_q_cnt++; 872 } 873 874 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 875 876 /* 877 * If the next block is free then we will dispatch it 878 * and return a good offset. 879 * Else, we will freeze the queue. 880 * So, caller must check the return value. 881 */ 882 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 883 struct packet_sock *po) 884 { 885 struct tpacket_block_desc *pbd; 886 887 smp_rmb(); 888 889 /* 1. Get current block num */ 890 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 891 892 /* 2. If this block is currently in_use then freeze the queue */ 893 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 894 prb_freeze_queue(pkc, po); 895 return NULL; 896 } 897 898 /* 899 * 3. 900 * open this block and return the offset where the first packet 901 * needs to get stored. 902 */ 903 prb_open_block(pkc, pbd); 904 return (void *)pkc->nxt_offset; 905 } 906 907 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 908 struct packet_sock *po, unsigned int status) 909 { 910 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 911 912 /* retire/close the current block */ 913 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 914 /* 915 * Plug the case where copy_bits() is in progress on 916 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 917 * have space to copy the pkt in the current block and 918 * called prb_retire_current_block() 919 * 920 * We don't need to worry about the TMO case because 921 * the timer-handler already handled this case. 922 */ 923 if (!(status & TP_STATUS_BLK_TMO)) { 924 while (atomic_read(&pkc->blk_fill_in_prog)) { 925 /* Waiting for skb_copy_bits to finish... */ 926 cpu_relax(); 927 } 928 } 929 prb_close_block(pkc, pbd, po, status); 930 return; 931 } 932 } 933 934 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) 935 { 936 return TP_STATUS_USER & BLOCK_STATUS(pbd); 937 } 938 939 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 940 { 941 return pkc->reset_pending_on_curr_blk; 942 } 943 944 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 945 { 946 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 947 atomic_dec(&pkc->blk_fill_in_prog); 948 } 949 950 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 951 struct tpacket3_hdr *ppd) 952 { 953 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 954 } 955 956 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 957 struct tpacket3_hdr *ppd) 958 { 959 ppd->hv1.tp_rxhash = 0; 960 } 961 962 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 963 struct tpacket3_hdr *ppd) 964 { 965 if (skb_vlan_tag_present(pkc->skb)) { 966 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 967 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 968 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 969 } else { 970 ppd->hv1.tp_vlan_tci = 0; 971 ppd->hv1.tp_vlan_tpid = 0; 972 ppd->tp_status = TP_STATUS_AVAILABLE; 973 } 974 } 975 976 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 977 struct tpacket3_hdr *ppd) 978 { 979 ppd->hv1.tp_padding = 0; 980 prb_fill_vlan_info(pkc, ppd); 981 982 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 983 prb_fill_rxhash(pkc, ppd); 984 else 985 prb_clear_rxhash(pkc, ppd); 986 } 987 988 static void prb_fill_curr_block(char *curr, 989 struct tpacket_kbdq_core *pkc, 990 struct tpacket_block_desc *pbd, 991 unsigned int len) 992 { 993 struct tpacket3_hdr *ppd; 994 995 ppd = (struct tpacket3_hdr *)curr; 996 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 997 pkc->prev = curr; 998 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 999 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1000 BLOCK_NUM_PKTS(pbd) += 1; 1001 atomic_inc(&pkc->blk_fill_in_prog); 1002 prb_run_all_ft_ops(pkc, ppd); 1003 } 1004 1005 /* Assumes caller has the sk->rx_queue.lock */ 1006 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1007 struct sk_buff *skb, 1008 unsigned int len 1009 ) 1010 { 1011 struct tpacket_kbdq_core *pkc; 1012 struct tpacket_block_desc *pbd; 1013 char *curr, *end; 1014 1015 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1016 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1017 1018 /* Queue is frozen when user space is lagging behind */ 1019 if (prb_queue_frozen(pkc)) { 1020 /* 1021 * Check if that last block which caused the queue to freeze, 1022 * is still in_use by user-space. 1023 */ 1024 if (prb_curr_blk_in_use(pbd)) { 1025 /* Can't record this packet */ 1026 return NULL; 1027 } else { 1028 /* 1029 * Ok, the block was released by user-space. 1030 * Now let's open that block. 1031 * opening a block also thaws the queue. 1032 * Thawing is a side effect. 1033 */ 1034 prb_open_block(pkc, pbd); 1035 } 1036 } 1037 1038 smp_mb(); 1039 curr = pkc->nxt_offset; 1040 pkc->skb = skb; 1041 end = (char *)pbd + pkc->kblk_size; 1042 1043 /* first try the current block */ 1044 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1045 prb_fill_curr_block(curr, pkc, pbd, len); 1046 return (void *)curr; 1047 } 1048 1049 /* Ok, close the current block */ 1050 prb_retire_current_block(pkc, po, 0); 1051 1052 /* Now, try to dispatch the next block */ 1053 curr = (char *)prb_dispatch_next_block(pkc, po); 1054 if (curr) { 1055 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1056 prb_fill_curr_block(curr, pkc, pbd, len); 1057 return (void *)curr; 1058 } 1059 1060 /* 1061 * No free blocks are available.user_space hasn't caught up yet. 1062 * Queue was just frozen and now this packet will get dropped. 1063 */ 1064 return NULL; 1065 } 1066 1067 static void *packet_current_rx_frame(struct packet_sock *po, 1068 struct sk_buff *skb, 1069 int status, unsigned int len) 1070 { 1071 char *curr = NULL; 1072 switch (po->tp_version) { 1073 case TPACKET_V1: 1074 case TPACKET_V2: 1075 curr = packet_lookup_frame(po, &po->rx_ring, 1076 po->rx_ring.head, status); 1077 return curr; 1078 case TPACKET_V3: 1079 return __packet_lookup_frame_in_block(po, skb, len); 1080 default: 1081 WARN(1, "TPACKET version not supported\n"); 1082 BUG(); 1083 return NULL; 1084 } 1085 } 1086 1087 static void *prb_lookup_block(const struct packet_sock *po, 1088 const struct packet_ring_buffer *rb, 1089 unsigned int idx, 1090 int status) 1091 { 1092 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1093 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1094 1095 if (status != BLOCK_STATUS(pbd)) 1096 return NULL; 1097 return pbd; 1098 } 1099 1100 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1101 { 1102 unsigned int prev; 1103 if (rb->prb_bdqc.kactive_blk_num) 1104 prev = rb->prb_bdqc.kactive_blk_num-1; 1105 else 1106 prev = rb->prb_bdqc.knum_blocks-1; 1107 return prev; 1108 } 1109 1110 /* Assumes caller has held the rx_queue.lock */ 1111 static void *__prb_previous_block(struct packet_sock *po, 1112 struct packet_ring_buffer *rb, 1113 int status) 1114 { 1115 unsigned int previous = prb_previous_blk_num(rb); 1116 return prb_lookup_block(po, rb, previous, status); 1117 } 1118 1119 static void *packet_previous_rx_frame(struct packet_sock *po, 1120 struct packet_ring_buffer *rb, 1121 int status) 1122 { 1123 if (po->tp_version <= TPACKET_V2) 1124 return packet_previous_frame(po, rb, status); 1125 1126 return __prb_previous_block(po, rb, status); 1127 } 1128 1129 static void packet_increment_rx_head(struct packet_sock *po, 1130 struct packet_ring_buffer *rb) 1131 { 1132 switch (po->tp_version) { 1133 case TPACKET_V1: 1134 case TPACKET_V2: 1135 return packet_increment_head(rb); 1136 case TPACKET_V3: 1137 default: 1138 WARN(1, "TPACKET version not supported.\n"); 1139 BUG(); 1140 return; 1141 } 1142 } 1143 1144 static void *packet_previous_frame(struct packet_sock *po, 1145 struct packet_ring_buffer *rb, 1146 int status) 1147 { 1148 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1149 return packet_lookup_frame(po, rb, previous, status); 1150 } 1151 1152 static void packet_increment_head(struct packet_ring_buffer *buff) 1153 { 1154 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1155 } 1156 1157 static void packet_inc_pending(struct packet_ring_buffer *rb) 1158 { 1159 this_cpu_inc(*rb->pending_refcnt); 1160 } 1161 1162 static void packet_dec_pending(struct packet_ring_buffer *rb) 1163 { 1164 this_cpu_dec(*rb->pending_refcnt); 1165 } 1166 1167 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1168 { 1169 unsigned int refcnt = 0; 1170 int cpu; 1171 1172 /* We don't use pending refcount in rx_ring. */ 1173 if (rb->pending_refcnt == NULL) 1174 return 0; 1175 1176 for_each_possible_cpu(cpu) 1177 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1178 1179 return refcnt; 1180 } 1181 1182 static int packet_alloc_pending(struct packet_sock *po) 1183 { 1184 po->rx_ring.pending_refcnt = NULL; 1185 1186 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1187 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1188 return -ENOBUFS; 1189 1190 return 0; 1191 } 1192 1193 static void packet_free_pending(struct packet_sock *po) 1194 { 1195 free_percpu(po->tx_ring.pending_refcnt); 1196 } 1197 1198 #define ROOM_POW_OFF 2 1199 #define ROOM_NONE 0x0 1200 #define ROOM_LOW 0x1 1201 #define ROOM_NORMAL 0x2 1202 1203 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off) 1204 { 1205 int idx, len; 1206 1207 len = READ_ONCE(po->rx_ring.frame_max) + 1; 1208 idx = READ_ONCE(po->rx_ring.head); 1209 if (pow_off) 1210 idx += len >> pow_off; 1211 if (idx >= len) 1212 idx -= len; 1213 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1214 } 1215 1216 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off) 1217 { 1218 int idx, len; 1219 1220 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks); 1221 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num); 1222 if (pow_off) 1223 idx += len >> pow_off; 1224 if (idx >= len) 1225 idx -= len; 1226 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1227 } 1228 1229 static int __packet_rcv_has_room(const struct packet_sock *po, 1230 const struct sk_buff *skb) 1231 { 1232 const struct sock *sk = &po->sk; 1233 int ret = ROOM_NONE; 1234 1235 if (po->prot_hook.func != tpacket_rcv) { 1236 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1237 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1238 - (skb ? skb->truesize : 0); 1239 1240 if (avail > (rcvbuf >> ROOM_POW_OFF)) 1241 return ROOM_NORMAL; 1242 else if (avail > 0) 1243 return ROOM_LOW; 1244 else 1245 return ROOM_NONE; 1246 } 1247 1248 if (po->tp_version == TPACKET_V3) { 1249 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1250 ret = ROOM_NORMAL; 1251 else if (__tpacket_v3_has_room(po, 0)) 1252 ret = ROOM_LOW; 1253 } else { 1254 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1255 ret = ROOM_NORMAL; 1256 else if (__tpacket_has_room(po, 0)) 1257 ret = ROOM_LOW; 1258 } 1259 1260 return ret; 1261 } 1262 1263 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1264 { 1265 int pressure, ret; 1266 1267 ret = __packet_rcv_has_room(po, skb); 1268 pressure = ret != ROOM_NORMAL; 1269 1270 if (READ_ONCE(po->pressure) != pressure) 1271 WRITE_ONCE(po->pressure, pressure); 1272 1273 return ret; 1274 } 1275 1276 static void packet_rcv_try_clear_pressure(struct packet_sock *po) 1277 { 1278 if (READ_ONCE(po->pressure) && 1279 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 1280 WRITE_ONCE(po->pressure, 0); 1281 } 1282 1283 static void packet_sock_destruct(struct sock *sk) 1284 { 1285 skb_queue_purge(&sk->sk_error_queue); 1286 1287 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1288 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 1289 1290 if (!sock_flag(sk, SOCK_DEAD)) { 1291 pr_err("Attempt to release alive packet socket: %p\n", sk); 1292 return; 1293 } 1294 1295 sk_refcnt_debug_dec(sk); 1296 } 1297 1298 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1299 { 1300 u32 *history = po->rollover->history; 1301 u32 victim, rxhash; 1302 int i, count = 0; 1303 1304 rxhash = skb_get_hash(skb); 1305 for (i = 0; i < ROLLOVER_HLEN; i++) 1306 if (READ_ONCE(history[i]) == rxhash) 1307 count++; 1308 1309 victim = prandom_u32() % ROLLOVER_HLEN; 1310 1311 /* Avoid dirtying the cache line if possible */ 1312 if (READ_ONCE(history[victim]) != rxhash) 1313 WRITE_ONCE(history[victim], rxhash); 1314 1315 return count > (ROLLOVER_HLEN >> 1); 1316 } 1317 1318 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1319 struct sk_buff *skb, 1320 unsigned int num) 1321 { 1322 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1323 } 1324 1325 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1326 struct sk_buff *skb, 1327 unsigned int num) 1328 { 1329 unsigned int val = atomic_inc_return(&f->rr_cur); 1330 1331 return val % num; 1332 } 1333 1334 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1335 struct sk_buff *skb, 1336 unsigned int num) 1337 { 1338 return smp_processor_id() % num; 1339 } 1340 1341 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1342 struct sk_buff *skb, 1343 unsigned int num) 1344 { 1345 return prandom_u32_max(num); 1346 } 1347 1348 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1349 struct sk_buff *skb, 1350 unsigned int idx, bool try_self, 1351 unsigned int num) 1352 { 1353 struct packet_sock *po, *po_next, *po_skip = NULL; 1354 unsigned int i, j, room = ROOM_NONE; 1355 1356 po = pkt_sk(f->arr[idx]); 1357 1358 if (try_self) { 1359 room = packet_rcv_has_room(po, skb); 1360 if (room == ROOM_NORMAL || 1361 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1362 return idx; 1363 po_skip = po; 1364 } 1365 1366 i = j = min_t(int, po->rollover->sock, num - 1); 1367 do { 1368 po_next = pkt_sk(f->arr[i]); 1369 if (po_next != po_skip && !READ_ONCE(po_next->pressure) && 1370 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1371 if (i != j) 1372 po->rollover->sock = i; 1373 atomic_long_inc(&po->rollover->num); 1374 if (room == ROOM_LOW) 1375 atomic_long_inc(&po->rollover->num_huge); 1376 return i; 1377 } 1378 1379 if (++i == num) 1380 i = 0; 1381 } while (i != j); 1382 1383 atomic_long_inc(&po->rollover->num_failed); 1384 return idx; 1385 } 1386 1387 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1388 struct sk_buff *skb, 1389 unsigned int num) 1390 { 1391 return skb_get_queue_mapping(skb) % num; 1392 } 1393 1394 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1395 struct sk_buff *skb, 1396 unsigned int num) 1397 { 1398 struct bpf_prog *prog; 1399 unsigned int ret = 0; 1400 1401 rcu_read_lock(); 1402 prog = rcu_dereference(f->bpf_prog); 1403 if (prog) 1404 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1405 rcu_read_unlock(); 1406 1407 return ret; 1408 } 1409 1410 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1411 { 1412 return f->flags & (flag >> 8); 1413 } 1414 1415 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1416 struct packet_type *pt, struct net_device *orig_dev) 1417 { 1418 struct packet_fanout *f = pt->af_packet_priv; 1419 unsigned int num = READ_ONCE(f->num_members); 1420 struct net *net = read_pnet(&f->net); 1421 struct packet_sock *po; 1422 unsigned int idx; 1423 1424 if (!net_eq(dev_net(dev), net) || !num) { 1425 kfree_skb(skb); 1426 return 0; 1427 } 1428 1429 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1430 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1431 if (!skb) 1432 return 0; 1433 } 1434 switch (f->type) { 1435 case PACKET_FANOUT_HASH: 1436 default: 1437 idx = fanout_demux_hash(f, skb, num); 1438 break; 1439 case PACKET_FANOUT_LB: 1440 idx = fanout_demux_lb(f, skb, num); 1441 break; 1442 case PACKET_FANOUT_CPU: 1443 idx = fanout_demux_cpu(f, skb, num); 1444 break; 1445 case PACKET_FANOUT_RND: 1446 idx = fanout_demux_rnd(f, skb, num); 1447 break; 1448 case PACKET_FANOUT_QM: 1449 idx = fanout_demux_qm(f, skb, num); 1450 break; 1451 case PACKET_FANOUT_ROLLOVER: 1452 idx = fanout_demux_rollover(f, skb, 0, false, num); 1453 break; 1454 case PACKET_FANOUT_CBPF: 1455 case PACKET_FANOUT_EBPF: 1456 idx = fanout_demux_bpf(f, skb, num); 1457 break; 1458 } 1459 1460 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1461 idx = fanout_demux_rollover(f, skb, idx, true, num); 1462 1463 po = pkt_sk(f->arr[idx]); 1464 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1465 } 1466 1467 DEFINE_MUTEX(fanout_mutex); 1468 EXPORT_SYMBOL_GPL(fanout_mutex); 1469 static LIST_HEAD(fanout_list); 1470 static u16 fanout_next_id; 1471 1472 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1473 { 1474 struct packet_fanout *f = po->fanout; 1475 1476 spin_lock(&f->lock); 1477 f->arr[f->num_members] = sk; 1478 smp_wmb(); 1479 f->num_members++; 1480 if (f->num_members == 1) 1481 dev_add_pack(&f->prot_hook); 1482 spin_unlock(&f->lock); 1483 } 1484 1485 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1486 { 1487 struct packet_fanout *f = po->fanout; 1488 int i; 1489 1490 spin_lock(&f->lock); 1491 for (i = 0; i < f->num_members; i++) { 1492 if (f->arr[i] == sk) 1493 break; 1494 } 1495 BUG_ON(i >= f->num_members); 1496 f->arr[i] = f->arr[f->num_members - 1]; 1497 f->num_members--; 1498 if (f->num_members == 0) 1499 __dev_remove_pack(&f->prot_hook); 1500 spin_unlock(&f->lock); 1501 } 1502 1503 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1504 { 1505 if (sk->sk_family != PF_PACKET) 1506 return false; 1507 1508 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1509 } 1510 1511 static void fanout_init_data(struct packet_fanout *f) 1512 { 1513 switch (f->type) { 1514 case PACKET_FANOUT_LB: 1515 atomic_set(&f->rr_cur, 0); 1516 break; 1517 case PACKET_FANOUT_CBPF: 1518 case PACKET_FANOUT_EBPF: 1519 RCU_INIT_POINTER(f->bpf_prog, NULL); 1520 break; 1521 } 1522 } 1523 1524 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1525 { 1526 struct bpf_prog *old; 1527 1528 spin_lock(&f->lock); 1529 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1530 rcu_assign_pointer(f->bpf_prog, new); 1531 spin_unlock(&f->lock); 1532 1533 if (old) { 1534 synchronize_net(); 1535 bpf_prog_destroy(old); 1536 } 1537 } 1538 1539 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data, 1540 unsigned int len) 1541 { 1542 struct bpf_prog *new; 1543 struct sock_fprog fprog; 1544 int ret; 1545 1546 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1547 return -EPERM; 1548 if (len != sizeof(fprog)) 1549 return -EINVAL; 1550 if (copy_from_user(&fprog, data, len)) 1551 return -EFAULT; 1552 1553 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1554 if (ret) 1555 return ret; 1556 1557 __fanout_set_data_bpf(po->fanout, new); 1558 return 0; 1559 } 1560 1561 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data, 1562 unsigned int len) 1563 { 1564 struct bpf_prog *new; 1565 u32 fd; 1566 1567 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1568 return -EPERM; 1569 if (len != sizeof(fd)) 1570 return -EINVAL; 1571 if (copy_from_user(&fd, data, len)) 1572 return -EFAULT; 1573 1574 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1575 if (IS_ERR(new)) 1576 return PTR_ERR(new); 1577 1578 __fanout_set_data_bpf(po->fanout, new); 1579 return 0; 1580 } 1581 1582 static int fanout_set_data(struct packet_sock *po, char __user *data, 1583 unsigned int len) 1584 { 1585 switch (po->fanout->type) { 1586 case PACKET_FANOUT_CBPF: 1587 return fanout_set_data_cbpf(po, data, len); 1588 case PACKET_FANOUT_EBPF: 1589 return fanout_set_data_ebpf(po, data, len); 1590 default: 1591 return -EINVAL; 1592 } 1593 } 1594 1595 static void fanout_release_data(struct packet_fanout *f) 1596 { 1597 switch (f->type) { 1598 case PACKET_FANOUT_CBPF: 1599 case PACKET_FANOUT_EBPF: 1600 __fanout_set_data_bpf(f, NULL); 1601 } 1602 } 1603 1604 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) 1605 { 1606 struct packet_fanout *f; 1607 1608 list_for_each_entry(f, &fanout_list, list) { 1609 if (f->id == candidate_id && 1610 read_pnet(&f->net) == sock_net(sk)) { 1611 return false; 1612 } 1613 } 1614 return true; 1615 } 1616 1617 static bool fanout_find_new_id(struct sock *sk, u16 *new_id) 1618 { 1619 u16 id = fanout_next_id; 1620 1621 do { 1622 if (__fanout_id_is_free(sk, id)) { 1623 *new_id = id; 1624 fanout_next_id = id + 1; 1625 return true; 1626 } 1627 1628 id++; 1629 } while (id != fanout_next_id); 1630 1631 return false; 1632 } 1633 1634 static int fanout_add(struct sock *sk, u16 id, u16 type_flags) 1635 { 1636 struct packet_rollover *rollover = NULL; 1637 struct packet_sock *po = pkt_sk(sk); 1638 struct packet_fanout *f, *match; 1639 u8 type = type_flags & 0xff; 1640 u8 flags = type_flags >> 8; 1641 int err; 1642 1643 switch (type) { 1644 case PACKET_FANOUT_ROLLOVER: 1645 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1646 return -EINVAL; 1647 case PACKET_FANOUT_HASH: 1648 case PACKET_FANOUT_LB: 1649 case PACKET_FANOUT_CPU: 1650 case PACKET_FANOUT_RND: 1651 case PACKET_FANOUT_QM: 1652 case PACKET_FANOUT_CBPF: 1653 case PACKET_FANOUT_EBPF: 1654 break; 1655 default: 1656 return -EINVAL; 1657 } 1658 1659 mutex_lock(&fanout_mutex); 1660 1661 err = -EALREADY; 1662 if (po->fanout) 1663 goto out; 1664 1665 if (type == PACKET_FANOUT_ROLLOVER || 1666 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1667 err = -ENOMEM; 1668 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); 1669 if (!rollover) 1670 goto out; 1671 atomic_long_set(&rollover->num, 0); 1672 atomic_long_set(&rollover->num_huge, 0); 1673 atomic_long_set(&rollover->num_failed, 0); 1674 } 1675 1676 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { 1677 if (id != 0) { 1678 err = -EINVAL; 1679 goto out; 1680 } 1681 if (!fanout_find_new_id(sk, &id)) { 1682 err = -ENOMEM; 1683 goto out; 1684 } 1685 /* ephemeral flag for the first socket in the group: drop it */ 1686 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); 1687 } 1688 1689 match = NULL; 1690 list_for_each_entry(f, &fanout_list, list) { 1691 if (f->id == id && 1692 read_pnet(&f->net) == sock_net(sk)) { 1693 match = f; 1694 break; 1695 } 1696 } 1697 err = -EINVAL; 1698 if (match && match->flags != flags) 1699 goto out; 1700 if (!match) { 1701 err = -ENOMEM; 1702 match = kzalloc(sizeof(*match), GFP_KERNEL); 1703 if (!match) 1704 goto out; 1705 write_pnet(&match->net, sock_net(sk)); 1706 match->id = id; 1707 match->type = type; 1708 match->flags = flags; 1709 INIT_LIST_HEAD(&match->list); 1710 spin_lock_init(&match->lock); 1711 refcount_set(&match->sk_ref, 0); 1712 fanout_init_data(match); 1713 match->prot_hook.type = po->prot_hook.type; 1714 match->prot_hook.dev = po->prot_hook.dev; 1715 match->prot_hook.func = packet_rcv_fanout; 1716 match->prot_hook.af_packet_priv = match; 1717 match->prot_hook.id_match = match_fanout_group; 1718 list_add(&match->list, &fanout_list); 1719 } 1720 err = -EINVAL; 1721 1722 spin_lock(&po->bind_lock); 1723 if (po->running && 1724 match->type == type && 1725 match->prot_hook.type == po->prot_hook.type && 1726 match->prot_hook.dev == po->prot_hook.dev) { 1727 err = -ENOSPC; 1728 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) { 1729 __dev_remove_pack(&po->prot_hook); 1730 po->fanout = match; 1731 po->rollover = rollover; 1732 rollover = NULL; 1733 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); 1734 __fanout_link(sk, po); 1735 err = 0; 1736 } 1737 } 1738 spin_unlock(&po->bind_lock); 1739 1740 if (err && !refcount_read(&match->sk_ref)) { 1741 list_del(&match->list); 1742 kfree(match); 1743 } 1744 1745 out: 1746 kfree(rollover); 1747 mutex_unlock(&fanout_mutex); 1748 return err; 1749 } 1750 1751 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes 1752 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. 1753 * It is the responsibility of the caller to call fanout_release_data() and 1754 * free the returned packet_fanout (after synchronize_net()) 1755 */ 1756 static struct packet_fanout *fanout_release(struct sock *sk) 1757 { 1758 struct packet_sock *po = pkt_sk(sk); 1759 struct packet_fanout *f; 1760 1761 mutex_lock(&fanout_mutex); 1762 f = po->fanout; 1763 if (f) { 1764 po->fanout = NULL; 1765 1766 if (refcount_dec_and_test(&f->sk_ref)) 1767 list_del(&f->list); 1768 else 1769 f = NULL; 1770 } 1771 mutex_unlock(&fanout_mutex); 1772 1773 return f; 1774 } 1775 1776 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1777 struct sk_buff *skb) 1778 { 1779 /* Earlier code assumed this would be a VLAN pkt, double-check 1780 * this now that we have the actual packet in hand. We can only 1781 * do this check on Ethernet devices. 1782 */ 1783 if (unlikely(dev->type != ARPHRD_ETHER)) 1784 return false; 1785 1786 skb_reset_mac_header(skb); 1787 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1788 } 1789 1790 static const struct proto_ops packet_ops; 1791 1792 static const struct proto_ops packet_ops_spkt; 1793 1794 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1795 struct packet_type *pt, struct net_device *orig_dev) 1796 { 1797 struct sock *sk; 1798 struct sockaddr_pkt *spkt; 1799 1800 /* 1801 * When we registered the protocol we saved the socket in the data 1802 * field for just this event. 1803 */ 1804 1805 sk = pt->af_packet_priv; 1806 1807 /* 1808 * Yank back the headers [hope the device set this 1809 * right or kerboom...] 1810 * 1811 * Incoming packets have ll header pulled, 1812 * push it back. 1813 * 1814 * For outgoing ones skb->data == skb_mac_header(skb) 1815 * so that this procedure is noop. 1816 */ 1817 1818 if (skb->pkt_type == PACKET_LOOPBACK) 1819 goto out; 1820 1821 if (!net_eq(dev_net(dev), sock_net(sk))) 1822 goto out; 1823 1824 skb = skb_share_check(skb, GFP_ATOMIC); 1825 if (skb == NULL) 1826 goto oom; 1827 1828 /* drop any routing info */ 1829 skb_dst_drop(skb); 1830 1831 /* drop conntrack reference */ 1832 nf_reset_ct(skb); 1833 1834 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1835 1836 skb_push(skb, skb->data - skb_mac_header(skb)); 1837 1838 /* 1839 * The SOCK_PACKET socket receives _all_ frames. 1840 */ 1841 1842 spkt->spkt_family = dev->type; 1843 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1844 spkt->spkt_protocol = skb->protocol; 1845 1846 /* 1847 * Charge the memory to the socket. This is done specifically 1848 * to prevent sockets using all the memory up. 1849 */ 1850 1851 if (sock_queue_rcv_skb(sk, skb) == 0) 1852 return 0; 1853 1854 out: 1855 kfree_skb(skb); 1856 oom: 1857 return 0; 1858 } 1859 1860 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock) 1861 { 1862 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && 1863 sock->type == SOCK_RAW) { 1864 skb_reset_mac_header(skb); 1865 skb->protocol = dev_parse_header_protocol(skb); 1866 } 1867 1868 skb_probe_transport_header(skb); 1869 } 1870 1871 /* 1872 * Output a raw packet to a device layer. This bypasses all the other 1873 * protocol layers and you must therefore supply it with a complete frame 1874 */ 1875 1876 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1877 size_t len) 1878 { 1879 struct sock *sk = sock->sk; 1880 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1881 struct sk_buff *skb = NULL; 1882 struct net_device *dev; 1883 struct sockcm_cookie sockc; 1884 __be16 proto = 0; 1885 int err; 1886 int extra_len = 0; 1887 1888 /* 1889 * Get and verify the address. 1890 */ 1891 1892 if (saddr) { 1893 if (msg->msg_namelen < sizeof(struct sockaddr)) 1894 return -EINVAL; 1895 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1896 proto = saddr->spkt_protocol; 1897 } else 1898 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1899 1900 /* 1901 * Find the device first to size check it 1902 */ 1903 1904 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1905 retry: 1906 rcu_read_lock(); 1907 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1908 err = -ENODEV; 1909 if (dev == NULL) 1910 goto out_unlock; 1911 1912 err = -ENETDOWN; 1913 if (!(dev->flags & IFF_UP)) 1914 goto out_unlock; 1915 1916 /* 1917 * You may not queue a frame bigger than the mtu. This is the lowest level 1918 * raw protocol and you must do your own fragmentation at this level. 1919 */ 1920 1921 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1922 if (!netif_supports_nofcs(dev)) { 1923 err = -EPROTONOSUPPORT; 1924 goto out_unlock; 1925 } 1926 extra_len = 4; /* We're doing our own CRC */ 1927 } 1928 1929 err = -EMSGSIZE; 1930 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 1931 goto out_unlock; 1932 1933 if (!skb) { 1934 size_t reserved = LL_RESERVED_SPACE(dev); 1935 int tlen = dev->needed_tailroom; 1936 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 1937 1938 rcu_read_unlock(); 1939 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 1940 if (skb == NULL) 1941 return -ENOBUFS; 1942 /* FIXME: Save some space for broken drivers that write a hard 1943 * header at transmission time by themselves. PPP is the notable 1944 * one here. This should really be fixed at the driver level. 1945 */ 1946 skb_reserve(skb, reserved); 1947 skb_reset_network_header(skb); 1948 1949 /* Try to align data part correctly */ 1950 if (hhlen) { 1951 skb->data -= hhlen; 1952 skb->tail -= hhlen; 1953 if (len < hhlen) 1954 skb_reset_network_header(skb); 1955 } 1956 err = memcpy_from_msg(skb_put(skb, len), msg, len); 1957 if (err) 1958 goto out_free; 1959 goto retry; 1960 } 1961 1962 if (!dev_validate_header(dev, skb->data, len)) { 1963 err = -EINVAL; 1964 goto out_unlock; 1965 } 1966 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 1967 !packet_extra_vlan_len_allowed(dev, skb)) { 1968 err = -EMSGSIZE; 1969 goto out_unlock; 1970 } 1971 1972 sockcm_init(&sockc, sk); 1973 if (msg->msg_controllen) { 1974 err = sock_cmsg_send(sk, msg, &sockc); 1975 if (unlikely(err)) 1976 goto out_unlock; 1977 } 1978 1979 skb->protocol = proto; 1980 skb->dev = dev; 1981 skb->priority = sk->sk_priority; 1982 skb->mark = sk->sk_mark; 1983 skb->tstamp = sockc.transmit_time; 1984 1985 skb_setup_tx_timestamp(skb, sockc.tsflags); 1986 1987 if (unlikely(extra_len == 4)) 1988 skb->no_fcs = 1; 1989 1990 packet_parse_headers(skb, sock); 1991 1992 dev_queue_xmit(skb); 1993 rcu_read_unlock(); 1994 return len; 1995 1996 out_unlock: 1997 rcu_read_unlock(); 1998 out_free: 1999 kfree_skb(skb); 2000 return err; 2001 } 2002 2003 static unsigned int run_filter(struct sk_buff *skb, 2004 const struct sock *sk, 2005 unsigned int res) 2006 { 2007 struct sk_filter *filter; 2008 2009 rcu_read_lock(); 2010 filter = rcu_dereference(sk->sk_filter); 2011 if (filter != NULL) 2012 res = bpf_prog_run_clear_cb(filter->prog, skb); 2013 rcu_read_unlock(); 2014 2015 return res; 2016 } 2017 2018 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2019 size_t *len) 2020 { 2021 struct virtio_net_hdr vnet_hdr; 2022 2023 if (*len < sizeof(vnet_hdr)) 2024 return -EINVAL; 2025 *len -= sizeof(vnet_hdr); 2026 2027 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0)) 2028 return -EINVAL; 2029 2030 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 2031 } 2032 2033 /* 2034 * This function makes lazy skb cloning in hope that most of packets 2035 * are discarded by BPF. 2036 * 2037 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2038 * and skb->cb are mangled. It works because (and until) packets 2039 * falling here are owned by current CPU. Output packets are cloned 2040 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2041 * sequencially, so that if we return skb to original state on exit, 2042 * we will not harm anyone. 2043 */ 2044 2045 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2046 struct packet_type *pt, struct net_device *orig_dev) 2047 { 2048 struct sock *sk; 2049 struct sockaddr_ll *sll; 2050 struct packet_sock *po; 2051 u8 *skb_head = skb->data; 2052 int skb_len = skb->len; 2053 unsigned int snaplen, res; 2054 bool is_drop_n_account = false; 2055 2056 if (skb->pkt_type == PACKET_LOOPBACK) 2057 goto drop; 2058 2059 sk = pt->af_packet_priv; 2060 po = pkt_sk(sk); 2061 2062 if (!net_eq(dev_net(dev), sock_net(sk))) 2063 goto drop; 2064 2065 skb->dev = dev; 2066 2067 if (dev->header_ops) { 2068 /* The device has an explicit notion of ll header, 2069 * exported to higher levels. 2070 * 2071 * Otherwise, the device hides details of its frame 2072 * structure, so that corresponding packet head is 2073 * never delivered to user. 2074 */ 2075 if (sk->sk_type != SOCK_DGRAM) 2076 skb_push(skb, skb->data - skb_mac_header(skb)); 2077 else if (skb->pkt_type == PACKET_OUTGOING) { 2078 /* Special case: outgoing packets have ll header at head */ 2079 skb_pull(skb, skb_network_offset(skb)); 2080 } 2081 } 2082 2083 snaplen = skb->len; 2084 2085 res = run_filter(skb, sk, snaplen); 2086 if (!res) 2087 goto drop_n_restore; 2088 if (snaplen > res) 2089 snaplen = res; 2090 2091 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2092 goto drop_n_acct; 2093 2094 if (skb_shared(skb)) { 2095 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2096 if (nskb == NULL) 2097 goto drop_n_acct; 2098 2099 if (skb_head != skb->data) { 2100 skb->data = skb_head; 2101 skb->len = skb_len; 2102 } 2103 consume_skb(skb); 2104 skb = nskb; 2105 } 2106 2107 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2108 2109 sll = &PACKET_SKB_CB(skb)->sa.ll; 2110 sll->sll_hatype = dev->type; 2111 sll->sll_pkttype = skb->pkt_type; 2112 if (unlikely(po->origdev)) 2113 sll->sll_ifindex = orig_dev->ifindex; 2114 else 2115 sll->sll_ifindex = dev->ifindex; 2116 2117 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2118 2119 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2120 * Use their space for storing the original skb length. 2121 */ 2122 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2123 2124 if (pskb_trim(skb, snaplen)) 2125 goto drop_n_acct; 2126 2127 skb_set_owner_r(skb, sk); 2128 skb->dev = NULL; 2129 skb_dst_drop(skb); 2130 2131 /* drop conntrack reference */ 2132 nf_reset_ct(skb); 2133 2134 spin_lock(&sk->sk_receive_queue.lock); 2135 po->stats.stats1.tp_packets++; 2136 sock_skb_set_dropcount(sk, skb); 2137 __skb_queue_tail(&sk->sk_receive_queue, skb); 2138 spin_unlock(&sk->sk_receive_queue.lock); 2139 sk->sk_data_ready(sk); 2140 return 0; 2141 2142 drop_n_acct: 2143 is_drop_n_account = true; 2144 atomic_inc(&po->tp_drops); 2145 atomic_inc(&sk->sk_drops); 2146 2147 drop_n_restore: 2148 if (skb_head != skb->data && skb_shared(skb)) { 2149 skb->data = skb_head; 2150 skb->len = skb_len; 2151 } 2152 drop: 2153 if (!is_drop_n_account) 2154 consume_skb(skb); 2155 else 2156 kfree_skb(skb); 2157 return 0; 2158 } 2159 2160 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2161 struct packet_type *pt, struct net_device *orig_dev) 2162 { 2163 struct sock *sk; 2164 struct packet_sock *po; 2165 struct sockaddr_ll *sll; 2166 union tpacket_uhdr h; 2167 u8 *skb_head = skb->data; 2168 int skb_len = skb->len; 2169 unsigned int snaplen, res; 2170 unsigned long status = TP_STATUS_USER; 2171 unsigned short macoff, netoff, hdrlen; 2172 struct sk_buff *copy_skb = NULL; 2173 struct timespec64 ts; 2174 __u32 ts_status; 2175 bool is_drop_n_account = false; 2176 bool do_vnet = false; 2177 2178 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2179 * We may add members to them until current aligned size without forcing 2180 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2181 */ 2182 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2183 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2184 2185 if (skb->pkt_type == PACKET_LOOPBACK) 2186 goto drop; 2187 2188 sk = pt->af_packet_priv; 2189 po = pkt_sk(sk); 2190 2191 if (!net_eq(dev_net(dev), sock_net(sk))) 2192 goto drop; 2193 2194 if (dev->header_ops) { 2195 if (sk->sk_type != SOCK_DGRAM) 2196 skb_push(skb, skb->data - skb_mac_header(skb)); 2197 else if (skb->pkt_type == PACKET_OUTGOING) { 2198 /* Special case: outgoing packets have ll header at head */ 2199 skb_pull(skb, skb_network_offset(skb)); 2200 } 2201 } 2202 2203 snaplen = skb->len; 2204 2205 res = run_filter(skb, sk, snaplen); 2206 if (!res) 2207 goto drop_n_restore; 2208 2209 /* If we are flooded, just give up */ 2210 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) { 2211 atomic_inc(&po->tp_drops); 2212 goto drop_n_restore; 2213 } 2214 2215 if (skb->ip_summed == CHECKSUM_PARTIAL) 2216 status |= TP_STATUS_CSUMNOTREADY; 2217 else if (skb->pkt_type != PACKET_OUTGOING && 2218 (skb->ip_summed == CHECKSUM_COMPLETE || 2219 skb_csum_unnecessary(skb))) 2220 status |= TP_STATUS_CSUM_VALID; 2221 2222 if (snaplen > res) 2223 snaplen = res; 2224 2225 if (sk->sk_type == SOCK_DGRAM) { 2226 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2227 po->tp_reserve; 2228 } else { 2229 unsigned int maclen = skb_network_offset(skb); 2230 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2231 (maclen < 16 ? 16 : maclen)) + 2232 po->tp_reserve; 2233 if (po->has_vnet_hdr) { 2234 netoff += sizeof(struct virtio_net_hdr); 2235 do_vnet = true; 2236 } 2237 macoff = netoff - maclen; 2238 } 2239 if (po->tp_version <= TPACKET_V2) { 2240 if (macoff + snaplen > po->rx_ring.frame_size) { 2241 if (po->copy_thresh && 2242 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2243 if (skb_shared(skb)) { 2244 copy_skb = skb_clone(skb, GFP_ATOMIC); 2245 } else { 2246 copy_skb = skb_get(skb); 2247 skb_head = skb->data; 2248 } 2249 if (copy_skb) 2250 skb_set_owner_r(copy_skb, sk); 2251 } 2252 snaplen = po->rx_ring.frame_size - macoff; 2253 if ((int)snaplen < 0) { 2254 snaplen = 0; 2255 do_vnet = false; 2256 } 2257 } 2258 } else if (unlikely(macoff + snaplen > 2259 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2260 u32 nval; 2261 2262 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2263 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2264 snaplen, nval, macoff); 2265 snaplen = nval; 2266 if (unlikely((int)snaplen < 0)) { 2267 snaplen = 0; 2268 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2269 do_vnet = false; 2270 } 2271 } 2272 spin_lock(&sk->sk_receive_queue.lock); 2273 h.raw = packet_current_rx_frame(po, skb, 2274 TP_STATUS_KERNEL, (macoff+snaplen)); 2275 if (!h.raw) 2276 goto drop_n_account; 2277 2278 if (do_vnet && 2279 virtio_net_hdr_from_skb(skb, h.raw + macoff - 2280 sizeof(struct virtio_net_hdr), 2281 vio_le(), true, 0)) 2282 goto drop_n_account; 2283 2284 if (po->tp_version <= TPACKET_V2) { 2285 packet_increment_rx_head(po, &po->rx_ring); 2286 /* 2287 * LOSING will be reported till you read the stats, 2288 * because it's COR - Clear On Read. 2289 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2290 * at packet level. 2291 */ 2292 if (atomic_read(&po->tp_drops)) 2293 status |= TP_STATUS_LOSING; 2294 } 2295 2296 po->stats.stats1.tp_packets++; 2297 if (copy_skb) { 2298 status |= TP_STATUS_COPY; 2299 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2300 } 2301 spin_unlock(&sk->sk_receive_queue.lock); 2302 2303 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2304 2305 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) 2306 ktime_get_real_ts64(&ts); 2307 2308 status |= ts_status; 2309 2310 switch (po->tp_version) { 2311 case TPACKET_V1: 2312 h.h1->tp_len = skb->len; 2313 h.h1->tp_snaplen = snaplen; 2314 h.h1->tp_mac = macoff; 2315 h.h1->tp_net = netoff; 2316 h.h1->tp_sec = ts.tv_sec; 2317 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2318 hdrlen = sizeof(*h.h1); 2319 break; 2320 case TPACKET_V2: 2321 h.h2->tp_len = skb->len; 2322 h.h2->tp_snaplen = snaplen; 2323 h.h2->tp_mac = macoff; 2324 h.h2->tp_net = netoff; 2325 h.h2->tp_sec = ts.tv_sec; 2326 h.h2->tp_nsec = ts.tv_nsec; 2327 if (skb_vlan_tag_present(skb)) { 2328 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2329 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2330 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2331 } else { 2332 h.h2->tp_vlan_tci = 0; 2333 h.h2->tp_vlan_tpid = 0; 2334 } 2335 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2336 hdrlen = sizeof(*h.h2); 2337 break; 2338 case TPACKET_V3: 2339 /* tp_nxt_offset,vlan are already populated above. 2340 * So DONT clear those fields here 2341 */ 2342 h.h3->tp_status |= status; 2343 h.h3->tp_len = skb->len; 2344 h.h3->tp_snaplen = snaplen; 2345 h.h3->tp_mac = macoff; 2346 h.h3->tp_net = netoff; 2347 h.h3->tp_sec = ts.tv_sec; 2348 h.h3->tp_nsec = ts.tv_nsec; 2349 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2350 hdrlen = sizeof(*h.h3); 2351 break; 2352 default: 2353 BUG(); 2354 } 2355 2356 sll = h.raw + TPACKET_ALIGN(hdrlen); 2357 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2358 sll->sll_family = AF_PACKET; 2359 sll->sll_hatype = dev->type; 2360 sll->sll_protocol = skb->protocol; 2361 sll->sll_pkttype = skb->pkt_type; 2362 if (unlikely(po->origdev)) 2363 sll->sll_ifindex = orig_dev->ifindex; 2364 else 2365 sll->sll_ifindex = dev->ifindex; 2366 2367 smp_mb(); 2368 2369 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2370 if (po->tp_version <= TPACKET_V2) { 2371 u8 *start, *end; 2372 2373 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2374 macoff + snaplen); 2375 2376 for (start = h.raw; start < end; start += PAGE_SIZE) 2377 flush_dcache_page(pgv_to_page(start)); 2378 } 2379 smp_wmb(); 2380 #endif 2381 2382 if (po->tp_version <= TPACKET_V2) { 2383 __packet_set_status(po, h.raw, status); 2384 sk->sk_data_ready(sk); 2385 } else { 2386 prb_clear_blk_fill_status(&po->rx_ring); 2387 } 2388 2389 drop_n_restore: 2390 if (skb_head != skb->data && skb_shared(skb)) { 2391 skb->data = skb_head; 2392 skb->len = skb_len; 2393 } 2394 drop: 2395 if (!is_drop_n_account) 2396 consume_skb(skb); 2397 else 2398 kfree_skb(skb); 2399 return 0; 2400 2401 drop_n_account: 2402 spin_unlock(&sk->sk_receive_queue.lock); 2403 atomic_inc(&po->tp_drops); 2404 is_drop_n_account = true; 2405 2406 sk->sk_data_ready(sk); 2407 kfree_skb(copy_skb); 2408 goto drop_n_restore; 2409 } 2410 2411 static void tpacket_destruct_skb(struct sk_buff *skb) 2412 { 2413 struct packet_sock *po = pkt_sk(skb->sk); 2414 2415 if (likely(po->tx_ring.pg_vec)) { 2416 void *ph; 2417 __u32 ts; 2418 2419 ph = skb_zcopy_get_nouarg(skb); 2420 packet_dec_pending(&po->tx_ring); 2421 2422 ts = __packet_set_timestamp(po, ph, skb); 2423 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2424 2425 if (!packet_read_pending(&po->tx_ring)) 2426 complete(&po->skb_completion); 2427 } 2428 2429 sock_wfree(skb); 2430 } 2431 2432 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2433 { 2434 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2435 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2436 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2437 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2438 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2439 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2440 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2441 2442 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2443 return -EINVAL; 2444 2445 return 0; 2446 } 2447 2448 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2449 struct virtio_net_hdr *vnet_hdr) 2450 { 2451 if (*len < sizeof(*vnet_hdr)) 2452 return -EINVAL; 2453 *len -= sizeof(*vnet_hdr); 2454 2455 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2456 return -EFAULT; 2457 2458 return __packet_snd_vnet_parse(vnet_hdr, *len); 2459 } 2460 2461 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2462 void *frame, struct net_device *dev, void *data, int tp_len, 2463 __be16 proto, unsigned char *addr, int hlen, int copylen, 2464 const struct sockcm_cookie *sockc) 2465 { 2466 union tpacket_uhdr ph; 2467 int to_write, offset, len, nr_frags, len_max; 2468 struct socket *sock = po->sk.sk_socket; 2469 struct page *page; 2470 int err; 2471 2472 ph.raw = frame; 2473 2474 skb->protocol = proto; 2475 skb->dev = dev; 2476 skb->priority = po->sk.sk_priority; 2477 skb->mark = po->sk.sk_mark; 2478 skb->tstamp = sockc->transmit_time; 2479 skb_setup_tx_timestamp(skb, sockc->tsflags); 2480 skb_zcopy_set_nouarg(skb, ph.raw); 2481 2482 skb_reserve(skb, hlen); 2483 skb_reset_network_header(skb); 2484 2485 to_write = tp_len; 2486 2487 if (sock->type == SOCK_DGRAM) { 2488 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2489 NULL, tp_len); 2490 if (unlikely(err < 0)) 2491 return -EINVAL; 2492 } else if (copylen) { 2493 int hdrlen = min_t(int, copylen, tp_len); 2494 2495 skb_push(skb, dev->hard_header_len); 2496 skb_put(skb, copylen - dev->hard_header_len); 2497 err = skb_store_bits(skb, 0, data, hdrlen); 2498 if (unlikely(err)) 2499 return err; 2500 if (!dev_validate_header(dev, skb->data, hdrlen)) 2501 return -EINVAL; 2502 2503 data += hdrlen; 2504 to_write -= hdrlen; 2505 } 2506 2507 offset = offset_in_page(data); 2508 len_max = PAGE_SIZE - offset; 2509 len = ((to_write > len_max) ? len_max : to_write); 2510 2511 skb->data_len = to_write; 2512 skb->len += to_write; 2513 skb->truesize += to_write; 2514 refcount_add(to_write, &po->sk.sk_wmem_alloc); 2515 2516 while (likely(to_write)) { 2517 nr_frags = skb_shinfo(skb)->nr_frags; 2518 2519 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2520 pr_err("Packet exceed the number of skb frags(%lu)\n", 2521 MAX_SKB_FRAGS); 2522 return -EFAULT; 2523 } 2524 2525 page = pgv_to_page(data); 2526 data += len; 2527 flush_dcache_page(page); 2528 get_page(page); 2529 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2530 to_write -= len; 2531 offset = 0; 2532 len_max = PAGE_SIZE; 2533 len = ((to_write > len_max) ? len_max : to_write); 2534 } 2535 2536 packet_parse_headers(skb, sock); 2537 2538 return tp_len; 2539 } 2540 2541 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2542 int size_max, void **data) 2543 { 2544 union tpacket_uhdr ph; 2545 int tp_len, off; 2546 2547 ph.raw = frame; 2548 2549 switch (po->tp_version) { 2550 case TPACKET_V3: 2551 if (ph.h3->tp_next_offset != 0) { 2552 pr_warn_once("variable sized slot not supported"); 2553 return -EINVAL; 2554 } 2555 tp_len = ph.h3->tp_len; 2556 break; 2557 case TPACKET_V2: 2558 tp_len = ph.h2->tp_len; 2559 break; 2560 default: 2561 tp_len = ph.h1->tp_len; 2562 break; 2563 } 2564 if (unlikely(tp_len > size_max)) { 2565 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2566 return -EMSGSIZE; 2567 } 2568 2569 if (unlikely(po->tp_tx_has_off)) { 2570 int off_min, off_max; 2571 2572 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2573 off_max = po->tx_ring.frame_size - tp_len; 2574 if (po->sk.sk_type == SOCK_DGRAM) { 2575 switch (po->tp_version) { 2576 case TPACKET_V3: 2577 off = ph.h3->tp_net; 2578 break; 2579 case TPACKET_V2: 2580 off = ph.h2->tp_net; 2581 break; 2582 default: 2583 off = ph.h1->tp_net; 2584 break; 2585 } 2586 } else { 2587 switch (po->tp_version) { 2588 case TPACKET_V3: 2589 off = ph.h3->tp_mac; 2590 break; 2591 case TPACKET_V2: 2592 off = ph.h2->tp_mac; 2593 break; 2594 default: 2595 off = ph.h1->tp_mac; 2596 break; 2597 } 2598 } 2599 if (unlikely((off < off_min) || (off_max < off))) 2600 return -EINVAL; 2601 } else { 2602 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2603 } 2604 2605 *data = frame + off; 2606 return tp_len; 2607 } 2608 2609 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2610 { 2611 struct sk_buff *skb = NULL; 2612 struct net_device *dev; 2613 struct virtio_net_hdr *vnet_hdr = NULL; 2614 struct sockcm_cookie sockc; 2615 __be16 proto; 2616 int err, reserve = 0; 2617 void *ph; 2618 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2619 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2620 unsigned char *addr = NULL; 2621 int tp_len, size_max; 2622 void *data; 2623 int len_sum = 0; 2624 int status = TP_STATUS_AVAILABLE; 2625 int hlen, tlen, copylen = 0; 2626 long timeo = 0; 2627 2628 mutex_lock(&po->pg_vec_lock); 2629 2630 /* packet_sendmsg() check on tx_ring.pg_vec was lockless, 2631 * we need to confirm it under protection of pg_vec_lock. 2632 */ 2633 if (unlikely(!po->tx_ring.pg_vec)) { 2634 err = -EBUSY; 2635 goto out; 2636 } 2637 if (likely(saddr == NULL)) { 2638 dev = packet_cached_dev_get(po); 2639 proto = po->num; 2640 } else { 2641 err = -EINVAL; 2642 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2643 goto out; 2644 if (msg->msg_namelen < (saddr->sll_halen 2645 + offsetof(struct sockaddr_ll, 2646 sll_addr))) 2647 goto out; 2648 proto = saddr->sll_protocol; 2649 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2650 if (po->sk.sk_socket->type == SOCK_DGRAM) { 2651 if (dev && msg->msg_namelen < dev->addr_len + 2652 offsetof(struct sockaddr_ll, sll_addr)) 2653 goto out_put; 2654 addr = saddr->sll_addr; 2655 } 2656 } 2657 2658 err = -ENXIO; 2659 if (unlikely(dev == NULL)) 2660 goto out; 2661 err = -ENETDOWN; 2662 if (unlikely(!(dev->flags & IFF_UP))) 2663 goto out_put; 2664 2665 sockcm_init(&sockc, &po->sk); 2666 if (msg->msg_controllen) { 2667 err = sock_cmsg_send(&po->sk, msg, &sockc); 2668 if (unlikely(err)) 2669 goto out_put; 2670 } 2671 2672 if (po->sk.sk_socket->type == SOCK_RAW) 2673 reserve = dev->hard_header_len; 2674 size_max = po->tx_ring.frame_size 2675 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2676 2677 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) 2678 size_max = dev->mtu + reserve + VLAN_HLEN; 2679 2680 reinit_completion(&po->skb_completion); 2681 2682 do { 2683 ph = packet_current_frame(po, &po->tx_ring, 2684 TP_STATUS_SEND_REQUEST); 2685 if (unlikely(ph == NULL)) { 2686 if (need_wait && skb) { 2687 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT); 2688 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo); 2689 if (timeo <= 0) { 2690 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS; 2691 goto out_put; 2692 } 2693 } 2694 /* check for additional frames */ 2695 continue; 2696 } 2697 2698 skb = NULL; 2699 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2700 if (tp_len < 0) 2701 goto tpacket_error; 2702 2703 status = TP_STATUS_SEND_REQUEST; 2704 hlen = LL_RESERVED_SPACE(dev); 2705 tlen = dev->needed_tailroom; 2706 if (po->has_vnet_hdr) { 2707 vnet_hdr = data; 2708 data += sizeof(*vnet_hdr); 2709 tp_len -= sizeof(*vnet_hdr); 2710 if (tp_len < 0 || 2711 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2712 tp_len = -EINVAL; 2713 goto tpacket_error; 2714 } 2715 copylen = __virtio16_to_cpu(vio_le(), 2716 vnet_hdr->hdr_len); 2717 } 2718 copylen = max_t(int, copylen, dev->hard_header_len); 2719 skb = sock_alloc_send_skb(&po->sk, 2720 hlen + tlen + sizeof(struct sockaddr_ll) + 2721 (copylen - dev->hard_header_len), 2722 !need_wait, &err); 2723 2724 if (unlikely(skb == NULL)) { 2725 /* we assume the socket was initially writeable ... */ 2726 if (likely(len_sum > 0)) 2727 err = len_sum; 2728 goto out_status; 2729 } 2730 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2731 addr, hlen, copylen, &sockc); 2732 if (likely(tp_len >= 0) && 2733 tp_len > dev->mtu + reserve && 2734 !po->has_vnet_hdr && 2735 !packet_extra_vlan_len_allowed(dev, skb)) 2736 tp_len = -EMSGSIZE; 2737 2738 if (unlikely(tp_len < 0)) { 2739 tpacket_error: 2740 if (po->tp_loss) { 2741 __packet_set_status(po, ph, 2742 TP_STATUS_AVAILABLE); 2743 packet_increment_head(&po->tx_ring); 2744 kfree_skb(skb); 2745 continue; 2746 } else { 2747 status = TP_STATUS_WRONG_FORMAT; 2748 err = tp_len; 2749 goto out_status; 2750 } 2751 } 2752 2753 if (po->has_vnet_hdr) { 2754 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) { 2755 tp_len = -EINVAL; 2756 goto tpacket_error; 2757 } 2758 virtio_net_hdr_set_proto(skb, vnet_hdr); 2759 } 2760 2761 skb->destructor = tpacket_destruct_skb; 2762 __packet_set_status(po, ph, TP_STATUS_SENDING); 2763 packet_inc_pending(&po->tx_ring); 2764 2765 status = TP_STATUS_SEND_REQUEST; 2766 err = po->xmit(skb); 2767 if (unlikely(err > 0)) { 2768 err = net_xmit_errno(err); 2769 if (err && __packet_get_status(po, ph) == 2770 TP_STATUS_AVAILABLE) { 2771 /* skb was destructed already */ 2772 skb = NULL; 2773 goto out_status; 2774 } 2775 /* 2776 * skb was dropped but not destructed yet; 2777 * let's treat it like congestion or err < 0 2778 */ 2779 err = 0; 2780 } 2781 packet_increment_head(&po->tx_ring); 2782 len_sum += tp_len; 2783 } while (likely((ph != NULL) || 2784 /* Note: packet_read_pending() might be slow if we have 2785 * to call it as it's per_cpu variable, but in fast-path 2786 * we already short-circuit the loop with the first 2787 * condition, and luckily don't have to go that path 2788 * anyway. 2789 */ 2790 (need_wait && packet_read_pending(&po->tx_ring)))); 2791 2792 err = len_sum; 2793 goto out_put; 2794 2795 out_status: 2796 __packet_set_status(po, ph, status); 2797 kfree_skb(skb); 2798 out_put: 2799 dev_put(dev); 2800 out: 2801 mutex_unlock(&po->pg_vec_lock); 2802 return err; 2803 } 2804 2805 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2806 size_t reserve, size_t len, 2807 size_t linear, int noblock, 2808 int *err) 2809 { 2810 struct sk_buff *skb; 2811 2812 /* Under a page? Don't bother with paged skb. */ 2813 if (prepad + len < PAGE_SIZE || !linear) 2814 linear = len; 2815 2816 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2817 err, 0); 2818 if (!skb) 2819 return NULL; 2820 2821 skb_reserve(skb, reserve); 2822 skb_put(skb, linear); 2823 skb->data_len = len - linear; 2824 skb->len += len - linear; 2825 2826 return skb; 2827 } 2828 2829 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2830 { 2831 struct sock *sk = sock->sk; 2832 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2833 struct sk_buff *skb; 2834 struct net_device *dev; 2835 __be16 proto; 2836 unsigned char *addr = NULL; 2837 int err, reserve = 0; 2838 struct sockcm_cookie sockc; 2839 struct virtio_net_hdr vnet_hdr = { 0 }; 2840 int offset = 0; 2841 struct packet_sock *po = pkt_sk(sk); 2842 bool has_vnet_hdr = false; 2843 int hlen, tlen, linear; 2844 int extra_len = 0; 2845 2846 /* 2847 * Get and verify the address. 2848 */ 2849 2850 if (likely(saddr == NULL)) { 2851 dev = packet_cached_dev_get(po); 2852 proto = po->num; 2853 } else { 2854 err = -EINVAL; 2855 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2856 goto out; 2857 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2858 goto out; 2859 proto = saddr->sll_protocol; 2860 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2861 if (sock->type == SOCK_DGRAM) { 2862 if (dev && msg->msg_namelen < dev->addr_len + 2863 offsetof(struct sockaddr_ll, sll_addr)) 2864 goto out_unlock; 2865 addr = saddr->sll_addr; 2866 } 2867 } 2868 2869 err = -ENXIO; 2870 if (unlikely(dev == NULL)) 2871 goto out_unlock; 2872 err = -ENETDOWN; 2873 if (unlikely(!(dev->flags & IFF_UP))) 2874 goto out_unlock; 2875 2876 sockcm_init(&sockc, sk); 2877 sockc.mark = sk->sk_mark; 2878 if (msg->msg_controllen) { 2879 err = sock_cmsg_send(sk, msg, &sockc); 2880 if (unlikely(err)) 2881 goto out_unlock; 2882 } 2883 2884 if (sock->type == SOCK_RAW) 2885 reserve = dev->hard_header_len; 2886 if (po->has_vnet_hdr) { 2887 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2888 if (err) 2889 goto out_unlock; 2890 has_vnet_hdr = true; 2891 } 2892 2893 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2894 if (!netif_supports_nofcs(dev)) { 2895 err = -EPROTONOSUPPORT; 2896 goto out_unlock; 2897 } 2898 extra_len = 4; /* We're doing our own CRC */ 2899 } 2900 2901 err = -EMSGSIZE; 2902 if (!vnet_hdr.gso_type && 2903 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 2904 goto out_unlock; 2905 2906 err = -ENOBUFS; 2907 hlen = LL_RESERVED_SPACE(dev); 2908 tlen = dev->needed_tailroom; 2909 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); 2910 linear = max(linear, min_t(int, len, dev->hard_header_len)); 2911 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, 2912 msg->msg_flags & MSG_DONTWAIT, &err); 2913 if (skb == NULL) 2914 goto out_unlock; 2915 2916 skb_reset_network_header(skb); 2917 2918 err = -EINVAL; 2919 if (sock->type == SOCK_DGRAM) { 2920 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 2921 if (unlikely(offset < 0)) 2922 goto out_free; 2923 } else if (reserve) { 2924 skb_reserve(skb, -reserve); 2925 if (len < reserve + sizeof(struct ipv6hdr) && 2926 dev->min_header_len != dev->hard_header_len) 2927 skb_reset_network_header(skb); 2928 } 2929 2930 /* Returns -EFAULT on error */ 2931 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 2932 if (err) 2933 goto out_free; 2934 2935 if (sock->type == SOCK_RAW && 2936 !dev_validate_header(dev, skb->data, len)) { 2937 err = -EINVAL; 2938 goto out_free; 2939 } 2940 2941 skb_setup_tx_timestamp(skb, sockc.tsflags); 2942 2943 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 2944 !packet_extra_vlan_len_allowed(dev, skb)) { 2945 err = -EMSGSIZE; 2946 goto out_free; 2947 } 2948 2949 skb->protocol = proto; 2950 skb->dev = dev; 2951 skb->priority = sk->sk_priority; 2952 skb->mark = sockc.mark; 2953 skb->tstamp = sockc.transmit_time; 2954 2955 if (has_vnet_hdr) { 2956 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 2957 if (err) 2958 goto out_free; 2959 len += sizeof(vnet_hdr); 2960 virtio_net_hdr_set_proto(skb, &vnet_hdr); 2961 } 2962 2963 packet_parse_headers(skb, sock); 2964 2965 if (unlikely(extra_len == 4)) 2966 skb->no_fcs = 1; 2967 2968 err = po->xmit(skb); 2969 if (err > 0 && (err = net_xmit_errno(err)) != 0) 2970 goto out_unlock; 2971 2972 dev_put(dev); 2973 2974 return len; 2975 2976 out_free: 2977 kfree_skb(skb); 2978 out_unlock: 2979 if (dev) 2980 dev_put(dev); 2981 out: 2982 return err; 2983 } 2984 2985 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2986 { 2987 struct sock *sk = sock->sk; 2988 struct packet_sock *po = pkt_sk(sk); 2989 2990 if (po->tx_ring.pg_vec) 2991 return tpacket_snd(po, msg); 2992 else 2993 return packet_snd(sock, msg, len); 2994 } 2995 2996 /* 2997 * Close a PACKET socket. This is fairly simple. We immediately go 2998 * to 'closed' state and remove our protocol entry in the device list. 2999 */ 3000 3001 static int packet_release(struct socket *sock) 3002 { 3003 struct sock *sk = sock->sk; 3004 struct packet_sock *po; 3005 struct packet_fanout *f; 3006 struct net *net; 3007 union tpacket_req_u req_u; 3008 3009 if (!sk) 3010 return 0; 3011 3012 net = sock_net(sk); 3013 po = pkt_sk(sk); 3014 3015 mutex_lock(&net->packet.sklist_lock); 3016 sk_del_node_init_rcu(sk); 3017 mutex_unlock(&net->packet.sklist_lock); 3018 3019 preempt_disable(); 3020 sock_prot_inuse_add(net, sk->sk_prot, -1); 3021 preempt_enable(); 3022 3023 spin_lock(&po->bind_lock); 3024 unregister_prot_hook(sk, false); 3025 packet_cached_dev_reset(po); 3026 3027 if (po->prot_hook.dev) { 3028 dev_put(po->prot_hook.dev); 3029 po->prot_hook.dev = NULL; 3030 } 3031 spin_unlock(&po->bind_lock); 3032 3033 packet_flush_mclist(sk); 3034 3035 lock_sock(sk); 3036 if (po->rx_ring.pg_vec) { 3037 memset(&req_u, 0, sizeof(req_u)); 3038 packet_set_ring(sk, &req_u, 1, 0); 3039 } 3040 3041 if (po->tx_ring.pg_vec) { 3042 memset(&req_u, 0, sizeof(req_u)); 3043 packet_set_ring(sk, &req_u, 1, 1); 3044 } 3045 release_sock(sk); 3046 3047 f = fanout_release(sk); 3048 3049 synchronize_net(); 3050 3051 kfree(po->rollover); 3052 if (f) { 3053 fanout_release_data(f); 3054 kfree(f); 3055 } 3056 /* 3057 * Now the socket is dead. No more input will appear. 3058 */ 3059 sock_orphan(sk); 3060 sock->sk = NULL; 3061 3062 /* Purge queues */ 3063 3064 skb_queue_purge(&sk->sk_receive_queue); 3065 packet_free_pending(po); 3066 sk_refcnt_debug_release(sk); 3067 3068 sock_put(sk); 3069 return 0; 3070 } 3071 3072 /* 3073 * Attach a packet hook. 3074 */ 3075 3076 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3077 __be16 proto) 3078 { 3079 struct packet_sock *po = pkt_sk(sk); 3080 struct net_device *dev_curr; 3081 __be16 proto_curr; 3082 bool need_rehook; 3083 struct net_device *dev = NULL; 3084 int ret = 0; 3085 bool unlisted = false; 3086 3087 lock_sock(sk); 3088 spin_lock(&po->bind_lock); 3089 rcu_read_lock(); 3090 3091 if (po->fanout) { 3092 ret = -EINVAL; 3093 goto out_unlock; 3094 } 3095 3096 if (name) { 3097 dev = dev_get_by_name_rcu(sock_net(sk), name); 3098 if (!dev) { 3099 ret = -ENODEV; 3100 goto out_unlock; 3101 } 3102 } else if (ifindex) { 3103 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3104 if (!dev) { 3105 ret = -ENODEV; 3106 goto out_unlock; 3107 } 3108 } 3109 3110 if (dev) 3111 dev_hold(dev); 3112 3113 proto_curr = po->prot_hook.type; 3114 dev_curr = po->prot_hook.dev; 3115 3116 need_rehook = proto_curr != proto || dev_curr != dev; 3117 3118 if (need_rehook) { 3119 if (po->running) { 3120 rcu_read_unlock(); 3121 /* prevents packet_notifier() from calling 3122 * register_prot_hook() 3123 */ 3124 po->num = 0; 3125 __unregister_prot_hook(sk, true); 3126 rcu_read_lock(); 3127 dev_curr = po->prot_hook.dev; 3128 if (dev) 3129 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3130 dev->ifindex); 3131 } 3132 3133 BUG_ON(po->running); 3134 po->num = proto; 3135 po->prot_hook.type = proto; 3136 3137 if (unlikely(unlisted)) { 3138 dev_put(dev); 3139 po->prot_hook.dev = NULL; 3140 po->ifindex = -1; 3141 packet_cached_dev_reset(po); 3142 } else { 3143 po->prot_hook.dev = dev; 3144 po->ifindex = dev ? dev->ifindex : 0; 3145 packet_cached_dev_assign(po, dev); 3146 } 3147 } 3148 if (dev_curr) 3149 dev_put(dev_curr); 3150 3151 if (proto == 0 || !need_rehook) 3152 goto out_unlock; 3153 3154 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3155 register_prot_hook(sk); 3156 } else { 3157 sk->sk_err = ENETDOWN; 3158 if (!sock_flag(sk, SOCK_DEAD)) 3159 sk->sk_error_report(sk); 3160 } 3161 3162 out_unlock: 3163 rcu_read_unlock(); 3164 spin_unlock(&po->bind_lock); 3165 release_sock(sk); 3166 return ret; 3167 } 3168 3169 /* 3170 * Bind a packet socket to a device 3171 */ 3172 3173 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3174 int addr_len) 3175 { 3176 struct sock *sk = sock->sk; 3177 char name[sizeof(uaddr->sa_data) + 1]; 3178 3179 /* 3180 * Check legality 3181 */ 3182 3183 if (addr_len != sizeof(struct sockaddr)) 3184 return -EINVAL; 3185 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be 3186 * zero-terminated. 3187 */ 3188 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); 3189 name[sizeof(uaddr->sa_data)] = 0; 3190 3191 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3192 } 3193 3194 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3195 { 3196 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3197 struct sock *sk = sock->sk; 3198 3199 /* 3200 * Check legality 3201 */ 3202 3203 if (addr_len < sizeof(struct sockaddr_ll)) 3204 return -EINVAL; 3205 if (sll->sll_family != AF_PACKET) 3206 return -EINVAL; 3207 3208 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3209 sll->sll_protocol ? : pkt_sk(sk)->num); 3210 } 3211 3212 static struct proto packet_proto = { 3213 .name = "PACKET", 3214 .owner = THIS_MODULE, 3215 .obj_size = sizeof(struct packet_sock), 3216 }; 3217 3218 /* 3219 * Create a packet of type SOCK_PACKET. 3220 */ 3221 3222 static int packet_create(struct net *net, struct socket *sock, int protocol, 3223 int kern) 3224 { 3225 struct sock *sk; 3226 struct packet_sock *po; 3227 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3228 int err; 3229 3230 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3231 return -EPERM; 3232 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3233 sock->type != SOCK_PACKET) 3234 return -ESOCKTNOSUPPORT; 3235 3236 sock->state = SS_UNCONNECTED; 3237 3238 err = -ENOBUFS; 3239 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3240 if (sk == NULL) 3241 goto out; 3242 3243 sock->ops = &packet_ops; 3244 if (sock->type == SOCK_PACKET) 3245 sock->ops = &packet_ops_spkt; 3246 3247 sock_init_data(sock, sk); 3248 3249 po = pkt_sk(sk); 3250 init_completion(&po->skb_completion); 3251 sk->sk_family = PF_PACKET; 3252 po->num = proto; 3253 po->xmit = dev_queue_xmit; 3254 3255 err = packet_alloc_pending(po); 3256 if (err) 3257 goto out2; 3258 3259 packet_cached_dev_reset(po); 3260 3261 sk->sk_destruct = packet_sock_destruct; 3262 sk_refcnt_debug_inc(sk); 3263 3264 /* 3265 * Attach a protocol block 3266 */ 3267 3268 spin_lock_init(&po->bind_lock); 3269 mutex_init(&po->pg_vec_lock); 3270 po->rollover = NULL; 3271 po->prot_hook.func = packet_rcv; 3272 3273 if (sock->type == SOCK_PACKET) 3274 po->prot_hook.func = packet_rcv_spkt; 3275 3276 po->prot_hook.af_packet_priv = sk; 3277 3278 if (proto) { 3279 po->prot_hook.type = proto; 3280 __register_prot_hook(sk); 3281 } 3282 3283 mutex_lock(&net->packet.sklist_lock); 3284 sk_add_node_tail_rcu(sk, &net->packet.sklist); 3285 mutex_unlock(&net->packet.sklist_lock); 3286 3287 preempt_disable(); 3288 sock_prot_inuse_add(net, &packet_proto, 1); 3289 preempt_enable(); 3290 3291 return 0; 3292 out2: 3293 sk_free(sk); 3294 out: 3295 return err; 3296 } 3297 3298 /* 3299 * Pull a packet from our receive queue and hand it to the user. 3300 * If necessary we block. 3301 */ 3302 3303 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3304 int flags) 3305 { 3306 struct sock *sk = sock->sk; 3307 struct sk_buff *skb; 3308 int copied, err; 3309 int vnet_hdr_len = 0; 3310 unsigned int origlen = 0; 3311 3312 err = -EINVAL; 3313 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3314 goto out; 3315 3316 #if 0 3317 /* What error should we return now? EUNATTACH? */ 3318 if (pkt_sk(sk)->ifindex < 0) 3319 return -ENODEV; 3320 #endif 3321 3322 if (flags & MSG_ERRQUEUE) { 3323 err = sock_recv_errqueue(sk, msg, len, 3324 SOL_PACKET, PACKET_TX_TIMESTAMP); 3325 goto out; 3326 } 3327 3328 /* 3329 * Call the generic datagram receiver. This handles all sorts 3330 * of horrible races and re-entrancy so we can forget about it 3331 * in the protocol layers. 3332 * 3333 * Now it will return ENETDOWN, if device have just gone down, 3334 * but then it will block. 3335 */ 3336 3337 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); 3338 3339 /* 3340 * An error occurred so return it. Because skb_recv_datagram() 3341 * handles the blocking we don't see and worry about blocking 3342 * retries. 3343 */ 3344 3345 if (skb == NULL) 3346 goto out; 3347 3348 packet_rcv_try_clear_pressure(pkt_sk(sk)); 3349 3350 if (pkt_sk(sk)->has_vnet_hdr) { 3351 err = packet_rcv_vnet(msg, skb, &len); 3352 if (err) 3353 goto out_free; 3354 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3355 } 3356 3357 /* You lose any data beyond the buffer you gave. If it worries 3358 * a user program they can ask the device for its MTU 3359 * anyway. 3360 */ 3361 copied = skb->len; 3362 if (copied > len) { 3363 copied = len; 3364 msg->msg_flags |= MSG_TRUNC; 3365 } 3366 3367 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3368 if (err) 3369 goto out_free; 3370 3371 if (sock->type != SOCK_PACKET) { 3372 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3373 3374 /* Original length was stored in sockaddr_ll fields */ 3375 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3376 sll->sll_family = AF_PACKET; 3377 sll->sll_protocol = skb->protocol; 3378 } 3379 3380 sock_recv_ts_and_drops(msg, sk, skb); 3381 3382 if (msg->msg_name) { 3383 int copy_len; 3384 3385 /* If the address length field is there to be filled 3386 * in, we fill it in now. 3387 */ 3388 if (sock->type == SOCK_PACKET) { 3389 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3390 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3391 copy_len = msg->msg_namelen; 3392 } else { 3393 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3394 3395 msg->msg_namelen = sll->sll_halen + 3396 offsetof(struct sockaddr_ll, sll_addr); 3397 copy_len = msg->msg_namelen; 3398 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { 3399 memset(msg->msg_name + 3400 offsetof(struct sockaddr_ll, sll_addr), 3401 0, sizeof(sll->sll_addr)); 3402 msg->msg_namelen = sizeof(struct sockaddr_ll); 3403 } 3404 } 3405 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); 3406 } 3407 3408 if (pkt_sk(sk)->auxdata) { 3409 struct tpacket_auxdata aux; 3410 3411 aux.tp_status = TP_STATUS_USER; 3412 if (skb->ip_summed == CHECKSUM_PARTIAL) 3413 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3414 else if (skb->pkt_type != PACKET_OUTGOING && 3415 (skb->ip_summed == CHECKSUM_COMPLETE || 3416 skb_csum_unnecessary(skb))) 3417 aux.tp_status |= TP_STATUS_CSUM_VALID; 3418 3419 aux.tp_len = origlen; 3420 aux.tp_snaplen = skb->len; 3421 aux.tp_mac = 0; 3422 aux.tp_net = skb_network_offset(skb); 3423 if (skb_vlan_tag_present(skb)) { 3424 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3425 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3426 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3427 } else { 3428 aux.tp_vlan_tci = 0; 3429 aux.tp_vlan_tpid = 0; 3430 } 3431 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3432 } 3433 3434 /* 3435 * Free or return the buffer as appropriate. Again this 3436 * hides all the races and re-entrancy issues from us. 3437 */ 3438 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3439 3440 out_free: 3441 skb_free_datagram(sk, skb); 3442 out: 3443 return err; 3444 } 3445 3446 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3447 int peer) 3448 { 3449 struct net_device *dev; 3450 struct sock *sk = sock->sk; 3451 3452 if (peer) 3453 return -EOPNOTSUPP; 3454 3455 uaddr->sa_family = AF_PACKET; 3456 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3457 rcu_read_lock(); 3458 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); 3459 if (dev) 3460 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3461 rcu_read_unlock(); 3462 3463 return sizeof(*uaddr); 3464 } 3465 3466 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3467 int peer) 3468 { 3469 struct net_device *dev; 3470 struct sock *sk = sock->sk; 3471 struct packet_sock *po = pkt_sk(sk); 3472 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3473 3474 if (peer) 3475 return -EOPNOTSUPP; 3476 3477 sll->sll_family = AF_PACKET; 3478 sll->sll_ifindex = po->ifindex; 3479 sll->sll_protocol = po->num; 3480 sll->sll_pkttype = 0; 3481 rcu_read_lock(); 3482 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); 3483 if (dev) { 3484 sll->sll_hatype = dev->type; 3485 sll->sll_halen = dev->addr_len; 3486 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3487 } else { 3488 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3489 sll->sll_halen = 0; 3490 } 3491 rcu_read_unlock(); 3492 3493 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3494 } 3495 3496 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3497 int what) 3498 { 3499 switch (i->type) { 3500 case PACKET_MR_MULTICAST: 3501 if (i->alen != dev->addr_len) 3502 return -EINVAL; 3503 if (what > 0) 3504 return dev_mc_add(dev, i->addr); 3505 else 3506 return dev_mc_del(dev, i->addr); 3507 break; 3508 case PACKET_MR_PROMISC: 3509 return dev_set_promiscuity(dev, what); 3510 case PACKET_MR_ALLMULTI: 3511 return dev_set_allmulti(dev, what); 3512 case PACKET_MR_UNICAST: 3513 if (i->alen != dev->addr_len) 3514 return -EINVAL; 3515 if (what > 0) 3516 return dev_uc_add(dev, i->addr); 3517 else 3518 return dev_uc_del(dev, i->addr); 3519 break; 3520 default: 3521 break; 3522 } 3523 return 0; 3524 } 3525 3526 static void packet_dev_mclist_delete(struct net_device *dev, 3527 struct packet_mclist **mlp) 3528 { 3529 struct packet_mclist *ml; 3530 3531 while ((ml = *mlp) != NULL) { 3532 if (ml->ifindex == dev->ifindex) { 3533 packet_dev_mc(dev, ml, -1); 3534 *mlp = ml->next; 3535 kfree(ml); 3536 } else 3537 mlp = &ml->next; 3538 } 3539 } 3540 3541 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3542 { 3543 struct packet_sock *po = pkt_sk(sk); 3544 struct packet_mclist *ml, *i; 3545 struct net_device *dev; 3546 int err; 3547 3548 rtnl_lock(); 3549 3550 err = -ENODEV; 3551 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3552 if (!dev) 3553 goto done; 3554 3555 err = -EINVAL; 3556 if (mreq->mr_alen > dev->addr_len) 3557 goto done; 3558 3559 err = -ENOBUFS; 3560 i = kmalloc(sizeof(*i), GFP_KERNEL); 3561 if (i == NULL) 3562 goto done; 3563 3564 err = 0; 3565 for (ml = po->mclist; ml; ml = ml->next) { 3566 if (ml->ifindex == mreq->mr_ifindex && 3567 ml->type == mreq->mr_type && 3568 ml->alen == mreq->mr_alen && 3569 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3570 ml->count++; 3571 /* Free the new element ... */ 3572 kfree(i); 3573 goto done; 3574 } 3575 } 3576 3577 i->type = mreq->mr_type; 3578 i->ifindex = mreq->mr_ifindex; 3579 i->alen = mreq->mr_alen; 3580 memcpy(i->addr, mreq->mr_address, i->alen); 3581 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3582 i->count = 1; 3583 i->next = po->mclist; 3584 po->mclist = i; 3585 err = packet_dev_mc(dev, i, 1); 3586 if (err) { 3587 po->mclist = i->next; 3588 kfree(i); 3589 } 3590 3591 done: 3592 rtnl_unlock(); 3593 return err; 3594 } 3595 3596 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3597 { 3598 struct packet_mclist *ml, **mlp; 3599 3600 rtnl_lock(); 3601 3602 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3603 if (ml->ifindex == mreq->mr_ifindex && 3604 ml->type == mreq->mr_type && 3605 ml->alen == mreq->mr_alen && 3606 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3607 if (--ml->count == 0) { 3608 struct net_device *dev; 3609 *mlp = ml->next; 3610 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3611 if (dev) 3612 packet_dev_mc(dev, ml, -1); 3613 kfree(ml); 3614 } 3615 break; 3616 } 3617 } 3618 rtnl_unlock(); 3619 return 0; 3620 } 3621 3622 static void packet_flush_mclist(struct sock *sk) 3623 { 3624 struct packet_sock *po = pkt_sk(sk); 3625 struct packet_mclist *ml; 3626 3627 if (!po->mclist) 3628 return; 3629 3630 rtnl_lock(); 3631 while ((ml = po->mclist) != NULL) { 3632 struct net_device *dev; 3633 3634 po->mclist = ml->next; 3635 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3636 if (dev != NULL) 3637 packet_dev_mc(dev, ml, -1); 3638 kfree(ml); 3639 } 3640 rtnl_unlock(); 3641 } 3642 3643 static int 3644 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) 3645 { 3646 struct sock *sk = sock->sk; 3647 struct packet_sock *po = pkt_sk(sk); 3648 int ret; 3649 3650 if (level != SOL_PACKET) 3651 return -ENOPROTOOPT; 3652 3653 switch (optname) { 3654 case PACKET_ADD_MEMBERSHIP: 3655 case PACKET_DROP_MEMBERSHIP: 3656 { 3657 struct packet_mreq_max mreq; 3658 int len = optlen; 3659 memset(&mreq, 0, sizeof(mreq)); 3660 if (len < sizeof(struct packet_mreq)) 3661 return -EINVAL; 3662 if (len > sizeof(mreq)) 3663 len = sizeof(mreq); 3664 if (copy_from_user(&mreq, optval, len)) 3665 return -EFAULT; 3666 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3667 return -EINVAL; 3668 if (optname == PACKET_ADD_MEMBERSHIP) 3669 ret = packet_mc_add(sk, &mreq); 3670 else 3671 ret = packet_mc_drop(sk, &mreq); 3672 return ret; 3673 } 3674 3675 case PACKET_RX_RING: 3676 case PACKET_TX_RING: 3677 { 3678 union tpacket_req_u req_u; 3679 int len; 3680 3681 lock_sock(sk); 3682 switch (po->tp_version) { 3683 case TPACKET_V1: 3684 case TPACKET_V2: 3685 len = sizeof(req_u.req); 3686 break; 3687 case TPACKET_V3: 3688 default: 3689 len = sizeof(req_u.req3); 3690 break; 3691 } 3692 if (optlen < len) { 3693 ret = -EINVAL; 3694 } else { 3695 if (copy_from_user(&req_u.req, optval, len)) 3696 ret = -EFAULT; 3697 else 3698 ret = packet_set_ring(sk, &req_u, 0, 3699 optname == PACKET_TX_RING); 3700 } 3701 release_sock(sk); 3702 return ret; 3703 } 3704 case PACKET_COPY_THRESH: 3705 { 3706 int val; 3707 3708 if (optlen != sizeof(val)) 3709 return -EINVAL; 3710 if (copy_from_user(&val, optval, sizeof(val))) 3711 return -EFAULT; 3712 3713 pkt_sk(sk)->copy_thresh = val; 3714 return 0; 3715 } 3716 case PACKET_VERSION: 3717 { 3718 int val; 3719 3720 if (optlen != sizeof(val)) 3721 return -EINVAL; 3722 if (copy_from_user(&val, optval, sizeof(val))) 3723 return -EFAULT; 3724 switch (val) { 3725 case TPACKET_V1: 3726 case TPACKET_V2: 3727 case TPACKET_V3: 3728 break; 3729 default: 3730 return -EINVAL; 3731 } 3732 lock_sock(sk); 3733 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3734 ret = -EBUSY; 3735 } else { 3736 po->tp_version = val; 3737 ret = 0; 3738 } 3739 release_sock(sk); 3740 return ret; 3741 } 3742 case PACKET_RESERVE: 3743 { 3744 unsigned int val; 3745 3746 if (optlen != sizeof(val)) 3747 return -EINVAL; 3748 if (copy_from_user(&val, optval, sizeof(val))) 3749 return -EFAULT; 3750 if (val > INT_MAX) 3751 return -EINVAL; 3752 lock_sock(sk); 3753 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3754 ret = -EBUSY; 3755 } else { 3756 po->tp_reserve = val; 3757 ret = 0; 3758 } 3759 release_sock(sk); 3760 return ret; 3761 } 3762 case PACKET_LOSS: 3763 { 3764 unsigned int val; 3765 3766 if (optlen != sizeof(val)) 3767 return -EINVAL; 3768 if (copy_from_user(&val, optval, sizeof(val))) 3769 return -EFAULT; 3770 3771 lock_sock(sk); 3772 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3773 ret = -EBUSY; 3774 } else { 3775 po->tp_loss = !!val; 3776 ret = 0; 3777 } 3778 release_sock(sk); 3779 return ret; 3780 } 3781 case PACKET_AUXDATA: 3782 { 3783 int val; 3784 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 po->auxdata = !!val; 3792 release_sock(sk); 3793 return 0; 3794 } 3795 case PACKET_ORIGDEV: 3796 { 3797 int val; 3798 3799 if (optlen < sizeof(val)) 3800 return -EINVAL; 3801 if (copy_from_user(&val, optval, sizeof(val))) 3802 return -EFAULT; 3803 3804 lock_sock(sk); 3805 po->origdev = !!val; 3806 release_sock(sk); 3807 return 0; 3808 } 3809 case PACKET_VNET_HDR: 3810 { 3811 int val; 3812 3813 if (sock->type != SOCK_RAW) 3814 return -EINVAL; 3815 if (optlen < sizeof(val)) 3816 return -EINVAL; 3817 if (copy_from_user(&val, optval, sizeof(val))) 3818 return -EFAULT; 3819 3820 lock_sock(sk); 3821 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3822 ret = -EBUSY; 3823 } else { 3824 po->has_vnet_hdr = !!val; 3825 ret = 0; 3826 } 3827 release_sock(sk); 3828 return ret; 3829 } 3830 case PACKET_TIMESTAMP: 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 3839 po->tp_tstamp = val; 3840 return 0; 3841 } 3842 case PACKET_FANOUT: 3843 { 3844 int val; 3845 3846 if (optlen != sizeof(val)) 3847 return -EINVAL; 3848 if (copy_from_user(&val, optval, sizeof(val))) 3849 return -EFAULT; 3850 3851 return fanout_add(sk, val & 0xffff, val >> 16); 3852 } 3853 case PACKET_FANOUT_DATA: 3854 { 3855 if (!po->fanout) 3856 return -EINVAL; 3857 3858 return fanout_set_data(po, optval, optlen); 3859 } 3860 case PACKET_IGNORE_OUTGOING: 3861 { 3862 int val; 3863 3864 if (optlen != sizeof(val)) 3865 return -EINVAL; 3866 if (copy_from_user(&val, optval, sizeof(val))) 3867 return -EFAULT; 3868 if (val < 0 || val > 1) 3869 return -EINVAL; 3870 3871 po->prot_hook.ignore_outgoing = !!val; 3872 return 0; 3873 } 3874 case PACKET_TX_HAS_OFF: 3875 { 3876 unsigned int val; 3877 3878 if (optlen != sizeof(val)) 3879 return -EINVAL; 3880 if (copy_from_user(&val, optval, sizeof(val))) 3881 return -EFAULT; 3882 3883 lock_sock(sk); 3884 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3885 ret = -EBUSY; 3886 } else { 3887 po->tp_tx_has_off = !!val; 3888 ret = 0; 3889 } 3890 release_sock(sk); 3891 return 0; 3892 } 3893 case PACKET_QDISC_BYPASS: 3894 { 3895 int val; 3896 3897 if (optlen != sizeof(val)) 3898 return -EINVAL; 3899 if (copy_from_user(&val, optval, sizeof(val))) 3900 return -EFAULT; 3901 3902 po->xmit = val ? packet_direct_xmit : dev_queue_xmit; 3903 return 0; 3904 } 3905 default: 3906 return -ENOPROTOOPT; 3907 } 3908 } 3909 3910 static int packet_getsockopt(struct socket *sock, int level, int optname, 3911 char __user *optval, int __user *optlen) 3912 { 3913 int len; 3914 int val, lv = sizeof(val); 3915 struct sock *sk = sock->sk; 3916 struct packet_sock *po = pkt_sk(sk); 3917 void *data = &val; 3918 union tpacket_stats_u st; 3919 struct tpacket_rollover_stats rstats; 3920 int drops; 3921 3922 if (level != SOL_PACKET) 3923 return -ENOPROTOOPT; 3924 3925 if (get_user(len, optlen)) 3926 return -EFAULT; 3927 3928 if (len < 0) 3929 return -EINVAL; 3930 3931 switch (optname) { 3932 case PACKET_STATISTICS: 3933 spin_lock_bh(&sk->sk_receive_queue.lock); 3934 memcpy(&st, &po->stats, sizeof(st)); 3935 memset(&po->stats, 0, sizeof(po->stats)); 3936 spin_unlock_bh(&sk->sk_receive_queue.lock); 3937 drops = atomic_xchg(&po->tp_drops, 0); 3938 3939 if (po->tp_version == TPACKET_V3) { 3940 lv = sizeof(struct tpacket_stats_v3); 3941 st.stats3.tp_drops = drops; 3942 st.stats3.tp_packets += drops; 3943 data = &st.stats3; 3944 } else { 3945 lv = sizeof(struct tpacket_stats); 3946 st.stats1.tp_drops = drops; 3947 st.stats1.tp_packets += drops; 3948 data = &st.stats1; 3949 } 3950 3951 break; 3952 case PACKET_AUXDATA: 3953 val = po->auxdata; 3954 break; 3955 case PACKET_ORIGDEV: 3956 val = po->origdev; 3957 break; 3958 case PACKET_VNET_HDR: 3959 val = po->has_vnet_hdr; 3960 break; 3961 case PACKET_VERSION: 3962 val = po->tp_version; 3963 break; 3964 case PACKET_HDRLEN: 3965 if (len > sizeof(int)) 3966 len = sizeof(int); 3967 if (len < sizeof(int)) 3968 return -EINVAL; 3969 if (copy_from_user(&val, optval, len)) 3970 return -EFAULT; 3971 switch (val) { 3972 case TPACKET_V1: 3973 val = sizeof(struct tpacket_hdr); 3974 break; 3975 case TPACKET_V2: 3976 val = sizeof(struct tpacket2_hdr); 3977 break; 3978 case TPACKET_V3: 3979 val = sizeof(struct tpacket3_hdr); 3980 break; 3981 default: 3982 return -EINVAL; 3983 } 3984 break; 3985 case PACKET_RESERVE: 3986 val = po->tp_reserve; 3987 break; 3988 case PACKET_LOSS: 3989 val = po->tp_loss; 3990 break; 3991 case PACKET_TIMESTAMP: 3992 val = po->tp_tstamp; 3993 break; 3994 case PACKET_FANOUT: 3995 val = (po->fanout ? 3996 ((u32)po->fanout->id | 3997 ((u32)po->fanout->type << 16) | 3998 ((u32)po->fanout->flags << 24)) : 3999 0); 4000 break; 4001 case PACKET_IGNORE_OUTGOING: 4002 val = po->prot_hook.ignore_outgoing; 4003 break; 4004 case PACKET_ROLLOVER_STATS: 4005 if (!po->rollover) 4006 return -EINVAL; 4007 rstats.tp_all = atomic_long_read(&po->rollover->num); 4008 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 4009 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 4010 data = &rstats; 4011 lv = sizeof(rstats); 4012 break; 4013 case PACKET_TX_HAS_OFF: 4014 val = po->tp_tx_has_off; 4015 break; 4016 case PACKET_QDISC_BYPASS: 4017 val = packet_use_direct_xmit(po); 4018 break; 4019 default: 4020 return -ENOPROTOOPT; 4021 } 4022 4023 if (len > lv) 4024 len = lv; 4025 if (put_user(len, optlen)) 4026 return -EFAULT; 4027 if (copy_to_user(optval, data, len)) 4028 return -EFAULT; 4029 return 0; 4030 } 4031 4032 4033 #ifdef CONFIG_COMPAT 4034 static int compat_packet_setsockopt(struct socket *sock, int level, int optname, 4035 char __user *optval, unsigned int optlen) 4036 { 4037 struct packet_sock *po = pkt_sk(sock->sk); 4038 4039 if (level != SOL_PACKET) 4040 return -ENOPROTOOPT; 4041 4042 if (optname == PACKET_FANOUT_DATA && 4043 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) { 4044 optval = (char __user *)get_compat_bpf_fprog(optval); 4045 if (!optval) 4046 return -EFAULT; 4047 optlen = sizeof(struct sock_fprog); 4048 } 4049 4050 return packet_setsockopt(sock, level, optname, optval, optlen); 4051 } 4052 #endif 4053 4054 static int packet_notifier(struct notifier_block *this, 4055 unsigned long msg, void *ptr) 4056 { 4057 struct sock *sk; 4058 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4059 struct net *net = dev_net(dev); 4060 4061 rcu_read_lock(); 4062 sk_for_each_rcu(sk, &net->packet.sklist) { 4063 struct packet_sock *po = pkt_sk(sk); 4064 4065 switch (msg) { 4066 case NETDEV_UNREGISTER: 4067 if (po->mclist) 4068 packet_dev_mclist_delete(dev, &po->mclist); 4069 /* fallthrough */ 4070 4071 case NETDEV_DOWN: 4072 if (dev->ifindex == po->ifindex) { 4073 spin_lock(&po->bind_lock); 4074 if (po->running) { 4075 __unregister_prot_hook(sk, false); 4076 sk->sk_err = ENETDOWN; 4077 if (!sock_flag(sk, SOCK_DEAD)) 4078 sk->sk_error_report(sk); 4079 } 4080 if (msg == NETDEV_UNREGISTER) { 4081 packet_cached_dev_reset(po); 4082 po->ifindex = -1; 4083 if (po->prot_hook.dev) 4084 dev_put(po->prot_hook.dev); 4085 po->prot_hook.dev = NULL; 4086 } 4087 spin_unlock(&po->bind_lock); 4088 } 4089 break; 4090 case NETDEV_UP: 4091 if (dev->ifindex == po->ifindex) { 4092 spin_lock(&po->bind_lock); 4093 if (po->num) 4094 register_prot_hook(sk); 4095 spin_unlock(&po->bind_lock); 4096 } 4097 break; 4098 } 4099 } 4100 rcu_read_unlock(); 4101 return NOTIFY_DONE; 4102 } 4103 4104 4105 static int packet_ioctl(struct socket *sock, unsigned int cmd, 4106 unsigned long arg) 4107 { 4108 struct sock *sk = sock->sk; 4109 4110 switch (cmd) { 4111 case SIOCOUTQ: 4112 { 4113 int amount = sk_wmem_alloc_get(sk); 4114 4115 return put_user(amount, (int __user *)arg); 4116 } 4117 case SIOCINQ: 4118 { 4119 struct sk_buff *skb; 4120 int amount = 0; 4121 4122 spin_lock_bh(&sk->sk_receive_queue.lock); 4123 skb = skb_peek(&sk->sk_receive_queue); 4124 if (skb) 4125 amount = skb->len; 4126 spin_unlock_bh(&sk->sk_receive_queue.lock); 4127 return put_user(amount, (int __user *)arg); 4128 } 4129 #ifdef CONFIG_INET 4130 case SIOCADDRT: 4131 case SIOCDELRT: 4132 case SIOCDARP: 4133 case SIOCGARP: 4134 case SIOCSARP: 4135 case SIOCGIFADDR: 4136 case SIOCSIFADDR: 4137 case SIOCGIFBRDADDR: 4138 case SIOCSIFBRDADDR: 4139 case SIOCGIFNETMASK: 4140 case SIOCSIFNETMASK: 4141 case SIOCGIFDSTADDR: 4142 case SIOCSIFDSTADDR: 4143 case SIOCSIFFLAGS: 4144 return inet_dgram_ops.ioctl(sock, cmd, arg); 4145 #endif 4146 4147 default: 4148 return -ENOIOCTLCMD; 4149 } 4150 return 0; 4151 } 4152 4153 static __poll_t packet_poll(struct file *file, struct socket *sock, 4154 poll_table *wait) 4155 { 4156 struct sock *sk = sock->sk; 4157 struct packet_sock *po = pkt_sk(sk); 4158 __poll_t mask = datagram_poll(file, sock, wait); 4159 4160 spin_lock_bh(&sk->sk_receive_queue.lock); 4161 if (po->rx_ring.pg_vec) { 4162 if (!packet_previous_rx_frame(po, &po->rx_ring, 4163 TP_STATUS_KERNEL)) 4164 mask |= EPOLLIN | EPOLLRDNORM; 4165 } 4166 packet_rcv_try_clear_pressure(po); 4167 spin_unlock_bh(&sk->sk_receive_queue.lock); 4168 spin_lock_bh(&sk->sk_write_queue.lock); 4169 if (po->tx_ring.pg_vec) { 4170 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4171 mask |= EPOLLOUT | EPOLLWRNORM; 4172 } 4173 spin_unlock_bh(&sk->sk_write_queue.lock); 4174 return mask; 4175 } 4176 4177 4178 /* Dirty? Well, I still did not learn better way to account 4179 * for user mmaps. 4180 */ 4181 4182 static void packet_mm_open(struct vm_area_struct *vma) 4183 { 4184 struct file *file = vma->vm_file; 4185 struct socket *sock = file->private_data; 4186 struct sock *sk = sock->sk; 4187 4188 if (sk) 4189 atomic_inc(&pkt_sk(sk)->mapped); 4190 } 4191 4192 static void packet_mm_close(struct vm_area_struct *vma) 4193 { 4194 struct file *file = vma->vm_file; 4195 struct socket *sock = file->private_data; 4196 struct sock *sk = sock->sk; 4197 4198 if (sk) 4199 atomic_dec(&pkt_sk(sk)->mapped); 4200 } 4201 4202 static const struct vm_operations_struct packet_mmap_ops = { 4203 .open = packet_mm_open, 4204 .close = packet_mm_close, 4205 }; 4206 4207 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4208 unsigned int len) 4209 { 4210 int i; 4211 4212 for (i = 0; i < len; i++) { 4213 if (likely(pg_vec[i].buffer)) { 4214 if (is_vmalloc_addr(pg_vec[i].buffer)) 4215 vfree(pg_vec[i].buffer); 4216 else 4217 free_pages((unsigned long)pg_vec[i].buffer, 4218 order); 4219 pg_vec[i].buffer = NULL; 4220 } 4221 } 4222 kfree(pg_vec); 4223 } 4224 4225 static char *alloc_one_pg_vec_page(unsigned long order) 4226 { 4227 char *buffer; 4228 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4229 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4230 4231 buffer = (char *) __get_free_pages(gfp_flags, order); 4232 if (buffer) 4233 return buffer; 4234 4235 /* __get_free_pages failed, fall back to vmalloc */ 4236 buffer = vzalloc(array_size((1 << order), PAGE_SIZE)); 4237 if (buffer) 4238 return buffer; 4239 4240 /* vmalloc failed, lets dig into swap here */ 4241 gfp_flags &= ~__GFP_NORETRY; 4242 buffer = (char *) __get_free_pages(gfp_flags, order); 4243 if (buffer) 4244 return buffer; 4245 4246 /* complete and utter failure */ 4247 return NULL; 4248 } 4249 4250 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4251 { 4252 unsigned int block_nr = req->tp_block_nr; 4253 struct pgv *pg_vec; 4254 int i; 4255 4256 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN); 4257 if (unlikely(!pg_vec)) 4258 goto out; 4259 4260 for (i = 0; i < block_nr; i++) { 4261 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4262 if (unlikely(!pg_vec[i].buffer)) 4263 goto out_free_pgvec; 4264 } 4265 4266 out: 4267 return pg_vec; 4268 4269 out_free_pgvec: 4270 free_pg_vec(pg_vec, order, block_nr); 4271 pg_vec = NULL; 4272 goto out; 4273 } 4274 4275 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4276 int closing, int tx_ring) 4277 { 4278 struct pgv *pg_vec = NULL; 4279 struct packet_sock *po = pkt_sk(sk); 4280 int was_running, order = 0; 4281 struct packet_ring_buffer *rb; 4282 struct sk_buff_head *rb_queue; 4283 __be16 num; 4284 int err = -EINVAL; 4285 /* Added to avoid minimal code churn */ 4286 struct tpacket_req *req = &req_u->req; 4287 4288 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4289 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4290 4291 err = -EBUSY; 4292 if (!closing) { 4293 if (atomic_read(&po->mapped)) 4294 goto out; 4295 if (packet_read_pending(rb)) 4296 goto out; 4297 } 4298 4299 if (req->tp_block_nr) { 4300 unsigned int min_frame_size; 4301 4302 /* Sanity tests and some calculations */ 4303 err = -EBUSY; 4304 if (unlikely(rb->pg_vec)) 4305 goto out; 4306 4307 switch (po->tp_version) { 4308 case TPACKET_V1: 4309 po->tp_hdrlen = TPACKET_HDRLEN; 4310 break; 4311 case TPACKET_V2: 4312 po->tp_hdrlen = TPACKET2_HDRLEN; 4313 break; 4314 case TPACKET_V3: 4315 po->tp_hdrlen = TPACKET3_HDRLEN; 4316 break; 4317 } 4318 4319 err = -EINVAL; 4320 if (unlikely((int)req->tp_block_size <= 0)) 4321 goto out; 4322 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4323 goto out; 4324 min_frame_size = po->tp_hdrlen + po->tp_reserve; 4325 if (po->tp_version >= TPACKET_V3 && 4326 req->tp_block_size < 4327 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) 4328 goto out; 4329 if (unlikely(req->tp_frame_size < min_frame_size)) 4330 goto out; 4331 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4332 goto out; 4333 4334 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4335 if (unlikely(rb->frames_per_block == 0)) 4336 goto out; 4337 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) 4338 goto out; 4339 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4340 req->tp_frame_nr)) 4341 goto out; 4342 4343 err = -ENOMEM; 4344 order = get_order(req->tp_block_size); 4345 pg_vec = alloc_pg_vec(req, order); 4346 if (unlikely(!pg_vec)) 4347 goto out; 4348 switch (po->tp_version) { 4349 case TPACKET_V3: 4350 /* Block transmit is not supported yet */ 4351 if (!tx_ring) { 4352 init_prb_bdqc(po, rb, pg_vec, req_u); 4353 } else { 4354 struct tpacket_req3 *req3 = &req_u->req3; 4355 4356 if (req3->tp_retire_blk_tov || 4357 req3->tp_sizeof_priv || 4358 req3->tp_feature_req_word) { 4359 err = -EINVAL; 4360 goto out_free_pg_vec; 4361 } 4362 } 4363 break; 4364 default: 4365 break; 4366 } 4367 } 4368 /* Done */ 4369 else { 4370 err = -EINVAL; 4371 if (unlikely(req->tp_frame_nr)) 4372 goto out; 4373 } 4374 4375 4376 /* Detach socket from network */ 4377 spin_lock(&po->bind_lock); 4378 was_running = po->running; 4379 num = po->num; 4380 if (was_running) { 4381 po->num = 0; 4382 __unregister_prot_hook(sk, false); 4383 } 4384 spin_unlock(&po->bind_lock); 4385 4386 synchronize_net(); 4387 4388 err = -EBUSY; 4389 mutex_lock(&po->pg_vec_lock); 4390 if (closing || atomic_read(&po->mapped) == 0) { 4391 err = 0; 4392 spin_lock_bh(&rb_queue->lock); 4393 swap(rb->pg_vec, pg_vec); 4394 rb->frame_max = (req->tp_frame_nr - 1); 4395 rb->head = 0; 4396 rb->frame_size = req->tp_frame_size; 4397 spin_unlock_bh(&rb_queue->lock); 4398 4399 swap(rb->pg_vec_order, order); 4400 swap(rb->pg_vec_len, req->tp_block_nr); 4401 4402 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4403 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4404 tpacket_rcv : packet_rcv; 4405 skb_queue_purge(rb_queue); 4406 if (atomic_read(&po->mapped)) 4407 pr_err("packet_mmap: vma is busy: %d\n", 4408 atomic_read(&po->mapped)); 4409 } 4410 mutex_unlock(&po->pg_vec_lock); 4411 4412 spin_lock(&po->bind_lock); 4413 if (was_running) { 4414 po->num = num; 4415 register_prot_hook(sk); 4416 } 4417 spin_unlock(&po->bind_lock); 4418 if (pg_vec && (po->tp_version > TPACKET_V2)) { 4419 /* Because we don't support block-based V3 on tx-ring */ 4420 if (!tx_ring) 4421 prb_shutdown_retire_blk_timer(po, rb_queue); 4422 } 4423 4424 out_free_pg_vec: 4425 if (pg_vec) 4426 free_pg_vec(pg_vec, order, req->tp_block_nr); 4427 out: 4428 return err; 4429 } 4430 4431 static int packet_mmap(struct file *file, struct socket *sock, 4432 struct vm_area_struct *vma) 4433 { 4434 struct sock *sk = sock->sk; 4435 struct packet_sock *po = pkt_sk(sk); 4436 unsigned long size, expected_size; 4437 struct packet_ring_buffer *rb; 4438 unsigned long start; 4439 int err = -EINVAL; 4440 int i; 4441 4442 if (vma->vm_pgoff) 4443 return -EINVAL; 4444 4445 mutex_lock(&po->pg_vec_lock); 4446 4447 expected_size = 0; 4448 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4449 if (rb->pg_vec) { 4450 expected_size += rb->pg_vec_len 4451 * rb->pg_vec_pages 4452 * PAGE_SIZE; 4453 } 4454 } 4455 4456 if (expected_size == 0) 4457 goto out; 4458 4459 size = vma->vm_end - vma->vm_start; 4460 if (size != expected_size) 4461 goto out; 4462 4463 start = vma->vm_start; 4464 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4465 if (rb->pg_vec == NULL) 4466 continue; 4467 4468 for (i = 0; i < rb->pg_vec_len; i++) { 4469 struct page *page; 4470 void *kaddr = rb->pg_vec[i].buffer; 4471 int pg_num; 4472 4473 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4474 page = pgv_to_page(kaddr); 4475 err = vm_insert_page(vma, start, page); 4476 if (unlikely(err)) 4477 goto out; 4478 start += PAGE_SIZE; 4479 kaddr += PAGE_SIZE; 4480 } 4481 } 4482 } 4483 4484 atomic_inc(&po->mapped); 4485 vma->vm_ops = &packet_mmap_ops; 4486 err = 0; 4487 4488 out: 4489 mutex_unlock(&po->pg_vec_lock); 4490 return err; 4491 } 4492 4493 static const struct proto_ops packet_ops_spkt = { 4494 .family = PF_PACKET, 4495 .owner = THIS_MODULE, 4496 .release = packet_release, 4497 .bind = packet_bind_spkt, 4498 .connect = sock_no_connect, 4499 .socketpair = sock_no_socketpair, 4500 .accept = sock_no_accept, 4501 .getname = packet_getname_spkt, 4502 .poll = datagram_poll, 4503 .ioctl = packet_ioctl, 4504 .gettstamp = sock_gettstamp, 4505 .listen = sock_no_listen, 4506 .shutdown = sock_no_shutdown, 4507 .setsockopt = sock_no_setsockopt, 4508 .getsockopt = sock_no_getsockopt, 4509 .sendmsg = packet_sendmsg_spkt, 4510 .recvmsg = packet_recvmsg, 4511 .mmap = sock_no_mmap, 4512 .sendpage = sock_no_sendpage, 4513 }; 4514 4515 static const struct proto_ops packet_ops = { 4516 .family = PF_PACKET, 4517 .owner = THIS_MODULE, 4518 .release = packet_release, 4519 .bind = packet_bind, 4520 .connect = sock_no_connect, 4521 .socketpair = sock_no_socketpair, 4522 .accept = sock_no_accept, 4523 .getname = packet_getname, 4524 .poll = packet_poll, 4525 .ioctl = packet_ioctl, 4526 .gettstamp = sock_gettstamp, 4527 .listen = sock_no_listen, 4528 .shutdown = sock_no_shutdown, 4529 .setsockopt = packet_setsockopt, 4530 .getsockopt = packet_getsockopt, 4531 #ifdef CONFIG_COMPAT 4532 .compat_setsockopt = compat_packet_setsockopt, 4533 #endif 4534 .sendmsg = packet_sendmsg, 4535 .recvmsg = packet_recvmsg, 4536 .mmap = packet_mmap, 4537 .sendpage = sock_no_sendpage, 4538 }; 4539 4540 static const struct net_proto_family packet_family_ops = { 4541 .family = PF_PACKET, 4542 .create = packet_create, 4543 .owner = THIS_MODULE, 4544 }; 4545 4546 static struct notifier_block packet_netdev_notifier = { 4547 .notifier_call = packet_notifier, 4548 }; 4549 4550 #ifdef CONFIG_PROC_FS 4551 4552 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4553 __acquires(RCU) 4554 { 4555 struct net *net = seq_file_net(seq); 4556 4557 rcu_read_lock(); 4558 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4559 } 4560 4561 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4562 { 4563 struct net *net = seq_file_net(seq); 4564 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4565 } 4566 4567 static void packet_seq_stop(struct seq_file *seq, void *v) 4568 __releases(RCU) 4569 { 4570 rcu_read_unlock(); 4571 } 4572 4573 static int packet_seq_show(struct seq_file *seq, void *v) 4574 { 4575 if (v == SEQ_START_TOKEN) 4576 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4577 else { 4578 struct sock *s = sk_entry(v); 4579 const struct packet_sock *po = pkt_sk(s); 4580 4581 seq_printf(seq, 4582 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4583 s, 4584 refcount_read(&s->sk_refcnt), 4585 s->sk_type, 4586 ntohs(po->num), 4587 po->ifindex, 4588 po->running, 4589 atomic_read(&s->sk_rmem_alloc), 4590 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4591 sock_i_ino(s)); 4592 } 4593 4594 return 0; 4595 } 4596 4597 static const struct seq_operations packet_seq_ops = { 4598 .start = packet_seq_start, 4599 .next = packet_seq_next, 4600 .stop = packet_seq_stop, 4601 .show = packet_seq_show, 4602 }; 4603 #endif 4604 4605 static int __net_init packet_net_init(struct net *net) 4606 { 4607 mutex_init(&net->packet.sklist_lock); 4608 INIT_HLIST_HEAD(&net->packet.sklist); 4609 4610 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, 4611 sizeof(struct seq_net_private))) 4612 return -ENOMEM; 4613 4614 return 0; 4615 } 4616 4617 static void __net_exit packet_net_exit(struct net *net) 4618 { 4619 remove_proc_entry("packet", net->proc_net); 4620 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); 4621 } 4622 4623 static struct pernet_operations packet_net_ops = { 4624 .init = packet_net_init, 4625 .exit = packet_net_exit, 4626 }; 4627 4628 4629 static void __exit packet_exit(void) 4630 { 4631 unregister_netdevice_notifier(&packet_netdev_notifier); 4632 unregister_pernet_subsys(&packet_net_ops); 4633 sock_unregister(PF_PACKET); 4634 proto_unregister(&packet_proto); 4635 } 4636 4637 static int __init packet_init(void) 4638 { 4639 int rc; 4640 4641 rc = proto_register(&packet_proto, 0); 4642 if (rc) 4643 goto out; 4644 rc = sock_register(&packet_family_ops); 4645 if (rc) 4646 goto out_proto; 4647 rc = register_pernet_subsys(&packet_net_ops); 4648 if (rc) 4649 goto out_sock; 4650 rc = register_netdevice_notifier(&packet_netdev_notifier); 4651 if (rc) 4652 goto out_pernet; 4653 4654 return 0; 4655 4656 out_pernet: 4657 unregister_pernet_subsys(&packet_net_ops); 4658 out_sock: 4659 sock_unregister(PF_PACKET); 4660 out_proto: 4661 proto_unregister(&packet_proto); 4662 out: 4663 return rc; 4664 } 4665 4666 module_init(packet_init); 4667 module_exit(packet_exit); 4668 MODULE_LICENSE("GPL"); 4669 MODULE_ALIAS_NETPROTO(PF_PACKET); 4670