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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 50 51 #include <linux/ethtool.h> 52 #include <linux/uio.h> 53 #include <linux/filter.h> 54 #include <linux/types.h> 55 #include <linux/mm.h> 56 #include <linux/capability.h> 57 #include <linux/fcntl.h> 58 #include <linux/socket.h> 59 #include <linux/in.h> 60 #include <linux/inet.h> 61 #include <linux/netdevice.h> 62 #include <linux/if_packet.h> 63 #include <linux/wireless.h> 64 #include <linux/kernel.h> 65 #include <linux/kmod.h> 66 #include <linux/slab.h> 67 #include <linux/vmalloc.h> 68 #include <net/net_namespace.h> 69 #include <net/ip.h> 70 #include <net/protocol.h> 71 #include <linux/skbuff.h> 72 #include <net/sock.h> 73 #include <linux/errno.h> 74 #include <linux/timer.h> 75 #include <linux/uaccess.h> 76 #include <asm/ioctls.h> 77 #include <asm/page.h> 78 #include <asm/cacheflush.h> 79 #include <asm/io.h> 80 #include <linux/proc_fs.h> 81 #include <linux/seq_file.h> 82 #include <linux/poll.h> 83 #include <linux/module.h> 84 #include <linux/init.h> 85 #include <linux/mutex.h> 86 #include <linux/if_vlan.h> 87 #include <linux/virtio_net.h> 88 #include <linux/errqueue.h> 89 #include <linux/net_tstamp.h> 90 #include <linux/percpu.h> 91 #include <linux/workqueue.h> 92 #ifdef CONFIG_INET 93 #include <net/inet_common.h> 94 #endif 95 #include <linux/bpf.h> 96 #include <net/compat.h> 97 #include <linux/netfilter_netdev.h> 98 99 #include "internal.h" 100 101 /* 102 Assumptions: 103 - If the device has no dev->header_ops->create, there is no LL header 104 visible above the device. In this case, its hard_header_len should be 0. 105 The device may prepend its own header internally. In this case, its 106 needed_headroom should be set to the space needed for it to add its 107 internal header. 108 For example, a WiFi driver pretending to be an Ethernet driver should 109 set its hard_header_len to be the Ethernet header length, and set its 110 needed_headroom to be (the real WiFi header length - the fake Ethernet 111 header length). 112 - packet socket receives packets with pulled ll header, 113 so that SOCK_RAW should push it back. 114 115 On receive: 116 ----------- 117 118 Incoming, dev_has_header(dev) == true 119 mac_header -> ll header 120 data -> data 121 122 Outgoing, dev_has_header(dev) == true 123 mac_header -> ll header 124 data -> ll header 125 126 Incoming, dev_has_header(dev) == false 127 mac_header -> data 128 However drivers often make it point to the ll header. 129 This is incorrect because the ll header should be invisible to us. 130 data -> data 131 132 Outgoing, dev_has_header(dev) == false 133 mac_header -> data. ll header is invisible to us. 134 data -> data 135 136 Resume 137 If dev_has_header(dev) == false we are unable to restore the ll header, 138 because it is invisible to us. 139 140 141 On transmit: 142 ------------ 143 144 dev_has_header(dev) == true 145 mac_header -> ll header 146 data -> ll header 147 148 dev_has_header(dev) == false (ll header is invisible to us) 149 mac_header -> data 150 data -> data 151 152 We should set network_header on output to the correct position, 153 packet classifier depends on it. 154 */ 155 156 /* Private packet socket structures. */ 157 158 /* identical to struct packet_mreq except it has 159 * a longer address field. 160 */ 161 struct packet_mreq_max { 162 int mr_ifindex; 163 unsigned short mr_type; 164 unsigned short mr_alen; 165 unsigned char mr_address[MAX_ADDR_LEN]; 166 }; 167 168 union tpacket_uhdr { 169 struct tpacket_hdr *h1; 170 struct tpacket2_hdr *h2; 171 struct tpacket3_hdr *h3; 172 void *raw; 173 }; 174 175 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 176 int closing, int tx_ring); 177 178 #define V3_ALIGNMENT (8) 179 180 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) 181 182 #define BLK_PLUS_PRIV(sz_of_priv) \ 183 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) 184 185 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) 186 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) 187 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) 188 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) 189 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) 190 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) 191 192 struct packet_sock; 193 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 194 struct packet_type *pt, struct net_device *orig_dev); 195 196 static void *packet_previous_frame(struct packet_sock *po, 197 struct packet_ring_buffer *rb, 198 int status); 199 static void packet_increment_head(struct packet_ring_buffer *buff); 200 static int prb_curr_blk_in_use(struct tpacket_block_desc *); 201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 202 struct packet_sock *); 203 static void prb_retire_current_block(struct tpacket_kbdq_core *, 204 struct packet_sock *, unsigned int status); 205 static int prb_queue_frozen(struct tpacket_kbdq_core *); 206 static void prb_open_block(struct tpacket_kbdq_core *, 207 struct tpacket_block_desc *); 208 static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *); 209 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 210 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 211 struct tpacket3_hdr *); 212 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 213 struct tpacket3_hdr *); 214 static void packet_flush_mclist(struct sock *sk); 215 static u16 packet_pick_tx_queue(struct sk_buff *skb); 216 217 struct packet_skb_cb { 218 union { 219 struct sockaddr_pkt pkt; 220 union { 221 /* Trick: alias skb original length with 222 * ll.sll_family and ll.protocol in order 223 * to save room. 224 */ 225 unsigned int origlen; 226 struct sockaddr_ll ll; 227 }; 228 } sa; 229 }; 230 231 #define vio_le() virtio_legacy_is_little_endian() 232 233 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 234 235 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 236 #define GET_PBLOCK_DESC(x, bid) \ 237 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 238 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 239 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 240 #define GET_NEXT_PRB_BLK_NUM(x) \ 241 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 242 ((x)->kactive_blk_num+1) : 0) 243 244 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 245 static void __fanout_link(struct sock *sk, struct packet_sock *po); 246 247 #ifdef CONFIG_NETFILTER_EGRESS 248 static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb) 249 { 250 struct sk_buff *next, *head = NULL, *tail; 251 int rc; 252 253 rcu_read_lock(); 254 for (; skb != NULL; skb = next) { 255 next = skb->next; 256 skb_mark_not_on_list(skb); 257 258 if (!nf_hook_egress(skb, &rc, skb->dev)) 259 continue; 260 261 if (!head) 262 head = skb; 263 else 264 tail->next = skb; 265 266 tail = skb; 267 } 268 rcu_read_unlock(); 269 270 return head; 271 } 272 #endif 273 274 static int packet_xmit(const struct packet_sock *po, struct sk_buff *skb) 275 { 276 if (!packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS)) 277 return dev_queue_xmit(skb); 278 279 #ifdef CONFIG_NETFILTER_EGRESS 280 if (nf_hook_egress_active()) { 281 skb = nf_hook_direct_egress(skb); 282 if (!skb) 283 return NET_XMIT_DROP; 284 } 285 #endif 286 return dev_direct_xmit(skb, packet_pick_tx_queue(skb)); 287 } 288 289 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 290 { 291 struct net_device *dev; 292 293 rcu_read_lock(); 294 dev = rcu_dereference(po->cached_dev); 295 dev_hold(dev); 296 rcu_read_unlock(); 297 298 return dev; 299 } 300 301 static void packet_cached_dev_assign(struct packet_sock *po, 302 struct net_device *dev) 303 { 304 rcu_assign_pointer(po->cached_dev, dev); 305 } 306 307 static void packet_cached_dev_reset(struct packet_sock *po) 308 { 309 RCU_INIT_POINTER(po->cached_dev, NULL); 310 } 311 312 static u16 packet_pick_tx_queue(struct sk_buff *skb) 313 { 314 struct net_device *dev = skb->dev; 315 const struct net_device_ops *ops = dev->netdev_ops; 316 int cpu = raw_smp_processor_id(); 317 u16 queue_index; 318 319 #ifdef CONFIG_XPS 320 skb->sender_cpu = cpu + 1; 321 #endif 322 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues); 323 if (ops->ndo_select_queue) { 324 queue_index = ops->ndo_select_queue(dev, skb, NULL); 325 queue_index = netdev_cap_txqueue(dev, queue_index); 326 } else { 327 queue_index = netdev_pick_tx(dev, skb, NULL); 328 } 329 330 return queue_index; 331 } 332 333 /* __register_prot_hook must be invoked through register_prot_hook 334 * or from a context in which asynchronous accesses to the packet 335 * socket is not possible (packet_create()). 336 */ 337 static void __register_prot_hook(struct sock *sk) 338 { 339 struct packet_sock *po = pkt_sk(sk); 340 341 if (!packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 342 if (po->fanout) 343 __fanout_link(sk, po); 344 else 345 dev_add_pack(&po->prot_hook); 346 347 sock_hold(sk); 348 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 1); 349 } 350 } 351 352 static void register_prot_hook(struct sock *sk) 353 { 354 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); 355 __register_prot_hook(sk); 356 } 357 358 /* If the sync parameter is true, we will temporarily drop 359 * the po->bind_lock and do a synchronize_net to make sure no 360 * asynchronous packet processing paths still refer to the elements 361 * of po->prot_hook. If the sync parameter is false, it is the 362 * callers responsibility to take care of this. 363 */ 364 static void __unregister_prot_hook(struct sock *sk, bool sync) 365 { 366 struct packet_sock *po = pkt_sk(sk); 367 368 lockdep_assert_held_once(&po->bind_lock); 369 370 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 0); 371 372 if (po->fanout) 373 __fanout_unlink(sk, po); 374 else 375 __dev_remove_pack(&po->prot_hook); 376 377 __sock_put(sk); 378 379 if (sync) { 380 spin_unlock(&po->bind_lock); 381 synchronize_net(); 382 spin_lock(&po->bind_lock); 383 } 384 } 385 386 static void unregister_prot_hook(struct sock *sk, bool sync) 387 { 388 struct packet_sock *po = pkt_sk(sk); 389 390 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) 391 __unregister_prot_hook(sk, sync); 392 } 393 394 static inline struct page * __pure pgv_to_page(void *addr) 395 { 396 if (is_vmalloc_addr(addr)) 397 return vmalloc_to_page(addr); 398 return virt_to_page(addr); 399 } 400 401 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 402 { 403 union tpacket_uhdr h; 404 405 /* WRITE_ONCE() are paired with READ_ONCE() in __packet_get_status */ 406 407 h.raw = frame; 408 switch (po->tp_version) { 409 case TPACKET_V1: 410 WRITE_ONCE(h.h1->tp_status, status); 411 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 412 break; 413 case TPACKET_V2: 414 WRITE_ONCE(h.h2->tp_status, status); 415 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 416 break; 417 case TPACKET_V3: 418 WRITE_ONCE(h.h3->tp_status, status); 419 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 420 break; 421 default: 422 WARN(1, "TPACKET version not supported.\n"); 423 BUG(); 424 } 425 426 smp_wmb(); 427 } 428 429 static int __packet_get_status(const struct packet_sock *po, void *frame) 430 { 431 union tpacket_uhdr h; 432 433 smp_rmb(); 434 435 /* READ_ONCE() are paired with WRITE_ONCE() in __packet_set_status */ 436 437 h.raw = frame; 438 switch (po->tp_version) { 439 case TPACKET_V1: 440 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 441 return READ_ONCE(h.h1->tp_status); 442 case TPACKET_V2: 443 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 444 return READ_ONCE(h.h2->tp_status); 445 case TPACKET_V3: 446 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 447 return READ_ONCE(h.h3->tp_status); 448 default: 449 WARN(1, "TPACKET version not supported.\n"); 450 BUG(); 451 return 0; 452 } 453 } 454 455 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts, 456 unsigned int flags) 457 { 458 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 459 460 if (shhwtstamps && 461 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 462 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts)) 463 return TP_STATUS_TS_RAW_HARDWARE; 464 465 if ((flags & SOF_TIMESTAMPING_SOFTWARE) && 466 ktime_to_timespec64_cond(skb_tstamp(skb), ts)) 467 return TP_STATUS_TS_SOFTWARE; 468 469 return 0; 470 } 471 472 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 473 struct sk_buff *skb) 474 { 475 union tpacket_uhdr h; 476 struct timespec64 ts; 477 __u32 ts_status; 478 479 if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp)))) 480 return 0; 481 482 h.raw = frame; 483 /* 484 * versions 1 through 3 overflow the timestamps in y2106, since they 485 * all store the seconds in a 32-bit unsigned integer. 486 * If we create a version 4, that should have a 64-bit timestamp, 487 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit 488 * nanoseconds. 489 */ 490 switch (po->tp_version) { 491 case TPACKET_V1: 492 h.h1->tp_sec = ts.tv_sec; 493 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 494 break; 495 case TPACKET_V2: 496 h.h2->tp_sec = ts.tv_sec; 497 h.h2->tp_nsec = ts.tv_nsec; 498 break; 499 case TPACKET_V3: 500 h.h3->tp_sec = ts.tv_sec; 501 h.h3->tp_nsec = ts.tv_nsec; 502 break; 503 default: 504 WARN(1, "TPACKET version not supported.\n"); 505 BUG(); 506 } 507 508 /* one flush is safe, as both fields always lie on the same cacheline */ 509 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 510 smp_wmb(); 511 512 return ts_status; 513 } 514 515 static void *packet_lookup_frame(const struct packet_sock *po, 516 const struct packet_ring_buffer *rb, 517 unsigned int position, 518 int status) 519 { 520 unsigned int pg_vec_pos, frame_offset; 521 union tpacket_uhdr h; 522 523 pg_vec_pos = position / rb->frames_per_block; 524 frame_offset = position % rb->frames_per_block; 525 526 h.raw = rb->pg_vec[pg_vec_pos].buffer + 527 (frame_offset * rb->frame_size); 528 529 if (status != __packet_get_status(po, h.raw)) 530 return NULL; 531 532 return h.raw; 533 } 534 535 static void *packet_current_frame(struct packet_sock *po, 536 struct packet_ring_buffer *rb, 537 int status) 538 { 539 return packet_lookup_frame(po, rb, rb->head, status); 540 } 541 542 static u16 vlan_get_tci(const struct sk_buff *skb, struct net_device *dev) 543 { 544 struct vlan_hdr vhdr, *vh; 545 unsigned int header_len; 546 547 if (!dev) 548 return 0; 549 550 /* In the SOCK_DGRAM scenario, skb data starts at the network 551 * protocol, which is after the VLAN headers. The outer VLAN 552 * header is at the hard_header_len offset in non-variable 553 * length link layer headers. If it's a VLAN device, the 554 * min_header_len should be used to exclude the VLAN header 555 * size. 556 */ 557 if (dev->min_header_len == dev->hard_header_len) 558 header_len = dev->hard_header_len; 559 else if (is_vlan_dev(dev)) 560 header_len = dev->min_header_len; 561 else 562 return 0; 563 564 vh = skb_header_pointer(skb, skb_mac_offset(skb) + header_len, 565 sizeof(vhdr), &vhdr); 566 if (unlikely(!vh)) 567 return 0; 568 569 return ntohs(vh->h_vlan_TCI); 570 } 571 572 static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb) 573 { 574 __be16 proto = skb->protocol; 575 576 if (unlikely(eth_type_vlan(proto))) 577 proto = vlan_get_protocol_offset_inline(skb, proto, 578 skb_mac_offset(skb), 579 NULL); 580 581 return proto; 582 } 583 584 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 585 struct sk_buff_head *rb_queue) 586 { 587 struct tpacket_kbdq_core *pkc; 588 589 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 590 hrtimer_cancel(&pkc->retire_blk_timer); 591 } 592 593 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 594 int blk_size_in_bytes) 595 { 596 struct net_device *dev; 597 unsigned int mbits, div; 598 struct ethtool_link_ksettings ecmd; 599 int err; 600 601 rtnl_lock(); 602 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 603 if (unlikely(!dev)) { 604 rtnl_unlock(); 605 return DEFAULT_PRB_RETIRE_TOV; 606 } 607 err = __ethtool_get_link_ksettings(dev, &ecmd); 608 rtnl_unlock(); 609 if (err) 610 return DEFAULT_PRB_RETIRE_TOV; 611 612 /* If the link speed is so slow you don't really 613 * need to worry about perf anyways 614 */ 615 if (ecmd.base.speed < SPEED_1000 || 616 ecmd.base.speed == SPEED_UNKNOWN) 617 return DEFAULT_PRB_RETIRE_TOV; 618 619 div = ecmd.base.speed / 1000; 620 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 621 622 if (div) 623 mbits /= div; 624 625 if (div) 626 return mbits + 1; 627 return mbits; 628 } 629 630 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 631 union tpacket_req_u *req_u) 632 { 633 p1->feature_req_word = req_u->req3.tp_feature_req_word; 634 } 635 636 static void init_prb_bdqc(struct packet_sock *po, 637 struct packet_ring_buffer *rb, 638 struct pgv *pg_vec, 639 union tpacket_req_u *req_u) 640 { 641 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 642 struct tpacket_block_desc *pbd; 643 644 memset(p1, 0x0, sizeof(*p1)); 645 646 p1->knxt_seq_num = 1; 647 p1->pkbdq = pg_vec; 648 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 649 p1->pkblk_start = pg_vec[0].buffer; 650 p1->kblk_size = req_u->req3.tp_block_size; 651 p1->knum_blocks = req_u->req3.tp_block_nr; 652 p1->hdrlen = po->tp_hdrlen; 653 p1->version = po->tp_version; 654 po->stats.stats3.tp_freeze_q_cnt = 0; 655 if (req_u->req3.tp_retire_blk_tov) 656 p1->interval_ktime = ms_to_ktime(req_u->req3.tp_retire_blk_tov); 657 else 658 p1->interval_ktime = ms_to_ktime(prb_calc_retire_blk_tmo(po, 659 req_u->req3.tp_block_size)); 660 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 661 rwlock_init(&p1->blk_fill_in_prog_lock); 662 663 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 664 prb_init_ft_ops(p1, req_u); 665 hrtimer_setup(&p1->retire_blk_timer, prb_retire_rx_blk_timer_expired, 666 CLOCK_MONOTONIC, HRTIMER_MODE_REL_SOFT); 667 hrtimer_start(&p1->retire_blk_timer, p1->interval_ktime, 668 HRTIMER_MODE_REL_SOFT); 669 prb_open_block(p1, pbd); 670 } 671 672 /* 673 * With a 1MB block-size, on a 1Gbps line, it will take 674 * i) ~8 ms to fill a block + ii) memcpy etc. 675 * In this cut we are not accounting for the memcpy time. 676 * 677 * Since the tmo granularity is in msecs, it is not too expensive 678 * to refresh the timer, lets say every '8' msecs. 679 * Either the user can set the 'tmo' or we can derive it based on 680 * a) line-speed and b) block-size. 681 * prb_calc_retire_blk_tmo() calculates the tmo. 682 */ 683 static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *t) 684 { 685 struct packet_sock *po = 686 timer_container_of(po, t, rx_ring.prb_bdqc.retire_blk_timer); 687 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 688 unsigned int frozen; 689 struct tpacket_block_desc *pbd; 690 691 spin_lock(&po->sk.sk_receive_queue.lock); 692 693 frozen = prb_queue_frozen(pkc); 694 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 695 696 /* We only need to plug the race when the block is partially filled. 697 * tpacket_rcv: 698 * lock(); increment BLOCK_NUM_PKTS; unlock() 699 * copy_bits() is in progress ... 700 * timer fires on other cpu: 701 * we can't retire the current block because copy_bits 702 * is in progress. 703 * 704 */ 705 if (BLOCK_NUM_PKTS(pbd)) { 706 /* Waiting for skb_copy_bits to finish... */ 707 write_lock(&pkc->blk_fill_in_prog_lock); 708 write_unlock(&pkc->blk_fill_in_prog_lock); 709 } 710 711 if (!frozen) { 712 if (BLOCK_NUM_PKTS(pbd)) { 713 /* Not an empty block. Need retire the block. */ 714 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 715 prb_dispatch_next_block(pkc, po); 716 } 717 } else { 718 /* Case 1. Queue was frozen because user-space was 719 * lagging behind. 720 */ 721 if (!prb_curr_blk_in_use(pbd)) { 722 /* Case 2. queue was frozen,user-space caught up, 723 * now the link went idle && the timer fired. 724 * We don't have a block to close.So we open this 725 * block and restart the timer. 726 * opening a block thaws the queue,restarts timer 727 * Thawing/timer-refresh is a side effect. 728 */ 729 prb_open_block(pkc, pbd); 730 } 731 } 732 733 hrtimer_forward_now(&pkc->retire_blk_timer, pkc->interval_ktime); 734 spin_unlock(&po->sk.sk_receive_queue.lock); 735 return HRTIMER_RESTART; 736 } 737 738 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 739 struct tpacket_block_desc *pbd1, __u32 status) 740 { 741 /* Flush everything minus the block header */ 742 743 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 744 u8 *start, *end; 745 746 start = (u8 *)pbd1; 747 748 /* Skip the block header(we know header WILL fit in 4K) */ 749 start += PAGE_SIZE; 750 751 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 752 for (; start < end; start += PAGE_SIZE) 753 flush_dcache_page(pgv_to_page(start)); 754 755 smp_wmb(); 756 #endif 757 758 /* Now update the block status. */ 759 760 BLOCK_STATUS(pbd1) = status; 761 762 /* Flush the block header */ 763 764 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 765 start = (u8 *)pbd1; 766 flush_dcache_page(pgv_to_page(start)); 767 768 smp_wmb(); 769 #endif 770 } 771 772 /* 773 * Side effect: 774 * 775 * 1) flush the block 776 * 2) Increment active_blk_num 777 * 778 * Note:We DONT refresh the timer on purpose. 779 * Because almost always the next block will be opened. 780 */ 781 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 782 struct tpacket_block_desc *pbd1, 783 struct packet_sock *po, unsigned int stat) 784 { 785 __u32 status = TP_STATUS_USER | stat; 786 787 struct tpacket3_hdr *last_pkt; 788 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 789 struct sock *sk = &po->sk; 790 791 if (atomic_read(&po->tp_drops)) 792 status |= TP_STATUS_LOSING; 793 794 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 795 last_pkt->tp_next_offset = 0; 796 797 /* Get the ts of the last pkt */ 798 if (BLOCK_NUM_PKTS(pbd1)) { 799 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 800 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 801 } else { 802 /* Ok, we tmo'd - so get the current time. 803 * 804 * It shouldn't really happen as we don't close empty 805 * blocks. See prb_retire_rx_blk_timer_expired(). 806 */ 807 struct timespec64 ts; 808 ktime_get_real_ts64(&ts); 809 h1->ts_last_pkt.ts_sec = ts.tv_sec; 810 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 811 } 812 813 smp_wmb(); 814 815 /* Flush the block */ 816 prb_flush_block(pkc1, pbd1, status); 817 818 sk->sk_data_ready(sk); 819 820 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 821 } 822 823 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 824 { 825 pkc->reset_pending_on_curr_blk = 0; 826 } 827 828 /* 829 * prb_open_block is called by tpacket_rcv or timer callback. 830 * 831 * Reasons why NOT update hrtimer in prb_open_block: 832 * 1) It will increase complexity to distinguish the two caller scenario. 833 * 2) hrtimer_cancel and hrtimer_start need to be called if you want to update 834 * TMO of an already enqueued hrtimer, leading to complex shutdown logic. 835 * 836 * One side effect of NOT update hrtimer when called by tpacket_rcv is that 837 * a newly opened block triggered by tpacket_rcv may be retired earlier than 838 * expected. On the other hand, if timeout is updated in prb_open_block, the 839 * frequent reception of network packets that leads to prb_open_block being 840 * called may cause hrtimer to be removed and enqueued repeatedly. 841 */ 842 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 843 struct tpacket_block_desc *pbd1) 844 { 845 struct timespec64 ts; 846 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 847 848 smp_rmb(); 849 850 /* We could have just memset this but we will lose the 851 * flexibility of making the priv area sticky 852 */ 853 854 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 855 BLOCK_NUM_PKTS(pbd1) = 0; 856 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 857 858 ktime_get_real_ts64(&ts); 859 860 h1->ts_first_pkt.ts_sec = ts.tv_sec; 861 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 862 863 pkc1->pkblk_start = (char *)pbd1; 864 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 865 866 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 867 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 868 869 pbd1->version = pkc1->version; 870 pkc1->prev = pkc1->nxt_offset; 871 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 872 873 prb_thaw_queue(pkc1); 874 875 smp_wmb(); 876 } 877 878 /* 879 * Queue freeze logic: 880 * 1) Assume tp_block_nr = 8 blocks. 881 * 2) At time 't0', user opens Rx ring. 882 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 883 * 4) user-space is either sleeping or processing block '0'. 884 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 885 * it will close block-7,loop around and try to fill block '0'. 886 * call-flow: 887 * __packet_lookup_frame_in_block 888 * prb_retire_current_block() 889 * prb_dispatch_next_block() 890 * |->(BLOCK_STATUS == USER) evaluates to true 891 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 892 * 6) Now there are two cases: 893 * 6.1) Link goes idle right after the queue is frozen. 894 * But remember, the last open_block() refreshed the timer. 895 * When this timer expires,it will refresh itself so that we can 896 * re-open block-0 in near future. 897 * 6.2) Link is busy and keeps on receiving packets. This is a simple 898 * case and __packet_lookup_frame_in_block will check if block-0 899 * is free and can now be re-used. 900 */ 901 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 902 struct packet_sock *po) 903 { 904 pkc->reset_pending_on_curr_blk = 1; 905 po->stats.stats3.tp_freeze_q_cnt++; 906 } 907 908 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 909 910 /* 911 * If the next block is free then we will dispatch it 912 * and return a good offset. 913 * Else, we will freeze the queue. 914 * So, caller must check the return value. 915 */ 916 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 917 struct packet_sock *po) 918 { 919 struct tpacket_block_desc *pbd; 920 921 smp_rmb(); 922 923 /* 1. Get current block num */ 924 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 925 926 /* 2. If this block is currently in_use then freeze the queue */ 927 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 928 prb_freeze_queue(pkc, po); 929 return NULL; 930 } 931 932 /* 933 * 3. 934 * open this block and return the offset where the first packet 935 * needs to get stored. 936 */ 937 prb_open_block(pkc, pbd); 938 return (void *)pkc->nxt_offset; 939 } 940 941 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 942 struct packet_sock *po, unsigned int status) 943 { 944 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 945 946 /* retire/close the current block */ 947 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 948 /* 949 * Plug the case where copy_bits() is in progress on 950 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 951 * have space to copy the pkt in the current block and 952 * called prb_retire_current_block() 953 * 954 * We don't need to worry about the TMO case because 955 * the timer-handler already handled this case. 956 */ 957 if (!(status & TP_STATUS_BLK_TMO)) { 958 /* Waiting for skb_copy_bits to finish... */ 959 write_lock(&pkc->blk_fill_in_prog_lock); 960 write_unlock(&pkc->blk_fill_in_prog_lock); 961 } 962 prb_close_block(pkc, pbd, po, status); 963 return; 964 } 965 } 966 967 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) 968 { 969 return TP_STATUS_USER & BLOCK_STATUS(pbd); 970 } 971 972 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 973 { 974 return pkc->reset_pending_on_curr_blk; 975 } 976 977 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 978 __releases(&pkc->blk_fill_in_prog_lock) 979 { 980 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 981 982 read_unlock(&pkc->blk_fill_in_prog_lock); 983 } 984 985 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 986 struct tpacket3_hdr *ppd) 987 { 988 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 989 } 990 991 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 992 struct tpacket3_hdr *ppd) 993 { 994 ppd->hv1.tp_rxhash = 0; 995 } 996 997 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 998 struct tpacket3_hdr *ppd) 999 { 1000 struct packet_sock *po = container_of(pkc, struct packet_sock, rx_ring.prb_bdqc); 1001 1002 if (skb_vlan_tag_present(pkc->skb)) { 1003 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 1004 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 1005 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 1006 } else if (unlikely(po->sk.sk_type == SOCK_DGRAM && eth_type_vlan(pkc->skb->protocol))) { 1007 ppd->hv1.tp_vlan_tci = vlan_get_tci(pkc->skb, pkc->skb->dev); 1008 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->protocol); 1009 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 1010 } else { 1011 ppd->hv1.tp_vlan_tci = 0; 1012 ppd->hv1.tp_vlan_tpid = 0; 1013 ppd->tp_status = TP_STATUS_AVAILABLE; 1014 } 1015 } 1016 1017 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 1018 struct tpacket3_hdr *ppd) 1019 { 1020 ppd->hv1.tp_padding = 0; 1021 prb_fill_vlan_info(pkc, ppd); 1022 1023 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 1024 prb_fill_rxhash(pkc, ppd); 1025 else 1026 prb_clear_rxhash(pkc, ppd); 1027 } 1028 1029 static void prb_fill_curr_block(char *curr, 1030 struct tpacket_kbdq_core *pkc, 1031 struct tpacket_block_desc *pbd, 1032 unsigned int len) 1033 __acquires(&pkc->blk_fill_in_prog_lock) 1034 { 1035 struct tpacket3_hdr *ppd; 1036 1037 ppd = (struct tpacket3_hdr *)curr; 1038 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 1039 pkc->prev = curr; 1040 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 1041 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1042 BLOCK_NUM_PKTS(pbd) += 1; 1043 read_lock(&pkc->blk_fill_in_prog_lock); 1044 prb_run_all_ft_ops(pkc, ppd); 1045 } 1046 1047 /* Assumes caller has the sk->rx_queue.lock */ 1048 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1049 struct sk_buff *skb, 1050 unsigned int len 1051 ) 1052 { 1053 struct tpacket_kbdq_core *pkc; 1054 struct tpacket_block_desc *pbd; 1055 char *curr, *end; 1056 1057 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1058 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1059 1060 /* Queue is frozen when user space is lagging behind */ 1061 if (prb_queue_frozen(pkc)) { 1062 /* 1063 * Check if that last block which caused the queue to freeze, 1064 * is still in_use by user-space. 1065 */ 1066 if (prb_curr_blk_in_use(pbd)) { 1067 /* Can't record this packet */ 1068 return NULL; 1069 } else { 1070 /* 1071 * Ok, the block was released by user-space. 1072 * Now let's open that block. 1073 * opening a block also thaws the queue. 1074 * Thawing is a side effect. 1075 */ 1076 prb_open_block(pkc, pbd); 1077 } 1078 } 1079 1080 smp_mb(); 1081 curr = pkc->nxt_offset; 1082 pkc->skb = skb; 1083 end = (char *)pbd + pkc->kblk_size; 1084 1085 /* first try the current block */ 1086 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1087 prb_fill_curr_block(curr, pkc, pbd, len); 1088 return (void *)curr; 1089 } 1090 1091 /* Ok, close the current block */ 1092 prb_retire_current_block(pkc, po, 0); 1093 1094 /* Now, try to dispatch the next block */ 1095 curr = (char *)prb_dispatch_next_block(pkc, po); 1096 if (curr) { 1097 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1098 prb_fill_curr_block(curr, pkc, pbd, len); 1099 return (void *)curr; 1100 } 1101 1102 /* 1103 * No free blocks are available.user_space hasn't caught up yet. 1104 * Queue was just frozen and now this packet will get dropped. 1105 */ 1106 return NULL; 1107 } 1108 1109 static void *packet_current_rx_frame(struct packet_sock *po, 1110 struct sk_buff *skb, 1111 int status, unsigned int len) 1112 { 1113 char *curr = NULL; 1114 switch (po->tp_version) { 1115 case TPACKET_V1: 1116 case TPACKET_V2: 1117 curr = packet_lookup_frame(po, &po->rx_ring, 1118 po->rx_ring.head, status); 1119 return curr; 1120 case TPACKET_V3: 1121 return __packet_lookup_frame_in_block(po, skb, len); 1122 default: 1123 WARN(1, "TPACKET version not supported\n"); 1124 BUG(); 1125 return NULL; 1126 } 1127 } 1128 1129 static void *prb_lookup_block(const struct packet_sock *po, 1130 const struct packet_ring_buffer *rb, 1131 unsigned int idx, 1132 int status) 1133 { 1134 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1135 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1136 1137 if (status != BLOCK_STATUS(pbd)) 1138 return NULL; 1139 return pbd; 1140 } 1141 1142 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1143 { 1144 unsigned int prev; 1145 if (rb->prb_bdqc.kactive_blk_num) 1146 prev = rb->prb_bdqc.kactive_blk_num-1; 1147 else 1148 prev = rb->prb_bdqc.knum_blocks-1; 1149 return prev; 1150 } 1151 1152 /* Assumes caller has held the rx_queue.lock */ 1153 static void *__prb_previous_block(struct packet_sock *po, 1154 struct packet_ring_buffer *rb, 1155 int status) 1156 { 1157 unsigned int previous = prb_previous_blk_num(rb); 1158 return prb_lookup_block(po, rb, previous, status); 1159 } 1160 1161 static void *packet_previous_rx_frame(struct packet_sock *po, 1162 struct packet_ring_buffer *rb, 1163 int status) 1164 { 1165 if (po->tp_version <= TPACKET_V2) 1166 return packet_previous_frame(po, rb, status); 1167 1168 return __prb_previous_block(po, rb, status); 1169 } 1170 1171 static void packet_increment_rx_head(struct packet_sock *po, 1172 struct packet_ring_buffer *rb) 1173 { 1174 switch (po->tp_version) { 1175 case TPACKET_V1: 1176 case TPACKET_V2: 1177 return packet_increment_head(rb); 1178 case TPACKET_V3: 1179 default: 1180 WARN(1, "TPACKET version not supported.\n"); 1181 BUG(); 1182 return; 1183 } 1184 } 1185 1186 static void *packet_previous_frame(struct packet_sock *po, 1187 struct packet_ring_buffer *rb, 1188 int status) 1189 { 1190 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1191 return packet_lookup_frame(po, rb, previous, status); 1192 } 1193 1194 static void packet_increment_head(struct packet_ring_buffer *buff) 1195 { 1196 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1197 } 1198 1199 static void packet_inc_pending(struct packet_ring_buffer *rb) 1200 { 1201 this_cpu_inc(*rb->pending_refcnt); 1202 } 1203 1204 static void packet_dec_pending(struct packet_ring_buffer *rb) 1205 { 1206 this_cpu_dec(*rb->pending_refcnt); 1207 } 1208 1209 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1210 { 1211 unsigned int refcnt = 0; 1212 int cpu; 1213 1214 /* We don't use pending refcount in rx_ring. */ 1215 if (rb->pending_refcnt == NULL) 1216 return 0; 1217 1218 for_each_possible_cpu(cpu) 1219 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1220 1221 return refcnt; 1222 } 1223 1224 static int packet_alloc_pending(struct packet_sock *po) 1225 { 1226 po->rx_ring.pending_refcnt = NULL; 1227 1228 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1229 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1230 return -ENOBUFS; 1231 1232 return 0; 1233 } 1234 1235 static void packet_free_pending(struct packet_sock *po) 1236 { 1237 free_percpu(po->tx_ring.pending_refcnt); 1238 } 1239 1240 #define ROOM_POW_OFF 2 1241 #define ROOM_NONE 0x0 1242 #define ROOM_LOW 0x1 1243 #define ROOM_NORMAL 0x2 1244 1245 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off) 1246 { 1247 int idx, len; 1248 1249 len = READ_ONCE(po->rx_ring.frame_max) + 1; 1250 idx = READ_ONCE(po->rx_ring.head); 1251 if (pow_off) 1252 idx += len >> pow_off; 1253 if (idx >= len) 1254 idx -= len; 1255 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1256 } 1257 1258 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off) 1259 { 1260 int idx, len; 1261 1262 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks); 1263 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num); 1264 if (pow_off) 1265 idx += len >> pow_off; 1266 if (idx >= len) 1267 idx -= len; 1268 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1269 } 1270 1271 static int __packet_rcv_has_room(const struct packet_sock *po, 1272 const struct sk_buff *skb) 1273 { 1274 const struct sock *sk = &po->sk; 1275 int ret = ROOM_NONE; 1276 1277 if (po->prot_hook.func != tpacket_rcv) { 1278 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1279 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1280 - (skb ? skb->truesize : 0); 1281 1282 if (avail > (rcvbuf >> ROOM_POW_OFF)) 1283 return ROOM_NORMAL; 1284 else if (avail > 0) 1285 return ROOM_LOW; 1286 else 1287 return ROOM_NONE; 1288 } 1289 1290 if (po->tp_version == TPACKET_V3) { 1291 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1292 ret = ROOM_NORMAL; 1293 else if (__tpacket_v3_has_room(po, 0)) 1294 ret = ROOM_LOW; 1295 } else { 1296 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1297 ret = ROOM_NORMAL; 1298 else if (__tpacket_has_room(po, 0)) 1299 ret = ROOM_LOW; 1300 } 1301 1302 return ret; 1303 } 1304 1305 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1306 { 1307 bool pressure; 1308 int ret; 1309 1310 ret = __packet_rcv_has_room(po, skb); 1311 pressure = ret != ROOM_NORMAL; 1312 1313 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure) 1314 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure); 1315 1316 return ret; 1317 } 1318 1319 static void __packet_rcv_try_clear_pressure(struct packet_sock *po) 1320 { 1321 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) && 1322 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 1323 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false); 1324 } 1325 1326 static void packet_rcv_try_clear_pressure(struct packet_sock *po) 1327 { 1328 struct sock *sk = &po->sk; 1329 1330 if (!packet_sock_flag(po, PACKET_SOCK_PRESSURE)) 1331 return; 1332 1333 spin_lock_bh(&sk->sk_receive_queue.lock); 1334 __packet_rcv_try_clear_pressure(po); 1335 spin_unlock_bh(&sk->sk_receive_queue.lock); 1336 } 1337 1338 static void packet_sock_destruct(struct sock *sk) 1339 { 1340 skb_queue_purge(&sk->sk_error_queue); 1341 1342 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1343 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 1344 1345 packet_free_pending(pkt_sk(sk)); 1346 1347 if (!sock_flag(sk, SOCK_DEAD)) { 1348 pr_err("Attempt to release alive packet socket: %p\n", sk); 1349 return; 1350 } 1351 } 1352 1353 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1354 { 1355 u32 *history = po->rollover->history; 1356 u32 victim, rxhash; 1357 int i, count = 0; 1358 1359 rxhash = skb_get_hash(skb); 1360 for (i = 0; i < ROLLOVER_HLEN; i++) 1361 if (READ_ONCE(history[i]) == rxhash) 1362 count++; 1363 1364 victim = get_random_u32_below(ROLLOVER_HLEN); 1365 1366 /* Avoid dirtying the cache line if possible */ 1367 if (READ_ONCE(history[victim]) != rxhash) 1368 WRITE_ONCE(history[victim], rxhash); 1369 1370 return count > (ROLLOVER_HLEN >> 1); 1371 } 1372 1373 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1374 struct sk_buff *skb, 1375 unsigned int num) 1376 { 1377 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1378 } 1379 1380 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1381 struct sk_buff *skb, 1382 unsigned int num) 1383 { 1384 unsigned int val = atomic_inc_return(&f->rr_cur); 1385 1386 return val % num; 1387 } 1388 1389 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1390 struct sk_buff *skb, 1391 unsigned int num) 1392 { 1393 return smp_processor_id() % num; 1394 } 1395 1396 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1397 struct sk_buff *skb, 1398 unsigned int num) 1399 { 1400 return get_random_u32_below(num); 1401 } 1402 1403 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1404 struct sk_buff *skb, 1405 unsigned int idx, bool try_self, 1406 unsigned int num) 1407 { 1408 struct packet_sock *po, *po_next, *po_skip = NULL; 1409 unsigned int i, j, room = ROOM_NONE; 1410 1411 po = pkt_sk(rcu_dereference(f->arr[idx])); 1412 1413 if (try_self) { 1414 room = packet_rcv_has_room(po, skb); 1415 if (room == ROOM_NORMAL || 1416 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1417 return idx; 1418 po_skip = po; 1419 } 1420 1421 i = j = min_t(int, po->rollover->sock, num - 1); 1422 do { 1423 po_next = pkt_sk(rcu_dereference(f->arr[i])); 1424 if (po_next != po_skip && 1425 !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) && 1426 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1427 if (i != j) 1428 po->rollover->sock = i; 1429 atomic_long_inc(&po->rollover->num); 1430 if (room == ROOM_LOW) 1431 atomic_long_inc(&po->rollover->num_huge); 1432 return i; 1433 } 1434 1435 if (++i == num) 1436 i = 0; 1437 } while (i != j); 1438 1439 atomic_long_inc(&po->rollover->num_failed); 1440 return idx; 1441 } 1442 1443 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1444 struct sk_buff *skb, 1445 unsigned int num) 1446 { 1447 return skb_get_queue_mapping(skb) % num; 1448 } 1449 1450 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1451 struct sk_buff *skb, 1452 unsigned int num) 1453 { 1454 struct bpf_prog *prog; 1455 unsigned int ret = 0; 1456 1457 rcu_read_lock(); 1458 prog = rcu_dereference(f->bpf_prog); 1459 if (prog) 1460 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1461 rcu_read_unlock(); 1462 1463 return ret; 1464 } 1465 1466 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1467 { 1468 return f->flags & (flag >> 8); 1469 } 1470 1471 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1472 struct packet_type *pt, struct net_device *orig_dev) 1473 { 1474 struct packet_fanout *f = pt->af_packet_priv; 1475 unsigned int num = READ_ONCE(f->num_members); 1476 struct net *net = read_pnet(&f->net); 1477 struct packet_sock *po; 1478 unsigned int idx; 1479 1480 if (!net_eq(dev_net(dev), net) || !num) { 1481 kfree_skb(skb); 1482 return 0; 1483 } 1484 1485 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1486 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1487 if (!skb) 1488 return 0; 1489 } 1490 switch (f->type) { 1491 case PACKET_FANOUT_HASH: 1492 default: 1493 idx = fanout_demux_hash(f, skb, num); 1494 break; 1495 case PACKET_FANOUT_LB: 1496 idx = fanout_demux_lb(f, skb, num); 1497 break; 1498 case PACKET_FANOUT_CPU: 1499 idx = fanout_demux_cpu(f, skb, num); 1500 break; 1501 case PACKET_FANOUT_RND: 1502 idx = fanout_demux_rnd(f, skb, num); 1503 break; 1504 case PACKET_FANOUT_QM: 1505 idx = fanout_demux_qm(f, skb, num); 1506 break; 1507 case PACKET_FANOUT_ROLLOVER: 1508 idx = fanout_demux_rollover(f, skb, 0, false, num); 1509 break; 1510 case PACKET_FANOUT_CBPF: 1511 case PACKET_FANOUT_EBPF: 1512 idx = fanout_demux_bpf(f, skb, num); 1513 break; 1514 } 1515 1516 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1517 idx = fanout_demux_rollover(f, skb, idx, true, num); 1518 1519 po = pkt_sk(rcu_dereference(f->arr[idx])); 1520 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1521 } 1522 1523 DEFINE_MUTEX(fanout_mutex); 1524 EXPORT_SYMBOL_GPL(fanout_mutex); 1525 static LIST_HEAD(fanout_list); 1526 static u16 fanout_next_id; 1527 1528 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1529 { 1530 struct packet_fanout *f = po->fanout; 1531 1532 spin_lock(&f->lock); 1533 rcu_assign_pointer(f->arr[f->num_members], sk); 1534 smp_wmb(); 1535 f->num_members++; 1536 if (f->num_members == 1) 1537 dev_add_pack(&f->prot_hook); 1538 spin_unlock(&f->lock); 1539 } 1540 1541 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1542 { 1543 struct packet_fanout *f = po->fanout; 1544 int i; 1545 1546 spin_lock(&f->lock); 1547 for (i = 0; i < f->num_members; i++) { 1548 if (rcu_dereference_protected(f->arr[i], 1549 lockdep_is_held(&f->lock)) == sk) 1550 break; 1551 } 1552 BUG_ON(i >= f->num_members); 1553 rcu_assign_pointer(f->arr[i], 1554 rcu_dereference_protected(f->arr[f->num_members - 1], 1555 lockdep_is_held(&f->lock))); 1556 f->num_members--; 1557 if (f->num_members == 0) 1558 __dev_remove_pack(&f->prot_hook); 1559 spin_unlock(&f->lock); 1560 } 1561 1562 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1563 { 1564 if (sk->sk_family != PF_PACKET) 1565 return false; 1566 1567 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1568 } 1569 1570 static void fanout_init_data(struct packet_fanout *f) 1571 { 1572 switch (f->type) { 1573 case PACKET_FANOUT_LB: 1574 atomic_set(&f->rr_cur, 0); 1575 break; 1576 case PACKET_FANOUT_CBPF: 1577 case PACKET_FANOUT_EBPF: 1578 RCU_INIT_POINTER(f->bpf_prog, NULL); 1579 break; 1580 } 1581 } 1582 1583 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1584 { 1585 struct bpf_prog *old; 1586 1587 spin_lock(&f->lock); 1588 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1589 rcu_assign_pointer(f->bpf_prog, new); 1590 spin_unlock(&f->lock); 1591 1592 if (old) { 1593 synchronize_net(); 1594 bpf_prog_destroy(old); 1595 } 1596 } 1597 1598 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data, 1599 unsigned int len) 1600 { 1601 struct bpf_prog *new; 1602 struct sock_fprog fprog; 1603 int ret; 1604 1605 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1606 return -EPERM; 1607 1608 ret = copy_bpf_fprog_from_user(&fprog, data, len); 1609 if (ret) 1610 return ret; 1611 1612 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1613 if (ret) 1614 return ret; 1615 1616 __fanout_set_data_bpf(po->fanout, new); 1617 return 0; 1618 } 1619 1620 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data, 1621 unsigned int len) 1622 { 1623 struct bpf_prog *new; 1624 u32 fd; 1625 1626 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1627 return -EPERM; 1628 if (len != sizeof(fd)) 1629 return -EINVAL; 1630 if (copy_from_sockptr(&fd, data, len)) 1631 return -EFAULT; 1632 1633 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1634 if (IS_ERR(new)) 1635 return PTR_ERR(new); 1636 1637 __fanout_set_data_bpf(po->fanout, new); 1638 return 0; 1639 } 1640 1641 static int fanout_set_data(struct packet_sock *po, sockptr_t data, 1642 unsigned int len) 1643 { 1644 switch (po->fanout->type) { 1645 case PACKET_FANOUT_CBPF: 1646 return fanout_set_data_cbpf(po, data, len); 1647 case PACKET_FANOUT_EBPF: 1648 return fanout_set_data_ebpf(po, data, len); 1649 default: 1650 return -EINVAL; 1651 } 1652 } 1653 1654 static void fanout_release_data(struct packet_fanout *f) 1655 { 1656 switch (f->type) { 1657 case PACKET_FANOUT_CBPF: 1658 case PACKET_FANOUT_EBPF: 1659 __fanout_set_data_bpf(f, NULL); 1660 } 1661 } 1662 1663 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) 1664 { 1665 struct packet_fanout *f; 1666 1667 list_for_each_entry(f, &fanout_list, list) { 1668 if (f->id == candidate_id && 1669 read_pnet(&f->net) == sock_net(sk)) { 1670 return false; 1671 } 1672 } 1673 return true; 1674 } 1675 1676 static bool fanout_find_new_id(struct sock *sk, u16 *new_id) 1677 { 1678 u16 id = fanout_next_id; 1679 1680 do { 1681 if (__fanout_id_is_free(sk, id)) { 1682 *new_id = id; 1683 fanout_next_id = id + 1; 1684 return true; 1685 } 1686 1687 id++; 1688 } while (id != fanout_next_id); 1689 1690 return false; 1691 } 1692 1693 static int fanout_add(struct sock *sk, struct fanout_args *args) 1694 { 1695 struct packet_rollover *rollover = NULL; 1696 struct packet_sock *po = pkt_sk(sk); 1697 u16 type_flags = args->type_flags; 1698 struct packet_fanout *f, *match; 1699 u8 type = type_flags & 0xff; 1700 u8 flags = type_flags >> 8; 1701 u16 id = args->id; 1702 int err; 1703 1704 switch (type) { 1705 case PACKET_FANOUT_ROLLOVER: 1706 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1707 return -EINVAL; 1708 break; 1709 case PACKET_FANOUT_HASH: 1710 case PACKET_FANOUT_LB: 1711 case PACKET_FANOUT_CPU: 1712 case PACKET_FANOUT_RND: 1713 case PACKET_FANOUT_QM: 1714 case PACKET_FANOUT_CBPF: 1715 case PACKET_FANOUT_EBPF: 1716 break; 1717 default: 1718 return -EINVAL; 1719 } 1720 1721 mutex_lock(&fanout_mutex); 1722 1723 err = -EALREADY; 1724 if (po->fanout) 1725 goto out; 1726 1727 if (type == PACKET_FANOUT_ROLLOVER || 1728 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1729 err = -ENOMEM; 1730 rollover = kzalloc_obj(*rollover); 1731 if (!rollover) 1732 goto out; 1733 atomic_long_set(&rollover->num, 0); 1734 atomic_long_set(&rollover->num_huge, 0); 1735 atomic_long_set(&rollover->num_failed, 0); 1736 } 1737 1738 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { 1739 if (id != 0) { 1740 err = -EINVAL; 1741 goto out; 1742 } 1743 if (!fanout_find_new_id(sk, &id)) { 1744 err = -ENOMEM; 1745 goto out; 1746 } 1747 /* ephemeral flag for the first socket in the group: drop it */ 1748 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); 1749 } 1750 1751 match = NULL; 1752 list_for_each_entry(f, &fanout_list, list) { 1753 if (f->id == id && 1754 read_pnet(&f->net) == sock_net(sk)) { 1755 match = f; 1756 break; 1757 } 1758 } 1759 err = -EINVAL; 1760 if (match) { 1761 if (match->flags != flags) 1762 goto out; 1763 if (args->max_num_members && 1764 args->max_num_members != match->max_num_members) 1765 goto out; 1766 } else { 1767 if (args->max_num_members > PACKET_FANOUT_MAX) 1768 goto out; 1769 if (!args->max_num_members) 1770 /* legacy PACKET_FANOUT_MAX */ 1771 args->max_num_members = 256; 1772 err = -ENOMEM; 1773 match = kvzalloc_flex(*match, arr, args->max_num_members); 1774 if (!match) 1775 goto out; 1776 write_pnet(&match->net, sock_net(sk)); 1777 match->id = id; 1778 match->type = type; 1779 match->flags = flags; 1780 INIT_LIST_HEAD(&match->list); 1781 spin_lock_init(&match->lock); 1782 refcount_set(&match->sk_ref, 0); 1783 fanout_init_data(match); 1784 match->prot_hook.type = po->prot_hook.type; 1785 match->prot_hook.dev = po->prot_hook.dev; 1786 match->prot_hook.func = packet_rcv_fanout; 1787 match->prot_hook.af_packet_priv = match; 1788 match->prot_hook.af_packet_net = read_pnet(&match->net); 1789 match->prot_hook.id_match = match_fanout_group; 1790 match->max_num_members = args->max_num_members; 1791 match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING; 1792 list_add(&match->list, &fanout_list); 1793 } 1794 err = -EINVAL; 1795 1796 spin_lock(&po->bind_lock); 1797 if (po->num && 1798 match->type == type && 1799 match->prot_hook.type == po->prot_hook.type && 1800 match->prot_hook.dev == po->prot_hook.dev) { 1801 err = -ENOSPC; 1802 if (refcount_read(&match->sk_ref) < match->max_num_members) { 1803 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */ 1804 WRITE_ONCE(po->fanout, match); 1805 1806 po->rollover = rollover; 1807 rollover = NULL; 1808 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); 1809 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 1810 __dev_remove_pack(&po->prot_hook); 1811 __fanout_link(sk, po); 1812 } 1813 err = 0; 1814 } 1815 } 1816 spin_unlock(&po->bind_lock); 1817 1818 if (err && !refcount_read(&match->sk_ref)) { 1819 list_del(&match->list); 1820 kvfree(match); 1821 } 1822 1823 out: 1824 kfree(rollover); 1825 mutex_unlock(&fanout_mutex); 1826 return err; 1827 } 1828 1829 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes 1830 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. 1831 * It is the responsibility of the caller to call fanout_release_data() and 1832 * free the returned packet_fanout (after synchronize_net()) 1833 */ 1834 static struct packet_fanout *fanout_release(struct sock *sk) 1835 { 1836 struct packet_sock *po = pkt_sk(sk); 1837 struct packet_fanout *f; 1838 1839 mutex_lock(&fanout_mutex); 1840 f = po->fanout; 1841 if (f) { 1842 po->fanout = NULL; 1843 1844 if (refcount_dec_and_test(&f->sk_ref)) 1845 list_del(&f->list); 1846 else 1847 f = NULL; 1848 } 1849 mutex_unlock(&fanout_mutex); 1850 1851 return f; 1852 } 1853 1854 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1855 struct sk_buff *skb) 1856 { 1857 /* Earlier code assumed this would be a VLAN pkt, double-check 1858 * this now that we have the actual packet in hand. We can only 1859 * do this check on Ethernet devices. 1860 */ 1861 if (unlikely(dev->type != ARPHRD_ETHER)) 1862 return false; 1863 1864 skb_reset_mac_header(skb); 1865 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1866 } 1867 1868 static const struct proto_ops packet_ops; 1869 1870 static const struct proto_ops packet_ops_spkt; 1871 1872 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1873 struct packet_type *pt, struct net_device *orig_dev) 1874 { 1875 struct sock *sk; 1876 struct sockaddr_pkt *spkt; 1877 1878 /* 1879 * When we registered the protocol we saved the socket in the data 1880 * field for just this event. 1881 */ 1882 1883 sk = pt->af_packet_priv; 1884 1885 /* 1886 * Yank back the headers [hope the device set this 1887 * right or kerboom...] 1888 * 1889 * Incoming packets have ll header pulled, 1890 * push it back. 1891 * 1892 * For outgoing ones skb->data == skb_mac_header(skb) 1893 * so that this procedure is noop. 1894 */ 1895 1896 if (skb->pkt_type == PACKET_LOOPBACK) 1897 goto out; 1898 1899 if (!net_eq(dev_net(dev), sock_net(sk))) 1900 goto out; 1901 1902 skb = skb_share_check(skb, GFP_ATOMIC); 1903 if (skb == NULL) 1904 goto oom; 1905 1906 /* drop any routing info */ 1907 skb_dst_drop(skb); 1908 1909 /* drop conntrack reference */ 1910 nf_reset_ct(skb); 1911 1912 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1913 1914 skb_push(skb, skb->data - skb_mac_header(skb)); 1915 1916 /* 1917 * The SOCK_PACKET socket receives _all_ frames. 1918 */ 1919 1920 spkt->spkt_family = dev->type; 1921 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1922 spkt->spkt_protocol = skb->protocol; 1923 1924 /* 1925 * Charge the memory to the socket. This is done specifically 1926 * to prevent sockets using all the memory up. 1927 */ 1928 1929 if (sock_queue_rcv_skb(sk, skb) == 0) 1930 return 0; 1931 1932 out: 1933 kfree_skb(skb); 1934 oom: 1935 return 0; 1936 } 1937 1938 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock) 1939 { 1940 int depth; 1941 1942 /* On TX skb->data is the L2 header; anchor it for all socket types. */ 1943 skb_reset_mac_header(skb); 1944 1945 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && 1946 sock->type == SOCK_RAW) 1947 skb->protocol = dev_parse_header_protocol(skb); 1948 1949 skb_probe_transport_header(skb); 1950 1951 /* Move network header to the right position for VLAN tagged packets */ 1952 if (likely(skb->dev->type == ARPHRD_ETHER) && 1953 eth_type_vlan(skb->protocol) && 1954 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0) 1955 skb_set_network_header(skb, depth); 1956 } 1957 1958 /* 1959 * Output a raw packet to a device layer. This bypasses all the other 1960 * protocol layers and you must therefore supply it with a complete frame 1961 */ 1962 1963 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1964 size_t len) 1965 { 1966 struct sock *sk = sock->sk; 1967 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1968 struct sk_buff *skb = NULL; 1969 struct net_device *dev; 1970 struct sockcm_cookie sockc; 1971 __be16 proto = 0; 1972 int hard_header_len; 1973 int extra_len = 0; 1974 int err; 1975 1976 /* 1977 * Get and verify the address. 1978 */ 1979 1980 if (saddr) { 1981 if (msg->msg_namelen < sizeof(struct sockaddr)) 1982 return -EINVAL; 1983 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1984 proto = saddr->spkt_protocol; 1985 } else 1986 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1987 1988 /* 1989 * Find the device first to size check it 1990 */ 1991 1992 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1993 retry: 1994 rcu_read_lock(); 1995 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1996 err = -ENODEV; 1997 if (dev == NULL) 1998 goto out_unlock; 1999 2000 err = -ENETDOWN; 2001 if (!(dev->flags & IFF_UP)) 2002 goto out_unlock; 2003 2004 /* 2005 * You may not queue a frame bigger than the mtu. This is the lowest level 2006 * raw protocol and you must do your own fragmentation at this level. 2007 */ 2008 2009 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2010 if (!netif_supports_nofcs(dev)) { 2011 err = -EPROTONOSUPPORT; 2012 goto out_unlock; 2013 } 2014 extra_len = 4; /* We're doing our own CRC */ 2015 } 2016 2017 /* Keep the allocation-time header length across retry. */ 2018 if (!skb) 2019 hard_header_len = READ_ONCE(dev->hard_header_len); 2020 2021 err = -EMSGSIZE; 2022 if (len > dev->mtu + hard_header_len + VLAN_HLEN + extra_len) 2023 goto out_unlock; 2024 2025 if (!skb) { 2026 size_t reserved = LL_RESERVED_SPACE_EX(dev, hard_header_len); 2027 int tlen = dev->needed_tailroom; 2028 unsigned int hhlen = dev->header_ops ? hard_header_len : 0; 2029 2030 rcu_read_unlock(); 2031 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 2032 if (skb == NULL) 2033 return -ENOBUFS; 2034 /* FIXME: Save some space for broken drivers that write a hard 2035 * header at transmission time by themselves. PPP is the notable 2036 * one here. This should really be fixed at the driver level. 2037 */ 2038 skb_reserve(skb, reserved); 2039 skb_reset_network_header(skb); 2040 2041 /* Try to align data part correctly */ 2042 if (hhlen) { 2043 skb->data -= hhlen; 2044 skb->tail -= hhlen; 2045 if (len < hhlen) 2046 skb_reset_network_header(skb); 2047 } 2048 err = memcpy_from_msg(skb_put(skb, len), msg, len); 2049 if (err) 2050 goto out_free; 2051 goto retry; 2052 } 2053 2054 if (!dev_validate_header(dev, skb->data, len) || !skb->len) { 2055 err = -EINVAL; 2056 goto out_unlock; 2057 } 2058 if (len > (dev->mtu + hard_header_len + extra_len) && 2059 !packet_extra_vlan_len_allowed(dev, skb)) { 2060 err = -EMSGSIZE; 2061 goto out_unlock; 2062 } 2063 2064 sockcm_init(&sockc, sk); 2065 if (msg->msg_controllen) { 2066 err = sock_cmsg_send(sk, msg, &sockc); 2067 if (unlikely(err)) 2068 goto out_unlock; 2069 } 2070 2071 skb->protocol = proto; 2072 skb->dev = dev; 2073 skb->priority = sockc.priority; 2074 skb->mark = sockc.mark; 2075 skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid); 2076 skb_setup_tx_timestamp(skb, &sockc); 2077 2078 if (unlikely(extra_len == 4)) 2079 skb->no_fcs = 1; 2080 2081 packet_parse_headers(skb, sock); 2082 2083 dev_queue_xmit(skb); 2084 rcu_read_unlock(); 2085 return len; 2086 2087 out_unlock: 2088 rcu_read_unlock(); 2089 out_free: 2090 kfree_skb(skb); 2091 return err; 2092 } 2093 2094 static unsigned int run_filter(struct sk_buff *skb, 2095 const struct sock *sk, 2096 unsigned int res) 2097 { 2098 struct sk_filter *filter; 2099 2100 rcu_read_lock(); 2101 filter = rcu_dereference(sk->sk_filter); 2102 if (filter != NULL) 2103 res = bpf_prog_run_clear_cb(filter->prog, skb); 2104 rcu_read_unlock(); 2105 2106 return res; 2107 } 2108 2109 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2110 size_t *len, int vnet_hdr_sz) 2111 { 2112 struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 }; 2113 2114 if (*len < vnet_hdr_sz) 2115 return -EINVAL; 2116 *len -= vnet_hdr_sz; 2117 2118 if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0)) 2119 return -EINVAL; 2120 2121 return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz); 2122 } 2123 2124 /* 2125 * This function makes lazy skb cloning in hope that most of packets 2126 * are discarded by BPF. 2127 * 2128 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2129 * and skb->cb are mangled. It works because (and until) packets 2130 * falling here are owned by current CPU. Output packets are cloned 2131 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2132 * sequentially, so that if we return skb to original state on exit, 2133 * we will not harm anyone. 2134 */ 2135 2136 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2137 struct packet_type *pt, struct net_device *orig_dev) 2138 { 2139 enum skb_drop_reason drop_reason = SKB_CONSUMED; 2140 struct sock *sk = NULL; 2141 struct sockaddr_ll *sll; 2142 struct packet_sock *po; 2143 u8 *skb_head = skb->data; 2144 int skb_len = skb->len; 2145 unsigned int snaplen, res; 2146 2147 if (skb->pkt_type == PACKET_LOOPBACK) 2148 goto drop; 2149 2150 sk = pt->af_packet_priv; 2151 po = pkt_sk(sk); 2152 2153 if (!net_eq(dev_net(dev), sock_net(sk))) 2154 goto drop; 2155 2156 skb->dev = dev; 2157 2158 if (dev_has_header(dev)) { 2159 /* The device has an explicit notion of ll header, 2160 * exported to higher levels. 2161 * 2162 * Otherwise, the device hides details of its frame 2163 * structure, so that corresponding packet head is 2164 * never delivered to user. 2165 */ 2166 if (sk->sk_type != SOCK_DGRAM) 2167 skb_push(skb, skb->data - skb_mac_header(skb)); 2168 else if (skb->pkt_type == PACKET_OUTGOING) { 2169 /* Special case: outgoing packets have ll header at head */ 2170 skb_pull(skb, skb_network_offset(skb)); 2171 } 2172 } 2173 2174 snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb); 2175 2176 res = run_filter(skb, sk, snaplen); 2177 if (!res) 2178 goto drop_n_restore; 2179 if (snaplen > res) 2180 snaplen = res; 2181 2182 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2183 goto drop_n_acct; 2184 2185 if (skb_shared(skb)) { 2186 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2187 if (nskb == NULL) 2188 goto drop_n_acct; 2189 2190 if (skb_head != skb->data) { 2191 skb->data = skb_head; 2192 skb->len = skb_len; 2193 } 2194 consume_skb(skb); 2195 skb = nskb; 2196 } 2197 2198 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2199 2200 sll = &PACKET_SKB_CB(skb)->sa.ll; 2201 sll->sll_hatype = dev->type; 2202 sll->sll_pkttype = skb->pkt_type; 2203 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV))) 2204 sll->sll_ifindex = orig_dev->ifindex; 2205 else 2206 sll->sll_ifindex = dev->ifindex; 2207 2208 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2209 2210 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2211 * Use their space for storing the original skb length. 2212 */ 2213 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2214 2215 if (pskb_trim(skb, snaplen)) 2216 goto drop_n_acct; 2217 2218 skb_set_owner_r(skb, sk); 2219 skb->dev = NULL; 2220 skb_dst_drop(skb); 2221 2222 /* drop conntrack reference */ 2223 nf_reset_ct(skb); 2224 2225 spin_lock(&sk->sk_receive_queue.lock); 2226 po->stats.stats1.tp_packets++; 2227 sock_skb_set_dropcount(sk, skb); 2228 skb_clear_delivery_time(skb); 2229 __skb_queue_tail(&sk->sk_receive_queue, skb); 2230 spin_unlock(&sk->sk_receive_queue.lock); 2231 sk->sk_data_ready(sk); 2232 return 0; 2233 2234 drop_n_acct: 2235 atomic_inc(&po->tp_drops); 2236 sk_drops_inc(sk); 2237 drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR; 2238 2239 drop_n_restore: 2240 if (skb_head != skb->data && skb_shared(skb)) { 2241 skb->data = skb_head; 2242 skb->len = skb_len; 2243 } 2244 drop: 2245 sk_skb_reason_drop(sk, skb, drop_reason); 2246 return 0; 2247 } 2248 2249 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2250 struct packet_type *pt, struct net_device *orig_dev) 2251 { 2252 enum skb_drop_reason drop_reason = SKB_CONSUMED; 2253 struct sock *sk = NULL; 2254 struct packet_sock *po; 2255 struct sockaddr_ll *sll; 2256 union tpacket_uhdr h; 2257 u8 *skb_head = skb->data; 2258 int skb_len = skb->len; 2259 unsigned int snaplen, res; 2260 unsigned long status = TP_STATUS_USER; 2261 unsigned short macoff, hdrlen; 2262 unsigned int netoff; 2263 struct sk_buff *copy_skb = NULL; 2264 struct timespec64 ts; 2265 __u32 ts_status; 2266 unsigned int slot_id = 0; 2267 int vnet_hdr_sz = 0; 2268 2269 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2270 * We may add members to them until current aligned size without forcing 2271 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2272 */ 2273 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2274 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2275 2276 if (skb->pkt_type == PACKET_LOOPBACK) 2277 goto drop; 2278 2279 sk = pt->af_packet_priv; 2280 po = pkt_sk(sk); 2281 2282 if (!net_eq(dev_net(dev), sock_net(sk))) 2283 goto drop; 2284 2285 if (dev_has_header(dev)) { 2286 if (sk->sk_type != SOCK_DGRAM) 2287 skb_push(skb, skb->data - skb_mac_header(skb)); 2288 else if (skb->pkt_type == PACKET_OUTGOING) { 2289 /* Special case: outgoing packets have ll header at head */ 2290 skb_pull(skb, skb_network_offset(skb)); 2291 } 2292 } 2293 2294 snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb); 2295 2296 res = run_filter(skb, sk, snaplen); 2297 if (!res) 2298 goto drop_n_restore; 2299 2300 /* If we are flooded, just give up */ 2301 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) { 2302 atomic_inc(&po->tp_drops); 2303 goto drop_n_restore; 2304 } 2305 2306 if (skb->ip_summed == CHECKSUM_PARTIAL) 2307 status |= TP_STATUS_CSUMNOTREADY; 2308 else if (skb->pkt_type != PACKET_OUTGOING && 2309 skb_csum_unnecessary(skb)) 2310 status |= TP_STATUS_CSUM_VALID; 2311 if (skb_is_gso(skb) && skb_is_gso_tcp(skb)) 2312 status |= TP_STATUS_GSO_TCP; 2313 2314 if (snaplen > res) 2315 snaplen = res; 2316 2317 if (sk->sk_type == SOCK_DGRAM) { 2318 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2319 po->tp_reserve; 2320 } else { 2321 unsigned int maclen = skb_network_offset(skb); 2322 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2323 (maclen < 16 ? 16 : maclen)) + 2324 po->tp_reserve; 2325 vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz); 2326 if (vnet_hdr_sz) 2327 netoff += vnet_hdr_sz; 2328 macoff = netoff - maclen; 2329 } 2330 if (netoff > USHRT_MAX) { 2331 atomic_inc(&po->tp_drops); 2332 goto drop_n_restore; 2333 } 2334 if (po->tp_version <= TPACKET_V2) { 2335 if (macoff + snaplen > po->rx_ring.frame_size) { 2336 if (READ_ONCE(po->copy_thresh) && 2337 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2338 if (skb_shared(skb)) { 2339 copy_skb = skb_clone(skb, GFP_ATOMIC); 2340 } else { 2341 copy_skb = skb_get(skb); 2342 skb_head = skb->data; 2343 } 2344 if (copy_skb) { 2345 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0, 2346 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll)); 2347 skb_set_owner_r(copy_skb, sk); 2348 } 2349 } 2350 snaplen = po->rx_ring.frame_size - macoff; 2351 if ((int)snaplen < 0) { 2352 snaplen = 0; 2353 vnet_hdr_sz = 0; 2354 } 2355 } 2356 } else if (unlikely(macoff + snaplen > 2357 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2358 u32 nval; 2359 2360 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2361 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2362 snaplen, nval, macoff); 2363 snaplen = nval; 2364 if (unlikely((int)snaplen < 0)) { 2365 snaplen = 0; 2366 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2367 vnet_hdr_sz = 0; 2368 } 2369 } 2370 spin_lock(&sk->sk_receive_queue.lock); 2371 h.raw = packet_current_rx_frame(po, skb, 2372 TP_STATUS_KERNEL, (macoff+snaplen)); 2373 if (!h.raw) 2374 goto drop_n_account; 2375 2376 if (po->tp_version <= TPACKET_V2) { 2377 slot_id = po->rx_ring.head; 2378 if (test_bit(slot_id, po->rx_ring.rx_owner_map)) 2379 goto drop_n_account; 2380 __set_bit(slot_id, po->rx_ring.rx_owner_map); 2381 } 2382 2383 if (vnet_hdr_sz && 2384 virtio_net_hdr_from_skb(skb, h.raw + macoff - 2385 sizeof(struct virtio_net_hdr), 2386 vio_le(), true, 0)) { 2387 if (po->tp_version == TPACKET_V3) 2388 prb_clear_blk_fill_status(&po->rx_ring); 2389 goto drop_n_account; 2390 } 2391 2392 if (po->tp_version <= TPACKET_V2) { 2393 packet_increment_rx_head(po, &po->rx_ring); 2394 /* 2395 * LOSING will be reported till you read the stats, 2396 * because it's COR - Clear On Read. 2397 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2398 * at packet level. 2399 */ 2400 if (atomic_read(&po->tp_drops)) 2401 status |= TP_STATUS_LOSING; 2402 } 2403 2404 po->stats.stats1.tp_packets++; 2405 if (copy_skb) { 2406 status |= TP_STATUS_COPY; 2407 skb_clear_delivery_time(copy_skb); 2408 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2409 } 2410 spin_unlock(&sk->sk_receive_queue.lock); 2411 2412 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2413 2414 /* Always timestamp; prefer an existing software timestamp taken 2415 * closer to the time of capture. 2416 */ 2417 ts_status = tpacket_get_timestamp(skb, &ts, 2418 READ_ONCE(po->tp_tstamp) | 2419 SOF_TIMESTAMPING_SOFTWARE); 2420 if (!ts_status) 2421 ktime_get_real_ts64(&ts); 2422 2423 status |= ts_status; 2424 2425 switch (po->tp_version) { 2426 case TPACKET_V1: 2427 h.h1->tp_len = skb->len; 2428 h.h1->tp_snaplen = snaplen; 2429 h.h1->tp_mac = macoff; 2430 h.h1->tp_net = netoff; 2431 h.h1->tp_sec = ts.tv_sec; 2432 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2433 hdrlen = sizeof(*h.h1); 2434 break; 2435 case TPACKET_V2: 2436 h.h2->tp_len = skb->len; 2437 h.h2->tp_snaplen = snaplen; 2438 h.h2->tp_mac = macoff; 2439 h.h2->tp_net = netoff; 2440 h.h2->tp_sec = ts.tv_sec; 2441 h.h2->tp_nsec = ts.tv_nsec; 2442 if (skb_vlan_tag_present(skb)) { 2443 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2444 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2445 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2446 } else if (unlikely(sk->sk_type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) { 2447 h.h2->tp_vlan_tci = vlan_get_tci(skb, skb->dev); 2448 h.h2->tp_vlan_tpid = ntohs(skb->protocol); 2449 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2450 } else { 2451 h.h2->tp_vlan_tci = 0; 2452 h.h2->tp_vlan_tpid = 0; 2453 } 2454 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2455 hdrlen = sizeof(*h.h2); 2456 break; 2457 case TPACKET_V3: 2458 /* tp_nxt_offset,vlan are already populated above. 2459 * So DONT clear those fields here 2460 */ 2461 h.h3->tp_status |= status; 2462 h.h3->tp_len = skb->len; 2463 h.h3->tp_snaplen = snaplen; 2464 h.h3->tp_mac = macoff; 2465 h.h3->tp_net = netoff; 2466 h.h3->tp_sec = ts.tv_sec; 2467 h.h3->tp_nsec = ts.tv_nsec; 2468 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2469 hdrlen = sizeof(*h.h3); 2470 break; 2471 default: 2472 BUG(); 2473 } 2474 2475 sll = h.raw + TPACKET_ALIGN(hdrlen); 2476 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2477 sll->sll_family = AF_PACKET; 2478 sll->sll_hatype = dev->type; 2479 sll->sll_protocol = (sk->sk_type == SOCK_DGRAM) ? 2480 vlan_get_protocol_dgram(skb) : skb->protocol; 2481 sll->sll_pkttype = skb->pkt_type; 2482 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV))) 2483 sll->sll_ifindex = orig_dev->ifindex; 2484 else 2485 sll->sll_ifindex = dev->ifindex; 2486 2487 smp_mb(); 2488 2489 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2490 if (po->tp_version <= TPACKET_V2) { 2491 u8 *start, *end; 2492 2493 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2494 macoff + snaplen); 2495 2496 for (start = h.raw; start < end; start += PAGE_SIZE) 2497 flush_dcache_page(pgv_to_page(start)); 2498 } 2499 smp_wmb(); 2500 #endif 2501 2502 if (po->tp_version <= TPACKET_V2) { 2503 spin_lock(&sk->sk_receive_queue.lock); 2504 __packet_set_status(po, h.raw, status); 2505 __clear_bit(slot_id, po->rx_ring.rx_owner_map); 2506 spin_unlock(&sk->sk_receive_queue.lock); 2507 sk->sk_data_ready(sk); 2508 } else if (po->tp_version == TPACKET_V3) { 2509 prb_clear_blk_fill_status(&po->rx_ring); 2510 } 2511 2512 drop_n_restore: 2513 if (skb_head != skb->data && skb_shared(skb)) { 2514 skb->data = skb_head; 2515 skb->len = skb_len; 2516 } 2517 drop: 2518 sk_skb_reason_drop(sk, skb, drop_reason); 2519 return 0; 2520 2521 drop_n_account: 2522 spin_unlock(&sk->sk_receive_queue.lock); 2523 atomic_inc(&po->tp_drops); 2524 drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR; 2525 2526 sk->sk_data_ready(sk); 2527 sk_skb_reason_drop(sk, copy_skb, drop_reason); 2528 goto drop_n_restore; 2529 } 2530 2531 static void tpacket_destruct_skb(struct sk_buff *skb) 2532 { 2533 struct packet_sock *po = pkt_sk(skb->sk); 2534 2535 if (likely(po->tx_ring.pg_vec)) { 2536 void *ph; 2537 __u32 ts; 2538 2539 ph = skb_zcopy_get_nouarg(skb); 2540 2541 ts = __packet_set_timestamp(po, ph, skb); 2542 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2543 2544 packet_dec_pending(&po->tx_ring); 2545 complete(&po->skb_completion); 2546 } 2547 2548 sock_wfree(skb); 2549 } 2550 2551 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2552 { 2553 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2554 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2555 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2556 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2557 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2558 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2559 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2560 2561 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2562 return -EINVAL; 2563 2564 return 0; 2565 } 2566 2567 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2568 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz) 2569 { 2570 int ret; 2571 2572 if (*len < vnet_hdr_sz) 2573 return -EINVAL; 2574 *len -= vnet_hdr_sz; 2575 2576 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2577 return -EFAULT; 2578 2579 ret = __packet_snd_vnet_parse(vnet_hdr, *len); 2580 if (ret) 2581 return ret; 2582 2583 /* move iter to point to the start of mac header */ 2584 if (vnet_hdr_sz != sizeof(struct virtio_net_hdr)) 2585 iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr)); 2586 2587 return 0; 2588 } 2589 2590 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2591 void *frame, struct net_device *dev, void *data, int tp_len, 2592 __be16 proto, unsigned char *addr, int hlen, int copylen, 2593 int hard_header_len, 2594 const struct sockcm_cookie *sockc) 2595 { 2596 union tpacket_uhdr ph; 2597 int to_write, offset, len, nr_frags, len_max; 2598 struct socket *sock = po->sk.sk_socket; 2599 struct page *page; 2600 int err; 2601 2602 ph.raw = frame; 2603 2604 skb->protocol = proto; 2605 skb->dev = dev; 2606 skb->priority = sockc->priority; 2607 skb->mark = sockc->mark; 2608 skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, po->sk.sk_clockid); 2609 skb_setup_tx_timestamp(skb, sockc); 2610 skb_zcopy_set_nouarg(skb, ph.raw); 2611 2612 skb_reserve(skb, hlen); 2613 skb_reset_network_header(skb); 2614 2615 to_write = tp_len; 2616 2617 if (sock->type == SOCK_DGRAM) { 2618 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2619 NULL, tp_len); 2620 if (unlikely(err < 0)) 2621 return -EINVAL; 2622 } else if (copylen) { 2623 int hdrlen = min_t(int, copylen, tp_len); 2624 2625 skb_push(skb, hard_header_len); 2626 skb_put(skb, copylen - hard_header_len); 2627 err = skb_store_bits(skb, 0, data, hdrlen); 2628 if (unlikely(err)) 2629 return err; 2630 if (!dev_validate_header(dev, skb->data, hdrlen)) 2631 return -EINVAL; 2632 2633 data += hdrlen; 2634 to_write -= hdrlen; 2635 } 2636 2637 offset = offset_in_page(data); 2638 len_max = PAGE_SIZE - offset; 2639 len = ((to_write > len_max) ? len_max : to_write); 2640 2641 skb->data_len = to_write; 2642 skb->len += to_write; 2643 skb->truesize += to_write; 2644 refcount_add(to_write, &po->sk.sk_wmem_alloc); 2645 2646 while (likely(to_write)) { 2647 nr_frags = skb_shinfo(skb)->nr_frags; 2648 2649 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2650 pr_err("Packet exceed the number of skb frags(%u)\n", 2651 (unsigned int)MAX_SKB_FRAGS); 2652 return -EFAULT; 2653 } 2654 2655 page = pgv_to_page(data); 2656 data += len; 2657 flush_dcache_page(page); 2658 get_page(page); 2659 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2660 to_write -= len; 2661 offset = 0; 2662 len_max = PAGE_SIZE; 2663 len = ((to_write > len_max) ? len_max : to_write); 2664 } 2665 2666 if (unlikely(!skb->len)) 2667 return -EINVAL; 2668 2669 packet_parse_headers(skb, sock); 2670 2671 return tp_len; 2672 } 2673 2674 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2675 int size_max, void **data) 2676 { 2677 union tpacket_uhdr ph; 2678 u32 tp_len; 2679 int off; 2680 2681 ph.raw = frame; 2682 2683 switch (po->tp_version) { 2684 case TPACKET_V3: 2685 if (ph.h3->tp_next_offset != 0) { 2686 pr_warn_once("variable sized slot not supported"); 2687 return -EINVAL; 2688 } 2689 tp_len = ph.h3->tp_len; 2690 break; 2691 case TPACKET_V2: 2692 tp_len = ph.h2->tp_len; 2693 break; 2694 default: 2695 tp_len = ph.h1->tp_len; 2696 break; 2697 } 2698 if (unlikely(tp_len > size_max)) { 2699 pr_err("packet size is too long (%u > %d)\n", tp_len, size_max); 2700 return -EMSGSIZE; 2701 } 2702 2703 if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) { 2704 int off_min, off_max; 2705 2706 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2707 off_max = po->tx_ring.frame_size - tp_len; 2708 if (po->sk.sk_type == SOCK_DGRAM) { 2709 switch (po->tp_version) { 2710 case TPACKET_V3: 2711 off = ph.h3->tp_net; 2712 break; 2713 case TPACKET_V2: 2714 off = ph.h2->tp_net; 2715 break; 2716 default: 2717 off = ph.h1->tp_net; 2718 break; 2719 } 2720 } else { 2721 switch (po->tp_version) { 2722 case TPACKET_V3: 2723 off = ph.h3->tp_mac; 2724 break; 2725 case TPACKET_V2: 2726 off = ph.h2->tp_mac; 2727 break; 2728 default: 2729 off = ph.h1->tp_mac; 2730 break; 2731 } 2732 } 2733 if (unlikely((off < off_min) || (off_max < off))) 2734 return -EINVAL; 2735 } else { 2736 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2737 } 2738 2739 *data = frame + off; 2740 return tp_len; 2741 } 2742 2743 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2744 { 2745 struct sk_buff *skb = NULL; 2746 struct net_device *dev; 2747 struct virtio_net_hdr vnet_hdr; 2748 bool has_vnet_hdr = false; 2749 struct sockcm_cookie sockc; 2750 __be16 proto; 2751 int err, reserve = 0; 2752 void *ph; 2753 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2754 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2755 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz); 2756 unsigned char *addr = NULL; 2757 int tp_len, size_max; 2758 void *data; 2759 int len_sum = 0; 2760 int status = TP_STATUS_AVAILABLE; 2761 int hard_header_len, hlen, tlen, copylen = 0; 2762 long timeo; 2763 2764 mutex_lock(&po->pg_vec_lock); 2765 2766 /* packet_sendmsg() check on tx_ring.pg_vec was lockless, 2767 * we need to confirm it under protection of pg_vec_lock. 2768 */ 2769 if (unlikely(!po->tx_ring.pg_vec)) { 2770 err = -EBUSY; 2771 goto out; 2772 } 2773 if (likely(saddr == NULL)) { 2774 dev = packet_cached_dev_get(po); 2775 proto = READ_ONCE(po->num); 2776 } else { 2777 err = -EINVAL; 2778 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2779 goto out; 2780 if (msg->msg_namelen < (saddr->sll_halen 2781 + offsetof(struct sockaddr_ll, 2782 sll_addr))) 2783 goto out; 2784 proto = saddr->sll_protocol; 2785 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2786 if (po->sk.sk_socket->type == SOCK_DGRAM) { 2787 if (dev && msg->msg_namelen < dev->addr_len + 2788 offsetof(struct sockaddr_ll, sll_addr)) 2789 goto out_put; 2790 addr = saddr->sll_addr; 2791 } 2792 } 2793 2794 err = -ENXIO; 2795 if (unlikely(dev == NULL)) 2796 goto out; 2797 err = -ENETDOWN; 2798 if (unlikely(!(dev->flags & IFF_UP))) 2799 goto out_put; 2800 2801 sockcm_init(&sockc, &po->sk); 2802 if (msg->msg_controllen) { 2803 err = sock_cmsg_send(&po->sk, msg, &sockc); 2804 if (unlikely(err)) 2805 goto out_put; 2806 } 2807 2808 hard_header_len = READ_ONCE(dev->hard_header_len); 2809 if (po->sk.sk_socket->type == SOCK_RAW) 2810 reserve = hard_header_len; 2811 size_max = po->tx_ring.frame_size 2812 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2813 2814 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz) 2815 size_max = dev->mtu + reserve + VLAN_HLEN; 2816 2817 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT); 2818 reinit_completion(&po->skb_completion); 2819 2820 do { 2821 ph = packet_current_frame(po, &po->tx_ring, 2822 TP_STATUS_SEND_REQUEST); 2823 if (unlikely(ph == NULL)) { 2824 /* Note: packet_read_pending() might be slow if we 2825 * have to call it as it's per_cpu variable, but in 2826 * fast-path we don't have to call it, only when ph 2827 * is NULL, we need to check the pending_refcnt. 2828 */ 2829 if (need_wait && packet_read_pending(&po->tx_ring)) { 2830 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo); 2831 if (timeo <= 0) { 2832 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS; 2833 goto out_put; 2834 } 2835 /* check for additional frames */ 2836 continue; 2837 } else 2838 break; 2839 } 2840 2841 skb = NULL; 2842 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2843 if (tp_len < 0) 2844 goto tpacket_error; 2845 2846 status = TP_STATUS_SEND_REQUEST; 2847 hlen = LL_RESERVED_SPACE_EX(dev, hard_header_len); 2848 tlen = dev->needed_tailroom; 2849 if (vnet_hdr_sz) { 2850 data += vnet_hdr_sz; 2851 tp_len -= vnet_hdr_sz; 2852 if (tp_len < 0) { 2853 tp_len = -EINVAL; 2854 goto tpacket_error; 2855 } 2856 memcpy(&vnet_hdr, data - vnet_hdr_sz, sizeof(vnet_hdr)); 2857 if (__packet_snd_vnet_parse(&vnet_hdr, tp_len)) { 2858 tp_len = -EINVAL; 2859 goto tpacket_error; 2860 } 2861 copylen = __virtio16_to_cpu(vio_le(), 2862 vnet_hdr.hdr_len); 2863 has_vnet_hdr = true; 2864 } 2865 copylen = max_t(int, copylen, hard_header_len); 2866 skb = sock_alloc_send_skb(&po->sk, 2867 hlen + tlen + sizeof(struct sockaddr_ll) + 2868 (copylen - hard_header_len), 2869 !need_wait, &err); 2870 2871 if (unlikely(skb == NULL)) { 2872 /* we assume the socket was initially writeable ... */ 2873 if (likely(len_sum > 0)) 2874 err = len_sum; 2875 goto out_status; 2876 } 2877 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2878 addr, hlen, copylen, hard_header_len, 2879 &sockc); 2880 if (likely(tp_len >= 0) && 2881 tp_len > dev->mtu + reserve && 2882 !vnet_hdr_sz && 2883 !packet_extra_vlan_len_allowed(dev, skb)) 2884 tp_len = -EMSGSIZE; 2885 2886 if (unlikely(tp_len < 0)) { 2887 tpacket_error: 2888 if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) { 2889 __packet_set_status(po, ph, 2890 TP_STATUS_AVAILABLE); 2891 packet_increment_head(&po->tx_ring); 2892 kfree_skb(skb); 2893 continue; 2894 } else { 2895 status = TP_STATUS_WRONG_FORMAT; 2896 err = tp_len; 2897 goto out_status; 2898 } 2899 } 2900 2901 if (has_vnet_hdr) { 2902 if (virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le())) { 2903 tp_len = -EINVAL; 2904 goto tpacket_error; 2905 } 2906 virtio_net_hdr_set_proto(skb, &vnet_hdr); 2907 } 2908 2909 skb->destructor = tpacket_destruct_skb; 2910 __packet_set_status(po, ph, TP_STATUS_SENDING); 2911 packet_inc_pending(&po->tx_ring); 2912 2913 status = TP_STATUS_SEND_REQUEST; 2914 err = packet_xmit(po, skb); 2915 if (unlikely(err != 0)) { 2916 if (err > 0) 2917 err = net_xmit_errno(err); 2918 if (err && __packet_get_status(po, ph) == 2919 TP_STATUS_AVAILABLE) { 2920 /* skb was destructed already */ 2921 skb = NULL; 2922 goto out_status; 2923 } 2924 /* 2925 * skb was dropped but not destructed yet; 2926 * let's treat it like congestion or err < 0 2927 */ 2928 err = 0; 2929 } 2930 packet_increment_head(&po->tx_ring); 2931 len_sum += tp_len; 2932 } while (1); 2933 2934 err = len_sum; 2935 goto out_put; 2936 2937 out_status: 2938 __packet_set_status(po, ph, status); 2939 kfree_skb(skb); 2940 out_put: 2941 dev_put(dev); 2942 out: 2943 mutex_unlock(&po->pg_vec_lock); 2944 return err; 2945 } 2946 2947 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2948 size_t reserve, size_t len, 2949 size_t linear, int noblock, 2950 int *err) 2951 { 2952 struct sk_buff *skb; 2953 2954 /* Under a page? Don't bother with paged skb. */ 2955 if (prepad + len < PAGE_SIZE || !linear) 2956 linear = len; 2957 2958 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) 2959 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER); 2960 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2961 err, PAGE_ALLOC_COSTLY_ORDER); 2962 if (!skb) 2963 return NULL; 2964 2965 skb_reserve(skb, reserve); 2966 skb_put(skb, linear); 2967 skb->data_len = len - linear; 2968 skb->len += len - linear; 2969 2970 return skb; 2971 } 2972 2973 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2974 { 2975 struct sock *sk = sock->sk; 2976 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2977 struct sk_buff *skb; 2978 struct net_device *dev; 2979 __be16 proto; 2980 unsigned char *addr = NULL; 2981 int err, reserve = 0; 2982 struct sockcm_cookie sockc; 2983 struct virtio_net_hdr vnet_hdr = { 0 }; 2984 int offset = 0; 2985 struct packet_sock *po = pkt_sk(sk); 2986 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz); 2987 int hard_header_len, hlen, tlen, linear; 2988 int extra_len = 0; 2989 2990 /* 2991 * Get and verify the address. 2992 */ 2993 2994 if (likely(saddr == NULL)) { 2995 dev = packet_cached_dev_get(po); 2996 proto = READ_ONCE(po->num); 2997 } else { 2998 err = -EINVAL; 2999 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 3000 goto out; 3001 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 3002 goto out; 3003 proto = saddr->sll_protocol; 3004 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 3005 if (sock->type == SOCK_DGRAM) { 3006 if (dev && msg->msg_namelen < dev->addr_len + 3007 offsetof(struct sockaddr_ll, sll_addr)) 3008 goto out_unlock; 3009 addr = saddr->sll_addr; 3010 } 3011 } 3012 3013 err = -ENXIO; 3014 if (unlikely(dev == NULL)) 3015 goto out_unlock; 3016 err = -ENETDOWN; 3017 if (unlikely(!(dev->flags & IFF_UP))) 3018 goto out_unlock; 3019 3020 sockcm_init(&sockc, sk); 3021 if (msg->msg_controllen) { 3022 err = sock_cmsg_send(sk, msg, &sockc); 3023 if (unlikely(err)) 3024 goto out_unlock; 3025 } 3026 3027 hard_header_len = READ_ONCE(dev->hard_header_len); 3028 if (sock->type == SOCK_RAW) 3029 reserve = hard_header_len; 3030 if (vnet_hdr_sz) { 3031 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz); 3032 if (err) 3033 goto out_unlock; 3034 } 3035 3036 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 3037 if (!netif_supports_nofcs(dev)) { 3038 err = -EPROTONOSUPPORT; 3039 goto out_unlock; 3040 } 3041 extra_len = 4; /* We're doing our own CRC */ 3042 } 3043 3044 err = -EMSGSIZE; 3045 if (!vnet_hdr.gso_type && 3046 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 3047 goto out_unlock; 3048 3049 err = -ENOBUFS; 3050 hlen = LL_RESERVED_SPACE_EX(dev, hard_header_len); 3051 tlen = dev->needed_tailroom; 3052 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); 3053 linear = max(linear, min_t(int, len, hard_header_len)); 3054 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, 3055 msg->msg_flags & MSG_DONTWAIT, &err); 3056 if (skb == NULL) 3057 goto out_unlock; 3058 3059 skb_reset_network_header(skb); 3060 3061 err = -EINVAL; 3062 if (sock->type == SOCK_DGRAM) { 3063 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 3064 if (unlikely(offset < 0)) 3065 goto out_free; 3066 } else if (reserve) { 3067 skb_reserve(skb, -reserve); 3068 if (len < reserve + sizeof(struct ipv6hdr) && 3069 dev->min_header_len != hard_header_len) 3070 skb_reset_network_header(skb); 3071 } 3072 3073 /* Returns -EFAULT on error */ 3074 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 3075 if (err) 3076 goto out_free; 3077 3078 if ((sock->type == SOCK_RAW && 3079 !dev_validate_header(dev, skb->data, len)) || !skb->len) { 3080 err = -EINVAL; 3081 goto out_free; 3082 } 3083 3084 skb_setup_tx_timestamp(skb, &sockc); 3085 3086 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 3087 !packet_extra_vlan_len_allowed(dev, skb)) { 3088 err = -EMSGSIZE; 3089 goto out_free; 3090 } 3091 3092 skb->protocol = proto; 3093 skb->dev = dev; 3094 skb->priority = sockc.priority; 3095 skb->mark = sockc.mark; 3096 skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid); 3097 3098 if (unlikely(extra_len == 4)) 3099 skb->no_fcs = 1; 3100 3101 packet_parse_headers(skb, sock); 3102 3103 if (vnet_hdr_sz) { 3104 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 3105 if (err) 3106 goto out_free; 3107 len += vnet_hdr_sz; 3108 virtio_net_hdr_set_proto(skb, &vnet_hdr); 3109 } 3110 3111 err = packet_xmit(po, skb); 3112 3113 if (unlikely(err != 0)) { 3114 if (err > 0) 3115 err = net_xmit_errno(err); 3116 if (err) 3117 goto out_unlock; 3118 } 3119 3120 dev_put(dev); 3121 3122 return len; 3123 3124 out_free: 3125 kfree_skb(skb); 3126 out_unlock: 3127 dev_put(dev); 3128 out: 3129 return err; 3130 } 3131 3132 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 3133 { 3134 struct sock *sk = sock->sk; 3135 struct packet_sock *po = pkt_sk(sk); 3136 3137 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy. 3138 * tpacket_snd() will redo the check safely. 3139 */ 3140 if (data_race(po->tx_ring.pg_vec)) 3141 return tpacket_snd(po, msg); 3142 3143 return packet_snd(sock, msg, len); 3144 } 3145 3146 /* 3147 * Close a PACKET socket. This is fairly simple. We immediately go 3148 * to 'closed' state and remove our protocol entry in the device list. 3149 */ 3150 3151 static int packet_release(struct socket *sock) 3152 { 3153 struct sock *sk = sock->sk; 3154 struct packet_sock *po; 3155 struct packet_fanout *f; 3156 struct net *net; 3157 union tpacket_req_u req_u; 3158 3159 if (!sk) 3160 return 0; 3161 3162 net = sock_net(sk); 3163 po = pkt_sk(sk); 3164 3165 mutex_lock(&net->packet.sklist_lock); 3166 sk_del_node_init_rcu(sk); 3167 mutex_unlock(&net->packet.sklist_lock); 3168 3169 sock_prot_inuse_add(net, sk->sk_prot, -1); 3170 3171 spin_lock(&po->bind_lock); 3172 unregister_prot_hook(sk, false); 3173 WRITE_ONCE(po->num, 0); 3174 packet_cached_dev_reset(po); 3175 3176 if (po->prot_hook.dev) { 3177 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker); 3178 po->prot_hook.dev = NULL; 3179 } 3180 spin_unlock(&po->bind_lock); 3181 3182 packet_flush_mclist(sk); 3183 3184 lock_sock(sk); 3185 if (po->rx_ring.pg_vec) { 3186 memset(&req_u, 0, sizeof(req_u)); 3187 packet_set_ring(sk, &req_u, 1, 0); 3188 } 3189 3190 if (po->tx_ring.pg_vec) { 3191 memset(&req_u, 0, sizeof(req_u)); 3192 packet_set_ring(sk, &req_u, 1, 1); 3193 } 3194 release_sock(sk); 3195 3196 f = fanout_release(sk); 3197 3198 synchronize_net(); 3199 3200 kfree(po->rollover); 3201 if (f) { 3202 fanout_release_data(f); 3203 kvfree(f); 3204 } 3205 /* 3206 * Now the socket is dead. No more input will appear. 3207 */ 3208 sock_orphan(sk); 3209 sock->sk = NULL; 3210 3211 /* Purge queues */ 3212 3213 skb_queue_purge(&sk->sk_receive_queue); 3214 3215 sock_put(sk); 3216 return 0; 3217 } 3218 3219 /* 3220 * Attach a packet hook. 3221 */ 3222 3223 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3224 __be16 proto) 3225 { 3226 struct packet_sock *po = pkt_sk(sk); 3227 struct net_device *dev = NULL; 3228 bool unlisted = false; 3229 bool need_rehook; 3230 int ret = 0; 3231 3232 lock_sock(sk); 3233 spin_lock(&po->bind_lock); 3234 if (!proto) 3235 proto = po->num; 3236 3237 rcu_read_lock(); 3238 3239 if (po->fanout) { 3240 ret = -EINVAL; 3241 goto out_unlock; 3242 } 3243 3244 if (name) { 3245 dev = dev_get_by_name_rcu(sock_net(sk), name); 3246 if (!dev) { 3247 ret = -ENODEV; 3248 goto out_unlock; 3249 } 3250 } else if (ifindex) { 3251 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3252 if (!dev) { 3253 ret = -ENODEV; 3254 goto out_unlock; 3255 } 3256 } 3257 3258 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev; 3259 3260 if (need_rehook) { 3261 dev_hold(dev); 3262 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 3263 rcu_read_unlock(); 3264 /* prevents packet_notifier() from calling 3265 * register_prot_hook() 3266 */ 3267 WRITE_ONCE(po->num, 0); 3268 __unregister_prot_hook(sk, true); 3269 rcu_read_lock(); 3270 if (dev) 3271 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3272 dev->ifindex); 3273 } 3274 3275 BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING)); 3276 WRITE_ONCE(po->num, proto); 3277 po->prot_hook.type = proto; 3278 3279 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker); 3280 3281 if (unlikely(unlisted)) { 3282 po->prot_hook.dev = NULL; 3283 WRITE_ONCE(po->ifindex, -1); 3284 packet_cached_dev_reset(po); 3285 } else { 3286 netdev_hold(dev, &po->prot_hook.dev_tracker, 3287 GFP_ATOMIC); 3288 po->prot_hook.dev = dev; 3289 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0); 3290 packet_cached_dev_assign(po, dev); 3291 } 3292 dev_put(dev); 3293 } 3294 3295 if (proto == 0 || !need_rehook) 3296 goto out_unlock; 3297 3298 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3299 register_prot_hook(sk); 3300 } else { 3301 sk->sk_err = ENETDOWN; 3302 if (!sock_flag(sk, SOCK_DEAD)) 3303 sk_error_report(sk); 3304 } 3305 3306 out_unlock: 3307 rcu_read_unlock(); 3308 spin_unlock(&po->bind_lock); 3309 release_sock(sk); 3310 return ret; 3311 } 3312 3313 /* 3314 * Bind a packet socket to a device 3315 */ 3316 3317 static int packet_bind_spkt(struct socket *sock, struct sockaddr_unsized *uaddr, 3318 int addr_len) 3319 { 3320 struct sock *sk = sock->sk; 3321 struct sockaddr *sa = (struct sockaddr *)uaddr; 3322 char name[sizeof(sa->sa_data) + 1]; 3323 3324 /* 3325 * Check legality 3326 */ 3327 3328 if (addr_len != sizeof(struct sockaddr)) 3329 return -EINVAL; 3330 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be 3331 * zero-terminated. 3332 */ 3333 memcpy(name, sa->sa_data, sizeof(sa->sa_data)); 3334 name[sizeof(sa->sa_data)] = 0; 3335 3336 return packet_do_bind(sk, name, 0, 0); 3337 } 3338 3339 static int packet_bind(struct socket *sock, struct sockaddr_unsized *uaddr, int addr_len) 3340 { 3341 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3342 struct sock *sk = sock->sk; 3343 3344 /* 3345 * Check legality 3346 */ 3347 3348 if (addr_len < sizeof(struct sockaddr_ll)) 3349 return -EINVAL; 3350 if (sll->sll_family != AF_PACKET) 3351 return -EINVAL; 3352 3353 return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol); 3354 } 3355 3356 static struct proto packet_proto = { 3357 .name = "PACKET", 3358 .owner = THIS_MODULE, 3359 .obj_size = sizeof(struct packet_sock), 3360 }; 3361 3362 /* 3363 * Create a packet of type SOCK_PACKET. 3364 */ 3365 3366 static int packet_create(struct net *net, struct socket *sock, int protocol, 3367 int kern) 3368 { 3369 struct sock *sk; 3370 struct packet_sock *po; 3371 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3372 int err; 3373 3374 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3375 return -EPERM; 3376 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3377 sock->type != SOCK_PACKET) 3378 return -ESOCKTNOSUPPORT; 3379 3380 sock->state = SS_UNCONNECTED; 3381 3382 err = -ENOBUFS; 3383 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3384 if (sk == NULL) 3385 goto out; 3386 3387 sock->ops = &packet_ops; 3388 if (sock->type == SOCK_PACKET) 3389 sock->ops = &packet_ops_spkt; 3390 3391 po = pkt_sk(sk); 3392 err = packet_alloc_pending(po); 3393 if (err) 3394 goto out_sk_free; 3395 3396 sock_init_data(sock, sk); 3397 3398 init_completion(&po->skb_completion); 3399 sk->sk_family = PF_PACKET; 3400 po->num = proto; 3401 3402 packet_cached_dev_reset(po); 3403 3404 sk->sk_destruct = packet_sock_destruct; 3405 3406 /* 3407 * Attach a protocol block 3408 */ 3409 3410 spin_lock_init(&po->bind_lock); 3411 mutex_init(&po->pg_vec_lock); 3412 po->rollover = NULL; 3413 po->prot_hook.func = packet_rcv; 3414 3415 if (sock->type == SOCK_PACKET) 3416 po->prot_hook.func = packet_rcv_spkt; 3417 3418 po->prot_hook.af_packet_priv = sk; 3419 po->prot_hook.af_packet_net = sock_net(sk); 3420 3421 if (proto) { 3422 po->prot_hook.type = proto; 3423 __register_prot_hook(sk); 3424 } 3425 3426 mutex_lock(&net->packet.sklist_lock); 3427 sk_add_node_tail_rcu(sk, &net->packet.sklist); 3428 mutex_unlock(&net->packet.sklist_lock); 3429 3430 sock_prot_inuse_add(net, &packet_proto, 1); 3431 3432 return 0; 3433 out_sk_free: 3434 sk_free(sk); 3435 out: 3436 return err; 3437 } 3438 3439 /* 3440 * Pull a packet from our receive queue and hand it to the user. 3441 * If necessary we block. 3442 */ 3443 3444 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3445 int flags) 3446 { 3447 struct sock *sk = sock->sk; 3448 struct sk_buff *skb; 3449 int copied, err; 3450 int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz); 3451 unsigned int origlen = 0; 3452 3453 err = -EINVAL; 3454 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3455 goto out; 3456 3457 #if 0 3458 /* What error should we return now? EUNATTACH? */ 3459 if (pkt_sk(sk)->ifindex < 0) 3460 return -ENODEV; 3461 #endif 3462 3463 if (flags & MSG_ERRQUEUE) { 3464 err = sock_recv_errqueue(sk, msg, len, 3465 SOL_PACKET, PACKET_TX_TIMESTAMP); 3466 goto out; 3467 } 3468 3469 /* 3470 * Call the generic datagram receiver. This handles all sorts 3471 * of horrible races and re-entrancy so we can forget about it 3472 * in the protocol layers. 3473 * 3474 * Now it will return ENETDOWN, if device have just gone down, 3475 * but then it will block. 3476 */ 3477 3478 skb = skb_recv_datagram(sk, flags, &err); 3479 3480 /* 3481 * An error occurred so return it. Because skb_recv_datagram() 3482 * handles the blocking we don't see and worry about blocking 3483 * retries. 3484 */ 3485 3486 if (skb == NULL) 3487 goto out; 3488 3489 packet_rcv_try_clear_pressure(pkt_sk(sk)); 3490 3491 if (vnet_hdr_len) { 3492 err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len); 3493 if (err) 3494 goto out_free; 3495 } 3496 3497 /* You lose any data beyond the buffer you gave. If it worries 3498 * a user program they can ask the device for its MTU 3499 * anyway. 3500 */ 3501 copied = skb->len; 3502 if (copied > len) { 3503 copied = len; 3504 msg->msg_flags |= MSG_TRUNC; 3505 } 3506 3507 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3508 if (err) 3509 goto out_free; 3510 3511 if (sock->type != SOCK_PACKET) { 3512 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3513 3514 /* Original length was stored in sockaddr_ll fields */ 3515 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3516 sll->sll_family = AF_PACKET; 3517 sll->sll_protocol = (sock->type == SOCK_DGRAM) ? 3518 vlan_get_protocol_dgram(skb) : skb->protocol; 3519 } 3520 3521 sock_recv_cmsgs(msg, sk, skb); 3522 3523 if (msg->msg_name) { 3524 const size_t max_len = min(sizeof(skb->cb), 3525 sizeof(struct sockaddr_storage)); 3526 int copy_len; 3527 3528 /* If the address length field is there to be filled 3529 * in, we fill it in now. 3530 */ 3531 if (sock->type == SOCK_PACKET) { 3532 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3533 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3534 copy_len = msg->msg_namelen; 3535 } else { 3536 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3537 3538 msg->msg_namelen = sll->sll_halen + 3539 offsetof(struct sockaddr_ll, sll_addr); 3540 copy_len = msg->msg_namelen; 3541 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { 3542 memset(msg->msg_name + 3543 offsetof(struct sockaddr_ll, sll_addr), 3544 0, sizeof(sll->sll_addr)); 3545 msg->msg_namelen = sizeof(struct sockaddr_ll); 3546 } 3547 } 3548 if (WARN_ON_ONCE(copy_len > max_len)) { 3549 copy_len = max_len; 3550 msg->msg_namelen = copy_len; 3551 } 3552 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); 3553 } 3554 3555 if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) { 3556 struct tpacket_auxdata aux; 3557 3558 aux.tp_status = TP_STATUS_USER; 3559 if (skb->ip_summed == CHECKSUM_PARTIAL) 3560 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3561 else if (skb->pkt_type != PACKET_OUTGOING && 3562 skb_csum_unnecessary(skb)) 3563 aux.tp_status |= TP_STATUS_CSUM_VALID; 3564 if (skb_is_gso(skb) && skb_is_gso_tcp(skb)) 3565 aux.tp_status |= TP_STATUS_GSO_TCP; 3566 3567 aux.tp_len = origlen; 3568 aux.tp_snaplen = skb->len; 3569 aux.tp_mac = 0; 3570 aux.tp_net = skb_network_offset(skb); 3571 if (skb_vlan_tag_present(skb)) { 3572 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3573 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3574 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3575 } else if (unlikely(sock->type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) { 3576 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3577 struct net_device *dev; 3578 3579 rcu_read_lock(); 3580 dev = dev_get_by_index_rcu(sock_net(sk), sll->sll_ifindex); 3581 if (dev) { 3582 aux.tp_vlan_tci = vlan_get_tci(skb, dev); 3583 aux.tp_vlan_tpid = ntohs(skb->protocol); 3584 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3585 } else { 3586 aux.tp_vlan_tci = 0; 3587 aux.tp_vlan_tpid = 0; 3588 } 3589 rcu_read_unlock(); 3590 } else { 3591 aux.tp_vlan_tci = 0; 3592 aux.tp_vlan_tpid = 0; 3593 } 3594 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3595 } 3596 3597 /* 3598 * Free or return the buffer as appropriate. Again this 3599 * hides all the races and re-entrancy issues from us. 3600 */ 3601 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3602 3603 out_free: 3604 skb_free_datagram(sk, skb); 3605 out: 3606 return err; 3607 } 3608 3609 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3610 int peer) 3611 { 3612 struct net_device *dev; 3613 struct sock *sk = sock->sk; 3614 3615 if (peer) 3616 return -EOPNOTSUPP; 3617 3618 uaddr->sa_family = AF_PACKET; 3619 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3620 rcu_read_lock(); 3621 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex)); 3622 if (dev) 3623 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3624 rcu_read_unlock(); 3625 3626 return sizeof(*uaddr); 3627 } 3628 3629 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3630 int peer) 3631 { 3632 struct net_device *dev; 3633 struct sock *sk = sock->sk; 3634 struct packet_sock *po = pkt_sk(sk); 3635 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3636 int ifindex; 3637 3638 if (peer) 3639 return -EOPNOTSUPP; 3640 3641 ifindex = READ_ONCE(po->ifindex); 3642 sll->sll_family = AF_PACKET; 3643 sll->sll_ifindex = ifindex; 3644 sll->sll_protocol = READ_ONCE(po->num); 3645 sll->sll_pkttype = 0; 3646 rcu_read_lock(); 3647 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3648 if (dev) { 3649 sll->sll_hatype = dev->type; 3650 sll->sll_halen = dev->addr_len; 3651 3652 /* Let __fortify_memcpy_chk() know the actual buffer size. */ 3653 memcpy(((struct sockaddr_storage *)sll)->__data + 3654 offsetof(struct sockaddr_ll, sll_addr) - 3655 offsetofend(struct sockaddr_ll, sll_family), 3656 dev->dev_addr, dev->addr_len); 3657 } else { 3658 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3659 sll->sll_halen = 0; 3660 } 3661 rcu_read_unlock(); 3662 3663 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3664 } 3665 3666 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3667 int what) 3668 { 3669 switch (i->type) { 3670 case PACKET_MR_MULTICAST: 3671 if (i->alen != dev->addr_len) 3672 return -EINVAL; 3673 if (what > 0) 3674 return dev_mc_add(dev, i->addr); 3675 else 3676 return dev_mc_del(dev, i->addr); 3677 break; 3678 case PACKET_MR_PROMISC: 3679 return dev_set_promiscuity(dev, what); 3680 case PACKET_MR_ALLMULTI: 3681 return dev_set_allmulti(dev, what); 3682 case PACKET_MR_UNICAST: 3683 if (i->alen != dev->addr_len) 3684 return -EINVAL; 3685 if (what > 0) 3686 return dev_uc_add(dev, i->addr); 3687 else 3688 return dev_uc_del(dev, i->addr); 3689 break; 3690 default: 3691 break; 3692 } 3693 return 0; 3694 } 3695 3696 static void packet_dev_mclist_delete(struct net_device *dev, 3697 struct packet_mclist **mlp, 3698 struct list_head *list) 3699 { 3700 struct packet_mclist *ml; 3701 3702 while ((ml = *mlp) != NULL) { 3703 if (ml->ifindex == dev->ifindex) { 3704 list_add(&ml->remove_list, list); 3705 *mlp = ml->next; 3706 } else 3707 mlp = &ml->next; 3708 } 3709 } 3710 3711 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3712 { 3713 struct packet_sock *po = pkt_sk(sk); 3714 struct packet_mclist *ml, *i; 3715 struct net_device *dev; 3716 int err; 3717 3718 rtnl_lock(); 3719 3720 err = -ENODEV; 3721 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3722 if (!dev) 3723 goto done; 3724 3725 err = -EINVAL; 3726 if (mreq->mr_alen > dev->addr_len) 3727 goto done; 3728 3729 err = -ENOBUFS; 3730 i = kmalloc_obj(*i); 3731 if (i == NULL) 3732 goto done; 3733 3734 err = 0; 3735 for (ml = po->mclist; ml; ml = ml->next) { 3736 if (ml->ifindex == mreq->mr_ifindex && 3737 ml->type == mreq->mr_type && 3738 ml->alen == mreq->mr_alen && 3739 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3740 ml->count++; 3741 /* Free the new element ... */ 3742 kfree(i); 3743 goto done; 3744 } 3745 } 3746 3747 i->type = mreq->mr_type; 3748 i->ifindex = mreq->mr_ifindex; 3749 i->alen = mreq->mr_alen; 3750 memcpy(i->addr, mreq->mr_address, i->alen); 3751 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3752 i->count = 1; 3753 INIT_LIST_HEAD(&i->remove_list); 3754 i->next = po->mclist; 3755 po->mclist = i; 3756 err = packet_dev_mc(dev, i, 1); 3757 if (err) { 3758 po->mclist = i->next; 3759 kfree(i); 3760 } 3761 3762 done: 3763 rtnl_unlock(); 3764 return err; 3765 } 3766 3767 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3768 { 3769 struct packet_mclist *ml, **mlp; 3770 3771 rtnl_lock(); 3772 3773 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3774 if (ml->ifindex == mreq->mr_ifindex && 3775 ml->type == mreq->mr_type && 3776 ml->alen == mreq->mr_alen && 3777 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3778 if (--ml->count == 0) { 3779 struct net_device *dev; 3780 *mlp = ml->next; 3781 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3782 if (dev) 3783 packet_dev_mc(dev, ml, -1); 3784 kfree(ml); 3785 } 3786 break; 3787 } 3788 } 3789 rtnl_unlock(); 3790 return 0; 3791 } 3792 3793 static void packet_flush_mclist(struct sock *sk) 3794 { 3795 struct packet_sock *po = pkt_sk(sk); 3796 struct packet_mclist *ml; 3797 3798 if (!po->mclist) 3799 return; 3800 3801 rtnl_lock(); 3802 while ((ml = po->mclist) != NULL) { 3803 struct net_device *dev; 3804 3805 po->mclist = ml->next; 3806 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3807 if (dev != NULL) 3808 packet_dev_mc(dev, ml, -1); 3809 kfree(ml); 3810 } 3811 rtnl_unlock(); 3812 } 3813 3814 static int 3815 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval, 3816 unsigned int optlen) 3817 { 3818 struct sock *sk = sock->sk; 3819 struct packet_sock *po = pkt_sk(sk); 3820 int ret; 3821 3822 if (level != SOL_PACKET) 3823 return -ENOPROTOOPT; 3824 3825 switch (optname) { 3826 case PACKET_ADD_MEMBERSHIP: 3827 case PACKET_DROP_MEMBERSHIP: 3828 { 3829 struct packet_mreq_max mreq; 3830 int len = optlen; 3831 memset(&mreq, 0, sizeof(mreq)); 3832 if (len < sizeof(struct packet_mreq)) 3833 return -EINVAL; 3834 if (len > sizeof(mreq)) 3835 len = sizeof(mreq); 3836 if (copy_from_sockptr(&mreq, optval, len)) 3837 return -EFAULT; 3838 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3839 return -EINVAL; 3840 if (optname == PACKET_ADD_MEMBERSHIP) 3841 ret = packet_mc_add(sk, &mreq); 3842 else 3843 ret = packet_mc_drop(sk, &mreq); 3844 return ret; 3845 } 3846 3847 case PACKET_RX_RING: 3848 case PACKET_TX_RING: 3849 { 3850 union tpacket_req_u req_u; 3851 3852 ret = -EINVAL; 3853 lock_sock(sk); 3854 switch (po->tp_version) { 3855 case TPACKET_V1: 3856 case TPACKET_V2: 3857 if (optlen < sizeof(req_u.req)) 3858 break; 3859 ret = copy_from_sockptr(&req_u.req, optval, 3860 sizeof(req_u.req)) ? 3861 -EINVAL : 0; 3862 break; 3863 case TPACKET_V3: 3864 default: 3865 if (optlen < sizeof(req_u.req3)) 3866 break; 3867 ret = copy_from_sockptr(&req_u.req3, optval, 3868 sizeof(req_u.req3)) ? 3869 -EINVAL : 0; 3870 break; 3871 } 3872 if (!ret) 3873 ret = packet_set_ring(sk, &req_u, 0, 3874 optname == PACKET_TX_RING); 3875 release_sock(sk); 3876 return ret; 3877 } 3878 case PACKET_COPY_THRESH: 3879 { 3880 int val; 3881 3882 if (optlen != sizeof(val)) 3883 return -EINVAL; 3884 if (copy_from_sockptr(&val, optval, sizeof(val))) 3885 return -EFAULT; 3886 3887 WRITE_ONCE(pkt_sk(sk)->copy_thresh, val); 3888 return 0; 3889 } 3890 case PACKET_VERSION: 3891 { 3892 int val; 3893 3894 if (optlen != sizeof(val)) 3895 return -EINVAL; 3896 if (copy_from_sockptr(&val, optval, sizeof(val))) 3897 return -EFAULT; 3898 switch (val) { 3899 case TPACKET_V1: 3900 case TPACKET_V2: 3901 case TPACKET_V3: 3902 break; 3903 default: 3904 return -EINVAL; 3905 } 3906 lock_sock(sk); 3907 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3908 ret = -EBUSY; 3909 } else { 3910 po->tp_version = val; 3911 ret = 0; 3912 } 3913 release_sock(sk); 3914 return ret; 3915 } 3916 case PACKET_RESERVE: 3917 { 3918 unsigned int val; 3919 3920 if (optlen != sizeof(val)) 3921 return -EINVAL; 3922 if (copy_from_sockptr(&val, optval, sizeof(val))) 3923 return -EFAULT; 3924 if (val > INT_MAX) 3925 return -EINVAL; 3926 lock_sock(sk); 3927 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3928 ret = -EBUSY; 3929 } else { 3930 po->tp_reserve = val; 3931 ret = 0; 3932 } 3933 release_sock(sk); 3934 return ret; 3935 } 3936 case PACKET_LOSS: 3937 { 3938 unsigned int val; 3939 3940 if (optlen != sizeof(val)) 3941 return -EINVAL; 3942 if (copy_from_sockptr(&val, optval, sizeof(val))) 3943 return -EFAULT; 3944 3945 lock_sock(sk); 3946 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3947 ret = -EBUSY; 3948 } else { 3949 packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val); 3950 ret = 0; 3951 } 3952 release_sock(sk); 3953 return ret; 3954 } 3955 case PACKET_AUXDATA: 3956 { 3957 int val; 3958 3959 if (optlen < sizeof(val)) 3960 return -EINVAL; 3961 if (copy_from_sockptr(&val, optval, sizeof(val))) 3962 return -EFAULT; 3963 3964 packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val); 3965 return 0; 3966 } 3967 case PACKET_ORIGDEV: 3968 { 3969 int val; 3970 3971 if (optlen < sizeof(val)) 3972 return -EINVAL; 3973 if (copy_from_sockptr(&val, optval, sizeof(val))) 3974 return -EFAULT; 3975 3976 packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val); 3977 return 0; 3978 } 3979 case PACKET_VNET_HDR: 3980 case PACKET_VNET_HDR_SZ: 3981 { 3982 int val, hdr_len; 3983 3984 if (sock->type != SOCK_RAW) 3985 return -EINVAL; 3986 if (optlen < sizeof(val)) 3987 return -EINVAL; 3988 if (copy_from_sockptr(&val, optval, sizeof(val))) 3989 return -EFAULT; 3990 3991 if (optname == PACKET_VNET_HDR_SZ) { 3992 if (val && val != sizeof(struct virtio_net_hdr) && 3993 val != sizeof(struct virtio_net_hdr_mrg_rxbuf)) 3994 return -EINVAL; 3995 hdr_len = val; 3996 } else { 3997 hdr_len = val ? sizeof(struct virtio_net_hdr) : 0; 3998 } 3999 lock_sock(sk); 4000 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 4001 ret = -EBUSY; 4002 } else { 4003 WRITE_ONCE(po->vnet_hdr_sz, hdr_len); 4004 ret = 0; 4005 } 4006 release_sock(sk); 4007 return ret; 4008 } 4009 case PACKET_TIMESTAMP: 4010 { 4011 int val; 4012 4013 if (optlen != sizeof(val)) 4014 return -EINVAL; 4015 if (copy_from_sockptr(&val, optval, sizeof(val))) 4016 return -EFAULT; 4017 4018 WRITE_ONCE(po->tp_tstamp, val); 4019 return 0; 4020 } 4021 case PACKET_FANOUT: 4022 { 4023 struct fanout_args args = { 0 }; 4024 4025 if (optlen != sizeof(int) && optlen != sizeof(args)) 4026 return -EINVAL; 4027 if (copy_from_sockptr(&args, optval, optlen)) 4028 return -EFAULT; 4029 4030 return fanout_add(sk, &args); 4031 } 4032 case PACKET_FANOUT_DATA: 4033 { 4034 /* Paired with the WRITE_ONCE() in fanout_add() */ 4035 if (!READ_ONCE(po->fanout)) 4036 return -EINVAL; 4037 4038 return fanout_set_data(po, optval, optlen); 4039 } 4040 case PACKET_IGNORE_OUTGOING: 4041 { 4042 int val; 4043 4044 if (optlen != sizeof(val)) 4045 return -EINVAL; 4046 if (copy_from_sockptr(&val, optval, sizeof(val))) 4047 return -EFAULT; 4048 if (val < 0 || val > 1) 4049 return -EINVAL; 4050 4051 WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val); 4052 return 0; 4053 } 4054 case PACKET_TX_HAS_OFF: 4055 { 4056 unsigned int val; 4057 4058 if (optlen != sizeof(val)) 4059 return -EINVAL; 4060 if (copy_from_sockptr(&val, optval, sizeof(val))) 4061 return -EFAULT; 4062 4063 lock_sock(sk); 4064 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec) 4065 packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val); 4066 4067 release_sock(sk); 4068 return 0; 4069 } 4070 case PACKET_QDISC_BYPASS: 4071 { 4072 int val; 4073 4074 if (optlen != sizeof(val)) 4075 return -EINVAL; 4076 if (copy_from_sockptr(&val, optval, sizeof(val))) 4077 return -EFAULT; 4078 4079 packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val); 4080 return 0; 4081 } 4082 default: 4083 return -ENOPROTOOPT; 4084 } 4085 } 4086 4087 static int packet_getsockopt(struct socket *sock, int level, int optname, 4088 sockopt_t *opt) 4089 { 4090 int len; 4091 int val, lv = sizeof(val); 4092 struct sock *sk = sock->sk; 4093 struct packet_sock *po = pkt_sk(sk); 4094 void *data = &val; 4095 union tpacket_stats_u st; 4096 struct tpacket_rollover_stats rstats; 4097 int drops; 4098 4099 if (level != SOL_PACKET) 4100 return -ENOPROTOOPT; 4101 4102 len = opt->optlen; 4103 4104 if (len < 0) 4105 return -EINVAL; 4106 4107 switch (optname) { 4108 case PACKET_STATISTICS: 4109 spin_lock_bh(&sk->sk_receive_queue.lock); 4110 memcpy(&st, &po->stats, sizeof(st)); 4111 memset(&po->stats, 0, sizeof(po->stats)); 4112 spin_unlock_bh(&sk->sk_receive_queue.lock); 4113 drops = atomic_xchg(&po->tp_drops, 0); 4114 4115 if (po->tp_version == TPACKET_V3) { 4116 lv = sizeof(struct tpacket_stats_v3); 4117 st.stats3.tp_drops = drops; 4118 st.stats3.tp_packets += drops; 4119 data = &st.stats3; 4120 } else { 4121 lv = sizeof(struct tpacket_stats); 4122 st.stats1.tp_drops = drops; 4123 st.stats1.tp_packets += drops; 4124 data = &st.stats1; 4125 } 4126 4127 break; 4128 case PACKET_AUXDATA: 4129 val = packet_sock_flag(po, PACKET_SOCK_AUXDATA); 4130 break; 4131 case PACKET_ORIGDEV: 4132 val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV); 4133 break; 4134 case PACKET_VNET_HDR: 4135 val = !!READ_ONCE(po->vnet_hdr_sz); 4136 break; 4137 case PACKET_VNET_HDR_SZ: 4138 val = READ_ONCE(po->vnet_hdr_sz); 4139 break; 4140 case PACKET_COPY_THRESH: 4141 val = READ_ONCE(pkt_sk(sk)->copy_thresh); 4142 break; 4143 case PACKET_VERSION: 4144 val = po->tp_version; 4145 break; 4146 case PACKET_HDRLEN: 4147 if (len > sizeof(int)) 4148 len = sizeof(int); 4149 if (len < sizeof(int)) 4150 return -EINVAL; 4151 if (copy_from_iter(&val, len, &opt->iter_in) != len) 4152 return -EFAULT; 4153 switch (val) { 4154 case TPACKET_V1: 4155 val = sizeof(struct tpacket_hdr); 4156 break; 4157 case TPACKET_V2: 4158 val = sizeof(struct tpacket2_hdr); 4159 break; 4160 case TPACKET_V3: 4161 val = sizeof(struct tpacket3_hdr); 4162 break; 4163 default: 4164 return -EINVAL; 4165 } 4166 break; 4167 case PACKET_RESERVE: 4168 val = po->tp_reserve; 4169 break; 4170 case PACKET_LOSS: 4171 val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS); 4172 break; 4173 case PACKET_TIMESTAMP: 4174 val = READ_ONCE(po->tp_tstamp); 4175 break; 4176 case PACKET_FANOUT: 4177 val = (po->fanout ? 4178 ((u32)po->fanout->id | 4179 ((u32)po->fanout->type << 16) | 4180 ((u32)po->fanout->flags << 24)) : 4181 0); 4182 break; 4183 case PACKET_IGNORE_OUTGOING: 4184 val = READ_ONCE(po->prot_hook.ignore_outgoing); 4185 break; 4186 case PACKET_ROLLOVER_STATS: 4187 if (!po->rollover) 4188 return -EINVAL; 4189 rstats.tp_all = atomic_long_read(&po->rollover->num); 4190 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 4191 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 4192 data = &rstats; 4193 lv = sizeof(rstats); 4194 break; 4195 case PACKET_TX_HAS_OFF: 4196 val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF); 4197 break; 4198 case PACKET_QDISC_BYPASS: 4199 val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS); 4200 break; 4201 default: 4202 return -ENOPROTOOPT; 4203 } 4204 4205 if (len > lv) 4206 len = lv; 4207 opt->optlen = len; 4208 if (copy_to_iter(data, len, &opt->iter_out) != len) 4209 return -EFAULT; 4210 return 0; 4211 } 4212 4213 static int packet_notifier(struct notifier_block *this, 4214 unsigned long msg, void *ptr) 4215 { 4216 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4217 struct net *net = dev_net(dev); 4218 struct packet_mclist *ml, *tmp; 4219 LIST_HEAD(mclist); 4220 struct sock *sk; 4221 4222 rcu_read_lock(); 4223 sk_for_each_rcu(sk, &net->packet.sklist) { 4224 struct packet_sock *po = pkt_sk(sk); 4225 4226 switch (msg) { 4227 case NETDEV_UNREGISTER: 4228 if (po->mclist) 4229 packet_dev_mclist_delete(dev, &po->mclist, 4230 &mclist); 4231 fallthrough; 4232 4233 case NETDEV_DOWN: 4234 if (dev->ifindex == po->ifindex) { 4235 spin_lock(&po->bind_lock); 4236 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 4237 __unregister_prot_hook(sk, false); 4238 sk->sk_err = ENETDOWN; 4239 if (!sock_flag(sk, SOCK_DEAD)) 4240 sk_error_report(sk); 4241 } 4242 if (msg == NETDEV_UNREGISTER) { 4243 packet_cached_dev_reset(po); 4244 WRITE_ONCE(po->ifindex, -1); 4245 netdev_put(po->prot_hook.dev, 4246 &po->prot_hook.dev_tracker); 4247 po->prot_hook.dev = NULL; 4248 } 4249 spin_unlock(&po->bind_lock); 4250 } 4251 break; 4252 case NETDEV_UP: 4253 if (dev->ifindex == po->ifindex) { 4254 spin_lock(&po->bind_lock); 4255 if (po->num) 4256 register_prot_hook(sk); 4257 spin_unlock(&po->bind_lock); 4258 } 4259 break; 4260 } 4261 } 4262 rcu_read_unlock(); 4263 4264 /* packet_dev_mc might grab instance locks so can't run under rcu */ 4265 list_for_each_entry_safe(ml, tmp, &mclist, remove_list) { 4266 packet_dev_mc(dev, ml, -1); 4267 kfree(ml); 4268 } 4269 4270 return NOTIFY_DONE; 4271 } 4272 4273 4274 static int packet_ioctl(struct socket *sock, unsigned int cmd, 4275 unsigned long arg) 4276 { 4277 struct sock *sk = sock->sk; 4278 4279 switch (cmd) { 4280 case SIOCOUTQ: 4281 { 4282 int amount = sk_wmem_alloc_get(sk); 4283 4284 return put_user(amount, (int __user *)arg); 4285 } 4286 case SIOCINQ: 4287 { 4288 struct sk_buff *skb; 4289 int amount = 0; 4290 4291 spin_lock_bh(&sk->sk_receive_queue.lock); 4292 skb = skb_peek(&sk->sk_receive_queue); 4293 if (skb) 4294 amount = skb->len; 4295 spin_unlock_bh(&sk->sk_receive_queue.lock); 4296 return put_user(amount, (int __user *)arg); 4297 } 4298 #ifdef CONFIG_INET 4299 case SIOCADDRT: 4300 case SIOCDELRT: 4301 case SIOCDARP: 4302 case SIOCGARP: 4303 case SIOCSARP: 4304 case SIOCGIFADDR: 4305 case SIOCSIFADDR: 4306 case SIOCGIFBRDADDR: 4307 case SIOCSIFBRDADDR: 4308 case SIOCGIFNETMASK: 4309 case SIOCSIFNETMASK: 4310 case SIOCGIFDSTADDR: 4311 case SIOCSIFDSTADDR: 4312 case SIOCSIFFLAGS: 4313 return inet_dgram_ops.ioctl(sock, cmd, arg); 4314 #endif 4315 4316 default: 4317 return -ENOIOCTLCMD; 4318 } 4319 return 0; 4320 } 4321 4322 static __poll_t packet_poll(struct file *file, struct socket *sock, 4323 poll_table *wait) 4324 { 4325 struct sock *sk = sock->sk; 4326 struct packet_sock *po = pkt_sk(sk); 4327 __poll_t mask = datagram_poll(file, sock, wait); 4328 4329 spin_lock_bh(&sk->sk_receive_queue.lock); 4330 if (po->rx_ring.pg_vec) { 4331 if (!packet_previous_rx_frame(po, &po->rx_ring, 4332 TP_STATUS_KERNEL)) 4333 mask |= EPOLLIN | EPOLLRDNORM; 4334 } 4335 __packet_rcv_try_clear_pressure(po); 4336 spin_unlock_bh(&sk->sk_receive_queue.lock); 4337 spin_lock_bh(&sk->sk_write_queue.lock); 4338 if (po->tx_ring.pg_vec) { 4339 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4340 mask |= EPOLLOUT | EPOLLWRNORM; 4341 } 4342 spin_unlock_bh(&sk->sk_write_queue.lock); 4343 return mask; 4344 } 4345 4346 4347 /* Dirty? Well, I still did not learn better way to account 4348 * for user mmaps. 4349 */ 4350 4351 static void packet_mm_open(struct vm_area_struct *vma) 4352 { 4353 struct file *file = vma->vm_file; 4354 struct socket *sock = file->private_data; 4355 struct sock *sk = sock->sk; 4356 4357 if (sk) 4358 atomic_long_inc(&pkt_sk(sk)->mapped); 4359 } 4360 4361 static void packet_mm_close(struct vm_area_struct *vma) 4362 { 4363 struct file *file = vma->vm_file; 4364 struct socket *sock = file->private_data; 4365 struct sock *sk = sock->sk; 4366 4367 if (sk) 4368 atomic_long_dec(&pkt_sk(sk)->mapped); 4369 } 4370 4371 static const struct vm_operations_struct packet_mmap_ops = { 4372 .open = packet_mm_open, 4373 .close = packet_mm_close, 4374 }; 4375 4376 struct packet_pg_vec { 4377 struct packet_pg_vec_free *deferred; 4378 unsigned int order; 4379 unsigned int len; 4380 struct pgv pg_vec[] __counted_by(len); 4381 }; 4382 4383 struct packet_pg_vec_free { 4384 struct delayed_work work; 4385 struct sock *sk; 4386 struct packet_pg_vec *vec; 4387 }; 4388 4389 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4390 unsigned int len) 4391 { 4392 struct packet_pg_vec *vec; 4393 int i; 4394 4395 vec = container_of_const(pg_vec, struct packet_pg_vec, pg_vec[0]); 4396 for (i = 0; i < len; i++) { 4397 if (likely(pg_vec[i].buffer)) { 4398 if (is_vmalloc_addr(pg_vec[i].buffer)) 4399 vfree(pg_vec[i].buffer); 4400 else 4401 free_pages((unsigned long)pg_vec[i].buffer, 4402 order); 4403 pg_vec[i].buffer = NULL; 4404 } 4405 } 4406 kfree(vec->deferred); 4407 kfree(vec); 4408 } 4409 4410 static void packet_free_pg_vec_work(struct work_struct *work) 4411 { 4412 struct packet_pg_vec_free *deferred; 4413 struct packet_pg_vec *vec; 4414 struct sock *sk; 4415 4416 deferred = container_of_const(to_delayed_work(work), 4417 struct packet_pg_vec_free, work); 4418 vec = deferred->vec; 4419 sk = deferred->sk; 4420 if (sk_wmem_alloc_get(sk)) { 4421 queue_delayed_work(system_long_wq, &deferred->work, 1); 4422 return; 4423 } 4424 4425 free_pg_vec(vec->pg_vec, vec->order, vec->len); 4426 sock_put(sk); 4427 } 4428 4429 static void packet_free_tx_ring(struct sock *sk, struct pgv *pg_vec, 4430 unsigned int order, unsigned int len) 4431 { 4432 struct packet_pg_vec_free *deferred; 4433 struct packet_pg_vec *vec; 4434 4435 vec = container_of_const(pg_vec, struct packet_pg_vec, pg_vec[0]); 4436 deferred = vec->deferred; 4437 if (!deferred || !sk_wmem_alloc_get(sk)) { 4438 free_pg_vec(pg_vec, order, len); 4439 return; 4440 } 4441 4442 /* A detached ring's pending count can miss late skb destructors. */ 4443 deferred->sk = sk; 4444 sock_hold(sk); 4445 queue_delayed_work(system_long_wq, &deferred->work, 0); 4446 } 4447 4448 static char *alloc_one_pg_vec_page(unsigned long order) 4449 { 4450 char *buffer; 4451 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4452 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4453 4454 buffer = (char *) __get_free_pages(gfp_flags, order); 4455 if (buffer) 4456 return buffer; 4457 4458 /* __get_free_pages failed, fall back to vmalloc */ 4459 buffer = vzalloc(array_size((1 << order), PAGE_SIZE)); 4460 if (buffer) 4461 return buffer; 4462 4463 /* vmalloc failed, lets dig into swap here */ 4464 gfp_flags &= ~__GFP_NORETRY; 4465 buffer = (char *) __get_free_pages(gfp_flags, order); 4466 if (buffer) 4467 return buffer; 4468 4469 /* complete and utter failure */ 4470 return NULL; 4471 } 4472 4473 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order, bool tx_ring) 4474 { 4475 unsigned int block_nr = req->tp_block_nr; 4476 struct packet_pg_vec *vec; 4477 struct pgv *pg_vec; 4478 int i; 4479 4480 vec = kzalloc_flex(*vec, pg_vec, block_nr, GFP_KERNEL | __GFP_NOWARN); 4481 if (unlikely(!vec)) 4482 return NULL; 4483 vec->order = order; 4484 vec->len = block_nr; 4485 pg_vec = vec->pg_vec; 4486 4487 for (i = 0; i < block_nr; i++) { 4488 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4489 if (unlikely(!pg_vec[i].buffer)) 4490 goto out_free_pgvec; 4491 4492 if (tx_ring && !vec->deferred && 4493 is_vmalloc_addr(pg_vec[i].buffer)) { 4494 vec->deferred = kzalloc_obj(*vec->deferred, 4495 GFP_KERNEL | __GFP_NOWARN); 4496 if (!vec->deferred) 4497 goto out_free_pgvec; 4498 vec->deferred->vec = vec; 4499 INIT_DELAYED_WORK(&vec->deferred->work, 4500 packet_free_pg_vec_work); 4501 } 4502 } 4503 4504 out: 4505 return pg_vec; 4506 4507 out_free_pgvec: 4508 free_pg_vec(pg_vec, order, block_nr); 4509 pg_vec = NULL; 4510 goto out; 4511 } 4512 4513 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4514 int closing, int tx_ring) 4515 { 4516 struct pgv *pg_vec = NULL; 4517 struct packet_sock *po = pkt_sk(sk); 4518 unsigned long *rx_owner_map = NULL; 4519 int was_running, order = 0; 4520 struct packet_ring_buffer *rb; 4521 struct sk_buff_head *rb_queue; 4522 __be16 num; 4523 int err; 4524 /* Added to avoid minimal code churn */ 4525 struct tpacket_req *req = &req_u->req; 4526 4527 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4528 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4529 4530 err = -EBUSY; 4531 if (!closing) { 4532 if (atomic_long_read(&po->mapped)) 4533 goto out; 4534 if (packet_read_pending(rb)) 4535 goto out; 4536 } 4537 4538 if (req->tp_block_nr) { 4539 unsigned int min_frame_size; 4540 4541 /* Sanity tests and some calculations */ 4542 err = -EBUSY; 4543 if (unlikely(rb->pg_vec)) 4544 goto out; 4545 4546 switch (po->tp_version) { 4547 case TPACKET_V1: 4548 po->tp_hdrlen = TPACKET_HDRLEN; 4549 break; 4550 case TPACKET_V2: 4551 po->tp_hdrlen = TPACKET2_HDRLEN; 4552 break; 4553 case TPACKET_V3: 4554 po->tp_hdrlen = TPACKET3_HDRLEN; 4555 break; 4556 } 4557 4558 err = -EINVAL; 4559 if (unlikely((int)req->tp_block_size <= 0)) 4560 goto out; 4561 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4562 goto out; 4563 min_frame_size = po->tp_hdrlen + po->tp_reserve; 4564 if (po->tp_version >= TPACKET_V3 && 4565 req->tp_block_size < 4566 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) 4567 goto out; 4568 if (unlikely(req->tp_frame_size < min_frame_size)) 4569 goto out; 4570 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4571 goto out; 4572 4573 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4574 if (unlikely(rb->frames_per_block == 0)) 4575 goto out; 4576 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) 4577 goto out; 4578 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4579 req->tp_frame_nr)) 4580 goto out; 4581 4582 err = -ENOMEM; 4583 order = get_order(req->tp_block_size); 4584 pg_vec = alloc_pg_vec(req, order, tx_ring); 4585 if (unlikely(!pg_vec)) 4586 goto out; 4587 switch (po->tp_version) { 4588 case TPACKET_V3: 4589 /* Block transmit is not supported yet */ 4590 if (!tx_ring) { 4591 init_prb_bdqc(po, rb, pg_vec, req_u); 4592 } else { 4593 struct tpacket_req3 *req3 = &req_u->req3; 4594 4595 if (req3->tp_retire_blk_tov || 4596 req3->tp_sizeof_priv || 4597 req3->tp_feature_req_word) { 4598 err = -EINVAL; 4599 goto out_free_pg_vec; 4600 } 4601 } 4602 break; 4603 default: 4604 if (!tx_ring) { 4605 rx_owner_map = bitmap_alloc(req->tp_frame_nr, 4606 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO); 4607 if (!rx_owner_map) 4608 goto out_free_pg_vec; 4609 } 4610 break; 4611 } 4612 } 4613 /* Done */ 4614 else { 4615 err = -EINVAL; 4616 if (unlikely(req->tp_frame_nr)) 4617 goto out; 4618 } 4619 4620 4621 /* Detach socket from network */ 4622 spin_lock(&po->bind_lock); 4623 was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING); 4624 num = po->num; 4625 WRITE_ONCE(po->num, 0); 4626 if (was_running) 4627 __unregister_prot_hook(sk, false); 4628 4629 spin_unlock(&po->bind_lock); 4630 4631 synchronize_net(); 4632 4633 err = -EBUSY; 4634 mutex_lock(&po->pg_vec_lock); 4635 if (closing || atomic_long_read(&po->mapped) == 0) { 4636 if (tx_ring && !closing && packet_read_pending(rb)) 4637 goto out_unlock; 4638 4639 err = 0; 4640 spin_lock_bh(&rb_queue->lock); 4641 swap(rb->pg_vec, pg_vec); 4642 if (po->tp_version <= TPACKET_V2) 4643 swap(rb->rx_owner_map, rx_owner_map); 4644 rb->frame_max = (req->tp_frame_nr - 1); 4645 rb->head = 0; 4646 rb->frame_size = req->tp_frame_size; 4647 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4648 tpacket_rcv : packet_rcv; 4649 spin_unlock_bh(&rb_queue->lock); 4650 4651 swap(rb->pg_vec_order, order); 4652 swap(rb->pg_vec_len, req->tp_block_nr); 4653 4654 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4655 skb_queue_purge(rb_queue); 4656 if (atomic_long_read(&po->mapped)) 4657 pr_err("packet_mmap: vma is busy: %ld\n", 4658 atomic_long_read(&po->mapped)); 4659 } 4660 out_unlock: 4661 mutex_unlock(&po->pg_vec_lock); 4662 4663 spin_lock(&po->bind_lock); 4664 WRITE_ONCE(po->num, num); 4665 /* 4666 * NETDEV_UNREGISTER may have invalidated the binding while bind_lock 4667 * was dropped above. Do not re-add a fanout hook to a dead device. 4668 */ 4669 if (was_running && READ_ONCE(po->ifindex) != -1) 4670 register_prot_hook(sk); 4671 4672 spin_unlock(&po->bind_lock); 4673 if (pg_vec && (po->tp_version > TPACKET_V2)) { 4674 /* Because we don't support block-based V3 on tx-ring */ 4675 if (!tx_ring) 4676 prb_shutdown_retire_blk_timer(po, rb_queue); 4677 } 4678 4679 out_free_pg_vec: 4680 if (pg_vec) { 4681 bitmap_free(rx_owner_map); 4682 if (tx_ring && closing) 4683 packet_free_tx_ring(sk, pg_vec, order, req->tp_block_nr); 4684 else 4685 free_pg_vec(pg_vec, order, req->tp_block_nr); 4686 } 4687 out: 4688 return err; 4689 } 4690 4691 static int packet_mmap(struct file *file, struct socket *sock, 4692 struct vm_area_struct *vma) 4693 { 4694 struct sock *sk = sock->sk; 4695 struct packet_sock *po = pkt_sk(sk); 4696 unsigned long size, expected_size; 4697 struct packet_ring_buffer *rb; 4698 unsigned long start; 4699 int err = -EINVAL; 4700 int i; 4701 4702 if (vma->vm_pgoff) 4703 return -EINVAL; 4704 4705 mutex_lock(&po->pg_vec_lock); 4706 4707 expected_size = 0; 4708 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4709 if (rb->pg_vec) { 4710 expected_size += rb->pg_vec_len 4711 * rb->pg_vec_pages 4712 * PAGE_SIZE; 4713 } 4714 } 4715 4716 if (expected_size == 0) 4717 goto out; 4718 4719 size = vma->vm_end - vma->vm_start; 4720 if (size != expected_size) 4721 goto out; 4722 4723 start = vma->vm_start; 4724 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4725 if (rb->pg_vec == NULL) 4726 continue; 4727 4728 for (i = 0; i < rb->pg_vec_len; i++) { 4729 struct page *page; 4730 void *kaddr = rb->pg_vec[i].buffer; 4731 int pg_num; 4732 4733 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4734 page = pgv_to_page(kaddr); 4735 err = vm_insert_page(vma, start, page); 4736 if (unlikely(err)) 4737 goto out; 4738 start += PAGE_SIZE; 4739 kaddr += PAGE_SIZE; 4740 } 4741 } 4742 } 4743 4744 atomic_long_inc(&po->mapped); 4745 vma->vm_ops = &packet_mmap_ops; 4746 err = 0; 4747 4748 out: 4749 mutex_unlock(&po->pg_vec_lock); 4750 return err; 4751 } 4752 4753 static const struct proto_ops packet_ops_spkt = { 4754 .family = PF_PACKET, 4755 .owner = THIS_MODULE, 4756 .release = packet_release, 4757 .bind = packet_bind_spkt, 4758 .connect = sock_no_connect, 4759 .socketpair = sock_no_socketpair, 4760 .accept = sock_no_accept, 4761 .getname = packet_getname_spkt, 4762 .poll = datagram_poll, 4763 .ioctl = packet_ioctl, 4764 .gettstamp = sock_gettstamp, 4765 .listen = sock_no_listen, 4766 .shutdown = sock_no_shutdown, 4767 .sendmsg = packet_sendmsg_spkt, 4768 .recvmsg = packet_recvmsg, 4769 .mmap = sock_no_mmap, 4770 }; 4771 4772 static const struct proto_ops packet_ops = { 4773 .family = PF_PACKET, 4774 .owner = THIS_MODULE, 4775 .release = packet_release, 4776 .bind = packet_bind, 4777 .connect = sock_no_connect, 4778 .socketpair = sock_no_socketpair, 4779 .accept = sock_no_accept, 4780 .getname = packet_getname, 4781 .poll = packet_poll, 4782 .ioctl = packet_ioctl, 4783 .gettstamp = sock_gettstamp, 4784 .listen = sock_no_listen, 4785 .shutdown = sock_no_shutdown, 4786 .setsockopt = packet_setsockopt, 4787 .getsockopt_iter = packet_getsockopt, 4788 .sendmsg = packet_sendmsg, 4789 .recvmsg = packet_recvmsg, 4790 .mmap = packet_mmap, 4791 }; 4792 4793 static const struct net_proto_family packet_family_ops = { 4794 .family = PF_PACKET, 4795 .create = packet_create, 4796 .owner = THIS_MODULE, 4797 }; 4798 4799 static struct notifier_block packet_netdev_notifier = { 4800 .notifier_call = packet_notifier, 4801 }; 4802 4803 #ifdef CONFIG_PROC_FS 4804 4805 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4806 __acquires(RCU) 4807 { 4808 struct net *net = seq_file_net(seq); 4809 4810 rcu_read_lock(); 4811 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4812 } 4813 4814 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4815 { 4816 struct net *net = seq_file_net(seq); 4817 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4818 } 4819 4820 static void packet_seq_stop(struct seq_file *seq, void *v) 4821 __releases(RCU) 4822 { 4823 rcu_read_unlock(); 4824 } 4825 4826 static int packet_seq_show(struct seq_file *seq, void *v) 4827 { 4828 if (v == SEQ_START_TOKEN) 4829 seq_printf(seq, 4830 "%*sRefCnt Type Proto Iface R Rmem User Inode\n", 4831 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk"); 4832 else { 4833 struct sock *s = sk_entry(v); 4834 const struct packet_sock *po = pkt_sk(s); 4835 4836 seq_printf(seq, 4837 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6llu\n", 4838 s, 4839 refcount_read(&s->sk_refcnt), 4840 s->sk_type, 4841 ntohs(READ_ONCE(po->num)), 4842 READ_ONCE(po->ifindex), 4843 packet_sock_flag(po, PACKET_SOCK_RUNNING), 4844 atomic_read(&s->sk_rmem_alloc), 4845 from_kuid_munged(seq_user_ns(seq), sk_uid(s)), 4846 sock_i_ino(s)); 4847 } 4848 4849 return 0; 4850 } 4851 4852 static const struct seq_operations packet_seq_ops = { 4853 .start = packet_seq_start, 4854 .next = packet_seq_next, 4855 .stop = packet_seq_stop, 4856 .show = packet_seq_show, 4857 }; 4858 #endif 4859 4860 static int __net_init packet_net_init(struct net *net) 4861 { 4862 mutex_init(&net->packet.sklist_lock); 4863 INIT_HLIST_HEAD(&net->packet.sklist); 4864 4865 #ifdef CONFIG_PROC_FS 4866 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, 4867 sizeof(struct seq_net_private))) 4868 return -ENOMEM; 4869 #endif /* CONFIG_PROC_FS */ 4870 4871 return 0; 4872 } 4873 4874 static void __net_exit packet_net_exit(struct net *net) 4875 { 4876 remove_proc_entry("packet", net->proc_net); 4877 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); 4878 } 4879 4880 static struct pernet_operations packet_net_ops = { 4881 .init = packet_net_init, 4882 .exit = packet_net_exit, 4883 }; 4884 4885 4886 static void __exit packet_exit(void) 4887 { 4888 sock_unregister(PF_PACKET); 4889 proto_unregister(&packet_proto); 4890 unregister_netdevice_notifier(&packet_netdev_notifier); 4891 unregister_pernet_subsys(&packet_net_ops); 4892 } 4893 4894 static int __init packet_init(void) 4895 { 4896 int rc; 4897 4898 rc = register_pernet_subsys(&packet_net_ops); 4899 if (rc) 4900 goto out; 4901 rc = register_netdevice_notifier(&packet_netdev_notifier); 4902 if (rc) 4903 goto out_pernet; 4904 rc = proto_register(&packet_proto, 0); 4905 if (rc) 4906 goto out_notifier; 4907 rc = sock_register(&packet_family_ops); 4908 if (rc) 4909 goto out_proto; 4910 4911 return 0; 4912 4913 out_proto: 4914 proto_unregister(&packet_proto); 4915 out_notifier: 4916 unregister_netdevice_notifier(&packet_netdev_notifier); 4917 out_pernet: 4918 unregister_pernet_subsys(&packet_net_ops); 4919 out: 4920 return rc; 4921 } 4922 4923 module_init(packet_init); 4924 module_exit(packet_exit); 4925 MODULE_DESCRIPTION("Packet socket support (AF_PACKET)"); 4926 MODULE_LICENSE("GPL"); 4927 MODULE_ALIAS_NETPROTO(PF_PACKET); 4928