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/filter.h> 53 #include <linux/types.h> 54 #include <linux/mm.h> 55 #include <linux/capability.h> 56 #include <linux/fcntl.h> 57 #include <linux/socket.h> 58 #include <linux/in.h> 59 #include <linux/inet.h> 60 #include <linux/netdevice.h> 61 #include <linux/if_packet.h> 62 #include <linux/wireless.h> 63 #include <linux/kernel.h> 64 #include <linux/kmod.h> 65 #include <linux/slab.h> 66 #include <linux/vmalloc.h> 67 #include <net/net_namespace.h> 68 #include <net/ip.h> 69 #include <net/protocol.h> 70 #include <linux/skbuff.h> 71 #include <net/sock.h> 72 #include <linux/errno.h> 73 #include <linux/timer.h> 74 #include <linux/uaccess.h> 75 #include <asm/ioctls.h> 76 #include <asm/page.h> 77 #include <asm/cacheflush.h> 78 #include <asm/io.h> 79 #include <linux/proc_fs.h> 80 #include <linux/seq_file.h> 81 #include <linux/poll.h> 82 #include <linux/module.h> 83 #include <linux/init.h> 84 #include <linux/mutex.h> 85 #include <linux/if_vlan.h> 86 #include <linux/virtio_net.h> 87 #include <linux/errqueue.h> 88 #include <linux/net_tstamp.h> 89 #include <linux/percpu.h> 90 #ifdef CONFIG_INET 91 #include <net/inet_common.h> 92 #endif 93 #include <linux/bpf.h> 94 #include <net/compat.h> 95 #include <linux/netfilter_netdev.h> 96 97 #include "internal.h" 98 99 /* 100 Assumptions: 101 - If the device has no dev->header_ops->create, there is no LL header 102 visible above the device. In this case, its hard_header_len should be 0. 103 The device may prepend its own header internally. In this case, its 104 needed_headroom should be set to the space needed for it to add its 105 internal header. 106 For example, a WiFi driver pretending to be an Ethernet driver should 107 set its hard_header_len to be the Ethernet header length, and set its 108 needed_headroom to be (the real WiFi header length - the fake Ethernet 109 header length). 110 - packet socket receives packets with pulled ll header, 111 so that SOCK_RAW should push it back. 112 113 On receive: 114 ----------- 115 116 Incoming, dev_has_header(dev) == true 117 mac_header -> ll header 118 data -> data 119 120 Outgoing, dev_has_header(dev) == true 121 mac_header -> ll header 122 data -> ll header 123 124 Incoming, dev_has_header(dev) == false 125 mac_header -> data 126 However drivers often make it point to the ll header. 127 This is incorrect because the ll header should be invisible to us. 128 data -> data 129 130 Outgoing, dev_has_header(dev) == false 131 mac_header -> data. ll header is invisible to us. 132 data -> data 133 134 Resume 135 If dev_has_header(dev) == false we are unable to restore the ll header, 136 because it is invisible to us. 137 138 139 On transmit: 140 ------------ 141 142 dev_has_header(dev) == true 143 mac_header -> ll header 144 data -> ll header 145 146 dev_has_header(dev) == false (ll header is invisible to us) 147 mac_header -> data 148 data -> data 149 150 We should set network_header on output to the correct position, 151 packet classifier depends on it. 152 */ 153 154 /* Private packet socket structures. */ 155 156 /* identical to struct packet_mreq except it has 157 * a longer address field. 158 */ 159 struct packet_mreq_max { 160 int mr_ifindex; 161 unsigned short mr_type; 162 unsigned short mr_alen; 163 unsigned char mr_address[MAX_ADDR_LEN]; 164 }; 165 166 union tpacket_uhdr { 167 struct tpacket_hdr *h1; 168 struct tpacket2_hdr *h2; 169 struct tpacket3_hdr *h3; 170 void *raw; 171 }; 172 173 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 174 int closing, int tx_ring); 175 176 #define V3_ALIGNMENT (8) 177 178 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) 179 180 #define BLK_PLUS_PRIV(sz_of_priv) \ 181 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) 182 183 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) 184 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) 185 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) 186 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) 187 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) 188 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) 189 190 struct packet_sock; 191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 192 struct packet_type *pt, struct net_device *orig_dev); 193 194 static void *packet_previous_frame(struct packet_sock *po, 195 struct packet_ring_buffer *rb, 196 int status); 197 static void packet_increment_head(struct packet_ring_buffer *buff); 198 static int prb_curr_blk_in_use(struct tpacket_block_desc *); 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 200 struct packet_sock *); 201 static void prb_retire_current_block(struct tpacket_kbdq_core *, 202 struct packet_sock *, unsigned int status); 203 static int prb_queue_frozen(struct tpacket_kbdq_core *); 204 static void prb_open_block(struct tpacket_kbdq_core *, 205 struct tpacket_block_desc *); 206 static void prb_retire_rx_blk_timer_expired(struct timer_list *); 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); 208 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 209 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 210 struct tpacket3_hdr *); 211 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 212 struct tpacket3_hdr *); 213 static void packet_flush_mclist(struct sock *sk); 214 static u16 packet_pick_tx_queue(struct sk_buff *skb); 215 216 struct packet_skb_cb { 217 union { 218 struct sockaddr_pkt pkt; 219 union { 220 /* Trick: alias skb original length with 221 * ll.sll_family and ll.protocol in order 222 * to save room. 223 */ 224 unsigned int origlen; 225 struct sockaddr_ll ll; 226 }; 227 } sa; 228 }; 229 230 #define vio_le() virtio_legacy_is_little_endian() 231 232 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 233 234 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 235 #define GET_PBLOCK_DESC(x, bid) \ 236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 239 #define GET_NEXT_PRB_BLK_NUM(x) \ 240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 241 ((x)->kactive_blk_num+1) : 0) 242 243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 244 static void __fanout_link(struct sock *sk, struct packet_sock *po); 245 246 #ifdef CONFIG_NETFILTER_EGRESS 247 static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb) 248 { 249 struct sk_buff *next, *head = NULL, *tail; 250 int rc; 251 252 rcu_read_lock(); 253 for (; skb != NULL; skb = next) { 254 next = skb->next; 255 skb_mark_not_on_list(skb); 256 257 if (!nf_hook_egress(skb, &rc, skb->dev)) 258 continue; 259 260 if (!head) 261 head = skb; 262 else 263 tail->next = skb; 264 265 tail = skb; 266 } 267 rcu_read_unlock(); 268 269 return head; 270 } 271 #endif 272 273 static int packet_xmit(const struct packet_sock *po, struct sk_buff *skb) 274 { 275 if (!packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS)) 276 return dev_queue_xmit(skb); 277 278 #ifdef CONFIG_NETFILTER_EGRESS 279 if (nf_hook_egress_active()) { 280 skb = nf_hook_direct_egress(skb); 281 if (!skb) 282 return NET_XMIT_DROP; 283 } 284 #endif 285 return dev_direct_xmit(skb, packet_pick_tx_queue(skb)); 286 } 287 288 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 289 { 290 struct net_device *dev; 291 292 rcu_read_lock(); 293 dev = rcu_dereference(po->cached_dev); 294 dev_hold(dev); 295 rcu_read_unlock(); 296 297 return dev; 298 } 299 300 static void packet_cached_dev_assign(struct packet_sock *po, 301 struct net_device *dev) 302 { 303 rcu_assign_pointer(po->cached_dev, dev); 304 } 305 306 static void packet_cached_dev_reset(struct packet_sock *po) 307 { 308 RCU_INIT_POINTER(po->cached_dev, NULL); 309 } 310 311 static u16 packet_pick_tx_queue(struct sk_buff *skb) 312 { 313 struct net_device *dev = skb->dev; 314 const struct net_device_ops *ops = dev->netdev_ops; 315 int cpu = raw_smp_processor_id(); 316 u16 queue_index; 317 318 #ifdef CONFIG_XPS 319 skb->sender_cpu = cpu + 1; 320 #endif 321 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues); 322 if (ops->ndo_select_queue) { 323 queue_index = ops->ndo_select_queue(dev, skb, NULL); 324 queue_index = netdev_cap_txqueue(dev, queue_index); 325 } else { 326 queue_index = netdev_pick_tx(dev, skb, NULL); 327 } 328 329 return queue_index; 330 } 331 332 /* __register_prot_hook must be invoked through register_prot_hook 333 * or from a context in which asynchronous accesses to the packet 334 * socket is not possible (packet_create()). 335 */ 336 static void __register_prot_hook(struct sock *sk) 337 { 338 struct packet_sock *po = pkt_sk(sk); 339 340 if (!packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 341 if (po->fanout) 342 __fanout_link(sk, po); 343 else 344 dev_add_pack(&po->prot_hook); 345 346 sock_hold(sk); 347 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 1); 348 } 349 } 350 351 static void register_prot_hook(struct sock *sk) 352 { 353 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); 354 __register_prot_hook(sk); 355 } 356 357 /* If the sync parameter is true, we will temporarily drop 358 * the po->bind_lock and do a synchronize_net to make sure no 359 * asynchronous packet processing paths still refer to the elements 360 * of po->prot_hook. If the sync parameter is false, it is the 361 * callers responsibility to take care of this. 362 */ 363 static void __unregister_prot_hook(struct sock *sk, bool sync) 364 { 365 struct packet_sock *po = pkt_sk(sk); 366 367 lockdep_assert_held_once(&po->bind_lock); 368 369 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 0); 370 371 if (po->fanout) 372 __fanout_unlink(sk, po); 373 else 374 __dev_remove_pack(&po->prot_hook); 375 376 __sock_put(sk); 377 378 if (sync) { 379 spin_unlock(&po->bind_lock); 380 synchronize_net(); 381 spin_lock(&po->bind_lock); 382 } 383 } 384 385 static void unregister_prot_hook(struct sock *sk, bool sync) 386 { 387 struct packet_sock *po = pkt_sk(sk); 388 389 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) 390 __unregister_prot_hook(sk, sync); 391 } 392 393 static inline struct page * __pure pgv_to_page(void *addr) 394 { 395 if (is_vmalloc_addr(addr)) 396 return vmalloc_to_page(addr); 397 return virt_to_page(addr); 398 } 399 400 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 401 { 402 union tpacket_uhdr h; 403 404 h.raw = frame; 405 switch (po->tp_version) { 406 case TPACKET_V1: 407 h.h1->tp_status = status; 408 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 409 break; 410 case TPACKET_V2: 411 h.h2->tp_status = status; 412 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 413 break; 414 case TPACKET_V3: 415 h.h3->tp_status = status; 416 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 417 break; 418 default: 419 WARN(1, "TPACKET version not supported.\n"); 420 BUG(); 421 } 422 423 smp_wmb(); 424 } 425 426 static int __packet_get_status(const struct packet_sock *po, void *frame) 427 { 428 union tpacket_uhdr h; 429 430 smp_rmb(); 431 432 h.raw = frame; 433 switch (po->tp_version) { 434 case TPACKET_V1: 435 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 436 return h.h1->tp_status; 437 case TPACKET_V2: 438 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 439 return h.h2->tp_status; 440 case TPACKET_V3: 441 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 442 return h.h3->tp_status; 443 default: 444 WARN(1, "TPACKET version not supported.\n"); 445 BUG(); 446 return 0; 447 } 448 } 449 450 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts, 451 unsigned int flags) 452 { 453 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 454 455 if (shhwtstamps && 456 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 457 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts)) 458 return TP_STATUS_TS_RAW_HARDWARE; 459 460 if ((flags & SOF_TIMESTAMPING_SOFTWARE) && 461 ktime_to_timespec64_cond(skb_tstamp(skb), ts)) 462 return TP_STATUS_TS_SOFTWARE; 463 464 return 0; 465 } 466 467 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 468 struct sk_buff *skb) 469 { 470 union tpacket_uhdr h; 471 struct timespec64 ts; 472 __u32 ts_status; 473 474 if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp)))) 475 return 0; 476 477 h.raw = frame; 478 /* 479 * versions 1 through 3 overflow the timestamps in y2106, since they 480 * all store the seconds in a 32-bit unsigned integer. 481 * If we create a version 4, that should have a 64-bit timestamp, 482 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit 483 * nanoseconds. 484 */ 485 switch (po->tp_version) { 486 case TPACKET_V1: 487 h.h1->tp_sec = ts.tv_sec; 488 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 489 break; 490 case TPACKET_V2: 491 h.h2->tp_sec = ts.tv_sec; 492 h.h2->tp_nsec = ts.tv_nsec; 493 break; 494 case TPACKET_V3: 495 h.h3->tp_sec = ts.tv_sec; 496 h.h3->tp_nsec = ts.tv_nsec; 497 break; 498 default: 499 WARN(1, "TPACKET version not supported.\n"); 500 BUG(); 501 } 502 503 /* one flush is safe, as both fields always lie on the same cacheline */ 504 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 505 smp_wmb(); 506 507 return ts_status; 508 } 509 510 static void *packet_lookup_frame(const struct packet_sock *po, 511 const struct packet_ring_buffer *rb, 512 unsigned int position, 513 int status) 514 { 515 unsigned int pg_vec_pos, frame_offset; 516 union tpacket_uhdr h; 517 518 pg_vec_pos = position / rb->frames_per_block; 519 frame_offset = position % rb->frames_per_block; 520 521 h.raw = rb->pg_vec[pg_vec_pos].buffer + 522 (frame_offset * rb->frame_size); 523 524 if (status != __packet_get_status(po, h.raw)) 525 return NULL; 526 527 return h.raw; 528 } 529 530 static void *packet_current_frame(struct packet_sock *po, 531 struct packet_ring_buffer *rb, 532 int status) 533 { 534 return packet_lookup_frame(po, rb, rb->head, status); 535 } 536 537 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) 538 { 539 del_timer_sync(&pkc->retire_blk_timer); 540 } 541 542 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 543 struct sk_buff_head *rb_queue) 544 { 545 struct tpacket_kbdq_core *pkc; 546 547 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 548 549 spin_lock_bh(&rb_queue->lock); 550 pkc->delete_blk_timer = 1; 551 spin_unlock_bh(&rb_queue->lock); 552 553 prb_del_retire_blk_timer(pkc); 554 } 555 556 static void prb_setup_retire_blk_timer(struct packet_sock *po) 557 { 558 struct tpacket_kbdq_core *pkc; 559 560 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 561 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, 562 0); 563 pkc->retire_blk_timer.expires = jiffies; 564 } 565 566 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 567 int blk_size_in_bytes) 568 { 569 struct net_device *dev; 570 unsigned int mbits, div; 571 struct ethtool_link_ksettings ecmd; 572 int err; 573 574 rtnl_lock(); 575 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 576 if (unlikely(!dev)) { 577 rtnl_unlock(); 578 return DEFAULT_PRB_RETIRE_TOV; 579 } 580 err = __ethtool_get_link_ksettings(dev, &ecmd); 581 rtnl_unlock(); 582 if (err) 583 return DEFAULT_PRB_RETIRE_TOV; 584 585 /* If the link speed is so slow you don't really 586 * need to worry about perf anyways 587 */ 588 if (ecmd.base.speed < SPEED_1000 || 589 ecmd.base.speed == SPEED_UNKNOWN) 590 return DEFAULT_PRB_RETIRE_TOV; 591 592 div = ecmd.base.speed / 1000; 593 mbits = (blk_size_in_bytes * 8) / (1024 * 1024); 594 595 if (div) 596 mbits /= div; 597 598 if (div) 599 return mbits + 1; 600 return mbits; 601 } 602 603 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 604 union tpacket_req_u *req_u) 605 { 606 p1->feature_req_word = req_u->req3.tp_feature_req_word; 607 } 608 609 static void init_prb_bdqc(struct packet_sock *po, 610 struct packet_ring_buffer *rb, 611 struct pgv *pg_vec, 612 union tpacket_req_u *req_u) 613 { 614 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 615 struct tpacket_block_desc *pbd; 616 617 memset(p1, 0x0, sizeof(*p1)); 618 619 p1->knxt_seq_num = 1; 620 p1->pkbdq = pg_vec; 621 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 622 p1->pkblk_start = pg_vec[0].buffer; 623 p1->kblk_size = req_u->req3.tp_block_size; 624 p1->knum_blocks = req_u->req3.tp_block_nr; 625 p1->hdrlen = po->tp_hdrlen; 626 p1->version = po->tp_version; 627 p1->last_kactive_blk_num = 0; 628 po->stats.stats3.tp_freeze_q_cnt = 0; 629 if (req_u->req3.tp_retire_blk_tov) 630 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; 631 else 632 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, 633 req_u->req3.tp_block_size); 634 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); 635 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 636 rwlock_init(&p1->blk_fill_in_prog_lock); 637 638 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 639 prb_init_ft_ops(p1, req_u); 640 prb_setup_retire_blk_timer(po); 641 prb_open_block(p1, pbd); 642 } 643 644 /* Do NOT update the last_blk_num first. 645 * Assumes sk_buff_head lock is held. 646 */ 647 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) 648 { 649 mod_timer(&pkc->retire_blk_timer, 650 jiffies + pkc->tov_in_jiffies); 651 pkc->last_kactive_blk_num = pkc->kactive_blk_num; 652 } 653 654 /* 655 * Timer logic: 656 * 1) We refresh the timer only when we open a block. 657 * By doing this we don't waste cycles refreshing the timer 658 * on packet-by-packet basis. 659 * 660 * With a 1MB block-size, on a 1Gbps line, it will take 661 * i) ~8 ms to fill a block + ii) memcpy etc. 662 * In this cut we are not accounting for the memcpy time. 663 * 664 * So, if the user sets the 'tmo' to 10ms then the timer 665 * will never fire while the block is still getting filled 666 * (which is what we want). However, the user could choose 667 * to close a block early and that's fine. 668 * 669 * But when the timer does fire, we check whether or not to refresh it. 670 * Since the tmo granularity is in msecs, it is not too expensive 671 * to refresh the timer, lets say every '8' msecs. 672 * Either the user can set the 'tmo' or we can derive it based on 673 * a) line-speed and b) block-size. 674 * prb_calc_retire_blk_tmo() calculates the tmo. 675 * 676 */ 677 static void prb_retire_rx_blk_timer_expired(struct timer_list *t) 678 { 679 struct packet_sock *po = 680 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer); 681 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 682 unsigned int frozen; 683 struct tpacket_block_desc *pbd; 684 685 spin_lock(&po->sk.sk_receive_queue.lock); 686 687 frozen = prb_queue_frozen(pkc); 688 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 689 690 if (unlikely(pkc->delete_blk_timer)) 691 goto out; 692 693 /* We only need to plug the race when the block is partially filled. 694 * tpacket_rcv: 695 * lock(); increment BLOCK_NUM_PKTS; unlock() 696 * copy_bits() is in progress ... 697 * timer fires on other cpu: 698 * we can't retire the current block because copy_bits 699 * is in progress. 700 * 701 */ 702 if (BLOCK_NUM_PKTS(pbd)) { 703 /* Waiting for skb_copy_bits to finish... */ 704 write_lock(&pkc->blk_fill_in_prog_lock); 705 write_unlock(&pkc->blk_fill_in_prog_lock); 706 } 707 708 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { 709 if (!frozen) { 710 if (!BLOCK_NUM_PKTS(pbd)) { 711 /* An empty block. Just refresh the timer. */ 712 goto refresh_timer; 713 } 714 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 715 if (!prb_dispatch_next_block(pkc, po)) 716 goto refresh_timer; 717 else 718 goto out; 719 } else { 720 /* Case 1. Queue was frozen because user-space was 721 * lagging behind. 722 */ 723 if (prb_curr_blk_in_use(pbd)) { 724 /* 725 * Ok, user-space is still behind. 726 * So just refresh the timer. 727 */ 728 goto refresh_timer; 729 } else { 730 /* Case 2. queue was frozen,user-space caught up, 731 * now the link went idle && the timer fired. 732 * We don't have a block to close.So we open this 733 * block and restart the timer. 734 * opening a block thaws the queue,restarts timer 735 * Thawing/timer-refresh is a side effect. 736 */ 737 prb_open_block(pkc, pbd); 738 goto out; 739 } 740 } 741 } 742 743 refresh_timer: 744 _prb_refresh_rx_retire_blk_timer(pkc); 745 746 out: 747 spin_unlock(&po->sk.sk_receive_queue.lock); 748 } 749 750 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 751 struct tpacket_block_desc *pbd1, __u32 status) 752 { 753 /* Flush everything minus the block header */ 754 755 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 756 u8 *start, *end; 757 758 start = (u8 *)pbd1; 759 760 /* Skip the block header(we know header WILL fit in 4K) */ 761 start += PAGE_SIZE; 762 763 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 764 for (; start < end; start += PAGE_SIZE) 765 flush_dcache_page(pgv_to_page(start)); 766 767 smp_wmb(); 768 #endif 769 770 /* Now update the block status. */ 771 772 BLOCK_STATUS(pbd1) = status; 773 774 /* Flush the block header */ 775 776 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 777 start = (u8 *)pbd1; 778 flush_dcache_page(pgv_to_page(start)); 779 780 smp_wmb(); 781 #endif 782 } 783 784 /* 785 * Side effect: 786 * 787 * 1) flush the block 788 * 2) Increment active_blk_num 789 * 790 * Note:We DONT refresh the timer on purpose. 791 * Because almost always the next block will be opened. 792 */ 793 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 794 struct tpacket_block_desc *pbd1, 795 struct packet_sock *po, unsigned int stat) 796 { 797 __u32 status = TP_STATUS_USER | stat; 798 799 struct tpacket3_hdr *last_pkt; 800 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 801 struct sock *sk = &po->sk; 802 803 if (atomic_read(&po->tp_drops)) 804 status |= TP_STATUS_LOSING; 805 806 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 807 last_pkt->tp_next_offset = 0; 808 809 /* Get the ts of the last pkt */ 810 if (BLOCK_NUM_PKTS(pbd1)) { 811 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 812 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 813 } else { 814 /* Ok, we tmo'd - so get the current time. 815 * 816 * It shouldn't really happen as we don't close empty 817 * blocks. See prb_retire_rx_blk_timer_expired(). 818 */ 819 struct timespec64 ts; 820 ktime_get_real_ts64(&ts); 821 h1->ts_last_pkt.ts_sec = ts.tv_sec; 822 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 823 } 824 825 smp_wmb(); 826 827 /* Flush the block */ 828 prb_flush_block(pkc1, pbd1, status); 829 830 sk->sk_data_ready(sk); 831 832 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 833 } 834 835 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 836 { 837 pkc->reset_pending_on_curr_blk = 0; 838 } 839 840 /* 841 * Side effect of opening a block: 842 * 843 * 1) prb_queue is thawed. 844 * 2) retire_blk_timer is refreshed. 845 * 846 */ 847 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 848 struct tpacket_block_desc *pbd1) 849 { 850 struct timespec64 ts; 851 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 852 853 smp_rmb(); 854 855 /* We could have just memset this but we will lose the 856 * flexibility of making the priv area sticky 857 */ 858 859 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 860 BLOCK_NUM_PKTS(pbd1) = 0; 861 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 862 863 ktime_get_real_ts64(&ts); 864 865 h1->ts_first_pkt.ts_sec = ts.tv_sec; 866 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 867 868 pkc1->pkblk_start = (char *)pbd1; 869 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 870 871 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 872 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 873 874 pbd1->version = pkc1->version; 875 pkc1->prev = pkc1->nxt_offset; 876 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 877 878 prb_thaw_queue(pkc1); 879 _prb_refresh_rx_retire_blk_timer(pkc1); 880 881 smp_wmb(); 882 } 883 884 /* 885 * Queue freeze logic: 886 * 1) Assume tp_block_nr = 8 blocks. 887 * 2) At time 't0', user opens Rx ring. 888 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 889 * 4) user-space is either sleeping or processing block '0'. 890 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 891 * it will close block-7,loop around and try to fill block '0'. 892 * call-flow: 893 * __packet_lookup_frame_in_block 894 * prb_retire_current_block() 895 * prb_dispatch_next_block() 896 * |->(BLOCK_STATUS == USER) evaluates to true 897 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 898 * 6) Now there are two cases: 899 * 6.1) Link goes idle right after the queue is frozen. 900 * But remember, the last open_block() refreshed the timer. 901 * When this timer expires,it will refresh itself so that we can 902 * re-open block-0 in near future. 903 * 6.2) Link is busy and keeps on receiving packets. This is a simple 904 * case and __packet_lookup_frame_in_block will check if block-0 905 * is free and can now be re-used. 906 */ 907 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 908 struct packet_sock *po) 909 { 910 pkc->reset_pending_on_curr_blk = 1; 911 po->stats.stats3.tp_freeze_q_cnt++; 912 } 913 914 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 915 916 /* 917 * If the next block is free then we will dispatch it 918 * and return a good offset. 919 * Else, we will freeze the queue. 920 * So, caller must check the return value. 921 */ 922 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 923 struct packet_sock *po) 924 { 925 struct tpacket_block_desc *pbd; 926 927 smp_rmb(); 928 929 /* 1. Get current block num */ 930 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 931 932 /* 2. If this block is currently in_use then freeze the queue */ 933 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 934 prb_freeze_queue(pkc, po); 935 return NULL; 936 } 937 938 /* 939 * 3. 940 * open this block and return the offset where the first packet 941 * needs to get stored. 942 */ 943 prb_open_block(pkc, pbd); 944 return (void *)pkc->nxt_offset; 945 } 946 947 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 948 struct packet_sock *po, unsigned int status) 949 { 950 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 951 952 /* retire/close the current block */ 953 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 954 /* 955 * Plug the case where copy_bits() is in progress on 956 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 957 * have space to copy the pkt in the current block and 958 * called prb_retire_current_block() 959 * 960 * We don't need to worry about the TMO case because 961 * the timer-handler already handled this case. 962 */ 963 if (!(status & TP_STATUS_BLK_TMO)) { 964 /* Waiting for skb_copy_bits to finish... */ 965 write_lock(&pkc->blk_fill_in_prog_lock); 966 write_unlock(&pkc->blk_fill_in_prog_lock); 967 } 968 prb_close_block(pkc, pbd, po, status); 969 return; 970 } 971 } 972 973 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) 974 { 975 return TP_STATUS_USER & BLOCK_STATUS(pbd); 976 } 977 978 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 979 { 980 return pkc->reset_pending_on_curr_blk; 981 } 982 983 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 984 __releases(&pkc->blk_fill_in_prog_lock) 985 { 986 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 987 988 read_unlock(&pkc->blk_fill_in_prog_lock); 989 } 990 991 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 992 struct tpacket3_hdr *ppd) 993 { 994 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 995 } 996 997 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 998 struct tpacket3_hdr *ppd) 999 { 1000 ppd->hv1.tp_rxhash = 0; 1001 } 1002 1003 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 1004 struct tpacket3_hdr *ppd) 1005 { 1006 if (skb_vlan_tag_present(pkc->skb)) { 1007 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 1008 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 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_sock_destruct(struct sock *sk) 1327 { 1328 skb_queue_purge(&sk->sk_error_queue); 1329 1330 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1331 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 1332 1333 if (!sock_flag(sk, SOCK_DEAD)) { 1334 pr_err("Attempt to release alive packet socket: %p\n", sk); 1335 return; 1336 } 1337 } 1338 1339 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1340 { 1341 u32 *history = po->rollover->history; 1342 u32 victim, rxhash; 1343 int i, count = 0; 1344 1345 rxhash = skb_get_hash(skb); 1346 for (i = 0; i < ROLLOVER_HLEN; i++) 1347 if (READ_ONCE(history[i]) == rxhash) 1348 count++; 1349 1350 victim = get_random_u32_below(ROLLOVER_HLEN); 1351 1352 /* Avoid dirtying the cache line if possible */ 1353 if (READ_ONCE(history[victim]) != rxhash) 1354 WRITE_ONCE(history[victim], rxhash); 1355 1356 return count > (ROLLOVER_HLEN >> 1); 1357 } 1358 1359 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1360 struct sk_buff *skb, 1361 unsigned int num) 1362 { 1363 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1364 } 1365 1366 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1367 struct sk_buff *skb, 1368 unsigned int num) 1369 { 1370 unsigned int val = atomic_inc_return(&f->rr_cur); 1371 1372 return val % num; 1373 } 1374 1375 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1376 struct sk_buff *skb, 1377 unsigned int num) 1378 { 1379 return smp_processor_id() % num; 1380 } 1381 1382 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1383 struct sk_buff *skb, 1384 unsigned int num) 1385 { 1386 return get_random_u32_below(num); 1387 } 1388 1389 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1390 struct sk_buff *skb, 1391 unsigned int idx, bool try_self, 1392 unsigned int num) 1393 { 1394 struct packet_sock *po, *po_next, *po_skip = NULL; 1395 unsigned int i, j, room = ROOM_NONE; 1396 1397 po = pkt_sk(rcu_dereference(f->arr[idx])); 1398 1399 if (try_self) { 1400 room = packet_rcv_has_room(po, skb); 1401 if (room == ROOM_NORMAL || 1402 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1403 return idx; 1404 po_skip = po; 1405 } 1406 1407 i = j = min_t(int, po->rollover->sock, num - 1); 1408 do { 1409 po_next = pkt_sk(rcu_dereference(f->arr[i])); 1410 if (po_next != po_skip && 1411 !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) && 1412 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1413 if (i != j) 1414 po->rollover->sock = i; 1415 atomic_long_inc(&po->rollover->num); 1416 if (room == ROOM_LOW) 1417 atomic_long_inc(&po->rollover->num_huge); 1418 return i; 1419 } 1420 1421 if (++i == num) 1422 i = 0; 1423 } while (i != j); 1424 1425 atomic_long_inc(&po->rollover->num_failed); 1426 return idx; 1427 } 1428 1429 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1430 struct sk_buff *skb, 1431 unsigned int num) 1432 { 1433 return skb_get_queue_mapping(skb) % num; 1434 } 1435 1436 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1437 struct sk_buff *skb, 1438 unsigned int num) 1439 { 1440 struct bpf_prog *prog; 1441 unsigned int ret = 0; 1442 1443 rcu_read_lock(); 1444 prog = rcu_dereference(f->bpf_prog); 1445 if (prog) 1446 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1447 rcu_read_unlock(); 1448 1449 return ret; 1450 } 1451 1452 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1453 { 1454 return f->flags & (flag >> 8); 1455 } 1456 1457 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1458 struct packet_type *pt, struct net_device *orig_dev) 1459 { 1460 struct packet_fanout *f = pt->af_packet_priv; 1461 unsigned int num = READ_ONCE(f->num_members); 1462 struct net *net = read_pnet(&f->net); 1463 struct packet_sock *po; 1464 unsigned int idx; 1465 1466 if (!net_eq(dev_net(dev), net) || !num) { 1467 kfree_skb(skb); 1468 return 0; 1469 } 1470 1471 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1472 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1473 if (!skb) 1474 return 0; 1475 } 1476 switch (f->type) { 1477 case PACKET_FANOUT_HASH: 1478 default: 1479 idx = fanout_demux_hash(f, skb, num); 1480 break; 1481 case PACKET_FANOUT_LB: 1482 idx = fanout_demux_lb(f, skb, num); 1483 break; 1484 case PACKET_FANOUT_CPU: 1485 idx = fanout_demux_cpu(f, skb, num); 1486 break; 1487 case PACKET_FANOUT_RND: 1488 idx = fanout_demux_rnd(f, skb, num); 1489 break; 1490 case PACKET_FANOUT_QM: 1491 idx = fanout_demux_qm(f, skb, num); 1492 break; 1493 case PACKET_FANOUT_ROLLOVER: 1494 idx = fanout_demux_rollover(f, skb, 0, false, num); 1495 break; 1496 case PACKET_FANOUT_CBPF: 1497 case PACKET_FANOUT_EBPF: 1498 idx = fanout_demux_bpf(f, skb, num); 1499 break; 1500 } 1501 1502 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1503 idx = fanout_demux_rollover(f, skb, idx, true, num); 1504 1505 po = pkt_sk(rcu_dereference(f->arr[idx])); 1506 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1507 } 1508 1509 DEFINE_MUTEX(fanout_mutex); 1510 EXPORT_SYMBOL_GPL(fanout_mutex); 1511 static LIST_HEAD(fanout_list); 1512 static u16 fanout_next_id; 1513 1514 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1515 { 1516 struct packet_fanout *f = po->fanout; 1517 1518 spin_lock(&f->lock); 1519 rcu_assign_pointer(f->arr[f->num_members], sk); 1520 smp_wmb(); 1521 f->num_members++; 1522 if (f->num_members == 1) 1523 dev_add_pack(&f->prot_hook); 1524 spin_unlock(&f->lock); 1525 } 1526 1527 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1528 { 1529 struct packet_fanout *f = po->fanout; 1530 int i; 1531 1532 spin_lock(&f->lock); 1533 for (i = 0; i < f->num_members; i++) { 1534 if (rcu_dereference_protected(f->arr[i], 1535 lockdep_is_held(&f->lock)) == sk) 1536 break; 1537 } 1538 BUG_ON(i >= f->num_members); 1539 rcu_assign_pointer(f->arr[i], 1540 rcu_dereference_protected(f->arr[f->num_members - 1], 1541 lockdep_is_held(&f->lock))); 1542 f->num_members--; 1543 if (f->num_members == 0) 1544 __dev_remove_pack(&f->prot_hook); 1545 spin_unlock(&f->lock); 1546 } 1547 1548 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1549 { 1550 if (sk->sk_family != PF_PACKET) 1551 return false; 1552 1553 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1554 } 1555 1556 static void fanout_init_data(struct packet_fanout *f) 1557 { 1558 switch (f->type) { 1559 case PACKET_FANOUT_LB: 1560 atomic_set(&f->rr_cur, 0); 1561 break; 1562 case PACKET_FANOUT_CBPF: 1563 case PACKET_FANOUT_EBPF: 1564 RCU_INIT_POINTER(f->bpf_prog, NULL); 1565 break; 1566 } 1567 } 1568 1569 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1570 { 1571 struct bpf_prog *old; 1572 1573 spin_lock(&f->lock); 1574 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1575 rcu_assign_pointer(f->bpf_prog, new); 1576 spin_unlock(&f->lock); 1577 1578 if (old) { 1579 synchronize_net(); 1580 bpf_prog_destroy(old); 1581 } 1582 } 1583 1584 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data, 1585 unsigned int len) 1586 { 1587 struct bpf_prog *new; 1588 struct sock_fprog fprog; 1589 int ret; 1590 1591 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1592 return -EPERM; 1593 1594 ret = copy_bpf_fprog_from_user(&fprog, data, len); 1595 if (ret) 1596 return ret; 1597 1598 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1599 if (ret) 1600 return ret; 1601 1602 __fanout_set_data_bpf(po->fanout, new); 1603 return 0; 1604 } 1605 1606 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data, 1607 unsigned int len) 1608 { 1609 struct bpf_prog *new; 1610 u32 fd; 1611 1612 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1613 return -EPERM; 1614 if (len != sizeof(fd)) 1615 return -EINVAL; 1616 if (copy_from_sockptr(&fd, data, len)) 1617 return -EFAULT; 1618 1619 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1620 if (IS_ERR(new)) 1621 return PTR_ERR(new); 1622 1623 __fanout_set_data_bpf(po->fanout, new); 1624 return 0; 1625 } 1626 1627 static int fanout_set_data(struct packet_sock *po, sockptr_t data, 1628 unsigned int len) 1629 { 1630 switch (po->fanout->type) { 1631 case PACKET_FANOUT_CBPF: 1632 return fanout_set_data_cbpf(po, data, len); 1633 case PACKET_FANOUT_EBPF: 1634 return fanout_set_data_ebpf(po, data, len); 1635 default: 1636 return -EINVAL; 1637 } 1638 } 1639 1640 static void fanout_release_data(struct packet_fanout *f) 1641 { 1642 switch (f->type) { 1643 case PACKET_FANOUT_CBPF: 1644 case PACKET_FANOUT_EBPF: 1645 __fanout_set_data_bpf(f, NULL); 1646 } 1647 } 1648 1649 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) 1650 { 1651 struct packet_fanout *f; 1652 1653 list_for_each_entry(f, &fanout_list, list) { 1654 if (f->id == candidate_id && 1655 read_pnet(&f->net) == sock_net(sk)) { 1656 return false; 1657 } 1658 } 1659 return true; 1660 } 1661 1662 static bool fanout_find_new_id(struct sock *sk, u16 *new_id) 1663 { 1664 u16 id = fanout_next_id; 1665 1666 do { 1667 if (__fanout_id_is_free(sk, id)) { 1668 *new_id = id; 1669 fanout_next_id = id + 1; 1670 return true; 1671 } 1672 1673 id++; 1674 } while (id != fanout_next_id); 1675 1676 return false; 1677 } 1678 1679 static int fanout_add(struct sock *sk, struct fanout_args *args) 1680 { 1681 struct packet_rollover *rollover = NULL; 1682 struct packet_sock *po = pkt_sk(sk); 1683 u16 type_flags = args->type_flags; 1684 struct packet_fanout *f, *match; 1685 u8 type = type_flags & 0xff; 1686 u8 flags = type_flags >> 8; 1687 u16 id = args->id; 1688 int err; 1689 1690 switch (type) { 1691 case PACKET_FANOUT_ROLLOVER: 1692 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1693 return -EINVAL; 1694 break; 1695 case PACKET_FANOUT_HASH: 1696 case PACKET_FANOUT_LB: 1697 case PACKET_FANOUT_CPU: 1698 case PACKET_FANOUT_RND: 1699 case PACKET_FANOUT_QM: 1700 case PACKET_FANOUT_CBPF: 1701 case PACKET_FANOUT_EBPF: 1702 break; 1703 default: 1704 return -EINVAL; 1705 } 1706 1707 mutex_lock(&fanout_mutex); 1708 1709 err = -EALREADY; 1710 if (po->fanout) 1711 goto out; 1712 1713 if (type == PACKET_FANOUT_ROLLOVER || 1714 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1715 err = -ENOMEM; 1716 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); 1717 if (!rollover) 1718 goto out; 1719 atomic_long_set(&rollover->num, 0); 1720 atomic_long_set(&rollover->num_huge, 0); 1721 atomic_long_set(&rollover->num_failed, 0); 1722 } 1723 1724 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { 1725 if (id != 0) { 1726 err = -EINVAL; 1727 goto out; 1728 } 1729 if (!fanout_find_new_id(sk, &id)) { 1730 err = -ENOMEM; 1731 goto out; 1732 } 1733 /* ephemeral flag for the first socket in the group: drop it */ 1734 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); 1735 } 1736 1737 match = NULL; 1738 list_for_each_entry(f, &fanout_list, list) { 1739 if (f->id == id && 1740 read_pnet(&f->net) == sock_net(sk)) { 1741 match = f; 1742 break; 1743 } 1744 } 1745 err = -EINVAL; 1746 if (match) { 1747 if (match->flags != flags) 1748 goto out; 1749 if (args->max_num_members && 1750 args->max_num_members != match->max_num_members) 1751 goto out; 1752 } else { 1753 if (args->max_num_members > PACKET_FANOUT_MAX) 1754 goto out; 1755 if (!args->max_num_members) 1756 /* legacy PACKET_FANOUT_MAX */ 1757 args->max_num_members = 256; 1758 err = -ENOMEM; 1759 match = kvzalloc(struct_size(match, arr, args->max_num_members), 1760 GFP_KERNEL); 1761 if (!match) 1762 goto out; 1763 write_pnet(&match->net, sock_net(sk)); 1764 match->id = id; 1765 match->type = type; 1766 match->flags = flags; 1767 INIT_LIST_HEAD(&match->list); 1768 spin_lock_init(&match->lock); 1769 refcount_set(&match->sk_ref, 0); 1770 fanout_init_data(match); 1771 match->prot_hook.type = po->prot_hook.type; 1772 match->prot_hook.dev = po->prot_hook.dev; 1773 match->prot_hook.func = packet_rcv_fanout; 1774 match->prot_hook.af_packet_priv = match; 1775 match->prot_hook.af_packet_net = read_pnet(&match->net); 1776 match->prot_hook.id_match = match_fanout_group; 1777 match->max_num_members = args->max_num_members; 1778 match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING; 1779 list_add(&match->list, &fanout_list); 1780 } 1781 err = -EINVAL; 1782 1783 spin_lock(&po->bind_lock); 1784 if (packet_sock_flag(po, PACKET_SOCK_RUNNING) && 1785 match->type == type && 1786 match->prot_hook.type == po->prot_hook.type && 1787 match->prot_hook.dev == po->prot_hook.dev) { 1788 err = -ENOSPC; 1789 if (refcount_read(&match->sk_ref) < match->max_num_members) { 1790 __dev_remove_pack(&po->prot_hook); 1791 1792 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */ 1793 WRITE_ONCE(po->fanout, match); 1794 1795 po->rollover = rollover; 1796 rollover = NULL; 1797 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); 1798 __fanout_link(sk, po); 1799 err = 0; 1800 } 1801 } 1802 spin_unlock(&po->bind_lock); 1803 1804 if (err && !refcount_read(&match->sk_ref)) { 1805 list_del(&match->list); 1806 kvfree(match); 1807 } 1808 1809 out: 1810 kfree(rollover); 1811 mutex_unlock(&fanout_mutex); 1812 return err; 1813 } 1814 1815 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes 1816 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. 1817 * It is the responsibility of the caller to call fanout_release_data() and 1818 * free the returned packet_fanout (after synchronize_net()) 1819 */ 1820 static struct packet_fanout *fanout_release(struct sock *sk) 1821 { 1822 struct packet_sock *po = pkt_sk(sk); 1823 struct packet_fanout *f; 1824 1825 mutex_lock(&fanout_mutex); 1826 f = po->fanout; 1827 if (f) { 1828 po->fanout = NULL; 1829 1830 if (refcount_dec_and_test(&f->sk_ref)) 1831 list_del(&f->list); 1832 else 1833 f = NULL; 1834 } 1835 mutex_unlock(&fanout_mutex); 1836 1837 return f; 1838 } 1839 1840 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1841 struct sk_buff *skb) 1842 { 1843 /* Earlier code assumed this would be a VLAN pkt, double-check 1844 * this now that we have the actual packet in hand. We can only 1845 * do this check on Ethernet devices. 1846 */ 1847 if (unlikely(dev->type != ARPHRD_ETHER)) 1848 return false; 1849 1850 skb_reset_mac_header(skb); 1851 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1852 } 1853 1854 static const struct proto_ops packet_ops; 1855 1856 static const struct proto_ops packet_ops_spkt; 1857 1858 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1859 struct packet_type *pt, struct net_device *orig_dev) 1860 { 1861 struct sock *sk; 1862 struct sockaddr_pkt *spkt; 1863 1864 /* 1865 * When we registered the protocol we saved the socket in the data 1866 * field for just this event. 1867 */ 1868 1869 sk = pt->af_packet_priv; 1870 1871 /* 1872 * Yank back the headers [hope the device set this 1873 * right or kerboom...] 1874 * 1875 * Incoming packets have ll header pulled, 1876 * push it back. 1877 * 1878 * For outgoing ones skb->data == skb_mac_header(skb) 1879 * so that this procedure is noop. 1880 */ 1881 1882 if (skb->pkt_type == PACKET_LOOPBACK) 1883 goto out; 1884 1885 if (!net_eq(dev_net(dev), sock_net(sk))) 1886 goto out; 1887 1888 skb = skb_share_check(skb, GFP_ATOMIC); 1889 if (skb == NULL) 1890 goto oom; 1891 1892 /* drop any routing info */ 1893 skb_dst_drop(skb); 1894 1895 /* drop conntrack reference */ 1896 nf_reset_ct(skb); 1897 1898 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1899 1900 skb_push(skb, skb->data - skb_mac_header(skb)); 1901 1902 /* 1903 * The SOCK_PACKET socket receives _all_ frames. 1904 */ 1905 1906 spkt->spkt_family = dev->type; 1907 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1908 spkt->spkt_protocol = skb->protocol; 1909 1910 /* 1911 * Charge the memory to the socket. This is done specifically 1912 * to prevent sockets using all the memory up. 1913 */ 1914 1915 if (sock_queue_rcv_skb(sk, skb) == 0) 1916 return 0; 1917 1918 out: 1919 kfree_skb(skb); 1920 oom: 1921 return 0; 1922 } 1923 1924 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock) 1925 { 1926 int depth; 1927 1928 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && 1929 sock->type == SOCK_RAW) { 1930 skb_reset_mac_header(skb); 1931 skb->protocol = dev_parse_header_protocol(skb); 1932 } 1933 1934 /* Move network header to the right position for VLAN tagged packets */ 1935 if (likely(skb->dev->type == ARPHRD_ETHER) && 1936 eth_type_vlan(skb->protocol) && 1937 __vlan_get_protocol(skb, skb->protocol, &depth) != 0) { 1938 if (pskb_may_pull(skb, depth)) 1939 skb_set_network_header(skb, depth); 1940 } 1941 1942 skb_probe_transport_header(skb); 1943 } 1944 1945 /* 1946 * Output a raw packet to a device layer. This bypasses all the other 1947 * protocol layers and you must therefore supply it with a complete frame 1948 */ 1949 1950 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1951 size_t len) 1952 { 1953 struct sock *sk = sock->sk; 1954 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1955 struct sk_buff *skb = NULL; 1956 struct net_device *dev; 1957 struct sockcm_cookie sockc; 1958 __be16 proto = 0; 1959 int err; 1960 int extra_len = 0; 1961 1962 /* 1963 * Get and verify the address. 1964 */ 1965 1966 if (saddr) { 1967 if (msg->msg_namelen < sizeof(struct sockaddr)) 1968 return -EINVAL; 1969 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1970 proto = saddr->spkt_protocol; 1971 } else 1972 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1973 1974 /* 1975 * Find the device first to size check it 1976 */ 1977 1978 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 1979 retry: 1980 rcu_read_lock(); 1981 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 1982 err = -ENODEV; 1983 if (dev == NULL) 1984 goto out_unlock; 1985 1986 err = -ENETDOWN; 1987 if (!(dev->flags & IFF_UP)) 1988 goto out_unlock; 1989 1990 /* 1991 * You may not queue a frame bigger than the mtu. This is the lowest level 1992 * raw protocol and you must do your own fragmentation at this level. 1993 */ 1994 1995 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 1996 if (!netif_supports_nofcs(dev)) { 1997 err = -EPROTONOSUPPORT; 1998 goto out_unlock; 1999 } 2000 extra_len = 4; /* We're doing our own CRC */ 2001 } 2002 2003 err = -EMSGSIZE; 2004 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) 2005 goto out_unlock; 2006 2007 if (!skb) { 2008 size_t reserved = LL_RESERVED_SPACE(dev); 2009 int tlen = dev->needed_tailroom; 2010 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; 2011 2012 rcu_read_unlock(); 2013 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 2014 if (skb == NULL) 2015 return -ENOBUFS; 2016 /* FIXME: Save some space for broken drivers that write a hard 2017 * header at transmission time by themselves. PPP is the notable 2018 * one here. This should really be fixed at the driver level. 2019 */ 2020 skb_reserve(skb, reserved); 2021 skb_reset_network_header(skb); 2022 2023 /* Try to align data part correctly */ 2024 if (hhlen) { 2025 skb->data -= hhlen; 2026 skb->tail -= hhlen; 2027 if (len < hhlen) 2028 skb_reset_network_header(skb); 2029 } 2030 err = memcpy_from_msg(skb_put(skb, len), msg, len); 2031 if (err) 2032 goto out_free; 2033 goto retry; 2034 } 2035 2036 if (!dev_validate_header(dev, skb->data, len)) { 2037 err = -EINVAL; 2038 goto out_unlock; 2039 } 2040 if (len > (dev->mtu + dev->hard_header_len + extra_len) && 2041 !packet_extra_vlan_len_allowed(dev, skb)) { 2042 err = -EMSGSIZE; 2043 goto out_unlock; 2044 } 2045 2046 sockcm_init(&sockc, sk); 2047 if (msg->msg_controllen) { 2048 err = sock_cmsg_send(sk, msg, &sockc); 2049 if (unlikely(err)) 2050 goto out_unlock; 2051 } 2052 2053 skb->protocol = proto; 2054 skb->dev = dev; 2055 skb->priority = sk->sk_priority; 2056 skb->mark = sk->sk_mark; 2057 skb->tstamp = sockc.transmit_time; 2058 2059 skb_setup_tx_timestamp(skb, sockc.tsflags); 2060 2061 if (unlikely(extra_len == 4)) 2062 skb->no_fcs = 1; 2063 2064 packet_parse_headers(skb, sock); 2065 2066 dev_queue_xmit(skb); 2067 rcu_read_unlock(); 2068 return len; 2069 2070 out_unlock: 2071 rcu_read_unlock(); 2072 out_free: 2073 kfree_skb(skb); 2074 return err; 2075 } 2076 2077 static unsigned int run_filter(struct sk_buff *skb, 2078 const struct sock *sk, 2079 unsigned int res) 2080 { 2081 struct sk_filter *filter; 2082 2083 rcu_read_lock(); 2084 filter = rcu_dereference(sk->sk_filter); 2085 if (filter != NULL) 2086 res = bpf_prog_run_clear_cb(filter->prog, skb); 2087 rcu_read_unlock(); 2088 2089 return res; 2090 } 2091 2092 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2093 size_t *len) 2094 { 2095 struct virtio_net_hdr vnet_hdr; 2096 2097 if (*len < sizeof(vnet_hdr)) 2098 return -EINVAL; 2099 *len -= sizeof(vnet_hdr); 2100 2101 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0)) 2102 return -EINVAL; 2103 2104 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); 2105 } 2106 2107 /* 2108 * This function makes lazy skb cloning in hope that most of packets 2109 * are discarded by BPF. 2110 * 2111 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2112 * and skb->cb are mangled. It works because (and until) packets 2113 * falling here are owned by current CPU. Output packets are cloned 2114 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2115 * sequentially, so that if we return skb to original state on exit, 2116 * we will not harm anyone. 2117 */ 2118 2119 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2120 struct packet_type *pt, struct net_device *orig_dev) 2121 { 2122 struct sock *sk; 2123 struct sockaddr_ll *sll; 2124 struct packet_sock *po; 2125 u8 *skb_head = skb->data; 2126 int skb_len = skb->len; 2127 unsigned int snaplen, res; 2128 bool is_drop_n_account = false; 2129 2130 if (skb->pkt_type == PACKET_LOOPBACK) 2131 goto drop; 2132 2133 sk = pt->af_packet_priv; 2134 po = pkt_sk(sk); 2135 2136 if (!net_eq(dev_net(dev), sock_net(sk))) 2137 goto drop; 2138 2139 skb->dev = dev; 2140 2141 if (dev_has_header(dev)) { 2142 /* The device has an explicit notion of ll header, 2143 * exported to higher levels. 2144 * 2145 * Otherwise, the device hides details of its frame 2146 * structure, so that corresponding packet head is 2147 * never delivered to user. 2148 */ 2149 if (sk->sk_type != SOCK_DGRAM) 2150 skb_push(skb, skb->data - skb_mac_header(skb)); 2151 else if (skb->pkt_type == PACKET_OUTGOING) { 2152 /* Special case: outgoing packets have ll header at head */ 2153 skb_pull(skb, skb_network_offset(skb)); 2154 } 2155 } 2156 2157 snaplen = skb->len; 2158 2159 res = run_filter(skb, sk, snaplen); 2160 if (!res) 2161 goto drop_n_restore; 2162 if (snaplen > res) 2163 snaplen = res; 2164 2165 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2166 goto drop_n_acct; 2167 2168 if (skb_shared(skb)) { 2169 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2170 if (nskb == NULL) 2171 goto drop_n_acct; 2172 2173 if (skb_head != skb->data) { 2174 skb->data = skb_head; 2175 skb->len = skb_len; 2176 } 2177 consume_skb(skb); 2178 skb = nskb; 2179 } 2180 2181 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2182 2183 sll = &PACKET_SKB_CB(skb)->sa.ll; 2184 sll->sll_hatype = dev->type; 2185 sll->sll_pkttype = skb->pkt_type; 2186 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV))) 2187 sll->sll_ifindex = orig_dev->ifindex; 2188 else 2189 sll->sll_ifindex = dev->ifindex; 2190 2191 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2192 2193 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2194 * Use their space for storing the original skb length. 2195 */ 2196 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2197 2198 if (pskb_trim(skb, snaplen)) 2199 goto drop_n_acct; 2200 2201 skb_set_owner_r(skb, sk); 2202 skb->dev = NULL; 2203 skb_dst_drop(skb); 2204 2205 /* drop conntrack reference */ 2206 nf_reset_ct(skb); 2207 2208 spin_lock(&sk->sk_receive_queue.lock); 2209 po->stats.stats1.tp_packets++; 2210 sock_skb_set_dropcount(sk, skb); 2211 skb_clear_delivery_time(skb); 2212 __skb_queue_tail(&sk->sk_receive_queue, skb); 2213 spin_unlock(&sk->sk_receive_queue.lock); 2214 sk->sk_data_ready(sk); 2215 return 0; 2216 2217 drop_n_acct: 2218 is_drop_n_account = true; 2219 atomic_inc(&po->tp_drops); 2220 atomic_inc(&sk->sk_drops); 2221 2222 drop_n_restore: 2223 if (skb_head != skb->data && skb_shared(skb)) { 2224 skb->data = skb_head; 2225 skb->len = skb_len; 2226 } 2227 drop: 2228 if (!is_drop_n_account) 2229 consume_skb(skb); 2230 else 2231 kfree_skb(skb); 2232 return 0; 2233 } 2234 2235 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2236 struct packet_type *pt, struct net_device *orig_dev) 2237 { 2238 struct sock *sk; 2239 struct packet_sock *po; 2240 struct sockaddr_ll *sll; 2241 union tpacket_uhdr h; 2242 u8 *skb_head = skb->data; 2243 int skb_len = skb->len; 2244 unsigned int snaplen, res; 2245 unsigned long status = TP_STATUS_USER; 2246 unsigned short macoff, hdrlen; 2247 unsigned int netoff; 2248 struct sk_buff *copy_skb = NULL; 2249 struct timespec64 ts; 2250 __u32 ts_status; 2251 bool is_drop_n_account = false; 2252 unsigned int slot_id = 0; 2253 bool do_vnet = false; 2254 2255 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2256 * We may add members to them until current aligned size without forcing 2257 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2258 */ 2259 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2260 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2261 2262 if (skb->pkt_type == PACKET_LOOPBACK) 2263 goto drop; 2264 2265 sk = pt->af_packet_priv; 2266 po = pkt_sk(sk); 2267 2268 if (!net_eq(dev_net(dev), sock_net(sk))) 2269 goto drop; 2270 2271 if (dev_has_header(dev)) { 2272 if (sk->sk_type != SOCK_DGRAM) 2273 skb_push(skb, skb->data - skb_mac_header(skb)); 2274 else if (skb->pkt_type == PACKET_OUTGOING) { 2275 /* Special case: outgoing packets have ll header at head */ 2276 skb_pull(skb, skb_network_offset(skb)); 2277 } 2278 } 2279 2280 snaplen = skb->len; 2281 2282 res = run_filter(skb, sk, snaplen); 2283 if (!res) 2284 goto drop_n_restore; 2285 2286 /* If we are flooded, just give up */ 2287 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) { 2288 atomic_inc(&po->tp_drops); 2289 goto drop_n_restore; 2290 } 2291 2292 if (skb->ip_summed == CHECKSUM_PARTIAL) 2293 status |= TP_STATUS_CSUMNOTREADY; 2294 else if (skb->pkt_type != PACKET_OUTGOING && 2295 skb_csum_unnecessary(skb)) 2296 status |= TP_STATUS_CSUM_VALID; 2297 if (skb_is_gso(skb) && skb_is_gso_tcp(skb)) 2298 status |= TP_STATUS_GSO_TCP; 2299 2300 if (snaplen > res) 2301 snaplen = res; 2302 2303 if (sk->sk_type == SOCK_DGRAM) { 2304 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2305 po->tp_reserve; 2306 } else { 2307 unsigned int maclen = skb_network_offset(skb); 2308 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2309 (maclen < 16 ? 16 : maclen)) + 2310 po->tp_reserve; 2311 if (packet_sock_flag(po, PACKET_SOCK_HAS_VNET_HDR)) { 2312 netoff += sizeof(struct virtio_net_hdr); 2313 do_vnet = true; 2314 } 2315 macoff = netoff - maclen; 2316 } 2317 if (netoff > USHRT_MAX) { 2318 atomic_inc(&po->tp_drops); 2319 goto drop_n_restore; 2320 } 2321 if (po->tp_version <= TPACKET_V2) { 2322 if (macoff + snaplen > po->rx_ring.frame_size) { 2323 if (po->copy_thresh && 2324 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2325 if (skb_shared(skb)) { 2326 copy_skb = skb_clone(skb, GFP_ATOMIC); 2327 } else { 2328 copy_skb = skb_get(skb); 2329 skb_head = skb->data; 2330 } 2331 if (copy_skb) { 2332 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0, 2333 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll)); 2334 skb_set_owner_r(copy_skb, sk); 2335 } 2336 } 2337 snaplen = po->rx_ring.frame_size - macoff; 2338 if ((int)snaplen < 0) { 2339 snaplen = 0; 2340 do_vnet = false; 2341 } 2342 } 2343 } else if (unlikely(macoff + snaplen > 2344 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2345 u32 nval; 2346 2347 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2348 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2349 snaplen, nval, macoff); 2350 snaplen = nval; 2351 if (unlikely((int)snaplen < 0)) { 2352 snaplen = 0; 2353 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2354 do_vnet = false; 2355 } 2356 } 2357 spin_lock(&sk->sk_receive_queue.lock); 2358 h.raw = packet_current_rx_frame(po, skb, 2359 TP_STATUS_KERNEL, (macoff+snaplen)); 2360 if (!h.raw) 2361 goto drop_n_account; 2362 2363 if (po->tp_version <= TPACKET_V2) { 2364 slot_id = po->rx_ring.head; 2365 if (test_bit(slot_id, po->rx_ring.rx_owner_map)) 2366 goto drop_n_account; 2367 __set_bit(slot_id, po->rx_ring.rx_owner_map); 2368 } 2369 2370 if (do_vnet && 2371 virtio_net_hdr_from_skb(skb, h.raw + macoff - 2372 sizeof(struct virtio_net_hdr), 2373 vio_le(), true, 0)) { 2374 if (po->tp_version == TPACKET_V3) 2375 prb_clear_blk_fill_status(&po->rx_ring); 2376 goto drop_n_account; 2377 } 2378 2379 if (po->tp_version <= TPACKET_V2) { 2380 packet_increment_rx_head(po, &po->rx_ring); 2381 /* 2382 * LOSING will be reported till you read the stats, 2383 * because it's COR - Clear On Read. 2384 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2385 * at packet level. 2386 */ 2387 if (atomic_read(&po->tp_drops)) 2388 status |= TP_STATUS_LOSING; 2389 } 2390 2391 po->stats.stats1.tp_packets++; 2392 if (copy_skb) { 2393 status |= TP_STATUS_COPY; 2394 skb_clear_delivery_time(copy_skb); 2395 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2396 } 2397 spin_unlock(&sk->sk_receive_queue.lock); 2398 2399 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2400 2401 /* Always timestamp; prefer an existing software timestamp taken 2402 * closer to the time of capture. 2403 */ 2404 ts_status = tpacket_get_timestamp(skb, &ts, 2405 READ_ONCE(po->tp_tstamp) | 2406 SOF_TIMESTAMPING_SOFTWARE); 2407 if (!ts_status) 2408 ktime_get_real_ts64(&ts); 2409 2410 status |= ts_status; 2411 2412 switch (po->tp_version) { 2413 case TPACKET_V1: 2414 h.h1->tp_len = skb->len; 2415 h.h1->tp_snaplen = snaplen; 2416 h.h1->tp_mac = macoff; 2417 h.h1->tp_net = netoff; 2418 h.h1->tp_sec = ts.tv_sec; 2419 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2420 hdrlen = sizeof(*h.h1); 2421 break; 2422 case TPACKET_V2: 2423 h.h2->tp_len = skb->len; 2424 h.h2->tp_snaplen = snaplen; 2425 h.h2->tp_mac = macoff; 2426 h.h2->tp_net = netoff; 2427 h.h2->tp_sec = ts.tv_sec; 2428 h.h2->tp_nsec = ts.tv_nsec; 2429 if (skb_vlan_tag_present(skb)) { 2430 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2431 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2432 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2433 } else { 2434 h.h2->tp_vlan_tci = 0; 2435 h.h2->tp_vlan_tpid = 0; 2436 } 2437 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2438 hdrlen = sizeof(*h.h2); 2439 break; 2440 case TPACKET_V3: 2441 /* tp_nxt_offset,vlan are already populated above. 2442 * So DONT clear those fields here 2443 */ 2444 h.h3->tp_status |= status; 2445 h.h3->tp_len = skb->len; 2446 h.h3->tp_snaplen = snaplen; 2447 h.h3->tp_mac = macoff; 2448 h.h3->tp_net = netoff; 2449 h.h3->tp_sec = ts.tv_sec; 2450 h.h3->tp_nsec = ts.tv_nsec; 2451 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2452 hdrlen = sizeof(*h.h3); 2453 break; 2454 default: 2455 BUG(); 2456 } 2457 2458 sll = h.raw + TPACKET_ALIGN(hdrlen); 2459 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2460 sll->sll_family = AF_PACKET; 2461 sll->sll_hatype = dev->type; 2462 sll->sll_protocol = skb->protocol; 2463 sll->sll_pkttype = skb->pkt_type; 2464 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV))) 2465 sll->sll_ifindex = orig_dev->ifindex; 2466 else 2467 sll->sll_ifindex = dev->ifindex; 2468 2469 smp_mb(); 2470 2471 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2472 if (po->tp_version <= TPACKET_V2) { 2473 u8 *start, *end; 2474 2475 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2476 macoff + snaplen); 2477 2478 for (start = h.raw; start < end; start += PAGE_SIZE) 2479 flush_dcache_page(pgv_to_page(start)); 2480 } 2481 smp_wmb(); 2482 #endif 2483 2484 if (po->tp_version <= TPACKET_V2) { 2485 spin_lock(&sk->sk_receive_queue.lock); 2486 __packet_set_status(po, h.raw, status); 2487 __clear_bit(slot_id, po->rx_ring.rx_owner_map); 2488 spin_unlock(&sk->sk_receive_queue.lock); 2489 sk->sk_data_ready(sk); 2490 } else if (po->tp_version == TPACKET_V3) { 2491 prb_clear_blk_fill_status(&po->rx_ring); 2492 } 2493 2494 drop_n_restore: 2495 if (skb_head != skb->data && skb_shared(skb)) { 2496 skb->data = skb_head; 2497 skb->len = skb_len; 2498 } 2499 drop: 2500 if (!is_drop_n_account) 2501 consume_skb(skb); 2502 else 2503 kfree_skb(skb); 2504 return 0; 2505 2506 drop_n_account: 2507 spin_unlock(&sk->sk_receive_queue.lock); 2508 atomic_inc(&po->tp_drops); 2509 is_drop_n_account = true; 2510 2511 sk->sk_data_ready(sk); 2512 kfree_skb(copy_skb); 2513 goto drop_n_restore; 2514 } 2515 2516 static void tpacket_destruct_skb(struct sk_buff *skb) 2517 { 2518 struct packet_sock *po = pkt_sk(skb->sk); 2519 2520 if (likely(po->tx_ring.pg_vec)) { 2521 void *ph; 2522 __u32 ts; 2523 2524 ph = skb_zcopy_get_nouarg(skb); 2525 packet_dec_pending(&po->tx_ring); 2526 2527 ts = __packet_set_timestamp(po, ph, skb); 2528 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2529 2530 if (!packet_read_pending(&po->tx_ring)) 2531 complete(&po->skb_completion); 2532 } 2533 2534 sock_wfree(skb); 2535 } 2536 2537 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2538 { 2539 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2540 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2541 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2542 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2543 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2544 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2545 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2546 2547 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2548 return -EINVAL; 2549 2550 return 0; 2551 } 2552 2553 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2554 struct virtio_net_hdr *vnet_hdr) 2555 { 2556 if (*len < sizeof(*vnet_hdr)) 2557 return -EINVAL; 2558 *len -= sizeof(*vnet_hdr); 2559 2560 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2561 return -EFAULT; 2562 2563 return __packet_snd_vnet_parse(vnet_hdr, *len); 2564 } 2565 2566 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2567 void *frame, struct net_device *dev, void *data, int tp_len, 2568 __be16 proto, unsigned char *addr, int hlen, int copylen, 2569 const struct sockcm_cookie *sockc) 2570 { 2571 union tpacket_uhdr ph; 2572 int to_write, offset, len, nr_frags, len_max; 2573 struct socket *sock = po->sk.sk_socket; 2574 struct page *page; 2575 int err; 2576 2577 ph.raw = frame; 2578 2579 skb->protocol = proto; 2580 skb->dev = dev; 2581 skb->priority = po->sk.sk_priority; 2582 skb->mark = po->sk.sk_mark; 2583 skb->tstamp = sockc->transmit_time; 2584 skb_setup_tx_timestamp(skb, sockc->tsflags); 2585 skb_zcopy_set_nouarg(skb, ph.raw); 2586 2587 skb_reserve(skb, hlen); 2588 skb_reset_network_header(skb); 2589 2590 to_write = tp_len; 2591 2592 if (sock->type == SOCK_DGRAM) { 2593 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2594 NULL, tp_len); 2595 if (unlikely(err < 0)) 2596 return -EINVAL; 2597 } else if (copylen) { 2598 int hdrlen = min_t(int, copylen, tp_len); 2599 2600 skb_push(skb, dev->hard_header_len); 2601 skb_put(skb, copylen - dev->hard_header_len); 2602 err = skb_store_bits(skb, 0, data, hdrlen); 2603 if (unlikely(err)) 2604 return err; 2605 if (!dev_validate_header(dev, skb->data, hdrlen)) 2606 return -EINVAL; 2607 2608 data += hdrlen; 2609 to_write -= hdrlen; 2610 } 2611 2612 offset = offset_in_page(data); 2613 len_max = PAGE_SIZE - offset; 2614 len = ((to_write > len_max) ? len_max : to_write); 2615 2616 skb->data_len = to_write; 2617 skb->len += to_write; 2618 skb->truesize += to_write; 2619 refcount_add(to_write, &po->sk.sk_wmem_alloc); 2620 2621 while (likely(to_write)) { 2622 nr_frags = skb_shinfo(skb)->nr_frags; 2623 2624 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2625 pr_err("Packet exceed the number of skb frags(%u)\n", 2626 (unsigned int)MAX_SKB_FRAGS); 2627 return -EFAULT; 2628 } 2629 2630 page = pgv_to_page(data); 2631 data += len; 2632 flush_dcache_page(page); 2633 get_page(page); 2634 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2635 to_write -= len; 2636 offset = 0; 2637 len_max = PAGE_SIZE; 2638 len = ((to_write > len_max) ? len_max : to_write); 2639 } 2640 2641 packet_parse_headers(skb, sock); 2642 2643 return tp_len; 2644 } 2645 2646 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2647 int size_max, void **data) 2648 { 2649 union tpacket_uhdr ph; 2650 int tp_len, off; 2651 2652 ph.raw = frame; 2653 2654 switch (po->tp_version) { 2655 case TPACKET_V3: 2656 if (ph.h3->tp_next_offset != 0) { 2657 pr_warn_once("variable sized slot not supported"); 2658 return -EINVAL; 2659 } 2660 tp_len = ph.h3->tp_len; 2661 break; 2662 case TPACKET_V2: 2663 tp_len = ph.h2->tp_len; 2664 break; 2665 default: 2666 tp_len = ph.h1->tp_len; 2667 break; 2668 } 2669 if (unlikely(tp_len > size_max)) { 2670 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); 2671 return -EMSGSIZE; 2672 } 2673 2674 if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) { 2675 int off_min, off_max; 2676 2677 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2678 off_max = po->tx_ring.frame_size - tp_len; 2679 if (po->sk.sk_type == SOCK_DGRAM) { 2680 switch (po->tp_version) { 2681 case TPACKET_V3: 2682 off = ph.h3->tp_net; 2683 break; 2684 case TPACKET_V2: 2685 off = ph.h2->tp_net; 2686 break; 2687 default: 2688 off = ph.h1->tp_net; 2689 break; 2690 } 2691 } else { 2692 switch (po->tp_version) { 2693 case TPACKET_V3: 2694 off = ph.h3->tp_mac; 2695 break; 2696 case TPACKET_V2: 2697 off = ph.h2->tp_mac; 2698 break; 2699 default: 2700 off = ph.h1->tp_mac; 2701 break; 2702 } 2703 } 2704 if (unlikely((off < off_min) || (off_max < off))) 2705 return -EINVAL; 2706 } else { 2707 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2708 } 2709 2710 *data = frame + off; 2711 return tp_len; 2712 } 2713 2714 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2715 { 2716 struct sk_buff *skb = NULL; 2717 struct net_device *dev; 2718 struct virtio_net_hdr *vnet_hdr = NULL; 2719 struct sockcm_cookie sockc; 2720 __be16 proto; 2721 int err, reserve = 0; 2722 void *ph; 2723 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2724 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2725 unsigned char *addr = NULL; 2726 int tp_len, size_max; 2727 void *data; 2728 int len_sum = 0; 2729 int status = TP_STATUS_AVAILABLE; 2730 int hlen, tlen, copylen = 0; 2731 long timeo = 0; 2732 2733 mutex_lock(&po->pg_vec_lock); 2734 2735 /* packet_sendmsg() check on tx_ring.pg_vec was lockless, 2736 * we need to confirm it under protection of pg_vec_lock. 2737 */ 2738 if (unlikely(!po->tx_ring.pg_vec)) { 2739 err = -EBUSY; 2740 goto out; 2741 } 2742 if (likely(saddr == NULL)) { 2743 dev = packet_cached_dev_get(po); 2744 proto = READ_ONCE(po->num); 2745 } else { 2746 err = -EINVAL; 2747 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2748 goto out; 2749 if (msg->msg_namelen < (saddr->sll_halen 2750 + offsetof(struct sockaddr_ll, 2751 sll_addr))) 2752 goto out; 2753 proto = saddr->sll_protocol; 2754 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2755 if (po->sk.sk_socket->type == SOCK_DGRAM) { 2756 if (dev && msg->msg_namelen < dev->addr_len + 2757 offsetof(struct sockaddr_ll, sll_addr)) 2758 goto out_put; 2759 addr = saddr->sll_addr; 2760 } 2761 } 2762 2763 err = -ENXIO; 2764 if (unlikely(dev == NULL)) 2765 goto out; 2766 err = -ENETDOWN; 2767 if (unlikely(!(dev->flags & IFF_UP))) 2768 goto out_put; 2769 2770 sockcm_init(&sockc, &po->sk); 2771 if (msg->msg_controllen) { 2772 err = sock_cmsg_send(&po->sk, msg, &sockc); 2773 if (unlikely(err)) 2774 goto out_put; 2775 } 2776 2777 if (po->sk.sk_socket->type == SOCK_RAW) 2778 reserve = dev->hard_header_len; 2779 size_max = po->tx_ring.frame_size 2780 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2781 2782 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && 2783 !packet_sock_flag(po, PACKET_SOCK_HAS_VNET_HDR)) 2784 size_max = dev->mtu + reserve + VLAN_HLEN; 2785 2786 reinit_completion(&po->skb_completion); 2787 2788 do { 2789 ph = packet_current_frame(po, &po->tx_ring, 2790 TP_STATUS_SEND_REQUEST); 2791 if (unlikely(ph == NULL)) { 2792 if (need_wait && skb) { 2793 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT); 2794 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo); 2795 if (timeo <= 0) { 2796 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS; 2797 goto out_put; 2798 } 2799 } 2800 /* check for additional frames */ 2801 continue; 2802 } 2803 2804 skb = NULL; 2805 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2806 if (tp_len < 0) 2807 goto tpacket_error; 2808 2809 status = TP_STATUS_SEND_REQUEST; 2810 hlen = LL_RESERVED_SPACE(dev); 2811 tlen = dev->needed_tailroom; 2812 if (packet_sock_flag(po, PACKET_SOCK_HAS_VNET_HDR)) { 2813 vnet_hdr = data; 2814 data += sizeof(*vnet_hdr); 2815 tp_len -= sizeof(*vnet_hdr); 2816 if (tp_len < 0 || 2817 __packet_snd_vnet_parse(vnet_hdr, tp_len)) { 2818 tp_len = -EINVAL; 2819 goto tpacket_error; 2820 } 2821 copylen = __virtio16_to_cpu(vio_le(), 2822 vnet_hdr->hdr_len); 2823 } 2824 copylen = max_t(int, copylen, dev->hard_header_len); 2825 skb = sock_alloc_send_skb(&po->sk, 2826 hlen + tlen + sizeof(struct sockaddr_ll) + 2827 (copylen - dev->hard_header_len), 2828 !need_wait, &err); 2829 2830 if (unlikely(skb == NULL)) { 2831 /* we assume the socket was initially writeable ... */ 2832 if (likely(len_sum > 0)) 2833 err = len_sum; 2834 goto out_status; 2835 } 2836 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, 2837 addr, hlen, copylen, &sockc); 2838 if (likely(tp_len >= 0) && 2839 tp_len > dev->mtu + reserve && 2840 !packet_sock_flag(po, PACKET_SOCK_HAS_VNET_HDR) && 2841 !packet_extra_vlan_len_allowed(dev, skb)) 2842 tp_len = -EMSGSIZE; 2843 2844 if (unlikely(tp_len < 0)) { 2845 tpacket_error: 2846 if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) { 2847 __packet_set_status(po, ph, 2848 TP_STATUS_AVAILABLE); 2849 packet_increment_head(&po->tx_ring); 2850 kfree_skb(skb); 2851 continue; 2852 } else { 2853 status = TP_STATUS_WRONG_FORMAT; 2854 err = tp_len; 2855 goto out_status; 2856 } 2857 } 2858 2859 if (packet_sock_flag(po, PACKET_SOCK_HAS_VNET_HDR)) { 2860 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) { 2861 tp_len = -EINVAL; 2862 goto tpacket_error; 2863 } 2864 virtio_net_hdr_set_proto(skb, vnet_hdr); 2865 } 2866 2867 skb->destructor = tpacket_destruct_skb; 2868 __packet_set_status(po, ph, TP_STATUS_SENDING); 2869 packet_inc_pending(&po->tx_ring); 2870 2871 status = TP_STATUS_SEND_REQUEST; 2872 err = packet_xmit(po, skb); 2873 if (unlikely(err != 0)) { 2874 if (err > 0) 2875 err = net_xmit_errno(err); 2876 if (err && __packet_get_status(po, ph) == 2877 TP_STATUS_AVAILABLE) { 2878 /* skb was destructed already */ 2879 skb = NULL; 2880 goto out_status; 2881 } 2882 /* 2883 * skb was dropped but not destructed yet; 2884 * let's treat it like congestion or err < 0 2885 */ 2886 err = 0; 2887 } 2888 packet_increment_head(&po->tx_ring); 2889 len_sum += tp_len; 2890 } while (likely((ph != NULL) || 2891 /* Note: packet_read_pending() might be slow if we have 2892 * to call it as it's per_cpu variable, but in fast-path 2893 * we already short-circuit the loop with the first 2894 * condition, and luckily don't have to go that path 2895 * anyway. 2896 */ 2897 (need_wait && packet_read_pending(&po->tx_ring)))); 2898 2899 err = len_sum; 2900 goto out_put; 2901 2902 out_status: 2903 __packet_set_status(po, ph, status); 2904 kfree_skb(skb); 2905 out_put: 2906 dev_put(dev); 2907 out: 2908 mutex_unlock(&po->pg_vec_lock); 2909 return err; 2910 } 2911 2912 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2913 size_t reserve, size_t len, 2914 size_t linear, int noblock, 2915 int *err) 2916 { 2917 struct sk_buff *skb; 2918 2919 /* Under a page? Don't bother with paged skb. */ 2920 if (prepad + len < PAGE_SIZE || !linear) 2921 linear = len; 2922 2923 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 2924 err, 0); 2925 if (!skb) 2926 return NULL; 2927 2928 skb_reserve(skb, reserve); 2929 skb_put(skb, linear); 2930 skb->data_len = len - linear; 2931 skb->len += len - linear; 2932 2933 return skb; 2934 } 2935 2936 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 2937 { 2938 struct sock *sk = sock->sk; 2939 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2940 struct sk_buff *skb; 2941 struct net_device *dev; 2942 __be16 proto; 2943 unsigned char *addr = NULL; 2944 int err, reserve = 0; 2945 struct sockcm_cookie sockc; 2946 struct virtio_net_hdr vnet_hdr = { 0 }; 2947 int offset = 0; 2948 struct packet_sock *po = pkt_sk(sk); 2949 bool has_vnet_hdr = false; 2950 int hlen, tlen, linear; 2951 int extra_len = 0; 2952 2953 /* 2954 * Get and verify the address. 2955 */ 2956 2957 if (likely(saddr == NULL)) { 2958 dev = packet_cached_dev_get(po); 2959 proto = READ_ONCE(po->num); 2960 } else { 2961 err = -EINVAL; 2962 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2963 goto out; 2964 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 2965 goto out; 2966 proto = saddr->sll_protocol; 2967 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 2968 if (sock->type == SOCK_DGRAM) { 2969 if (dev && msg->msg_namelen < dev->addr_len + 2970 offsetof(struct sockaddr_ll, sll_addr)) 2971 goto out_unlock; 2972 addr = saddr->sll_addr; 2973 } 2974 } 2975 2976 err = -ENXIO; 2977 if (unlikely(dev == NULL)) 2978 goto out_unlock; 2979 err = -ENETDOWN; 2980 if (unlikely(!(dev->flags & IFF_UP))) 2981 goto out_unlock; 2982 2983 sockcm_init(&sockc, sk); 2984 sockc.mark = sk->sk_mark; 2985 if (msg->msg_controllen) { 2986 err = sock_cmsg_send(sk, msg, &sockc); 2987 if (unlikely(err)) 2988 goto out_unlock; 2989 } 2990 2991 if (sock->type == SOCK_RAW) 2992 reserve = dev->hard_header_len; 2993 if (packet_sock_flag(po, PACKET_SOCK_HAS_VNET_HDR)) { 2994 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); 2995 if (err) 2996 goto out_unlock; 2997 has_vnet_hdr = true; 2998 } 2999 3000 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 3001 if (!netif_supports_nofcs(dev)) { 3002 err = -EPROTONOSUPPORT; 3003 goto out_unlock; 3004 } 3005 extra_len = 4; /* We're doing our own CRC */ 3006 } 3007 3008 err = -EMSGSIZE; 3009 if (!vnet_hdr.gso_type && 3010 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 3011 goto out_unlock; 3012 3013 err = -ENOBUFS; 3014 hlen = LL_RESERVED_SPACE(dev); 3015 tlen = dev->needed_tailroom; 3016 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); 3017 linear = max(linear, min_t(int, len, dev->hard_header_len)); 3018 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, 3019 msg->msg_flags & MSG_DONTWAIT, &err); 3020 if (skb == NULL) 3021 goto out_unlock; 3022 3023 skb_reset_network_header(skb); 3024 3025 err = -EINVAL; 3026 if (sock->type == SOCK_DGRAM) { 3027 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 3028 if (unlikely(offset < 0)) 3029 goto out_free; 3030 } else if (reserve) { 3031 skb_reserve(skb, -reserve); 3032 if (len < reserve + sizeof(struct ipv6hdr) && 3033 dev->min_header_len != dev->hard_header_len) 3034 skb_reset_network_header(skb); 3035 } 3036 3037 /* Returns -EFAULT on error */ 3038 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 3039 if (err) 3040 goto out_free; 3041 3042 if ((sock->type == SOCK_RAW && 3043 !dev_validate_header(dev, skb->data, len)) || !skb->len) { 3044 err = -EINVAL; 3045 goto out_free; 3046 } 3047 3048 skb_setup_tx_timestamp(skb, sockc.tsflags); 3049 3050 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 3051 !packet_extra_vlan_len_allowed(dev, skb)) { 3052 err = -EMSGSIZE; 3053 goto out_free; 3054 } 3055 3056 skb->protocol = proto; 3057 skb->dev = dev; 3058 skb->priority = sk->sk_priority; 3059 skb->mark = sockc.mark; 3060 skb->tstamp = sockc.transmit_time; 3061 3062 if (unlikely(extra_len == 4)) 3063 skb->no_fcs = 1; 3064 3065 packet_parse_headers(skb, sock); 3066 3067 if (has_vnet_hdr) { 3068 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 3069 if (err) 3070 goto out_free; 3071 len += sizeof(vnet_hdr); 3072 virtio_net_hdr_set_proto(skb, &vnet_hdr); 3073 } 3074 3075 err = packet_xmit(po, skb); 3076 3077 if (unlikely(err != 0)) { 3078 if (err > 0) 3079 err = net_xmit_errno(err); 3080 if (err) 3081 goto out_unlock; 3082 } 3083 3084 dev_put(dev); 3085 3086 return len; 3087 3088 out_free: 3089 kfree_skb(skb); 3090 out_unlock: 3091 dev_put(dev); 3092 out: 3093 return err; 3094 } 3095 3096 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 3097 { 3098 struct sock *sk = sock->sk; 3099 struct packet_sock *po = pkt_sk(sk); 3100 3101 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy. 3102 * tpacket_snd() will redo the check safely. 3103 */ 3104 if (data_race(po->tx_ring.pg_vec)) 3105 return tpacket_snd(po, msg); 3106 3107 return packet_snd(sock, msg, len); 3108 } 3109 3110 /* 3111 * Close a PACKET socket. This is fairly simple. We immediately go 3112 * to 'closed' state and remove our protocol entry in the device list. 3113 */ 3114 3115 static int packet_release(struct socket *sock) 3116 { 3117 struct sock *sk = sock->sk; 3118 struct packet_sock *po; 3119 struct packet_fanout *f; 3120 struct net *net; 3121 union tpacket_req_u req_u; 3122 3123 if (!sk) 3124 return 0; 3125 3126 net = sock_net(sk); 3127 po = pkt_sk(sk); 3128 3129 mutex_lock(&net->packet.sklist_lock); 3130 sk_del_node_init_rcu(sk); 3131 mutex_unlock(&net->packet.sklist_lock); 3132 3133 sock_prot_inuse_add(net, sk->sk_prot, -1); 3134 3135 spin_lock(&po->bind_lock); 3136 unregister_prot_hook(sk, false); 3137 packet_cached_dev_reset(po); 3138 3139 if (po->prot_hook.dev) { 3140 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker); 3141 po->prot_hook.dev = NULL; 3142 } 3143 spin_unlock(&po->bind_lock); 3144 3145 packet_flush_mclist(sk); 3146 3147 lock_sock(sk); 3148 if (po->rx_ring.pg_vec) { 3149 memset(&req_u, 0, sizeof(req_u)); 3150 packet_set_ring(sk, &req_u, 1, 0); 3151 } 3152 3153 if (po->tx_ring.pg_vec) { 3154 memset(&req_u, 0, sizeof(req_u)); 3155 packet_set_ring(sk, &req_u, 1, 1); 3156 } 3157 release_sock(sk); 3158 3159 f = fanout_release(sk); 3160 3161 synchronize_net(); 3162 3163 kfree(po->rollover); 3164 if (f) { 3165 fanout_release_data(f); 3166 kvfree(f); 3167 } 3168 /* 3169 * Now the socket is dead. No more input will appear. 3170 */ 3171 sock_orphan(sk); 3172 sock->sk = NULL; 3173 3174 /* Purge queues */ 3175 3176 skb_queue_purge(&sk->sk_receive_queue); 3177 packet_free_pending(po); 3178 3179 sock_put(sk); 3180 return 0; 3181 } 3182 3183 /* 3184 * Attach a packet hook. 3185 */ 3186 3187 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3188 __be16 proto) 3189 { 3190 struct packet_sock *po = pkt_sk(sk); 3191 struct net_device *dev = NULL; 3192 bool unlisted = false; 3193 bool need_rehook; 3194 int ret = 0; 3195 3196 lock_sock(sk); 3197 spin_lock(&po->bind_lock); 3198 rcu_read_lock(); 3199 3200 if (po->fanout) { 3201 ret = -EINVAL; 3202 goto out_unlock; 3203 } 3204 3205 if (name) { 3206 dev = dev_get_by_name_rcu(sock_net(sk), name); 3207 if (!dev) { 3208 ret = -ENODEV; 3209 goto out_unlock; 3210 } 3211 } else if (ifindex) { 3212 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3213 if (!dev) { 3214 ret = -ENODEV; 3215 goto out_unlock; 3216 } 3217 } 3218 3219 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev; 3220 3221 if (need_rehook) { 3222 dev_hold(dev); 3223 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 3224 rcu_read_unlock(); 3225 /* prevents packet_notifier() from calling 3226 * register_prot_hook() 3227 */ 3228 WRITE_ONCE(po->num, 0); 3229 __unregister_prot_hook(sk, true); 3230 rcu_read_lock(); 3231 if (dev) 3232 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3233 dev->ifindex); 3234 } 3235 3236 BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING)); 3237 WRITE_ONCE(po->num, proto); 3238 po->prot_hook.type = proto; 3239 3240 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker); 3241 3242 if (unlikely(unlisted)) { 3243 po->prot_hook.dev = NULL; 3244 WRITE_ONCE(po->ifindex, -1); 3245 packet_cached_dev_reset(po); 3246 } else { 3247 netdev_hold(dev, &po->prot_hook.dev_tracker, 3248 GFP_ATOMIC); 3249 po->prot_hook.dev = dev; 3250 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0); 3251 packet_cached_dev_assign(po, dev); 3252 } 3253 dev_put(dev); 3254 } 3255 3256 if (proto == 0 || !need_rehook) 3257 goto out_unlock; 3258 3259 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3260 register_prot_hook(sk); 3261 } else { 3262 sk->sk_err = ENETDOWN; 3263 if (!sock_flag(sk, SOCK_DEAD)) 3264 sk_error_report(sk); 3265 } 3266 3267 out_unlock: 3268 rcu_read_unlock(); 3269 spin_unlock(&po->bind_lock); 3270 release_sock(sk); 3271 return ret; 3272 } 3273 3274 /* 3275 * Bind a packet socket to a device 3276 */ 3277 3278 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, 3279 int addr_len) 3280 { 3281 struct sock *sk = sock->sk; 3282 char name[sizeof(uaddr->sa_data_min) + 1]; 3283 3284 /* 3285 * Check legality 3286 */ 3287 3288 if (addr_len != sizeof(struct sockaddr)) 3289 return -EINVAL; 3290 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be 3291 * zero-terminated. 3292 */ 3293 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min)); 3294 name[sizeof(uaddr->sa_data_min)] = 0; 3295 3296 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); 3297 } 3298 3299 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 3300 { 3301 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3302 struct sock *sk = sock->sk; 3303 3304 /* 3305 * Check legality 3306 */ 3307 3308 if (addr_len < sizeof(struct sockaddr_ll)) 3309 return -EINVAL; 3310 if (sll->sll_family != AF_PACKET) 3311 return -EINVAL; 3312 3313 return packet_do_bind(sk, NULL, sll->sll_ifindex, 3314 sll->sll_protocol ? : pkt_sk(sk)->num); 3315 } 3316 3317 static struct proto packet_proto = { 3318 .name = "PACKET", 3319 .owner = THIS_MODULE, 3320 .obj_size = sizeof(struct packet_sock), 3321 }; 3322 3323 /* 3324 * Create a packet of type SOCK_PACKET. 3325 */ 3326 3327 static int packet_create(struct net *net, struct socket *sock, int protocol, 3328 int kern) 3329 { 3330 struct sock *sk; 3331 struct packet_sock *po; 3332 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3333 int err; 3334 3335 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3336 return -EPERM; 3337 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3338 sock->type != SOCK_PACKET) 3339 return -ESOCKTNOSUPPORT; 3340 3341 sock->state = SS_UNCONNECTED; 3342 3343 err = -ENOBUFS; 3344 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3345 if (sk == NULL) 3346 goto out; 3347 3348 sock->ops = &packet_ops; 3349 if (sock->type == SOCK_PACKET) 3350 sock->ops = &packet_ops_spkt; 3351 3352 sock_init_data(sock, sk); 3353 3354 po = pkt_sk(sk); 3355 init_completion(&po->skb_completion); 3356 sk->sk_family = PF_PACKET; 3357 po->num = proto; 3358 3359 err = packet_alloc_pending(po); 3360 if (err) 3361 goto out2; 3362 3363 packet_cached_dev_reset(po); 3364 3365 sk->sk_destruct = packet_sock_destruct; 3366 3367 /* 3368 * Attach a protocol block 3369 */ 3370 3371 spin_lock_init(&po->bind_lock); 3372 mutex_init(&po->pg_vec_lock); 3373 po->rollover = NULL; 3374 po->prot_hook.func = packet_rcv; 3375 3376 if (sock->type == SOCK_PACKET) 3377 po->prot_hook.func = packet_rcv_spkt; 3378 3379 po->prot_hook.af_packet_priv = sk; 3380 po->prot_hook.af_packet_net = sock_net(sk); 3381 3382 if (proto) { 3383 po->prot_hook.type = proto; 3384 __register_prot_hook(sk); 3385 } 3386 3387 mutex_lock(&net->packet.sklist_lock); 3388 sk_add_node_tail_rcu(sk, &net->packet.sklist); 3389 mutex_unlock(&net->packet.sklist_lock); 3390 3391 sock_prot_inuse_add(net, &packet_proto, 1); 3392 3393 return 0; 3394 out2: 3395 sk_free(sk); 3396 out: 3397 return err; 3398 } 3399 3400 /* 3401 * Pull a packet from our receive queue and hand it to the user. 3402 * If necessary we block. 3403 */ 3404 3405 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3406 int flags) 3407 { 3408 struct sock *sk = sock->sk; 3409 struct sk_buff *skb; 3410 int copied, err; 3411 int vnet_hdr_len = 0; 3412 unsigned int origlen = 0; 3413 3414 err = -EINVAL; 3415 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3416 goto out; 3417 3418 #if 0 3419 /* What error should we return now? EUNATTACH? */ 3420 if (pkt_sk(sk)->ifindex < 0) 3421 return -ENODEV; 3422 #endif 3423 3424 if (flags & MSG_ERRQUEUE) { 3425 err = sock_recv_errqueue(sk, msg, len, 3426 SOL_PACKET, PACKET_TX_TIMESTAMP); 3427 goto out; 3428 } 3429 3430 /* 3431 * Call the generic datagram receiver. This handles all sorts 3432 * of horrible races and re-entrancy so we can forget about it 3433 * in the protocol layers. 3434 * 3435 * Now it will return ENETDOWN, if device have just gone down, 3436 * but then it will block. 3437 */ 3438 3439 skb = skb_recv_datagram(sk, flags, &err); 3440 3441 /* 3442 * An error occurred so return it. Because skb_recv_datagram() 3443 * handles the blocking we don't see and worry about blocking 3444 * retries. 3445 */ 3446 3447 if (skb == NULL) 3448 goto out; 3449 3450 packet_rcv_try_clear_pressure(pkt_sk(sk)); 3451 3452 if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_HAS_VNET_HDR)) { 3453 err = packet_rcv_vnet(msg, skb, &len); 3454 if (err) 3455 goto out_free; 3456 vnet_hdr_len = sizeof(struct virtio_net_hdr); 3457 } 3458 3459 /* You lose any data beyond the buffer you gave. If it worries 3460 * a user program they can ask the device for its MTU 3461 * anyway. 3462 */ 3463 copied = skb->len; 3464 if (copied > len) { 3465 copied = len; 3466 msg->msg_flags |= MSG_TRUNC; 3467 } 3468 3469 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3470 if (err) 3471 goto out_free; 3472 3473 if (sock->type != SOCK_PACKET) { 3474 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3475 3476 /* Original length was stored in sockaddr_ll fields */ 3477 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3478 sll->sll_family = AF_PACKET; 3479 sll->sll_protocol = skb->protocol; 3480 } 3481 3482 sock_recv_cmsgs(msg, sk, skb); 3483 3484 if (msg->msg_name) { 3485 const size_t max_len = min(sizeof(skb->cb), 3486 sizeof(struct sockaddr_storage)); 3487 int copy_len; 3488 3489 /* If the address length field is there to be filled 3490 * in, we fill it in now. 3491 */ 3492 if (sock->type == SOCK_PACKET) { 3493 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3494 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3495 copy_len = msg->msg_namelen; 3496 } else { 3497 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3498 3499 msg->msg_namelen = sll->sll_halen + 3500 offsetof(struct sockaddr_ll, sll_addr); 3501 copy_len = msg->msg_namelen; 3502 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { 3503 memset(msg->msg_name + 3504 offsetof(struct sockaddr_ll, sll_addr), 3505 0, sizeof(sll->sll_addr)); 3506 msg->msg_namelen = sizeof(struct sockaddr_ll); 3507 } 3508 } 3509 if (WARN_ON_ONCE(copy_len > max_len)) { 3510 copy_len = max_len; 3511 msg->msg_namelen = copy_len; 3512 } 3513 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); 3514 } 3515 3516 if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) { 3517 struct tpacket_auxdata aux; 3518 3519 aux.tp_status = TP_STATUS_USER; 3520 if (skb->ip_summed == CHECKSUM_PARTIAL) 3521 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3522 else if (skb->pkt_type != PACKET_OUTGOING && 3523 skb_csum_unnecessary(skb)) 3524 aux.tp_status |= TP_STATUS_CSUM_VALID; 3525 if (skb_is_gso(skb) && skb_is_gso_tcp(skb)) 3526 aux.tp_status |= TP_STATUS_GSO_TCP; 3527 3528 aux.tp_len = origlen; 3529 aux.tp_snaplen = skb->len; 3530 aux.tp_mac = 0; 3531 aux.tp_net = skb_network_offset(skb); 3532 if (skb_vlan_tag_present(skb)) { 3533 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3534 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3535 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3536 } else { 3537 aux.tp_vlan_tci = 0; 3538 aux.tp_vlan_tpid = 0; 3539 } 3540 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3541 } 3542 3543 /* 3544 * Free or return the buffer as appropriate. Again this 3545 * hides all the races and re-entrancy issues from us. 3546 */ 3547 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3548 3549 out_free: 3550 skb_free_datagram(sk, skb); 3551 out: 3552 return err; 3553 } 3554 3555 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3556 int peer) 3557 { 3558 struct net_device *dev; 3559 struct sock *sk = sock->sk; 3560 3561 if (peer) 3562 return -EOPNOTSUPP; 3563 3564 uaddr->sa_family = AF_PACKET; 3565 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min)); 3566 rcu_read_lock(); 3567 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex)); 3568 if (dev) 3569 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min)); 3570 rcu_read_unlock(); 3571 3572 return sizeof(*uaddr); 3573 } 3574 3575 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3576 int peer) 3577 { 3578 struct net_device *dev; 3579 struct sock *sk = sock->sk; 3580 struct packet_sock *po = pkt_sk(sk); 3581 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3582 int ifindex; 3583 3584 if (peer) 3585 return -EOPNOTSUPP; 3586 3587 ifindex = READ_ONCE(po->ifindex); 3588 sll->sll_family = AF_PACKET; 3589 sll->sll_ifindex = ifindex; 3590 sll->sll_protocol = READ_ONCE(po->num); 3591 sll->sll_pkttype = 0; 3592 rcu_read_lock(); 3593 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3594 if (dev) { 3595 sll->sll_hatype = dev->type; 3596 sll->sll_halen = dev->addr_len; 3597 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); 3598 } else { 3599 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3600 sll->sll_halen = 0; 3601 } 3602 rcu_read_unlock(); 3603 3604 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3605 } 3606 3607 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3608 int what) 3609 { 3610 switch (i->type) { 3611 case PACKET_MR_MULTICAST: 3612 if (i->alen != dev->addr_len) 3613 return -EINVAL; 3614 if (what > 0) 3615 return dev_mc_add(dev, i->addr); 3616 else 3617 return dev_mc_del(dev, i->addr); 3618 break; 3619 case PACKET_MR_PROMISC: 3620 return dev_set_promiscuity(dev, what); 3621 case PACKET_MR_ALLMULTI: 3622 return dev_set_allmulti(dev, what); 3623 case PACKET_MR_UNICAST: 3624 if (i->alen != dev->addr_len) 3625 return -EINVAL; 3626 if (what > 0) 3627 return dev_uc_add(dev, i->addr); 3628 else 3629 return dev_uc_del(dev, i->addr); 3630 break; 3631 default: 3632 break; 3633 } 3634 return 0; 3635 } 3636 3637 static void packet_dev_mclist_delete(struct net_device *dev, 3638 struct packet_mclist **mlp) 3639 { 3640 struct packet_mclist *ml; 3641 3642 while ((ml = *mlp) != NULL) { 3643 if (ml->ifindex == dev->ifindex) { 3644 packet_dev_mc(dev, ml, -1); 3645 *mlp = ml->next; 3646 kfree(ml); 3647 } else 3648 mlp = &ml->next; 3649 } 3650 } 3651 3652 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3653 { 3654 struct packet_sock *po = pkt_sk(sk); 3655 struct packet_mclist *ml, *i; 3656 struct net_device *dev; 3657 int err; 3658 3659 rtnl_lock(); 3660 3661 err = -ENODEV; 3662 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3663 if (!dev) 3664 goto done; 3665 3666 err = -EINVAL; 3667 if (mreq->mr_alen > dev->addr_len) 3668 goto done; 3669 3670 err = -ENOBUFS; 3671 i = kmalloc(sizeof(*i), GFP_KERNEL); 3672 if (i == NULL) 3673 goto done; 3674 3675 err = 0; 3676 for (ml = po->mclist; ml; ml = ml->next) { 3677 if (ml->ifindex == mreq->mr_ifindex && 3678 ml->type == mreq->mr_type && 3679 ml->alen == mreq->mr_alen && 3680 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3681 ml->count++; 3682 /* Free the new element ... */ 3683 kfree(i); 3684 goto done; 3685 } 3686 } 3687 3688 i->type = mreq->mr_type; 3689 i->ifindex = mreq->mr_ifindex; 3690 i->alen = mreq->mr_alen; 3691 memcpy(i->addr, mreq->mr_address, i->alen); 3692 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3693 i->count = 1; 3694 i->next = po->mclist; 3695 po->mclist = i; 3696 err = packet_dev_mc(dev, i, 1); 3697 if (err) { 3698 po->mclist = i->next; 3699 kfree(i); 3700 } 3701 3702 done: 3703 rtnl_unlock(); 3704 return err; 3705 } 3706 3707 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3708 { 3709 struct packet_mclist *ml, **mlp; 3710 3711 rtnl_lock(); 3712 3713 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3714 if (ml->ifindex == mreq->mr_ifindex && 3715 ml->type == mreq->mr_type && 3716 ml->alen == mreq->mr_alen && 3717 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3718 if (--ml->count == 0) { 3719 struct net_device *dev; 3720 *mlp = ml->next; 3721 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3722 if (dev) 3723 packet_dev_mc(dev, ml, -1); 3724 kfree(ml); 3725 } 3726 break; 3727 } 3728 } 3729 rtnl_unlock(); 3730 return 0; 3731 } 3732 3733 static void packet_flush_mclist(struct sock *sk) 3734 { 3735 struct packet_sock *po = pkt_sk(sk); 3736 struct packet_mclist *ml; 3737 3738 if (!po->mclist) 3739 return; 3740 3741 rtnl_lock(); 3742 while ((ml = po->mclist) != NULL) { 3743 struct net_device *dev; 3744 3745 po->mclist = ml->next; 3746 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3747 if (dev != NULL) 3748 packet_dev_mc(dev, ml, -1); 3749 kfree(ml); 3750 } 3751 rtnl_unlock(); 3752 } 3753 3754 static int 3755 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval, 3756 unsigned int optlen) 3757 { 3758 struct sock *sk = sock->sk; 3759 struct packet_sock *po = pkt_sk(sk); 3760 int ret; 3761 3762 if (level != SOL_PACKET) 3763 return -ENOPROTOOPT; 3764 3765 switch (optname) { 3766 case PACKET_ADD_MEMBERSHIP: 3767 case PACKET_DROP_MEMBERSHIP: 3768 { 3769 struct packet_mreq_max mreq; 3770 int len = optlen; 3771 memset(&mreq, 0, sizeof(mreq)); 3772 if (len < sizeof(struct packet_mreq)) 3773 return -EINVAL; 3774 if (len > sizeof(mreq)) 3775 len = sizeof(mreq); 3776 if (copy_from_sockptr(&mreq, optval, len)) 3777 return -EFAULT; 3778 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3779 return -EINVAL; 3780 if (optname == PACKET_ADD_MEMBERSHIP) 3781 ret = packet_mc_add(sk, &mreq); 3782 else 3783 ret = packet_mc_drop(sk, &mreq); 3784 return ret; 3785 } 3786 3787 case PACKET_RX_RING: 3788 case PACKET_TX_RING: 3789 { 3790 union tpacket_req_u req_u; 3791 int len; 3792 3793 lock_sock(sk); 3794 switch (po->tp_version) { 3795 case TPACKET_V1: 3796 case TPACKET_V2: 3797 len = sizeof(req_u.req); 3798 break; 3799 case TPACKET_V3: 3800 default: 3801 len = sizeof(req_u.req3); 3802 break; 3803 } 3804 if (optlen < len) { 3805 ret = -EINVAL; 3806 } else { 3807 if (copy_from_sockptr(&req_u.req, optval, len)) 3808 ret = -EFAULT; 3809 else 3810 ret = packet_set_ring(sk, &req_u, 0, 3811 optname == PACKET_TX_RING); 3812 } 3813 release_sock(sk); 3814 return ret; 3815 } 3816 case PACKET_COPY_THRESH: 3817 { 3818 int val; 3819 3820 if (optlen != sizeof(val)) 3821 return -EINVAL; 3822 if (copy_from_sockptr(&val, optval, sizeof(val))) 3823 return -EFAULT; 3824 3825 pkt_sk(sk)->copy_thresh = val; 3826 return 0; 3827 } 3828 case PACKET_VERSION: 3829 { 3830 int val; 3831 3832 if (optlen != sizeof(val)) 3833 return -EINVAL; 3834 if (copy_from_sockptr(&val, optval, sizeof(val))) 3835 return -EFAULT; 3836 switch (val) { 3837 case TPACKET_V1: 3838 case TPACKET_V2: 3839 case TPACKET_V3: 3840 break; 3841 default: 3842 return -EINVAL; 3843 } 3844 lock_sock(sk); 3845 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3846 ret = -EBUSY; 3847 } else { 3848 po->tp_version = val; 3849 ret = 0; 3850 } 3851 release_sock(sk); 3852 return ret; 3853 } 3854 case PACKET_RESERVE: 3855 { 3856 unsigned int val; 3857 3858 if (optlen != sizeof(val)) 3859 return -EINVAL; 3860 if (copy_from_sockptr(&val, optval, sizeof(val))) 3861 return -EFAULT; 3862 if (val > INT_MAX) 3863 return -EINVAL; 3864 lock_sock(sk); 3865 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3866 ret = -EBUSY; 3867 } else { 3868 po->tp_reserve = val; 3869 ret = 0; 3870 } 3871 release_sock(sk); 3872 return ret; 3873 } 3874 case PACKET_LOSS: 3875 { 3876 unsigned int val; 3877 3878 if (optlen != sizeof(val)) 3879 return -EINVAL; 3880 if (copy_from_sockptr(&val, optval, sizeof(val))) 3881 return -EFAULT; 3882 3883 lock_sock(sk); 3884 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3885 ret = -EBUSY; 3886 } else { 3887 packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val); 3888 ret = 0; 3889 } 3890 release_sock(sk); 3891 return ret; 3892 } 3893 case PACKET_AUXDATA: 3894 { 3895 int val; 3896 3897 if (optlen < sizeof(val)) 3898 return -EINVAL; 3899 if (copy_from_sockptr(&val, optval, sizeof(val))) 3900 return -EFAULT; 3901 3902 packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val); 3903 return 0; 3904 } 3905 case PACKET_ORIGDEV: 3906 { 3907 int val; 3908 3909 if (optlen < sizeof(val)) 3910 return -EINVAL; 3911 if (copy_from_sockptr(&val, optval, sizeof(val))) 3912 return -EFAULT; 3913 3914 packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val); 3915 return 0; 3916 } 3917 case PACKET_VNET_HDR: 3918 { 3919 int val; 3920 3921 if (sock->type != SOCK_RAW) 3922 return -EINVAL; 3923 if (optlen < sizeof(val)) 3924 return -EINVAL; 3925 if (copy_from_sockptr(&val, optval, sizeof(val))) 3926 return -EFAULT; 3927 3928 lock_sock(sk); 3929 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3930 ret = -EBUSY; 3931 } else { 3932 packet_sock_flag_set(po, PACKET_SOCK_HAS_VNET_HDR, val); 3933 ret = 0; 3934 } 3935 release_sock(sk); 3936 return ret; 3937 } 3938 case PACKET_TIMESTAMP: 3939 { 3940 int val; 3941 3942 if (optlen != sizeof(val)) 3943 return -EINVAL; 3944 if (copy_from_sockptr(&val, optval, sizeof(val))) 3945 return -EFAULT; 3946 3947 WRITE_ONCE(po->tp_tstamp, val); 3948 return 0; 3949 } 3950 case PACKET_FANOUT: 3951 { 3952 struct fanout_args args = { 0 }; 3953 3954 if (optlen != sizeof(int) && optlen != sizeof(args)) 3955 return -EINVAL; 3956 if (copy_from_sockptr(&args, optval, optlen)) 3957 return -EFAULT; 3958 3959 return fanout_add(sk, &args); 3960 } 3961 case PACKET_FANOUT_DATA: 3962 { 3963 /* Paired with the WRITE_ONCE() in fanout_add() */ 3964 if (!READ_ONCE(po->fanout)) 3965 return -EINVAL; 3966 3967 return fanout_set_data(po, optval, optlen); 3968 } 3969 case PACKET_IGNORE_OUTGOING: 3970 { 3971 int val; 3972 3973 if (optlen != sizeof(val)) 3974 return -EINVAL; 3975 if (copy_from_sockptr(&val, optval, sizeof(val))) 3976 return -EFAULT; 3977 if (val < 0 || val > 1) 3978 return -EINVAL; 3979 3980 po->prot_hook.ignore_outgoing = !!val; 3981 return 0; 3982 } 3983 case PACKET_TX_HAS_OFF: 3984 { 3985 unsigned int val; 3986 3987 if (optlen != sizeof(val)) 3988 return -EINVAL; 3989 if (copy_from_sockptr(&val, optval, sizeof(val))) 3990 return -EFAULT; 3991 3992 lock_sock(sk); 3993 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec) 3994 packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val); 3995 3996 release_sock(sk); 3997 return 0; 3998 } 3999 case PACKET_QDISC_BYPASS: 4000 { 4001 int val; 4002 4003 if (optlen != sizeof(val)) 4004 return -EINVAL; 4005 if (copy_from_sockptr(&val, optval, sizeof(val))) 4006 return -EFAULT; 4007 4008 packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val); 4009 return 0; 4010 } 4011 default: 4012 return -ENOPROTOOPT; 4013 } 4014 } 4015 4016 static int packet_getsockopt(struct socket *sock, int level, int optname, 4017 char __user *optval, int __user *optlen) 4018 { 4019 int len; 4020 int val, lv = sizeof(val); 4021 struct sock *sk = sock->sk; 4022 struct packet_sock *po = pkt_sk(sk); 4023 void *data = &val; 4024 union tpacket_stats_u st; 4025 struct tpacket_rollover_stats rstats; 4026 int drops; 4027 4028 if (level != SOL_PACKET) 4029 return -ENOPROTOOPT; 4030 4031 if (get_user(len, optlen)) 4032 return -EFAULT; 4033 4034 if (len < 0) 4035 return -EINVAL; 4036 4037 switch (optname) { 4038 case PACKET_STATISTICS: 4039 spin_lock_bh(&sk->sk_receive_queue.lock); 4040 memcpy(&st, &po->stats, sizeof(st)); 4041 memset(&po->stats, 0, sizeof(po->stats)); 4042 spin_unlock_bh(&sk->sk_receive_queue.lock); 4043 drops = atomic_xchg(&po->tp_drops, 0); 4044 4045 if (po->tp_version == TPACKET_V3) { 4046 lv = sizeof(struct tpacket_stats_v3); 4047 st.stats3.tp_drops = drops; 4048 st.stats3.tp_packets += drops; 4049 data = &st.stats3; 4050 } else { 4051 lv = sizeof(struct tpacket_stats); 4052 st.stats1.tp_drops = drops; 4053 st.stats1.tp_packets += drops; 4054 data = &st.stats1; 4055 } 4056 4057 break; 4058 case PACKET_AUXDATA: 4059 val = packet_sock_flag(po, PACKET_SOCK_AUXDATA); 4060 break; 4061 case PACKET_ORIGDEV: 4062 val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV); 4063 break; 4064 case PACKET_VNET_HDR: 4065 val = packet_sock_flag(po, PACKET_SOCK_HAS_VNET_HDR); 4066 break; 4067 case PACKET_VERSION: 4068 val = po->tp_version; 4069 break; 4070 case PACKET_HDRLEN: 4071 if (len > sizeof(int)) 4072 len = sizeof(int); 4073 if (len < sizeof(int)) 4074 return -EINVAL; 4075 if (copy_from_user(&val, optval, len)) 4076 return -EFAULT; 4077 switch (val) { 4078 case TPACKET_V1: 4079 val = sizeof(struct tpacket_hdr); 4080 break; 4081 case TPACKET_V2: 4082 val = sizeof(struct tpacket2_hdr); 4083 break; 4084 case TPACKET_V3: 4085 val = sizeof(struct tpacket3_hdr); 4086 break; 4087 default: 4088 return -EINVAL; 4089 } 4090 break; 4091 case PACKET_RESERVE: 4092 val = po->tp_reserve; 4093 break; 4094 case PACKET_LOSS: 4095 val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS); 4096 break; 4097 case PACKET_TIMESTAMP: 4098 val = READ_ONCE(po->tp_tstamp); 4099 break; 4100 case PACKET_FANOUT: 4101 val = (po->fanout ? 4102 ((u32)po->fanout->id | 4103 ((u32)po->fanout->type << 16) | 4104 ((u32)po->fanout->flags << 24)) : 4105 0); 4106 break; 4107 case PACKET_IGNORE_OUTGOING: 4108 val = po->prot_hook.ignore_outgoing; 4109 break; 4110 case PACKET_ROLLOVER_STATS: 4111 if (!po->rollover) 4112 return -EINVAL; 4113 rstats.tp_all = atomic_long_read(&po->rollover->num); 4114 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 4115 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 4116 data = &rstats; 4117 lv = sizeof(rstats); 4118 break; 4119 case PACKET_TX_HAS_OFF: 4120 val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF); 4121 break; 4122 case PACKET_QDISC_BYPASS: 4123 val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS); 4124 break; 4125 default: 4126 return -ENOPROTOOPT; 4127 } 4128 4129 if (len > lv) 4130 len = lv; 4131 if (put_user(len, optlen)) 4132 return -EFAULT; 4133 if (copy_to_user(optval, data, len)) 4134 return -EFAULT; 4135 return 0; 4136 } 4137 4138 static int packet_notifier(struct notifier_block *this, 4139 unsigned long msg, void *ptr) 4140 { 4141 struct sock *sk; 4142 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4143 struct net *net = dev_net(dev); 4144 4145 rcu_read_lock(); 4146 sk_for_each_rcu(sk, &net->packet.sklist) { 4147 struct packet_sock *po = pkt_sk(sk); 4148 4149 switch (msg) { 4150 case NETDEV_UNREGISTER: 4151 if (po->mclist) 4152 packet_dev_mclist_delete(dev, &po->mclist); 4153 fallthrough; 4154 4155 case NETDEV_DOWN: 4156 if (dev->ifindex == po->ifindex) { 4157 spin_lock(&po->bind_lock); 4158 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 4159 __unregister_prot_hook(sk, false); 4160 sk->sk_err = ENETDOWN; 4161 if (!sock_flag(sk, SOCK_DEAD)) 4162 sk_error_report(sk); 4163 } 4164 if (msg == NETDEV_UNREGISTER) { 4165 packet_cached_dev_reset(po); 4166 WRITE_ONCE(po->ifindex, -1); 4167 netdev_put(po->prot_hook.dev, 4168 &po->prot_hook.dev_tracker); 4169 po->prot_hook.dev = NULL; 4170 } 4171 spin_unlock(&po->bind_lock); 4172 } 4173 break; 4174 case NETDEV_UP: 4175 if (dev->ifindex == po->ifindex) { 4176 spin_lock(&po->bind_lock); 4177 if (po->num) 4178 register_prot_hook(sk); 4179 spin_unlock(&po->bind_lock); 4180 } 4181 break; 4182 } 4183 } 4184 rcu_read_unlock(); 4185 return NOTIFY_DONE; 4186 } 4187 4188 4189 static int packet_ioctl(struct socket *sock, unsigned int cmd, 4190 unsigned long arg) 4191 { 4192 struct sock *sk = sock->sk; 4193 4194 switch (cmd) { 4195 case SIOCOUTQ: 4196 { 4197 int amount = sk_wmem_alloc_get(sk); 4198 4199 return put_user(amount, (int __user *)arg); 4200 } 4201 case SIOCINQ: 4202 { 4203 struct sk_buff *skb; 4204 int amount = 0; 4205 4206 spin_lock_bh(&sk->sk_receive_queue.lock); 4207 skb = skb_peek(&sk->sk_receive_queue); 4208 if (skb) 4209 amount = skb->len; 4210 spin_unlock_bh(&sk->sk_receive_queue.lock); 4211 return put_user(amount, (int __user *)arg); 4212 } 4213 #ifdef CONFIG_INET 4214 case SIOCADDRT: 4215 case SIOCDELRT: 4216 case SIOCDARP: 4217 case SIOCGARP: 4218 case SIOCSARP: 4219 case SIOCGIFADDR: 4220 case SIOCSIFADDR: 4221 case SIOCGIFBRDADDR: 4222 case SIOCSIFBRDADDR: 4223 case SIOCGIFNETMASK: 4224 case SIOCSIFNETMASK: 4225 case SIOCGIFDSTADDR: 4226 case SIOCSIFDSTADDR: 4227 case SIOCSIFFLAGS: 4228 return inet_dgram_ops.ioctl(sock, cmd, arg); 4229 #endif 4230 4231 default: 4232 return -ENOIOCTLCMD; 4233 } 4234 return 0; 4235 } 4236 4237 static __poll_t packet_poll(struct file *file, struct socket *sock, 4238 poll_table *wait) 4239 { 4240 struct sock *sk = sock->sk; 4241 struct packet_sock *po = pkt_sk(sk); 4242 __poll_t mask = datagram_poll(file, sock, wait); 4243 4244 spin_lock_bh(&sk->sk_receive_queue.lock); 4245 if (po->rx_ring.pg_vec) { 4246 if (!packet_previous_rx_frame(po, &po->rx_ring, 4247 TP_STATUS_KERNEL)) 4248 mask |= EPOLLIN | EPOLLRDNORM; 4249 } 4250 packet_rcv_try_clear_pressure(po); 4251 spin_unlock_bh(&sk->sk_receive_queue.lock); 4252 spin_lock_bh(&sk->sk_write_queue.lock); 4253 if (po->tx_ring.pg_vec) { 4254 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4255 mask |= EPOLLOUT | EPOLLWRNORM; 4256 } 4257 spin_unlock_bh(&sk->sk_write_queue.lock); 4258 return mask; 4259 } 4260 4261 4262 /* Dirty? Well, I still did not learn better way to account 4263 * for user mmaps. 4264 */ 4265 4266 static void packet_mm_open(struct vm_area_struct *vma) 4267 { 4268 struct file *file = vma->vm_file; 4269 struct socket *sock = file->private_data; 4270 struct sock *sk = sock->sk; 4271 4272 if (sk) 4273 atomic_inc(&pkt_sk(sk)->mapped); 4274 } 4275 4276 static void packet_mm_close(struct vm_area_struct *vma) 4277 { 4278 struct file *file = vma->vm_file; 4279 struct socket *sock = file->private_data; 4280 struct sock *sk = sock->sk; 4281 4282 if (sk) 4283 atomic_dec(&pkt_sk(sk)->mapped); 4284 } 4285 4286 static const struct vm_operations_struct packet_mmap_ops = { 4287 .open = packet_mm_open, 4288 .close = packet_mm_close, 4289 }; 4290 4291 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4292 unsigned int len) 4293 { 4294 int i; 4295 4296 for (i = 0; i < len; i++) { 4297 if (likely(pg_vec[i].buffer)) { 4298 if (is_vmalloc_addr(pg_vec[i].buffer)) 4299 vfree(pg_vec[i].buffer); 4300 else 4301 free_pages((unsigned long)pg_vec[i].buffer, 4302 order); 4303 pg_vec[i].buffer = NULL; 4304 } 4305 } 4306 kfree(pg_vec); 4307 } 4308 4309 static char *alloc_one_pg_vec_page(unsigned long order) 4310 { 4311 char *buffer; 4312 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4313 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4314 4315 buffer = (char *) __get_free_pages(gfp_flags, order); 4316 if (buffer) 4317 return buffer; 4318 4319 /* __get_free_pages failed, fall back to vmalloc */ 4320 buffer = vzalloc(array_size((1 << order), PAGE_SIZE)); 4321 if (buffer) 4322 return buffer; 4323 4324 /* vmalloc failed, lets dig into swap here */ 4325 gfp_flags &= ~__GFP_NORETRY; 4326 buffer = (char *) __get_free_pages(gfp_flags, order); 4327 if (buffer) 4328 return buffer; 4329 4330 /* complete and utter failure */ 4331 return NULL; 4332 } 4333 4334 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) 4335 { 4336 unsigned int block_nr = req->tp_block_nr; 4337 struct pgv *pg_vec; 4338 int i; 4339 4340 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN); 4341 if (unlikely(!pg_vec)) 4342 goto out; 4343 4344 for (i = 0; i < block_nr; i++) { 4345 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4346 if (unlikely(!pg_vec[i].buffer)) 4347 goto out_free_pgvec; 4348 } 4349 4350 out: 4351 return pg_vec; 4352 4353 out_free_pgvec: 4354 free_pg_vec(pg_vec, order, block_nr); 4355 pg_vec = NULL; 4356 goto out; 4357 } 4358 4359 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4360 int closing, int tx_ring) 4361 { 4362 struct pgv *pg_vec = NULL; 4363 struct packet_sock *po = pkt_sk(sk); 4364 unsigned long *rx_owner_map = NULL; 4365 int was_running, order = 0; 4366 struct packet_ring_buffer *rb; 4367 struct sk_buff_head *rb_queue; 4368 __be16 num; 4369 int err; 4370 /* Added to avoid minimal code churn */ 4371 struct tpacket_req *req = &req_u->req; 4372 4373 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4374 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4375 4376 err = -EBUSY; 4377 if (!closing) { 4378 if (atomic_read(&po->mapped)) 4379 goto out; 4380 if (packet_read_pending(rb)) 4381 goto out; 4382 } 4383 4384 if (req->tp_block_nr) { 4385 unsigned int min_frame_size; 4386 4387 /* Sanity tests and some calculations */ 4388 err = -EBUSY; 4389 if (unlikely(rb->pg_vec)) 4390 goto out; 4391 4392 switch (po->tp_version) { 4393 case TPACKET_V1: 4394 po->tp_hdrlen = TPACKET_HDRLEN; 4395 break; 4396 case TPACKET_V2: 4397 po->tp_hdrlen = TPACKET2_HDRLEN; 4398 break; 4399 case TPACKET_V3: 4400 po->tp_hdrlen = TPACKET3_HDRLEN; 4401 break; 4402 } 4403 4404 err = -EINVAL; 4405 if (unlikely((int)req->tp_block_size <= 0)) 4406 goto out; 4407 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4408 goto out; 4409 min_frame_size = po->tp_hdrlen + po->tp_reserve; 4410 if (po->tp_version >= TPACKET_V3 && 4411 req->tp_block_size < 4412 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) 4413 goto out; 4414 if (unlikely(req->tp_frame_size < min_frame_size)) 4415 goto out; 4416 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4417 goto out; 4418 4419 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4420 if (unlikely(rb->frames_per_block == 0)) 4421 goto out; 4422 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) 4423 goto out; 4424 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4425 req->tp_frame_nr)) 4426 goto out; 4427 4428 err = -ENOMEM; 4429 order = get_order(req->tp_block_size); 4430 pg_vec = alloc_pg_vec(req, order); 4431 if (unlikely(!pg_vec)) 4432 goto out; 4433 switch (po->tp_version) { 4434 case TPACKET_V3: 4435 /* Block transmit is not supported yet */ 4436 if (!tx_ring) { 4437 init_prb_bdqc(po, rb, pg_vec, req_u); 4438 } else { 4439 struct tpacket_req3 *req3 = &req_u->req3; 4440 4441 if (req3->tp_retire_blk_tov || 4442 req3->tp_sizeof_priv || 4443 req3->tp_feature_req_word) { 4444 err = -EINVAL; 4445 goto out_free_pg_vec; 4446 } 4447 } 4448 break; 4449 default: 4450 if (!tx_ring) { 4451 rx_owner_map = bitmap_alloc(req->tp_frame_nr, 4452 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO); 4453 if (!rx_owner_map) 4454 goto out_free_pg_vec; 4455 } 4456 break; 4457 } 4458 } 4459 /* Done */ 4460 else { 4461 err = -EINVAL; 4462 if (unlikely(req->tp_frame_nr)) 4463 goto out; 4464 } 4465 4466 4467 /* Detach socket from network */ 4468 spin_lock(&po->bind_lock); 4469 was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING); 4470 num = po->num; 4471 if (was_running) { 4472 WRITE_ONCE(po->num, 0); 4473 __unregister_prot_hook(sk, false); 4474 } 4475 spin_unlock(&po->bind_lock); 4476 4477 synchronize_net(); 4478 4479 err = -EBUSY; 4480 mutex_lock(&po->pg_vec_lock); 4481 if (closing || atomic_read(&po->mapped) == 0) { 4482 err = 0; 4483 spin_lock_bh(&rb_queue->lock); 4484 swap(rb->pg_vec, pg_vec); 4485 if (po->tp_version <= TPACKET_V2) 4486 swap(rb->rx_owner_map, rx_owner_map); 4487 rb->frame_max = (req->tp_frame_nr - 1); 4488 rb->head = 0; 4489 rb->frame_size = req->tp_frame_size; 4490 spin_unlock_bh(&rb_queue->lock); 4491 4492 swap(rb->pg_vec_order, order); 4493 swap(rb->pg_vec_len, req->tp_block_nr); 4494 4495 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4496 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4497 tpacket_rcv : packet_rcv; 4498 skb_queue_purge(rb_queue); 4499 if (atomic_read(&po->mapped)) 4500 pr_err("packet_mmap: vma is busy: %d\n", 4501 atomic_read(&po->mapped)); 4502 } 4503 mutex_unlock(&po->pg_vec_lock); 4504 4505 spin_lock(&po->bind_lock); 4506 if (was_running) { 4507 WRITE_ONCE(po->num, num); 4508 register_prot_hook(sk); 4509 } 4510 spin_unlock(&po->bind_lock); 4511 if (pg_vec && (po->tp_version > TPACKET_V2)) { 4512 /* Because we don't support block-based V3 on tx-ring */ 4513 if (!tx_ring) 4514 prb_shutdown_retire_blk_timer(po, rb_queue); 4515 } 4516 4517 out_free_pg_vec: 4518 if (pg_vec) { 4519 bitmap_free(rx_owner_map); 4520 free_pg_vec(pg_vec, order, req->tp_block_nr); 4521 } 4522 out: 4523 return err; 4524 } 4525 4526 static int packet_mmap(struct file *file, struct socket *sock, 4527 struct vm_area_struct *vma) 4528 { 4529 struct sock *sk = sock->sk; 4530 struct packet_sock *po = pkt_sk(sk); 4531 unsigned long size, expected_size; 4532 struct packet_ring_buffer *rb; 4533 unsigned long start; 4534 int err = -EINVAL; 4535 int i; 4536 4537 if (vma->vm_pgoff) 4538 return -EINVAL; 4539 4540 mutex_lock(&po->pg_vec_lock); 4541 4542 expected_size = 0; 4543 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4544 if (rb->pg_vec) { 4545 expected_size += rb->pg_vec_len 4546 * rb->pg_vec_pages 4547 * PAGE_SIZE; 4548 } 4549 } 4550 4551 if (expected_size == 0) 4552 goto out; 4553 4554 size = vma->vm_end - vma->vm_start; 4555 if (size != expected_size) 4556 goto out; 4557 4558 start = vma->vm_start; 4559 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4560 if (rb->pg_vec == NULL) 4561 continue; 4562 4563 for (i = 0; i < rb->pg_vec_len; i++) { 4564 struct page *page; 4565 void *kaddr = rb->pg_vec[i].buffer; 4566 int pg_num; 4567 4568 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4569 page = pgv_to_page(kaddr); 4570 err = vm_insert_page(vma, start, page); 4571 if (unlikely(err)) 4572 goto out; 4573 start += PAGE_SIZE; 4574 kaddr += PAGE_SIZE; 4575 } 4576 } 4577 } 4578 4579 atomic_inc(&po->mapped); 4580 vma->vm_ops = &packet_mmap_ops; 4581 err = 0; 4582 4583 out: 4584 mutex_unlock(&po->pg_vec_lock); 4585 return err; 4586 } 4587 4588 static const struct proto_ops packet_ops_spkt = { 4589 .family = PF_PACKET, 4590 .owner = THIS_MODULE, 4591 .release = packet_release, 4592 .bind = packet_bind_spkt, 4593 .connect = sock_no_connect, 4594 .socketpair = sock_no_socketpair, 4595 .accept = sock_no_accept, 4596 .getname = packet_getname_spkt, 4597 .poll = datagram_poll, 4598 .ioctl = packet_ioctl, 4599 .gettstamp = sock_gettstamp, 4600 .listen = sock_no_listen, 4601 .shutdown = sock_no_shutdown, 4602 .sendmsg = packet_sendmsg_spkt, 4603 .recvmsg = packet_recvmsg, 4604 .mmap = sock_no_mmap, 4605 .sendpage = sock_no_sendpage, 4606 }; 4607 4608 static const struct proto_ops packet_ops = { 4609 .family = PF_PACKET, 4610 .owner = THIS_MODULE, 4611 .release = packet_release, 4612 .bind = packet_bind, 4613 .connect = sock_no_connect, 4614 .socketpair = sock_no_socketpair, 4615 .accept = sock_no_accept, 4616 .getname = packet_getname, 4617 .poll = packet_poll, 4618 .ioctl = packet_ioctl, 4619 .gettstamp = sock_gettstamp, 4620 .listen = sock_no_listen, 4621 .shutdown = sock_no_shutdown, 4622 .setsockopt = packet_setsockopt, 4623 .getsockopt = packet_getsockopt, 4624 .sendmsg = packet_sendmsg, 4625 .recvmsg = packet_recvmsg, 4626 .mmap = packet_mmap, 4627 .sendpage = sock_no_sendpage, 4628 }; 4629 4630 static const struct net_proto_family packet_family_ops = { 4631 .family = PF_PACKET, 4632 .create = packet_create, 4633 .owner = THIS_MODULE, 4634 }; 4635 4636 static struct notifier_block packet_netdev_notifier = { 4637 .notifier_call = packet_notifier, 4638 }; 4639 4640 #ifdef CONFIG_PROC_FS 4641 4642 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4643 __acquires(RCU) 4644 { 4645 struct net *net = seq_file_net(seq); 4646 4647 rcu_read_lock(); 4648 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4649 } 4650 4651 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4652 { 4653 struct net *net = seq_file_net(seq); 4654 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4655 } 4656 4657 static void packet_seq_stop(struct seq_file *seq, void *v) 4658 __releases(RCU) 4659 { 4660 rcu_read_unlock(); 4661 } 4662 4663 static int packet_seq_show(struct seq_file *seq, void *v) 4664 { 4665 if (v == SEQ_START_TOKEN) 4666 seq_printf(seq, 4667 "%*sRefCnt Type Proto Iface R Rmem User Inode\n", 4668 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk"); 4669 else { 4670 struct sock *s = sk_entry(v); 4671 const struct packet_sock *po = pkt_sk(s); 4672 4673 seq_printf(seq, 4674 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", 4675 s, 4676 refcount_read(&s->sk_refcnt), 4677 s->sk_type, 4678 ntohs(READ_ONCE(po->num)), 4679 READ_ONCE(po->ifindex), 4680 packet_sock_flag(po, PACKET_SOCK_RUNNING), 4681 atomic_read(&s->sk_rmem_alloc), 4682 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), 4683 sock_i_ino(s)); 4684 } 4685 4686 return 0; 4687 } 4688 4689 static const struct seq_operations packet_seq_ops = { 4690 .start = packet_seq_start, 4691 .next = packet_seq_next, 4692 .stop = packet_seq_stop, 4693 .show = packet_seq_show, 4694 }; 4695 #endif 4696 4697 static int __net_init packet_net_init(struct net *net) 4698 { 4699 mutex_init(&net->packet.sklist_lock); 4700 INIT_HLIST_HEAD(&net->packet.sklist); 4701 4702 #ifdef CONFIG_PROC_FS 4703 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, 4704 sizeof(struct seq_net_private))) 4705 return -ENOMEM; 4706 #endif /* CONFIG_PROC_FS */ 4707 4708 return 0; 4709 } 4710 4711 static void __net_exit packet_net_exit(struct net *net) 4712 { 4713 remove_proc_entry("packet", net->proc_net); 4714 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); 4715 } 4716 4717 static struct pernet_operations packet_net_ops = { 4718 .init = packet_net_init, 4719 .exit = packet_net_exit, 4720 }; 4721 4722 4723 static void __exit packet_exit(void) 4724 { 4725 sock_unregister(PF_PACKET); 4726 proto_unregister(&packet_proto); 4727 unregister_netdevice_notifier(&packet_netdev_notifier); 4728 unregister_pernet_subsys(&packet_net_ops); 4729 } 4730 4731 static int __init packet_init(void) 4732 { 4733 int rc; 4734 4735 rc = register_pernet_subsys(&packet_net_ops); 4736 if (rc) 4737 goto out; 4738 rc = register_netdevice_notifier(&packet_netdev_notifier); 4739 if (rc) 4740 goto out_pernet; 4741 rc = proto_register(&packet_proto, 0); 4742 if (rc) 4743 goto out_notifier; 4744 rc = sock_register(&packet_family_ops); 4745 if (rc) 4746 goto out_proto; 4747 4748 return 0; 4749 4750 out_proto: 4751 proto_unregister(&packet_proto); 4752 out_notifier: 4753 unregister_netdevice_notifier(&packet_netdev_notifier); 4754 out_pernet: 4755 unregister_pernet_subsys(&packet_net_ops); 4756 out: 4757 return rc; 4758 } 4759 4760 module_init(packet_init); 4761 module_exit(packet_exit); 4762 MODULE_LICENSE("GPL"); 4763 MODULE_ALIAS_NETPROTO(PF_PACKET); 4764