1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * PACKET - implements raw packet sockets. 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Alan Cox, <gw4pts@gw4pts.ampr.org> 12 * 13 * Fixes: 14 * Alan Cox : verify_area() now used correctly 15 * Alan Cox : new skbuff lists, look ma no backlogs! 16 * Alan Cox : tidied skbuff lists. 17 * Alan Cox : Now uses generic datagram routines I 18 * added. Also fixed the peek/read crash 19 * from all old Linux datagram code. 20 * Alan Cox : Uses the improved datagram code. 21 * Alan Cox : Added NULL's for socket options. 22 * Alan Cox : Re-commented the code. 23 * Alan Cox : Use new kernel side addressing 24 * Rob Janssen : Correct MTU usage. 25 * Dave Platt : Counter leaks caused by incorrect 26 * interrupt locking and some slightly 27 * dubious gcc output. Can you read 28 * compiler: it said _VOLATILE_ 29 * Richard Kooijman : Timestamp fixes. 30 * Alan Cox : New buffers. Use sk->mac.raw. 31 * Alan Cox : sendmsg/recvmsg support. 32 * Alan Cox : Protocol setting support 33 * Alexey Kuznetsov : Untied from IPv4 stack. 34 * Cyrus Durgin : Fixed kerneld for kmod. 35 * Michal Ostrowski : Module initialization cleanup. 36 * Ulises Alonso : Frame number limit removal and 37 * packet_set_ring memory leak. 38 * Eric Biederman : Allow for > 8 byte hardware addresses. 39 * The convention is that longer addresses 40 * will simply extend the hardware address 41 * byte arrays at the end of sockaddr_ll 42 * and packet_mreq. 43 * Johann Baudy : Added TX RING. 44 * Chetan Loke : Implemented TPACKET_V3 block abstraction 45 * layer. 46 * Copyright (C) 2011, <lokec@ccs.neu.edu> 47 */ 48 49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 50 51 #include <linux/ethtool.h> 52 #include <linux/uio.h> 53 #include <linux/filter.h> 54 #include <linux/types.h> 55 #include <linux/mm.h> 56 #include <linux/capability.h> 57 #include <linux/fcntl.h> 58 #include <linux/socket.h> 59 #include <linux/in.h> 60 #include <linux/inet.h> 61 #include <linux/netdevice.h> 62 #include <linux/if_packet.h> 63 #include <linux/wireless.h> 64 #include <linux/kernel.h> 65 #include <linux/kmod.h> 66 #include <linux/slab.h> 67 #include <linux/vmalloc.h> 68 #include <net/net_namespace.h> 69 #include <net/ip.h> 70 #include <net/protocol.h> 71 #include <linux/skbuff.h> 72 #include <net/sock.h> 73 #include <linux/errno.h> 74 #include <linux/timer.h> 75 #include <linux/uaccess.h> 76 #include <asm/ioctls.h> 77 #include <asm/page.h> 78 #include <asm/cacheflush.h> 79 #include <asm/io.h> 80 #include <linux/proc_fs.h> 81 #include <linux/seq_file.h> 82 #include <linux/poll.h> 83 #include <linux/module.h> 84 #include <linux/init.h> 85 #include <linux/mutex.h> 86 #include <linux/if_vlan.h> 87 #include <linux/virtio_net.h> 88 #include <linux/errqueue.h> 89 #include <linux/net_tstamp.h> 90 #include <linux/percpu.h> 91 #include <linux/workqueue.h> 92 #ifdef CONFIG_INET 93 #include <net/inet_common.h> 94 #endif 95 #include <linux/bpf.h> 96 #include <net/compat.h> 97 #include <linux/netfilter_netdev.h> 98 99 #include "internal.h" 100 101 /* 102 Assumptions: 103 - If the device has no dev->header_ops->create, there is no LL header 104 visible above the device. In this case, its hard_header_len should be 0. 105 The device may prepend its own header internally. In this case, its 106 needed_headroom should be set to the space needed for it to add its 107 internal header. 108 For example, a WiFi driver pretending to be an Ethernet driver should 109 set its hard_header_len to be the Ethernet header length, and set its 110 needed_headroom to be (the real WiFi header length - the fake Ethernet 111 header length). 112 - packet socket receives packets with pulled ll header, 113 so that SOCK_RAW should push it back. 114 115 On receive: 116 ----------- 117 118 Incoming, dev_has_header(dev) == true 119 mac_header -> ll header 120 data -> data 121 122 Outgoing, dev_has_header(dev) == true 123 mac_header -> ll header 124 data -> ll header 125 126 Incoming, dev_has_header(dev) == false 127 mac_header -> data 128 However drivers often make it point to the ll header. 129 This is incorrect because the ll header should be invisible to us. 130 data -> data 131 132 Outgoing, dev_has_header(dev) == false 133 mac_header -> data. ll header is invisible to us. 134 data -> data 135 136 Resume 137 If dev_has_header(dev) == false we are unable to restore the ll header, 138 because it is invisible to us. 139 140 141 On transmit: 142 ------------ 143 144 dev_has_header(dev) == true 145 mac_header -> ll header 146 data -> ll header 147 148 dev_has_header(dev) == false (ll header is invisible to us) 149 mac_header -> data 150 data -> data 151 152 We should set network_header on output to the correct position, 153 packet classifier depends on it. 154 */ 155 156 /* Private packet socket structures. */ 157 158 /* identical to struct packet_mreq except it has 159 * a longer address field. 160 */ 161 struct packet_mreq_max { 162 int mr_ifindex; 163 unsigned short mr_type; 164 unsigned short mr_alen; 165 unsigned char mr_address[MAX_ADDR_LEN]; 166 }; 167 168 union tpacket_uhdr { 169 struct tpacket_hdr *h1; 170 struct tpacket2_hdr *h2; 171 struct tpacket3_hdr *h3; 172 void *raw; 173 }; 174 175 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 176 int closing, int tx_ring); 177 178 #define V3_ALIGNMENT (8) 179 180 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) 181 182 #define BLK_PLUS_PRIV(sz_of_priv) \ 183 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) 184 185 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) 186 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) 187 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) 188 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) 189 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) 190 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) 191 192 struct packet_sock; 193 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 194 struct packet_type *pt, struct net_device *orig_dev); 195 196 static void *packet_previous_frame(struct packet_sock *po, 197 struct packet_ring_buffer *rb, 198 int status); 199 static void packet_increment_head(struct packet_ring_buffer *buff); 200 static int prb_curr_blk_in_use(struct tpacket_block_desc *); 201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, 202 struct packet_sock *); 203 static void prb_retire_current_block(struct tpacket_kbdq_core *, 204 struct packet_sock *, unsigned int status); 205 static int prb_queue_frozen(struct tpacket_kbdq_core *); 206 static void prb_open_block(struct tpacket_kbdq_core *, 207 struct tpacket_block_desc *); 208 static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *); 209 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); 210 static void prb_clear_rxhash(struct tpacket_kbdq_core *, 211 struct tpacket3_hdr *); 212 static void prb_fill_vlan_info(struct tpacket_kbdq_core *, 213 struct tpacket3_hdr *); 214 static void packet_flush_mclist(struct sock *sk); 215 static u16 packet_pick_tx_queue(struct sk_buff *skb); 216 217 struct packet_skb_cb { 218 union { 219 struct sockaddr_pkt pkt; 220 union { 221 /* Trick: alias skb original length with 222 * ll.sll_family and ll.protocol in order 223 * to save room. 224 */ 225 unsigned int origlen; 226 struct sockaddr_ll ll; 227 }; 228 } sa; 229 }; 230 231 #define vio_le() virtio_legacy_is_little_endian() 232 233 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) 234 235 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) 236 #define GET_PBLOCK_DESC(x, bid) \ 237 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) 238 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ 239 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) 240 #define GET_NEXT_PRB_BLK_NUM(x) \ 241 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ 242 ((x)->kactive_blk_num+1) : 0) 243 244 static void __fanout_unlink(struct sock *sk, struct packet_sock *po); 245 static void __fanout_link(struct sock *sk, struct packet_sock *po); 246 247 #ifdef CONFIG_NETFILTER_EGRESS 248 static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb) 249 { 250 struct sk_buff *next, *head = NULL, *tail; 251 int rc; 252 253 rcu_read_lock(); 254 for (; skb != NULL; skb = next) { 255 next = skb->next; 256 skb_mark_not_on_list(skb); 257 258 if (!nf_hook_egress(skb, &rc, skb->dev)) 259 continue; 260 261 if (!head) 262 head = skb; 263 else 264 tail->next = skb; 265 266 tail = skb; 267 } 268 rcu_read_unlock(); 269 270 return head; 271 } 272 #endif 273 274 static int packet_xmit(const struct packet_sock *po, struct sk_buff *skb) 275 { 276 if (!packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS)) 277 return dev_queue_xmit(skb); 278 279 #ifdef CONFIG_NETFILTER_EGRESS 280 if (nf_hook_egress_active()) { 281 skb = nf_hook_direct_egress(skb); 282 if (!skb) 283 return NET_XMIT_DROP; 284 } 285 #endif 286 return dev_direct_xmit(skb, packet_pick_tx_queue(skb)); 287 } 288 289 static struct net_device *packet_cached_dev_get(struct packet_sock *po) 290 { 291 struct net_device *dev; 292 293 rcu_read_lock(); 294 dev = rcu_dereference(po->cached_dev); 295 dev_hold(dev); 296 rcu_read_unlock(); 297 298 return dev; 299 } 300 301 static void packet_cached_dev_assign(struct packet_sock *po, 302 struct net_device *dev) 303 { 304 rcu_assign_pointer(po->cached_dev, dev); 305 } 306 307 static void packet_cached_dev_reset(struct packet_sock *po) 308 { 309 RCU_INIT_POINTER(po->cached_dev, NULL); 310 } 311 312 static u16 packet_pick_tx_queue(struct sk_buff *skb) 313 { 314 struct net_device *dev = skb->dev; 315 const struct net_device_ops *ops = dev->netdev_ops; 316 int cpu = raw_smp_processor_id(); 317 u16 queue_index; 318 319 #ifdef CONFIG_XPS 320 skb->sender_cpu = cpu + 1; 321 #endif 322 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues); 323 if (ops->ndo_select_queue) { 324 queue_index = ops->ndo_select_queue(dev, skb, NULL); 325 queue_index = netdev_cap_txqueue(dev, queue_index); 326 } else { 327 queue_index = netdev_pick_tx(dev, skb, NULL); 328 } 329 330 return queue_index; 331 } 332 333 /* __register_prot_hook must be invoked through register_prot_hook 334 * or from a context in which asynchronous accesses to the packet 335 * socket is not possible (packet_create()). 336 */ 337 static void __register_prot_hook(struct sock *sk) 338 { 339 struct packet_sock *po = pkt_sk(sk); 340 341 if (!packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 342 if (po->fanout) 343 __fanout_link(sk, po); 344 else 345 dev_add_pack(&po->prot_hook); 346 347 sock_hold(sk); 348 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 1); 349 } 350 } 351 352 static void register_prot_hook(struct sock *sk) 353 { 354 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); 355 __register_prot_hook(sk); 356 } 357 358 /* If the sync parameter is true, we will temporarily drop 359 * the po->bind_lock and do a synchronize_net to make sure no 360 * asynchronous packet processing paths still refer to the elements 361 * of po->prot_hook. If the sync parameter is false, it is the 362 * callers responsibility to take care of this. 363 */ 364 static void __unregister_prot_hook(struct sock *sk, bool sync) 365 { 366 struct packet_sock *po = pkt_sk(sk); 367 368 lockdep_assert_held_once(&po->bind_lock); 369 370 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 0); 371 372 if (po->fanout) 373 __fanout_unlink(sk, po); 374 else 375 __dev_remove_pack(&po->prot_hook); 376 377 __sock_put(sk); 378 379 if (sync) { 380 spin_unlock(&po->bind_lock); 381 synchronize_net(); 382 spin_lock(&po->bind_lock); 383 } 384 } 385 386 static void unregister_prot_hook(struct sock *sk, bool sync) 387 { 388 struct packet_sock *po = pkt_sk(sk); 389 390 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) 391 __unregister_prot_hook(sk, sync); 392 } 393 394 static inline struct page * __pure pgv_to_page(void *addr) 395 { 396 if (is_vmalloc_addr(addr)) 397 return vmalloc_to_page(addr); 398 return virt_to_page(addr); 399 } 400 401 /* Timestamp bits that may be set in addition to the frame state */ 402 static const u32 tp_status_ts_mask = TP_STATUS_TS_SOFTWARE | 403 TP_STATUS_TS_RAW_HARDWARE; 404 405 static void __packet_set_status(struct packet_sock *po, void *frame, int status) 406 { 407 union tpacket_uhdr h; 408 409 /* WRITE_ONCE() are paired with READ_ONCE() in __packet_get_status */ 410 411 h.raw = frame; 412 switch (po->tp_version) { 413 case TPACKET_V1: 414 WRITE_ONCE(h.h1->tp_status, status); 415 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 416 break; 417 case TPACKET_V2: 418 WRITE_ONCE(h.h2->tp_status, status); 419 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 420 break; 421 case TPACKET_V3: 422 WRITE_ONCE(h.h3->tp_status, status); 423 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 424 break; 425 default: 426 WARN(1, "TPACKET version not supported.\n"); 427 BUG(); 428 } 429 430 smp_wmb(); 431 } 432 433 static int __packet_get_status(const struct packet_sock *po, void *frame) 434 { 435 union tpacket_uhdr h; 436 437 smp_rmb(); 438 439 /* READ_ONCE() are paired with WRITE_ONCE() in __packet_set_status */ 440 441 h.raw = frame; 442 switch (po->tp_version) { 443 case TPACKET_V1: 444 flush_dcache_page(pgv_to_page(&h.h1->tp_status)); 445 return READ_ONCE(h.h1->tp_status); 446 case TPACKET_V2: 447 flush_dcache_page(pgv_to_page(&h.h2->tp_status)); 448 return READ_ONCE(h.h2->tp_status); 449 case TPACKET_V3: 450 flush_dcache_page(pgv_to_page(&h.h3->tp_status)); 451 return READ_ONCE(h.h3->tp_status); 452 default: 453 WARN(1, "TPACKET version not supported.\n"); 454 BUG(); 455 return 0; 456 } 457 } 458 459 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts, 460 unsigned int flags) 461 { 462 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); 463 464 if (shhwtstamps && 465 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && 466 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts)) 467 return TP_STATUS_TS_RAW_HARDWARE; 468 469 if ((flags & SOF_TIMESTAMPING_SOFTWARE) && 470 ktime_to_timespec64_cond(skb_tstamp(skb), ts)) 471 return TP_STATUS_TS_SOFTWARE; 472 473 return 0; 474 } 475 476 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, 477 struct sk_buff *skb) 478 { 479 union tpacket_uhdr h; 480 struct timespec64 ts; 481 __u32 ts_status; 482 483 if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp)))) 484 return 0; 485 486 h.raw = frame; 487 /* 488 * versions 1 through 3 overflow the timestamps in y2106, since they 489 * all store the seconds in a 32-bit unsigned integer. 490 * If we create a version 4, that should have a 64-bit timestamp, 491 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit 492 * nanoseconds. 493 */ 494 switch (po->tp_version) { 495 case TPACKET_V1: 496 h.h1->tp_sec = ts.tv_sec; 497 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 498 break; 499 case TPACKET_V2: 500 h.h2->tp_sec = ts.tv_sec; 501 h.h2->tp_nsec = ts.tv_nsec; 502 break; 503 case TPACKET_V3: 504 h.h3->tp_sec = ts.tv_sec; 505 h.h3->tp_nsec = ts.tv_nsec; 506 break; 507 default: 508 WARN(1, "TPACKET version not supported.\n"); 509 BUG(); 510 } 511 512 /* one flush is safe, as both fields always lie on the same cacheline */ 513 flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); 514 smp_wmb(); 515 516 return ts_status; 517 } 518 519 static void *packet_lookup_frame(const struct packet_sock *po, 520 const struct packet_ring_buffer *rb, 521 unsigned int position, 522 int status) 523 { 524 unsigned int pg_vec_pos, frame_offset; 525 union tpacket_uhdr h; 526 u32 mask = 0; 527 528 pg_vec_pos = position / rb->frames_per_block; 529 frame_offset = position % rb->frames_per_block; 530 531 h.raw = rb->pg_vec[pg_vec_pos].buffer + 532 (frame_offset * rb->frame_size); 533 534 if (rb == &po->tx_ring && status == TP_STATUS_AVAILABLE) 535 mask = tp_status_ts_mask; 536 537 if (status != (__packet_get_status(po, h.raw) & ~mask)) 538 return NULL; 539 540 return h.raw; 541 } 542 543 static void *packet_current_frame(struct packet_sock *po, 544 struct packet_ring_buffer *rb, 545 int status) 546 { 547 return packet_lookup_frame(po, rb, rb->head, status); 548 } 549 550 static u16 vlan_get_tci(const struct sk_buff *skb, struct net_device *dev) 551 { 552 struct vlan_hdr vhdr, *vh; 553 unsigned int header_len; 554 555 if (!dev) 556 return 0; 557 558 /* In the SOCK_DGRAM scenario, skb data starts at the network 559 * protocol, which is after the VLAN headers. The outer VLAN 560 * header is at the hard_header_len offset in non-variable 561 * length link layer headers. If it's a VLAN device, the 562 * min_header_len should be used to exclude the VLAN header 563 * size. 564 */ 565 if (dev->min_header_len == dev->hard_header_len) 566 header_len = dev->hard_header_len; 567 else if (is_vlan_dev(dev)) 568 header_len = dev->min_header_len; 569 else 570 return 0; 571 572 vh = skb_header_pointer(skb, skb_mac_offset(skb) + header_len, 573 sizeof(vhdr), &vhdr); 574 if (unlikely(!vh)) 575 return 0; 576 577 return ntohs(vh->h_vlan_TCI); 578 } 579 580 static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb) 581 { 582 __be16 proto = skb->protocol; 583 584 if (unlikely(eth_type_vlan(proto))) 585 proto = vlan_get_protocol_offset_inline(skb, proto, 586 skb_mac_offset(skb), 587 NULL); 588 589 return proto; 590 } 591 592 static void prb_shutdown_retire_blk_timer(struct packet_sock *po, 593 struct sk_buff_head *rb_queue) 594 { 595 struct tpacket_kbdq_core *pkc; 596 597 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 598 hrtimer_cancel(&pkc->retire_blk_timer); 599 } 600 601 static int prb_calc_retire_blk_tmo(struct packet_sock *po, 602 int blk_size_in_bytes) 603 { 604 struct net_device *dev; 605 unsigned int mbits, div; 606 struct ethtool_link_ksettings ecmd; 607 int err; 608 609 rtnl_lock(); 610 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); 611 if (unlikely(!dev)) { 612 rtnl_unlock(); 613 return DEFAULT_PRB_RETIRE_TOV; 614 } 615 err = __ethtool_get_link_ksettings(dev, &ecmd); 616 rtnl_unlock(); 617 if (err) 618 return DEFAULT_PRB_RETIRE_TOV; 619 620 /* If the link speed is so slow you don't really 621 * need to worry about perf anyways 622 */ 623 if (ecmd.base.speed < SPEED_1000 || 624 ecmd.base.speed == SPEED_UNKNOWN) 625 return DEFAULT_PRB_RETIRE_TOV; 626 627 div = ecmd.base.speed / 1000; 628 mbits = (u64)blk_size_in_bytes * 8 / (1024 * 1024); 629 630 if (div) 631 mbits /= div; 632 633 if (div) 634 return mbits + 1; 635 return mbits; 636 } 637 638 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, 639 union tpacket_req_u *req_u) 640 { 641 p1->feature_req_word = req_u->req3.tp_feature_req_word; 642 } 643 644 static void init_prb_bdqc(struct packet_sock *po, 645 struct packet_ring_buffer *rb, 646 struct pgv *pg_vec, 647 union tpacket_req_u *req_u) 648 { 649 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); 650 struct tpacket_block_desc *pbd; 651 652 memset(p1, 0x0, sizeof(*p1)); 653 654 p1->knxt_seq_num = 1; 655 p1->pkbdq = pg_vec; 656 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; 657 p1->pkblk_start = pg_vec[0].buffer; 658 p1->kblk_size = req_u->req3.tp_block_size; 659 p1->knum_blocks = req_u->req3.tp_block_nr; 660 p1->hdrlen = po->tp_hdrlen; 661 p1->version = po->tp_version; 662 po->stats.stats3.tp_freeze_q_cnt = 0; 663 if (req_u->req3.tp_retire_blk_tov) 664 p1->interval_ktime = ms_to_ktime(req_u->req3.tp_retire_blk_tov); 665 else 666 p1->interval_ktime = ms_to_ktime(prb_calc_retire_blk_tmo(po, 667 req_u->req3.tp_block_size)); 668 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; 669 rwlock_init(&p1->blk_fill_in_prog_lock); 670 671 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); 672 prb_init_ft_ops(p1, req_u); 673 hrtimer_setup(&p1->retire_blk_timer, prb_retire_rx_blk_timer_expired, 674 CLOCK_MONOTONIC, HRTIMER_MODE_REL_SOFT); 675 hrtimer_start(&p1->retire_blk_timer, p1->interval_ktime, 676 HRTIMER_MODE_REL_SOFT); 677 prb_open_block(p1, pbd); 678 } 679 680 /* 681 * With a 1MB block-size, on a 1Gbps line, it will take 682 * i) ~8 ms to fill a block + ii) memcpy etc. 683 * In this cut we are not accounting for the memcpy time. 684 * 685 * Since the tmo granularity is in msecs, it is not too expensive 686 * to refresh the timer, lets say every '8' msecs. 687 * Either the user can set the 'tmo' or we can derive it based on 688 * a) line-speed and b) block-size. 689 * prb_calc_retire_blk_tmo() calculates the tmo. 690 */ 691 static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *t) 692 { 693 struct packet_sock *po = 694 timer_container_of(po, t, rx_ring.prb_bdqc.retire_blk_timer); 695 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 696 unsigned int frozen; 697 struct tpacket_block_desc *pbd; 698 699 spin_lock(&po->sk.sk_receive_queue.lock); 700 701 frozen = prb_queue_frozen(pkc); 702 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 703 704 /* We only need to plug the race when the block is partially filled. 705 * tpacket_rcv: 706 * lock(); increment BLOCK_NUM_PKTS; unlock() 707 * copy_bits() is in progress ... 708 * timer fires on other cpu: 709 * we can't retire the current block because copy_bits 710 * is in progress. 711 * 712 */ 713 if (BLOCK_NUM_PKTS(pbd)) { 714 /* Waiting for skb_copy_bits to finish... */ 715 write_lock(&pkc->blk_fill_in_prog_lock); 716 write_unlock(&pkc->blk_fill_in_prog_lock); 717 } 718 719 if (!frozen) { 720 if (BLOCK_NUM_PKTS(pbd)) { 721 /* Not an empty block. Need retire the block. */ 722 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); 723 prb_dispatch_next_block(pkc, po); 724 } 725 } else { 726 /* Case 1. Queue was frozen because user-space was 727 * lagging behind. 728 */ 729 if (!prb_curr_blk_in_use(pbd)) { 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 } 739 } 740 741 hrtimer_forward_now(&pkc->retire_blk_timer, pkc->interval_ktime); 742 spin_unlock(&po->sk.sk_receive_queue.lock); 743 return HRTIMER_RESTART; 744 } 745 746 static void prb_flush_block(struct tpacket_kbdq_core *pkc1, 747 struct tpacket_block_desc *pbd1, __u32 status) 748 { 749 /* Flush everything minus the block header */ 750 751 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 752 u8 *start, *end; 753 754 start = (u8 *)pbd1; 755 756 /* Skip the block header(we know header WILL fit in 4K) */ 757 start += PAGE_SIZE; 758 759 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); 760 for (; start < end; start += PAGE_SIZE) 761 flush_dcache_page(pgv_to_page(start)); 762 763 smp_wmb(); 764 #endif 765 766 /* Now update the block status. */ 767 768 BLOCK_STATUS(pbd1) = status; 769 770 /* Flush the block header */ 771 772 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 773 start = (u8 *)pbd1; 774 flush_dcache_page(pgv_to_page(start)); 775 776 smp_wmb(); 777 #endif 778 } 779 780 /* 781 * Side effect: 782 * 783 * 1) flush the block 784 * 2) Increment active_blk_num 785 * 786 * Note:We DONT refresh the timer on purpose. 787 * Because almost always the next block will be opened. 788 */ 789 static void prb_close_block(struct tpacket_kbdq_core *pkc1, 790 struct tpacket_block_desc *pbd1, 791 struct packet_sock *po, unsigned int stat) 792 { 793 __u32 status = TP_STATUS_USER | stat; 794 795 struct tpacket3_hdr *last_pkt; 796 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 797 struct sock *sk = &po->sk; 798 799 if (atomic_read(&po->tp_drops)) 800 status |= TP_STATUS_LOSING; 801 802 last_pkt = (struct tpacket3_hdr *)pkc1->prev; 803 last_pkt->tp_next_offset = 0; 804 805 /* Get the ts of the last pkt */ 806 if (BLOCK_NUM_PKTS(pbd1)) { 807 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; 808 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; 809 } else { 810 /* Ok, we tmo'd - so get the current time. 811 * 812 * It shouldn't really happen as we don't close empty 813 * blocks. See prb_retire_rx_blk_timer_expired(). 814 */ 815 struct timespec64 ts; 816 ktime_get_real_ts64(&ts); 817 h1->ts_last_pkt.ts_sec = ts.tv_sec; 818 h1->ts_last_pkt.ts_nsec = ts.tv_nsec; 819 } 820 821 smp_wmb(); 822 823 /* Flush the block */ 824 prb_flush_block(pkc1, pbd1, status); 825 826 sk->sk_data_ready(sk); 827 828 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); 829 } 830 831 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) 832 { 833 pkc->reset_pending_on_curr_blk = 0; 834 } 835 836 /* 837 * prb_open_block is called by tpacket_rcv or timer callback. 838 * 839 * Reasons why NOT update hrtimer in prb_open_block: 840 * 1) It will increase complexity to distinguish the two caller scenario. 841 * 2) hrtimer_cancel and hrtimer_start need to be called if you want to update 842 * TMO of an already enqueued hrtimer, leading to complex shutdown logic. 843 * 844 * One side effect of NOT update hrtimer when called by tpacket_rcv is that 845 * a newly opened block triggered by tpacket_rcv may be retired earlier than 846 * expected. On the other hand, if timeout is updated in prb_open_block, the 847 * frequent reception of network packets that leads to prb_open_block being 848 * called may cause hrtimer to be removed and enqueued repeatedly. 849 */ 850 static void prb_open_block(struct tpacket_kbdq_core *pkc1, 851 struct tpacket_block_desc *pbd1) 852 { 853 struct timespec64 ts; 854 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; 855 856 smp_rmb(); 857 858 /* We could have just memset this but we will lose the 859 * flexibility of making the priv area sticky 860 */ 861 862 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; 863 BLOCK_NUM_PKTS(pbd1) = 0; 864 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 865 866 ktime_get_real_ts64(&ts); 867 868 h1->ts_first_pkt.ts_sec = ts.tv_sec; 869 h1->ts_first_pkt.ts_nsec = ts.tv_nsec; 870 871 pkc1->pkblk_start = (char *)pbd1; 872 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 873 874 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); 875 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; 876 877 pbd1->version = pkc1->version; 878 pkc1->prev = pkc1->nxt_offset; 879 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; 880 881 prb_thaw_queue(pkc1); 882 883 smp_wmb(); 884 } 885 886 /* 887 * Queue freeze logic: 888 * 1) Assume tp_block_nr = 8 blocks. 889 * 2) At time 't0', user opens Rx ring. 890 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 891 * 4) user-space is either sleeping or processing block '0'. 892 * 5) tpacket_rcv is currently filling block '7', since there is no space left, 893 * it will close block-7,loop around and try to fill block '0'. 894 * call-flow: 895 * __packet_lookup_frame_in_block 896 * prb_retire_current_block() 897 * prb_dispatch_next_block() 898 * |->(BLOCK_STATUS == USER) evaluates to true 899 * 5.1) Since block-0 is currently in-use, we just freeze the queue. 900 * 6) Now there are two cases: 901 * 6.1) Link goes idle right after the queue is frozen. 902 * But remember, the last open_block() refreshed the timer. 903 * When this timer expires,it will refresh itself so that we can 904 * re-open block-0 in near future. 905 * 6.2) Link is busy and keeps on receiving packets. This is a simple 906 * case and __packet_lookup_frame_in_block will check if block-0 907 * is free and can now be re-used. 908 */ 909 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, 910 struct packet_sock *po) 911 { 912 pkc->reset_pending_on_curr_blk = 1; 913 po->stats.stats3.tp_freeze_q_cnt++; 914 } 915 916 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) 917 918 /* 919 * If the next block is free then we will dispatch it 920 * and return a good offset. 921 * Else, we will freeze the queue. 922 * So, caller must check the return value. 923 */ 924 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, 925 struct packet_sock *po) 926 { 927 struct tpacket_block_desc *pbd; 928 929 smp_rmb(); 930 931 /* 1. Get current block num */ 932 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 933 934 /* 2. If this block is currently in_use then freeze the queue */ 935 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { 936 prb_freeze_queue(pkc, po); 937 return NULL; 938 } 939 940 /* 941 * 3. 942 * open this block and return the offset where the first packet 943 * needs to get stored. 944 */ 945 prb_open_block(pkc, pbd); 946 return (void *)pkc->nxt_offset; 947 } 948 949 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, 950 struct packet_sock *po, unsigned int status) 951 { 952 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 953 954 /* retire/close the current block */ 955 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { 956 /* 957 * Plug the case where copy_bits() is in progress on 958 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't 959 * have space to copy the pkt in the current block and 960 * called prb_retire_current_block() 961 * 962 * We don't need to worry about the TMO case because 963 * the timer-handler already handled this case. 964 */ 965 if (!(status & TP_STATUS_BLK_TMO)) { 966 /* Waiting for skb_copy_bits to finish... */ 967 write_lock(&pkc->blk_fill_in_prog_lock); 968 write_unlock(&pkc->blk_fill_in_prog_lock); 969 } 970 prb_close_block(pkc, pbd, po, status); 971 return; 972 } 973 } 974 975 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) 976 { 977 return TP_STATUS_USER & BLOCK_STATUS(pbd); 978 } 979 980 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) 981 { 982 return pkc->reset_pending_on_curr_blk; 983 } 984 985 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) 986 __releases(&pkc->blk_fill_in_prog_lock) 987 { 988 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 989 990 read_unlock(&pkc->blk_fill_in_prog_lock); 991 } 992 993 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, 994 struct tpacket3_hdr *ppd) 995 { 996 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); 997 } 998 999 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, 1000 struct tpacket3_hdr *ppd) 1001 { 1002 ppd->hv1.tp_rxhash = 0; 1003 } 1004 1005 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, 1006 struct tpacket3_hdr *ppd) 1007 { 1008 struct packet_sock *po = container_of(pkc, struct packet_sock, rx_ring.prb_bdqc); 1009 1010 if (skb_vlan_tag_present(pkc->skb)) { 1011 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); 1012 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); 1013 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 1014 } else if (unlikely(po->sk.sk_type == SOCK_DGRAM && eth_type_vlan(pkc->skb->protocol))) { 1015 ppd->hv1.tp_vlan_tci = vlan_get_tci(pkc->skb, pkc->skb->dev); 1016 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->protocol); 1017 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 1018 } else { 1019 ppd->hv1.tp_vlan_tci = 0; 1020 ppd->hv1.tp_vlan_tpid = 0; 1021 ppd->tp_status = TP_STATUS_AVAILABLE; 1022 } 1023 } 1024 1025 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, 1026 struct tpacket3_hdr *ppd) 1027 { 1028 ppd->hv1.tp_padding = 0; 1029 prb_fill_vlan_info(pkc, ppd); 1030 1031 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) 1032 prb_fill_rxhash(pkc, ppd); 1033 else 1034 prb_clear_rxhash(pkc, ppd); 1035 } 1036 1037 static void prb_fill_curr_block(char *curr, 1038 struct tpacket_kbdq_core *pkc, 1039 struct tpacket_block_desc *pbd, 1040 unsigned int len) 1041 __acquires(&pkc->blk_fill_in_prog_lock) 1042 { 1043 struct tpacket3_hdr *ppd; 1044 1045 ppd = (struct tpacket3_hdr *)curr; 1046 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); 1047 pkc->prev = curr; 1048 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); 1049 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); 1050 BLOCK_NUM_PKTS(pbd) += 1; 1051 read_lock(&pkc->blk_fill_in_prog_lock); 1052 prb_run_all_ft_ops(pkc, ppd); 1053 } 1054 1055 /* Assumes caller has the sk->rx_queue.lock */ 1056 static void *__packet_lookup_frame_in_block(struct packet_sock *po, 1057 struct sk_buff *skb, 1058 unsigned int len 1059 ) 1060 { 1061 struct tpacket_kbdq_core *pkc; 1062 struct tpacket_block_desc *pbd; 1063 char *curr, *end; 1064 1065 pkc = GET_PBDQC_FROM_RB(&po->rx_ring); 1066 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1067 1068 /* Queue is frozen when user space is lagging behind */ 1069 if (prb_queue_frozen(pkc)) { 1070 /* 1071 * Check if that last block which caused the queue to freeze, 1072 * is still in_use by user-space. 1073 */ 1074 if (prb_curr_blk_in_use(pbd)) { 1075 /* Can't record this packet */ 1076 return NULL; 1077 } else { 1078 /* 1079 * Ok, the block was released by user-space. 1080 * Now let's open that block. 1081 * opening a block also thaws the queue. 1082 * Thawing is a side effect. 1083 */ 1084 prb_open_block(pkc, pbd); 1085 } 1086 } 1087 1088 smp_mb(); 1089 curr = pkc->nxt_offset; 1090 pkc->skb = skb; 1091 end = (char *)pbd + pkc->kblk_size; 1092 1093 /* first try the current block */ 1094 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { 1095 prb_fill_curr_block(curr, pkc, pbd, len); 1096 return (void *)curr; 1097 } 1098 1099 /* Ok, close the current block */ 1100 prb_retire_current_block(pkc, po, 0); 1101 1102 /* Now, try to dispatch the next block */ 1103 curr = (char *)prb_dispatch_next_block(pkc, po); 1104 if (curr) { 1105 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); 1106 prb_fill_curr_block(curr, pkc, pbd, len); 1107 return (void *)curr; 1108 } 1109 1110 /* 1111 * No free blocks are available.user_space hasn't caught up yet. 1112 * Queue was just frozen and now this packet will get dropped. 1113 */ 1114 return NULL; 1115 } 1116 1117 static void *packet_current_rx_frame(struct packet_sock *po, 1118 struct sk_buff *skb, 1119 int status, unsigned int len) 1120 { 1121 char *curr = NULL; 1122 switch (po->tp_version) { 1123 case TPACKET_V1: 1124 case TPACKET_V2: 1125 curr = packet_lookup_frame(po, &po->rx_ring, 1126 po->rx_ring.head, status); 1127 return curr; 1128 case TPACKET_V3: 1129 return __packet_lookup_frame_in_block(po, skb, len); 1130 default: 1131 WARN(1, "TPACKET version not supported\n"); 1132 BUG(); 1133 return NULL; 1134 } 1135 } 1136 1137 static void *prb_lookup_block(const struct packet_sock *po, 1138 const struct packet_ring_buffer *rb, 1139 unsigned int idx, 1140 int status) 1141 { 1142 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); 1143 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); 1144 1145 if (status != BLOCK_STATUS(pbd)) 1146 return NULL; 1147 return pbd; 1148 } 1149 1150 static int prb_previous_blk_num(struct packet_ring_buffer *rb) 1151 { 1152 unsigned int prev; 1153 if (rb->prb_bdqc.kactive_blk_num) 1154 prev = rb->prb_bdqc.kactive_blk_num-1; 1155 else 1156 prev = rb->prb_bdqc.knum_blocks-1; 1157 return prev; 1158 } 1159 1160 /* Assumes caller has held the rx_queue.lock */ 1161 static void *__prb_previous_block(struct packet_sock *po, 1162 struct packet_ring_buffer *rb, 1163 int status) 1164 { 1165 unsigned int previous = prb_previous_blk_num(rb); 1166 return prb_lookup_block(po, rb, previous, status); 1167 } 1168 1169 static void *packet_previous_rx_frame(struct packet_sock *po, 1170 struct packet_ring_buffer *rb, 1171 int status) 1172 { 1173 if (po->tp_version <= TPACKET_V2) 1174 return packet_previous_frame(po, rb, status); 1175 1176 return __prb_previous_block(po, rb, status); 1177 } 1178 1179 static void packet_increment_rx_head(struct packet_sock *po, 1180 struct packet_ring_buffer *rb) 1181 { 1182 switch (po->tp_version) { 1183 case TPACKET_V1: 1184 case TPACKET_V2: 1185 return packet_increment_head(rb); 1186 case TPACKET_V3: 1187 default: 1188 WARN(1, "TPACKET version not supported.\n"); 1189 BUG(); 1190 return; 1191 } 1192 } 1193 1194 static void *packet_previous_frame(struct packet_sock *po, 1195 struct packet_ring_buffer *rb, 1196 int status) 1197 { 1198 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; 1199 return packet_lookup_frame(po, rb, previous, status); 1200 } 1201 1202 static void packet_increment_head(struct packet_ring_buffer *buff) 1203 { 1204 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; 1205 } 1206 1207 static void packet_inc_pending(struct packet_ring_buffer *rb) 1208 { 1209 this_cpu_inc(*rb->pending_refcnt); 1210 } 1211 1212 static void packet_dec_pending(struct packet_ring_buffer *rb) 1213 { 1214 this_cpu_dec(*rb->pending_refcnt); 1215 } 1216 1217 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) 1218 { 1219 unsigned int refcnt = 0; 1220 int cpu; 1221 1222 /* We don't use pending refcount in rx_ring. */ 1223 if (rb->pending_refcnt == NULL) 1224 return 0; 1225 1226 for_each_possible_cpu(cpu) 1227 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); 1228 1229 return refcnt; 1230 } 1231 1232 static int packet_alloc_pending(struct packet_sock *po) 1233 { 1234 po->rx_ring.pending_refcnt = NULL; 1235 1236 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); 1237 if (unlikely(po->tx_ring.pending_refcnt == NULL)) 1238 return -ENOBUFS; 1239 1240 return 0; 1241 } 1242 1243 static void packet_free_pending(struct packet_sock *po) 1244 { 1245 free_percpu(po->tx_ring.pending_refcnt); 1246 } 1247 1248 #define ROOM_POW_OFF 2 1249 #define ROOM_NONE 0x0 1250 #define ROOM_LOW 0x1 1251 #define ROOM_NORMAL 0x2 1252 1253 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off) 1254 { 1255 int idx, len; 1256 1257 len = READ_ONCE(po->rx_ring.frame_max) + 1; 1258 idx = READ_ONCE(po->rx_ring.head); 1259 if (pow_off) 1260 idx += len >> pow_off; 1261 if (idx >= len) 1262 idx -= len; 1263 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1264 } 1265 1266 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off) 1267 { 1268 int idx, len; 1269 1270 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks); 1271 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num); 1272 if (pow_off) 1273 idx += len >> pow_off; 1274 if (idx >= len) 1275 idx -= len; 1276 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); 1277 } 1278 1279 static int __packet_rcv_has_room(const struct packet_sock *po, 1280 const struct sk_buff *skb) 1281 { 1282 const struct sock *sk = &po->sk; 1283 int ret = ROOM_NONE; 1284 1285 if (po->prot_hook.func != tpacket_rcv) { 1286 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1287 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc) 1288 - (skb ? skb->truesize : 0); 1289 1290 if (avail > (rcvbuf >> ROOM_POW_OFF)) 1291 return ROOM_NORMAL; 1292 else if (avail > 0) 1293 return ROOM_LOW; 1294 else 1295 return ROOM_NONE; 1296 } 1297 1298 if (po->tp_version == TPACKET_V3) { 1299 if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) 1300 ret = ROOM_NORMAL; 1301 else if (__tpacket_v3_has_room(po, 0)) 1302 ret = ROOM_LOW; 1303 } else { 1304 if (__tpacket_has_room(po, ROOM_POW_OFF)) 1305 ret = ROOM_NORMAL; 1306 else if (__tpacket_has_room(po, 0)) 1307 ret = ROOM_LOW; 1308 } 1309 1310 return ret; 1311 } 1312 1313 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) 1314 { 1315 bool pressure; 1316 int ret; 1317 1318 ret = __packet_rcv_has_room(po, skb); 1319 pressure = ret != ROOM_NORMAL; 1320 1321 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure) 1322 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure); 1323 1324 return ret; 1325 } 1326 1327 static void __packet_rcv_try_clear_pressure(struct packet_sock *po) 1328 { 1329 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) && 1330 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) 1331 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false); 1332 } 1333 1334 static void packet_rcv_try_clear_pressure(struct packet_sock *po) 1335 { 1336 struct sock *sk = &po->sk; 1337 1338 if (!packet_sock_flag(po, PACKET_SOCK_PRESSURE)) 1339 return; 1340 1341 spin_lock_bh(&sk->sk_receive_queue.lock); 1342 __packet_rcv_try_clear_pressure(po); 1343 spin_unlock_bh(&sk->sk_receive_queue.lock); 1344 } 1345 1346 static void packet_sock_destruct(struct sock *sk) 1347 { 1348 skb_queue_purge(&sk->sk_error_queue); 1349 1350 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 1351 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 1352 1353 packet_free_pending(pkt_sk(sk)); 1354 1355 if (!sock_flag(sk, SOCK_DEAD)) { 1356 pr_err("Attempt to release alive packet socket: %p\n", sk); 1357 return; 1358 } 1359 } 1360 1361 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) 1362 { 1363 u32 *history = po->rollover->history; 1364 u32 victim, rxhash; 1365 int i, count = 0; 1366 1367 rxhash = skb_get_hash(skb); 1368 for (i = 0; i < ROLLOVER_HLEN; i++) 1369 if (READ_ONCE(history[i]) == rxhash) 1370 count++; 1371 1372 victim = get_random_u32_below(ROLLOVER_HLEN); 1373 1374 /* Avoid dirtying the cache line if possible */ 1375 if (READ_ONCE(history[victim]) != rxhash) 1376 WRITE_ONCE(history[victim], rxhash); 1377 1378 return count > (ROLLOVER_HLEN >> 1); 1379 } 1380 1381 static unsigned int fanout_demux_hash(struct packet_fanout *f, 1382 struct sk_buff *skb, 1383 unsigned int num) 1384 { 1385 return reciprocal_scale(__skb_get_hash_symmetric(skb), num); 1386 } 1387 1388 static unsigned int fanout_demux_lb(struct packet_fanout *f, 1389 struct sk_buff *skb, 1390 unsigned int num) 1391 { 1392 unsigned int val = atomic_inc_return(&f->rr_cur); 1393 1394 return val % num; 1395 } 1396 1397 static unsigned int fanout_demux_cpu(struct packet_fanout *f, 1398 struct sk_buff *skb, 1399 unsigned int num) 1400 { 1401 return smp_processor_id() % num; 1402 } 1403 1404 static unsigned int fanout_demux_rnd(struct packet_fanout *f, 1405 struct sk_buff *skb, 1406 unsigned int num) 1407 { 1408 return get_random_u32_below(num); 1409 } 1410 1411 static unsigned int fanout_demux_rollover(struct packet_fanout *f, 1412 struct sk_buff *skb, 1413 unsigned int idx, bool try_self, 1414 unsigned int num) 1415 { 1416 struct packet_sock *po, *po_next, *po_skip = NULL; 1417 unsigned int i, j, room = ROOM_NONE; 1418 1419 po = pkt_sk(rcu_dereference(f->arr[idx])); 1420 1421 if (try_self) { 1422 room = packet_rcv_has_room(po, skb); 1423 if (room == ROOM_NORMAL || 1424 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) 1425 return idx; 1426 po_skip = po; 1427 } 1428 1429 i = j = min_t(int, po->rollover->sock, num - 1); 1430 do { 1431 po_next = pkt_sk(rcu_dereference(f->arr[i])); 1432 if (po_next != po_skip && 1433 !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) && 1434 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { 1435 if (i != j) 1436 po->rollover->sock = i; 1437 atomic_long_inc(&po->rollover->num); 1438 if (room == ROOM_LOW) 1439 atomic_long_inc(&po->rollover->num_huge); 1440 return i; 1441 } 1442 1443 if (++i == num) 1444 i = 0; 1445 } while (i != j); 1446 1447 atomic_long_inc(&po->rollover->num_failed); 1448 return idx; 1449 } 1450 1451 static unsigned int fanout_demux_qm(struct packet_fanout *f, 1452 struct sk_buff *skb, 1453 unsigned int num) 1454 { 1455 return skb_get_queue_mapping(skb) % num; 1456 } 1457 1458 static unsigned int fanout_demux_bpf(struct packet_fanout *f, 1459 struct sk_buff *skb, 1460 unsigned int num) 1461 { 1462 struct bpf_prog *prog; 1463 unsigned int ret = 0; 1464 1465 rcu_read_lock(); 1466 prog = rcu_dereference(f->bpf_prog); 1467 if (prog) 1468 ret = bpf_prog_run_clear_cb(prog, skb) % num; 1469 rcu_read_unlock(); 1470 1471 return ret; 1472 } 1473 1474 static bool fanout_has_flag(struct packet_fanout *f, u16 flag) 1475 { 1476 return f->flags & (flag >> 8); 1477 } 1478 1479 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, 1480 struct packet_type *pt, struct net_device *orig_dev) 1481 { 1482 struct packet_fanout *f = pt->af_packet_priv; 1483 unsigned int num = READ_ONCE(f->num_members); 1484 struct net *net = read_pnet(&f->net); 1485 struct packet_sock *po; 1486 unsigned int idx; 1487 1488 if (!net_eq(dev_net(dev), net) || !num) { 1489 kfree_skb(skb); 1490 return 0; 1491 } 1492 1493 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { 1494 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); 1495 if (!skb) 1496 return 0; 1497 } 1498 switch (f->type) { 1499 case PACKET_FANOUT_HASH: 1500 default: 1501 idx = fanout_demux_hash(f, skb, num); 1502 break; 1503 case PACKET_FANOUT_LB: 1504 idx = fanout_demux_lb(f, skb, num); 1505 break; 1506 case PACKET_FANOUT_CPU: 1507 idx = fanout_demux_cpu(f, skb, num); 1508 break; 1509 case PACKET_FANOUT_RND: 1510 idx = fanout_demux_rnd(f, skb, num); 1511 break; 1512 case PACKET_FANOUT_QM: 1513 idx = fanout_demux_qm(f, skb, num); 1514 break; 1515 case PACKET_FANOUT_ROLLOVER: 1516 idx = fanout_demux_rollover(f, skb, 0, false, num); 1517 break; 1518 case PACKET_FANOUT_CBPF: 1519 case PACKET_FANOUT_EBPF: 1520 idx = fanout_demux_bpf(f, skb, num); 1521 break; 1522 } 1523 1524 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) 1525 idx = fanout_demux_rollover(f, skb, idx, true, num); 1526 1527 po = pkt_sk(rcu_dereference(f->arr[idx])); 1528 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); 1529 } 1530 1531 DEFINE_MUTEX(fanout_mutex); 1532 EXPORT_SYMBOL_GPL(fanout_mutex); 1533 static LIST_HEAD(fanout_list); 1534 static u16 fanout_next_id; 1535 1536 static void __fanout_link(struct sock *sk, struct packet_sock *po) 1537 { 1538 struct packet_fanout *f = po->fanout; 1539 1540 spin_lock(&f->lock); 1541 rcu_assign_pointer(f->arr[f->num_members], sk); 1542 smp_wmb(); 1543 f->num_members++; 1544 if (f->num_members == 1) 1545 dev_add_pack(&f->prot_hook); 1546 spin_unlock(&f->lock); 1547 } 1548 1549 static void __fanout_unlink(struct sock *sk, struct packet_sock *po) 1550 { 1551 struct packet_fanout *f = po->fanout; 1552 int i; 1553 1554 spin_lock(&f->lock); 1555 for (i = 0; i < f->num_members; i++) { 1556 if (rcu_dereference_protected(f->arr[i], 1557 lockdep_is_held(&f->lock)) == sk) 1558 break; 1559 } 1560 BUG_ON(i >= f->num_members); 1561 rcu_assign_pointer(f->arr[i], 1562 rcu_dereference_protected(f->arr[f->num_members - 1], 1563 lockdep_is_held(&f->lock))); 1564 f->num_members--; 1565 if (f->num_members == 0) 1566 __dev_remove_pack(&f->prot_hook); 1567 spin_unlock(&f->lock); 1568 } 1569 1570 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) 1571 { 1572 if (sk->sk_family != PF_PACKET) 1573 return false; 1574 1575 return ptype->af_packet_priv == pkt_sk(sk)->fanout; 1576 } 1577 1578 static void fanout_init_data(struct packet_fanout *f) 1579 { 1580 switch (f->type) { 1581 case PACKET_FANOUT_LB: 1582 atomic_set(&f->rr_cur, 0); 1583 break; 1584 case PACKET_FANOUT_CBPF: 1585 case PACKET_FANOUT_EBPF: 1586 RCU_INIT_POINTER(f->bpf_prog, NULL); 1587 break; 1588 } 1589 } 1590 1591 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) 1592 { 1593 struct bpf_prog *old; 1594 1595 spin_lock(&f->lock); 1596 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); 1597 rcu_assign_pointer(f->bpf_prog, new); 1598 spin_unlock(&f->lock); 1599 1600 if (old) { 1601 synchronize_net(); 1602 bpf_prog_destroy(old); 1603 } 1604 } 1605 1606 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data, 1607 unsigned int len) 1608 { 1609 struct bpf_prog *new; 1610 struct sock_fprog fprog; 1611 int ret; 1612 1613 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1614 return -EPERM; 1615 1616 ret = copy_bpf_fprog_from_user(&fprog, data, len); 1617 if (ret) 1618 return ret; 1619 1620 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); 1621 if (ret) 1622 return ret; 1623 1624 __fanout_set_data_bpf(po->fanout, new); 1625 return 0; 1626 } 1627 1628 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data, 1629 unsigned int len) 1630 { 1631 struct bpf_prog *new; 1632 u32 fd; 1633 1634 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) 1635 return -EPERM; 1636 if (len != sizeof(fd)) 1637 return -EINVAL; 1638 if (copy_from_sockptr(&fd, data, len)) 1639 return -EFAULT; 1640 1641 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); 1642 if (IS_ERR(new)) 1643 return PTR_ERR(new); 1644 1645 __fanout_set_data_bpf(po->fanout, new); 1646 return 0; 1647 } 1648 1649 static int fanout_set_data(struct packet_sock *po, sockptr_t data, 1650 unsigned int len) 1651 { 1652 switch (po->fanout->type) { 1653 case PACKET_FANOUT_CBPF: 1654 return fanout_set_data_cbpf(po, data, len); 1655 case PACKET_FANOUT_EBPF: 1656 return fanout_set_data_ebpf(po, data, len); 1657 default: 1658 return -EINVAL; 1659 } 1660 } 1661 1662 static void fanout_release_data(struct packet_fanout *f) 1663 { 1664 switch (f->type) { 1665 case PACKET_FANOUT_CBPF: 1666 case PACKET_FANOUT_EBPF: 1667 __fanout_set_data_bpf(f, NULL); 1668 } 1669 } 1670 1671 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) 1672 { 1673 struct packet_fanout *f; 1674 1675 list_for_each_entry(f, &fanout_list, list) { 1676 if (f->id == candidate_id && 1677 read_pnet(&f->net) == sock_net(sk)) { 1678 return false; 1679 } 1680 } 1681 return true; 1682 } 1683 1684 static bool fanout_find_new_id(struct sock *sk, u16 *new_id) 1685 { 1686 u16 id = fanout_next_id; 1687 1688 do { 1689 if (__fanout_id_is_free(sk, id)) { 1690 *new_id = id; 1691 fanout_next_id = id + 1; 1692 return true; 1693 } 1694 1695 id++; 1696 } while (id != fanout_next_id); 1697 1698 return false; 1699 } 1700 1701 static int fanout_add(struct sock *sk, struct fanout_args *args) 1702 { 1703 struct packet_rollover *rollover = NULL; 1704 struct packet_sock *po = pkt_sk(sk); 1705 u16 type_flags = args->type_flags; 1706 struct packet_fanout *f, *match; 1707 u8 type = type_flags & 0xff; 1708 u8 flags = type_flags >> 8; 1709 u16 id = args->id; 1710 int err; 1711 1712 switch (type) { 1713 case PACKET_FANOUT_ROLLOVER: 1714 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) 1715 return -EINVAL; 1716 break; 1717 case PACKET_FANOUT_HASH: 1718 case PACKET_FANOUT_LB: 1719 case PACKET_FANOUT_CPU: 1720 case PACKET_FANOUT_RND: 1721 case PACKET_FANOUT_QM: 1722 case PACKET_FANOUT_CBPF: 1723 case PACKET_FANOUT_EBPF: 1724 break; 1725 default: 1726 return -EINVAL; 1727 } 1728 1729 mutex_lock(&fanout_mutex); 1730 1731 err = -EALREADY; 1732 if (po->fanout) 1733 goto out; 1734 1735 if (type == PACKET_FANOUT_ROLLOVER || 1736 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { 1737 err = -ENOMEM; 1738 rollover = kzalloc_obj(*rollover); 1739 if (!rollover) 1740 goto out; 1741 atomic_long_set(&rollover->num, 0); 1742 atomic_long_set(&rollover->num_huge, 0); 1743 atomic_long_set(&rollover->num_failed, 0); 1744 } 1745 1746 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { 1747 if (id != 0) { 1748 err = -EINVAL; 1749 goto out; 1750 } 1751 if (!fanout_find_new_id(sk, &id)) { 1752 err = -ENOMEM; 1753 goto out; 1754 } 1755 /* ephemeral flag for the first socket in the group: drop it */ 1756 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); 1757 } 1758 1759 match = NULL; 1760 list_for_each_entry(f, &fanout_list, list) { 1761 if (f->id == id && 1762 read_pnet(&f->net) == sock_net(sk)) { 1763 match = f; 1764 break; 1765 } 1766 } 1767 err = -EINVAL; 1768 if (match) { 1769 if (match->flags != flags) 1770 goto out; 1771 if (args->max_num_members && 1772 args->max_num_members != match->max_num_members) 1773 goto out; 1774 } else { 1775 if (args->max_num_members > PACKET_FANOUT_MAX) 1776 goto out; 1777 if (!args->max_num_members) 1778 /* legacy PACKET_FANOUT_MAX */ 1779 args->max_num_members = 256; 1780 err = -ENOMEM; 1781 match = kvzalloc_flex(*match, arr, args->max_num_members); 1782 if (!match) 1783 goto out; 1784 write_pnet(&match->net, sock_net(sk)); 1785 match->id = id; 1786 match->type = type; 1787 match->flags = flags; 1788 INIT_LIST_HEAD(&match->list); 1789 spin_lock_init(&match->lock); 1790 refcount_set(&match->sk_ref, 0); 1791 fanout_init_data(match); 1792 match->prot_hook.type = po->prot_hook.type; 1793 match->prot_hook.dev = po->prot_hook.dev; 1794 match->prot_hook.func = packet_rcv_fanout; 1795 match->prot_hook.af_packet_priv = match; 1796 match->prot_hook.af_packet_net = read_pnet(&match->net); 1797 match->prot_hook.id_match = match_fanout_group; 1798 match->max_num_members = args->max_num_members; 1799 match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING; 1800 list_add(&match->list, &fanout_list); 1801 } 1802 err = -EINVAL; 1803 1804 spin_lock(&po->bind_lock); 1805 if (po->num && 1806 match->type == type && 1807 match->prot_hook.type == po->prot_hook.type && 1808 match->prot_hook.dev == po->prot_hook.dev) { 1809 err = -ENOSPC; 1810 if (refcount_read(&match->sk_ref) < match->max_num_members) { 1811 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */ 1812 WRITE_ONCE(po->fanout, match); 1813 1814 po->rollover = rollover; 1815 rollover = NULL; 1816 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); 1817 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 1818 __dev_remove_pack(&po->prot_hook); 1819 __fanout_link(sk, po); 1820 } 1821 err = 0; 1822 } 1823 } 1824 spin_unlock(&po->bind_lock); 1825 1826 if (err && !refcount_read(&match->sk_ref)) { 1827 list_del(&match->list); 1828 kvfree(match); 1829 } 1830 1831 out: 1832 kfree(rollover); 1833 mutex_unlock(&fanout_mutex); 1834 return err; 1835 } 1836 1837 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes 1838 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. 1839 * It is the responsibility of the caller to call fanout_release_data() and 1840 * free the returned packet_fanout (after synchronize_net()) 1841 */ 1842 static struct packet_fanout *fanout_release(struct sock *sk) 1843 { 1844 struct packet_sock *po = pkt_sk(sk); 1845 struct packet_fanout *f; 1846 1847 mutex_lock(&fanout_mutex); 1848 f = po->fanout; 1849 if (f) { 1850 po->fanout = NULL; 1851 1852 if (refcount_dec_and_test(&f->sk_ref)) 1853 list_del(&f->list); 1854 else 1855 f = NULL; 1856 } 1857 mutex_unlock(&fanout_mutex); 1858 1859 return f; 1860 } 1861 1862 static bool packet_extra_vlan_len_allowed(const struct net_device *dev, 1863 struct sk_buff *skb) 1864 { 1865 /* Earlier code assumed this would be a VLAN pkt, double-check 1866 * this now that we have the actual packet in hand. We can only 1867 * do this check on Ethernet devices. 1868 */ 1869 if (unlikely(dev->type != ARPHRD_ETHER)) 1870 return false; 1871 1872 skb_reset_mac_header(skb); 1873 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); 1874 } 1875 1876 static const struct proto_ops packet_ops; 1877 1878 static const struct proto_ops packet_ops_spkt; 1879 1880 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, 1881 struct packet_type *pt, struct net_device *orig_dev) 1882 { 1883 struct sock *sk; 1884 struct sockaddr_pkt *spkt; 1885 1886 /* 1887 * When we registered the protocol we saved the socket in the data 1888 * field for just this event. 1889 */ 1890 1891 sk = pt->af_packet_priv; 1892 1893 /* 1894 * Yank back the headers [hope the device set this 1895 * right or kerboom...] 1896 * 1897 * Incoming packets have ll header pulled, 1898 * push it back. 1899 * 1900 * For outgoing ones skb->data == skb_mac_header(skb) 1901 * so that this procedure is noop. 1902 */ 1903 1904 if (skb->pkt_type == PACKET_LOOPBACK) 1905 goto out; 1906 1907 if (!net_eq(dev_net(dev), sock_net(sk))) 1908 goto out; 1909 1910 skb = skb_share_check(skb, GFP_ATOMIC); 1911 if (skb == NULL) 1912 goto oom; 1913 1914 /* drop any routing info */ 1915 skb_dst_drop(skb); 1916 1917 /* drop conntrack reference */ 1918 nf_reset_ct(skb); 1919 1920 spkt = &PACKET_SKB_CB(skb)->sa.pkt; 1921 1922 if (dev_has_header(dev)) 1923 skb_push(skb, skb->data - skb_mac_header(skb)); 1924 1925 /* 1926 * The SOCK_PACKET socket receives _all_ frames. 1927 */ 1928 1929 spkt->spkt_family = dev->type; 1930 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); 1931 spkt->spkt_protocol = skb->protocol; 1932 1933 /* 1934 * Charge the memory to the socket. This is done specifically 1935 * to prevent sockets using all the memory up. 1936 */ 1937 1938 if (sock_queue_rcv_skb(sk, skb) == 0) 1939 return 0; 1940 1941 out: 1942 kfree_skb(skb); 1943 oom: 1944 return 0; 1945 } 1946 1947 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock) 1948 { 1949 int depth; 1950 1951 /* On TX skb->data is the L2 header; anchor it for all socket types. */ 1952 skb_reset_mac_header(skb); 1953 1954 if (!skb->protocol && sock->type == SOCK_RAW) 1955 skb->protocol = dev_parse_header_protocol(skb); 1956 1957 skb_probe_transport_header(skb); 1958 1959 /* Move network header to the right position for VLAN tagged packets */ 1960 if (likely(skb->dev->type == ARPHRD_ETHER) && 1961 eth_type_vlan(skb->protocol) && 1962 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0) 1963 skb_set_network_header(skb, depth); 1964 } 1965 1966 /* 1967 * Output a raw packet to a device layer. This bypasses all the other 1968 * protocol layers and you must therefore supply it with a complete frame 1969 */ 1970 1971 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, 1972 size_t len) 1973 { 1974 struct sock *sk = sock->sk; 1975 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); 1976 struct sk_buff *skb = NULL; 1977 struct net_device *dev; 1978 struct sockcm_cookie sockc; 1979 __be16 proto = 0; 1980 int hard_header_len; 1981 int extra_len = 0; 1982 int err; 1983 1984 /* 1985 * Get and verify the address. 1986 */ 1987 1988 if (saddr) { 1989 if (msg->msg_namelen < sizeof(struct sockaddr)) 1990 return -EINVAL; 1991 if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) 1992 proto = saddr->spkt_protocol; 1993 } else 1994 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ 1995 1996 /* 1997 * Find the device first to size check it 1998 */ 1999 2000 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; 2001 retry: 2002 rcu_read_lock(); 2003 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); 2004 err = -ENODEV; 2005 if (dev == NULL) 2006 goto out_unlock; 2007 2008 err = -ENETDOWN; 2009 if (!(dev->flags & IFF_UP)) 2010 goto out_unlock; 2011 2012 /* 2013 * You may not queue a frame bigger than the mtu. This is the lowest level 2014 * raw protocol and you must do your own fragmentation at this level. 2015 */ 2016 2017 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 2018 if (!netif_supports_nofcs(dev)) { 2019 err = -EPROTONOSUPPORT; 2020 goto out_unlock; 2021 } 2022 extra_len = 4; /* We're doing our own CRC */ 2023 } 2024 2025 /* Keep the allocation-time header length across retry. */ 2026 if (!skb) 2027 hard_header_len = READ_ONCE(dev->hard_header_len); 2028 2029 err = -EMSGSIZE; 2030 if (len > dev->mtu + hard_header_len + VLAN_HLEN + extra_len) 2031 goto out_unlock; 2032 2033 if (!skb) { 2034 size_t reserved = LL_RESERVED_SPACE_EX(dev, hard_header_len); 2035 int tlen = dev->needed_tailroom; 2036 unsigned int hhlen = dev->header_ops ? hard_header_len : 0; 2037 2038 rcu_read_unlock(); 2039 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); 2040 if (skb == NULL) 2041 return -ENOBUFS; 2042 /* FIXME: Save some space for broken drivers that write a hard 2043 * header at transmission time by themselves. PPP is the notable 2044 * one here. This should really be fixed at the driver level. 2045 */ 2046 skb_reserve(skb, reserved); 2047 skb_reset_network_header(skb); 2048 2049 /* Try to align data part correctly */ 2050 if (hhlen) { 2051 skb->data -= hhlen; 2052 skb->tail -= hhlen; 2053 if (len < hhlen) 2054 skb_reset_network_header(skb); 2055 } 2056 err = memcpy_from_msg(skb_put(skb, len), msg, len); 2057 if (err) 2058 goto out_free; 2059 goto retry; 2060 } 2061 2062 if (!dev_validate_header(dev, skb->data, len) || !skb->len) { 2063 err = -EINVAL; 2064 goto out_unlock; 2065 } 2066 if (len > (dev->mtu + hard_header_len + extra_len) && 2067 !packet_extra_vlan_len_allowed(dev, skb)) { 2068 err = -EMSGSIZE; 2069 goto out_unlock; 2070 } 2071 2072 sockcm_init(&sockc, sk); 2073 if (msg->msg_controllen) { 2074 err = sock_cmsg_send(sk, msg, &sockc); 2075 if (unlikely(err)) 2076 goto out_unlock; 2077 } 2078 2079 skb->protocol = proto; 2080 skb->dev = dev; 2081 skb->priority = sockc.priority; 2082 skb->mark = sockc.mark; 2083 skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid); 2084 skb_setup_tx_timestamp(skb, &sockc); 2085 2086 if (unlikely(extra_len == 4)) 2087 skb->no_fcs = 1; 2088 2089 packet_parse_headers(skb, sock); 2090 2091 dev_queue_xmit(skb); 2092 rcu_read_unlock(); 2093 return len; 2094 2095 out_unlock: 2096 rcu_read_unlock(); 2097 out_free: 2098 kfree_skb(skb); 2099 return err; 2100 } 2101 2102 static unsigned int run_filter(struct sk_buff *skb, 2103 const struct sock *sk, 2104 unsigned int res) 2105 { 2106 struct sk_filter *filter; 2107 2108 rcu_read_lock(); 2109 filter = rcu_dereference(sk->sk_filter); 2110 if (filter != NULL) 2111 res = bpf_prog_run_clear_cb(filter->prog, skb); 2112 rcu_read_unlock(); 2113 2114 return res; 2115 } 2116 2117 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, 2118 size_t *len, int vnet_hdr_sz) 2119 { 2120 struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 }; 2121 2122 if (*len < vnet_hdr_sz) 2123 return -EINVAL; 2124 *len -= vnet_hdr_sz; 2125 2126 if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0)) 2127 return -EINVAL; 2128 2129 return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz); 2130 } 2131 2132 /* 2133 * This function makes lazy skb cloning in hope that most of packets 2134 * are discarded by BPF. 2135 * 2136 * Note tricky part: we DO mangle shared skb! skb->data, skb->len 2137 * and skb->cb are mangled. It works because (and until) packets 2138 * falling here are owned by current CPU. Output packets are cloned 2139 * by dev_queue_xmit_nit(), input packets are processed by net_bh 2140 * sequentially, so that if we return skb to original state on exit, 2141 * we will not harm anyone. 2142 */ 2143 2144 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, 2145 struct packet_type *pt, struct net_device *orig_dev) 2146 { 2147 enum skb_drop_reason drop_reason = SKB_CONSUMED; 2148 struct sock *sk = NULL; 2149 struct sockaddr_ll *sll; 2150 struct packet_sock *po; 2151 u8 *skb_head = skb->data; 2152 int skb_len = skb->len; 2153 unsigned int snaplen, res; 2154 2155 if (skb->pkt_type == PACKET_LOOPBACK) 2156 goto drop; 2157 2158 sk = pt->af_packet_priv; 2159 po = pkt_sk(sk); 2160 2161 if (!net_eq(dev_net(dev), sock_net(sk))) 2162 goto drop; 2163 2164 skb->dev = dev; 2165 2166 if (dev_has_header(dev)) { 2167 /* The device has an explicit notion of ll header, 2168 * exported to higher levels. 2169 * 2170 * Otherwise, the device hides details of its frame 2171 * structure, so that corresponding packet head is 2172 * never delivered to user. 2173 */ 2174 if (sk->sk_type != SOCK_DGRAM) 2175 skb_push(skb, skb->data - skb_mac_header(skb)); 2176 else if (skb->pkt_type == PACKET_OUTGOING) { 2177 /* Special case: outgoing packets have ll header at head */ 2178 skb_pull(skb, skb_network_offset(skb)); 2179 } 2180 } 2181 2182 snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb); 2183 2184 res = run_filter(skb, sk, snaplen); 2185 if (!res) 2186 goto drop_n_restore; 2187 if (snaplen > res) 2188 snaplen = res; 2189 2190 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2191 goto drop_n_acct; 2192 2193 if (skb_shared(skb)) { 2194 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2195 if (nskb == NULL) 2196 goto drop_n_acct; 2197 2198 if (skb_head != skb->data) { 2199 skb->data = skb_head; 2200 skb->len = skb_len; 2201 } 2202 consume_skb(skb); 2203 skb = nskb; 2204 } 2205 2206 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); 2207 2208 sll = &PACKET_SKB_CB(skb)->sa.ll; 2209 sll->sll_hatype = dev->type; 2210 sll->sll_pkttype = skb->pkt_type; 2211 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV))) 2212 sll->sll_ifindex = orig_dev->ifindex; 2213 else 2214 sll->sll_ifindex = dev->ifindex; 2215 2216 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2217 2218 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). 2219 * Use their space for storing the original skb length. 2220 */ 2221 PACKET_SKB_CB(skb)->sa.origlen = skb->len; 2222 2223 if (pskb_trim(skb, snaplen)) 2224 goto drop_n_acct; 2225 2226 skb_set_owner_r(skb, sk); 2227 skb->dev = NULL; 2228 skb_dst_drop(skb); 2229 2230 /* drop conntrack reference */ 2231 nf_reset_ct(skb); 2232 2233 spin_lock(&sk->sk_receive_queue.lock); 2234 po->stats.stats1.tp_packets++; 2235 sock_skb_set_dropcount(sk, skb); 2236 skb_clear_delivery_time(skb); 2237 __skb_queue_tail(&sk->sk_receive_queue, skb); 2238 spin_unlock(&sk->sk_receive_queue.lock); 2239 sk->sk_data_ready(sk); 2240 return 0; 2241 2242 drop_n_acct: 2243 atomic_inc(&po->tp_drops); 2244 sk_drops_inc(sk); 2245 drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR; 2246 2247 drop_n_restore: 2248 if (skb_head != skb->data && skb_shared(skb)) { 2249 skb->data = skb_head; 2250 skb->len = skb_len; 2251 } 2252 drop: 2253 sk_skb_reason_drop(sk, skb, drop_reason); 2254 return 0; 2255 } 2256 2257 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, 2258 struct packet_type *pt, struct net_device *orig_dev) 2259 { 2260 enum skb_drop_reason drop_reason = SKB_CONSUMED; 2261 struct sock *sk = NULL; 2262 struct packet_sock *po; 2263 struct sockaddr_ll *sll; 2264 union tpacket_uhdr h; 2265 u8 *skb_head = skb->data; 2266 int skb_len = skb->len; 2267 unsigned int snaplen, res; 2268 unsigned long status = TP_STATUS_USER; 2269 unsigned short macoff, hdrlen; 2270 unsigned int netoff; 2271 struct sk_buff *copy_skb = NULL; 2272 struct timespec64 ts; 2273 __u32 ts_status; 2274 unsigned int slot_id = 0; 2275 int vnet_hdr_sz = 0; 2276 2277 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. 2278 * We may add members to them until current aligned size without forcing 2279 * userspace to call getsockopt(..., PACKET_HDRLEN, ...). 2280 */ 2281 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); 2282 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); 2283 2284 if (skb->pkt_type == PACKET_LOOPBACK) 2285 goto drop; 2286 2287 sk = pt->af_packet_priv; 2288 po = pkt_sk(sk); 2289 2290 if (!net_eq(dev_net(dev), sock_net(sk))) 2291 goto drop; 2292 2293 if (dev_has_header(dev)) { 2294 if (sk->sk_type != SOCK_DGRAM) 2295 skb_push(skb, skb->data - skb_mac_header(skb)); 2296 else if (skb->pkt_type == PACKET_OUTGOING) { 2297 /* Special case: outgoing packets have ll header at head */ 2298 skb_pull(skb, skb_network_offset(skb)); 2299 } 2300 } 2301 2302 snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb); 2303 2304 res = run_filter(skb, sk, snaplen); 2305 if (!res) 2306 goto drop_n_restore; 2307 2308 /* If we are flooded, just give up */ 2309 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) { 2310 atomic_inc(&po->tp_drops); 2311 goto drop_n_restore; 2312 } 2313 2314 if (skb->ip_summed == CHECKSUM_PARTIAL) 2315 status |= TP_STATUS_CSUMNOTREADY; 2316 else if (skb->pkt_type != PACKET_OUTGOING && 2317 skb_csum_unnecessary(skb)) 2318 status |= TP_STATUS_CSUM_VALID; 2319 if (skb_is_gso(skb) && skb_is_gso_tcp(skb)) 2320 status |= TP_STATUS_GSO_TCP; 2321 2322 if (snaplen > res) 2323 snaplen = res; 2324 2325 if (sk->sk_type == SOCK_DGRAM) { 2326 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + 2327 po->tp_reserve; 2328 } else { 2329 unsigned int maclen = skb_network_offset(skb); 2330 netoff = TPACKET_ALIGN(po->tp_hdrlen + 2331 (maclen < 16 ? 16 : maclen)) + 2332 po->tp_reserve; 2333 vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz); 2334 if (vnet_hdr_sz) 2335 netoff += vnet_hdr_sz; 2336 macoff = netoff - maclen; 2337 } 2338 if (netoff > USHRT_MAX) { 2339 atomic_inc(&po->tp_drops); 2340 goto drop_n_restore; 2341 } 2342 if (po->tp_version <= TPACKET_V2) { 2343 if (macoff + snaplen > po->rx_ring.frame_size) { 2344 if (READ_ONCE(po->copy_thresh) && 2345 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 2346 if (skb_shared(skb)) { 2347 copy_skb = skb_clone(skb, GFP_ATOMIC); 2348 } else { 2349 copy_skb = skb_get(skb); 2350 skb_head = skb->data; 2351 } 2352 if (copy_skb) { 2353 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0, 2354 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll)); 2355 skb_set_owner_r(copy_skb, sk); 2356 } 2357 } 2358 snaplen = po->rx_ring.frame_size - macoff; 2359 if ((int)snaplen < 0) { 2360 snaplen = 0; 2361 vnet_hdr_sz = 0; 2362 } 2363 } 2364 } else if (unlikely(macoff + snaplen > 2365 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { 2366 u32 nval; 2367 2368 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; 2369 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", 2370 snaplen, nval, macoff); 2371 snaplen = nval; 2372 if (unlikely((int)snaplen < 0)) { 2373 snaplen = 0; 2374 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; 2375 vnet_hdr_sz = 0; 2376 } 2377 } 2378 spin_lock(&sk->sk_receive_queue.lock); 2379 h.raw = packet_current_rx_frame(po, skb, 2380 TP_STATUS_KERNEL, (macoff+snaplen)); 2381 if (!h.raw) 2382 goto drop_n_account; 2383 2384 if (po->tp_version <= TPACKET_V2) { 2385 slot_id = po->rx_ring.head; 2386 if (test_bit(slot_id, po->rx_ring.rx_owner_map)) 2387 goto drop_n_account; 2388 __set_bit(slot_id, po->rx_ring.rx_owner_map); 2389 } 2390 2391 if (vnet_hdr_sz && 2392 virtio_net_hdr_from_skb(skb, h.raw + macoff - 2393 sizeof(struct virtio_net_hdr), 2394 vio_le(), true, 0)) { 2395 if (po->tp_version <= TPACKET_V2) 2396 __clear_bit(slot_id, po->rx_ring.rx_owner_map); 2397 else 2398 prb_clear_blk_fill_status(&po->rx_ring); 2399 goto drop_n_account; 2400 } 2401 2402 if (po->tp_version <= TPACKET_V2) { 2403 packet_increment_rx_head(po, &po->rx_ring); 2404 /* 2405 * LOSING will be reported till you read the stats, 2406 * because it's COR - Clear On Read. 2407 * Anyways, moving it for V1/V2 only as V3 doesn't need this 2408 * at packet level. 2409 */ 2410 if (atomic_read(&po->tp_drops)) 2411 status |= TP_STATUS_LOSING; 2412 } 2413 2414 po->stats.stats1.tp_packets++; 2415 if (copy_skb) { 2416 status |= TP_STATUS_COPY; 2417 skb_clear_delivery_time(copy_skb); 2418 __skb_queue_tail(&sk->sk_receive_queue, copy_skb); 2419 } 2420 spin_unlock(&sk->sk_receive_queue.lock); 2421 2422 skb_copy_bits(skb, 0, h.raw + macoff, snaplen); 2423 2424 /* Always timestamp; prefer an existing software timestamp taken 2425 * closer to the time of capture. 2426 */ 2427 ts_status = tpacket_get_timestamp(skb, &ts, 2428 READ_ONCE(po->tp_tstamp) | 2429 SOF_TIMESTAMPING_SOFTWARE); 2430 if (!ts_status) 2431 ktime_get_real_ts64(&ts); 2432 2433 status |= ts_status; 2434 2435 switch (po->tp_version) { 2436 case TPACKET_V1: 2437 h.h1->tp_len = skb->len; 2438 h.h1->tp_snaplen = snaplen; 2439 h.h1->tp_mac = macoff; 2440 h.h1->tp_net = netoff; 2441 h.h1->tp_sec = ts.tv_sec; 2442 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; 2443 hdrlen = sizeof(*h.h1); 2444 break; 2445 case TPACKET_V2: 2446 h.h2->tp_len = skb->len; 2447 h.h2->tp_snaplen = snaplen; 2448 h.h2->tp_mac = macoff; 2449 h.h2->tp_net = netoff; 2450 h.h2->tp_sec = ts.tv_sec; 2451 h.h2->tp_nsec = ts.tv_nsec; 2452 if (skb_vlan_tag_present(skb)) { 2453 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); 2454 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); 2455 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2456 } else if (unlikely(sk->sk_type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) { 2457 h.h2->tp_vlan_tci = vlan_get_tci(skb, skb->dev); 2458 h.h2->tp_vlan_tpid = ntohs(skb->protocol); 2459 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 2460 } else { 2461 h.h2->tp_vlan_tci = 0; 2462 h.h2->tp_vlan_tpid = 0; 2463 } 2464 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); 2465 hdrlen = sizeof(*h.h2); 2466 break; 2467 case TPACKET_V3: 2468 /* tp_nxt_offset,vlan are already populated above. 2469 * So DONT clear those fields here 2470 */ 2471 h.h3->tp_status |= status; 2472 h.h3->tp_len = skb->len; 2473 h.h3->tp_snaplen = snaplen; 2474 h.h3->tp_mac = macoff; 2475 h.h3->tp_net = netoff; 2476 h.h3->tp_sec = ts.tv_sec; 2477 h.h3->tp_nsec = ts.tv_nsec; 2478 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); 2479 hdrlen = sizeof(*h.h3); 2480 break; 2481 default: 2482 BUG(); 2483 } 2484 2485 sll = h.raw + TPACKET_ALIGN(hdrlen); 2486 sll->sll_halen = dev_parse_header(skb, sll->sll_addr); 2487 sll->sll_family = AF_PACKET; 2488 sll->sll_hatype = dev->type; 2489 sll->sll_protocol = (sk->sk_type == SOCK_DGRAM) ? 2490 vlan_get_protocol_dgram(skb) : skb->protocol; 2491 sll->sll_pkttype = skb->pkt_type; 2492 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV))) 2493 sll->sll_ifindex = orig_dev->ifindex; 2494 else 2495 sll->sll_ifindex = dev->ifindex; 2496 2497 smp_mb(); 2498 2499 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 2500 if (po->tp_version <= TPACKET_V2) { 2501 u8 *start, *end; 2502 2503 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + 2504 macoff + snaplen); 2505 2506 for (start = h.raw; start < end; start += PAGE_SIZE) 2507 flush_dcache_page(pgv_to_page(start)); 2508 } 2509 smp_wmb(); 2510 #endif 2511 2512 if (po->tp_version <= TPACKET_V2) { 2513 spin_lock(&sk->sk_receive_queue.lock); 2514 __packet_set_status(po, h.raw, status); 2515 __clear_bit(slot_id, po->rx_ring.rx_owner_map); 2516 spin_unlock(&sk->sk_receive_queue.lock); 2517 sk->sk_data_ready(sk); 2518 } else if (po->tp_version == TPACKET_V3) { 2519 prb_clear_blk_fill_status(&po->rx_ring); 2520 } 2521 2522 drop_n_restore: 2523 if (skb_head != skb->data && skb_shared(skb)) { 2524 skb->data = skb_head; 2525 skb->len = skb_len; 2526 } 2527 drop: 2528 sk_skb_reason_drop(sk, skb, drop_reason); 2529 return 0; 2530 2531 drop_n_account: 2532 spin_unlock(&sk->sk_receive_queue.lock); 2533 atomic_inc(&po->tp_drops); 2534 drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR; 2535 2536 sk->sk_data_ready(sk); 2537 sk_skb_reason_drop(sk, copy_skb, drop_reason); 2538 goto drop_n_restore; 2539 } 2540 2541 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) 2542 { 2543 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 2544 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2545 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > 2546 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) 2547 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), 2548 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + 2549 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); 2550 2551 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) 2552 return -EINVAL; 2553 2554 return 0; 2555 } 2556 2557 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, 2558 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz) 2559 { 2560 int ret; 2561 2562 if (*len < vnet_hdr_sz) 2563 return -EINVAL; 2564 *len -= vnet_hdr_sz; 2565 2566 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) 2567 return -EFAULT; 2568 2569 ret = __packet_snd_vnet_parse(vnet_hdr, *len); 2570 if (ret) 2571 return ret; 2572 2573 /* move iter to point to the start of mac header */ 2574 if (vnet_hdr_sz != sizeof(struct virtio_net_hdr)) 2575 iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr)); 2576 2577 return 0; 2578 } 2579 2580 struct tpacket_uarg { 2581 struct ubuf_info ubuf; 2582 struct packet_sock *po; 2583 void *ph; 2584 }; 2585 2586 static void tpacket_ubuf_complete(struct sk_buff *skb, struct ubuf_info *uarg, 2587 bool success) 2588 { 2589 struct tpacket_uarg *tu = container_of(uarg, struct tpacket_uarg, ubuf); 2590 struct packet_sock *po = tu->po; 2591 void *ph = tu->ph; 2592 __u32 ts; 2593 2594 DEBUG_NET_WARN_ON_ONCE(!skb); 2595 2596 if (!refcount_dec_and_test(&uarg->refcnt)) 2597 return; 2598 2599 ts = __packet_set_timestamp(po, ph, skb); 2600 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); 2601 2602 packet_dec_pending(&po->tx_ring); 2603 complete(&po->skb_completion); 2604 2605 kfree(tu); 2606 sk_free(&po->sk); 2607 } 2608 2609 static const struct ubuf_info_ops tpacket_ubuf_ops = { 2610 .complete = tpacket_ubuf_complete, 2611 }; 2612 2613 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, 2614 struct net_device *dev, void *data, int tp_len, 2615 __be16 proto, unsigned char *addr, int hlen, int copylen, 2616 int hard_header_len, 2617 const struct sockcm_cookie *sockc) 2618 { 2619 int to_write, offset, len, nr_frags, len_max; 2620 struct socket *sock = po->sk.sk_socket; 2621 struct page *page; 2622 int err; 2623 2624 if (proto != htons(ETH_P_ALL)) 2625 skb->protocol = proto; 2626 skb->dev = dev; 2627 skb->priority = sockc->priority; 2628 skb->mark = sockc->mark; 2629 skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, po->sk.sk_clockid); 2630 skb_setup_tx_timestamp(skb, sockc); 2631 2632 skb_reserve(skb, hlen); 2633 skb_reset_network_header(skb); 2634 2635 to_write = tp_len; 2636 2637 if (sock->type == SOCK_DGRAM) { 2638 err = dev_hard_header(skb, dev, ntohs(proto), addr, 2639 NULL, tp_len); 2640 if (unlikely(err < 0)) 2641 return -EINVAL; 2642 } else if (copylen) { 2643 int hdrlen = min_t(int, copylen, tp_len); 2644 2645 skb_push(skb, hard_header_len); 2646 skb_put(skb, copylen - hard_header_len); 2647 err = skb_store_bits(skb, 0, data, hdrlen); 2648 if (unlikely(err)) 2649 return err; 2650 if (!dev_validate_header(dev, skb->data, hdrlen)) 2651 return -EINVAL; 2652 2653 data += hdrlen; 2654 to_write -= hdrlen; 2655 } 2656 2657 offset = offset_in_page(data); 2658 len_max = PAGE_SIZE - offset; 2659 len = ((to_write > len_max) ? len_max : to_write); 2660 2661 skb->data_len = to_write; 2662 skb->len += to_write; 2663 skb->truesize += to_write; 2664 refcount_add(to_write, &po->sk.sk_wmem_alloc); 2665 2666 while (likely(to_write)) { 2667 nr_frags = skb_shinfo(skb)->nr_frags; 2668 2669 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { 2670 pr_err("Packet exceed the number of skb frags(%u)\n", 2671 (unsigned int)MAX_SKB_FRAGS); 2672 return -EFAULT; 2673 } 2674 2675 page = pgv_to_page(data); 2676 data += len; 2677 flush_dcache_page(page); 2678 get_page(page); 2679 skb_fill_page_desc(skb, nr_frags, page, offset, len); 2680 to_write -= len; 2681 offset = 0; 2682 len_max = PAGE_SIZE; 2683 len = ((to_write > len_max) ? len_max : to_write); 2684 } 2685 2686 if (unlikely(!skb->len)) 2687 return -EINVAL; 2688 2689 return tp_len; 2690 } 2691 2692 static int tpacket_parse_header(struct packet_sock *po, void *frame, 2693 int size_max, void **data) 2694 { 2695 union tpacket_uhdr ph; 2696 u32 tp_len; 2697 int off; 2698 2699 ph.raw = frame; 2700 2701 switch (po->tp_version) { 2702 case TPACKET_V3: 2703 if (ph.h3->tp_next_offset != 0) { 2704 pr_warn_once("variable sized slot not supported"); 2705 return -EINVAL; 2706 } 2707 tp_len = ph.h3->tp_len; 2708 break; 2709 case TPACKET_V2: 2710 tp_len = ph.h2->tp_len; 2711 break; 2712 default: 2713 tp_len = ph.h1->tp_len; 2714 break; 2715 } 2716 if (unlikely(tp_len > size_max)) { 2717 pr_err("packet size is too long (%u > %d)\n", tp_len, size_max); 2718 return -EMSGSIZE; 2719 } 2720 2721 if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) { 2722 int off_min, off_max; 2723 2724 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2725 off_max = po->tx_ring.frame_size - tp_len; 2726 if (po->sk.sk_type == SOCK_DGRAM) { 2727 switch (po->tp_version) { 2728 case TPACKET_V3: 2729 off = ph.h3->tp_net; 2730 break; 2731 case TPACKET_V2: 2732 off = ph.h2->tp_net; 2733 break; 2734 default: 2735 off = ph.h1->tp_net; 2736 break; 2737 } 2738 } else { 2739 switch (po->tp_version) { 2740 case TPACKET_V3: 2741 off = ph.h3->tp_mac; 2742 break; 2743 case TPACKET_V2: 2744 off = ph.h2->tp_mac; 2745 break; 2746 default: 2747 off = ph.h1->tp_mac; 2748 break; 2749 } 2750 } 2751 if (unlikely((off < off_min) || (off_max < off))) 2752 return -EINVAL; 2753 } else { 2754 off = po->tp_hdrlen - sizeof(struct sockaddr_ll); 2755 } 2756 2757 *data = frame + off; 2758 return tp_len; 2759 } 2760 2761 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) 2762 { 2763 struct sk_buff *skb = NULL; 2764 struct net_device *dev; 2765 struct virtio_net_hdr vnet_hdr; 2766 bool has_vnet_hdr = false; 2767 struct sockcm_cookie sockc; 2768 struct tpacket_uarg *uarg; 2769 __be16 proto; 2770 int err, reserve = 0; 2771 void *ph; 2772 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 2773 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); 2774 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz); 2775 unsigned char *addr = NULL; 2776 int tp_len, size_max; 2777 void *data; 2778 int len_sum = 0; 2779 int status = TP_STATUS_AVAILABLE; 2780 int hard_header_len, hlen, tlen, copylen = 0; 2781 long timeo; 2782 2783 mutex_lock(&po->pg_vec_lock); 2784 2785 /* packet_sendmsg() check on tx_ring.pg_vec was lockless, 2786 * we need to confirm it under protection of pg_vec_lock. 2787 */ 2788 if (unlikely(!po->tx_ring.pg_vec)) { 2789 err = -EBUSY; 2790 goto out; 2791 } 2792 if (likely(saddr == NULL)) { 2793 dev = packet_cached_dev_get(po); 2794 proto = READ_ONCE(po->num); 2795 } else { 2796 err = -EINVAL; 2797 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 2798 goto out; 2799 if (msg->msg_namelen < (saddr->sll_halen 2800 + offsetof(struct sockaddr_ll, 2801 sll_addr))) 2802 goto out; 2803 proto = saddr->sll_protocol; 2804 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); 2805 if (po->sk.sk_socket->type == SOCK_DGRAM) { 2806 if (dev && msg->msg_namelen < dev->addr_len + 2807 offsetof(struct sockaddr_ll, sll_addr)) 2808 goto out_put; 2809 addr = saddr->sll_addr; 2810 } 2811 } 2812 2813 err = -ENXIO; 2814 if (unlikely(dev == NULL)) 2815 goto out; 2816 err = -ENETDOWN; 2817 if (unlikely(!(dev->flags & IFF_UP))) 2818 goto out_put; 2819 2820 sockcm_init(&sockc, &po->sk); 2821 if (msg->msg_controllen) { 2822 err = sock_cmsg_send(&po->sk, msg, &sockc); 2823 if (unlikely(err)) 2824 goto out_put; 2825 } 2826 2827 hard_header_len = READ_ONCE(dev->hard_header_len); 2828 if (po->sk.sk_socket->type == SOCK_RAW) 2829 reserve = hard_header_len; 2830 size_max = po->tx_ring.frame_size 2831 - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); 2832 2833 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz) 2834 size_max = dev->mtu + reserve + VLAN_HLEN; 2835 2836 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT); 2837 reinit_completion(&po->skb_completion); 2838 2839 do { 2840 ph = packet_current_frame(po, &po->tx_ring, 2841 TP_STATUS_SEND_REQUEST); 2842 if (unlikely(ph == NULL)) { 2843 /* Note: packet_read_pending() might be slow if we 2844 * have to call it as it's per_cpu variable, but in 2845 * fast-path we don't have to call it, only when ph 2846 * is NULL, we need to check the pending_refcnt. 2847 */ 2848 if (need_wait && packet_read_pending(&po->tx_ring)) { 2849 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo); 2850 if (timeo <= 0) { 2851 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS; 2852 goto out_put; 2853 } 2854 /* check for additional frames */ 2855 continue; 2856 } else 2857 break; 2858 } 2859 2860 skb = NULL; 2861 tp_len = tpacket_parse_header(po, ph, size_max, &data); 2862 if (tp_len < 0) 2863 goto tpacket_error; 2864 2865 status = TP_STATUS_SEND_REQUEST; 2866 hlen = LL_RESERVED_SPACE_EX(dev, hard_header_len); 2867 tlen = dev->needed_tailroom; 2868 if (vnet_hdr_sz) { 2869 data += vnet_hdr_sz; 2870 tp_len -= vnet_hdr_sz; 2871 if (tp_len < 0) { 2872 tp_len = -EINVAL; 2873 goto tpacket_error; 2874 } 2875 memcpy(&vnet_hdr, data - vnet_hdr_sz, sizeof(vnet_hdr)); 2876 if (__packet_snd_vnet_parse(&vnet_hdr, tp_len)) { 2877 tp_len = -EINVAL; 2878 goto tpacket_error; 2879 } 2880 copylen = __virtio16_to_cpu(vio_le(), 2881 vnet_hdr.hdr_len); 2882 has_vnet_hdr = true; 2883 } 2884 /* Prevent the byte count from wrapping into a negative errno */ 2885 if (unlikely(tp_len > INT_MAX - len_sum)) 2886 break; 2887 copylen = max_t(int, copylen, hard_header_len); 2888 skb = sock_alloc_send_skb(&po->sk, 2889 hlen + tlen + sizeof(struct sockaddr_ll) + 2890 (copylen - hard_header_len), 2891 !need_wait, &err); 2892 2893 if (unlikely(skb == NULL)) { 2894 /* we assume the socket was initially writeable ... */ 2895 if (likely(len_sum > 0)) 2896 err = len_sum; 2897 goto out_status; 2898 } 2899 tp_len = tpacket_fill_skb(po, skb, dev, data, tp_len, proto, 2900 addr, hlen, copylen, hard_header_len, 2901 &sockc); 2902 if (likely(tp_len >= 0) && 2903 tp_len > dev->mtu + reserve && 2904 !vnet_hdr_sz && 2905 !packet_extra_vlan_len_allowed(dev, skb)) 2906 tp_len = -EMSGSIZE; 2907 2908 if (unlikely(tp_len < 0)) { 2909 tpacket_error: 2910 if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) { 2911 __packet_set_status(po, ph, 2912 TP_STATUS_AVAILABLE); 2913 packet_increment_head(&po->tx_ring); 2914 kfree_skb(skb); 2915 continue; 2916 } else { 2917 status = TP_STATUS_WRONG_FORMAT; 2918 err = tp_len; 2919 goto out_status; 2920 } 2921 } 2922 2923 if (has_vnet_hdr) { 2924 if (virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le())) { 2925 tp_len = -EINVAL; 2926 goto tpacket_error; 2927 } 2928 } 2929 2930 packet_parse_headers(skb, po->sk.sk_socket); 2931 2932 if (has_vnet_hdr) 2933 virtio_net_hdr_set_proto(skb, &vnet_hdr); 2934 2935 uarg = kmalloc(sizeof(*uarg), GFP_KERNEL); 2936 if (unlikely(!uarg)) { 2937 if (likely(len_sum > 0)) 2938 err = len_sum; 2939 else 2940 err = -ENOMEM; 2941 goto out_status; 2942 } 2943 uarg->po = po; 2944 uarg->ph = ph; 2945 uarg->ubuf.ops = &tpacket_ubuf_ops; 2946 uarg->ubuf.flags = SKBFL_ZEROCOPY_FRAG; 2947 refcount_set(&uarg->ubuf.refcnt, 1); 2948 2949 /* Hold a sk_wmem_alloc reference until completion */ 2950 refcount_inc(&po->sk.sk_wmem_alloc); 2951 skb_zcopy_init(skb, &uarg->ubuf); 2952 2953 __packet_set_status(po, ph, TP_STATUS_SENDING); 2954 packet_inc_pending(&po->tx_ring); 2955 2956 status = TP_STATUS_SEND_REQUEST; 2957 err = packet_xmit(po, skb); 2958 if (unlikely(err != 0)) { 2959 if (err > 0) 2960 err = net_xmit_errno(err); 2961 if (err && (__packet_get_status(po, ph) & ~tp_status_ts_mask) == 2962 TP_STATUS_AVAILABLE) { 2963 /* skb was destructed already */ 2964 skb = NULL; 2965 goto out_status; 2966 } 2967 /* 2968 * skb was dropped but not destructed yet; 2969 * let's treat it like congestion or err < 0 2970 */ 2971 err = 0; 2972 } 2973 packet_increment_head(&po->tx_ring); 2974 len_sum += tp_len; 2975 } while (1); 2976 2977 err = len_sum; 2978 goto out_put; 2979 2980 out_status: 2981 __packet_set_status(po, ph, status); 2982 kfree_skb(skb); 2983 out_put: 2984 dev_put(dev); 2985 out: 2986 mutex_unlock(&po->pg_vec_lock); 2987 return err; 2988 } 2989 2990 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, 2991 size_t reserve, size_t len, 2992 size_t linear, int noblock, 2993 int *err) 2994 { 2995 struct sk_buff *skb; 2996 2997 /* Under a page? Don't bother with paged skb. */ 2998 if (prepad + len < PAGE_SIZE || !linear) 2999 linear = len; 3000 3001 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) 3002 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER); 3003 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 3004 err, PAGE_ALLOC_COSTLY_ORDER); 3005 if (!skb) 3006 return NULL; 3007 3008 skb_reserve(skb, reserve); 3009 skb_put(skb, linear); 3010 skb->data_len = len - linear; 3011 skb->len += len - linear; 3012 3013 return skb; 3014 } 3015 3016 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) 3017 { 3018 struct sock *sk = sock->sk; 3019 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); 3020 struct sk_buff *skb; 3021 struct net_device *dev; 3022 __be16 proto; 3023 unsigned char *addr = NULL; 3024 int err, reserve = 0; 3025 struct sockcm_cookie sockc; 3026 struct virtio_net_hdr vnet_hdr = { 0 }; 3027 int offset = 0; 3028 struct packet_sock *po = pkt_sk(sk); 3029 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz); 3030 int hard_header_len, hlen, tlen, linear; 3031 int extra_len = 0; 3032 3033 /* 3034 * Get and verify the address. 3035 */ 3036 3037 if (likely(saddr == NULL)) { 3038 dev = packet_cached_dev_get(po); 3039 proto = READ_ONCE(po->num); 3040 } else { 3041 err = -EINVAL; 3042 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) 3043 goto out; 3044 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) 3045 goto out; 3046 proto = saddr->sll_protocol; 3047 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); 3048 if (sock->type == SOCK_DGRAM) { 3049 if (dev && msg->msg_namelen < dev->addr_len + 3050 offsetof(struct sockaddr_ll, sll_addr)) 3051 goto out_unlock; 3052 addr = saddr->sll_addr; 3053 } 3054 } 3055 3056 err = -ENXIO; 3057 if (unlikely(dev == NULL)) 3058 goto out_unlock; 3059 err = -ENETDOWN; 3060 if (unlikely(!(dev->flags & IFF_UP))) 3061 goto out_unlock; 3062 3063 sockcm_init(&sockc, sk); 3064 if (msg->msg_controllen) { 3065 err = sock_cmsg_send(sk, msg, &sockc); 3066 if (unlikely(err)) 3067 goto out_unlock; 3068 } 3069 3070 hard_header_len = READ_ONCE(dev->hard_header_len); 3071 if (sock->type == SOCK_RAW) 3072 reserve = hard_header_len; 3073 if (vnet_hdr_sz) { 3074 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz); 3075 if (err) 3076 goto out_unlock; 3077 } 3078 3079 if (unlikely(sock_flag(sk, SOCK_NOFCS))) { 3080 if (!netif_supports_nofcs(dev)) { 3081 err = -EPROTONOSUPPORT; 3082 goto out_unlock; 3083 } 3084 extra_len = 4; /* We're doing our own CRC */ 3085 } 3086 3087 err = -EMSGSIZE; 3088 if (!vnet_hdr.gso_type && 3089 (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) 3090 goto out_unlock; 3091 3092 err = -ENOBUFS; 3093 hlen = LL_RESERVED_SPACE_EX(dev, hard_header_len); 3094 tlen = dev->needed_tailroom; 3095 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); 3096 linear = max(linear, min_t(int, len, hard_header_len)); 3097 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, 3098 msg->msg_flags & MSG_DONTWAIT, &err); 3099 if (skb == NULL) 3100 goto out_unlock; 3101 3102 skb_reset_network_header(skb); 3103 3104 err = -EINVAL; 3105 if (sock->type == SOCK_DGRAM) { 3106 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); 3107 if (unlikely(offset < 0)) 3108 goto out_free; 3109 } else if (reserve) { 3110 skb_reserve(skb, -reserve); 3111 if (len < reserve + sizeof(struct ipv6hdr) && 3112 dev->min_header_len != hard_header_len) 3113 skb_reset_network_header(skb); 3114 } 3115 3116 /* Returns -EFAULT on error */ 3117 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); 3118 if (err) 3119 goto out_free; 3120 3121 if ((sock->type == SOCK_RAW && 3122 !dev_validate_header(dev, skb->data, len)) || !skb->len) { 3123 err = -EINVAL; 3124 goto out_free; 3125 } 3126 3127 skb_setup_tx_timestamp(skb, &sockc); 3128 3129 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && 3130 !packet_extra_vlan_len_allowed(dev, skb)) { 3131 err = -EMSGSIZE; 3132 goto out_free; 3133 } 3134 3135 if (proto != htons(ETH_P_ALL)) 3136 skb->protocol = proto; 3137 skb->dev = dev; 3138 skb->priority = sockc.priority; 3139 skb->mark = sockc.mark; 3140 skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid); 3141 3142 if (unlikely(extra_len == 4)) 3143 skb->no_fcs = 1; 3144 3145 if (vnet_hdr_sz) { 3146 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); 3147 if (err) 3148 goto out_free; 3149 len += vnet_hdr_sz; 3150 } 3151 3152 packet_parse_headers(skb, sock); 3153 3154 if (vnet_hdr_sz) 3155 virtio_net_hdr_set_proto(skb, &vnet_hdr); 3156 3157 err = packet_xmit(po, skb); 3158 3159 if (unlikely(err != 0)) { 3160 if (err > 0) 3161 err = net_xmit_errno(err); 3162 if (err) 3163 goto out_unlock; 3164 } 3165 3166 dev_put(dev); 3167 3168 return len; 3169 3170 out_free: 3171 kfree_skb(skb); 3172 out_unlock: 3173 dev_put(dev); 3174 out: 3175 return err; 3176 } 3177 3178 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 3179 { 3180 struct sock *sk = sock->sk; 3181 struct packet_sock *po = pkt_sk(sk); 3182 3183 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy. 3184 * tpacket_snd() will redo the check safely. 3185 */ 3186 if (data_race(po->tx_ring.pg_vec)) 3187 return tpacket_snd(po, msg); 3188 3189 return packet_snd(sock, msg, len); 3190 } 3191 3192 /* 3193 * Close a PACKET socket. This is fairly simple. We immediately go 3194 * to 'closed' state and remove our protocol entry in the device list. 3195 */ 3196 3197 static int packet_release(struct socket *sock) 3198 { 3199 struct sock *sk = sock->sk; 3200 struct packet_sock *po; 3201 struct packet_fanout *f; 3202 struct net *net; 3203 union tpacket_req_u req_u; 3204 3205 if (!sk) 3206 return 0; 3207 3208 net = sock_net(sk); 3209 po = pkt_sk(sk); 3210 3211 mutex_lock(&net->packet.sklist_lock); 3212 sk_del_node_init_rcu(sk); 3213 mutex_unlock(&net->packet.sklist_lock); 3214 3215 sock_prot_inuse_add(net, sk->sk_prot, -1); 3216 3217 spin_lock(&po->bind_lock); 3218 unregister_prot_hook(sk, false); 3219 WRITE_ONCE(po->num, 0); 3220 packet_cached_dev_reset(po); 3221 3222 if (po->prot_hook.dev) { 3223 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker); 3224 po->prot_hook.dev = NULL; 3225 } 3226 spin_unlock(&po->bind_lock); 3227 3228 packet_flush_mclist(sk); 3229 3230 lock_sock(sk); 3231 if (po->rx_ring.pg_vec) { 3232 memset(&req_u, 0, sizeof(req_u)); 3233 packet_set_ring(sk, &req_u, 1, 0); 3234 } 3235 3236 if (po->tx_ring.pg_vec) { 3237 memset(&req_u, 0, sizeof(req_u)); 3238 packet_set_ring(sk, &req_u, 1, 1); 3239 } 3240 release_sock(sk); 3241 3242 f = fanout_release(sk); 3243 3244 synchronize_net(); 3245 3246 kfree(po->rollover); 3247 if (f) { 3248 fanout_release_data(f); 3249 kvfree(f); 3250 } 3251 /* 3252 * Now the socket is dead. No more input will appear. 3253 */ 3254 sock_orphan(sk); 3255 sock->sk = NULL; 3256 3257 /* Purge queues */ 3258 3259 skb_queue_purge(&sk->sk_receive_queue); 3260 3261 sock_put(sk); 3262 return 0; 3263 } 3264 3265 /* 3266 * Attach a packet hook. 3267 */ 3268 3269 static int packet_do_bind(struct sock *sk, const char *name, int ifindex, 3270 __be16 proto) 3271 { 3272 struct packet_sock *po = pkt_sk(sk); 3273 struct net_device *dev = NULL; 3274 bool unlisted = false; 3275 bool need_rehook; 3276 int ret = 0; 3277 3278 lock_sock(sk); 3279 spin_lock(&po->bind_lock); 3280 if (!proto) 3281 proto = po->num; 3282 3283 rcu_read_lock(); 3284 3285 if (po->fanout) { 3286 ret = -EINVAL; 3287 goto out_unlock; 3288 } 3289 3290 if (name) { 3291 dev = dev_get_by_name_rcu(sock_net(sk), name); 3292 if (!dev) { 3293 ret = -ENODEV; 3294 goto out_unlock; 3295 } 3296 } else if (ifindex) { 3297 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3298 if (!dev) { 3299 ret = -ENODEV; 3300 goto out_unlock; 3301 } 3302 } 3303 3304 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev; 3305 3306 if (need_rehook) { 3307 dev_hold(dev); 3308 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 3309 rcu_read_unlock(); 3310 /* prevents packet_notifier() from calling 3311 * register_prot_hook() 3312 */ 3313 WRITE_ONCE(po->num, 0); 3314 __unregister_prot_hook(sk, true); 3315 rcu_read_lock(); 3316 if (dev) 3317 unlisted = !dev_get_by_index_rcu(sock_net(sk), 3318 dev->ifindex); 3319 } 3320 3321 BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING)); 3322 WRITE_ONCE(po->num, proto); 3323 po->prot_hook.type = proto; 3324 3325 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker); 3326 3327 if (unlikely(unlisted)) { 3328 po->prot_hook.dev = NULL; 3329 WRITE_ONCE(po->ifindex, -1); 3330 packet_cached_dev_reset(po); 3331 } else { 3332 netdev_hold(dev, &po->prot_hook.dev_tracker, 3333 GFP_ATOMIC); 3334 po->prot_hook.dev = dev; 3335 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0); 3336 packet_cached_dev_assign(po, dev); 3337 } 3338 dev_put(dev); 3339 } 3340 3341 if (proto == 0 || !need_rehook) 3342 goto out_unlock; 3343 3344 if (!unlisted && (!dev || (dev->flags & IFF_UP))) { 3345 register_prot_hook(sk); 3346 } else { 3347 sk->sk_err = ENETDOWN; 3348 if (!sock_flag(sk, SOCK_DEAD)) 3349 sk_error_report(sk); 3350 } 3351 3352 out_unlock: 3353 rcu_read_unlock(); 3354 spin_unlock(&po->bind_lock); 3355 release_sock(sk); 3356 return ret; 3357 } 3358 3359 /* 3360 * Bind a packet socket to a device 3361 */ 3362 3363 static int packet_bind_spkt(struct socket *sock, struct sockaddr_unsized *uaddr, 3364 int addr_len) 3365 { 3366 struct sock *sk = sock->sk; 3367 struct sockaddr *sa = (struct sockaddr *)uaddr; 3368 char name[sizeof(sa->sa_data) + 1]; 3369 3370 /* 3371 * Check legality 3372 */ 3373 3374 if (addr_len != sizeof(struct sockaddr)) 3375 return -EINVAL; 3376 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be 3377 * zero-terminated. 3378 */ 3379 memcpy(name, sa->sa_data, sizeof(sa->sa_data)); 3380 name[sizeof(sa->sa_data)] = 0; 3381 3382 return packet_do_bind(sk, name, 0, 0); 3383 } 3384 3385 static int packet_bind(struct socket *sock, struct sockaddr_unsized *uaddr, int addr_len) 3386 { 3387 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; 3388 struct sock *sk = sock->sk; 3389 3390 /* 3391 * Check legality 3392 */ 3393 3394 if (addr_len < sizeof(struct sockaddr_ll)) 3395 return -EINVAL; 3396 if (sll->sll_family != AF_PACKET) 3397 return -EINVAL; 3398 3399 return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol); 3400 } 3401 3402 static struct proto packet_proto = { 3403 .name = "PACKET", 3404 .owner = THIS_MODULE, 3405 .obj_size = sizeof(struct packet_sock), 3406 }; 3407 3408 /* 3409 * Create a packet of type SOCK_PACKET. 3410 */ 3411 3412 static int packet_create(struct net *net, struct socket *sock, int protocol, 3413 int kern) 3414 { 3415 struct sock *sk; 3416 struct packet_sock *po; 3417 __be16 proto = (__force __be16)protocol; /* weird, but documented */ 3418 int err; 3419 3420 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 3421 return -EPERM; 3422 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && 3423 sock->type != SOCK_PACKET) 3424 return -ESOCKTNOSUPPORT; 3425 3426 sock->state = SS_UNCONNECTED; 3427 3428 err = -ENOBUFS; 3429 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); 3430 if (sk == NULL) 3431 goto out; 3432 3433 sock->ops = &packet_ops; 3434 if (sock->type == SOCK_PACKET) 3435 sock->ops = &packet_ops_spkt; 3436 3437 po = pkt_sk(sk); 3438 err = packet_alloc_pending(po); 3439 if (err) 3440 goto out_sk_free; 3441 3442 sock_init_data(sock, sk); 3443 3444 init_completion(&po->skb_completion); 3445 sk->sk_family = PF_PACKET; 3446 po->num = proto; 3447 3448 packet_cached_dev_reset(po); 3449 3450 sk->sk_destruct = packet_sock_destruct; 3451 3452 /* 3453 * Attach a protocol block 3454 */ 3455 3456 spin_lock_init(&po->bind_lock); 3457 mutex_init(&po->pg_vec_lock); 3458 po->rollover = NULL; 3459 po->prot_hook.func = packet_rcv; 3460 3461 if (sock->type == SOCK_PACKET) 3462 po->prot_hook.func = packet_rcv_spkt; 3463 3464 po->prot_hook.af_packet_priv = sk; 3465 po->prot_hook.af_packet_net = sock_net(sk); 3466 3467 if (proto) { 3468 po->prot_hook.type = proto; 3469 __register_prot_hook(sk); 3470 } 3471 3472 mutex_lock(&net->packet.sklist_lock); 3473 sk_add_node_tail_rcu(sk, &net->packet.sklist); 3474 mutex_unlock(&net->packet.sklist_lock); 3475 3476 sock_prot_inuse_add(net, &packet_proto, 1); 3477 3478 return 0; 3479 out_sk_free: 3480 sk_free(sk); 3481 out: 3482 return err; 3483 } 3484 3485 /* 3486 * Pull a packet from our receive queue and hand it to the user. 3487 * If necessary we block. 3488 */ 3489 3490 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 3491 int flags) 3492 { 3493 struct sock *sk = sock->sk; 3494 struct sk_buff *skb; 3495 int copied, err; 3496 int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz); 3497 unsigned int origlen = 0; 3498 3499 err = -EINVAL; 3500 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) 3501 goto out; 3502 3503 #if 0 3504 /* What error should we return now? EUNATTACH? */ 3505 if (pkt_sk(sk)->ifindex < 0) 3506 return -ENODEV; 3507 #endif 3508 3509 if (flags & MSG_ERRQUEUE) { 3510 err = sock_recv_errqueue(sk, msg, len, 3511 SOL_PACKET, PACKET_TX_TIMESTAMP); 3512 goto out; 3513 } 3514 3515 /* 3516 * Call the generic datagram receiver. This handles all sorts 3517 * of horrible races and re-entrancy so we can forget about it 3518 * in the protocol layers. 3519 * 3520 * Now it will return ENETDOWN, if device have just gone down, 3521 * but then it will block. 3522 */ 3523 3524 skb = skb_recv_datagram(sk, flags, &err); 3525 3526 /* 3527 * An error occurred so return it. Because skb_recv_datagram() 3528 * handles the blocking we don't see and worry about blocking 3529 * retries. 3530 */ 3531 3532 if (skb == NULL) 3533 goto out; 3534 3535 packet_rcv_try_clear_pressure(pkt_sk(sk)); 3536 3537 if (vnet_hdr_len) { 3538 err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len); 3539 if (err) 3540 goto out_free; 3541 } 3542 3543 /* You lose any data beyond the buffer you gave. If it worries 3544 * a user program they can ask the device for its MTU 3545 * anyway. 3546 */ 3547 copied = skb->len; 3548 if (copied > len) { 3549 copied = len; 3550 msg->msg_flags |= MSG_TRUNC; 3551 } 3552 3553 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3554 if (err) 3555 goto out_free; 3556 3557 if (sock->type != SOCK_PACKET) { 3558 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3559 3560 /* Original length was stored in sockaddr_ll fields */ 3561 origlen = PACKET_SKB_CB(skb)->sa.origlen; 3562 sll->sll_family = AF_PACKET; 3563 sll->sll_protocol = (sock->type == SOCK_DGRAM) ? 3564 vlan_get_protocol_dgram(skb) : skb->protocol; 3565 } 3566 3567 sock_recv_cmsgs(msg, sk, skb); 3568 3569 if (msg->msg_name) { 3570 const size_t max_len = min(sizeof(skb->cb), 3571 sizeof(struct sockaddr_storage)); 3572 int copy_len; 3573 3574 /* If the address length field is there to be filled 3575 * in, we fill it in now. 3576 */ 3577 if (sock->type == SOCK_PACKET) { 3578 __sockaddr_check_size(sizeof(struct sockaddr_pkt)); 3579 msg->msg_namelen = sizeof(struct sockaddr_pkt); 3580 copy_len = msg->msg_namelen; 3581 } else { 3582 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3583 3584 msg->msg_namelen = sll->sll_halen + 3585 offsetof(struct sockaddr_ll, sll_addr); 3586 copy_len = msg->msg_namelen; 3587 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { 3588 memset(msg->msg_name + 3589 offsetof(struct sockaddr_ll, sll_addr), 3590 0, sizeof(sll->sll_addr)); 3591 msg->msg_namelen = sizeof(struct sockaddr_ll); 3592 } 3593 } 3594 if (WARN_ON_ONCE(copy_len > max_len)) { 3595 copy_len = max_len; 3596 msg->msg_namelen = copy_len; 3597 } 3598 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); 3599 } 3600 3601 if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) { 3602 struct tpacket_auxdata aux; 3603 3604 aux.tp_status = TP_STATUS_USER; 3605 if (skb->ip_summed == CHECKSUM_PARTIAL) 3606 aux.tp_status |= TP_STATUS_CSUMNOTREADY; 3607 else if (skb->pkt_type != PACKET_OUTGOING && 3608 skb_csum_unnecessary(skb)) 3609 aux.tp_status |= TP_STATUS_CSUM_VALID; 3610 if (skb_is_gso(skb) && skb_is_gso_tcp(skb)) 3611 aux.tp_status |= TP_STATUS_GSO_TCP; 3612 3613 aux.tp_len = origlen; 3614 aux.tp_snaplen = skb->len; 3615 aux.tp_mac = 0; 3616 aux.tp_net = skb_network_offset(skb); 3617 if (skb_vlan_tag_present(skb)) { 3618 aux.tp_vlan_tci = skb_vlan_tag_get(skb); 3619 aux.tp_vlan_tpid = ntohs(skb->vlan_proto); 3620 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3621 } else if (unlikely(sock->type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) { 3622 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; 3623 struct net_device *dev; 3624 3625 rcu_read_lock(); 3626 dev = dev_get_by_index_rcu(sock_net(sk), sll->sll_ifindex); 3627 if (dev) { 3628 aux.tp_vlan_tci = vlan_get_tci(skb, dev); 3629 aux.tp_vlan_tpid = ntohs(skb->protocol); 3630 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; 3631 } else { 3632 aux.tp_vlan_tci = 0; 3633 aux.tp_vlan_tpid = 0; 3634 } 3635 rcu_read_unlock(); 3636 } else { 3637 aux.tp_vlan_tci = 0; 3638 aux.tp_vlan_tpid = 0; 3639 } 3640 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); 3641 } 3642 3643 /* 3644 * Free or return the buffer as appropriate. Again this 3645 * hides all the races and re-entrancy issues from us. 3646 */ 3647 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); 3648 3649 out_free: 3650 skb_free_datagram(sk, skb); 3651 out: 3652 return err; 3653 } 3654 3655 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, 3656 int peer) 3657 { 3658 struct net_device *dev; 3659 struct sock *sk = sock->sk; 3660 3661 if (peer) 3662 return -EOPNOTSUPP; 3663 3664 uaddr->sa_family = AF_PACKET; 3665 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); 3666 rcu_read_lock(); 3667 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex)); 3668 if (dev) 3669 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); 3670 rcu_read_unlock(); 3671 3672 return sizeof(*uaddr); 3673 } 3674 3675 static int packet_getname(struct socket *sock, struct sockaddr *uaddr, 3676 int peer) 3677 { 3678 struct net_device *dev; 3679 struct sock *sk = sock->sk; 3680 struct packet_sock *po = pkt_sk(sk); 3681 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); 3682 int ifindex; 3683 3684 if (peer) 3685 return -EOPNOTSUPP; 3686 3687 ifindex = READ_ONCE(po->ifindex); 3688 sll->sll_family = AF_PACKET; 3689 sll->sll_ifindex = ifindex; 3690 sll->sll_protocol = READ_ONCE(po->num); 3691 sll->sll_pkttype = 0; 3692 rcu_read_lock(); 3693 dev = dev_get_by_index_rcu(sock_net(sk), ifindex); 3694 if (dev) { 3695 sll->sll_hatype = dev->type; 3696 sll->sll_halen = dev->addr_len; 3697 3698 /* Let __fortify_memcpy_chk() know the actual buffer size. */ 3699 memcpy(((struct sockaddr_storage *)sll)->__data + 3700 offsetof(struct sockaddr_ll, sll_addr) - 3701 offsetofend(struct sockaddr_ll, sll_family), 3702 dev->dev_addr, dev->addr_len); 3703 } else { 3704 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ 3705 sll->sll_halen = 0; 3706 } 3707 rcu_read_unlock(); 3708 3709 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; 3710 } 3711 3712 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, 3713 int what) 3714 { 3715 switch (i->type) { 3716 case PACKET_MR_MULTICAST: 3717 if (i->alen != dev->addr_len) 3718 return -EINVAL; 3719 if (what > 0) 3720 return dev_mc_add(dev, i->addr); 3721 else 3722 return dev_mc_del(dev, i->addr); 3723 break; 3724 case PACKET_MR_PROMISC: 3725 return dev_set_promiscuity(dev, what); 3726 case PACKET_MR_ALLMULTI: 3727 return dev_set_allmulti(dev, what); 3728 case PACKET_MR_UNICAST: 3729 if (i->alen != dev->addr_len) 3730 return -EINVAL; 3731 if (what > 0) 3732 return dev_uc_add(dev, i->addr); 3733 else 3734 return dev_uc_del(dev, i->addr); 3735 break; 3736 default: 3737 break; 3738 } 3739 return 0; 3740 } 3741 3742 static void packet_dev_mclist_delete(struct net_device *dev, 3743 struct packet_mclist **mlp, 3744 struct list_head *list) 3745 { 3746 struct packet_mclist *ml; 3747 3748 while ((ml = *mlp) != NULL) { 3749 if (ml->ifindex == dev->ifindex) { 3750 list_add(&ml->remove_list, list); 3751 *mlp = ml->next; 3752 } else 3753 mlp = &ml->next; 3754 } 3755 } 3756 3757 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) 3758 { 3759 struct packet_sock *po = pkt_sk(sk); 3760 struct packet_mclist *ml, *i; 3761 struct net_device *dev; 3762 int err; 3763 3764 rtnl_lock(); 3765 3766 err = -ENODEV; 3767 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); 3768 if (!dev) 3769 goto done; 3770 3771 err = -EINVAL; 3772 if (mreq->mr_alen > dev->addr_len) 3773 goto done; 3774 3775 err = -ENOBUFS; 3776 i = kmalloc_obj(*i); 3777 if (i == NULL) 3778 goto done; 3779 3780 err = 0; 3781 for (ml = po->mclist; ml; ml = ml->next) { 3782 if (ml->ifindex == mreq->mr_ifindex && 3783 ml->type == mreq->mr_type && 3784 ml->alen == mreq->mr_alen && 3785 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3786 ml->count++; 3787 /* Free the new element ... */ 3788 kfree(i); 3789 goto done; 3790 } 3791 } 3792 3793 i->type = mreq->mr_type; 3794 i->ifindex = mreq->mr_ifindex; 3795 i->alen = mreq->mr_alen; 3796 memcpy(i->addr, mreq->mr_address, i->alen); 3797 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); 3798 i->count = 1; 3799 INIT_LIST_HEAD(&i->remove_list); 3800 i->next = po->mclist; 3801 po->mclist = i; 3802 err = packet_dev_mc(dev, i, 1); 3803 if (err) { 3804 po->mclist = i->next; 3805 kfree(i); 3806 } 3807 3808 done: 3809 rtnl_unlock(); 3810 return err; 3811 } 3812 3813 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) 3814 { 3815 struct packet_mclist *ml, **mlp; 3816 3817 rtnl_lock(); 3818 3819 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { 3820 if (ml->ifindex == mreq->mr_ifindex && 3821 ml->type == mreq->mr_type && 3822 ml->alen == mreq->mr_alen && 3823 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { 3824 if (--ml->count == 0) { 3825 struct net_device *dev; 3826 *mlp = ml->next; 3827 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3828 if (dev) 3829 packet_dev_mc(dev, ml, -1); 3830 kfree(ml); 3831 } 3832 break; 3833 } 3834 } 3835 rtnl_unlock(); 3836 return 0; 3837 } 3838 3839 static void packet_flush_mclist(struct sock *sk) 3840 { 3841 struct packet_sock *po = pkt_sk(sk); 3842 struct packet_mclist *ml; 3843 3844 if (!po->mclist) 3845 return; 3846 3847 rtnl_lock(); 3848 while ((ml = po->mclist) != NULL) { 3849 struct net_device *dev; 3850 3851 po->mclist = ml->next; 3852 dev = __dev_get_by_index(sock_net(sk), ml->ifindex); 3853 if (dev != NULL) 3854 packet_dev_mc(dev, ml, -1); 3855 kfree(ml); 3856 } 3857 rtnl_unlock(); 3858 } 3859 3860 static int 3861 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval, 3862 unsigned int optlen) 3863 { 3864 struct sock *sk = sock->sk; 3865 struct packet_sock *po = pkt_sk(sk); 3866 int ret; 3867 3868 if (level != SOL_PACKET) 3869 return -ENOPROTOOPT; 3870 3871 switch (optname) { 3872 case PACKET_ADD_MEMBERSHIP: 3873 case PACKET_DROP_MEMBERSHIP: 3874 { 3875 struct packet_mreq_max mreq; 3876 int len = optlen; 3877 memset(&mreq, 0, sizeof(mreq)); 3878 if (len < sizeof(struct packet_mreq)) 3879 return -EINVAL; 3880 if (len > sizeof(mreq)) 3881 len = sizeof(mreq); 3882 if (copy_from_sockptr(&mreq, optval, len)) 3883 return -EFAULT; 3884 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) 3885 return -EINVAL; 3886 if (optname == PACKET_ADD_MEMBERSHIP) 3887 ret = packet_mc_add(sk, &mreq); 3888 else 3889 ret = packet_mc_drop(sk, &mreq); 3890 return ret; 3891 } 3892 3893 case PACKET_RX_RING: 3894 case PACKET_TX_RING: 3895 { 3896 union tpacket_req_u req_u; 3897 3898 ret = -EINVAL; 3899 lock_sock(sk); 3900 switch (po->tp_version) { 3901 case TPACKET_V1: 3902 case TPACKET_V2: 3903 if (optlen < sizeof(req_u.req)) 3904 break; 3905 ret = copy_from_sockptr(&req_u.req, optval, 3906 sizeof(req_u.req)) ? 3907 -EINVAL : 0; 3908 break; 3909 case TPACKET_V3: 3910 default: 3911 if (optlen < sizeof(req_u.req3)) 3912 break; 3913 ret = copy_from_sockptr(&req_u.req3, optval, 3914 sizeof(req_u.req3)) ? 3915 -EINVAL : 0; 3916 break; 3917 } 3918 if (!ret) 3919 ret = packet_set_ring(sk, &req_u, 0, 3920 optname == PACKET_TX_RING); 3921 release_sock(sk); 3922 return ret; 3923 } 3924 case PACKET_COPY_THRESH: 3925 { 3926 int val; 3927 3928 if (optlen != sizeof(val)) 3929 return -EINVAL; 3930 if (copy_from_sockptr(&val, optval, sizeof(val))) 3931 return -EFAULT; 3932 3933 WRITE_ONCE(pkt_sk(sk)->copy_thresh, val); 3934 return 0; 3935 } 3936 case PACKET_VERSION: 3937 { 3938 int val; 3939 3940 if (optlen != sizeof(val)) 3941 return -EINVAL; 3942 if (copy_from_sockptr(&val, optval, sizeof(val))) 3943 return -EFAULT; 3944 switch (val) { 3945 case TPACKET_V1: 3946 case TPACKET_V2: 3947 case TPACKET_V3: 3948 break; 3949 default: 3950 return -EINVAL; 3951 } 3952 lock_sock(sk); 3953 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3954 ret = -EBUSY; 3955 } else { 3956 po->tp_version = val; 3957 ret = 0; 3958 } 3959 release_sock(sk); 3960 return ret; 3961 } 3962 case PACKET_RESERVE: 3963 { 3964 unsigned int val; 3965 3966 if (optlen != sizeof(val)) 3967 return -EINVAL; 3968 if (copy_from_sockptr(&val, optval, sizeof(val))) 3969 return -EFAULT; 3970 if (val > INT_MAX) 3971 return -EINVAL; 3972 lock_sock(sk); 3973 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3974 ret = -EBUSY; 3975 } else { 3976 po->tp_reserve = val; 3977 ret = 0; 3978 } 3979 release_sock(sk); 3980 return ret; 3981 } 3982 case PACKET_LOSS: 3983 { 3984 unsigned int val; 3985 3986 if (optlen != sizeof(val)) 3987 return -EINVAL; 3988 if (copy_from_sockptr(&val, optval, sizeof(val))) 3989 return -EFAULT; 3990 3991 lock_sock(sk); 3992 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 3993 ret = -EBUSY; 3994 } else { 3995 packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val); 3996 ret = 0; 3997 } 3998 release_sock(sk); 3999 return ret; 4000 } 4001 case PACKET_AUXDATA: 4002 { 4003 int val; 4004 4005 if (optlen < sizeof(val)) 4006 return -EINVAL; 4007 if (copy_from_sockptr(&val, optval, sizeof(val))) 4008 return -EFAULT; 4009 4010 packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val); 4011 return 0; 4012 } 4013 case PACKET_ORIGDEV: 4014 { 4015 int val; 4016 4017 if (optlen < sizeof(val)) 4018 return -EINVAL; 4019 if (copy_from_sockptr(&val, optval, sizeof(val))) 4020 return -EFAULT; 4021 4022 packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val); 4023 return 0; 4024 } 4025 case PACKET_VNET_HDR: 4026 case PACKET_VNET_HDR_SZ: 4027 { 4028 int val, hdr_len; 4029 4030 if (sock->type != SOCK_RAW) 4031 return -EINVAL; 4032 if (optlen < sizeof(val)) 4033 return -EINVAL; 4034 if (copy_from_sockptr(&val, optval, sizeof(val))) 4035 return -EFAULT; 4036 4037 if (optname == PACKET_VNET_HDR_SZ) { 4038 if (val && val != sizeof(struct virtio_net_hdr) && 4039 val != sizeof(struct virtio_net_hdr_mrg_rxbuf)) 4040 return -EINVAL; 4041 hdr_len = val; 4042 } else { 4043 hdr_len = val ? sizeof(struct virtio_net_hdr) : 0; 4044 } 4045 lock_sock(sk); 4046 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { 4047 ret = -EBUSY; 4048 } else { 4049 WRITE_ONCE(po->vnet_hdr_sz, hdr_len); 4050 ret = 0; 4051 } 4052 release_sock(sk); 4053 return ret; 4054 } 4055 case PACKET_TIMESTAMP: 4056 { 4057 int val; 4058 4059 if (optlen != sizeof(val)) 4060 return -EINVAL; 4061 if (copy_from_sockptr(&val, optval, sizeof(val))) 4062 return -EFAULT; 4063 4064 WRITE_ONCE(po->tp_tstamp, val); 4065 return 0; 4066 } 4067 case PACKET_FANOUT: 4068 { 4069 struct fanout_args args = { 0 }; 4070 4071 if (optlen != sizeof(int) && optlen != sizeof(args)) 4072 return -EINVAL; 4073 if (copy_from_sockptr(&args, optval, optlen)) 4074 return -EFAULT; 4075 4076 return fanout_add(sk, &args); 4077 } 4078 case PACKET_FANOUT_DATA: 4079 { 4080 /* Paired with the WRITE_ONCE() in fanout_add() */ 4081 if (!READ_ONCE(po->fanout)) 4082 return -EINVAL; 4083 4084 return fanout_set_data(po, optval, optlen); 4085 } 4086 case PACKET_IGNORE_OUTGOING: 4087 { 4088 int val; 4089 4090 if (optlen != sizeof(val)) 4091 return -EINVAL; 4092 if (copy_from_sockptr(&val, optval, sizeof(val))) 4093 return -EFAULT; 4094 if (val < 0 || val > 1) 4095 return -EINVAL; 4096 4097 WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val); 4098 return 0; 4099 } 4100 case PACKET_TX_HAS_OFF: 4101 { 4102 unsigned int val; 4103 4104 if (optlen != sizeof(val)) 4105 return -EINVAL; 4106 if (copy_from_sockptr(&val, optval, sizeof(val))) 4107 return -EFAULT; 4108 4109 lock_sock(sk); 4110 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec) 4111 packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val); 4112 4113 release_sock(sk); 4114 return 0; 4115 } 4116 case PACKET_QDISC_BYPASS: 4117 { 4118 int val; 4119 4120 if (optlen != sizeof(val)) 4121 return -EINVAL; 4122 if (copy_from_sockptr(&val, optval, sizeof(val))) 4123 return -EFAULT; 4124 4125 packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val); 4126 return 0; 4127 } 4128 default: 4129 return -ENOPROTOOPT; 4130 } 4131 } 4132 4133 static int packet_getsockopt(struct socket *sock, int level, int optname, 4134 sockopt_t *opt) 4135 { 4136 int len; 4137 int val, lv = sizeof(val); 4138 struct sock *sk = sock->sk; 4139 struct packet_sock *po = pkt_sk(sk); 4140 void *data = &val; 4141 union tpacket_stats_u st; 4142 struct tpacket_rollover_stats rstats; 4143 int drops; 4144 4145 if (level != SOL_PACKET) 4146 return -ENOPROTOOPT; 4147 4148 len = opt->optlen; 4149 4150 if (len < 0) 4151 return -EINVAL; 4152 4153 switch (optname) { 4154 case PACKET_STATISTICS: 4155 spin_lock_bh(&sk->sk_receive_queue.lock); 4156 memcpy(&st, &po->stats, sizeof(st)); 4157 memset(&po->stats, 0, sizeof(po->stats)); 4158 spin_unlock_bh(&sk->sk_receive_queue.lock); 4159 drops = atomic_xchg(&po->tp_drops, 0); 4160 4161 if (po->tp_version == TPACKET_V3) { 4162 lv = sizeof(struct tpacket_stats_v3); 4163 st.stats3.tp_drops = drops; 4164 st.stats3.tp_packets += drops; 4165 data = &st.stats3; 4166 } else { 4167 lv = sizeof(struct tpacket_stats); 4168 st.stats1.tp_drops = drops; 4169 st.stats1.tp_packets += drops; 4170 data = &st.stats1; 4171 } 4172 4173 break; 4174 case PACKET_AUXDATA: 4175 val = packet_sock_flag(po, PACKET_SOCK_AUXDATA); 4176 break; 4177 case PACKET_ORIGDEV: 4178 val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV); 4179 break; 4180 case PACKET_VNET_HDR: 4181 val = !!READ_ONCE(po->vnet_hdr_sz); 4182 break; 4183 case PACKET_VNET_HDR_SZ: 4184 val = READ_ONCE(po->vnet_hdr_sz); 4185 break; 4186 case PACKET_COPY_THRESH: 4187 val = READ_ONCE(pkt_sk(sk)->copy_thresh); 4188 break; 4189 case PACKET_VERSION: 4190 val = po->tp_version; 4191 break; 4192 case PACKET_HDRLEN: 4193 if (len > sizeof(int)) 4194 len = sizeof(int); 4195 if (len < sizeof(int)) 4196 return -EINVAL; 4197 if (copy_from_iter(&val, len, &opt->iter_in) != len) 4198 return -EFAULT; 4199 switch (val) { 4200 case TPACKET_V1: 4201 val = sizeof(struct tpacket_hdr); 4202 break; 4203 case TPACKET_V2: 4204 val = sizeof(struct tpacket2_hdr); 4205 break; 4206 case TPACKET_V3: 4207 val = sizeof(struct tpacket3_hdr); 4208 break; 4209 default: 4210 return -EINVAL; 4211 } 4212 break; 4213 case PACKET_RESERVE: 4214 val = po->tp_reserve; 4215 break; 4216 case PACKET_LOSS: 4217 val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS); 4218 break; 4219 case PACKET_TIMESTAMP: 4220 val = READ_ONCE(po->tp_tstamp); 4221 break; 4222 case PACKET_FANOUT: 4223 val = (po->fanout ? 4224 ((u32)po->fanout->id | 4225 ((u32)po->fanout->type << 16) | 4226 ((u32)po->fanout->flags << 24)) : 4227 0); 4228 break; 4229 case PACKET_IGNORE_OUTGOING: 4230 val = READ_ONCE(po->prot_hook.ignore_outgoing); 4231 break; 4232 case PACKET_ROLLOVER_STATS: 4233 if (!po->rollover) 4234 return -EINVAL; 4235 rstats.tp_all = atomic_long_read(&po->rollover->num); 4236 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); 4237 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); 4238 data = &rstats; 4239 lv = sizeof(rstats); 4240 break; 4241 case PACKET_TX_HAS_OFF: 4242 val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF); 4243 break; 4244 case PACKET_QDISC_BYPASS: 4245 val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS); 4246 break; 4247 default: 4248 return -ENOPROTOOPT; 4249 } 4250 4251 if (len > lv) 4252 len = lv; 4253 opt->optlen = len; 4254 if (copy_to_iter(data, len, &opt->iter_out) != len) 4255 return -EFAULT; 4256 return 0; 4257 } 4258 4259 static int packet_notifier(struct notifier_block *this, 4260 unsigned long msg, void *ptr) 4261 { 4262 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4263 struct net *net = dev_net(dev); 4264 struct packet_mclist *ml, *tmp; 4265 LIST_HEAD(mclist); 4266 struct sock *sk; 4267 4268 rcu_read_lock(); 4269 sk_for_each_rcu(sk, &net->packet.sklist) { 4270 struct packet_sock *po = pkt_sk(sk); 4271 4272 switch (msg) { 4273 case NETDEV_UNREGISTER: 4274 if (po->mclist) 4275 packet_dev_mclist_delete(dev, &po->mclist, 4276 &mclist); 4277 fallthrough; 4278 4279 case NETDEV_DOWN: 4280 if (dev->ifindex == po->ifindex) { 4281 spin_lock(&po->bind_lock); 4282 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) { 4283 __unregister_prot_hook(sk, false); 4284 sk->sk_err = ENETDOWN; 4285 if (!sock_flag(sk, SOCK_DEAD)) 4286 sk_error_report(sk); 4287 } 4288 if (msg == NETDEV_UNREGISTER) { 4289 packet_cached_dev_reset(po); 4290 WRITE_ONCE(po->ifindex, -1); 4291 netdev_put(po->prot_hook.dev, 4292 &po->prot_hook.dev_tracker); 4293 po->prot_hook.dev = NULL; 4294 } 4295 spin_unlock(&po->bind_lock); 4296 } 4297 break; 4298 case NETDEV_UP: 4299 if (dev->ifindex == po->ifindex) { 4300 spin_lock(&po->bind_lock); 4301 if (po->num) 4302 register_prot_hook(sk); 4303 spin_unlock(&po->bind_lock); 4304 } 4305 break; 4306 } 4307 } 4308 rcu_read_unlock(); 4309 4310 /* packet_dev_mc might grab instance locks so can't run under rcu */ 4311 list_for_each_entry_safe(ml, tmp, &mclist, remove_list) { 4312 packet_dev_mc(dev, ml, -1); 4313 kfree(ml); 4314 } 4315 4316 return NOTIFY_DONE; 4317 } 4318 4319 4320 static int packet_ioctl(struct socket *sock, unsigned int cmd, 4321 unsigned long arg) 4322 { 4323 struct sock *sk = sock->sk; 4324 4325 switch (cmd) { 4326 case SIOCOUTQ: 4327 { 4328 int amount = sk_wmem_alloc_get(sk); 4329 4330 return put_user(amount, (int __user *)arg); 4331 } 4332 case SIOCINQ: 4333 { 4334 struct sk_buff *skb; 4335 int amount = 0; 4336 4337 spin_lock_bh(&sk->sk_receive_queue.lock); 4338 skb = skb_peek(&sk->sk_receive_queue); 4339 if (skb) 4340 amount = skb->len; 4341 spin_unlock_bh(&sk->sk_receive_queue.lock); 4342 return put_user(amount, (int __user *)arg); 4343 } 4344 #ifdef CONFIG_INET 4345 case SIOCADDRT: 4346 case SIOCDELRT: 4347 case SIOCDARP: 4348 case SIOCGARP: 4349 case SIOCSARP: 4350 case SIOCGIFADDR: 4351 case SIOCSIFADDR: 4352 case SIOCGIFBRDADDR: 4353 case SIOCSIFBRDADDR: 4354 case SIOCGIFNETMASK: 4355 case SIOCSIFNETMASK: 4356 case SIOCGIFDSTADDR: 4357 case SIOCSIFDSTADDR: 4358 case SIOCSIFFLAGS: 4359 return inet_dgram_ops.ioctl(sock, cmd, arg); 4360 #endif 4361 4362 default: 4363 return -ENOIOCTLCMD; 4364 } 4365 return 0; 4366 } 4367 4368 static __poll_t packet_poll(struct file *file, struct socket *sock, 4369 poll_table *wait) 4370 { 4371 struct sock *sk = sock->sk; 4372 struct packet_sock *po = pkt_sk(sk); 4373 __poll_t mask = datagram_poll(file, sock, wait); 4374 4375 spin_lock_bh(&sk->sk_receive_queue.lock); 4376 if (po->rx_ring.pg_vec) { 4377 if (!packet_previous_rx_frame(po, &po->rx_ring, 4378 TP_STATUS_KERNEL)) 4379 mask |= EPOLLIN | EPOLLRDNORM; 4380 } 4381 __packet_rcv_try_clear_pressure(po); 4382 spin_unlock_bh(&sk->sk_receive_queue.lock); 4383 spin_lock_bh(&sk->sk_write_queue.lock); 4384 if (po->tx_ring.pg_vec) { 4385 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) 4386 mask |= EPOLLOUT | EPOLLWRNORM; 4387 } 4388 spin_unlock_bh(&sk->sk_write_queue.lock); 4389 return mask; 4390 } 4391 4392 4393 /* Dirty? Well, I still did not learn better way to account 4394 * for user mmaps. 4395 */ 4396 4397 static void packet_mm_open(struct vm_area_struct *vma) 4398 { 4399 struct file *file = vma->vm_file; 4400 struct socket *sock = file->private_data; 4401 struct sock *sk = sock->sk; 4402 4403 if (sk) 4404 atomic_long_inc(&pkt_sk(sk)->mapped); 4405 } 4406 4407 static void packet_mm_close(struct vm_area_struct *vma) 4408 { 4409 struct file *file = vma->vm_file; 4410 struct socket *sock = file->private_data; 4411 struct sock *sk = sock->sk; 4412 4413 if (sk) 4414 atomic_long_dec(&pkt_sk(sk)->mapped); 4415 } 4416 4417 static const struct vm_operations_struct packet_mmap_ops = { 4418 .open = packet_mm_open, 4419 .close = packet_mm_close, 4420 }; 4421 4422 struct packet_pg_vec { 4423 struct packet_pg_vec_free *deferred; 4424 unsigned int order; 4425 unsigned int len; 4426 struct pgv pg_vec[] __counted_by(len); 4427 }; 4428 4429 struct packet_pg_vec_free { 4430 struct delayed_work work; 4431 struct sock *sk; 4432 struct packet_pg_vec *vec; 4433 }; 4434 4435 static void free_pg_vec(struct pgv *pg_vec, unsigned int order, 4436 unsigned int len) 4437 { 4438 struct packet_pg_vec *vec; 4439 int i; 4440 4441 vec = container_of_const(pg_vec, struct packet_pg_vec, pg_vec[0]); 4442 for (i = 0; i < len; i++) { 4443 if (likely(pg_vec[i].buffer)) { 4444 if (is_vmalloc_addr(pg_vec[i].buffer)) 4445 vfree(pg_vec[i].buffer); 4446 else 4447 free_pages((unsigned long)pg_vec[i].buffer, 4448 order); 4449 pg_vec[i].buffer = NULL; 4450 } 4451 } 4452 kfree(vec->deferred); 4453 kfree(vec); 4454 } 4455 4456 static void packet_free_pg_vec_work(struct work_struct *work) 4457 { 4458 struct packet_pg_vec_free *deferred; 4459 struct packet_pg_vec *vec; 4460 struct sock *sk; 4461 4462 deferred = container_of_const(to_delayed_work(work), 4463 struct packet_pg_vec_free, work); 4464 vec = deferred->vec; 4465 sk = deferred->sk; 4466 if (sk_wmem_alloc_get(sk)) { 4467 queue_delayed_work(system_long_wq, &deferred->work, 1); 4468 return; 4469 } 4470 4471 free_pg_vec(vec->pg_vec, vec->order, vec->len); 4472 sock_put(sk); 4473 } 4474 4475 static void packet_free_tx_ring(struct sock *sk, struct pgv *pg_vec, 4476 unsigned int order, unsigned int len) 4477 { 4478 struct packet_pg_vec_free *deferred; 4479 struct packet_pg_vec *vec; 4480 4481 vec = container_of_const(pg_vec, struct packet_pg_vec, pg_vec[0]); 4482 deferred = vec->deferred; 4483 if (!deferred || !sk_wmem_alloc_get(sk)) { 4484 free_pg_vec(pg_vec, order, len); 4485 return; 4486 } 4487 4488 /* A detached ring's pending count can miss late skb destructors. */ 4489 deferred->sk = sk; 4490 sock_hold(sk); 4491 queue_delayed_work(system_long_wq, &deferred->work, 0); 4492 } 4493 4494 static char *alloc_one_pg_vec_page(unsigned long order) 4495 { 4496 char *buffer; 4497 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | 4498 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; 4499 4500 buffer = (char *) __get_free_pages(gfp_flags, order); 4501 if (buffer) 4502 return buffer; 4503 4504 /* __get_free_pages failed, fall back to vmalloc */ 4505 buffer = vzalloc(array_size((1 << order), PAGE_SIZE)); 4506 if (buffer) 4507 return buffer; 4508 4509 /* vmalloc failed, lets dig into swap here */ 4510 gfp_flags &= ~__GFP_NORETRY; 4511 buffer = (char *) __get_free_pages(gfp_flags, order); 4512 if (buffer) 4513 return buffer; 4514 4515 /* complete and utter failure */ 4516 return NULL; 4517 } 4518 4519 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order, bool tx_ring) 4520 { 4521 unsigned int block_nr = req->tp_block_nr; 4522 struct packet_pg_vec *vec; 4523 struct pgv *pg_vec; 4524 int i; 4525 4526 vec = kzalloc_flex(*vec, pg_vec, block_nr, GFP_KERNEL | __GFP_NOWARN); 4527 if (unlikely(!vec)) 4528 return NULL; 4529 vec->order = order; 4530 vec->len = block_nr; 4531 pg_vec = vec->pg_vec; 4532 4533 if (tx_ring) { 4534 vec->deferred = kzalloc_obj(*vec->deferred, 4535 GFP_KERNEL | __GFP_NOWARN); 4536 if (!vec->deferred) 4537 goto out_free_pgvec; 4538 vec->deferred->vec = vec; 4539 INIT_DELAYED_WORK(&vec->deferred->work, 4540 packet_free_pg_vec_work); 4541 } 4542 4543 for (i = 0; i < block_nr; i++) { 4544 pg_vec[i].buffer = alloc_one_pg_vec_page(order); 4545 if (unlikely(!pg_vec[i].buffer)) 4546 goto out_free_pgvec; 4547 } 4548 4549 out: 4550 return pg_vec; 4551 4552 out_free_pgvec: 4553 free_pg_vec(pg_vec, order, block_nr); 4554 pg_vec = NULL; 4555 goto out; 4556 } 4557 4558 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, 4559 int closing, int tx_ring) 4560 { 4561 struct pgv *pg_vec = NULL; 4562 struct packet_sock *po = pkt_sk(sk); 4563 unsigned long *rx_owner_map = NULL; 4564 int was_running, order = 0; 4565 struct packet_ring_buffer *rb; 4566 struct sk_buff_head *rb_queue; 4567 __be16 num; 4568 int err; 4569 /* Added to avoid minimal code churn */ 4570 struct tpacket_req *req = &req_u->req; 4571 4572 rb = tx_ring ? &po->tx_ring : &po->rx_ring; 4573 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; 4574 4575 err = -EBUSY; 4576 if (!closing) { 4577 if (atomic_long_read(&po->mapped)) 4578 goto out; 4579 if (packet_read_pending(rb)) 4580 goto out; 4581 } 4582 4583 if (req->tp_block_nr) { 4584 unsigned int min_frame_size; 4585 4586 /* Sanity tests and some calculations */ 4587 err = -EBUSY; 4588 if (unlikely(rb->pg_vec)) 4589 goto out; 4590 4591 switch (po->tp_version) { 4592 case TPACKET_V1: 4593 po->tp_hdrlen = TPACKET_HDRLEN; 4594 break; 4595 case TPACKET_V2: 4596 po->tp_hdrlen = TPACKET2_HDRLEN; 4597 break; 4598 case TPACKET_V3: 4599 po->tp_hdrlen = TPACKET3_HDRLEN; 4600 break; 4601 } 4602 4603 err = -EINVAL; 4604 if (unlikely((int)req->tp_block_size <= 0)) 4605 goto out; 4606 if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) 4607 goto out; 4608 min_frame_size = po->tp_hdrlen + po->tp_reserve; 4609 if (po->tp_version >= TPACKET_V3 && 4610 req->tp_block_size < 4611 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) 4612 goto out; 4613 if (unlikely(req->tp_frame_size < min_frame_size)) 4614 goto out; 4615 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) 4616 goto out; 4617 4618 rb->frames_per_block = req->tp_block_size / req->tp_frame_size; 4619 if (unlikely(rb->frames_per_block == 0)) 4620 goto out; 4621 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) 4622 goto out; 4623 if (unlikely((rb->frames_per_block * req->tp_block_nr) != 4624 req->tp_frame_nr)) 4625 goto out; 4626 4627 err = -ENOMEM; 4628 order = get_order(req->tp_block_size); 4629 pg_vec = alloc_pg_vec(req, order, tx_ring); 4630 if (unlikely(!pg_vec)) 4631 goto out; 4632 switch (po->tp_version) { 4633 case TPACKET_V3: 4634 /* Block transmit is not supported yet */ 4635 if (!tx_ring) { 4636 init_prb_bdqc(po, rb, pg_vec, req_u); 4637 } else { 4638 struct tpacket_req3 *req3 = &req_u->req3; 4639 4640 if (req3->tp_retire_blk_tov || 4641 req3->tp_sizeof_priv || 4642 req3->tp_feature_req_word) { 4643 err = -EINVAL; 4644 goto out_free_pg_vec; 4645 } 4646 } 4647 break; 4648 default: 4649 if (!tx_ring) { 4650 rx_owner_map = bitmap_alloc(req->tp_frame_nr, 4651 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO); 4652 if (!rx_owner_map) 4653 goto out_free_pg_vec; 4654 } 4655 break; 4656 } 4657 } 4658 /* Done */ 4659 else { 4660 err = -EINVAL; 4661 if (unlikely(req->tp_frame_nr)) 4662 goto out; 4663 } 4664 4665 4666 /* Detach socket from network */ 4667 spin_lock(&po->bind_lock); 4668 was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING); 4669 num = po->num; 4670 WRITE_ONCE(po->num, 0); 4671 if (was_running) 4672 __unregister_prot_hook(sk, false); 4673 4674 spin_unlock(&po->bind_lock); 4675 4676 synchronize_net(); 4677 4678 err = -EBUSY; 4679 mutex_lock(&po->pg_vec_lock); 4680 if (closing || atomic_long_read(&po->mapped) == 0) { 4681 if (tx_ring && !closing && packet_read_pending(rb)) 4682 goto out_unlock; 4683 4684 err = 0; 4685 spin_lock_bh(&rb_queue->lock); 4686 swap(rb->pg_vec, pg_vec); 4687 if (po->tp_version <= TPACKET_V2) 4688 swap(rb->rx_owner_map, rx_owner_map); 4689 rb->frame_max = (req->tp_frame_nr - 1); 4690 rb->head = 0; 4691 rb->frame_size = req->tp_frame_size; 4692 po->prot_hook.func = (po->rx_ring.pg_vec) ? 4693 tpacket_rcv : packet_rcv; 4694 spin_unlock_bh(&rb_queue->lock); 4695 4696 swap(rb->pg_vec_order, order); 4697 swap(rb->pg_vec_len, req->tp_block_nr); 4698 4699 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; 4700 skb_queue_purge(rb_queue); 4701 if (atomic_long_read(&po->mapped)) 4702 pr_err("packet_mmap: vma is busy: %ld\n", 4703 atomic_long_read(&po->mapped)); 4704 } 4705 out_unlock: 4706 mutex_unlock(&po->pg_vec_lock); 4707 4708 spin_lock(&po->bind_lock); 4709 WRITE_ONCE(po->num, num); 4710 /* 4711 * NETDEV_UNREGISTER may have invalidated the binding while bind_lock 4712 * was dropped above. Do not re-add a fanout hook to a dead device. 4713 */ 4714 if (was_running && READ_ONCE(po->ifindex) != -1) 4715 register_prot_hook(sk); 4716 4717 spin_unlock(&po->bind_lock); 4718 if (pg_vec && (po->tp_version > TPACKET_V2)) { 4719 /* Because we don't support block-based V3 on tx-ring */ 4720 if (!tx_ring) 4721 prb_shutdown_retire_blk_timer(po, rb_queue); 4722 } 4723 4724 out_free_pg_vec: 4725 if (pg_vec) { 4726 bitmap_free(rx_owner_map); 4727 if (tx_ring && closing) 4728 packet_free_tx_ring(sk, pg_vec, order, req->tp_block_nr); 4729 else 4730 free_pg_vec(pg_vec, order, req->tp_block_nr); 4731 } 4732 out: 4733 return err; 4734 } 4735 4736 static int packet_mmap(struct file *file, struct socket *sock, 4737 struct vm_area_struct *vma) 4738 { 4739 struct sock *sk = sock->sk; 4740 struct packet_sock *po = pkt_sk(sk); 4741 unsigned long size, expected_size; 4742 struct packet_ring_buffer *rb; 4743 unsigned long start; 4744 int err = -EINVAL; 4745 int i; 4746 4747 if (vma->vm_pgoff) 4748 return -EINVAL; 4749 4750 mutex_lock(&po->pg_vec_lock); 4751 4752 expected_size = 0; 4753 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4754 if (rb->pg_vec) { 4755 expected_size += rb->pg_vec_len 4756 * rb->pg_vec_pages 4757 * PAGE_SIZE; 4758 } 4759 } 4760 4761 if (expected_size == 0) 4762 goto out; 4763 4764 size = vma->vm_end - vma->vm_start; 4765 if (size != expected_size) 4766 goto out; 4767 4768 start = vma->vm_start; 4769 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { 4770 if (rb->pg_vec == NULL) 4771 continue; 4772 4773 for (i = 0; i < rb->pg_vec_len; i++) { 4774 struct page *page; 4775 void *kaddr = rb->pg_vec[i].buffer; 4776 int pg_num; 4777 4778 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { 4779 page = pgv_to_page(kaddr); 4780 err = vm_insert_page(vma, start, page); 4781 if (unlikely(err)) 4782 goto out; 4783 start += PAGE_SIZE; 4784 kaddr += PAGE_SIZE; 4785 } 4786 } 4787 } 4788 4789 atomic_long_inc(&po->mapped); 4790 vma->vm_ops = &packet_mmap_ops; 4791 err = 0; 4792 4793 out: 4794 mutex_unlock(&po->pg_vec_lock); 4795 return err; 4796 } 4797 4798 static const struct proto_ops packet_ops_spkt = { 4799 .family = PF_PACKET, 4800 .owner = THIS_MODULE, 4801 .release = packet_release, 4802 .bind = packet_bind_spkt, 4803 .connect = sock_no_connect, 4804 .socketpair = sock_no_socketpair, 4805 .accept = sock_no_accept, 4806 .getname = packet_getname_spkt, 4807 .poll = datagram_poll, 4808 .ioctl = packet_ioctl, 4809 .gettstamp = sock_gettstamp, 4810 .listen = sock_no_listen, 4811 .shutdown = sock_no_shutdown, 4812 .sendmsg = packet_sendmsg_spkt, 4813 .recvmsg = packet_recvmsg, 4814 .mmap = sock_no_mmap, 4815 }; 4816 4817 static const struct proto_ops packet_ops = { 4818 .family = PF_PACKET, 4819 .owner = THIS_MODULE, 4820 .release = packet_release, 4821 .bind = packet_bind, 4822 .connect = sock_no_connect, 4823 .socketpair = sock_no_socketpair, 4824 .accept = sock_no_accept, 4825 .getname = packet_getname, 4826 .poll = packet_poll, 4827 .ioctl = packet_ioctl, 4828 .gettstamp = sock_gettstamp, 4829 .listen = sock_no_listen, 4830 .shutdown = sock_no_shutdown, 4831 .setsockopt = packet_setsockopt, 4832 .getsockopt_iter = packet_getsockopt, 4833 .sendmsg = packet_sendmsg, 4834 .recvmsg = packet_recvmsg, 4835 .mmap = packet_mmap, 4836 }; 4837 4838 static const struct net_proto_family packet_family_ops = { 4839 .family = PF_PACKET, 4840 .create = packet_create, 4841 .owner = THIS_MODULE, 4842 }; 4843 4844 static struct notifier_block packet_netdev_notifier = { 4845 .notifier_call = packet_notifier, 4846 }; 4847 4848 #ifdef CONFIG_PROC_FS 4849 4850 static void *packet_seq_start(struct seq_file *seq, loff_t *pos) 4851 __acquires(RCU) 4852 { 4853 struct net *net = seq_file_net(seq); 4854 4855 rcu_read_lock(); 4856 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); 4857 } 4858 4859 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4860 { 4861 struct net *net = seq_file_net(seq); 4862 return seq_hlist_next_rcu(v, &net->packet.sklist, pos); 4863 } 4864 4865 static void packet_seq_stop(struct seq_file *seq, void *v) 4866 __releases(RCU) 4867 { 4868 rcu_read_unlock(); 4869 } 4870 4871 static int packet_seq_show(struct seq_file *seq, void *v) 4872 { 4873 if (v == SEQ_START_TOKEN) 4874 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); 4875 else { 4876 struct sock *s = sk_entry(v); 4877 const struct packet_sock *po = pkt_sk(s); 4878 4879 seq_printf(seq, 4880 "0 %-6d %-4d %04x %-5d %1d %-6u %-6u %-6llu\n", 4881 refcount_read(&s->sk_refcnt), 4882 s->sk_type, 4883 ntohs(READ_ONCE(po->num)), 4884 READ_ONCE(po->ifindex), 4885 packet_sock_flag(po, PACKET_SOCK_RUNNING), 4886 atomic_read(&s->sk_rmem_alloc), 4887 from_kuid_munged(seq_user_ns(seq), sk_uid(s)), 4888 sock_i_ino(s)); 4889 } 4890 4891 return 0; 4892 } 4893 4894 static const struct seq_operations packet_seq_ops = { 4895 .start = packet_seq_start, 4896 .next = packet_seq_next, 4897 .stop = packet_seq_stop, 4898 .show = packet_seq_show, 4899 }; 4900 #endif 4901 4902 static int __net_init packet_net_init(struct net *net) 4903 { 4904 mutex_init(&net->packet.sklist_lock); 4905 INIT_HLIST_HEAD(&net->packet.sklist); 4906 4907 #ifdef CONFIG_PROC_FS 4908 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, 4909 sizeof(struct seq_net_private))) 4910 return -ENOMEM; 4911 #endif /* CONFIG_PROC_FS */ 4912 4913 return 0; 4914 } 4915 4916 static void __net_exit packet_net_exit(struct net *net) 4917 { 4918 remove_proc_entry("packet", net->proc_net); 4919 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); 4920 } 4921 4922 static struct pernet_operations packet_net_ops = { 4923 .init = packet_net_init, 4924 .exit = packet_net_exit, 4925 }; 4926 4927 4928 static void __exit packet_exit(void) 4929 { 4930 sock_unregister(PF_PACKET); 4931 proto_unregister(&packet_proto); 4932 unregister_netdevice_notifier(&packet_netdev_notifier); 4933 unregister_pernet_subsys(&packet_net_ops); 4934 } 4935 4936 static int __init packet_init(void) 4937 { 4938 int rc; 4939 4940 rc = register_pernet_subsys(&packet_net_ops); 4941 if (rc) 4942 goto out; 4943 rc = register_netdevice_notifier(&packet_netdev_notifier); 4944 if (rc) 4945 goto out_pernet; 4946 rc = proto_register(&packet_proto, 0); 4947 if (rc) 4948 goto out_notifier; 4949 rc = sock_register(&packet_family_ops); 4950 if (rc) 4951 goto out_proto; 4952 4953 return 0; 4954 4955 out_proto: 4956 proto_unregister(&packet_proto); 4957 out_notifier: 4958 unregister_netdevice_notifier(&packet_netdev_notifier); 4959 out_pernet: 4960 unregister_pernet_subsys(&packet_net_ops); 4961 out: 4962 return rc; 4963 } 4964 4965 module_init(packet_init); 4966 module_exit(packet_exit); 4967 MODULE_DESCRIPTION("Packet socket support (AF_PACKET)"); 4968 MODULE_LICENSE("GPL"); 4969 MODULE_ALIAS_NETPROTO(PF_PACKET); 4970