Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * PACKET - implements raw packet sockets.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 *
13 * Fixes:
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
42 * and packet_mreq.
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * layer.
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
47 */
48
49#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
50
51#include <linux/ethtool.h>
52#include <linux/filter.h>
53#include <linux/types.h>
54#include <linux/mm.h>
55#include <linux/capability.h>
56#include <linux/fcntl.h>
57#include <linux/socket.h>
58#include <linux/in.h>
59#include <linux/inet.h>
60#include <linux/netdevice.h>
61#include <linux/if_packet.h>
62#include <linux/wireless.h>
63#include <linux/kernel.h>
64#include <linux/kmod.h>
65#include <linux/slab.h>
66#include <linux/vmalloc.h>
67#include <net/net_namespace.h>
68#include <net/ip.h>
69#include <net/protocol.h>
70#include <linux/skbuff.h>
71#include <net/sock.h>
72#include <linux/errno.h>
73#include <linux/timer.h>
74#include <linux/uaccess.h>
75#include <asm/ioctls.h>
76#include <asm/page.h>
77#include <asm/cacheflush.h>
78#include <asm/io.h>
79#include <linux/proc_fs.h>
80#include <linux/seq_file.h>
81#include <linux/poll.h>
82#include <linux/module.h>
83#include <linux/init.h>
84#include <linux/mutex.h>
85#include <linux/if_vlan.h>
86#include <linux/virtio_net.h>
87#include <linux/errqueue.h>
88#include <linux/net_tstamp.h>
89#include <linux/percpu.h>
90#ifdef CONFIG_INET
91#include <net/inet_common.h>
92#endif
93#include <linux/bpf.h>
94#include <net/compat.h>
95#include <linux/netfilter_netdev.h>
96
97#include "internal.h"
98
99/*
100 Assumptions:
101 - If the device has no dev->header_ops->create, there is no LL header
102 visible above the device. In this case, its hard_header_len should be 0.
103 The device may prepend its own header internally. In this case, its
104 needed_headroom should be set to the space needed for it to add its
105 internal header.
106 For example, a WiFi driver pretending to be an Ethernet driver should
107 set its hard_header_len to be the Ethernet header length, and set its
108 needed_headroom to be (the real WiFi header length - the fake Ethernet
109 header length).
110 - packet socket receives packets with pulled ll header,
111 so that SOCK_RAW should push it back.
112
113On receive:
114-----------
115
116Incoming, dev_has_header(dev) == true
117 mac_header -> ll header
118 data -> data
119
120Outgoing, dev_has_header(dev) == true
121 mac_header -> ll header
122 data -> ll header
123
124Incoming, dev_has_header(dev) == false
125 mac_header -> data
126 However drivers often make it point to the ll header.
127 This is incorrect because the ll header should be invisible to us.
128 data -> data
129
130Outgoing, dev_has_header(dev) == false
131 mac_header -> data. ll header is invisible to us.
132 data -> data
133
134Resume
135 If dev_has_header(dev) == false we are unable to restore the ll header,
136 because it is invisible to us.
137
138
139On transmit:
140------------
141
142dev_has_header(dev) == true
143 mac_header -> ll header
144 data -> ll header
145
146dev_has_header(dev) == false (ll header is invisible to us)
147 mac_header -> data
148 data -> data
149
150 We should set network_header on output to the correct position,
151 packet classifier depends on it.
152 */
153
154/* Private packet socket structures. */
155
156/* identical to struct packet_mreq except it has
157 * a longer address field.
158 */
159struct packet_mreq_max {
160 int mr_ifindex;
161 unsigned short mr_type;
162 unsigned short mr_alen;
163 unsigned char mr_address[MAX_ADDR_LEN];
164};
165
166union tpacket_uhdr {
167 struct tpacket_hdr *h1;
168 struct tpacket2_hdr *h2;
169 struct tpacket3_hdr *h3;
170 void *raw;
171};
172
173static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
174 int closing, int tx_ring);
175
176#define V3_ALIGNMENT (8)
177
178#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
179
180#define BLK_PLUS_PRIV(sz_of_priv) \
181 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
182
183#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
184#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
185#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
186#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
187#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
188#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
189
190struct packet_sock;
191static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 struct packet_type *pt, struct net_device *orig_dev);
193
194static void *packet_previous_frame(struct packet_sock *po,
195 struct packet_ring_buffer *rb,
196 int status);
197static void packet_increment_head(struct packet_ring_buffer *buff);
198static int prb_curr_blk_in_use(struct tpacket_block_desc *);
199static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203static int prb_queue_frozen(struct tpacket_kbdq_core *);
204static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206static void prb_retire_rx_blk_timer_expired(struct timer_list *);
207static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
209static void prb_clear_rxhash(struct tpacket_kbdq_core *,
210 struct tpacket3_hdr *);
211static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
212 struct tpacket3_hdr *);
213static void packet_flush_mclist(struct sock *sk);
214static u16 packet_pick_tx_queue(struct sk_buff *skb);
215
216struct packet_skb_cb {
217 union {
218 struct sockaddr_pkt pkt;
219 union {
220 /* Trick: alias skb original length with
221 * ll.sll_family and ll.protocol in order
222 * to save room.
223 */
224 unsigned int origlen;
225 struct sockaddr_ll ll;
226 };
227 } sa;
228};
229
230#define vio_le() virtio_legacy_is_little_endian()
231
232#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
233
234#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235#define GET_PBLOCK_DESC(x, bid) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239#define GET_NEXT_PRB_BLK_NUM(x) \
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
242
243static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244static void __fanout_link(struct sock *sk, struct packet_sock *po);
245
246#ifdef CONFIG_NETFILTER_EGRESS
247static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb)
248{
249 struct sk_buff *next, *head = NULL, *tail;
250 int rc;
251
252 rcu_read_lock();
253 for (; skb != NULL; skb = next) {
254 next = skb->next;
255 skb_mark_not_on_list(skb);
256
257 if (!nf_hook_egress(skb, &rc, skb->dev))
258 continue;
259
260 if (!head)
261 head = skb;
262 else
263 tail->next = skb;
264
265 tail = skb;
266 }
267 rcu_read_unlock();
268
269 return head;
270}
271#endif
272
273static int packet_xmit(const struct packet_sock *po, struct sk_buff *skb)
274{
275 if (!packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS))
276 return dev_queue_xmit(skb);
277
278#ifdef CONFIG_NETFILTER_EGRESS
279 if (nf_hook_egress_active()) {
280 skb = nf_hook_direct_egress(skb);
281 if (!skb)
282 return NET_XMIT_DROP;
283 }
284#endif
285 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
286}
287
288static struct net_device *packet_cached_dev_get(struct packet_sock *po)
289{
290 struct net_device *dev;
291
292 rcu_read_lock();
293 dev = rcu_dereference(po->cached_dev);
294 dev_hold(dev);
295 rcu_read_unlock();
296
297 return dev;
298}
299
300static void packet_cached_dev_assign(struct packet_sock *po,
301 struct net_device *dev)
302{
303 rcu_assign_pointer(po->cached_dev, dev);
304}
305
306static void packet_cached_dev_reset(struct packet_sock *po)
307{
308 RCU_INIT_POINTER(po->cached_dev, NULL);
309}
310
311static u16 packet_pick_tx_queue(struct sk_buff *skb)
312{
313 struct net_device *dev = skb->dev;
314 const struct net_device_ops *ops = dev->netdev_ops;
315 int cpu = raw_smp_processor_id();
316 u16 queue_index;
317
318#ifdef CONFIG_XPS
319 skb->sender_cpu = cpu + 1;
320#endif
321 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
322 if (ops->ndo_select_queue) {
323 queue_index = ops->ndo_select_queue(dev, skb, NULL);
324 queue_index = netdev_cap_txqueue(dev, queue_index);
325 } else {
326 queue_index = netdev_pick_tx(dev, skb, NULL);
327 }
328
329 return queue_index;
330}
331
332/* __register_prot_hook must be invoked through register_prot_hook
333 * or from a context in which asynchronous accesses to the packet
334 * socket is not possible (packet_create()).
335 */
336static void __register_prot_hook(struct sock *sk)
337{
338 struct packet_sock *po = pkt_sk(sk);
339
340 if (!packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
341 if (po->fanout)
342 __fanout_link(sk, po);
343 else
344 dev_add_pack(&po->prot_hook);
345
346 sock_hold(sk);
347 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 1);
348 }
349}
350
351static void register_prot_hook(struct sock *sk)
352{
353 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
354 __register_prot_hook(sk);
355}
356
357/* If the sync parameter is true, we will temporarily drop
358 * the po->bind_lock and do a synchronize_net to make sure no
359 * asynchronous packet processing paths still refer to the elements
360 * of po->prot_hook. If the sync parameter is false, it is the
361 * callers responsibility to take care of this.
362 */
363static void __unregister_prot_hook(struct sock *sk, bool sync)
364{
365 struct packet_sock *po = pkt_sk(sk);
366
367 lockdep_assert_held_once(&po->bind_lock);
368
369 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 0);
370
371 if (po->fanout)
372 __fanout_unlink(sk, po);
373 else
374 __dev_remove_pack(&po->prot_hook);
375
376 __sock_put(sk);
377
378 if (sync) {
379 spin_unlock(&po->bind_lock);
380 synchronize_net();
381 spin_lock(&po->bind_lock);
382 }
383}
384
385static void unregister_prot_hook(struct sock *sk, bool sync)
386{
387 struct packet_sock *po = pkt_sk(sk);
388
389 if (packet_sock_flag(po, PACKET_SOCK_RUNNING))
390 __unregister_prot_hook(sk, sync);
391}
392
393static inline struct page * __pure pgv_to_page(void *addr)
394{
395 if (is_vmalloc_addr(addr))
396 return vmalloc_to_page(addr);
397 return virt_to_page(addr);
398}
399
400static void __packet_set_status(struct packet_sock *po, void *frame, int status)
401{
402 union tpacket_uhdr h;
403
404 /* WRITE_ONCE() are paired with READ_ONCE() in __packet_get_status */
405
406 h.raw = frame;
407 switch (po->tp_version) {
408 case TPACKET_V1:
409 WRITE_ONCE(h.h1->tp_status, status);
410 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
411 break;
412 case TPACKET_V2:
413 WRITE_ONCE(h.h2->tp_status, status);
414 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
415 break;
416 case TPACKET_V3:
417 WRITE_ONCE(h.h3->tp_status, status);
418 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
419 break;
420 default:
421 WARN(1, "TPACKET version not supported.\n");
422 BUG();
423 }
424
425 smp_wmb();
426}
427
428static int __packet_get_status(const struct packet_sock *po, void *frame)
429{
430 union tpacket_uhdr h;
431
432 smp_rmb();
433
434 /* READ_ONCE() are paired with WRITE_ONCE() in __packet_set_status */
435
436 h.raw = frame;
437 switch (po->tp_version) {
438 case TPACKET_V1:
439 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
440 return READ_ONCE(h.h1->tp_status);
441 case TPACKET_V2:
442 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
443 return READ_ONCE(h.h2->tp_status);
444 case TPACKET_V3:
445 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
446 return READ_ONCE(h.h3->tp_status);
447 default:
448 WARN(1, "TPACKET version not supported.\n");
449 BUG();
450 return 0;
451 }
452}
453
454static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
455 unsigned int flags)
456{
457 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
458
459 if (shhwtstamps &&
460 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
461 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
462 return TP_STATUS_TS_RAW_HARDWARE;
463
464 if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
465 ktime_to_timespec64_cond(skb_tstamp(skb), ts))
466 return TP_STATUS_TS_SOFTWARE;
467
468 return 0;
469}
470
471static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
472 struct sk_buff *skb)
473{
474 union tpacket_uhdr h;
475 struct timespec64 ts;
476 __u32 ts_status;
477
478 if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp))))
479 return 0;
480
481 h.raw = frame;
482 /*
483 * versions 1 through 3 overflow the timestamps in y2106, since they
484 * all store the seconds in a 32-bit unsigned integer.
485 * If we create a version 4, that should have a 64-bit timestamp,
486 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
487 * nanoseconds.
488 */
489 switch (po->tp_version) {
490 case TPACKET_V1:
491 h.h1->tp_sec = ts.tv_sec;
492 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
493 break;
494 case TPACKET_V2:
495 h.h2->tp_sec = ts.tv_sec;
496 h.h2->tp_nsec = ts.tv_nsec;
497 break;
498 case TPACKET_V3:
499 h.h3->tp_sec = ts.tv_sec;
500 h.h3->tp_nsec = ts.tv_nsec;
501 break;
502 default:
503 WARN(1, "TPACKET version not supported.\n");
504 BUG();
505 }
506
507 /* one flush is safe, as both fields always lie on the same cacheline */
508 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
509 smp_wmb();
510
511 return ts_status;
512}
513
514static void *packet_lookup_frame(const struct packet_sock *po,
515 const struct packet_ring_buffer *rb,
516 unsigned int position,
517 int status)
518{
519 unsigned int pg_vec_pos, frame_offset;
520 union tpacket_uhdr h;
521
522 pg_vec_pos = position / rb->frames_per_block;
523 frame_offset = position % rb->frames_per_block;
524
525 h.raw = rb->pg_vec[pg_vec_pos].buffer +
526 (frame_offset * rb->frame_size);
527
528 if (status != __packet_get_status(po, h.raw))
529 return NULL;
530
531 return h.raw;
532}
533
534static void *packet_current_frame(struct packet_sock *po,
535 struct packet_ring_buffer *rb,
536 int status)
537{
538 return packet_lookup_frame(po, rb, rb->head, status);
539}
540
541static u16 vlan_get_tci(const struct sk_buff *skb, struct net_device *dev)
542{
543 struct vlan_hdr vhdr, *vh;
544 unsigned int header_len;
545
546 if (!dev)
547 return 0;
548
549 /* In the SOCK_DGRAM scenario, skb data starts at the network
550 * protocol, which is after the VLAN headers. The outer VLAN
551 * header is at the hard_header_len offset in non-variable
552 * length link layer headers. If it's a VLAN device, the
553 * min_header_len should be used to exclude the VLAN header
554 * size.
555 */
556 if (dev->min_header_len == dev->hard_header_len)
557 header_len = dev->hard_header_len;
558 else if (is_vlan_dev(dev))
559 header_len = dev->min_header_len;
560 else
561 return 0;
562
563 vh = skb_header_pointer(skb, skb_mac_offset(skb) + header_len,
564 sizeof(vhdr), &vhdr);
565 if (unlikely(!vh))
566 return 0;
567
568 return ntohs(vh->h_vlan_TCI);
569}
570
571static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb)
572{
573 __be16 proto = skb->protocol;
574
575 if (unlikely(eth_type_vlan(proto)))
576 proto = __vlan_get_protocol_offset(skb, proto,
577 skb_mac_offset(skb), NULL);
578
579 return proto;
580}
581
582static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
583{
584 del_timer_sync(&pkc->retire_blk_timer);
585}
586
587static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
588 struct sk_buff_head *rb_queue)
589{
590 struct tpacket_kbdq_core *pkc;
591
592 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
593
594 spin_lock_bh(&rb_queue->lock);
595 pkc->delete_blk_timer = 1;
596 spin_unlock_bh(&rb_queue->lock);
597
598 prb_del_retire_blk_timer(pkc);
599}
600
601static void prb_setup_retire_blk_timer(struct packet_sock *po)
602{
603 struct tpacket_kbdq_core *pkc;
604
605 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
606 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
607 0);
608 pkc->retire_blk_timer.expires = jiffies;
609}
610
611static int prb_calc_retire_blk_tmo(struct packet_sock *po,
612 int blk_size_in_bytes)
613{
614 struct net_device *dev;
615 unsigned int mbits, div;
616 struct ethtool_link_ksettings ecmd;
617 int err;
618
619 rtnl_lock();
620 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
621 if (unlikely(!dev)) {
622 rtnl_unlock();
623 return DEFAULT_PRB_RETIRE_TOV;
624 }
625 err = __ethtool_get_link_ksettings(dev, &ecmd);
626 rtnl_unlock();
627 if (err)
628 return DEFAULT_PRB_RETIRE_TOV;
629
630 /* If the link speed is so slow you don't really
631 * need to worry about perf anyways
632 */
633 if (ecmd.base.speed < SPEED_1000 ||
634 ecmd.base.speed == SPEED_UNKNOWN)
635 return DEFAULT_PRB_RETIRE_TOV;
636
637 div = ecmd.base.speed / 1000;
638 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
639
640 if (div)
641 mbits /= div;
642
643 if (div)
644 return mbits + 1;
645 return mbits;
646}
647
648static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
649 union tpacket_req_u *req_u)
650{
651 p1->feature_req_word = req_u->req3.tp_feature_req_word;
652}
653
654static void init_prb_bdqc(struct packet_sock *po,
655 struct packet_ring_buffer *rb,
656 struct pgv *pg_vec,
657 union tpacket_req_u *req_u)
658{
659 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
660 struct tpacket_block_desc *pbd;
661
662 memset(p1, 0x0, sizeof(*p1));
663
664 p1->knxt_seq_num = 1;
665 p1->pkbdq = pg_vec;
666 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
667 p1->pkblk_start = pg_vec[0].buffer;
668 p1->kblk_size = req_u->req3.tp_block_size;
669 p1->knum_blocks = req_u->req3.tp_block_nr;
670 p1->hdrlen = po->tp_hdrlen;
671 p1->version = po->tp_version;
672 p1->last_kactive_blk_num = 0;
673 po->stats.stats3.tp_freeze_q_cnt = 0;
674 if (req_u->req3.tp_retire_blk_tov)
675 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
676 else
677 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
678 req_u->req3.tp_block_size);
679 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
680 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
681 rwlock_init(&p1->blk_fill_in_prog_lock);
682
683 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
684 prb_init_ft_ops(p1, req_u);
685 prb_setup_retire_blk_timer(po);
686 prb_open_block(p1, pbd);
687}
688
689/* Do NOT update the last_blk_num first.
690 * Assumes sk_buff_head lock is held.
691 */
692static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
693{
694 mod_timer(&pkc->retire_blk_timer,
695 jiffies + pkc->tov_in_jiffies);
696 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
697}
698
699/*
700 * Timer logic:
701 * 1) We refresh the timer only when we open a block.
702 * By doing this we don't waste cycles refreshing the timer
703 * on packet-by-packet basis.
704 *
705 * With a 1MB block-size, on a 1Gbps line, it will take
706 * i) ~8 ms to fill a block + ii) memcpy etc.
707 * In this cut we are not accounting for the memcpy time.
708 *
709 * So, if the user sets the 'tmo' to 10ms then the timer
710 * will never fire while the block is still getting filled
711 * (which is what we want). However, the user could choose
712 * to close a block early and that's fine.
713 *
714 * But when the timer does fire, we check whether or not to refresh it.
715 * Since the tmo granularity is in msecs, it is not too expensive
716 * to refresh the timer, lets say every '8' msecs.
717 * Either the user can set the 'tmo' or we can derive it based on
718 * a) line-speed and b) block-size.
719 * prb_calc_retire_blk_tmo() calculates the tmo.
720 *
721 */
722static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
723{
724 struct packet_sock *po =
725 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
726 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
727 unsigned int frozen;
728 struct tpacket_block_desc *pbd;
729
730 spin_lock(&po->sk.sk_receive_queue.lock);
731
732 frozen = prb_queue_frozen(pkc);
733 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
734
735 if (unlikely(pkc->delete_blk_timer))
736 goto out;
737
738 /* We only need to plug the race when the block is partially filled.
739 * tpacket_rcv:
740 * lock(); increment BLOCK_NUM_PKTS; unlock()
741 * copy_bits() is in progress ...
742 * timer fires on other cpu:
743 * we can't retire the current block because copy_bits
744 * is in progress.
745 *
746 */
747 if (BLOCK_NUM_PKTS(pbd)) {
748 /* Waiting for skb_copy_bits to finish... */
749 write_lock(&pkc->blk_fill_in_prog_lock);
750 write_unlock(&pkc->blk_fill_in_prog_lock);
751 }
752
753 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
754 if (!frozen) {
755 if (!BLOCK_NUM_PKTS(pbd)) {
756 /* An empty block. Just refresh the timer. */
757 goto refresh_timer;
758 }
759 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
760 if (!prb_dispatch_next_block(pkc, po))
761 goto refresh_timer;
762 else
763 goto out;
764 } else {
765 /* Case 1. Queue was frozen because user-space was
766 * lagging behind.
767 */
768 if (prb_curr_blk_in_use(pbd)) {
769 /*
770 * Ok, user-space is still behind.
771 * So just refresh the timer.
772 */
773 goto refresh_timer;
774 } else {
775 /* Case 2. queue was frozen,user-space caught up,
776 * now the link went idle && the timer fired.
777 * We don't have a block to close.So we open this
778 * block and restart the timer.
779 * opening a block thaws the queue,restarts timer
780 * Thawing/timer-refresh is a side effect.
781 */
782 prb_open_block(pkc, pbd);
783 goto out;
784 }
785 }
786 }
787
788refresh_timer:
789 _prb_refresh_rx_retire_blk_timer(pkc);
790
791out:
792 spin_unlock(&po->sk.sk_receive_queue.lock);
793}
794
795static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
796 struct tpacket_block_desc *pbd1, __u32 status)
797{
798 /* Flush everything minus the block header */
799
800#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
801 u8 *start, *end;
802
803 start = (u8 *)pbd1;
804
805 /* Skip the block header(we know header WILL fit in 4K) */
806 start += PAGE_SIZE;
807
808 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
809 for (; start < end; start += PAGE_SIZE)
810 flush_dcache_page(pgv_to_page(start));
811
812 smp_wmb();
813#endif
814
815 /* Now update the block status. */
816
817 BLOCK_STATUS(pbd1) = status;
818
819 /* Flush the block header */
820
821#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
822 start = (u8 *)pbd1;
823 flush_dcache_page(pgv_to_page(start));
824
825 smp_wmb();
826#endif
827}
828
829/*
830 * Side effect:
831 *
832 * 1) flush the block
833 * 2) Increment active_blk_num
834 *
835 * Note:We DONT refresh the timer on purpose.
836 * Because almost always the next block will be opened.
837 */
838static void prb_close_block(struct tpacket_kbdq_core *pkc1,
839 struct tpacket_block_desc *pbd1,
840 struct packet_sock *po, unsigned int stat)
841{
842 __u32 status = TP_STATUS_USER | stat;
843
844 struct tpacket3_hdr *last_pkt;
845 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
846 struct sock *sk = &po->sk;
847
848 if (atomic_read(&po->tp_drops))
849 status |= TP_STATUS_LOSING;
850
851 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
852 last_pkt->tp_next_offset = 0;
853
854 /* Get the ts of the last pkt */
855 if (BLOCK_NUM_PKTS(pbd1)) {
856 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
857 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
858 } else {
859 /* Ok, we tmo'd - so get the current time.
860 *
861 * It shouldn't really happen as we don't close empty
862 * blocks. See prb_retire_rx_blk_timer_expired().
863 */
864 struct timespec64 ts;
865 ktime_get_real_ts64(&ts);
866 h1->ts_last_pkt.ts_sec = ts.tv_sec;
867 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
868 }
869
870 smp_wmb();
871
872 /* Flush the block */
873 prb_flush_block(pkc1, pbd1, status);
874
875 sk->sk_data_ready(sk);
876
877 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
878}
879
880static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
881{
882 pkc->reset_pending_on_curr_blk = 0;
883}
884
885/*
886 * Side effect of opening a block:
887 *
888 * 1) prb_queue is thawed.
889 * 2) retire_blk_timer is refreshed.
890 *
891 */
892static void prb_open_block(struct tpacket_kbdq_core *pkc1,
893 struct tpacket_block_desc *pbd1)
894{
895 struct timespec64 ts;
896 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
897
898 smp_rmb();
899
900 /* We could have just memset this but we will lose the
901 * flexibility of making the priv area sticky
902 */
903
904 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
905 BLOCK_NUM_PKTS(pbd1) = 0;
906 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
907
908 ktime_get_real_ts64(&ts);
909
910 h1->ts_first_pkt.ts_sec = ts.tv_sec;
911 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
912
913 pkc1->pkblk_start = (char *)pbd1;
914 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
915
916 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
917 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
918
919 pbd1->version = pkc1->version;
920 pkc1->prev = pkc1->nxt_offset;
921 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
922
923 prb_thaw_queue(pkc1);
924 _prb_refresh_rx_retire_blk_timer(pkc1);
925
926 smp_wmb();
927}
928
929/*
930 * Queue freeze logic:
931 * 1) Assume tp_block_nr = 8 blocks.
932 * 2) At time 't0', user opens Rx ring.
933 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
934 * 4) user-space is either sleeping or processing block '0'.
935 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
936 * it will close block-7,loop around and try to fill block '0'.
937 * call-flow:
938 * __packet_lookup_frame_in_block
939 * prb_retire_current_block()
940 * prb_dispatch_next_block()
941 * |->(BLOCK_STATUS == USER) evaluates to true
942 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
943 * 6) Now there are two cases:
944 * 6.1) Link goes idle right after the queue is frozen.
945 * But remember, the last open_block() refreshed the timer.
946 * When this timer expires,it will refresh itself so that we can
947 * re-open block-0 in near future.
948 * 6.2) Link is busy and keeps on receiving packets. This is a simple
949 * case and __packet_lookup_frame_in_block will check if block-0
950 * is free and can now be re-used.
951 */
952static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
953 struct packet_sock *po)
954{
955 pkc->reset_pending_on_curr_blk = 1;
956 po->stats.stats3.tp_freeze_q_cnt++;
957}
958
959#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
960
961/*
962 * If the next block is free then we will dispatch it
963 * and return a good offset.
964 * Else, we will freeze the queue.
965 * So, caller must check the return value.
966 */
967static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
968 struct packet_sock *po)
969{
970 struct tpacket_block_desc *pbd;
971
972 smp_rmb();
973
974 /* 1. Get current block num */
975 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
976
977 /* 2. If this block is currently in_use then freeze the queue */
978 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
979 prb_freeze_queue(pkc, po);
980 return NULL;
981 }
982
983 /*
984 * 3.
985 * open this block and return the offset where the first packet
986 * needs to get stored.
987 */
988 prb_open_block(pkc, pbd);
989 return (void *)pkc->nxt_offset;
990}
991
992static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
993 struct packet_sock *po, unsigned int status)
994{
995 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
996
997 /* retire/close the current block */
998 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
999 /*
1000 * Plug the case where copy_bits() is in progress on
1001 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
1002 * have space to copy the pkt in the current block and
1003 * called prb_retire_current_block()
1004 *
1005 * We don't need to worry about the TMO case because
1006 * the timer-handler already handled this case.
1007 */
1008 if (!(status & TP_STATUS_BLK_TMO)) {
1009 /* Waiting for skb_copy_bits to finish... */
1010 write_lock(&pkc->blk_fill_in_prog_lock);
1011 write_unlock(&pkc->blk_fill_in_prog_lock);
1012 }
1013 prb_close_block(pkc, pbd, po, status);
1014 return;
1015 }
1016}
1017
1018static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
1019{
1020 return TP_STATUS_USER & BLOCK_STATUS(pbd);
1021}
1022
1023static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
1024{
1025 return pkc->reset_pending_on_curr_blk;
1026}
1027
1028static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
1029 __releases(&pkc->blk_fill_in_prog_lock)
1030{
1031 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1032
1033 read_unlock(&pkc->blk_fill_in_prog_lock);
1034}
1035
1036static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
1037 struct tpacket3_hdr *ppd)
1038{
1039 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
1040}
1041
1042static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1043 struct tpacket3_hdr *ppd)
1044{
1045 ppd->hv1.tp_rxhash = 0;
1046}
1047
1048static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1049 struct tpacket3_hdr *ppd)
1050{
1051 struct packet_sock *po = container_of(pkc, struct packet_sock, rx_ring.prb_bdqc);
1052
1053 if (skb_vlan_tag_present(pkc->skb)) {
1054 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1055 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1056 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1057 } else if (unlikely(po->sk.sk_type == SOCK_DGRAM && eth_type_vlan(pkc->skb->protocol))) {
1058 ppd->hv1.tp_vlan_tci = vlan_get_tci(pkc->skb, pkc->skb->dev);
1059 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->protocol);
1060 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1061 } else {
1062 ppd->hv1.tp_vlan_tci = 0;
1063 ppd->hv1.tp_vlan_tpid = 0;
1064 ppd->tp_status = TP_STATUS_AVAILABLE;
1065 }
1066}
1067
1068static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1069 struct tpacket3_hdr *ppd)
1070{
1071 ppd->hv1.tp_padding = 0;
1072 prb_fill_vlan_info(pkc, ppd);
1073
1074 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1075 prb_fill_rxhash(pkc, ppd);
1076 else
1077 prb_clear_rxhash(pkc, ppd);
1078}
1079
1080static void prb_fill_curr_block(char *curr,
1081 struct tpacket_kbdq_core *pkc,
1082 struct tpacket_block_desc *pbd,
1083 unsigned int len)
1084 __acquires(&pkc->blk_fill_in_prog_lock)
1085{
1086 struct tpacket3_hdr *ppd;
1087
1088 ppd = (struct tpacket3_hdr *)curr;
1089 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1090 pkc->prev = curr;
1091 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1092 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1093 BLOCK_NUM_PKTS(pbd) += 1;
1094 read_lock(&pkc->blk_fill_in_prog_lock);
1095 prb_run_all_ft_ops(pkc, ppd);
1096}
1097
1098/* Assumes caller has the sk->rx_queue.lock */
1099static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1100 struct sk_buff *skb,
1101 unsigned int len
1102 )
1103{
1104 struct tpacket_kbdq_core *pkc;
1105 struct tpacket_block_desc *pbd;
1106 char *curr, *end;
1107
1108 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1109 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1110
1111 /* Queue is frozen when user space is lagging behind */
1112 if (prb_queue_frozen(pkc)) {
1113 /*
1114 * Check if that last block which caused the queue to freeze,
1115 * is still in_use by user-space.
1116 */
1117 if (prb_curr_blk_in_use(pbd)) {
1118 /* Can't record this packet */
1119 return NULL;
1120 } else {
1121 /*
1122 * Ok, the block was released by user-space.
1123 * Now let's open that block.
1124 * opening a block also thaws the queue.
1125 * Thawing is a side effect.
1126 */
1127 prb_open_block(pkc, pbd);
1128 }
1129 }
1130
1131 smp_mb();
1132 curr = pkc->nxt_offset;
1133 pkc->skb = skb;
1134 end = (char *)pbd + pkc->kblk_size;
1135
1136 /* first try the current block */
1137 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1138 prb_fill_curr_block(curr, pkc, pbd, len);
1139 return (void *)curr;
1140 }
1141
1142 /* Ok, close the current block */
1143 prb_retire_current_block(pkc, po, 0);
1144
1145 /* Now, try to dispatch the next block */
1146 curr = (char *)prb_dispatch_next_block(pkc, po);
1147 if (curr) {
1148 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1149 prb_fill_curr_block(curr, pkc, pbd, len);
1150 return (void *)curr;
1151 }
1152
1153 /*
1154 * No free blocks are available.user_space hasn't caught up yet.
1155 * Queue was just frozen and now this packet will get dropped.
1156 */
1157 return NULL;
1158}
1159
1160static void *packet_current_rx_frame(struct packet_sock *po,
1161 struct sk_buff *skb,
1162 int status, unsigned int len)
1163{
1164 char *curr = NULL;
1165 switch (po->tp_version) {
1166 case TPACKET_V1:
1167 case TPACKET_V2:
1168 curr = packet_lookup_frame(po, &po->rx_ring,
1169 po->rx_ring.head, status);
1170 return curr;
1171 case TPACKET_V3:
1172 return __packet_lookup_frame_in_block(po, skb, len);
1173 default:
1174 WARN(1, "TPACKET version not supported\n");
1175 BUG();
1176 return NULL;
1177 }
1178}
1179
1180static void *prb_lookup_block(const struct packet_sock *po,
1181 const struct packet_ring_buffer *rb,
1182 unsigned int idx,
1183 int status)
1184{
1185 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1186 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1187
1188 if (status != BLOCK_STATUS(pbd))
1189 return NULL;
1190 return pbd;
1191}
1192
1193static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1194{
1195 unsigned int prev;
1196 if (rb->prb_bdqc.kactive_blk_num)
1197 prev = rb->prb_bdqc.kactive_blk_num-1;
1198 else
1199 prev = rb->prb_bdqc.knum_blocks-1;
1200 return prev;
1201}
1202
1203/* Assumes caller has held the rx_queue.lock */
1204static void *__prb_previous_block(struct packet_sock *po,
1205 struct packet_ring_buffer *rb,
1206 int status)
1207{
1208 unsigned int previous = prb_previous_blk_num(rb);
1209 return prb_lookup_block(po, rb, previous, status);
1210}
1211
1212static void *packet_previous_rx_frame(struct packet_sock *po,
1213 struct packet_ring_buffer *rb,
1214 int status)
1215{
1216 if (po->tp_version <= TPACKET_V2)
1217 return packet_previous_frame(po, rb, status);
1218
1219 return __prb_previous_block(po, rb, status);
1220}
1221
1222static void packet_increment_rx_head(struct packet_sock *po,
1223 struct packet_ring_buffer *rb)
1224{
1225 switch (po->tp_version) {
1226 case TPACKET_V1:
1227 case TPACKET_V2:
1228 return packet_increment_head(rb);
1229 case TPACKET_V3:
1230 default:
1231 WARN(1, "TPACKET version not supported.\n");
1232 BUG();
1233 return;
1234 }
1235}
1236
1237static void *packet_previous_frame(struct packet_sock *po,
1238 struct packet_ring_buffer *rb,
1239 int status)
1240{
1241 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1242 return packet_lookup_frame(po, rb, previous, status);
1243}
1244
1245static void packet_increment_head(struct packet_ring_buffer *buff)
1246{
1247 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1248}
1249
1250static void packet_inc_pending(struct packet_ring_buffer *rb)
1251{
1252 this_cpu_inc(*rb->pending_refcnt);
1253}
1254
1255static void packet_dec_pending(struct packet_ring_buffer *rb)
1256{
1257 this_cpu_dec(*rb->pending_refcnt);
1258}
1259
1260static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1261{
1262 unsigned int refcnt = 0;
1263 int cpu;
1264
1265 /* We don't use pending refcount in rx_ring. */
1266 if (rb->pending_refcnt == NULL)
1267 return 0;
1268
1269 for_each_possible_cpu(cpu)
1270 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1271
1272 return refcnt;
1273}
1274
1275static int packet_alloc_pending(struct packet_sock *po)
1276{
1277 po->rx_ring.pending_refcnt = NULL;
1278
1279 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1280 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1281 return -ENOBUFS;
1282
1283 return 0;
1284}
1285
1286static void packet_free_pending(struct packet_sock *po)
1287{
1288 free_percpu(po->tx_ring.pending_refcnt);
1289}
1290
1291#define ROOM_POW_OFF 2
1292#define ROOM_NONE 0x0
1293#define ROOM_LOW 0x1
1294#define ROOM_NORMAL 0x2
1295
1296static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1297{
1298 int idx, len;
1299
1300 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1301 idx = READ_ONCE(po->rx_ring.head);
1302 if (pow_off)
1303 idx += len >> pow_off;
1304 if (idx >= len)
1305 idx -= len;
1306 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1307}
1308
1309static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1310{
1311 int idx, len;
1312
1313 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1314 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1315 if (pow_off)
1316 idx += len >> pow_off;
1317 if (idx >= len)
1318 idx -= len;
1319 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1320}
1321
1322static int __packet_rcv_has_room(const struct packet_sock *po,
1323 const struct sk_buff *skb)
1324{
1325 const struct sock *sk = &po->sk;
1326 int ret = ROOM_NONE;
1327
1328 if (po->prot_hook.func != tpacket_rcv) {
1329 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1330 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1331 - (skb ? skb->truesize : 0);
1332
1333 if (avail > (rcvbuf >> ROOM_POW_OFF))
1334 return ROOM_NORMAL;
1335 else if (avail > 0)
1336 return ROOM_LOW;
1337 else
1338 return ROOM_NONE;
1339 }
1340
1341 if (po->tp_version == TPACKET_V3) {
1342 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1343 ret = ROOM_NORMAL;
1344 else if (__tpacket_v3_has_room(po, 0))
1345 ret = ROOM_LOW;
1346 } else {
1347 if (__tpacket_has_room(po, ROOM_POW_OFF))
1348 ret = ROOM_NORMAL;
1349 else if (__tpacket_has_room(po, 0))
1350 ret = ROOM_LOW;
1351 }
1352
1353 return ret;
1354}
1355
1356static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1357{
1358 bool pressure;
1359 int ret;
1360
1361 ret = __packet_rcv_has_room(po, skb);
1362 pressure = ret != ROOM_NORMAL;
1363
1364 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure)
1365 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure);
1366
1367 return ret;
1368}
1369
1370static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1371{
1372 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) &&
1373 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1374 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false);
1375}
1376
1377static void packet_sock_destruct(struct sock *sk)
1378{
1379 skb_queue_purge(&sk->sk_error_queue);
1380
1381 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1382 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1383
1384 if (!sock_flag(sk, SOCK_DEAD)) {
1385 pr_err("Attempt to release alive packet socket: %p\n", sk);
1386 return;
1387 }
1388}
1389
1390static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1391{
1392 u32 *history = po->rollover->history;
1393 u32 victim, rxhash;
1394 int i, count = 0;
1395
1396 rxhash = skb_get_hash(skb);
1397 for (i = 0; i < ROLLOVER_HLEN; i++)
1398 if (READ_ONCE(history[i]) == rxhash)
1399 count++;
1400
1401 victim = get_random_u32_below(ROLLOVER_HLEN);
1402
1403 /* Avoid dirtying the cache line if possible */
1404 if (READ_ONCE(history[victim]) != rxhash)
1405 WRITE_ONCE(history[victim], rxhash);
1406
1407 return count > (ROLLOVER_HLEN >> 1);
1408}
1409
1410static unsigned int fanout_demux_hash(struct packet_fanout *f,
1411 struct sk_buff *skb,
1412 unsigned int num)
1413{
1414 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1415}
1416
1417static unsigned int fanout_demux_lb(struct packet_fanout *f,
1418 struct sk_buff *skb,
1419 unsigned int num)
1420{
1421 unsigned int val = atomic_inc_return(&f->rr_cur);
1422
1423 return val % num;
1424}
1425
1426static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1427 struct sk_buff *skb,
1428 unsigned int num)
1429{
1430 return smp_processor_id() % num;
1431}
1432
1433static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1434 struct sk_buff *skb,
1435 unsigned int num)
1436{
1437 return get_random_u32_below(num);
1438}
1439
1440static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1441 struct sk_buff *skb,
1442 unsigned int idx, bool try_self,
1443 unsigned int num)
1444{
1445 struct packet_sock *po, *po_next, *po_skip = NULL;
1446 unsigned int i, j, room = ROOM_NONE;
1447
1448 po = pkt_sk(rcu_dereference(f->arr[idx]));
1449
1450 if (try_self) {
1451 room = packet_rcv_has_room(po, skb);
1452 if (room == ROOM_NORMAL ||
1453 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1454 return idx;
1455 po_skip = po;
1456 }
1457
1458 i = j = min_t(int, po->rollover->sock, num - 1);
1459 do {
1460 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1461 if (po_next != po_skip &&
1462 !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1463 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1464 if (i != j)
1465 po->rollover->sock = i;
1466 atomic_long_inc(&po->rollover->num);
1467 if (room == ROOM_LOW)
1468 atomic_long_inc(&po->rollover->num_huge);
1469 return i;
1470 }
1471
1472 if (++i == num)
1473 i = 0;
1474 } while (i != j);
1475
1476 atomic_long_inc(&po->rollover->num_failed);
1477 return idx;
1478}
1479
1480static unsigned int fanout_demux_qm(struct packet_fanout *f,
1481 struct sk_buff *skb,
1482 unsigned int num)
1483{
1484 return skb_get_queue_mapping(skb) % num;
1485}
1486
1487static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1488 struct sk_buff *skb,
1489 unsigned int num)
1490{
1491 struct bpf_prog *prog;
1492 unsigned int ret = 0;
1493
1494 rcu_read_lock();
1495 prog = rcu_dereference(f->bpf_prog);
1496 if (prog)
1497 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1498 rcu_read_unlock();
1499
1500 return ret;
1501}
1502
1503static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1504{
1505 return f->flags & (flag >> 8);
1506}
1507
1508static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1509 struct packet_type *pt, struct net_device *orig_dev)
1510{
1511 struct packet_fanout *f = pt->af_packet_priv;
1512 unsigned int num = READ_ONCE(f->num_members);
1513 struct net *net = read_pnet(&f->net);
1514 struct packet_sock *po;
1515 unsigned int idx;
1516
1517 if (!net_eq(dev_net(dev), net) || !num) {
1518 kfree_skb(skb);
1519 return 0;
1520 }
1521
1522 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1523 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1524 if (!skb)
1525 return 0;
1526 }
1527 switch (f->type) {
1528 case PACKET_FANOUT_HASH:
1529 default:
1530 idx = fanout_demux_hash(f, skb, num);
1531 break;
1532 case PACKET_FANOUT_LB:
1533 idx = fanout_demux_lb(f, skb, num);
1534 break;
1535 case PACKET_FANOUT_CPU:
1536 idx = fanout_demux_cpu(f, skb, num);
1537 break;
1538 case PACKET_FANOUT_RND:
1539 idx = fanout_demux_rnd(f, skb, num);
1540 break;
1541 case PACKET_FANOUT_QM:
1542 idx = fanout_demux_qm(f, skb, num);
1543 break;
1544 case PACKET_FANOUT_ROLLOVER:
1545 idx = fanout_demux_rollover(f, skb, 0, false, num);
1546 break;
1547 case PACKET_FANOUT_CBPF:
1548 case PACKET_FANOUT_EBPF:
1549 idx = fanout_demux_bpf(f, skb, num);
1550 break;
1551 }
1552
1553 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1554 idx = fanout_demux_rollover(f, skb, idx, true, num);
1555
1556 po = pkt_sk(rcu_dereference(f->arr[idx]));
1557 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1558}
1559
1560DEFINE_MUTEX(fanout_mutex);
1561EXPORT_SYMBOL_GPL(fanout_mutex);
1562static LIST_HEAD(fanout_list);
1563static u16 fanout_next_id;
1564
1565static void __fanout_link(struct sock *sk, struct packet_sock *po)
1566{
1567 struct packet_fanout *f = po->fanout;
1568
1569 spin_lock(&f->lock);
1570 rcu_assign_pointer(f->arr[f->num_members], sk);
1571 smp_wmb();
1572 f->num_members++;
1573 if (f->num_members == 1)
1574 dev_add_pack(&f->prot_hook);
1575 spin_unlock(&f->lock);
1576}
1577
1578static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1579{
1580 struct packet_fanout *f = po->fanout;
1581 int i;
1582
1583 spin_lock(&f->lock);
1584 for (i = 0; i < f->num_members; i++) {
1585 if (rcu_dereference_protected(f->arr[i],
1586 lockdep_is_held(&f->lock)) == sk)
1587 break;
1588 }
1589 BUG_ON(i >= f->num_members);
1590 rcu_assign_pointer(f->arr[i],
1591 rcu_dereference_protected(f->arr[f->num_members - 1],
1592 lockdep_is_held(&f->lock)));
1593 f->num_members--;
1594 if (f->num_members == 0)
1595 __dev_remove_pack(&f->prot_hook);
1596 spin_unlock(&f->lock);
1597}
1598
1599static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1600{
1601 if (sk->sk_family != PF_PACKET)
1602 return false;
1603
1604 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1605}
1606
1607static void fanout_init_data(struct packet_fanout *f)
1608{
1609 switch (f->type) {
1610 case PACKET_FANOUT_LB:
1611 atomic_set(&f->rr_cur, 0);
1612 break;
1613 case PACKET_FANOUT_CBPF:
1614 case PACKET_FANOUT_EBPF:
1615 RCU_INIT_POINTER(f->bpf_prog, NULL);
1616 break;
1617 }
1618}
1619
1620static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1621{
1622 struct bpf_prog *old;
1623
1624 spin_lock(&f->lock);
1625 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1626 rcu_assign_pointer(f->bpf_prog, new);
1627 spin_unlock(&f->lock);
1628
1629 if (old) {
1630 synchronize_net();
1631 bpf_prog_destroy(old);
1632 }
1633}
1634
1635static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1636 unsigned int len)
1637{
1638 struct bpf_prog *new;
1639 struct sock_fprog fprog;
1640 int ret;
1641
1642 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1643 return -EPERM;
1644
1645 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1646 if (ret)
1647 return ret;
1648
1649 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1650 if (ret)
1651 return ret;
1652
1653 __fanout_set_data_bpf(po->fanout, new);
1654 return 0;
1655}
1656
1657static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1658 unsigned int len)
1659{
1660 struct bpf_prog *new;
1661 u32 fd;
1662
1663 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1664 return -EPERM;
1665 if (len != sizeof(fd))
1666 return -EINVAL;
1667 if (copy_from_sockptr(&fd, data, len))
1668 return -EFAULT;
1669
1670 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1671 if (IS_ERR(new))
1672 return PTR_ERR(new);
1673
1674 __fanout_set_data_bpf(po->fanout, new);
1675 return 0;
1676}
1677
1678static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1679 unsigned int len)
1680{
1681 switch (po->fanout->type) {
1682 case PACKET_FANOUT_CBPF:
1683 return fanout_set_data_cbpf(po, data, len);
1684 case PACKET_FANOUT_EBPF:
1685 return fanout_set_data_ebpf(po, data, len);
1686 default:
1687 return -EINVAL;
1688 }
1689}
1690
1691static void fanout_release_data(struct packet_fanout *f)
1692{
1693 switch (f->type) {
1694 case PACKET_FANOUT_CBPF:
1695 case PACKET_FANOUT_EBPF:
1696 __fanout_set_data_bpf(f, NULL);
1697 }
1698}
1699
1700static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1701{
1702 struct packet_fanout *f;
1703
1704 list_for_each_entry(f, &fanout_list, list) {
1705 if (f->id == candidate_id &&
1706 read_pnet(&f->net) == sock_net(sk)) {
1707 return false;
1708 }
1709 }
1710 return true;
1711}
1712
1713static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1714{
1715 u16 id = fanout_next_id;
1716
1717 do {
1718 if (__fanout_id_is_free(sk, id)) {
1719 *new_id = id;
1720 fanout_next_id = id + 1;
1721 return true;
1722 }
1723
1724 id++;
1725 } while (id != fanout_next_id);
1726
1727 return false;
1728}
1729
1730static int fanout_add(struct sock *sk, struct fanout_args *args)
1731{
1732 struct packet_rollover *rollover = NULL;
1733 struct packet_sock *po = pkt_sk(sk);
1734 u16 type_flags = args->type_flags;
1735 struct packet_fanout *f, *match;
1736 u8 type = type_flags & 0xff;
1737 u8 flags = type_flags >> 8;
1738 u16 id = args->id;
1739 int err;
1740
1741 switch (type) {
1742 case PACKET_FANOUT_ROLLOVER:
1743 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1744 return -EINVAL;
1745 break;
1746 case PACKET_FANOUT_HASH:
1747 case PACKET_FANOUT_LB:
1748 case PACKET_FANOUT_CPU:
1749 case PACKET_FANOUT_RND:
1750 case PACKET_FANOUT_QM:
1751 case PACKET_FANOUT_CBPF:
1752 case PACKET_FANOUT_EBPF:
1753 break;
1754 default:
1755 return -EINVAL;
1756 }
1757
1758 mutex_lock(&fanout_mutex);
1759
1760 err = -EALREADY;
1761 if (po->fanout)
1762 goto out;
1763
1764 if (type == PACKET_FANOUT_ROLLOVER ||
1765 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1766 err = -ENOMEM;
1767 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1768 if (!rollover)
1769 goto out;
1770 atomic_long_set(&rollover->num, 0);
1771 atomic_long_set(&rollover->num_huge, 0);
1772 atomic_long_set(&rollover->num_failed, 0);
1773 }
1774
1775 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1776 if (id != 0) {
1777 err = -EINVAL;
1778 goto out;
1779 }
1780 if (!fanout_find_new_id(sk, &id)) {
1781 err = -ENOMEM;
1782 goto out;
1783 }
1784 /* ephemeral flag for the first socket in the group: drop it */
1785 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1786 }
1787
1788 match = NULL;
1789 list_for_each_entry(f, &fanout_list, list) {
1790 if (f->id == id &&
1791 read_pnet(&f->net) == sock_net(sk)) {
1792 match = f;
1793 break;
1794 }
1795 }
1796 err = -EINVAL;
1797 if (match) {
1798 if (match->flags != flags)
1799 goto out;
1800 if (args->max_num_members &&
1801 args->max_num_members != match->max_num_members)
1802 goto out;
1803 } else {
1804 if (args->max_num_members > PACKET_FANOUT_MAX)
1805 goto out;
1806 if (!args->max_num_members)
1807 /* legacy PACKET_FANOUT_MAX */
1808 args->max_num_members = 256;
1809 err = -ENOMEM;
1810 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1811 GFP_KERNEL);
1812 if (!match)
1813 goto out;
1814 write_pnet(&match->net, sock_net(sk));
1815 match->id = id;
1816 match->type = type;
1817 match->flags = flags;
1818 INIT_LIST_HEAD(&match->list);
1819 spin_lock_init(&match->lock);
1820 refcount_set(&match->sk_ref, 0);
1821 fanout_init_data(match);
1822 match->prot_hook.type = po->prot_hook.type;
1823 match->prot_hook.dev = po->prot_hook.dev;
1824 match->prot_hook.func = packet_rcv_fanout;
1825 match->prot_hook.af_packet_priv = match;
1826 match->prot_hook.af_packet_net = read_pnet(&match->net);
1827 match->prot_hook.id_match = match_fanout_group;
1828 match->max_num_members = args->max_num_members;
1829 match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1830 list_add(&match->list, &fanout_list);
1831 }
1832 err = -EINVAL;
1833
1834 spin_lock(&po->bind_lock);
1835 if (po->num &&
1836 match->type == type &&
1837 match->prot_hook.type == po->prot_hook.type &&
1838 match->prot_hook.dev == po->prot_hook.dev) {
1839 err = -ENOSPC;
1840 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1841 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1842 WRITE_ONCE(po->fanout, match);
1843
1844 po->rollover = rollover;
1845 rollover = NULL;
1846 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1847 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
1848 __dev_remove_pack(&po->prot_hook);
1849 __fanout_link(sk, po);
1850 }
1851 err = 0;
1852 }
1853 }
1854 spin_unlock(&po->bind_lock);
1855
1856 if (err && !refcount_read(&match->sk_ref)) {
1857 list_del(&match->list);
1858 kvfree(match);
1859 }
1860
1861out:
1862 kfree(rollover);
1863 mutex_unlock(&fanout_mutex);
1864 return err;
1865}
1866
1867/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1868 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1869 * It is the responsibility of the caller to call fanout_release_data() and
1870 * free the returned packet_fanout (after synchronize_net())
1871 */
1872static struct packet_fanout *fanout_release(struct sock *sk)
1873{
1874 struct packet_sock *po = pkt_sk(sk);
1875 struct packet_fanout *f;
1876
1877 mutex_lock(&fanout_mutex);
1878 f = po->fanout;
1879 if (f) {
1880 po->fanout = NULL;
1881
1882 if (refcount_dec_and_test(&f->sk_ref))
1883 list_del(&f->list);
1884 else
1885 f = NULL;
1886 }
1887 mutex_unlock(&fanout_mutex);
1888
1889 return f;
1890}
1891
1892static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1893 struct sk_buff *skb)
1894{
1895 /* Earlier code assumed this would be a VLAN pkt, double-check
1896 * this now that we have the actual packet in hand. We can only
1897 * do this check on Ethernet devices.
1898 */
1899 if (unlikely(dev->type != ARPHRD_ETHER))
1900 return false;
1901
1902 skb_reset_mac_header(skb);
1903 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1904}
1905
1906static const struct proto_ops packet_ops;
1907
1908static const struct proto_ops packet_ops_spkt;
1909
1910static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1911 struct packet_type *pt, struct net_device *orig_dev)
1912{
1913 struct sock *sk;
1914 struct sockaddr_pkt *spkt;
1915
1916 /*
1917 * When we registered the protocol we saved the socket in the data
1918 * field for just this event.
1919 */
1920
1921 sk = pt->af_packet_priv;
1922
1923 /*
1924 * Yank back the headers [hope the device set this
1925 * right or kerboom...]
1926 *
1927 * Incoming packets have ll header pulled,
1928 * push it back.
1929 *
1930 * For outgoing ones skb->data == skb_mac_header(skb)
1931 * so that this procedure is noop.
1932 */
1933
1934 if (skb->pkt_type == PACKET_LOOPBACK)
1935 goto out;
1936
1937 if (!net_eq(dev_net(dev), sock_net(sk)))
1938 goto out;
1939
1940 skb = skb_share_check(skb, GFP_ATOMIC);
1941 if (skb == NULL)
1942 goto oom;
1943
1944 /* drop any routing info */
1945 skb_dst_drop(skb);
1946
1947 /* drop conntrack reference */
1948 nf_reset_ct(skb);
1949
1950 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1951
1952 skb_push(skb, skb->data - skb_mac_header(skb));
1953
1954 /*
1955 * The SOCK_PACKET socket receives _all_ frames.
1956 */
1957
1958 spkt->spkt_family = dev->type;
1959 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1960 spkt->spkt_protocol = skb->protocol;
1961
1962 /*
1963 * Charge the memory to the socket. This is done specifically
1964 * to prevent sockets using all the memory up.
1965 */
1966
1967 if (sock_queue_rcv_skb(sk, skb) == 0)
1968 return 0;
1969
1970out:
1971 kfree_skb(skb);
1972oom:
1973 return 0;
1974}
1975
1976static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1977{
1978 int depth;
1979
1980 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1981 sock->type == SOCK_RAW) {
1982 skb_reset_mac_header(skb);
1983 skb->protocol = dev_parse_header_protocol(skb);
1984 }
1985
1986 /* Move network header to the right position for VLAN tagged packets */
1987 if (likely(skb->dev->type == ARPHRD_ETHER) &&
1988 eth_type_vlan(skb->protocol) &&
1989 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1990 skb_set_network_header(skb, depth);
1991
1992 skb_probe_transport_header(skb);
1993}
1994
1995/*
1996 * Output a raw packet to a device layer. This bypasses all the other
1997 * protocol layers and you must therefore supply it with a complete frame
1998 */
1999
2000static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
2001 size_t len)
2002{
2003 struct sock *sk = sock->sk;
2004 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
2005 struct sk_buff *skb = NULL;
2006 struct net_device *dev;
2007 struct sockcm_cookie sockc;
2008 __be16 proto = 0;
2009 int err;
2010 int extra_len = 0;
2011
2012 /*
2013 * Get and verify the address.
2014 */
2015
2016 if (saddr) {
2017 if (msg->msg_namelen < sizeof(struct sockaddr))
2018 return -EINVAL;
2019 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
2020 proto = saddr->spkt_protocol;
2021 } else
2022 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
2023
2024 /*
2025 * Find the device first to size check it
2026 */
2027
2028 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
2029retry:
2030 rcu_read_lock();
2031 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
2032 err = -ENODEV;
2033 if (dev == NULL)
2034 goto out_unlock;
2035
2036 err = -ENETDOWN;
2037 if (!(dev->flags & IFF_UP))
2038 goto out_unlock;
2039
2040 /*
2041 * You may not queue a frame bigger than the mtu. This is the lowest level
2042 * raw protocol and you must do your own fragmentation at this level.
2043 */
2044
2045 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2046 if (!netif_supports_nofcs(dev)) {
2047 err = -EPROTONOSUPPORT;
2048 goto out_unlock;
2049 }
2050 extra_len = 4; /* We're doing our own CRC */
2051 }
2052
2053 err = -EMSGSIZE;
2054 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2055 goto out_unlock;
2056
2057 if (!skb) {
2058 size_t reserved = LL_RESERVED_SPACE(dev);
2059 int tlen = dev->needed_tailroom;
2060 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2061
2062 rcu_read_unlock();
2063 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2064 if (skb == NULL)
2065 return -ENOBUFS;
2066 /* FIXME: Save some space for broken drivers that write a hard
2067 * header at transmission time by themselves. PPP is the notable
2068 * one here. This should really be fixed at the driver level.
2069 */
2070 skb_reserve(skb, reserved);
2071 skb_reset_network_header(skb);
2072
2073 /* Try to align data part correctly */
2074 if (hhlen) {
2075 skb->data -= hhlen;
2076 skb->tail -= hhlen;
2077 if (len < hhlen)
2078 skb_reset_network_header(skb);
2079 }
2080 err = memcpy_from_msg(skb_put(skb, len), msg, len);
2081 if (err)
2082 goto out_free;
2083 goto retry;
2084 }
2085
2086 if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2087 err = -EINVAL;
2088 goto out_unlock;
2089 }
2090 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2091 !packet_extra_vlan_len_allowed(dev, skb)) {
2092 err = -EMSGSIZE;
2093 goto out_unlock;
2094 }
2095
2096 sockcm_init(&sockc, sk);
2097 if (msg->msg_controllen) {
2098 err = sock_cmsg_send(sk, msg, &sockc);
2099 if (unlikely(err))
2100 goto out_unlock;
2101 }
2102
2103 skb->protocol = proto;
2104 skb->dev = dev;
2105 skb->priority = READ_ONCE(sk->sk_priority);
2106 skb->mark = READ_ONCE(sk->sk_mark);
2107 skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
2108 skb_setup_tx_timestamp(skb, &sockc);
2109
2110 if (unlikely(extra_len == 4))
2111 skb->no_fcs = 1;
2112
2113 packet_parse_headers(skb, sock);
2114
2115 dev_queue_xmit(skb);
2116 rcu_read_unlock();
2117 return len;
2118
2119out_unlock:
2120 rcu_read_unlock();
2121out_free:
2122 kfree_skb(skb);
2123 return err;
2124}
2125
2126static unsigned int run_filter(struct sk_buff *skb,
2127 const struct sock *sk,
2128 unsigned int res)
2129{
2130 struct sk_filter *filter;
2131
2132 rcu_read_lock();
2133 filter = rcu_dereference(sk->sk_filter);
2134 if (filter != NULL)
2135 res = bpf_prog_run_clear_cb(filter->prog, skb);
2136 rcu_read_unlock();
2137
2138 return res;
2139}
2140
2141static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2142 size_t *len, int vnet_hdr_sz)
2143{
2144 struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2145
2146 if (*len < vnet_hdr_sz)
2147 return -EINVAL;
2148 *len -= vnet_hdr_sz;
2149
2150 if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2151 return -EINVAL;
2152
2153 return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
2154}
2155
2156/*
2157 * This function makes lazy skb cloning in hope that most of packets
2158 * are discarded by BPF.
2159 *
2160 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2161 * and skb->cb are mangled. It works because (and until) packets
2162 * falling here are owned by current CPU. Output packets are cloned
2163 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2164 * sequentially, so that if we return skb to original state on exit,
2165 * we will not harm anyone.
2166 */
2167
2168static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2169 struct packet_type *pt, struct net_device *orig_dev)
2170{
2171 enum skb_drop_reason drop_reason = SKB_CONSUMED;
2172 struct sock *sk = NULL;
2173 struct sockaddr_ll *sll;
2174 struct packet_sock *po;
2175 u8 *skb_head = skb->data;
2176 int skb_len = skb->len;
2177 unsigned int snaplen, res;
2178
2179 if (skb->pkt_type == PACKET_LOOPBACK)
2180 goto drop;
2181
2182 sk = pt->af_packet_priv;
2183 po = pkt_sk(sk);
2184
2185 if (!net_eq(dev_net(dev), sock_net(sk)))
2186 goto drop;
2187
2188 skb->dev = dev;
2189
2190 if (dev_has_header(dev)) {
2191 /* The device has an explicit notion of ll header,
2192 * exported to higher levels.
2193 *
2194 * Otherwise, the device hides details of its frame
2195 * structure, so that corresponding packet head is
2196 * never delivered to user.
2197 */
2198 if (sk->sk_type != SOCK_DGRAM)
2199 skb_push(skb, skb->data - skb_mac_header(skb));
2200 else if (skb->pkt_type == PACKET_OUTGOING) {
2201 /* Special case: outgoing packets have ll header at head */
2202 skb_pull(skb, skb_network_offset(skb));
2203 }
2204 }
2205
2206 snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2207
2208 res = run_filter(skb, sk, snaplen);
2209 if (!res)
2210 goto drop_n_restore;
2211 if (snaplen > res)
2212 snaplen = res;
2213
2214 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2215 goto drop_n_acct;
2216
2217 if (skb_shared(skb)) {
2218 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2219 if (nskb == NULL)
2220 goto drop_n_acct;
2221
2222 if (skb_head != skb->data) {
2223 skb->data = skb_head;
2224 skb->len = skb_len;
2225 }
2226 consume_skb(skb);
2227 skb = nskb;
2228 }
2229
2230 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2231
2232 sll = &PACKET_SKB_CB(skb)->sa.ll;
2233 sll->sll_hatype = dev->type;
2234 sll->sll_pkttype = skb->pkt_type;
2235 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2236 sll->sll_ifindex = orig_dev->ifindex;
2237 else
2238 sll->sll_ifindex = dev->ifindex;
2239
2240 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2241
2242 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2243 * Use their space for storing the original skb length.
2244 */
2245 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2246
2247 if (pskb_trim(skb, snaplen))
2248 goto drop_n_acct;
2249
2250 skb_set_owner_r(skb, sk);
2251 skb->dev = NULL;
2252 skb_dst_drop(skb);
2253
2254 /* drop conntrack reference */
2255 nf_reset_ct(skb);
2256
2257 spin_lock(&sk->sk_receive_queue.lock);
2258 po->stats.stats1.tp_packets++;
2259 sock_skb_set_dropcount(sk, skb);
2260 skb_clear_delivery_time(skb);
2261 __skb_queue_tail(&sk->sk_receive_queue, skb);
2262 spin_unlock(&sk->sk_receive_queue.lock);
2263 sk->sk_data_ready(sk);
2264 return 0;
2265
2266drop_n_acct:
2267 atomic_inc(&po->tp_drops);
2268 atomic_inc(&sk->sk_drops);
2269 drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2270
2271drop_n_restore:
2272 if (skb_head != skb->data && skb_shared(skb)) {
2273 skb->data = skb_head;
2274 skb->len = skb_len;
2275 }
2276drop:
2277 sk_skb_reason_drop(sk, skb, drop_reason);
2278 return 0;
2279}
2280
2281static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2282 struct packet_type *pt, struct net_device *orig_dev)
2283{
2284 enum skb_drop_reason drop_reason = SKB_CONSUMED;
2285 struct sock *sk = NULL;
2286 struct packet_sock *po;
2287 struct sockaddr_ll *sll;
2288 union tpacket_uhdr h;
2289 u8 *skb_head = skb->data;
2290 int skb_len = skb->len;
2291 unsigned int snaplen, res;
2292 unsigned long status = TP_STATUS_USER;
2293 unsigned short macoff, hdrlen;
2294 unsigned int netoff;
2295 struct sk_buff *copy_skb = NULL;
2296 struct timespec64 ts;
2297 __u32 ts_status;
2298 unsigned int slot_id = 0;
2299 int vnet_hdr_sz = 0;
2300
2301 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2302 * We may add members to them until current aligned size without forcing
2303 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2304 */
2305 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2306 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2307
2308 if (skb->pkt_type == PACKET_LOOPBACK)
2309 goto drop;
2310
2311 sk = pt->af_packet_priv;
2312 po = pkt_sk(sk);
2313
2314 if (!net_eq(dev_net(dev), sock_net(sk)))
2315 goto drop;
2316
2317 if (dev_has_header(dev)) {
2318 if (sk->sk_type != SOCK_DGRAM)
2319 skb_push(skb, skb->data - skb_mac_header(skb));
2320 else if (skb->pkt_type == PACKET_OUTGOING) {
2321 /* Special case: outgoing packets have ll header at head */
2322 skb_pull(skb, skb_network_offset(skb));
2323 }
2324 }
2325
2326 snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2327
2328 res = run_filter(skb, sk, snaplen);
2329 if (!res)
2330 goto drop_n_restore;
2331
2332 /* If we are flooded, just give up */
2333 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2334 atomic_inc(&po->tp_drops);
2335 goto drop_n_restore;
2336 }
2337
2338 if (skb->ip_summed == CHECKSUM_PARTIAL)
2339 status |= TP_STATUS_CSUMNOTREADY;
2340 else if (skb->pkt_type != PACKET_OUTGOING &&
2341 skb_csum_unnecessary(skb))
2342 status |= TP_STATUS_CSUM_VALID;
2343 if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2344 status |= TP_STATUS_GSO_TCP;
2345
2346 if (snaplen > res)
2347 snaplen = res;
2348
2349 if (sk->sk_type == SOCK_DGRAM) {
2350 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2351 po->tp_reserve;
2352 } else {
2353 unsigned int maclen = skb_network_offset(skb);
2354 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2355 (maclen < 16 ? 16 : maclen)) +
2356 po->tp_reserve;
2357 vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2358 if (vnet_hdr_sz)
2359 netoff += vnet_hdr_sz;
2360 macoff = netoff - maclen;
2361 }
2362 if (netoff > USHRT_MAX) {
2363 atomic_inc(&po->tp_drops);
2364 goto drop_n_restore;
2365 }
2366 if (po->tp_version <= TPACKET_V2) {
2367 if (macoff + snaplen > po->rx_ring.frame_size) {
2368 if (READ_ONCE(po->copy_thresh) &&
2369 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2370 if (skb_shared(skb)) {
2371 copy_skb = skb_clone(skb, GFP_ATOMIC);
2372 } else {
2373 copy_skb = skb_get(skb);
2374 skb_head = skb->data;
2375 }
2376 if (copy_skb) {
2377 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2378 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2379 skb_set_owner_r(copy_skb, sk);
2380 }
2381 }
2382 snaplen = po->rx_ring.frame_size - macoff;
2383 if ((int)snaplen < 0) {
2384 snaplen = 0;
2385 vnet_hdr_sz = 0;
2386 }
2387 }
2388 } else if (unlikely(macoff + snaplen >
2389 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2390 u32 nval;
2391
2392 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2393 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2394 snaplen, nval, macoff);
2395 snaplen = nval;
2396 if (unlikely((int)snaplen < 0)) {
2397 snaplen = 0;
2398 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2399 vnet_hdr_sz = 0;
2400 }
2401 }
2402 spin_lock(&sk->sk_receive_queue.lock);
2403 h.raw = packet_current_rx_frame(po, skb,
2404 TP_STATUS_KERNEL, (macoff+snaplen));
2405 if (!h.raw)
2406 goto drop_n_account;
2407
2408 if (po->tp_version <= TPACKET_V2) {
2409 slot_id = po->rx_ring.head;
2410 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2411 goto drop_n_account;
2412 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2413 }
2414
2415 if (vnet_hdr_sz &&
2416 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2417 sizeof(struct virtio_net_hdr),
2418 vio_le(), true, 0)) {
2419 if (po->tp_version == TPACKET_V3)
2420 prb_clear_blk_fill_status(&po->rx_ring);
2421 goto drop_n_account;
2422 }
2423
2424 if (po->tp_version <= TPACKET_V2) {
2425 packet_increment_rx_head(po, &po->rx_ring);
2426 /*
2427 * LOSING will be reported till you read the stats,
2428 * because it's COR - Clear On Read.
2429 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2430 * at packet level.
2431 */
2432 if (atomic_read(&po->tp_drops))
2433 status |= TP_STATUS_LOSING;
2434 }
2435
2436 po->stats.stats1.tp_packets++;
2437 if (copy_skb) {
2438 status |= TP_STATUS_COPY;
2439 skb_clear_delivery_time(copy_skb);
2440 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2441 }
2442 spin_unlock(&sk->sk_receive_queue.lock);
2443
2444 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2445
2446 /* Always timestamp; prefer an existing software timestamp taken
2447 * closer to the time of capture.
2448 */
2449 ts_status = tpacket_get_timestamp(skb, &ts,
2450 READ_ONCE(po->tp_tstamp) |
2451 SOF_TIMESTAMPING_SOFTWARE);
2452 if (!ts_status)
2453 ktime_get_real_ts64(&ts);
2454
2455 status |= ts_status;
2456
2457 switch (po->tp_version) {
2458 case TPACKET_V1:
2459 h.h1->tp_len = skb->len;
2460 h.h1->tp_snaplen = snaplen;
2461 h.h1->tp_mac = macoff;
2462 h.h1->tp_net = netoff;
2463 h.h1->tp_sec = ts.tv_sec;
2464 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2465 hdrlen = sizeof(*h.h1);
2466 break;
2467 case TPACKET_V2:
2468 h.h2->tp_len = skb->len;
2469 h.h2->tp_snaplen = snaplen;
2470 h.h2->tp_mac = macoff;
2471 h.h2->tp_net = netoff;
2472 h.h2->tp_sec = ts.tv_sec;
2473 h.h2->tp_nsec = ts.tv_nsec;
2474 if (skb_vlan_tag_present(skb)) {
2475 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2476 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2477 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2478 } else if (unlikely(sk->sk_type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
2479 h.h2->tp_vlan_tci = vlan_get_tci(skb, skb->dev);
2480 h.h2->tp_vlan_tpid = ntohs(skb->protocol);
2481 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2482 } else {
2483 h.h2->tp_vlan_tci = 0;
2484 h.h2->tp_vlan_tpid = 0;
2485 }
2486 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2487 hdrlen = sizeof(*h.h2);
2488 break;
2489 case TPACKET_V3:
2490 /* tp_nxt_offset,vlan are already populated above.
2491 * So DONT clear those fields here
2492 */
2493 h.h3->tp_status |= status;
2494 h.h3->tp_len = skb->len;
2495 h.h3->tp_snaplen = snaplen;
2496 h.h3->tp_mac = macoff;
2497 h.h3->tp_net = netoff;
2498 h.h3->tp_sec = ts.tv_sec;
2499 h.h3->tp_nsec = ts.tv_nsec;
2500 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2501 hdrlen = sizeof(*h.h3);
2502 break;
2503 default:
2504 BUG();
2505 }
2506
2507 sll = h.raw + TPACKET_ALIGN(hdrlen);
2508 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2509 sll->sll_family = AF_PACKET;
2510 sll->sll_hatype = dev->type;
2511 sll->sll_protocol = (sk->sk_type == SOCK_DGRAM) ?
2512 vlan_get_protocol_dgram(skb) : skb->protocol;
2513 sll->sll_pkttype = skb->pkt_type;
2514 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2515 sll->sll_ifindex = orig_dev->ifindex;
2516 else
2517 sll->sll_ifindex = dev->ifindex;
2518
2519 smp_mb();
2520
2521#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2522 if (po->tp_version <= TPACKET_V2) {
2523 u8 *start, *end;
2524
2525 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2526 macoff + snaplen);
2527
2528 for (start = h.raw; start < end; start += PAGE_SIZE)
2529 flush_dcache_page(pgv_to_page(start));
2530 }
2531 smp_wmb();
2532#endif
2533
2534 if (po->tp_version <= TPACKET_V2) {
2535 spin_lock(&sk->sk_receive_queue.lock);
2536 __packet_set_status(po, h.raw, status);
2537 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2538 spin_unlock(&sk->sk_receive_queue.lock);
2539 sk->sk_data_ready(sk);
2540 } else if (po->tp_version == TPACKET_V3) {
2541 prb_clear_blk_fill_status(&po->rx_ring);
2542 }
2543
2544drop_n_restore:
2545 if (skb_head != skb->data && skb_shared(skb)) {
2546 skb->data = skb_head;
2547 skb->len = skb_len;
2548 }
2549drop:
2550 sk_skb_reason_drop(sk, skb, drop_reason);
2551 return 0;
2552
2553drop_n_account:
2554 spin_unlock(&sk->sk_receive_queue.lock);
2555 atomic_inc(&po->tp_drops);
2556 drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2557
2558 sk->sk_data_ready(sk);
2559 sk_skb_reason_drop(sk, copy_skb, drop_reason);
2560 goto drop_n_restore;
2561}
2562
2563static void tpacket_destruct_skb(struct sk_buff *skb)
2564{
2565 struct packet_sock *po = pkt_sk(skb->sk);
2566
2567 if (likely(po->tx_ring.pg_vec)) {
2568 void *ph;
2569 __u32 ts;
2570
2571 ph = skb_zcopy_get_nouarg(skb);
2572 packet_dec_pending(&po->tx_ring);
2573
2574 ts = __packet_set_timestamp(po, ph, skb);
2575 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2576
2577 complete(&po->skb_completion);
2578 }
2579
2580 sock_wfree(skb);
2581}
2582
2583static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2584{
2585 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2586 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2587 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2588 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2589 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2590 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2591 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2592
2593 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2594 return -EINVAL;
2595
2596 return 0;
2597}
2598
2599static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2600 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz)
2601{
2602 int ret;
2603
2604 if (*len < vnet_hdr_sz)
2605 return -EINVAL;
2606 *len -= vnet_hdr_sz;
2607
2608 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2609 return -EFAULT;
2610
2611 ret = __packet_snd_vnet_parse(vnet_hdr, *len);
2612 if (ret)
2613 return ret;
2614
2615 /* move iter to point to the start of mac header */
2616 if (vnet_hdr_sz != sizeof(struct virtio_net_hdr))
2617 iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr));
2618
2619 return 0;
2620}
2621
2622static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2623 void *frame, struct net_device *dev, void *data, int tp_len,
2624 __be16 proto, unsigned char *addr, int hlen, int copylen,
2625 const struct sockcm_cookie *sockc)
2626{
2627 union tpacket_uhdr ph;
2628 int to_write, offset, len, nr_frags, len_max;
2629 struct socket *sock = po->sk.sk_socket;
2630 struct page *page;
2631 int err;
2632
2633 ph.raw = frame;
2634
2635 skb->protocol = proto;
2636 skb->dev = dev;
2637 skb->priority = READ_ONCE(po->sk.sk_priority);
2638 skb->mark = READ_ONCE(po->sk.sk_mark);
2639 skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, po->sk.sk_clockid);
2640 skb_setup_tx_timestamp(skb, sockc);
2641 skb_zcopy_set_nouarg(skb, ph.raw);
2642
2643 skb_reserve(skb, hlen);
2644 skb_reset_network_header(skb);
2645
2646 to_write = tp_len;
2647
2648 if (sock->type == SOCK_DGRAM) {
2649 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2650 NULL, tp_len);
2651 if (unlikely(err < 0))
2652 return -EINVAL;
2653 } else if (copylen) {
2654 int hdrlen = min_t(int, copylen, tp_len);
2655
2656 skb_push(skb, dev->hard_header_len);
2657 skb_put(skb, copylen - dev->hard_header_len);
2658 err = skb_store_bits(skb, 0, data, hdrlen);
2659 if (unlikely(err))
2660 return err;
2661 if (!dev_validate_header(dev, skb->data, hdrlen))
2662 return -EINVAL;
2663
2664 data += hdrlen;
2665 to_write -= hdrlen;
2666 }
2667
2668 offset = offset_in_page(data);
2669 len_max = PAGE_SIZE - offset;
2670 len = ((to_write > len_max) ? len_max : to_write);
2671
2672 skb->data_len = to_write;
2673 skb->len += to_write;
2674 skb->truesize += to_write;
2675 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2676
2677 while (likely(to_write)) {
2678 nr_frags = skb_shinfo(skb)->nr_frags;
2679
2680 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2681 pr_err("Packet exceed the number of skb frags(%u)\n",
2682 (unsigned int)MAX_SKB_FRAGS);
2683 return -EFAULT;
2684 }
2685
2686 page = pgv_to_page(data);
2687 data += len;
2688 flush_dcache_page(page);
2689 get_page(page);
2690 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2691 to_write -= len;
2692 offset = 0;
2693 len_max = PAGE_SIZE;
2694 len = ((to_write > len_max) ? len_max : to_write);
2695 }
2696
2697 packet_parse_headers(skb, sock);
2698
2699 return tp_len;
2700}
2701
2702static int tpacket_parse_header(struct packet_sock *po, void *frame,
2703 int size_max, void **data)
2704{
2705 union tpacket_uhdr ph;
2706 int tp_len, off;
2707
2708 ph.raw = frame;
2709
2710 switch (po->tp_version) {
2711 case TPACKET_V3:
2712 if (ph.h3->tp_next_offset != 0) {
2713 pr_warn_once("variable sized slot not supported");
2714 return -EINVAL;
2715 }
2716 tp_len = ph.h3->tp_len;
2717 break;
2718 case TPACKET_V2:
2719 tp_len = ph.h2->tp_len;
2720 break;
2721 default:
2722 tp_len = ph.h1->tp_len;
2723 break;
2724 }
2725 if (unlikely(tp_len > size_max)) {
2726 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2727 return -EMSGSIZE;
2728 }
2729
2730 if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) {
2731 int off_min, off_max;
2732
2733 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2734 off_max = po->tx_ring.frame_size - tp_len;
2735 if (po->sk.sk_type == SOCK_DGRAM) {
2736 switch (po->tp_version) {
2737 case TPACKET_V3:
2738 off = ph.h3->tp_net;
2739 break;
2740 case TPACKET_V2:
2741 off = ph.h2->tp_net;
2742 break;
2743 default:
2744 off = ph.h1->tp_net;
2745 break;
2746 }
2747 } else {
2748 switch (po->tp_version) {
2749 case TPACKET_V3:
2750 off = ph.h3->tp_mac;
2751 break;
2752 case TPACKET_V2:
2753 off = ph.h2->tp_mac;
2754 break;
2755 default:
2756 off = ph.h1->tp_mac;
2757 break;
2758 }
2759 }
2760 if (unlikely((off < off_min) || (off_max < off)))
2761 return -EINVAL;
2762 } else {
2763 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2764 }
2765
2766 *data = frame + off;
2767 return tp_len;
2768}
2769
2770static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2771{
2772 struct sk_buff *skb = NULL;
2773 struct net_device *dev;
2774 struct virtio_net_hdr *vnet_hdr = NULL;
2775 struct sockcm_cookie sockc;
2776 __be16 proto;
2777 int err, reserve = 0;
2778 void *ph;
2779 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2780 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2781 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2782 unsigned char *addr = NULL;
2783 int tp_len, size_max;
2784 void *data;
2785 int len_sum = 0;
2786 int status = TP_STATUS_AVAILABLE;
2787 int hlen, tlen, copylen = 0;
2788 long timeo = 0;
2789
2790 mutex_lock(&po->pg_vec_lock);
2791
2792 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2793 * we need to confirm it under protection of pg_vec_lock.
2794 */
2795 if (unlikely(!po->tx_ring.pg_vec)) {
2796 err = -EBUSY;
2797 goto out;
2798 }
2799 if (likely(saddr == NULL)) {
2800 dev = packet_cached_dev_get(po);
2801 proto = READ_ONCE(po->num);
2802 } else {
2803 err = -EINVAL;
2804 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2805 goto out;
2806 if (msg->msg_namelen < (saddr->sll_halen
2807 + offsetof(struct sockaddr_ll,
2808 sll_addr)))
2809 goto out;
2810 proto = saddr->sll_protocol;
2811 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2812 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2813 if (dev && msg->msg_namelen < dev->addr_len +
2814 offsetof(struct sockaddr_ll, sll_addr))
2815 goto out_put;
2816 addr = saddr->sll_addr;
2817 }
2818 }
2819
2820 err = -ENXIO;
2821 if (unlikely(dev == NULL))
2822 goto out;
2823 err = -ENETDOWN;
2824 if (unlikely(!(dev->flags & IFF_UP)))
2825 goto out_put;
2826
2827 sockcm_init(&sockc, &po->sk);
2828 if (msg->msg_controllen) {
2829 err = sock_cmsg_send(&po->sk, msg, &sockc);
2830 if (unlikely(err))
2831 goto out_put;
2832 }
2833
2834 if (po->sk.sk_socket->type == SOCK_RAW)
2835 reserve = dev->hard_header_len;
2836 size_max = po->tx_ring.frame_size
2837 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2838
2839 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2840 size_max = dev->mtu + reserve + VLAN_HLEN;
2841
2842 reinit_completion(&po->skb_completion);
2843
2844 do {
2845 ph = packet_current_frame(po, &po->tx_ring,
2846 TP_STATUS_SEND_REQUEST);
2847 if (unlikely(ph == NULL)) {
2848 if (need_wait && skb) {
2849 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2850 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2851 if (timeo <= 0) {
2852 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2853 goto out_put;
2854 }
2855 }
2856 /* check for additional frames */
2857 continue;
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(dev);
2867 tlen = dev->needed_tailroom;
2868 if (vnet_hdr_sz) {
2869 vnet_hdr = data;
2870 data += vnet_hdr_sz;
2871 tp_len -= vnet_hdr_sz;
2872 if (tp_len < 0 ||
2873 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2874 tp_len = -EINVAL;
2875 goto tpacket_error;
2876 }
2877 copylen = __virtio16_to_cpu(vio_le(),
2878 vnet_hdr->hdr_len);
2879 }
2880 copylen = max_t(int, copylen, dev->hard_header_len);
2881 skb = sock_alloc_send_skb(&po->sk,
2882 hlen + tlen + sizeof(struct sockaddr_ll) +
2883 (copylen - dev->hard_header_len),
2884 !need_wait, &err);
2885
2886 if (unlikely(skb == NULL)) {
2887 /* we assume the socket was initially writeable ... */
2888 if (likely(len_sum > 0))
2889 err = len_sum;
2890 goto out_status;
2891 }
2892 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2893 addr, hlen, copylen, &sockc);
2894 if (likely(tp_len >= 0) &&
2895 tp_len > dev->mtu + reserve &&
2896 !vnet_hdr_sz &&
2897 !packet_extra_vlan_len_allowed(dev, skb))
2898 tp_len = -EMSGSIZE;
2899
2900 if (unlikely(tp_len < 0)) {
2901tpacket_error:
2902 if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2903 __packet_set_status(po, ph,
2904 TP_STATUS_AVAILABLE);
2905 packet_increment_head(&po->tx_ring);
2906 kfree_skb(skb);
2907 continue;
2908 } else {
2909 status = TP_STATUS_WRONG_FORMAT;
2910 err = tp_len;
2911 goto out_status;
2912 }
2913 }
2914
2915 if (vnet_hdr_sz) {
2916 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2917 tp_len = -EINVAL;
2918 goto tpacket_error;
2919 }
2920 virtio_net_hdr_set_proto(skb, vnet_hdr);
2921 }
2922
2923 skb->destructor = tpacket_destruct_skb;
2924 __packet_set_status(po, ph, TP_STATUS_SENDING);
2925 packet_inc_pending(&po->tx_ring);
2926
2927 status = TP_STATUS_SEND_REQUEST;
2928 err = packet_xmit(po, skb);
2929 if (unlikely(err != 0)) {
2930 if (err > 0)
2931 err = net_xmit_errno(err);
2932 if (err && __packet_get_status(po, ph) ==
2933 TP_STATUS_AVAILABLE) {
2934 /* skb was destructed already */
2935 skb = NULL;
2936 goto out_status;
2937 }
2938 /*
2939 * skb was dropped but not destructed yet;
2940 * let's treat it like congestion or err < 0
2941 */
2942 err = 0;
2943 }
2944 packet_increment_head(&po->tx_ring);
2945 len_sum += tp_len;
2946 } while (likely((ph != NULL) ||
2947 /* Note: packet_read_pending() might be slow if we have
2948 * to call it as it's per_cpu variable, but in fast-path
2949 * we already short-circuit the loop with the first
2950 * condition, and luckily don't have to go that path
2951 * anyway.
2952 */
2953 (need_wait && packet_read_pending(&po->tx_ring))));
2954
2955 err = len_sum;
2956 goto out_put;
2957
2958out_status:
2959 __packet_set_status(po, ph, status);
2960 kfree_skb(skb);
2961out_put:
2962 dev_put(dev);
2963out:
2964 mutex_unlock(&po->pg_vec_lock);
2965 return err;
2966}
2967
2968static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2969 size_t reserve, size_t len,
2970 size_t linear, int noblock,
2971 int *err)
2972{
2973 struct sk_buff *skb;
2974
2975 /* Under a page? Don't bother with paged skb. */
2976 if (prepad + len < PAGE_SIZE || !linear)
2977 linear = len;
2978
2979 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
2980 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
2981 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2982 err, PAGE_ALLOC_COSTLY_ORDER);
2983 if (!skb)
2984 return NULL;
2985
2986 skb_reserve(skb, reserve);
2987 skb_put(skb, linear);
2988 skb->data_len = len - linear;
2989 skb->len += len - linear;
2990
2991 return skb;
2992}
2993
2994static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2995{
2996 struct sock *sk = sock->sk;
2997 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2998 struct sk_buff *skb;
2999 struct net_device *dev;
3000 __be16 proto;
3001 unsigned char *addr = NULL;
3002 int err, reserve = 0;
3003 struct sockcm_cookie sockc;
3004 struct virtio_net_hdr vnet_hdr = { 0 };
3005 int offset = 0;
3006 struct packet_sock *po = pkt_sk(sk);
3007 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
3008 int hlen, tlen, linear;
3009 int extra_len = 0;
3010
3011 /*
3012 * Get and verify the address.
3013 */
3014
3015 if (likely(saddr == NULL)) {
3016 dev = packet_cached_dev_get(po);
3017 proto = READ_ONCE(po->num);
3018 } else {
3019 err = -EINVAL;
3020 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
3021 goto out;
3022 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
3023 goto out;
3024 proto = saddr->sll_protocol;
3025 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
3026 if (sock->type == SOCK_DGRAM) {
3027 if (dev && msg->msg_namelen < dev->addr_len +
3028 offsetof(struct sockaddr_ll, sll_addr))
3029 goto out_unlock;
3030 addr = saddr->sll_addr;
3031 }
3032 }
3033
3034 err = -ENXIO;
3035 if (unlikely(dev == NULL))
3036 goto out_unlock;
3037 err = -ENETDOWN;
3038 if (unlikely(!(dev->flags & IFF_UP)))
3039 goto out_unlock;
3040
3041 sockcm_init(&sockc, sk);
3042 sockc.mark = READ_ONCE(sk->sk_mark);
3043 if (msg->msg_controllen) {
3044 err = sock_cmsg_send(sk, msg, &sockc);
3045 if (unlikely(err))
3046 goto out_unlock;
3047 }
3048
3049 if (sock->type == SOCK_RAW)
3050 reserve = dev->hard_header_len;
3051 if (vnet_hdr_sz) {
3052 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3053 if (err)
3054 goto out_unlock;
3055 }
3056
3057 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3058 if (!netif_supports_nofcs(dev)) {
3059 err = -EPROTONOSUPPORT;
3060 goto out_unlock;
3061 }
3062 extra_len = 4; /* We're doing our own CRC */
3063 }
3064
3065 err = -EMSGSIZE;
3066 if (!vnet_hdr.gso_type &&
3067 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3068 goto out_unlock;
3069
3070 err = -ENOBUFS;
3071 hlen = LL_RESERVED_SPACE(dev);
3072 tlen = dev->needed_tailroom;
3073 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3074 linear = max(linear, min_t(int, len, dev->hard_header_len));
3075 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3076 msg->msg_flags & MSG_DONTWAIT, &err);
3077 if (skb == NULL)
3078 goto out_unlock;
3079
3080 skb_reset_network_header(skb);
3081
3082 err = -EINVAL;
3083 if (sock->type == SOCK_DGRAM) {
3084 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3085 if (unlikely(offset < 0))
3086 goto out_free;
3087 } else if (reserve) {
3088 skb_reserve(skb, -reserve);
3089 if (len < reserve + sizeof(struct ipv6hdr) &&
3090 dev->min_header_len != dev->hard_header_len)
3091 skb_reset_network_header(skb);
3092 }
3093
3094 /* Returns -EFAULT on error */
3095 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3096 if (err)
3097 goto out_free;
3098
3099 if ((sock->type == SOCK_RAW &&
3100 !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3101 err = -EINVAL;
3102 goto out_free;
3103 }
3104
3105 skb_setup_tx_timestamp(skb, &sockc);
3106
3107 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3108 !packet_extra_vlan_len_allowed(dev, skb)) {
3109 err = -EMSGSIZE;
3110 goto out_free;
3111 }
3112
3113 skb->protocol = proto;
3114 skb->dev = dev;
3115 skb->priority = READ_ONCE(sk->sk_priority);
3116 skb->mark = sockc.mark;
3117 skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
3118
3119 if (unlikely(extra_len == 4))
3120 skb->no_fcs = 1;
3121
3122 packet_parse_headers(skb, sock);
3123
3124 if (vnet_hdr_sz) {
3125 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3126 if (err)
3127 goto out_free;
3128 len += vnet_hdr_sz;
3129 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3130 }
3131
3132 err = packet_xmit(po, skb);
3133
3134 if (unlikely(err != 0)) {
3135 if (err > 0)
3136 err = net_xmit_errno(err);
3137 if (err)
3138 goto out_unlock;
3139 }
3140
3141 dev_put(dev);
3142
3143 return len;
3144
3145out_free:
3146 kfree_skb(skb);
3147out_unlock:
3148 dev_put(dev);
3149out:
3150 return err;
3151}
3152
3153static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3154{
3155 struct sock *sk = sock->sk;
3156 struct packet_sock *po = pkt_sk(sk);
3157
3158 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3159 * tpacket_snd() will redo the check safely.
3160 */
3161 if (data_race(po->tx_ring.pg_vec))
3162 return tpacket_snd(po, msg);
3163
3164 return packet_snd(sock, msg, len);
3165}
3166
3167/*
3168 * Close a PACKET socket. This is fairly simple. We immediately go
3169 * to 'closed' state and remove our protocol entry in the device list.
3170 */
3171
3172static int packet_release(struct socket *sock)
3173{
3174 struct sock *sk = sock->sk;
3175 struct packet_sock *po;
3176 struct packet_fanout *f;
3177 struct net *net;
3178 union tpacket_req_u req_u;
3179
3180 if (!sk)
3181 return 0;
3182
3183 net = sock_net(sk);
3184 po = pkt_sk(sk);
3185
3186 mutex_lock(&net->packet.sklist_lock);
3187 sk_del_node_init_rcu(sk);
3188 mutex_unlock(&net->packet.sklist_lock);
3189
3190 sock_prot_inuse_add(net, sk->sk_prot, -1);
3191
3192 spin_lock(&po->bind_lock);
3193 unregister_prot_hook(sk, false);
3194 packet_cached_dev_reset(po);
3195
3196 if (po->prot_hook.dev) {
3197 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3198 po->prot_hook.dev = NULL;
3199 }
3200 spin_unlock(&po->bind_lock);
3201
3202 packet_flush_mclist(sk);
3203
3204 lock_sock(sk);
3205 if (po->rx_ring.pg_vec) {
3206 memset(&req_u, 0, sizeof(req_u));
3207 packet_set_ring(sk, &req_u, 1, 0);
3208 }
3209
3210 if (po->tx_ring.pg_vec) {
3211 memset(&req_u, 0, sizeof(req_u));
3212 packet_set_ring(sk, &req_u, 1, 1);
3213 }
3214 release_sock(sk);
3215
3216 f = fanout_release(sk);
3217
3218 synchronize_net();
3219
3220 kfree(po->rollover);
3221 if (f) {
3222 fanout_release_data(f);
3223 kvfree(f);
3224 }
3225 /*
3226 * Now the socket is dead. No more input will appear.
3227 */
3228 sock_orphan(sk);
3229 sock->sk = NULL;
3230
3231 /* Purge queues */
3232
3233 skb_queue_purge(&sk->sk_receive_queue);
3234 packet_free_pending(po);
3235
3236 sock_put(sk);
3237 return 0;
3238}
3239
3240/*
3241 * Attach a packet hook.
3242 */
3243
3244static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3245 __be16 proto)
3246{
3247 struct packet_sock *po = pkt_sk(sk);
3248 struct net_device *dev = NULL;
3249 bool unlisted = false;
3250 bool need_rehook;
3251 int ret = 0;
3252
3253 lock_sock(sk);
3254 spin_lock(&po->bind_lock);
3255 if (!proto)
3256 proto = po->num;
3257
3258 rcu_read_lock();
3259
3260 if (po->fanout) {
3261 ret = -EINVAL;
3262 goto out_unlock;
3263 }
3264
3265 if (name) {
3266 dev = dev_get_by_name_rcu(sock_net(sk), name);
3267 if (!dev) {
3268 ret = -ENODEV;
3269 goto out_unlock;
3270 }
3271 } else if (ifindex) {
3272 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3273 if (!dev) {
3274 ret = -ENODEV;
3275 goto out_unlock;
3276 }
3277 }
3278
3279 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3280
3281 if (need_rehook) {
3282 dev_hold(dev);
3283 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3284 rcu_read_unlock();
3285 /* prevents packet_notifier() from calling
3286 * register_prot_hook()
3287 */
3288 WRITE_ONCE(po->num, 0);
3289 __unregister_prot_hook(sk, true);
3290 rcu_read_lock();
3291 if (dev)
3292 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3293 dev->ifindex);
3294 }
3295
3296 BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3297 WRITE_ONCE(po->num, proto);
3298 po->prot_hook.type = proto;
3299
3300 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3301
3302 if (unlikely(unlisted)) {
3303 po->prot_hook.dev = NULL;
3304 WRITE_ONCE(po->ifindex, -1);
3305 packet_cached_dev_reset(po);
3306 } else {
3307 netdev_hold(dev, &po->prot_hook.dev_tracker,
3308 GFP_ATOMIC);
3309 po->prot_hook.dev = dev;
3310 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3311 packet_cached_dev_assign(po, dev);
3312 }
3313 dev_put(dev);
3314 }
3315
3316 if (proto == 0 || !need_rehook)
3317 goto out_unlock;
3318
3319 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3320 register_prot_hook(sk);
3321 } else {
3322 sk->sk_err = ENETDOWN;
3323 if (!sock_flag(sk, SOCK_DEAD))
3324 sk_error_report(sk);
3325 }
3326
3327out_unlock:
3328 rcu_read_unlock();
3329 spin_unlock(&po->bind_lock);
3330 release_sock(sk);
3331 return ret;
3332}
3333
3334/*
3335 * Bind a packet socket to a device
3336 */
3337
3338static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3339 int addr_len)
3340{
3341 struct sock *sk = sock->sk;
3342 char name[sizeof(uaddr->sa_data_min) + 1];
3343
3344 /*
3345 * Check legality
3346 */
3347
3348 if (addr_len != sizeof(struct sockaddr))
3349 return -EINVAL;
3350 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3351 * zero-terminated.
3352 */
3353 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3354 name[sizeof(uaddr->sa_data_min)] = 0;
3355
3356 return packet_do_bind(sk, name, 0, 0);
3357}
3358
3359static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3360{
3361 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3362 struct sock *sk = sock->sk;
3363
3364 /*
3365 * Check legality
3366 */
3367
3368 if (addr_len < sizeof(struct sockaddr_ll))
3369 return -EINVAL;
3370 if (sll->sll_family != AF_PACKET)
3371 return -EINVAL;
3372
3373 return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3374}
3375
3376static struct proto packet_proto = {
3377 .name = "PACKET",
3378 .owner = THIS_MODULE,
3379 .obj_size = sizeof(struct packet_sock),
3380};
3381
3382/*
3383 * Create a packet of type SOCK_PACKET.
3384 */
3385
3386static int packet_create(struct net *net, struct socket *sock, int protocol,
3387 int kern)
3388{
3389 struct sock *sk;
3390 struct packet_sock *po;
3391 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3392 int err;
3393
3394 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3395 return -EPERM;
3396 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3397 sock->type != SOCK_PACKET)
3398 return -ESOCKTNOSUPPORT;
3399
3400 sock->state = SS_UNCONNECTED;
3401
3402 err = -ENOBUFS;
3403 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3404 if (sk == NULL)
3405 goto out;
3406
3407 sock->ops = &packet_ops;
3408 if (sock->type == SOCK_PACKET)
3409 sock->ops = &packet_ops_spkt;
3410
3411 po = pkt_sk(sk);
3412 err = packet_alloc_pending(po);
3413 if (err)
3414 goto out_sk_free;
3415
3416 sock_init_data(sock, sk);
3417
3418 init_completion(&po->skb_completion);
3419 sk->sk_family = PF_PACKET;
3420 po->num = proto;
3421
3422 packet_cached_dev_reset(po);
3423
3424 sk->sk_destruct = packet_sock_destruct;
3425
3426 /*
3427 * Attach a protocol block
3428 */
3429
3430 spin_lock_init(&po->bind_lock);
3431 mutex_init(&po->pg_vec_lock);
3432 po->rollover = NULL;
3433 po->prot_hook.func = packet_rcv;
3434
3435 if (sock->type == SOCK_PACKET)
3436 po->prot_hook.func = packet_rcv_spkt;
3437
3438 po->prot_hook.af_packet_priv = sk;
3439 po->prot_hook.af_packet_net = sock_net(sk);
3440
3441 if (proto) {
3442 po->prot_hook.type = proto;
3443 __register_prot_hook(sk);
3444 }
3445
3446 mutex_lock(&net->packet.sklist_lock);
3447 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3448 mutex_unlock(&net->packet.sklist_lock);
3449
3450 sock_prot_inuse_add(net, &packet_proto, 1);
3451
3452 return 0;
3453out_sk_free:
3454 sk_free(sk);
3455out:
3456 return err;
3457}
3458
3459/*
3460 * Pull a packet from our receive queue and hand it to the user.
3461 * If necessary we block.
3462 */
3463
3464static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3465 int flags)
3466{
3467 struct sock *sk = sock->sk;
3468 struct sk_buff *skb;
3469 int copied, err;
3470 int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3471 unsigned int origlen = 0;
3472
3473 err = -EINVAL;
3474 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3475 goto out;
3476
3477#if 0
3478 /* What error should we return now? EUNATTACH? */
3479 if (pkt_sk(sk)->ifindex < 0)
3480 return -ENODEV;
3481#endif
3482
3483 if (flags & MSG_ERRQUEUE) {
3484 err = sock_recv_errqueue(sk, msg, len,
3485 SOL_PACKET, PACKET_TX_TIMESTAMP);
3486 goto out;
3487 }
3488
3489 /*
3490 * Call the generic datagram receiver. This handles all sorts
3491 * of horrible races and re-entrancy so we can forget about it
3492 * in the protocol layers.
3493 *
3494 * Now it will return ENETDOWN, if device have just gone down,
3495 * but then it will block.
3496 */
3497
3498 skb = skb_recv_datagram(sk, flags, &err);
3499
3500 /*
3501 * An error occurred so return it. Because skb_recv_datagram()
3502 * handles the blocking we don't see and worry about blocking
3503 * retries.
3504 */
3505
3506 if (skb == NULL)
3507 goto out;
3508
3509 packet_rcv_try_clear_pressure(pkt_sk(sk));
3510
3511 if (vnet_hdr_len) {
3512 err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3513 if (err)
3514 goto out_free;
3515 }
3516
3517 /* You lose any data beyond the buffer you gave. If it worries
3518 * a user program they can ask the device for its MTU
3519 * anyway.
3520 */
3521 copied = skb->len;
3522 if (copied > len) {
3523 copied = len;
3524 msg->msg_flags |= MSG_TRUNC;
3525 }
3526
3527 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3528 if (err)
3529 goto out_free;
3530
3531 if (sock->type != SOCK_PACKET) {
3532 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3533
3534 /* Original length was stored in sockaddr_ll fields */
3535 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3536 sll->sll_family = AF_PACKET;
3537 sll->sll_protocol = (sock->type == SOCK_DGRAM) ?
3538 vlan_get_protocol_dgram(skb) : skb->protocol;
3539 }
3540
3541 sock_recv_cmsgs(msg, sk, skb);
3542
3543 if (msg->msg_name) {
3544 const size_t max_len = min(sizeof(skb->cb),
3545 sizeof(struct sockaddr_storage));
3546 int copy_len;
3547
3548 /* If the address length field is there to be filled
3549 * in, we fill it in now.
3550 */
3551 if (sock->type == SOCK_PACKET) {
3552 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3553 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3554 copy_len = msg->msg_namelen;
3555 } else {
3556 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3557
3558 msg->msg_namelen = sll->sll_halen +
3559 offsetof(struct sockaddr_ll, sll_addr);
3560 copy_len = msg->msg_namelen;
3561 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3562 memset(msg->msg_name +
3563 offsetof(struct sockaddr_ll, sll_addr),
3564 0, sizeof(sll->sll_addr));
3565 msg->msg_namelen = sizeof(struct sockaddr_ll);
3566 }
3567 }
3568 if (WARN_ON_ONCE(copy_len > max_len)) {
3569 copy_len = max_len;
3570 msg->msg_namelen = copy_len;
3571 }
3572 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3573 }
3574
3575 if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3576 struct tpacket_auxdata aux;
3577
3578 aux.tp_status = TP_STATUS_USER;
3579 if (skb->ip_summed == CHECKSUM_PARTIAL)
3580 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3581 else if (skb->pkt_type != PACKET_OUTGOING &&
3582 skb_csum_unnecessary(skb))
3583 aux.tp_status |= TP_STATUS_CSUM_VALID;
3584 if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3585 aux.tp_status |= TP_STATUS_GSO_TCP;
3586
3587 aux.tp_len = origlen;
3588 aux.tp_snaplen = skb->len;
3589 aux.tp_mac = 0;
3590 aux.tp_net = skb_network_offset(skb);
3591 if (skb_vlan_tag_present(skb)) {
3592 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3593 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3594 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3595 } else if (unlikely(sock->type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
3596 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3597 struct net_device *dev;
3598
3599 rcu_read_lock();
3600 dev = dev_get_by_index_rcu(sock_net(sk), sll->sll_ifindex);
3601 if (dev) {
3602 aux.tp_vlan_tci = vlan_get_tci(skb, dev);
3603 aux.tp_vlan_tpid = ntohs(skb->protocol);
3604 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3605 } else {
3606 aux.tp_vlan_tci = 0;
3607 aux.tp_vlan_tpid = 0;
3608 }
3609 rcu_read_unlock();
3610 } else {
3611 aux.tp_vlan_tci = 0;
3612 aux.tp_vlan_tpid = 0;
3613 }
3614 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3615 }
3616
3617 /*
3618 * Free or return the buffer as appropriate. Again this
3619 * hides all the races and re-entrancy issues from us.
3620 */
3621 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3622
3623out_free:
3624 skb_free_datagram(sk, skb);
3625out:
3626 return err;
3627}
3628
3629static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3630 int peer)
3631{
3632 struct net_device *dev;
3633 struct sock *sk = sock->sk;
3634
3635 if (peer)
3636 return -EOPNOTSUPP;
3637
3638 uaddr->sa_family = AF_PACKET;
3639 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3640 rcu_read_lock();
3641 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3642 if (dev)
3643 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3644 rcu_read_unlock();
3645
3646 return sizeof(*uaddr);
3647}
3648
3649static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3650 int peer)
3651{
3652 struct net_device *dev;
3653 struct sock *sk = sock->sk;
3654 struct packet_sock *po = pkt_sk(sk);
3655 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3656 int ifindex;
3657
3658 if (peer)
3659 return -EOPNOTSUPP;
3660
3661 ifindex = READ_ONCE(po->ifindex);
3662 sll->sll_family = AF_PACKET;
3663 sll->sll_ifindex = ifindex;
3664 sll->sll_protocol = READ_ONCE(po->num);
3665 sll->sll_pkttype = 0;
3666 rcu_read_lock();
3667 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3668 if (dev) {
3669 sll->sll_hatype = dev->type;
3670 sll->sll_halen = dev->addr_len;
3671
3672 /* Let __fortify_memcpy_chk() know the actual buffer size. */
3673 memcpy(((struct sockaddr_storage *)sll)->__data +
3674 offsetof(struct sockaddr_ll, sll_addr) -
3675 offsetofend(struct sockaddr_ll, sll_family),
3676 dev->dev_addr, dev->addr_len);
3677 } else {
3678 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3679 sll->sll_halen = 0;
3680 }
3681 rcu_read_unlock();
3682
3683 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3684}
3685
3686static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3687 int what)
3688{
3689 switch (i->type) {
3690 case PACKET_MR_MULTICAST:
3691 if (i->alen != dev->addr_len)
3692 return -EINVAL;
3693 if (what > 0)
3694 return dev_mc_add(dev, i->addr);
3695 else
3696 return dev_mc_del(dev, i->addr);
3697 break;
3698 case PACKET_MR_PROMISC:
3699 return dev_set_promiscuity(dev, what);
3700 case PACKET_MR_ALLMULTI:
3701 return dev_set_allmulti(dev, what);
3702 case PACKET_MR_UNICAST:
3703 if (i->alen != dev->addr_len)
3704 return -EINVAL;
3705 if (what > 0)
3706 return dev_uc_add(dev, i->addr);
3707 else
3708 return dev_uc_del(dev, i->addr);
3709 break;
3710 default:
3711 break;
3712 }
3713 return 0;
3714}
3715
3716static void packet_dev_mclist_delete(struct net_device *dev,
3717 struct packet_mclist **mlp)
3718{
3719 struct packet_mclist *ml;
3720
3721 while ((ml = *mlp) != NULL) {
3722 if (ml->ifindex == dev->ifindex) {
3723 packet_dev_mc(dev, ml, -1);
3724 *mlp = ml->next;
3725 kfree(ml);
3726 } else
3727 mlp = &ml->next;
3728 }
3729}
3730
3731static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3732{
3733 struct packet_sock *po = pkt_sk(sk);
3734 struct packet_mclist *ml, *i;
3735 struct net_device *dev;
3736 int err;
3737
3738 rtnl_lock();
3739
3740 err = -ENODEV;
3741 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3742 if (!dev)
3743 goto done;
3744
3745 err = -EINVAL;
3746 if (mreq->mr_alen > dev->addr_len)
3747 goto done;
3748
3749 err = -ENOBUFS;
3750 i = kmalloc(sizeof(*i), GFP_KERNEL);
3751 if (i == NULL)
3752 goto done;
3753
3754 err = 0;
3755 for (ml = po->mclist; ml; ml = ml->next) {
3756 if (ml->ifindex == mreq->mr_ifindex &&
3757 ml->type == mreq->mr_type &&
3758 ml->alen == mreq->mr_alen &&
3759 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3760 ml->count++;
3761 /* Free the new element ... */
3762 kfree(i);
3763 goto done;
3764 }
3765 }
3766
3767 i->type = mreq->mr_type;
3768 i->ifindex = mreq->mr_ifindex;
3769 i->alen = mreq->mr_alen;
3770 memcpy(i->addr, mreq->mr_address, i->alen);
3771 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3772 i->count = 1;
3773 i->next = po->mclist;
3774 po->mclist = i;
3775 err = packet_dev_mc(dev, i, 1);
3776 if (err) {
3777 po->mclist = i->next;
3778 kfree(i);
3779 }
3780
3781done:
3782 rtnl_unlock();
3783 return err;
3784}
3785
3786static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3787{
3788 struct packet_mclist *ml, **mlp;
3789
3790 rtnl_lock();
3791
3792 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3793 if (ml->ifindex == mreq->mr_ifindex &&
3794 ml->type == mreq->mr_type &&
3795 ml->alen == mreq->mr_alen &&
3796 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3797 if (--ml->count == 0) {
3798 struct net_device *dev;
3799 *mlp = ml->next;
3800 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3801 if (dev)
3802 packet_dev_mc(dev, ml, -1);
3803 kfree(ml);
3804 }
3805 break;
3806 }
3807 }
3808 rtnl_unlock();
3809 return 0;
3810}
3811
3812static void packet_flush_mclist(struct sock *sk)
3813{
3814 struct packet_sock *po = pkt_sk(sk);
3815 struct packet_mclist *ml;
3816
3817 if (!po->mclist)
3818 return;
3819
3820 rtnl_lock();
3821 while ((ml = po->mclist) != NULL) {
3822 struct net_device *dev;
3823
3824 po->mclist = ml->next;
3825 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3826 if (dev != NULL)
3827 packet_dev_mc(dev, ml, -1);
3828 kfree(ml);
3829 }
3830 rtnl_unlock();
3831}
3832
3833static int
3834packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3835 unsigned int optlen)
3836{
3837 struct sock *sk = sock->sk;
3838 struct packet_sock *po = pkt_sk(sk);
3839 int ret;
3840
3841 if (level != SOL_PACKET)
3842 return -ENOPROTOOPT;
3843
3844 switch (optname) {
3845 case PACKET_ADD_MEMBERSHIP:
3846 case PACKET_DROP_MEMBERSHIP:
3847 {
3848 struct packet_mreq_max mreq;
3849 int len = optlen;
3850 memset(&mreq, 0, sizeof(mreq));
3851 if (len < sizeof(struct packet_mreq))
3852 return -EINVAL;
3853 if (len > sizeof(mreq))
3854 len = sizeof(mreq);
3855 if (copy_from_sockptr(&mreq, optval, len))
3856 return -EFAULT;
3857 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3858 return -EINVAL;
3859 if (optname == PACKET_ADD_MEMBERSHIP)
3860 ret = packet_mc_add(sk, &mreq);
3861 else
3862 ret = packet_mc_drop(sk, &mreq);
3863 return ret;
3864 }
3865
3866 case PACKET_RX_RING:
3867 case PACKET_TX_RING:
3868 {
3869 union tpacket_req_u req_u;
3870
3871 ret = -EINVAL;
3872 lock_sock(sk);
3873 switch (po->tp_version) {
3874 case TPACKET_V1:
3875 case TPACKET_V2:
3876 if (optlen < sizeof(req_u.req))
3877 break;
3878 ret = copy_from_sockptr(&req_u.req, optval,
3879 sizeof(req_u.req)) ?
3880 -EINVAL : 0;
3881 break;
3882 case TPACKET_V3:
3883 default:
3884 if (optlen < sizeof(req_u.req3))
3885 break;
3886 ret = copy_from_sockptr(&req_u.req3, optval,
3887 sizeof(req_u.req3)) ?
3888 -EINVAL : 0;
3889 break;
3890 }
3891 if (!ret)
3892 ret = packet_set_ring(sk, &req_u, 0,
3893 optname == PACKET_TX_RING);
3894 release_sock(sk);
3895 return ret;
3896 }
3897 case PACKET_COPY_THRESH:
3898 {
3899 int val;
3900
3901 if (optlen != sizeof(val))
3902 return -EINVAL;
3903 if (copy_from_sockptr(&val, optval, sizeof(val)))
3904 return -EFAULT;
3905
3906 WRITE_ONCE(pkt_sk(sk)->copy_thresh, val);
3907 return 0;
3908 }
3909 case PACKET_VERSION:
3910 {
3911 int val;
3912
3913 if (optlen != sizeof(val))
3914 return -EINVAL;
3915 if (copy_from_sockptr(&val, optval, sizeof(val)))
3916 return -EFAULT;
3917 switch (val) {
3918 case TPACKET_V1:
3919 case TPACKET_V2:
3920 case TPACKET_V3:
3921 break;
3922 default:
3923 return -EINVAL;
3924 }
3925 lock_sock(sk);
3926 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3927 ret = -EBUSY;
3928 } else {
3929 po->tp_version = val;
3930 ret = 0;
3931 }
3932 release_sock(sk);
3933 return ret;
3934 }
3935 case PACKET_RESERVE:
3936 {
3937 unsigned int val;
3938
3939 if (optlen != sizeof(val))
3940 return -EINVAL;
3941 if (copy_from_sockptr(&val, optval, sizeof(val)))
3942 return -EFAULT;
3943 if (val > INT_MAX)
3944 return -EINVAL;
3945 lock_sock(sk);
3946 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3947 ret = -EBUSY;
3948 } else {
3949 po->tp_reserve = val;
3950 ret = 0;
3951 }
3952 release_sock(sk);
3953 return ret;
3954 }
3955 case PACKET_LOSS:
3956 {
3957 unsigned int val;
3958
3959 if (optlen != sizeof(val))
3960 return -EINVAL;
3961 if (copy_from_sockptr(&val, optval, sizeof(val)))
3962 return -EFAULT;
3963
3964 lock_sock(sk);
3965 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3966 ret = -EBUSY;
3967 } else {
3968 packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3969 ret = 0;
3970 }
3971 release_sock(sk);
3972 return ret;
3973 }
3974 case PACKET_AUXDATA:
3975 {
3976 int val;
3977
3978 if (optlen < sizeof(val))
3979 return -EINVAL;
3980 if (copy_from_sockptr(&val, optval, sizeof(val)))
3981 return -EFAULT;
3982
3983 packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3984 return 0;
3985 }
3986 case PACKET_ORIGDEV:
3987 {
3988 int val;
3989
3990 if (optlen < sizeof(val))
3991 return -EINVAL;
3992 if (copy_from_sockptr(&val, optval, sizeof(val)))
3993 return -EFAULT;
3994
3995 packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3996 return 0;
3997 }
3998 case PACKET_VNET_HDR:
3999 case PACKET_VNET_HDR_SZ:
4000 {
4001 int val, hdr_len;
4002
4003 if (sock->type != SOCK_RAW)
4004 return -EINVAL;
4005 if (optlen < sizeof(val))
4006 return -EINVAL;
4007 if (copy_from_sockptr(&val, optval, sizeof(val)))
4008 return -EFAULT;
4009
4010 if (optname == PACKET_VNET_HDR_SZ) {
4011 if (val && val != sizeof(struct virtio_net_hdr) &&
4012 val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
4013 return -EINVAL;
4014 hdr_len = val;
4015 } else {
4016 hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
4017 }
4018 lock_sock(sk);
4019 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
4020 ret = -EBUSY;
4021 } else {
4022 WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
4023 ret = 0;
4024 }
4025 release_sock(sk);
4026 return ret;
4027 }
4028 case PACKET_TIMESTAMP:
4029 {
4030 int val;
4031
4032 if (optlen != sizeof(val))
4033 return -EINVAL;
4034 if (copy_from_sockptr(&val, optval, sizeof(val)))
4035 return -EFAULT;
4036
4037 WRITE_ONCE(po->tp_tstamp, val);
4038 return 0;
4039 }
4040 case PACKET_FANOUT:
4041 {
4042 struct fanout_args args = { 0 };
4043
4044 if (optlen != sizeof(int) && optlen != sizeof(args))
4045 return -EINVAL;
4046 if (copy_from_sockptr(&args, optval, optlen))
4047 return -EFAULT;
4048
4049 return fanout_add(sk, &args);
4050 }
4051 case PACKET_FANOUT_DATA:
4052 {
4053 /* Paired with the WRITE_ONCE() in fanout_add() */
4054 if (!READ_ONCE(po->fanout))
4055 return -EINVAL;
4056
4057 return fanout_set_data(po, optval, optlen);
4058 }
4059 case PACKET_IGNORE_OUTGOING:
4060 {
4061 int val;
4062
4063 if (optlen != sizeof(val))
4064 return -EINVAL;
4065 if (copy_from_sockptr(&val, optval, sizeof(val)))
4066 return -EFAULT;
4067 if (val < 0 || val > 1)
4068 return -EINVAL;
4069
4070 WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val);
4071 return 0;
4072 }
4073 case PACKET_TX_HAS_OFF:
4074 {
4075 unsigned int val;
4076
4077 if (optlen != sizeof(val))
4078 return -EINVAL;
4079 if (copy_from_sockptr(&val, optval, sizeof(val)))
4080 return -EFAULT;
4081
4082 lock_sock(sk);
4083 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4084 packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4085
4086 release_sock(sk);
4087 return 0;
4088 }
4089 case PACKET_QDISC_BYPASS:
4090 {
4091 int val;
4092
4093 if (optlen != sizeof(val))
4094 return -EINVAL;
4095 if (copy_from_sockptr(&val, optval, sizeof(val)))
4096 return -EFAULT;
4097
4098 packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4099 return 0;
4100 }
4101 default:
4102 return -ENOPROTOOPT;
4103 }
4104}
4105
4106static int packet_getsockopt(struct socket *sock, int level, int optname,
4107 char __user *optval, int __user *optlen)
4108{
4109 int len;
4110 int val, lv = sizeof(val);
4111 struct sock *sk = sock->sk;
4112 struct packet_sock *po = pkt_sk(sk);
4113 void *data = &val;
4114 union tpacket_stats_u st;
4115 struct tpacket_rollover_stats rstats;
4116 int drops;
4117
4118 if (level != SOL_PACKET)
4119 return -ENOPROTOOPT;
4120
4121 if (get_user(len, optlen))
4122 return -EFAULT;
4123
4124 if (len < 0)
4125 return -EINVAL;
4126
4127 switch (optname) {
4128 case PACKET_STATISTICS:
4129 spin_lock_bh(&sk->sk_receive_queue.lock);
4130 memcpy(&st, &po->stats, sizeof(st));
4131 memset(&po->stats, 0, sizeof(po->stats));
4132 spin_unlock_bh(&sk->sk_receive_queue.lock);
4133 drops = atomic_xchg(&po->tp_drops, 0);
4134
4135 if (po->tp_version == TPACKET_V3) {
4136 lv = sizeof(struct tpacket_stats_v3);
4137 st.stats3.tp_drops = drops;
4138 st.stats3.tp_packets += drops;
4139 data = &st.stats3;
4140 } else {
4141 lv = sizeof(struct tpacket_stats);
4142 st.stats1.tp_drops = drops;
4143 st.stats1.tp_packets += drops;
4144 data = &st.stats1;
4145 }
4146
4147 break;
4148 case PACKET_AUXDATA:
4149 val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4150 break;
4151 case PACKET_ORIGDEV:
4152 val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4153 break;
4154 case PACKET_VNET_HDR:
4155 val = !!READ_ONCE(po->vnet_hdr_sz);
4156 break;
4157 case PACKET_VNET_HDR_SZ:
4158 val = READ_ONCE(po->vnet_hdr_sz);
4159 break;
4160 case PACKET_COPY_THRESH:
4161 val = READ_ONCE(pkt_sk(sk)->copy_thresh);
4162 break;
4163 case PACKET_VERSION:
4164 val = po->tp_version;
4165 break;
4166 case PACKET_HDRLEN:
4167 if (len > sizeof(int))
4168 len = sizeof(int);
4169 if (len < sizeof(int))
4170 return -EINVAL;
4171 if (copy_from_user(&val, optval, len))
4172 return -EFAULT;
4173 switch (val) {
4174 case TPACKET_V1:
4175 val = sizeof(struct tpacket_hdr);
4176 break;
4177 case TPACKET_V2:
4178 val = sizeof(struct tpacket2_hdr);
4179 break;
4180 case TPACKET_V3:
4181 val = sizeof(struct tpacket3_hdr);
4182 break;
4183 default:
4184 return -EINVAL;
4185 }
4186 break;
4187 case PACKET_RESERVE:
4188 val = po->tp_reserve;
4189 break;
4190 case PACKET_LOSS:
4191 val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4192 break;
4193 case PACKET_TIMESTAMP:
4194 val = READ_ONCE(po->tp_tstamp);
4195 break;
4196 case PACKET_FANOUT:
4197 val = (po->fanout ?
4198 ((u32)po->fanout->id |
4199 ((u32)po->fanout->type << 16) |
4200 ((u32)po->fanout->flags << 24)) :
4201 0);
4202 break;
4203 case PACKET_IGNORE_OUTGOING:
4204 val = READ_ONCE(po->prot_hook.ignore_outgoing);
4205 break;
4206 case PACKET_ROLLOVER_STATS:
4207 if (!po->rollover)
4208 return -EINVAL;
4209 rstats.tp_all = atomic_long_read(&po->rollover->num);
4210 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4211 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4212 data = &rstats;
4213 lv = sizeof(rstats);
4214 break;
4215 case PACKET_TX_HAS_OFF:
4216 val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4217 break;
4218 case PACKET_QDISC_BYPASS:
4219 val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4220 break;
4221 default:
4222 return -ENOPROTOOPT;
4223 }
4224
4225 if (len > lv)
4226 len = lv;
4227 if (put_user(len, optlen))
4228 return -EFAULT;
4229 if (copy_to_user(optval, data, len))
4230 return -EFAULT;
4231 return 0;
4232}
4233
4234static int packet_notifier(struct notifier_block *this,
4235 unsigned long msg, void *ptr)
4236{
4237 struct sock *sk;
4238 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4239 struct net *net = dev_net(dev);
4240
4241 rcu_read_lock();
4242 sk_for_each_rcu(sk, &net->packet.sklist) {
4243 struct packet_sock *po = pkt_sk(sk);
4244
4245 switch (msg) {
4246 case NETDEV_UNREGISTER:
4247 if (po->mclist)
4248 packet_dev_mclist_delete(dev, &po->mclist);
4249 fallthrough;
4250
4251 case NETDEV_DOWN:
4252 if (dev->ifindex == po->ifindex) {
4253 spin_lock(&po->bind_lock);
4254 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4255 __unregister_prot_hook(sk, false);
4256 sk->sk_err = ENETDOWN;
4257 if (!sock_flag(sk, SOCK_DEAD))
4258 sk_error_report(sk);
4259 }
4260 if (msg == NETDEV_UNREGISTER) {
4261 packet_cached_dev_reset(po);
4262 WRITE_ONCE(po->ifindex, -1);
4263 netdev_put(po->prot_hook.dev,
4264 &po->prot_hook.dev_tracker);
4265 po->prot_hook.dev = NULL;
4266 }
4267 spin_unlock(&po->bind_lock);
4268 }
4269 break;
4270 case NETDEV_UP:
4271 if (dev->ifindex == po->ifindex) {
4272 spin_lock(&po->bind_lock);
4273 if (po->num)
4274 register_prot_hook(sk);
4275 spin_unlock(&po->bind_lock);
4276 }
4277 break;
4278 }
4279 }
4280 rcu_read_unlock();
4281 return NOTIFY_DONE;
4282}
4283
4284
4285static int packet_ioctl(struct socket *sock, unsigned int cmd,
4286 unsigned long arg)
4287{
4288 struct sock *sk = sock->sk;
4289
4290 switch (cmd) {
4291 case SIOCOUTQ:
4292 {
4293 int amount = sk_wmem_alloc_get(sk);
4294
4295 return put_user(amount, (int __user *)arg);
4296 }
4297 case SIOCINQ:
4298 {
4299 struct sk_buff *skb;
4300 int amount = 0;
4301
4302 spin_lock_bh(&sk->sk_receive_queue.lock);
4303 skb = skb_peek(&sk->sk_receive_queue);
4304 if (skb)
4305 amount = skb->len;
4306 spin_unlock_bh(&sk->sk_receive_queue.lock);
4307 return put_user(amount, (int __user *)arg);
4308 }
4309#ifdef CONFIG_INET
4310 case SIOCADDRT:
4311 case SIOCDELRT:
4312 case SIOCDARP:
4313 case SIOCGARP:
4314 case SIOCSARP:
4315 case SIOCGIFADDR:
4316 case SIOCSIFADDR:
4317 case SIOCGIFBRDADDR:
4318 case SIOCSIFBRDADDR:
4319 case SIOCGIFNETMASK:
4320 case SIOCSIFNETMASK:
4321 case SIOCGIFDSTADDR:
4322 case SIOCSIFDSTADDR:
4323 case SIOCSIFFLAGS:
4324 return inet_dgram_ops.ioctl(sock, cmd, arg);
4325#endif
4326
4327 default:
4328 return -ENOIOCTLCMD;
4329 }
4330 return 0;
4331}
4332
4333static __poll_t packet_poll(struct file *file, struct socket *sock,
4334 poll_table *wait)
4335{
4336 struct sock *sk = sock->sk;
4337 struct packet_sock *po = pkt_sk(sk);
4338 __poll_t mask = datagram_poll(file, sock, wait);
4339
4340 spin_lock_bh(&sk->sk_receive_queue.lock);
4341 if (po->rx_ring.pg_vec) {
4342 if (!packet_previous_rx_frame(po, &po->rx_ring,
4343 TP_STATUS_KERNEL))
4344 mask |= EPOLLIN | EPOLLRDNORM;
4345 }
4346 packet_rcv_try_clear_pressure(po);
4347 spin_unlock_bh(&sk->sk_receive_queue.lock);
4348 spin_lock_bh(&sk->sk_write_queue.lock);
4349 if (po->tx_ring.pg_vec) {
4350 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4351 mask |= EPOLLOUT | EPOLLWRNORM;
4352 }
4353 spin_unlock_bh(&sk->sk_write_queue.lock);
4354 return mask;
4355}
4356
4357
4358/* Dirty? Well, I still did not learn better way to account
4359 * for user mmaps.
4360 */
4361
4362static void packet_mm_open(struct vm_area_struct *vma)
4363{
4364 struct file *file = vma->vm_file;
4365 struct socket *sock = file->private_data;
4366 struct sock *sk = sock->sk;
4367
4368 if (sk)
4369 atomic_long_inc(&pkt_sk(sk)->mapped);
4370}
4371
4372static void packet_mm_close(struct vm_area_struct *vma)
4373{
4374 struct file *file = vma->vm_file;
4375 struct socket *sock = file->private_data;
4376 struct sock *sk = sock->sk;
4377
4378 if (sk)
4379 atomic_long_dec(&pkt_sk(sk)->mapped);
4380}
4381
4382static const struct vm_operations_struct packet_mmap_ops = {
4383 .open = packet_mm_open,
4384 .close = packet_mm_close,
4385};
4386
4387static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4388 unsigned int len)
4389{
4390 int i;
4391
4392 for (i = 0; i < len; i++) {
4393 if (likely(pg_vec[i].buffer)) {
4394 if (is_vmalloc_addr(pg_vec[i].buffer))
4395 vfree(pg_vec[i].buffer);
4396 else
4397 free_pages((unsigned long)pg_vec[i].buffer,
4398 order);
4399 pg_vec[i].buffer = NULL;
4400 }
4401 }
4402 kfree(pg_vec);
4403}
4404
4405static char *alloc_one_pg_vec_page(unsigned long order)
4406{
4407 char *buffer;
4408 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4409 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4410
4411 buffer = (char *) __get_free_pages(gfp_flags, order);
4412 if (buffer)
4413 return buffer;
4414
4415 /* __get_free_pages failed, fall back to vmalloc */
4416 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4417 if (buffer)
4418 return buffer;
4419
4420 /* vmalloc failed, lets dig into swap here */
4421 gfp_flags &= ~__GFP_NORETRY;
4422 buffer = (char *) __get_free_pages(gfp_flags, order);
4423 if (buffer)
4424 return buffer;
4425
4426 /* complete and utter failure */
4427 return NULL;
4428}
4429
4430static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4431{
4432 unsigned int block_nr = req->tp_block_nr;
4433 struct pgv *pg_vec;
4434 int i;
4435
4436 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4437 if (unlikely(!pg_vec))
4438 goto out;
4439
4440 for (i = 0; i < block_nr; i++) {
4441 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4442 if (unlikely(!pg_vec[i].buffer))
4443 goto out_free_pgvec;
4444 }
4445
4446out:
4447 return pg_vec;
4448
4449out_free_pgvec:
4450 free_pg_vec(pg_vec, order, block_nr);
4451 pg_vec = NULL;
4452 goto out;
4453}
4454
4455static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4456 int closing, int tx_ring)
4457{
4458 struct pgv *pg_vec = NULL;
4459 struct packet_sock *po = pkt_sk(sk);
4460 unsigned long *rx_owner_map = NULL;
4461 int was_running, order = 0;
4462 struct packet_ring_buffer *rb;
4463 struct sk_buff_head *rb_queue;
4464 __be16 num;
4465 int err;
4466 /* Added to avoid minimal code churn */
4467 struct tpacket_req *req = &req_u->req;
4468
4469 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4470 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4471
4472 err = -EBUSY;
4473 if (!closing) {
4474 if (atomic_long_read(&po->mapped))
4475 goto out;
4476 if (packet_read_pending(rb))
4477 goto out;
4478 }
4479
4480 if (req->tp_block_nr) {
4481 unsigned int min_frame_size;
4482
4483 /* Sanity tests and some calculations */
4484 err = -EBUSY;
4485 if (unlikely(rb->pg_vec))
4486 goto out;
4487
4488 switch (po->tp_version) {
4489 case TPACKET_V1:
4490 po->tp_hdrlen = TPACKET_HDRLEN;
4491 break;
4492 case TPACKET_V2:
4493 po->tp_hdrlen = TPACKET2_HDRLEN;
4494 break;
4495 case TPACKET_V3:
4496 po->tp_hdrlen = TPACKET3_HDRLEN;
4497 break;
4498 }
4499
4500 err = -EINVAL;
4501 if (unlikely((int)req->tp_block_size <= 0))
4502 goto out;
4503 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4504 goto out;
4505 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4506 if (po->tp_version >= TPACKET_V3 &&
4507 req->tp_block_size <
4508 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4509 goto out;
4510 if (unlikely(req->tp_frame_size < min_frame_size))
4511 goto out;
4512 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4513 goto out;
4514
4515 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4516 if (unlikely(rb->frames_per_block == 0))
4517 goto out;
4518 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4519 goto out;
4520 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4521 req->tp_frame_nr))
4522 goto out;
4523
4524 err = -ENOMEM;
4525 order = get_order(req->tp_block_size);
4526 pg_vec = alloc_pg_vec(req, order);
4527 if (unlikely(!pg_vec))
4528 goto out;
4529 switch (po->tp_version) {
4530 case TPACKET_V3:
4531 /* Block transmit is not supported yet */
4532 if (!tx_ring) {
4533 init_prb_bdqc(po, rb, pg_vec, req_u);
4534 } else {
4535 struct tpacket_req3 *req3 = &req_u->req3;
4536
4537 if (req3->tp_retire_blk_tov ||
4538 req3->tp_sizeof_priv ||
4539 req3->tp_feature_req_word) {
4540 err = -EINVAL;
4541 goto out_free_pg_vec;
4542 }
4543 }
4544 break;
4545 default:
4546 if (!tx_ring) {
4547 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4548 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4549 if (!rx_owner_map)
4550 goto out_free_pg_vec;
4551 }
4552 break;
4553 }
4554 }
4555 /* Done */
4556 else {
4557 err = -EINVAL;
4558 if (unlikely(req->tp_frame_nr))
4559 goto out;
4560 }
4561
4562
4563 /* Detach socket from network */
4564 spin_lock(&po->bind_lock);
4565 was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4566 num = po->num;
4567 if (was_running) {
4568 WRITE_ONCE(po->num, 0);
4569 __unregister_prot_hook(sk, false);
4570 }
4571 spin_unlock(&po->bind_lock);
4572
4573 synchronize_net();
4574
4575 err = -EBUSY;
4576 mutex_lock(&po->pg_vec_lock);
4577 if (closing || atomic_long_read(&po->mapped) == 0) {
4578 err = 0;
4579 spin_lock_bh(&rb_queue->lock);
4580 swap(rb->pg_vec, pg_vec);
4581 if (po->tp_version <= TPACKET_V2)
4582 swap(rb->rx_owner_map, rx_owner_map);
4583 rb->frame_max = (req->tp_frame_nr - 1);
4584 rb->head = 0;
4585 rb->frame_size = req->tp_frame_size;
4586 spin_unlock_bh(&rb_queue->lock);
4587
4588 swap(rb->pg_vec_order, order);
4589 swap(rb->pg_vec_len, req->tp_block_nr);
4590
4591 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4592 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4593 tpacket_rcv : packet_rcv;
4594 skb_queue_purge(rb_queue);
4595 if (atomic_long_read(&po->mapped))
4596 pr_err("packet_mmap: vma is busy: %ld\n",
4597 atomic_long_read(&po->mapped));
4598 }
4599 mutex_unlock(&po->pg_vec_lock);
4600
4601 spin_lock(&po->bind_lock);
4602 if (was_running) {
4603 WRITE_ONCE(po->num, num);
4604 register_prot_hook(sk);
4605 }
4606 spin_unlock(&po->bind_lock);
4607 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4608 /* Because we don't support block-based V3 on tx-ring */
4609 if (!tx_ring)
4610 prb_shutdown_retire_blk_timer(po, rb_queue);
4611 }
4612
4613out_free_pg_vec:
4614 if (pg_vec) {
4615 bitmap_free(rx_owner_map);
4616 free_pg_vec(pg_vec, order, req->tp_block_nr);
4617 }
4618out:
4619 return err;
4620}
4621
4622static int packet_mmap(struct file *file, struct socket *sock,
4623 struct vm_area_struct *vma)
4624{
4625 struct sock *sk = sock->sk;
4626 struct packet_sock *po = pkt_sk(sk);
4627 unsigned long size, expected_size;
4628 struct packet_ring_buffer *rb;
4629 unsigned long start;
4630 int err = -EINVAL;
4631 int i;
4632
4633 if (vma->vm_pgoff)
4634 return -EINVAL;
4635
4636 mutex_lock(&po->pg_vec_lock);
4637
4638 expected_size = 0;
4639 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4640 if (rb->pg_vec) {
4641 expected_size += rb->pg_vec_len
4642 * rb->pg_vec_pages
4643 * PAGE_SIZE;
4644 }
4645 }
4646
4647 if (expected_size == 0)
4648 goto out;
4649
4650 size = vma->vm_end - vma->vm_start;
4651 if (size != expected_size)
4652 goto out;
4653
4654 start = vma->vm_start;
4655 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4656 if (rb->pg_vec == NULL)
4657 continue;
4658
4659 for (i = 0; i < rb->pg_vec_len; i++) {
4660 struct page *page;
4661 void *kaddr = rb->pg_vec[i].buffer;
4662 int pg_num;
4663
4664 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4665 page = pgv_to_page(kaddr);
4666 err = vm_insert_page(vma, start, page);
4667 if (unlikely(err))
4668 goto out;
4669 start += PAGE_SIZE;
4670 kaddr += PAGE_SIZE;
4671 }
4672 }
4673 }
4674
4675 atomic_long_inc(&po->mapped);
4676 vma->vm_ops = &packet_mmap_ops;
4677 err = 0;
4678
4679out:
4680 mutex_unlock(&po->pg_vec_lock);
4681 return err;
4682}
4683
4684static const struct proto_ops packet_ops_spkt = {
4685 .family = PF_PACKET,
4686 .owner = THIS_MODULE,
4687 .release = packet_release,
4688 .bind = packet_bind_spkt,
4689 .connect = sock_no_connect,
4690 .socketpair = sock_no_socketpair,
4691 .accept = sock_no_accept,
4692 .getname = packet_getname_spkt,
4693 .poll = datagram_poll,
4694 .ioctl = packet_ioctl,
4695 .gettstamp = sock_gettstamp,
4696 .listen = sock_no_listen,
4697 .shutdown = sock_no_shutdown,
4698 .sendmsg = packet_sendmsg_spkt,
4699 .recvmsg = packet_recvmsg,
4700 .mmap = sock_no_mmap,
4701};
4702
4703static const struct proto_ops packet_ops = {
4704 .family = PF_PACKET,
4705 .owner = THIS_MODULE,
4706 .release = packet_release,
4707 .bind = packet_bind,
4708 .connect = sock_no_connect,
4709 .socketpair = sock_no_socketpair,
4710 .accept = sock_no_accept,
4711 .getname = packet_getname,
4712 .poll = packet_poll,
4713 .ioctl = packet_ioctl,
4714 .gettstamp = sock_gettstamp,
4715 .listen = sock_no_listen,
4716 .shutdown = sock_no_shutdown,
4717 .setsockopt = packet_setsockopt,
4718 .getsockopt = packet_getsockopt,
4719 .sendmsg = packet_sendmsg,
4720 .recvmsg = packet_recvmsg,
4721 .mmap = packet_mmap,
4722};
4723
4724static const struct net_proto_family packet_family_ops = {
4725 .family = PF_PACKET,
4726 .create = packet_create,
4727 .owner = THIS_MODULE,
4728};
4729
4730static struct notifier_block packet_netdev_notifier = {
4731 .notifier_call = packet_notifier,
4732};
4733
4734#ifdef CONFIG_PROC_FS
4735
4736static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4737 __acquires(RCU)
4738{
4739 struct net *net = seq_file_net(seq);
4740
4741 rcu_read_lock();
4742 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4743}
4744
4745static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4746{
4747 struct net *net = seq_file_net(seq);
4748 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4749}
4750
4751static void packet_seq_stop(struct seq_file *seq, void *v)
4752 __releases(RCU)
4753{
4754 rcu_read_unlock();
4755}
4756
4757static int packet_seq_show(struct seq_file *seq, void *v)
4758{
4759 if (v == SEQ_START_TOKEN)
4760 seq_printf(seq,
4761 "%*sRefCnt Type Proto Iface R Rmem User Inode\n",
4762 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4763 else {
4764 struct sock *s = sk_entry(v);
4765 const struct packet_sock *po = pkt_sk(s);
4766
4767 seq_printf(seq,
4768 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4769 s,
4770 refcount_read(&s->sk_refcnt),
4771 s->sk_type,
4772 ntohs(READ_ONCE(po->num)),
4773 READ_ONCE(po->ifindex),
4774 packet_sock_flag(po, PACKET_SOCK_RUNNING),
4775 atomic_read(&s->sk_rmem_alloc),
4776 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4777 sock_i_ino(s));
4778 }
4779
4780 return 0;
4781}
4782
4783static const struct seq_operations packet_seq_ops = {
4784 .start = packet_seq_start,
4785 .next = packet_seq_next,
4786 .stop = packet_seq_stop,
4787 .show = packet_seq_show,
4788};
4789#endif
4790
4791static int __net_init packet_net_init(struct net *net)
4792{
4793 mutex_init(&net->packet.sklist_lock);
4794 INIT_HLIST_HEAD(&net->packet.sklist);
4795
4796#ifdef CONFIG_PROC_FS
4797 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4798 sizeof(struct seq_net_private)))
4799 return -ENOMEM;
4800#endif /* CONFIG_PROC_FS */
4801
4802 return 0;
4803}
4804
4805static void __net_exit packet_net_exit(struct net *net)
4806{
4807 remove_proc_entry("packet", net->proc_net);
4808 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4809}
4810
4811static struct pernet_operations packet_net_ops = {
4812 .init = packet_net_init,
4813 .exit = packet_net_exit,
4814};
4815
4816
4817static void __exit packet_exit(void)
4818{
4819 sock_unregister(PF_PACKET);
4820 proto_unregister(&packet_proto);
4821 unregister_netdevice_notifier(&packet_netdev_notifier);
4822 unregister_pernet_subsys(&packet_net_ops);
4823}
4824
4825static int __init packet_init(void)
4826{
4827 int rc;
4828
4829 rc = register_pernet_subsys(&packet_net_ops);
4830 if (rc)
4831 goto out;
4832 rc = register_netdevice_notifier(&packet_netdev_notifier);
4833 if (rc)
4834 goto out_pernet;
4835 rc = proto_register(&packet_proto, 0);
4836 if (rc)
4837 goto out_notifier;
4838 rc = sock_register(&packet_family_ops);
4839 if (rc)
4840 goto out_proto;
4841
4842 return 0;
4843
4844out_proto:
4845 proto_unregister(&packet_proto);
4846out_notifier:
4847 unregister_netdevice_notifier(&packet_netdev_notifier);
4848out_pernet:
4849 unregister_pernet_subsys(&packet_net_ops);
4850out:
4851 return rc;
4852}
4853
4854module_init(packet_init);
4855module_exit(packet_exit);
4856MODULE_DESCRIPTION("Packet socket support (AF_PACKET)");
4857MODULE_LICENSE("GPL");
4858MODULE_ALIAS_NETPROTO(PF_PACKET);
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PACKET - implements raw packet sockets.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 *
12 * Fixes:
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
41 * and packet_mreq.
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
44 * layer.
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
46 *
47 *
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
52 *
53 */
54
55#include <linux/types.h>
56#include <linux/mm.h>
57#include <linux/capability.h>
58#include <linux/fcntl.h>
59#include <linux/socket.h>
60#include <linux/in.h>
61#include <linux/inet.h>
62#include <linux/netdevice.h>
63#include <linux/if_packet.h>
64#include <linux/wireless.h>
65#include <linux/kernel.h>
66#include <linux/kmod.h>
67#include <linux/slab.h>
68#include <linux/vmalloc.h>
69#include <net/net_namespace.h>
70#include <net/ip.h>
71#include <net/protocol.h>
72#include <linux/skbuff.h>
73#include <net/sock.h>
74#include <linux/errno.h>
75#include <linux/timer.h>
76#include <asm/uaccess.h>
77#include <asm/ioctls.h>
78#include <asm/page.h>
79#include <asm/cacheflush.h>
80#include <asm/io.h>
81#include <linux/proc_fs.h>
82#include <linux/seq_file.h>
83#include <linux/poll.h>
84#include <linux/module.h>
85#include <linux/init.h>
86#include <linux/mutex.h>
87#include <linux/if_vlan.h>
88#include <linux/virtio_net.h>
89#include <linux/errqueue.h>
90#include <linux/net_tstamp.h>
91
92#ifdef CONFIG_INET
93#include <net/inet_common.h>
94#endif
95
96/*
97 Assumptions:
98 - if device has no dev->hard_header routine, it adds and removes ll header
99 inside itself. In this case ll header is invisible outside of device,
100 but higher levels still should reserve dev->hard_header_len.
101 Some devices are enough clever to reallocate skb, when header
102 will not fit to reserved space (tunnel), another ones are silly
103 (PPP).
104 - packet socket receives packets with pulled ll header,
105 so that SOCK_RAW should push it back.
106
107On receive:
108-----------
109
110Incoming, dev->hard_header!=NULL
111 mac_header -> ll header
112 data -> data
113
114Outgoing, dev->hard_header!=NULL
115 mac_header -> ll header
116 data -> ll header
117
118Incoming, dev->hard_header==NULL
119 mac_header -> UNKNOWN position. It is very likely, that it points to ll
120 header. PPP makes it, that is wrong, because introduce
121 assymetry between rx and tx paths.
122 data -> data
123
124Outgoing, dev->hard_header==NULL
125 mac_header -> data. ll header is still not built!
126 data -> data
127
128Resume
129 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
130
131
132On transmit:
133------------
134
135dev->hard_header != NULL
136 mac_header -> ll header
137 data -> ll header
138
139dev->hard_header == NULL (ll header is added by device, we cannot control it)
140 mac_header -> data
141 data -> data
142
143 We should set nh.raw on output to correct posistion,
144 packet classifier depends on it.
145 */
146
147/* Private packet socket structures. */
148
149struct packet_mclist {
150 struct packet_mclist *next;
151 int ifindex;
152 int count;
153 unsigned short type;
154 unsigned short alen;
155 unsigned char addr[MAX_ADDR_LEN];
156};
157/* identical to struct packet_mreq except it has
158 * a longer address field.
159 */
160struct packet_mreq_max {
161 int mr_ifindex;
162 unsigned short mr_type;
163 unsigned short mr_alen;
164 unsigned char mr_address[MAX_ADDR_LEN];
165};
166
167static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
168 int closing, int tx_ring);
169
170
171#define V3_ALIGNMENT (8)
172
173#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
174
175#define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
177
178/* kbdq - kernel block descriptor queue */
179struct tpacket_kbdq_core {
180 struct pgv *pkbdq;
181 unsigned int feature_req_word;
182 unsigned int hdrlen;
183 unsigned char reset_pending_on_curr_blk;
184 unsigned char delete_blk_timer;
185 unsigned short kactive_blk_num;
186 unsigned short blk_sizeof_priv;
187
188 /* last_kactive_blk_num:
189 * trick to see if user-space has caught up
190 * in order to avoid refreshing timer when every single pkt arrives.
191 */
192 unsigned short last_kactive_blk_num;
193
194 char *pkblk_start;
195 char *pkblk_end;
196 int kblk_size;
197 unsigned int knum_blocks;
198 uint64_t knxt_seq_num;
199 char *prev;
200 char *nxt_offset;
201 struct sk_buff *skb;
202
203 atomic_t blk_fill_in_prog;
204
205 /* Default is set to 8ms */
206#define DEFAULT_PRB_RETIRE_TOV (8)
207
208 unsigned short retire_blk_tov;
209 unsigned short version;
210 unsigned long tov_in_jiffies;
211
212 /* timer to retire an outstanding block */
213 struct timer_list retire_blk_timer;
214};
215
216#define PGV_FROM_VMALLOC 1
217struct pgv {
218 char *buffer;
219};
220
221struct packet_ring_buffer {
222 struct pgv *pg_vec;
223 unsigned int head;
224 unsigned int frames_per_block;
225 unsigned int frame_size;
226 unsigned int frame_max;
227
228 unsigned int pg_vec_order;
229 unsigned int pg_vec_pages;
230 unsigned int pg_vec_len;
231
232 struct tpacket_kbdq_core prb_bdqc;
233 atomic_t pending;
234};
235
236#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
237#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
238#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
239#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
240#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
241#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
242#define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
243
244struct packet_sock;
245static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
246
247static void *packet_previous_frame(struct packet_sock *po,
248 struct packet_ring_buffer *rb,
249 int status);
250static void packet_increment_head(struct packet_ring_buffer *buff);
251static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
252 struct tpacket_block_desc *);
253static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
254 struct packet_sock *);
255static void prb_retire_current_block(struct tpacket_kbdq_core *,
256 struct packet_sock *, unsigned int status);
257static int prb_queue_frozen(struct tpacket_kbdq_core *);
258static void prb_open_block(struct tpacket_kbdq_core *,
259 struct tpacket_block_desc *);
260static void prb_retire_rx_blk_timer_expired(unsigned long);
261static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
262static void prb_init_blk_timer(struct packet_sock *,
263 struct tpacket_kbdq_core *,
264 void (*func) (unsigned long));
265static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
266static void prb_clear_rxhash(struct tpacket_kbdq_core *,
267 struct tpacket3_hdr *);
268static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
269 struct tpacket3_hdr *);
270static void packet_flush_mclist(struct sock *sk);
271
272struct packet_fanout;
273struct packet_sock {
274 /* struct sock has to be the first member of packet_sock */
275 struct sock sk;
276 struct packet_fanout *fanout;
277 struct tpacket_stats stats;
278 union tpacket_stats_u stats_u;
279 struct packet_ring_buffer rx_ring;
280 struct packet_ring_buffer tx_ring;
281 int copy_thresh;
282 spinlock_t bind_lock;
283 struct mutex pg_vec_lock;
284 unsigned int running:1, /* prot_hook is attached*/
285 auxdata:1,
286 origdev:1,
287 has_vnet_hdr:1;
288 int ifindex; /* bound device */
289 __be16 num;
290 struct packet_mclist *mclist;
291 atomic_t mapped;
292 enum tpacket_versions tp_version;
293 unsigned int tp_hdrlen;
294 unsigned int tp_reserve;
295 unsigned int tp_loss:1;
296 unsigned int tp_tstamp;
297 struct packet_type prot_hook ____cacheline_aligned_in_smp;
298};
299
300#define PACKET_FANOUT_MAX 256
301
302struct packet_fanout {
303#ifdef CONFIG_NET_NS
304 struct net *net;
305#endif
306 unsigned int num_members;
307 u16 id;
308 u8 type;
309 u8 defrag;
310 atomic_t rr_cur;
311 struct list_head list;
312 struct sock *arr[PACKET_FANOUT_MAX];
313 spinlock_t lock;
314 atomic_t sk_ref;
315 struct packet_type prot_hook ____cacheline_aligned_in_smp;
316};
317
318struct packet_skb_cb {
319 unsigned int origlen;
320 union {
321 struct sockaddr_pkt pkt;
322 struct sockaddr_ll ll;
323 } sa;
324};
325
326#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
327
328#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
329#define GET_PBLOCK_DESC(x, bid) \
330 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
331#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
332 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
333#define GET_NEXT_PRB_BLK_NUM(x) \
334 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
335 ((x)->kactive_blk_num+1) : 0)
336
337static struct packet_sock *pkt_sk(struct sock *sk)
338{
339 return (struct packet_sock *)sk;
340}
341
342static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
343static void __fanout_link(struct sock *sk, struct packet_sock *po);
344
345/* register_prot_hook must be invoked with the po->bind_lock held,
346 * or from a context in which asynchronous accesses to the packet
347 * socket is not possible (packet_create()).
348 */
349static void register_prot_hook(struct sock *sk)
350{
351 struct packet_sock *po = pkt_sk(sk);
352 if (!po->running) {
353 if (po->fanout)
354 __fanout_link(sk, po);
355 else
356 dev_add_pack(&po->prot_hook);
357 sock_hold(sk);
358 po->running = 1;
359 }
360}
361
362/* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
363 * held. If the sync parameter is true, we will temporarily drop
364 * the po->bind_lock and do a synchronize_net to make sure no
365 * asynchronous packet processing paths still refer to the elements
366 * of po->prot_hook. If the sync parameter is false, it is the
367 * callers responsibility to take care of this.
368 */
369static void __unregister_prot_hook(struct sock *sk, bool sync)
370{
371 struct packet_sock *po = pkt_sk(sk);
372
373 po->running = 0;
374 if (po->fanout)
375 __fanout_unlink(sk, po);
376 else
377 __dev_remove_pack(&po->prot_hook);
378 __sock_put(sk);
379
380 if (sync) {
381 spin_unlock(&po->bind_lock);
382 synchronize_net();
383 spin_lock(&po->bind_lock);
384 }
385}
386
387static void unregister_prot_hook(struct sock *sk, bool sync)
388{
389 struct packet_sock *po = pkt_sk(sk);
390
391 if (po->running)
392 __unregister_prot_hook(sk, sync);
393}
394
395static inline __pure struct page *pgv_to_page(void *addr)
396{
397 if (is_vmalloc_addr(addr))
398 return vmalloc_to_page(addr);
399 return virt_to_page(addr);
400}
401
402static void __packet_set_status(struct packet_sock *po, void *frame, int status)
403{
404 union {
405 struct tpacket_hdr *h1;
406 struct tpacket2_hdr *h2;
407 void *raw;
408 } h;
409
410 h.raw = frame;
411 switch (po->tp_version) {
412 case TPACKET_V1:
413 h.h1->tp_status = status;
414 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
415 break;
416 case TPACKET_V2:
417 h.h2->tp_status = status;
418 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
419 break;
420 case TPACKET_V3:
421 default:
422 WARN(1, "TPACKET version not supported.\n");
423 BUG();
424 }
425
426 smp_wmb();
427}
428
429static int __packet_get_status(struct packet_sock *po, void *frame)
430{
431 union {
432 struct tpacket_hdr *h1;
433 struct tpacket2_hdr *h2;
434 void *raw;
435 } h;
436
437 smp_rmb();
438
439 h.raw = frame;
440 switch (po->tp_version) {
441 case TPACKET_V1:
442 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
443 return h.h1->tp_status;
444 case TPACKET_V2:
445 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
446 return h.h2->tp_status;
447 case TPACKET_V3:
448 default:
449 WARN(1, "TPACKET version not supported.\n");
450 BUG();
451 return 0;
452 }
453}
454
455static void *packet_lookup_frame(struct packet_sock *po,
456 struct packet_ring_buffer *rb,
457 unsigned int position,
458 int status)
459{
460 unsigned int pg_vec_pos, frame_offset;
461 union {
462 struct tpacket_hdr *h1;
463 struct tpacket2_hdr *h2;
464 void *raw;
465 } h;
466
467 pg_vec_pos = position / rb->frames_per_block;
468 frame_offset = position % rb->frames_per_block;
469
470 h.raw = rb->pg_vec[pg_vec_pos].buffer +
471 (frame_offset * rb->frame_size);
472
473 if (status != __packet_get_status(po, h.raw))
474 return NULL;
475
476 return h.raw;
477}
478
479static void *packet_current_frame(struct packet_sock *po,
480 struct packet_ring_buffer *rb,
481 int status)
482{
483 return packet_lookup_frame(po, rb, rb->head, status);
484}
485
486static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
487{
488 del_timer_sync(&pkc->retire_blk_timer);
489}
490
491static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
492 int tx_ring,
493 struct sk_buff_head *rb_queue)
494{
495 struct tpacket_kbdq_core *pkc;
496
497 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
498
499 spin_lock(&rb_queue->lock);
500 pkc->delete_blk_timer = 1;
501 spin_unlock(&rb_queue->lock);
502
503 prb_del_retire_blk_timer(pkc);
504}
505
506static void prb_init_blk_timer(struct packet_sock *po,
507 struct tpacket_kbdq_core *pkc,
508 void (*func) (unsigned long))
509{
510 init_timer(&pkc->retire_blk_timer);
511 pkc->retire_blk_timer.data = (long)po;
512 pkc->retire_blk_timer.function = func;
513 pkc->retire_blk_timer.expires = jiffies;
514}
515
516static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
517{
518 struct tpacket_kbdq_core *pkc;
519
520 if (tx_ring)
521 BUG();
522
523 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
524 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
525}
526
527static int prb_calc_retire_blk_tmo(struct packet_sock *po,
528 int blk_size_in_bytes)
529{
530 struct net_device *dev;
531 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
532 struct ethtool_cmd ecmd;
533 int err;
534
535 rtnl_lock();
536 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
537 if (unlikely(!dev)) {
538 rtnl_unlock();
539 return DEFAULT_PRB_RETIRE_TOV;
540 }
541 err = __ethtool_get_settings(dev, &ecmd);
542 rtnl_unlock();
543 if (!err) {
544 switch (ecmd.speed) {
545 case SPEED_10000:
546 msec = 1;
547 div = 10000/1000;
548 break;
549 case SPEED_1000:
550 msec = 1;
551 div = 1000/1000;
552 break;
553 /*
554 * If the link speed is so slow you don't really
555 * need to worry about perf anyways
556 */
557 case SPEED_100:
558 case SPEED_10:
559 default:
560 return DEFAULT_PRB_RETIRE_TOV;
561 }
562 }
563
564 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
565
566 if (div)
567 mbits /= div;
568
569 tmo = mbits * msec;
570
571 if (div)
572 return tmo+1;
573 return tmo;
574}
575
576static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
577 union tpacket_req_u *req_u)
578{
579 p1->feature_req_word = req_u->req3.tp_feature_req_word;
580}
581
582static void init_prb_bdqc(struct packet_sock *po,
583 struct packet_ring_buffer *rb,
584 struct pgv *pg_vec,
585 union tpacket_req_u *req_u, int tx_ring)
586{
587 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
588 struct tpacket_block_desc *pbd;
589
590 memset(p1, 0x0, sizeof(*p1));
591
592 p1->knxt_seq_num = 1;
593 p1->pkbdq = pg_vec;
594 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
595 p1->pkblk_start = (char *)pg_vec[0].buffer;
596 p1->kblk_size = req_u->req3.tp_block_size;
597 p1->knum_blocks = req_u->req3.tp_block_nr;
598 p1->hdrlen = po->tp_hdrlen;
599 p1->version = po->tp_version;
600 p1->last_kactive_blk_num = 0;
601 po->stats_u.stats3.tp_freeze_q_cnt = 0;
602 if (req_u->req3.tp_retire_blk_tov)
603 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
604 else
605 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
606 req_u->req3.tp_block_size);
607 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
608 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
609
610 prb_init_ft_ops(p1, req_u);
611 prb_setup_retire_blk_timer(po, tx_ring);
612 prb_open_block(p1, pbd);
613}
614
615/* Do NOT update the last_blk_num first.
616 * Assumes sk_buff_head lock is held.
617 */
618static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
619{
620 mod_timer(&pkc->retire_blk_timer,
621 jiffies + pkc->tov_in_jiffies);
622 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
623}
624
625/*
626 * Timer logic:
627 * 1) We refresh the timer only when we open a block.
628 * By doing this we don't waste cycles refreshing the timer
629 * on packet-by-packet basis.
630 *
631 * With a 1MB block-size, on a 1Gbps line, it will take
632 * i) ~8 ms to fill a block + ii) memcpy etc.
633 * In this cut we are not accounting for the memcpy time.
634 *
635 * So, if the user sets the 'tmo' to 10ms then the timer
636 * will never fire while the block is still getting filled
637 * (which is what we want). However, the user could choose
638 * to close a block early and that's fine.
639 *
640 * But when the timer does fire, we check whether or not to refresh it.
641 * Since the tmo granularity is in msecs, it is not too expensive
642 * to refresh the timer, lets say every '8' msecs.
643 * Either the user can set the 'tmo' or we can derive it based on
644 * a) line-speed and b) block-size.
645 * prb_calc_retire_blk_tmo() calculates the tmo.
646 *
647 */
648static void prb_retire_rx_blk_timer_expired(unsigned long data)
649{
650 struct packet_sock *po = (struct packet_sock *)data;
651 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
652 unsigned int frozen;
653 struct tpacket_block_desc *pbd;
654
655 spin_lock(&po->sk.sk_receive_queue.lock);
656
657 frozen = prb_queue_frozen(pkc);
658 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
659
660 if (unlikely(pkc->delete_blk_timer))
661 goto out;
662
663 /* We only need to plug the race when the block is partially filled.
664 * tpacket_rcv:
665 * lock(); increment BLOCK_NUM_PKTS; unlock()
666 * copy_bits() is in progress ...
667 * timer fires on other cpu:
668 * we can't retire the current block because copy_bits
669 * is in progress.
670 *
671 */
672 if (BLOCK_NUM_PKTS(pbd)) {
673 while (atomic_read(&pkc->blk_fill_in_prog)) {
674 /* Waiting for skb_copy_bits to finish... */
675 cpu_relax();
676 }
677 }
678
679 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
680 if (!frozen) {
681 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
682 if (!prb_dispatch_next_block(pkc, po))
683 goto refresh_timer;
684 else
685 goto out;
686 } else {
687 /* Case 1. Queue was frozen because user-space was
688 * lagging behind.
689 */
690 if (prb_curr_blk_in_use(pkc, pbd)) {
691 /*
692 * Ok, user-space is still behind.
693 * So just refresh the timer.
694 */
695 goto refresh_timer;
696 } else {
697 /* Case 2. queue was frozen,user-space caught up,
698 * now the link went idle && the timer fired.
699 * We don't have a block to close.So we open this
700 * block and restart the timer.
701 * opening a block thaws the queue,restarts timer
702 * Thawing/timer-refresh is a side effect.
703 */
704 prb_open_block(pkc, pbd);
705 goto out;
706 }
707 }
708 }
709
710refresh_timer:
711 _prb_refresh_rx_retire_blk_timer(pkc);
712
713out:
714 spin_unlock(&po->sk.sk_receive_queue.lock);
715}
716
717static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
718 struct tpacket_block_desc *pbd1, __u32 status)
719{
720 /* Flush everything minus the block header */
721
722#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
723 u8 *start, *end;
724
725 start = (u8 *)pbd1;
726
727 /* Skip the block header(we know header WILL fit in 4K) */
728 start += PAGE_SIZE;
729
730 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
731 for (; start < end; start += PAGE_SIZE)
732 flush_dcache_page(pgv_to_page(start));
733
734 smp_wmb();
735#endif
736
737 /* Now update the block status. */
738
739 BLOCK_STATUS(pbd1) = status;
740
741 /* Flush the block header */
742
743#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
744 start = (u8 *)pbd1;
745 flush_dcache_page(pgv_to_page(start));
746
747 smp_wmb();
748#endif
749}
750
751/*
752 * Side effect:
753 *
754 * 1) flush the block
755 * 2) Increment active_blk_num
756 *
757 * Note:We DONT refresh the timer on purpose.
758 * Because almost always the next block will be opened.
759 */
760static void prb_close_block(struct tpacket_kbdq_core *pkc1,
761 struct tpacket_block_desc *pbd1,
762 struct packet_sock *po, unsigned int stat)
763{
764 __u32 status = TP_STATUS_USER | stat;
765
766 struct tpacket3_hdr *last_pkt;
767 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
768
769 if (po->stats.tp_drops)
770 status |= TP_STATUS_LOSING;
771
772 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
773 last_pkt->tp_next_offset = 0;
774
775 /* Get the ts of the last pkt */
776 if (BLOCK_NUM_PKTS(pbd1)) {
777 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
778 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
779 } else {
780 /* Ok, we tmo'd - so get the current time */
781 struct timespec ts;
782 getnstimeofday(&ts);
783 h1->ts_last_pkt.ts_sec = ts.tv_sec;
784 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
785 }
786
787 smp_wmb();
788
789 /* Flush the block */
790 prb_flush_block(pkc1, pbd1, status);
791
792 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
793}
794
795static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
796{
797 pkc->reset_pending_on_curr_blk = 0;
798}
799
800/*
801 * Side effect of opening a block:
802 *
803 * 1) prb_queue is thawed.
804 * 2) retire_blk_timer is refreshed.
805 *
806 */
807static void prb_open_block(struct tpacket_kbdq_core *pkc1,
808 struct tpacket_block_desc *pbd1)
809{
810 struct timespec ts;
811 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
812
813 smp_rmb();
814
815 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
816
817 /* We could have just memset this but we will lose the
818 * flexibility of making the priv area sticky
819 */
820 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
821 BLOCK_NUM_PKTS(pbd1) = 0;
822 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
823 getnstimeofday(&ts);
824 h1->ts_first_pkt.ts_sec = ts.tv_sec;
825 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
826 pkc1->pkblk_start = (char *)pbd1;
827 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
828 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
829 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
831 pbd1->version = pkc1->version;
832 pkc1->prev = pkc1->nxt_offset;
833 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
834 prb_thaw_queue(pkc1);
835 _prb_refresh_rx_retire_blk_timer(pkc1);
836
837 smp_wmb();
838
839 return;
840 }
841
842 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
843 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
844 dump_stack();
845 BUG();
846}
847
848/*
849 * Queue freeze logic:
850 * 1) Assume tp_block_nr = 8 blocks.
851 * 2) At time 't0', user opens Rx ring.
852 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
853 * 4) user-space is either sleeping or processing block '0'.
854 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
855 * it will close block-7,loop around and try to fill block '0'.
856 * call-flow:
857 * __packet_lookup_frame_in_block
858 * prb_retire_current_block()
859 * prb_dispatch_next_block()
860 * |->(BLOCK_STATUS == USER) evaluates to true
861 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
862 * 6) Now there are two cases:
863 * 6.1) Link goes idle right after the queue is frozen.
864 * But remember, the last open_block() refreshed the timer.
865 * When this timer expires,it will refresh itself so that we can
866 * re-open block-0 in near future.
867 * 6.2) Link is busy and keeps on receiving packets. This is a simple
868 * case and __packet_lookup_frame_in_block will check if block-0
869 * is free and can now be re-used.
870 */
871static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
872 struct packet_sock *po)
873{
874 pkc->reset_pending_on_curr_blk = 1;
875 po->stats_u.stats3.tp_freeze_q_cnt++;
876}
877
878#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
879
880/*
881 * If the next block is free then we will dispatch it
882 * and return a good offset.
883 * Else, we will freeze the queue.
884 * So, caller must check the return value.
885 */
886static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
887 struct packet_sock *po)
888{
889 struct tpacket_block_desc *pbd;
890
891 smp_rmb();
892
893 /* 1. Get current block num */
894 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
895
896 /* 2. If this block is currently in_use then freeze the queue */
897 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
898 prb_freeze_queue(pkc, po);
899 return NULL;
900 }
901
902 /*
903 * 3.
904 * open this block and return the offset where the first packet
905 * needs to get stored.
906 */
907 prb_open_block(pkc, pbd);
908 return (void *)pkc->nxt_offset;
909}
910
911static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
912 struct packet_sock *po, unsigned int status)
913{
914 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
915
916 /* retire/close the current block */
917 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
918 /*
919 * Plug the case where copy_bits() is in progress on
920 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
921 * have space to copy the pkt in the current block and
922 * called prb_retire_current_block()
923 *
924 * We don't need to worry about the TMO case because
925 * the timer-handler already handled this case.
926 */
927 if (!(status & TP_STATUS_BLK_TMO)) {
928 while (atomic_read(&pkc->blk_fill_in_prog)) {
929 /* Waiting for skb_copy_bits to finish... */
930 cpu_relax();
931 }
932 }
933 prb_close_block(pkc, pbd, po, status);
934 return;
935 }
936
937 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
938 dump_stack();
939 BUG();
940}
941
942static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
943 struct tpacket_block_desc *pbd)
944{
945 return TP_STATUS_USER & BLOCK_STATUS(pbd);
946}
947
948static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
949{
950 return pkc->reset_pending_on_curr_blk;
951}
952
953static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
954{
955 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
956 atomic_dec(&pkc->blk_fill_in_prog);
957}
958
959static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
960 struct tpacket3_hdr *ppd)
961{
962 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
963}
964
965static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
966 struct tpacket3_hdr *ppd)
967{
968 ppd->hv1.tp_rxhash = 0;
969}
970
971static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
972 struct tpacket3_hdr *ppd)
973{
974 if (vlan_tx_tag_present(pkc->skb)) {
975 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
976 ppd->tp_status = TP_STATUS_VLAN_VALID;
977 } else {
978 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
979 }
980}
981
982static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
983 struct tpacket3_hdr *ppd)
984{
985 prb_fill_vlan_info(pkc, ppd);
986
987 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
988 prb_fill_rxhash(pkc, ppd);
989 else
990 prb_clear_rxhash(pkc, ppd);
991}
992
993static void prb_fill_curr_block(char *curr,
994 struct tpacket_kbdq_core *pkc,
995 struct tpacket_block_desc *pbd,
996 unsigned int len)
997{
998 struct tpacket3_hdr *ppd;
999
1000 ppd = (struct tpacket3_hdr *)curr;
1001 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1002 pkc->prev = curr;
1003 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1004 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 BLOCK_NUM_PKTS(pbd) += 1;
1006 atomic_inc(&pkc->blk_fill_in_prog);
1007 prb_run_all_ft_ops(pkc, ppd);
1008}
1009
1010/* Assumes caller has the sk->rx_queue.lock */
1011static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1012 struct sk_buff *skb,
1013 int status,
1014 unsigned int len
1015 )
1016{
1017 struct tpacket_kbdq_core *pkc;
1018 struct tpacket_block_desc *pbd;
1019 char *curr, *end;
1020
1021 pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1022 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1023
1024 /* Queue is frozen when user space is lagging behind */
1025 if (prb_queue_frozen(pkc)) {
1026 /*
1027 * Check if that last block which caused the queue to freeze,
1028 * is still in_use by user-space.
1029 */
1030 if (prb_curr_blk_in_use(pkc, pbd)) {
1031 /* Can't record this packet */
1032 return NULL;
1033 } else {
1034 /*
1035 * Ok, the block was released by user-space.
1036 * Now let's open that block.
1037 * opening a block also thaws the queue.
1038 * Thawing is a side effect.
1039 */
1040 prb_open_block(pkc, pbd);
1041 }
1042 }
1043
1044 smp_mb();
1045 curr = pkc->nxt_offset;
1046 pkc->skb = skb;
1047 end = (char *) ((char *)pbd + pkc->kblk_size);
1048
1049 /* first try the current block */
1050 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1051 prb_fill_curr_block(curr, pkc, pbd, len);
1052 return (void *)curr;
1053 }
1054
1055 /* Ok, close the current block */
1056 prb_retire_current_block(pkc, po, 0);
1057
1058 /* Now, try to dispatch the next block */
1059 curr = (char *)prb_dispatch_next_block(pkc, po);
1060 if (curr) {
1061 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1062 prb_fill_curr_block(curr, pkc, pbd, len);
1063 return (void *)curr;
1064 }
1065
1066 /*
1067 * No free blocks are available.user_space hasn't caught up yet.
1068 * Queue was just frozen and now this packet will get dropped.
1069 */
1070 return NULL;
1071}
1072
1073static void *packet_current_rx_frame(struct packet_sock *po,
1074 struct sk_buff *skb,
1075 int status, unsigned int len)
1076{
1077 char *curr = NULL;
1078 switch (po->tp_version) {
1079 case TPACKET_V1:
1080 case TPACKET_V2:
1081 curr = packet_lookup_frame(po, &po->rx_ring,
1082 po->rx_ring.head, status);
1083 return curr;
1084 case TPACKET_V3:
1085 return __packet_lookup_frame_in_block(po, skb, status, len);
1086 default:
1087 WARN(1, "TPACKET version not supported\n");
1088 BUG();
1089 return 0;
1090 }
1091}
1092
1093static void *prb_lookup_block(struct packet_sock *po,
1094 struct packet_ring_buffer *rb,
1095 unsigned int previous,
1096 int status)
1097{
1098 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1099 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1100
1101 if (status != BLOCK_STATUS(pbd))
1102 return NULL;
1103 return pbd;
1104}
1105
1106static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1107{
1108 unsigned int prev;
1109 if (rb->prb_bdqc.kactive_blk_num)
1110 prev = rb->prb_bdqc.kactive_blk_num-1;
1111 else
1112 prev = rb->prb_bdqc.knum_blocks-1;
1113 return prev;
1114}
1115
1116/* Assumes caller has held the rx_queue.lock */
1117static void *__prb_previous_block(struct packet_sock *po,
1118 struct packet_ring_buffer *rb,
1119 int status)
1120{
1121 unsigned int previous = prb_previous_blk_num(rb);
1122 return prb_lookup_block(po, rb, previous, status);
1123}
1124
1125static void *packet_previous_rx_frame(struct packet_sock *po,
1126 struct packet_ring_buffer *rb,
1127 int status)
1128{
1129 if (po->tp_version <= TPACKET_V2)
1130 return packet_previous_frame(po, rb, status);
1131
1132 return __prb_previous_block(po, rb, status);
1133}
1134
1135static void packet_increment_rx_head(struct packet_sock *po,
1136 struct packet_ring_buffer *rb)
1137{
1138 switch (po->tp_version) {
1139 case TPACKET_V1:
1140 case TPACKET_V2:
1141 return packet_increment_head(rb);
1142 case TPACKET_V3:
1143 default:
1144 WARN(1, "TPACKET version not supported.\n");
1145 BUG();
1146 return;
1147 }
1148}
1149
1150static void *packet_previous_frame(struct packet_sock *po,
1151 struct packet_ring_buffer *rb,
1152 int status)
1153{
1154 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1155 return packet_lookup_frame(po, rb, previous, status);
1156}
1157
1158static void packet_increment_head(struct packet_ring_buffer *buff)
1159{
1160 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1161}
1162
1163static void packet_sock_destruct(struct sock *sk)
1164{
1165 skb_queue_purge(&sk->sk_error_queue);
1166
1167 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1168 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1169
1170 if (!sock_flag(sk, SOCK_DEAD)) {
1171 pr_err("Attempt to release alive packet socket: %p\n", sk);
1172 return;
1173 }
1174
1175 sk_refcnt_debug_dec(sk);
1176}
1177
1178static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1179{
1180 int x = atomic_read(&f->rr_cur) + 1;
1181
1182 if (x >= num)
1183 x = 0;
1184
1185 return x;
1186}
1187
1188static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1189{
1190 u32 idx, hash = skb->rxhash;
1191
1192 idx = ((u64)hash * num) >> 32;
1193
1194 return f->arr[idx];
1195}
1196
1197static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1198{
1199 int cur, old;
1200
1201 cur = atomic_read(&f->rr_cur);
1202 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1203 fanout_rr_next(f, num))) != cur)
1204 cur = old;
1205 return f->arr[cur];
1206}
1207
1208static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1209{
1210 unsigned int cpu = smp_processor_id();
1211
1212 return f->arr[cpu % num];
1213}
1214
1215static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1216 struct packet_type *pt, struct net_device *orig_dev)
1217{
1218 struct packet_fanout *f = pt->af_packet_priv;
1219 unsigned int num = f->num_members;
1220 struct packet_sock *po;
1221 struct sock *sk;
1222
1223 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1224 !num) {
1225 kfree_skb(skb);
1226 return 0;
1227 }
1228
1229 switch (f->type) {
1230 case PACKET_FANOUT_HASH:
1231 default:
1232 if (f->defrag) {
1233 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1234 if (!skb)
1235 return 0;
1236 }
1237 skb_get_rxhash(skb);
1238 sk = fanout_demux_hash(f, skb, num);
1239 break;
1240 case PACKET_FANOUT_LB:
1241 sk = fanout_demux_lb(f, skb, num);
1242 break;
1243 case PACKET_FANOUT_CPU:
1244 sk = fanout_demux_cpu(f, skb, num);
1245 break;
1246 }
1247
1248 po = pkt_sk(sk);
1249
1250 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1251}
1252
1253static DEFINE_MUTEX(fanout_mutex);
1254static LIST_HEAD(fanout_list);
1255
1256static void __fanout_link(struct sock *sk, struct packet_sock *po)
1257{
1258 struct packet_fanout *f = po->fanout;
1259
1260 spin_lock(&f->lock);
1261 f->arr[f->num_members] = sk;
1262 smp_wmb();
1263 f->num_members++;
1264 spin_unlock(&f->lock);
1265}
1266
1267static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1268{
1269 struct packet_fanout *f = po->fanout;
1270 int i;
1271
1272 spin_lock(&f->lock);
1273 for (i = 0; i < f->num_members; i++) {
1274 if (f->arr[i] == sk)
1275 break;
1276 }
1277 BUG_ON(i >= f->num_members);
1278 f->arr[i] = f->arr[f->num_members - 1];
1279 f->num_members--;
1280 spin_unlock(&f->lock);
1281}
1282
1283bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1284{
1285 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1286 return true;
1287
1288 return false;
1289}
1290
1291static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1292{
1293 struct packet_sock *po = pkt_sk(sk);
1294 struct packet_fanout *f, *match;
1295 u8 type = type_flags & 0xff;
1296 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1297 int err;
1298
1299 switch (type) {
1300 case PACKET_FANOUT_HASH:
1301 case PACKET_FANOUT_LB:
1302 case PACKET_FANOUT_CPU:
1303 break;
1304 default:
1305 return -EINVAL;
1306 }
1307
1308 if (!po->running)
1309 return -EINVAL;
1310
1311 if (po->fanout)
1312 return -EALREADY;
1313
1314 mutex_lock(&fanout_mutex);
1315 match = NULL;
1316 list_for_each_entry(f, &fanout_list, list) {
1317 if (f->id == id &&
1318 read_pnet(&f->net) == sock_net(sk)) {
1319 match = f;
1320 break;
1321 }
1322 }
1323 err = -EINVAL;
1324 if (match && match->defrag != defrag)
1325 goto out;
1326 if (!match) {
1327 err = -ENOMEM;
1328 match = kzalloc(sizeof(*match), GFP_KERNEL);
1329 if (!match)
1330 goto out;
1331 write_pnet(&match->net, sock_net(sk));
1332 match->id = id;
1333 match->type = type;
1334 match->defrag = defrag;
1335 atomic_set(&match->rr_cur, 0);
1336 INIT_LIST_HEAD(&match->list);
1337 spin_lock_init(&match->lock);
1338 atomic_set(&match->sk_ref, 0);
1339 match->prot_hook.type = po->prot_hook.type;
1340 match->prot_hook.dev = po->prot_hook.dev;
1341 match->prot_hook.func = packet_rcv_fanout;
1342 match->prot_hook.af_packet_priv = match;
1343 match->prot_hook.id_match = match_fanout_group;
1344 dev_add_pack(&match->prot_hook);
1345 list_add(&match->list, &fanout_list);
1346 }
1347 err = -EINVAL;
1348 if (match->type == type &&
1349 match->prot_hook.type == po->prot_hook.type &&
1350 match->prot_hook.dev == po->prot_hook.dev) {
1351 err = -ENOSPC;
1352 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1353 __dev_remove_pack(&po->prot_hook);
1354 po->fanout = match;
1355 atomic_inc(&match->sk_ref);
1356 __fanout_link(sk, po);
1357 err = 0;
1358 }
1359 }
1360out:
1361 mutex_unlock(&fanout_mutex);
1362 return err;
1363}
1364
1365static void fanout_release(struct sock *sk)
1366{
1367 struct packet_sock *po = pkt_sk(sk);
1368 struct packet_fanout *f;
1369
1370 f = po->fanout;
1371 if (!f)
1372 return;
1373
1374 po->fanout = NULL;
1375
1376 mutex_lock(&fanout_mutex);
1377 if (atomic_dec_and_test(&f->sk_ref)) {
1378 list_del(&f->list);
1379 dev_remove_pack(&f->prot_hook);
1380 kfree(f);
1381 }
1382 mutex_unlock(&fanout_mutex);
1383}
1384
1385static const struct proto_ops packet_ops;
1386
1387static const struct proto_ops packet_ops_spkt;
1388
1389static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1390 struct packet_type *pt, struct net_device *orig_dev)
1391{
1392 struct sock *sk;
1393 struct sockaddr_pkt *spkt;
1394
1395 /*
1396 * When we registered the protocol we saved the socket in the data
1397 * field for just this event.
1398 */
1399
1400 sk = pt->af_packet_priv;
1401
1402 /*
1403 * Yank back the headers [hope the device set this
1404 * right or kerboom...]
1405 *
1406 * Incoming packets have ll header pulled,
1407 * push it back.
1408 *
1409 * For outgoing ones skb->data == skb_mac_header(skb)
1410 * so that this procedure is noop.
1411 */
1412
1413 if (skb->pkt_type == PACKET_LOOPBACK)
1414 goto out;
1415
1416 if (!net_eq(dev_net(dev), sock_net(sk)))
1417 goto out;
1418
1419 skb = skb_share_check(skb, GFP_ATOMIC);
1420 if (skb == NULL)
1421 goto oom;
1422
1423 /* drop any routing info */
1424 skb_dst_drop(skb);
1425
1426 /* drop conntrack reference */
1427 nf_reset(skb);
1428
1429 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1430
1431 skb_push(skb, skb->data - skb_mac_header(skb));
1432
1433 /*
1434 * The SOCK_PACKET socket receives _all_ frames.
1435 */
1436
1437 spkt->spkt_family = dev->type;
1438 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1439 spkt->spkt_protocol = skb->protocol;
1440
1441 /*
1442 * Charge the memory to the socket. This is done specifically
1443 * to prevent sockets using all the memory up.
1444 */
1445
1446 if (sock_queue_rcv_skb(sk, skb) == 0)
1447 return 0;
1448
1449out:
1450 kfree_skb(skb);
1451oom:
1452 return 0;
1453}
1454
1455
1456/*
1457 * Output a raw packet to a device layer. This bypasses all the other
1458 * protocol layers and you must therefore supply it with a complete frame
1459 */
1460
1461static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1462 struct msghdr *msg, size_t len)
1463{
1464 struct sock *sk = sock->sk;
1465 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1466 struct sk_buff *skb = NULL;
1467 struct net_device *dev;
1468 __be16 proto = 0;
1469 int err;
1470 int extra_len = 0;
1471
1472 /*
1473 * Get and verify the address.
1474 */
1475
1476 if (saddr) {
1477 if (msg->msg_namelen < sizeof(struct sockaddr))
1478 return -EINVAL;
1479 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1480 proto = saddr->spkt_protocol;
1481 } else
1482 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1483
1484 /*
1485 * Find the device first to size check it
1486 */
1487
1488 saddr->spkt_device[13] = 0;
1489retry:
1490 rcu_read_lock();
1491 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1492 err = -ENODEV;
1493 if (dev == NULL)
1494 goto out_unlock;
1495
1496 err = -ENETDOWN;
1497 if (!(dev->flags & IFF_UP))
1498 goto out_unlock;
1499
1500 /*
1501 * You may not queue a frame bigger than the mtu. This is the lowest level
1502 * raw protocol and you must do your own fragmentation at this level.
1503 */
1504
1505 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1506 if (!netif_supports_nofcs(dev)) {
1507 err = -EPROTONOSUPPORT;
1508 goto out_unlock;
1509 }
1510 extra_len = 4; /* We're doing our own CRC */
1511 }
1512
1513 err = -EMSGSIZE;
1514 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1515 goto out_unlock;
1516
1517 if (!skb) {
1518 size_t reserved = LL_RESERVED_SPACE(dev);
1519 int tlen = dev->needed_tailroom;
1520 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1521
1522 rcu_read_unlock();
1523 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1524 if (skb == NULL)
1525 return -ENOBUFS;
1526 /* FIXME: Save some space for broken drivers that write a hard
1527 * header at transmission time by themselves. PPP is the notable
1528 * one here. This should really be fixed at the driver level.
1529 */
1530 skb_reserve(skb, reserved);
1531 skb_reset_network_header(skb);
1532
1533 /* Try to align data part correctly */
1534 if (hhlen) {
1535 skb->data -= hhlen;
1536 skb->tail -= hhlen;
1537 if (len < hhlen)
1538 skb_reset_network_header(skb);
1539 }
1540 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1541 if (err)
1542 goto out_free;
1543 goto retry;
1544 }
1545
1546 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1547 /* Earlier code assumed this would be a VLAN pkt,
1548 * double-check this now that we have the actual
1549 * packet in hand.
1550 */
1551 struct ethhdr *ehdr;
1552 skb_reset_mac_header(skb);
1553 ehdr = eth_hdr(skb);
1554 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1555 err = -EMSGSIZE;
1556 goto out_unlock;
1557 }
1558 }
1559
1560 skb->protocol = proto;
1561 skb->dev = dev;
1562 skb->priority = sk->sk_priority;
1563 skb->mark = sk->sk_mark;
1564 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1565 if (err < 0)
1566 goto out_unlock;
1567
1568 if (unlikely(extra_len == 4))
1569 skb->no_fcs = 1;
1570
1571 dev_queue_xmit(skb);
1572 rcu_read_unlock();
1573 return len;
1574
1575out_unlock:
1576 rcu_read_unlock();
1577out_free:
1578 kfree_skb(skb);
1579 return err;
1580}
1581
1582static unsigned int run_filter(const struct sk_buff *skb,
1583 const struct sock *sk,
1584 unsigned int res)
1585{
1586 struct sk_filter *filter;
1587
1588 rcu_read_lock();
1589 filter = rcu_dereference(sk->sk_filter);
1590 if (filter != NULL)
1591 res = SK_RUN_FILTER(filter, skb);
1592 rcu_read_unlock();
1593
1594 return res;
1595}
1596
1597/*
1598 * This function makes lazy skb cloning in hope that most of packets
1599 * are discarded by BPF.
1600 *
1601 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1602 * and skb->cb are mangled. It works because (and until) packets
1603 * falling here are owned by current CPU. Output packets are cloned
1604 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1605 * sequencially, so that if we return skb to original state on exit,
1606 * we will not harm anyone.
1607 */
1608
1609static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1610 struct packet_type *pt, struct net_device *orig_dev)
1611{
1612 struct sock *sk;
1613 struct sockaddr_ll *sll;
1614 struct packet_sock *po;
1615 u8 *skb_head = skb->data;
1616 int skb_len = skb->len;
1617 unsigned int snaplen, res;
1618
1619 if (skb->pkt_type == PACKET_LOOPBACK)
1620 goto drop;
1621
1622 sk = pt->af_packet_priv;
1623 po = pkt_sk(sk);
1624
1625 if (!net_eq(dev_net(dev), sock_net(sk)))
1626 goto drop;
1627
1628 skb->dev = dev;
1629
1630 if (dev->header_ops) {
1631 /* The device has an explicit notion of ll header,
1632 * exported to higher levels.
1633 *
1634 * Otherwise, the device hides details of its frame
1635 * structure, so that corresponding packet head is
1636 * never delivered to user.
1637 */
1638 if (sk->sk_type != SOCK_DGRAM)
1639 skb_push(skb, skb->data - skb_mac_header(skb));
1640 else if (skb->pkt_type == PACKET_OUTGOING) {
1641 /* Special case: outgoing packets have ll header at head */
1642 skb_pull(skb, skb_network_offset(skb));
1643 }
1644 }
1645
1646 snaplen = skb->len;
1647
1648 res = run_filter(skb, sk, snaplen);
1649 if (!res)
1650 goto drop_n_restore;
1651 if (snaplen > res)
1652 snaplen = res;
1653
1654 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1655 goto drop_n_acct;
1656
1657 if (skb_shared(skb)) {
1658 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1659 if (nskb == NULL)
1660 goto drop_n_acct;
1661
1662 if (skb_head != skb->data) {
1663 skb->data = skb_head;
1664 skb->len = skb_len;
1665 }
1666 consume_skb(skb);
1667 skb = nskb;
1668 }
1669
1670 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1671 sizeof(skb->cb));
1672
1673 sll = &PACKET_SKB_CB(skb)->sa.ll;
1674 sll->sll_family = AF_PACKET;
1675 sll->sll_hatype = dev->type;
1676 sll->sll_protocol = skb->protocol;
1677 sll->sll_pkttype = skb->pkt_type;
1678 if (unlikely(po->origdev))
1679 sll->sll_ifindex = orig_dev->ifindex;
1680 else
1681 sll->sll_ifindex = dev->ifindex;
1682
1683 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1684
1685 PACKET_SKB_CB(skb)->origlen = skb->len;
1686
1687 if (pskb_trim(skb, snaplen))
1688 goto drop_n_acct;
1689
1690 skb_set_owner_r(skb, sk);
1691 skb->dev = NULL;
1692 skb_dst_drop(skb);
1693
1694 /* drop conntrack reference */
1695 nf_reset(skb);
1696
1697 spin_lock(&sk->sk_receive_queue.lock);
1698 po->stats.tp_packets++;
1699 skb->dropcount = atomic_read(&sk->sk_drops);
1700 __skb_queue_tail(&sk->sk_receive_queue, skb);
1701 spin_unlock(&sk->sk_receive_queue.lock);
1702 sk->sk_data_ready(sk, skb->len);
1703 return 0;
1704
1705drop_n_acct:
1706 spin_lock(&sk->sk_receive_queue.lock);
1707 po->stats.tp_drops++;
1708 atomic_inc(&sk->sk_drops);
1709 spin_unlock(&sk->sk_receive_queue.lock);
1710
1711drop_n_restore:
1712 if (skb_head != skb->data && skb_shared(skb)) {
1713 skb->data = skb_head;
1714 skb->len = skb_len;
1715 }
1716drop:
1717 consume_skb(skb);
1718 return 0;
1719}
1720
1721static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1722 struct packet_type *pt, struct net_device *orig_dev)
1723{
1724 struct sock *sk;
1725 struct packet_sock *po;
1726 struct sockaddr_ll *sll;
1727 union {
1728 struct tpacket_hdr *h1;
1729 struct tpacket2_hdr *h2;
1730 struct tpacket3_hdr *h3;
1731 void *raw;
1732 } h;
1733 u8 *skb_head = skb->data;
1734 int skb_len = skb->len;
1735 unsigned int snaplen, res;
1736 unsigned long status = TP_STATUS_USER;
1737 unsigned short macoff, netoff, hdrlen;
1738 struct sk_buff *copy_skb = NULL;
1739 struct timeval tv;
1740 struct timespec ts;
1741 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1742
1743 if (skb->pkt_type == PACKET_LOOPBACK)
1744 goto drop;
1745
1746 sk = pt->af_packet_priv;
1747 po = pkt_sk(sk);
1748
1749 if (!net_eq(dev_net(dev), sock_net(sk)))
1750 goto drop;
1751
1752 if (dev->header_ops) {
1753 if (sk->sk_type != SOCK_DGRAM)
1754 skb_push(skb, skb->data - skb_mac_header(skb));
1755 else if (skb->pkt_type == PACKET_OUTGOING) {
1756 /* Special case: outgoing packets have ll header at head */
1757 skb_pull(skb, skb_network_offset(skb));
1758 }
1759 }
1760
1761 if (skb->ip_summed == CHECKSUM_PARTIAL)
1762 status |= TP_STATUS_CSUMNOTREADY;
1763
1764 snaplen = skb->len;
1765
1766 res = run_filter(skb, sk, snaplen);
1767 if (!res)
1768 goto drop_n_restore;
1769 if (snaplen > res)
1770 snaplen = res;
1771
1772 if (sk->sk_type == SOCK_DGRAM) {
1773 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1774 po->tp_reserve;
1775 } else {
1776 unsigned int maclen = skb_network_offset(skb);
1777 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1778 (maclen < 16 ? 16 : maclen)) +
1779 po->tp_reserve;
1780 macoff = netoff - maclen;
1781 }
1782 if (po->tp_version <= TPACKET_V2) {
1783 if (macoff + snaplen > po->rx_ring.frame_size) {
1784 if (po->copy_thresh &&
1785 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1786 if (skb_shared(skb)) {
1787 copy_skb = skb_clone(skb, GFP_ATOMIC);
1788 } else {
1789 copy_skb = skb_get(skb);
1790 skb_head = skb->data;
1791 }
1792 if (copy_skb)
1793 skb_set_owner_r(copy_skb, sk);
1794 }
1795 snaplen = po->rx_ring.frame_size - macoff;
1796 if ((int)snaplen < 0)
1797 snaplen = 0;
1798 }
1799 }
1800 spin_lock(&sk->sk_receive_queue.lock);
1801 h.raw = packet_current_rx_frame(po, skb,
1802 TP_STATUS_KERNEL, (macoff+snaplen));
1803 if (!h.raw)
1804 goto ring_is_full;
1805 if (po->tp_version <= TPACKET_V2) {
1806 packet_increment_rx_head(po, &po->rx_ring);
1807 /*
1808 * LOSING will be reported till you read the stats,
1809 * because it's COR - Clear On Read.
1810 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1811 * at packet level.
1812 */
1813 if (po->stats.tp_drops)
1814 status |= TP_STATUS_LOSING;
1815 }
1816 po->stats.tp_packets++;
1817 if (copy_skb) {
1818 status |= TP_STATUS_COPY;
1819 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1820 }
1821 spin_unlock(&sk->sk_receive_queue.lock);
1822
1823 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1824
1825 switch (po->tp_version) {
1826 case TPACKET_V1:
1827 h.h1->tp_len = skb->len;
1828 h.h1->tp_snaplen = snaplen;
1829 h.h1->tp_mac = macoff;
1830 h.h1->tp_net = netoff;
1831 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1832 && shhwtstamps->syststamp.tv64)
1833 tv = ktime_to_timeval(shhwtstamps->syststamp);
1834 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1835 && shhwtstamps->hwtstamp.tv64)
1836 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1837 else if (skb->tstamp.tv64)
1838 tv = ktime_to_timeval(skb->tstamp);
1839 else
1840 do_gettimeofday(&tv);
1841 h.h1->tp_sec = tv.tv_sec;
1842 h.h1->tp_usec = tv.tv_usec;
1843 hdrlen = sizeof(*h.h1);
1844 break;
1845 case TPACKET_V2:
1846 h.h2->tp_len = skb->len;
1847 h.h2->tp_snaplen = snaplen;
1848 h.h2->tp_mac = macoff;
1849 h.h2->tp_net = netoff;
1850 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1851 && shhwtstamps->syststamp.tv64)
1852 ts = ktime_to_timespec(shhwtstamps->syststamp);
1853 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1854 && shhwtstamps->hwtstamp.tv64)
1855 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1856 else if (skb->tstamp.tv64)
1857 ts = ktime_to_timespec(skb->tstamp);
1858 else
1859 getnstimeofday(&ts);
1860 h.h2->tp_sec = ts.tv_sec;
1861 h.h2->tp_nsec = ts.tv_nsec;
1862 if (vlan_tx_tag_present(skb)) {
1863 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1864 status |= TP_STATUS_VLAN_VALID;
1865 } else {
1866 h.h2->tp_vlan_tci = 0;
1867 }
1868 h.h2->tp_padding = 0;
1869 hdrlen = sizeof(*h.h2);
1870 break;
1871 case TPACKET_V3:
1872 /* tp_nxt_offset,vlan are already populated above.
1873 * So DONT clear those fields here
1874 */
1875 h.h3->tp_status |= status;
1876 h.h3->tp_len = skb->len;
1877 h.h3->tp_snaplen = snaplen;
1878 h.h3->tp_mac = macoff;
1879 h.h3->tp_net = netoff;
1880 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1881 && shhwtstamps->syststamp.tv64)
1882 ts = ktime_to_timespec(shhwtstamps->syststamp);
1883 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1884 && shhwtstamps->hwtstamp.tv64)
1885 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1886 else if (skb->tstamp.tv64)
1887 ts = ktime_to_timespec(skb->tstamp);
1888 else
1889 getnstimeofday(&ts);
1890 h.h3->tp_sec = ts.tv_sec;
1891 h.h3->tp_nsec = ts.tv_nsec;
1892 hdrlen = sizeof(*h.h3);
1893 break;
1894 default:
1895 BUG();
1896 }
1897
1898 sll = h.raw + TPACKET_ALIGN(hdrlen);
1899 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1900 sll->sll_family = AF_PACKET;
1901 sll->sll_hatype = dev->type;
1902 sll->sll_protocol = skb->protocol;
1903 sll->sll_pkttype = skb->pkt_type;
1904 if (unlikely(po->origdev))
1905 sll->sll_ifindex = orig_dev->ifindex;
1906 else
1907 sll->sll_ifindex = dev->ifindex;
1908
1909 smp_mb();
1910#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1911 {
1912 u8 *start, *end;
1913
1914 if (po->tp_version <= TPACKET_V2) {
1915 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1916 + macoff + snaplen);
1917 for (start = h.raw; start < end; start += PAGE_SIZE)
1918 flush_dcache_page(pgv_to_page(start));
1919 }
1920 smp_wmb();
1921 }
1922#endif
1923 if (po->tp_version <= TPACKET_V2)
1924 __packet_set_status(po, h.raw, status);
1925 else
1926 prb_clear_blk_fill_status(&po->rx_ring);
1927
1928 sk->sk_data_ready(sk, 0);
1929
1930drop_n_restore:
1931 if (skb_head != skb->data && skb_shared(skb)) {
1932 skb->data = skb_head;
1933 skb->len = skb_len;
1934 }
1935drop:
1936 kfree_skb(skb);
1937 return 0;
1938
1939ring_is_full:
1940 po->stats.tp_drops++;
1941 spin_unlock(&sk->sk_receive_queue.lock);
1942
1943 sk->sk_data_ready(sk, 0);
1944 kfree_skb(copy_skb);
1945 goto drop_n_restore;
1946}
1947
1948static void tpacket_destruct_skb(struct sk_buff *skb)
1949{
1950 struct packet_sock *po = pkt_sk(skb->sk);
1951 void *ph;
1952
1953 if (likely(po->tx_ring.pg_vec)) {
1954 ph = skb_shinfo(skb)->destructor_arg;
1955 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1956 atomic_dec(&po->tx_ring.pending);
1957 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1958 }
1959
1960 sock_wfree(skb);
1961}
1962
1963static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1964 void *frame, struct net_device *dev, int size_max,
1965 __be16 proto, unsigned char *addr, int hlen)
1966{
1967 union {
1968 struct tpacket_hdr *h1;
1969 struct tpacket2_hdr *h2;
1970 void *raw;
1971 } ph;
1972 int to_write, offset, len, tp_len, nr_frags, len_max;
1973 struct socket *sock = po->sk.sk_socket;
1974 struct page *page;
1975 void *data;
1976 int err;
1977
1978 ph.raw = frame;
1979
1980 skb->protocol = proto;
1981 skb->dev = dev;
1982 skb->priority = po->sk.sk_priority;
1983 skb->mark = po->sk.sk_mark;
1984 skb_shinfo(skb)->destructor_arg = ph.raw;
1985
1986 switch (po->tp_version) {
1987 case TPACKET_V2:
1988 tp_len = ph.h2->tp_len;
1989 break;
1990 default:
1991 tp_len = ph.h1->tp_len;
1992 break;
1993 }
1994 if (unlikely(tp_len > size_max)) {
1995 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1996 return -EMSGSIZE;
1997 }
1998
1999 skb_reserve(skb, hlen);
2000 skb_reset_network_header(skb);
2001
2002 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2003 to_write = tp_len;
2004
2005 if (sock->type == SOCK_DGRAM) {
2006 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2007 NULL, tp_len);
2008 if (unlikely(err < 0))
2009 return -EINVAL;
2010 } else if (dev->hard_header_len) {
2011 /* net device doesn't like empty head */
2012 if (unlikely(tp_len <= dev->hard_header_len)) {
2013 pr_err("packet size is too short (%d < %d)\n",
2014 tp_len, dev->hard_header_len);
2015 return -EINVAL;
2016 }
2017
2018 skb_push(skb, dev->hard_header_len);
2019 err = skb_store_bits(skb, 0, data,
2020 dev->hard_header_len);
2021 if (unlikely(err))
2022 return err;
2023
2024 data += dev->hard_header_len;
2025 to_write -= dev->hard_header_len;
2026 }
2027
2028 err = -EFAULT;
2029 offset = offset_in_page(data);
2030 len_max = PAGE_SIZE - offset;
2031 len = ((to_write > len_max) ? len_max : to_write);
2032
2033 skb->data_len = to_write;
2034 skb->len += to_write;
2035 skb->truesize += to_write;
2036 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2037
2038 while (likely(to_write)) {
2039 nr_frags = skb_shinfo(skb)->nr_frags;
2040
2041 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2042 pr_err("Packet exceed the number of skb frags(%lu)\n",
2043 MAX_SKB_FRAGS);
2044 return -EFAULT;
2045 }
2046
2047 page = pgv_to_page(data);
2048 data += len;
2049 flush_dcache_page(page);
2050 get_page(page);
2051 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2052 to_write -= len;
2053 offset = 0;
2054 len_max = PAGE_SIZE;
2055 len = ((to_write > len_max) ? len_max : to_write);
2056 }
2057
2058 return tp_len;
2059}
2060
2061static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2062{
2063 struct sk_buff *skb;
2064 struct net_device *dev;
2065 __be16 proto;
2066 bool need_rls_dev = false;
2067 int err, reserve = 0;
2068 void *ph;
2069 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2070 int tp_len, size_max;
2071 unsigned char *addr;
2072 int len_sum = 0;
2073 int status = 0;
2074 int hlen, tlen;
2075
2076 mutex_lock(&po->pg_vec_lock);
2077
2078 err = -EBUSY;
2079 if (saddr == NULL) {
2080 dev = po->prot_hook.dev;
2081 proto = po->num;
2082 addr = NULL;
2083 } else {
2084 err = -EINVAL;
2085 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2086 goto out;
2087 if (msg->msg_namelen < (saddr->sll_halen
2088 + offsetof(struct sockaddr_ll,
2089 sll_addr)))
2090 goto out;
2091 proto = saddr->sll_protocol;
2092 addr = saddr->sll_addr;
2093 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2094 need_rls_dev = true;
2095 }
2096
2097 err = -ENXIO;
2098 if (unlikely(dev == NULL))
2099 goto out;
2100
2101 reserve = dev->hard_header_len;
2102
2103 err = -ENETDOWN;
2104 if (unlikely(!(dev->flags & IFF_UP)))
2105 goto out_put;
2106
2107 size_max = po->tx_ring.frame_size
2108 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2109
2110 if (size_max > dev->mtu + reserve)
2111 size_max = dev->mtu + reserve;
2112
2113 do {
2114 ph = packet_current_frame(po, &po->tx_ring,
2115 TP_STATUS_SEND_REQUEST);
2116
2117 if (unlikely(ph == NULL)) {
2118 schedule();
2119 continue;
2120 }
2121
2122 status = TP_STATUS_SEND_REQUEST;
2123 hlen = LL_RESERVED_SPACE(dev);
2124 tlen = dev->needed_tailroom;
2125 skb = sock_alloc_send_skb(&po->sk,
2126 hlen + tlen + sizeof(struct sockaddr_ll),
2127 0, &err);
2128
2129 if (unlikely(skb == NULL))
2130 goto out_status;
2131
2132 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2133 addr, hlen);
2134
2135 if (unlikely(tp_len < 0)) {
2136 if (po->tp_loss) {
2137 __packet_set_status(po, ph,
2138 TP_STATUS_AVAILABLE);
2139 packet_increment_head(&po->tx_ring);
2140 kfree_skb(skb);
2141 continue;
2142 } else {
2143 status = TP_STATUS_WRONG_FORMAT;
2144 err = tp_len;
2145 goto out_status;
2146 }
2147 }
2148
2149 skb->destructor = tpacket_destruct_skb;
2150 __packet_set_status(po, ph, TP_STATUS_SENDING);
2151 atomic_inc(&po->tx_ring.pending);
2152
2153 status = TP_STATUS_SEND_REQUEST;
2154 err = dev_queue_xmit(skb);
2155 if (unlikely(err > 0)) {
2156 err = net_xmit_errno(err);
2157 if (err && __packet_get_status(po, ph) ==
2158 TP_STATUS_AVAILABLE) {
2159 /* skb was destructed already */
2160 skb = NULL;
2161 goto out_status;
2162 }
2163 /*
2164 * skb was dropped but not destructed yet;
2165 * let's treat it like congestion or err < 0
2166 */
2167 err = 0;
2168 }
2169 packet_increment_head(&po->tx_ring);
2170 len_sum += tp_len;
2171 } while (likely((ph != NULL) ||
2172 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2173 (atomic_read(&po->tx_ring.pending))))
2174 );
2175
2176 err = len_sum;
2177 goto out_put;
2178
2179out_status:
2180 __packet_set_status(po, ph, status);
2181 kfree_skb(skb);
2182out_put:
2183 if (need_rls_dev)
2184 dev_put(dev);
2185out:
2186 mutex_unlock(&po->pg_vec_lock);
2187 return err;
2188}
2189
2190static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2191 size_t reserve, size_t len,
2192 size_t linear, int noblock,
2193 int *err)
2194{
2195 struct sk_buff *skb;
2196
2197 /* Under a page? Don't bother with paged skb. */
2198 if (prepad + len < PAGE_SIZE || !linear)
2199 linear = len;
2200
2201 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2202 err);
2203 if (!skb)
2204 return NULL;
2205
2206 skb_reserve(skb, reserve);
2207 skb_put(skb, linear);
2208 skb->data_len = len - linear;
2209 skb->len += len - linear;
2210
2211 return skb;
2212}
2213
2214static int packet_snd(struct socket *sock,
2215 struct msghdr *msg, size_t len)
2216{
2217 struct sock *sk = sock->sk;
2218 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2219 struct sk_buff *skb;
2220 struct net_device *dev;
2221 __be16 proto;
2222 bool need_rls_dev = false;
2223 unsigned char *addr;
2224 int err, reserve = 0;
2225 struct virtio_net_hdr vnet_hdr = { 0 };
2226 int offset = 0;
2227 int vnet_hdr_len;
2228 struct packet_sock *po = pkt_sk(sk);
2229 unsigned short gso_type = 0;
2230 int hlen, tlen;
2231 int extra_len = 0;
2232
2233 /*
2234 * Get and verify the address.
2235 */
2236
2237 if (saddr == NULL) {
2238 dev = po->prot_hook.dev;
2239 proto = po->num;
2240 addr = NULL;
2241 } else {
2242 err = -EINVAL;
2243 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2244 goto out;
2245 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2246 goto out;
2247 proto = saddr->sll_protocol;
2248 addr = saddr->sll_addr;
2249 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2250 need_rls_dev = true;
2251 }
2252
2253 err = -ENXIO;
2254 if (dev == NULL)
2255 goto out_unlock;
2256 if (sock->type == SOCK_RAW)
2257 reserve = dev->hard_header_len;
2258
2259 err = -ENETDOWN;
2260 if (!(dev->flags & IFF_UP))
2261 goto out_unlock;
2262
2263 if (po->has_vnet_hdr) {
2264 vnet_hdr_len = sizeof(vnet_hdr);
2265
2266 err = -EINVAL;
2267 if (len < vnet_hdr_len)
2268 goto out_unlock;
2269
2270 len -= vnet_hdr_len;
2271
2272 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2273 vnet_hdr_len);
2274 if (err < 0)
2275 goto out_unlock;
2276
2277 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2278 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2279 vnet_hdr.hdr_len))
2280 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2281 vnet_hdr.csum_offset + 2;
2282
2283 err = -EINVAL;
2284 if (vnet_hdr.hdr_len > len)
2285 goto out_unlock;
2286
2287 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2288 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2289 case VIRTIO_NET_HDR_GSO_TCPV4:
2290 gso_type = SKB_GSO_TCPV4;
2291 break;
2292 case VIRTIO_NET_HDR_GSO_TCPV6:
2293 gso_type = SKB_GSO_TCPV6;
2294 break;
2295 case VIRTIO_NET_HDR_GSO_UDP:
2296 gso_type = SKB_GSO_UDP;
2297 break;
2298 default:
2299 goto out_unlock;
2300 }
2301
2302 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2303 gso_type |= SKB_GSO_TCP_ECN;
2304
2305 if (vnet_hdr.gso_size == 0)
2306 goto out_unlock;
2307
2308 }
2309 }
2310
2311 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2312 if (!netif_supports_nofcs(dev)) {
2313 err = -EPROTONOSUPPORT;
2314 goto out_unlock;
2315 }
2316 extra_len = 4; /* We're doing our own CRC */
2317 }
2318
2319 err = -EMSGSIZE;
2320 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2321 goto out_unlock;
2322
2323 err = -ENOBUFS;
2324 hlen = LL_RESERVED_SPACE(dev);
2325 tlen = dev->needed_tailroom;
2326 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2327 msg->msg_flags & MSG_DONTWAIT, &err);
2328 if (skb == NULL)
2329 goto out_unlock;
2330
2331 skb_set_network_header(skb, reserve);
2332
2333 err = -EINVAL;
2334 if (sock->type == SOCK_DGRAM &&
2335 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2336 goto out_free;
2337
2338 /* Returns -EFAULT on error */
2339 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2340 if (err)
2341 goto out_free;
2342 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2343 if (err < 0)
2344 goto out_free;
2345
2346 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2347 /* Earlier code assumed this would be a VLAN pkt,
2348 * double-check this now that we have the actual
2349 * packet in hand.
2350 */
2351 struct ethhdr *ehdr;
2352 skb_reset_mac_header(skb);
2353 ehdr = eth_hdr(skb);
2354 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2355 err = -EMSGSIZE;
2356 goto out_free;
2357 }
2358 }
2359
2360 skb->protocol = proto;
2361 skb->dev = dev;
2362 skb->priority = sk->sk_priority;
2363 skb->mark = sk->sk_mark;
2364
2365 if (po->has_vnet_hdr) {
2366 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2367 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2368 vnet_hdr.csum_offset)) {
2369 err = -EINVAL;
2370 goto out_free;
2371 }
2372 }
2373
2374 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2375 skb_shinfo(skb)->gso_type = gso_type;
2376
2377 /* Header must be checked, and gso_segs computed. */
2378 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2379 skb_shinfo(skb)->gso_segs = 0;
2380
2381 len += vnet_hdr_len;
2382 }
2383
2384 if (unlikely(extra_len == 4))
2385 skb->no_fcs = 1;
2386
2387 /*
2388 * Now send it
2389 */
2390
2391 err = dev_queue_xmit(skb);
2392 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2393 goto out_unlock;
2394
2395 if (need_rls_dev)
2396 dev_put(dev);
2397
2398 return len;
2399
2400out_free:
2401 kfree_skb(skb);
2402out_unlock:
2403 if (dev && need_rls_dev)
2404 dev_put(dev);
2405out:
2406 return err;
2407}
2408
2409static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2410 struct msghdr *msg, size_t len)
2411{
2412 struct sock *sk = sock->sk;
2413 struct packet_sock *po = pkt_sk(sk);
2414 if (po->tx_ring.pg_vec)
2415 return tpacket_snd(po, msg);
2416 else
2417 return packet_snd(sock, msg, len);
2418}
2419
2420/*
2421 * Close a PACKET socket. This is fairly simple. We immediately go
2422 * to 'closed' state and remove our protocol entry in the device list.
2423 */
2424
2425static int packet_release(struct socket *sock)
2426{
2427 struct sock *sk = sock->sk;
2428 struct packet_sock *po;
2429 struct net *net;
2430 union tpacket_req_u req_u;
2431
2432 if (!sk)
2433 return 0;
2434
2435 net = sock_net(sk);
2436 po = pkt_sk(sk);
2437
2438 spin_lock_bh(&net->packet.sklist_lock);
2439 sk_del_node_init_rcu(sk);
2440 sock_prot_inuse_add(net, sk->sk_prot, -1);
2441 spin_unlock_bh(&net->packet.sklist_lock);
2442
2443 spin_lock(&po->bind_lock);
2444 unregister_prot_hook(sk, false);
2445 if (po->prot_hook.dev) {
2446 dev_put(po->prot_hook.dev);
2447 po->prot_hook.dev = NULL;
2448 }
2449 spin_unlock(&po->bind_lock);
2450
2451 packet_flush_mclist(sk);
2452
2453 memset(&req_u, 0, sizeof(req_u));
2454
2455 if (po->rx_ring.pg_vec)
2456 packet_set_ring(sk, &req_u, 1, 0);
2457
2458 if (po->tx_ring.pg_vec)
2459 packet_set_ring(sk, &req_u, 1, 1);
2460
2461 fanout_release(sk);
2462
2463 synchronize_net();
2464 /*
2465 * Now the socket is dead. No more input will appear.
2466 */
2467 sock_orphan(sk);
2468 sock->sk = NULL;
2469
2470 /* Purge queues */
2471
2472 skb_queue_purge(&sk->sk_receive_queue);
2473 sk_refcnt_debug_release(sk);
2474
2475 sock_put(sk);
2476 return 0;
2477}
2478
2479/*
2480 * Attach a packet hook.
2481 */
2482
2483static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2484{
2485 struct packet_sock *po = pkt_sk(sk);
2486
2487 if (po->fanout) {
2488 if (dev)
2489 dev_put(dev);
2490
2491 return -EINVAL;
2492 }
2493
2494 lock_sock(sk);
2495
2496 spin_lock(&po->bind_lock);
2497 unregister_prot_hook(sk, true);
2498 po->num = protocol;
2499 po->prot_hook.type = protocol;
2500 if (po->prot_hook.dev)
2501 dev_put(po->prot_hook.dev);
2502 po->prot_hook.dev = dev;
2503
2504 po->ifindex = dev ? dev->ifindex : 0;
2505
2506 if (protocol == 0)
2507 goto out_unlock;
2508
2509 if (!dev || (dev->flags & IFF_UP)) {
2510 register_prot_hook(sk);
2511 } else {
2512 sk->sk_err = ENETDOWN;
2513 if (!sock_flag(sk, SOCK_DEAD))
2514 sk->sk_error_report(sk);
2515 }
2516
2517out_unlock:
2518 spin_unlock(&po->bind_lock);
2519 release_sock(sk);
2520 return 0;
2521}
2522
2523/*
2524 * Bind a packet socket to a device
2525 */
2526
2527static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2528 int addr_len)
2529{
2530 struct sock *sk = sock->sk;
2531 char name[15];
2532 struct net_device *dev;
2533 int err = -ENODEV;
2534
2535 /*
2536 * Check legality
2537 */
2538
2539 if (addr_len != sizeof(struct sockaddr))
2540 return -EINVAL;
2541 strlcpy(name, uaddr->sa_data, sizeof(name));
2542
2543 dev = dev_get_by_name(sock_net(sk), name);
2544 if (dev)
2545 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2546 return err;
2547}
2548
2549static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2550{
2551 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2552 struct sock *sk = sock->sk;
2553 struct net_device *dev = NULL;
2554 int err;
2555
2556
2557 /*
2558 * Check legality
2559 */
2560
2561 if (addr_len < sizeof(struct sockaddr_ll))
2562 return -EINVAL;
2563 if (sll->sll_family != AF_PACKET)
2564 return -EINVAL;
2565
2566 if (sll->sll_ifindex) {
2567 err = -ENODEV;
2568 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2569 if (dev == NULL)
2570 goto out;
2571 }
2572 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2573
2574out:
2575 return err;
2576}
2577
2578static struct proto packet_proto = {
2579 .name = "PACKET",
2580 .owner = THIS_MODULE,
2581 .obj_size = sizeof(struct packet_sock),
2582};
2583
2584/*
2585 * Create a packet of type SOCK_PACKET.
2586 */
2587
2588static int packet_create(struct net *net, struct socket *sock, int protocol,
2589 int kern)
2590{
2591 struct sock *sk;
2592 struct packet_sock *po;
2593 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2594 int err;
2595
2596 if (!capable(CAP_NET_RAW))
2597 return -EPERM;
2598 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2599 sock->type != SOCK_PACKET)
2600 return -ESOCKTNOSUPPORT;
2601
2602 sock->state = SS_UNCONNECTED;
2603
2604 err = -ENOBUFS;
2605 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2606 if (sk == NULL)
2607 goto out;
2608
2609 sock->ops = &packet_ops;
2610 if (sock->type == SOCK_PACKET)
2611 sock->ops = &packet_ops_spkt;
2612
2613 sock_init_data(sock, sk);
2614
2615 po = pkt_sk(sk);
2616 sk->sk_family = PF_PACKET;
2617 po->num = proto;
2618
2619 sk->sk_destruct = packet_sock_destruct;
2620 sk_refcnt_debug_inc(sk);
2621
2622 /*
2623 * Attach a protocol block
2624 */
2625
2626 spin_lock_init(&po->bind_lock);
2627 mutex_init(&po->pg_vec_lock);
2628 po->prot_hook.func = packet_rcv;
2629
2630 if (sock->type == SOCK_PACKET)
2631 po->prot_hook.func = packet_rcv_spkt;
2632
2633 po->prot_hook.af_packet_priv = sk;
2634
2635 if (proto) {
2636 po->prot_hook.type = proto;
2637 register_prot_hook(sk);
2638 }
2639
2640 spin_lock_bh(&net->packet.sklist_lock);
2641 sk_add_node_rcu(sk, &net->packet.sklist);
2642 sock_prot_inuse_add(net, &packet_proto, 1);
2643 spin_unlock_bh(&net->packet.sklist_lock);
2644
2645 return 0;
2646out:
2647 return err;
2648}
2649
2650static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2651{
2652 struct sock_exterr_skb *serr;
2653 struct sk_buff *skb, *skb2;
2654 int copied, err;
2655
2656 err = -EAGAIN;
2657 skb = skb_dequeue(&sk->sk_error_queue);
2658 if (skb == NULL)
2659 goto out;
2660
2661 copied = skb->len;
2662 if (copied > len) {
2663 msg->msg_flags |= MSG_TRUNC;
2664 copied = len;
2665 }
2666 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2667 if (err)
2668 goto out_free_skb;
2669
2670 sock_recv_timestamp(msg, sk, skb);
2671
2672 serr = SKB_EXT_ERR(skb);
2673 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2674 sizeof(serr->ee), &serr->ee);
2675
2676 msg->msg_flags |= MSG_ERRQUEUE;
2677 err = copied;
2678
2679 /* Reset and regenerate socket error */
2680 spin_lock_bh(&sk->sk_error_queue.lock);
2681 sk->sk_err = 0;
2682 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2683 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2684 spin_unlock_bh(&sk->sk_error_queue.lock);
2685 sk->sk_error_report(sk);
2686 } else
2687 spin_unlock_bh(&sk->sk_error_queue.lock);
2688
2689out_free_skb:
2690 kfree_skb(skb);
2691out:
2692 return err;
2693}
2694
2695/*
2696 * Pull a packet from our receive queue and hand it to the user.
2697 * If necessary we block.
2698 */
2699
2700static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2701 struct msghdr *msg, size_t len, int flags)
2702{
2703 struct sock *sk = sock->sk;
2704 struct sk_buff *skb;
2705 int copied, err;
2706 struct sockaddr_ll *sll;
2707 int vnet_hdr_len = 0;
2708
2709 err = -EINVAL;
2710 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2711 goto out;
2712
2713#if 0
2714 /* What error should we return now? EUNATTACH? */
2715 if (pkt_sk(sk)->ifindex < 0)
2716 return -ENODEV;
2717#endif
2718
2719 if (flags & MSG_ERRQUEUE) {
2720 err = packet_recv_error(sk, msg, len);
2721 goto out;
2722 }
2723
2724 /*
2725 * Call the generic datagram receiver. This handles all sorts
2726 * of horrible races and re-entrancy so we can forget about it
2727 * in the protocol layers.
2728 *
2729 * Now it will return ENETDOWN, if device have just gone down,
2730 * but then it will block.
2731 */
2732
2733 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2734
2735 /*
2736 * An error occurred so return it. Because skb_recv_datagram()
2737 * handles the blocking we don't see and worry about blocking
2738 * retries.
2739 */
2740
2741 if (skb == NULL)
2742 goto out;
2743
2744 if (pkt_sk(sk)->has_vnet_hdr) {
2745 struct virtio_net_hdr vnet_hdr = { 0 };
2746
2747 err = -EINVAL;
2748 vnet_hdr_len = sizeof(vnet_hdr);
2749 if (len < vnet_hdr_len)
2750 goto out_free;
2751
2752 len -= vnet_hdr_len;
2753
2754 if (skb_is_gso(skb)) {
2755 struct skb_shared_info *sinfo = skb_shinfo(skb);
2756
2757 /* This is a hint as to how much should be linear. */
2758 vnet_hdr.hdr_len = skb_headlen(skb);
2759 vnet_hdr.gso_size = sinfo->gso_size;
2760 if (sinfo->gso_type & SKB_GSO_TCPV4)
2761 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2762 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2763 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2764 else if (sinfo->gso_type & SKB_GSO_UDP)
2765 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2766 else if (sinfo->gso_type & SKB_GSO_FCOE)
2767 goto out_free;
2768 else
2769 BUG();
2770 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2771 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2772 } else
2773 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2774
2775 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2776 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2777 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2778 vnet_hdr.csum_offset = skb->csum_offset;
2779 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2780 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2781 } /* else everything is zero */
2782
2783 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2784 vnet_hdr_len);
2785 if (err < 0)
2786 goto out_free;
2787 }
2788
2789 /*
2790 * If the address length field is there to be filled in, we fill
2791 * it in now.
2792 */
2793
2794 sll = &PACKET_SKB_CB(skb)->sa.ll;
2795 if (sock->type == SOCK_PACKET)
2796 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2797 else
2798 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2799
2800 /*
2801 * You lose any data beyond the buffer you gave. If it worries a
2802 * user program they can ask the device for its MTU anyway.
2803 */
2804
2805 copied = skb->len;
2806 if (copied > len) {
2807 copied = len;
2808 msg->msg_flags |= MSG_TRUNC;
2809 }
2810
2811 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2812 if (err)
2813 goto out_free;
2814
2815 sock_recv_ts_and_drops(msg, sk, skb);
2816
2817 if (msg->msg_name)
2818 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2819 msg->msg_namelen);
2820
2821 if (pkt_sk(sk)->auxdata) {
2822 struct tpacket_auxdata aux;
2823
2824 aux.tp_status = TP_STATUS_USER;
2825 if (skb->ip_summed == CHECKSUM_PARTIAL)
2826 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2827 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2828 aux.tp_snaplen = skb->len;
2829 aux.tp_mac = 0;
2830 aux.tp_net = skb_network_offset(skb);
2831 if (vlan_tx_tag_present(skb)) {
2832 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2833 aux.tp_status |= TP_STATUS_VLAN_VALID;
2834 } else {
2835 aux.tp_vlan_tci = 0;
2836 }
2837 aux.tp_padding = 0;
2838 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2839 }
2840
2841 /*
2842 * Free or return the buffer as appropriate. Again this
2843 * hides all the races and re-entrancy issues from us.
2844 */
2845 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2846
2847out_free:
2848 skb_free_datagram(sk, skb);
2849out:
2850 return err;
2851}
2852
2853static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2854 int *uaddr_len, int peer)
2855{
2856 struct net_device *dev;
2857 struct sock *sk = sock->sk;
2858
2859 if (peer)
2860 return -EOPNOTSUPP;
2861
2862 uaddr->sa_family = AF_PACKET;
2863 rcu_read_lock();
2864 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2865 if (dev)
2866 strncpy(uaddr->sa_data, dev->name, 14);
2867 else
2868 memset(uaddr->sa_data, 0, 14);
2869 rcu_read_unlock();
2870 *uaddr_len = sizeof(*uaddr);
2871
2872 return 0;
2873}
2874
2875static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2876 int *uaddr_len, int peer)
2877{
2878 struct net_device *dev;
2879 struct sock *sk = sock->sk;
2880 struct packet_sock *po = pkt_sk(sk);
2881 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2882
2883 if (peer)
2884 return -EOPNOTSUPP;
2885
2886 sll->sll_family = AF_PACKET;
2887 sll->sll_ifindex = po->ifindex;
2888 sll->sll_protocol = po->num;
2889 sll->sll_pkttype = 0;
2890 rcu_read_lock();
2891 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2892 if (dev) {
2893 sll->sll_hatype = dev->type;
2894 sll->sll_halen = dev->addr_len;
2895 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2896 } else {
2897 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2898 sll->sll_halen = 0;
2899 }
2900 rcu_read_unlock();
2901 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2902
2903 return 0;
2904}
2905
2906static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2907 int what)
2908{
2909 switch (i->type) {
2910 case PACKET_MR_MULTICAST:
2911 if (i->alen != dev->addr_len)
2912 return -EINVAL;
2913 if (what > 0)
2914 return dev_mc_add(dev, i->addr);
2915 else
2916 return dev_mc_del(dev, i->addr);
2917 break;
2918 case PACKET_MR_PROMISC:
2919 return dev_set_promiscuity(dev, what);
2920 break;
2921 case PACKET_MR_ALLMULTI:
2922 return dev_set_allmulti(dev, what);
2923 break;
2924 case PACKET_MR_UNICAST:
2925 if (i->alen != dev->addr_len)
2926 return -EINVAL;
2927 if (what > 0)
2928 return dev_uc_add(dev, i->addr);
2929 else
2930 return dev_uc_del(dev, i->addr);
2931 break;
2932 default:
2933 break;
2934 }
2935 return 0;
2936}
2937
2938static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2939{
2940 for ( ; i; i = i->next) {
2941 if (i->ifindex == dev->ifindex)
2942 packet_dev_mc(dev, i, what);
2943 }
2944}
2945
2946static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2947{
2948 struct packet_sock *po = pkt_sk(sk);
2949 struct packet_mclist *ml, *i;
2950 struct net_device *dev;
2951 int err;
2952
2953 rtnl_lock();
2954
2955 err = -ENODEV;
2956 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2957 if (!dev)
2958 goto done;
2959
2960 err = -EINVAL;
2961 if (mreq->mr_alen > dev->addr_len)
2962 goto done;
2963
2964 err = -ENOBUFS;
2965 i = kmalloc(sizeof(*i), GFP_KERNEL);
2966 if (i == NULL)
2967 goto done;
2968
2969 err = 0;
2970 for (ml = po->mclist; ml; ml = ml->next) {
2971 if (ml->ifindex == mreq->mr_ifindex &&
2972 ml->type == mreq->mr_type &&
2973 ml->alen == mreq->mr_alen &&
2974 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2975 ml->count++;
2976 /* Free the new element ... */
2977 kfree(i);
2978 goto done;
2979 }
2980 }
2981
2982 i->type = mreq->mr_type;
2983 i->ifindex = mreq->mr_ifindex;
2984 i->alen = mreq->mr_alen;
2985 memcpy(i->addr, mreq->mr_address, i->alen);
2986 i->count = 1;
2987 i->next = po->mclist;
2988 po->mclist = i;
2989 err = packet_dev_mc(dev, i, 1);
2990 if (err) {
2991 po->mclist = i->next;
2992 kfree(i);
2993 }
2994
2995done:
2996 rtnl_unlock();
2997 return err;
2998}
2999
3000static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3001{
3002 struct packet_mclist *ml, **mlp;
3003
3004 rtnl_lock();
3005
3006 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3007 if (ml->ifindex == mreq->mr_ifindex &&
3008 ml->type == mreq->mr_type &&
3009 ml->alen == mreq->mr_alen &&
3010 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3011 if (--ml->count == 0) {
3012 struct net_device *dev;
3013 *mlp = ml->next;
3014 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3015 if (dev)
3016 packet_dev_mc(dev, ml, -1);
3017 kfree(ml);
3018 }
3019 rtnl_unlock();
3020 return 0;
3021 }
3022 }
3023 rtnl_unlock();
3024 return -EADDRNOTAVAIL;
3025}
3026
3027static void packet_flush_mclist(struct sock *sk)
3028{
3029 struct packet_sock *po = pkt_sk(sk);
3030 struct packet_mclist *ml;
3031
3032 if (!po->mclist)
3033 return;
3034
3035 rtnl_lock();
3036 while ((ml = po->mclist) != NULL) {
3037 struct net_device *dev;
3038
3039 po->mclist = ml->next;
3040 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3041 if (dev != NULL)
3042 packet_dev_mc(dev, ml, -1);
3043 kfree(ml);
3044 }
3045 rtnl_unlock();
3046}
3047
3048static int
3049packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3050{
3051 struct sock *sk = sock->sk;
3052 struct packet_sock *po = pkt_sk(sk);
3053 int ret;
3054
3055 if (level != SOL_PACKET)
3056 return -ENOPROTOOPT;
3057
3058 switch (optname) {
3059 case PACKET_ADD_MEMBERSHIP:
3060 case PACKET_DROP_MEMBERSHIP:
3061 {
3062 struct packet_mreq_max mreq;
3063 int len = optlen;
3064 memset(&mreq, 0, sizeof(mreq));
3065 if (len < sizeof(struct packet_mreq))
3066 return -EINVAL;
3067 if (len > sizeof(mreq))
3068 len = sizeof(mreq);
3069 if (copy_from_user(&mreq, optval, len))
3070 return -EFAULT;
3071 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3072 return -EINVAL;
3073 if (optname == PACKET_ADD_MEMBERSHIP)
3074 ret = packet_mc_add(sk, &mreq);
3075 else
3076 ret = packet_mc_drop(sk, &mreq);
3077 return ret;
3078 }
3079
3080 case PACKET_RX_RING:
3081 case PACKET_TX_RING:
3082 {
3083 union tpacket_req_u req_u;
3084 int len;
3085
3086 switch (po->tp_version) {
3087 case TPACKET_V1:
3088 case TPACKET_V2:
3089 len = sizeof(req_u.req);
3090 break;
3091 case TPACKET_V3:
3092 default:
3093 len = sizeof(req_u.req3);
3094 break;
3095 }
3096 if (optlen < len)
3097 return -EINVAL;
3098 if (pkt_sk(sk)->has_vnet_hdr)
3099 return -EINVAL;
3100 if (copy_from_user(&req_u.req, optval, len))
3101 return -EFAULT;
3102 return packet_set_ring(sk, &req_u, 0,
3103 optname == PACKET_TX_RING);
3104 }
3105 case PACKET_COPY_THRESH:
3106 {
3107 int val;
3108
3109 if (optlen != sizeof(val))
3110 return -EINVAL;
3111 if (copy_from_user(&val, optval, sizeof(val)))
3112 return -EFAULT;
3113
3114 pkt_sk(sk)->copy_thresh = val;
3115 return 0;
3116 }
3117 case PACKET_VERSION:
3118 {
3119 int val;
3120
3121 if (optlen != sizeof(val))
3122 return -EINVAL;
3123 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3124 return -EBUSY;
3125 if (copy_from_user(&val, optval, sizeof(val)))
3126 return -EFAULT;
3127 switch (val) {
3128 case TPACKET_V1:
3129 case TPACKET_V2:
3130 case TPACKET_V3:
3131 po->tp_version = val;
3132 return 0;
3133 default:
3134 return -EINVAL;
3135 }
3136 }
3137 case PACKET_RESERVE:
3138 {
3139 unsigned int val;
3140
3141 if (optlen != sizeof(val))
3142 return -EINVAL;
3143 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3144 return -EBUSY;
3145 if (copy_from_user(&val, optval, sizeof(val)))
3146 return -EFAULT;
3147 po->tp_reserve = val;
3148 return 0;
3149 }
3150 case PACKET_LOSS:
3151 {
3152 unsigned int val;
3153
3154 if (optlen != sizeof(val))
3155 return -EINVAL;
3156 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3157 return -EBUSY;
3158 if (copy_from_user(&val, optval, sizeof(val)))
3159 return -EFAULT;
3160 po->tp_loss = !!val;
3161 return 0;
3162 }
3163 case PACKET_AUXDATA:
3164 {
3165 int val;
3166
3167 if (optlen < sizeof(val))
3168 return -EINVAL;
3169 if (copy_from_user(&val, optval, sizeof(val)))
3170 return -EFAULT;
3171
3172 po->auxdata = !!val;
3173 return 0;
3174 }
3175 case PACKET_ORIGDEV:
3176 {
3177 int val;
3178
3179 if (optlen < sizeof(val))
3180 return -EINVAL;
3181 if (copy_from_user(&val, optval, sizeof(val)))
3182 return -EFAULT;
3183
3184 po->origdev = !!val;
3185 return 0;
3186 }
3187 case PACKET_VNET_HDR:
3188 {
3189 int val;
3190
3191 if (sock->type != SOCK_RAW)
3192 return -EINVAL;
3193 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3194 return -EBUSY;
3195 if (optlen < sizeof(val))
3196 return -EINVAL;
3197 if (copy_from_user(&val, optval, sizeof(val)))
3198 return -EFAULT;
3199
3200 po->has_vnet_hdr = !!val;
3201 return 0;
3202 }
3203 case PACKET_TIMESTAMP:
3204 {
3205 int val;
3206
3207 if (optlen != sizeof(val))
3208 return -EINVAL;
3209 if (copy_from_user(&val, optval, sizeof(val)))
3210 return -EFAULT;
3211
3212 po->tp_tstamp = val;
3213 return 0;
3214 }
3215 case PACKET_FANOUT:
3216 {
3217 int val;
3218
3219 if (optlen != sizeof(val))
3220 return -EINVAL;
3221 if (copy_from_user(&val, optval, sizeof(val)))
3222 return -EFAULT;
3223
3224 return fanout_add(sk, val & 0xffff, val >> 16);
3225 }
3226 default:
3227 return -ENOPROTOOPT;
3228 }
3229}
3230
3231static int packet_getsockopt(struct socket *sock, int level, int optname,
3232 char __user *optval, int __user *optlen)
3233{
3234 int len;
3235 int val, lv = sizeof(val);
3236 struct sock *sk = sock->sk;
3237 struct packet_sock *po = pkt_sk(sk);
3238 void *data = &val;
3239 struct tpacket_stats st;
3240 union tpacket_stats_u st_u;
3241
3242 if (level != SOL_PACKET)
3243 return -ENOPROTOOPT;
3244
3245 if (get_user(len, optlen))
3246 return -EFAULT;
3247
3248 if (len < 0)
3249 return -EINVAL;
3250
3251 switch (optname) {
3252 case PACKET_STATISTICS:
3253 spin_lock_bh(&sk->sk_receive_queue.lock);
3254 if (po->tp_version == TPACKET_V3) {
3255 lv = sizeof(struct tpacket_stats_v3);
3256 memcpy(&st_u.stats3, &po->stats,
3257 sizeof(struct tpacket_stats));
3258 st_u.stats3.tp_freeze_q_cnt =
3259 po->stats_u.stats3.tp_freeze_q_cnt;
3260 st_u.stats3.tp_packets += po->stats.tp_drops;
3261 data = &st_u.stats3;
3262 } else {
3263 lv = sizeof(struct tpacket_stats);
3264 st = po->stats;
3265 st.tp_packets += st.tp_drops;
3266 data = &st;
3267 }
3268 memset(&po->stats, 0, sizeof(st));
3269 spin_unlock_bh(&sk->sk_receive_queue.lock);
3270 break;
3271 case PACKET_AUXDATA:
3272 val = po->auxdata;
3273 break;
3274 case PACKET_ORIGDEV:
3275 val = po->origdev;
3276 break;
3277 case PACKET_VNET_HDR:
3278 val = po->has_vnet_hdr;
3279 break;
3280 case PACKET_VERSION:
3281 val = po->tp_version;
3282 break;
3283 case PACKET_HDRLEN:
3284 if (len > sizeof(int))
3285 len = sizeof(int);
3286 if (copy_from_user(&val, optval, len))
3287 return -EFAULT;
3288 switch (val) {
3289 case TPACKET_V1:
3290 val = sizeof(struct tpacket_hdr);
3291 break;
3292 case TPACKET_V2:
3293 val = sizeof(struct tpacket2_hdr);
3294 break;
3295 case TPACKET_V3:
3296 val = sizeof(struct tpacket3_hdr);
3297 break;
3298 default:
3299 return -EINVAL;
3300 }
3301 break;
3302 case PACKET_RESERVE:
3303 val = po->tp_reserve;
3304 break;
3305 case PACKET_LOSS:
3306 val = po->tp_loss;
3307 break;
3308 case PACKET_TIMESTAMP:
3309 val = po->tp_tstamp;
3310 break;
3311 case PACKET_FANOUT:
3312 val = (po->fanout ?
3313 ((u32)po->fanout->id |
3314 ((u32)po->fanout->type << 16)) :
3315 0);
3316 break;
3317 default:
3318 return -ENOPROTOOPT;
3319 }
3320
3321 if (len > lv)
3322 len = lv;
3323 if (put_user(len, optlen))
3324 return -EFAULT;
3325 if (copy_to_user(optval, data, len))
3326 return -EFAULT;
3327 return 0;
3328}
3329
3330
3331static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3332{
3333 struct sock *sk;
3334 struct hlist_node *node;
3335 struct net_device *dev = data;
3336 struct net *net = dev_net(dev);
3337
3338 rcu_read_lock();
3339 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3340 struct packet_sock *po = pkt_sk(sk);
3341
3342 switch (msg) {
3343 case NETDEV_UNREGISTER:
3344 if (po->mclist)
3345 packet_dev_mclist(dev, po->mclist, -1);
3346 /* fallthrough */
3347
3348 case NETDEV_DOWN:
3349 if (dev->ifindex == po->ifindex) {
3350 spin_lock(&po->bind_lock);
3351 if (po->running) {
3352 __unregister_prot_hook(sk, false);
3353 sk->sk_err = ENETDOWN;
3354 if (!sock_flag(sk, SOCK_DEAD))
3355 sk->sk_error_report(sk);
3356 }
3357 if (msg == NETDEV_UNREGISTER) {
3358 po->ifindex = -1;
3359 if (po->prot_hook.dev)
3360 dev_put(po->prot_hook.dev);
3361 po->prot_hook.dev = NULL;
3362 }
3363 spin_unlock(&po->bind_lock);
3364 }
3365 break;
3366 case NETDEV_UP:
3367 if (dev->ifindex == po->ifindex) {
3368 spin_lock(&po->bind_lock);
3369 if (po->num)
3370 register_prot_hook(sk);
3371 spin_unlock(&po->bind_lock);
3372 }
3373 break;
3374 }
3375 }
3376 rcu_read_unlock();
3377 return NOTIFY_DONE;
3378}
3379
3380
3381static int packet_ioctl(struct socket *sock, unsigned int cmd,
3382 unsigned long arg)
3383{
3384 struct sock *sk = sock->sk;
3385
3386 switch (cmd) {
3387 case SIOCOUTQ:
3388 {
3389 int amount = sk_wmem_alloc_get(sk);
3390
3391 return put_user(amount, (int __user *)arg);
3392 }
3393 case SIOCINQ:
3394 {
3395 struct sk_buff *skb;
3396 int amount = 0;
3397
3398 spin_lock_bh(&sk->sk_receive_queue.lock);
3399 skb = skb_peek(&sk->sk_receive_queue);
3400 if (skb)
3401 amount = skb->len;
3402 spin_unlock_bh(&sk->sk_receive_queue.lock);
3403 return put_user(amount, (int __user *)arg);
3404 }
3405 case SIOCGSTAMP:
3406 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3407 case SIOCGSTAMPNS:
3408 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3409
3410#ifdef CONFIG_INET
3411 case SIOCADDRT:
3412 case SIOCDELRT:
3413 case SIOCDARP:
3414 case SIOCGARP:
3415 case SIOCSARP:
3416 case SIOCGIFADDR:
3417 case SIOCSIFADDR:
3418 case SIOCGIFBRDADDR:
3419 case SIOCSIFBRDADDR:
3420 case SIOCGIFNETMASK:
3421 case SIOCSIFNETMASK:
3422 case SIOCGIFDSTADDR:
3423 case SIOCSIFDSTADDR:
3424 case SIOCSIFFLAGS:
3425 return inet_dgram_ops.ioctl(sock, cmd, arg);
3426#endif
3427
3428 default:
3429 return -ENOIOCTLCMD;
3430 }
3431 return 0;
3432}
3433
3434static unsigned int packet_poll(struct file *file, struct socket *sock,
3435 poll_table *wait)
3436{
3437 struct sock *sk = sock->sk;
3438 struct packet_sock *po = pkt_sk(sk);
3439 unsigned int mask = datagram_poll(file, sock, wait);
3440
3441 spin_lock_bh(&sk->sk_receive_queue.lock);
3442 if (po->rx_ring.pg_vec) {
3443 if (!packet_previous_rx_frame(po, &po->rx_ring,
3444 TP_STATUS_KERNEL))
3445 mask |= POLLIN | POLLRDNORM;
3446 }
3447 spin_unlock_bh(&sk->sk_receive_queue.lock);
3448 spin_lock_bh(&sk->sk_write_queue.lock);
3449 if (po->tx_ring.pg_vec) {
3450 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3451 mask |= POLLOUT | POLLWRNORM;
3452 }
3453 spin_unlock_bh(&sk->sk_write_queue.lock);
3454 return mask;
3455}
3456
3457
3458/* Dirty? Well, I still did not learn better way to account
3459 * for user mmaps.
3460 */
3461
3462static void packet_mm_open(struct vm_area_struct *vma)
3463{
3464 struct file *file = vma->vm_file;
3465 struct socket *sock = file->private_data;
3466 struct sock *sk = sock->sk;
3467
3468 if (sk)
3469 atomic_inc(&pkt_sk(sk)->mapped);
3470}
3471
3472static void packet_mm_close(struct vm_area_struct *vma)
3473{
3474 struct file *file = vma->vm_file;
3475 struct socket *sock = file->private_data;
3476 struct sock *sk = sock->sk;
3477
3478 if (sk)
3479 atomic_dec(&pkt_sk(sk)->mapped);
3480}
3481
3482static const struct vm_operations_struct packet_mmap_ops = {
3483 .open = packet_mm_open,
3484 .close = packet_mm_close,
3485};
3486
3487static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3488 unsigned int len)
3489{
3490 int i;
3491
3492 for (i = 0; i < len; i++) {
3493 if (likely(pg_vec[i].buffer)) {
3494 if (is_vmalloc_addr(pg_vec[i].buffer))
3495 vfree(pg_vec[i].buffer);
3496 else
3497 free_pages((unsigned long)pg_vec[i].buffer,
3498 order);
3499 pg_vec[i].buffer = NULL;
3500 }
3501 }
3502 kfree(pg_vec);
3503}
3504
3505static char *alloc_one_pg_vec_page(unsigned long order)
3506{
3507 char *buffer = NULL;
3508 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3509 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3510
3511 buffer = (char *) __get_free_pages(gfp_flags, order);
3512
3513 if (buffer)
3514 return buffer;
3515
3516 /*
3517 * __get_free_pages failed, fall back to vmalloc
3518 */
3519 buffer = vzalloc((1 << order) * PAGE_SIZE);
3520
3521 if (buffer)
3522 return buffer;
3523
3524 /*
3525 * vmalloc failed, lets dig into swap here
3526 */
3527 gfp_flags &= ~__GFP_NORETRY;
3528 buffer = (char *)__get_free_pages(gfp_flags, order);
3529 if (buffer)
3530 return buffer;
3531
3532 /*
3533 * complete and utter failure
3534 */
3535 return NULL;
3536}
3537
3538static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3539{
3540 unsigned int block_nr = req->tp_block_nr;
3541 struct pgv *pg_vec;
3542 int i;
3543
3544 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3545 if (unlikely(!pg_vec))
3546 goto out;
3547
3548 for (i = 0; i < block_nr; i++) {
3549 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3550 if (unlikely(!pg_vec[i].buffer))
3551 goto out_free_pgvec;
3552 }
3553
3554out:
3555 return pg_vec;
3556
3557out_free_pgvec:
3558 free_pg_vec(pg_vec, order, block_nr);
3559 pg_vec = NULL;
3560 goto out;
3561}
3562
3563static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3564 int closing, int tx_ring)
3565{
3566 struct pgv *pg_vec = NULL;
3567 struct packet_sock *po = pkt_sk(sk);
3568 int was_running, order = 0;
3569 struct packet_ring_buffer *rb;
3570 struct sk_buff_head *rb_queue;
3571 __be16 num;
3572 int err = -EINVAL;
3573 /* Added to avoid minimal code churn */
3574 struct tpacket_req *req = &req_u->req;
3575
3576 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3577 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3578 WARN(1, "Tx-ring is not supported.\n");
3579 goto out;
3580 }
3581
3582 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3583 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3584
3585 err = -EBUSY;
3586 if (!closing) {
3587 if (atomic_read(&po->mapped))
3588 goto out;
3589 if (atomic_read(&rb->pending))
3590 goto out;
3591 }
3592
3593 if (req->tp_block_nr) {
3594 /* Sanity tests and some calculations */
3595 err = -EBUSY;
3596 if (unlikely(rb->pg_vec))
3597 goto out;
3598
3599 switch (po->tp_version) {
3600 case TPACKET_V1:
3601 po->tp_hdrlen = TPACKET_HDRLEN;
3602 break;
3603 case TPACKET_V2:
3604 po->tp_hdrlen = TPACKET2_HDRLEN;
3605 break;
3606 case TPACKET_V3:
3607 po->tp_hdrlen = TPACKET3_HDRLEN;
3608 break;
3609 }
3610
3611 err = -EINVAL;
3612 if (unlikely((int)req->tp_block_size <= 0))
3613 goto out;
3614 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3615 goto out;
3616 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3617 po->tp_reserve))
3618 goto out;
3619 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3620 goto out;
3621
3622 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3623 if (unlikely(rb->frames_per_block <= 0))
3624 goto out;
3625 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3626 req->tp_frame_nr))
3627 goto out;
3628
3629 err = -ENOMEM;
3630 order = get_order(req->tp_block_size);
3631 pg_vec = alloc_pg_vec(req, order);
3632 if (unlikely(!pg_vec))
3633 goto out;
3634 switch (po->tp_version) {
3635 case TPACKET_V3:
3636 /* Transmit path is not supported. We checked
3637 * it above but just being paranoid
3638 */
3639 if (!tx_ring)
3640 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3641 break;
3642 default:
3643 break;
3644 }
3645 }
3646 /* Done */
3647 else {
3648 err = -EINVAL;
3649 if (unlikely(req->tp_frame_nr))
3650 goto out;
3651 }
3652
3653 lock_sock(sk);
3654
3655 /* Detach socket from network */
3656 spin_lock(&po->bind_lock);
3657 was_running = po->running;
3658 num = po->num;
3659 if (was_running) {
3660 po->num = 0;
3661 __unregister_prot_hook(sk, false);
3662 }
3663 spin_unlock(&po->bind_lock);
3664
3665 synchronize_net();
3666
3667 err = -EBUSY;
3668 mutex_lock(&po->pg_vec_lock);
3669 if (closing || atomic_read(&po->mapped) == 0) {
3670 err = 0;
3671 spin_lock_bh(&rb_queue->lock);
3672 swap(rb->pg_vec, pg_vec);
3673 rb->frame_max = (req->tp_frame_nr - 1);
3674 rb->head = 0;
3675 rb->frame_size = req->tp_frame_size;
3676 spin_unlock_bh(&rb_queue->lock);
3677
3678 swap(rb->pg_vec_order, order);
3679 swap(rb->pg_vec_len, req->tp_block_nr);
3680
3681 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3682 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3683 tpacket_rcv : packet_rcv;
3684 skb_queue_purge(rb_queue);
3685 if (atomic_read(&po->mapped))
3686 pr_err("packet_mmap: vma is busy: %d\n",
3687 atomic_read(&po->mapped));
3688 }
3689 mutex_unlock(&po->pg_vec_lock);
3690
3691 spin_lock(&po->bind_lock);
3692 if (was_running) {
3693 po->num = num;
3694 register_prot_hook(sk);
3695 }
3696 spin_unlock(&po->bind_lock);
3697 if (closing && (po->tp_version > TPACKET_V2)) {
3698 /* Because we don't support block-based V3 on tx-ring */
3699 if (!tx_ring)
3700 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3701 }
3702 release_sock(sk);
3703
3704 if (pg_vec)
3705 free_pg_vec(pg_vec, order, req->tp_block_nr);
3706out:
3707 return err;
3708}
3709
3710static int packet_mmap(struct file *file, struct socket *sock,
3711 struct vm_area_struct *vma)
3712{
3713 struct sock *sk = sock->sk;
3714 struct packet_sock *po = pkt_sk(sk);
3715 unsigned long size, expected_size;
3716 struct packet_ring_buffer *rb;
3717 unsigned long start;
3718 int err = -EINVAL;
3719 int i;
3720
3721 if (vma->vm_pgoff)
3722 return -EINVAL;
3723
3724 mutex_lock(&po->pg_vec_lock);
3725
3726 expected_size = 0;
3727 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3728 if (rb->pg_vec) {
3729 expected_size += rb->pg_vec_len
3730 * rb->pg_vec_pages
3731 * PAGE_SIZE;
3732 }
3733 }
3734
3735 if (expected_size == 0)
3736 goto out;
3737
3738 size = vma->vm_end - vma->vm_start;
3739 if (size != expected_size)
3740 goto out;
3741
3742 start = vma->vm_start;
3743 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3744 if (rb->pg_vec == NULL)
3745 continue;
3746
3747 for (i = 0; i < rb->pg_vec_len; i++) {
3748 struct page *page;
3749 void *kaddr = rb->pg_vec[i].buffer;
3750 int pg_num;
3751
3752 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3753 page = pgv_to_page(kaddr);
3754 err = vm_insert_page(vma, start, page);
3755 if (unlikely(err))
3756 goto out;
3757 start += PAGE_SIZE;
3758 kaddr += PAGE_SIZE;
3759 }
3760 }
3761 }
3762
3763 atomic_inc(&po->mapped);
3764 vma->vm_ops = &packet_mmap_ops;
3765 err = 0;
3766
3767out:
3768 mutex_unlock(&po->pg_vec_lock);
3769 return err;
3770}
3771
3772static const struct proto_ops packet_ops_spkt = {
3773 .family = PF_PACKET,
3774 .owner = THIS_MODULE,
3775 .release = packet_release,
3776 .bind = packet_bind_spkt,
3777 .connect = sock_no_connect,
3778 .socketpair = sock_no_socketpair,
3779 .accept = sock_no_accept,
3780 .getname = packet_getname_spkt,
3781 .poll = datagram_poll,
3782 .ioctl = packet_ioctl,
3783 .listen = sock_no_listen,
3784 .shutdown = sock_no_shutdown,
3785 .setsockopt = sock_no_setsockopt,
3786 .getsockopt = sock_no_getsockopt,
3787 .sendmsg = packet_sendmsg_spkt,
3788 .recvmsg = packet_recvmsg,
3789 .mmap = sock_no_mmap,
3790 .sendpage = sock_no_sendpage,
3791};
3792
3793static const struct proto_ops packet_ops = {
3794 .family = PF_PACKET,
3795 .owner = THIS_MODULE,
3796 .release = packet_release,
3797 .bind = packet_bind,
3798 .connect = sock_no_connect,
3799 .socketpair = sock_no_socketpair,
3800 .accept = sock_no_accept,
3801 .getname = packet_getname,
3802 .poll = packet_poll,
3803 .ioctl = packet_ioctl,
3804 .listen = sock_no_listen,
3805 .shutdown = sock_no_shutdown,
3806 .setsockopt = packet_setsockopt,
3807 .getsockopt = packet_getsockopt,
3808 .sendmsg = packet_sendmsg,
3809 .recvmsg = packet_recvmsg,
3810 .mmap = packet_mmap,
3811 .sendpage = sock_no_sendpage,
3812};
3813
3814static const struct net_proto_family packet_family_ops = {
3815 .family = PF_PACKET,
3816 .create = packet_create,
3817 .owner = THIS_MODULE,
3818};
3819
3820static struct notifier_block packet_netdev_notifier = {
3821 .notifier_call = packet_notifier,
3822};
3823
3824#ifdef CONFIG_PROC_FS
3825
3826static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3827 __acquires(RCU)
3828{
3829 struct net *net = seq_file_net(seq);
3830
3831 rcu_read_lock();
3832 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3833}
3834
3835static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3836{
3837 struct net *net = seq_file_net(seq);
3838 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3839}
3840
3841static void packet_seq_stop(struct seq_file *seq, void *v)
3842 __releases(RCU)
3843{
3844 rcu_read_unlock();
3845}
3846
3847static int packet_seq_show(struct seq_file *seq, void *v)
3848{
3849 if (v == SEQ_START_TOKEN)
3850 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3851 else {
3852 struct sock *s = sk_entry(v);
3853 const struct packet_sock *po = pkt_sk(s);
3854
3855 seq_printf(seq,
3856 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3857 s,
3858 atomic_read(&s->sk_refcnt),
3859 s->sk_type,
3860 ntohs(po->num),
3861 po->ifindex,
3862 po->running,
3863 atomic_read(&s->sk_rmem_alloc),
3864 sock_i_uid(s),
3865 sock_i_ino(s));
3866 }
3867
3868 return 0;
3869}
3870
3871static const struct seq_operations packet_seq_ops = {
3872 .start = packet_seq_start,
3873 .next = packet_seq_next,
3874 .stop = packet_seq_stop,
3875 .show = packet_seq_show,
3876};
3877
3878static int packet_seq_open(struct inode *inode, struct file *file)
3879{
3880 return seq_open_net(inode, file, &packet_seq_ops,
3881 sizeof(struct seq_net_private));
3882}
3883
3884static const struct file_operations packet_seq_fops = {
3885 .owner = THIS_MODULE,
3886 .open = packet_seq_open,
3887 .read = seq_read,
3888 .llseek = seq_lseek,
3889 .release = seq_release_net,
3890};
3891
3892#endif
3893
3894static int __net_init packet_net_init(struct net *net)
3895{
3896 spin_lock_init(&net->packet.sklist_lock);
3897 INIT_HLIST_HEAD(&net->packet.sklist);
3898
3899 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3900 return -ENOMEM;
3901
3902 return 0;
3903}
3904
3905static void __net_exit packet_net_exit(struct net *net)
3906{
3907 proc_net_remove(net, "packet");
3908}
3909
3910static struct pernet_operations packet_net_ops = {
3911 .init = packet_net_init,
3912 .exit = packet_net_exit,
3913};
3914
3915
3916static void __exit packet_exit(void)
3917{
3918 unregister_netdevice_notifier(&packet_netdev_notifier);
3919 unregister_pernet_subsys(&packet_net_ops);
3920 sock_unregister(PF_PACKET);
3921 proto_unregister(&packet_proto);
3922}
3923
3924static int __init packet_init(void)
3925{
3926 int rc = proto_register(&packet_proto, 0);
3927
3928 if (rc != 0)
3929 goto out;
3930
3931 sock_register(&packet_family_ops);
3932 register_pernet_subsys(&packet_net_ops);
3933 register_netdevice_notifier(&packet_netdev_notifier);
3934out:
3935 return rc;
3936}
3937
3938module_init(packet_init);
3939module_exit(packet_exit);
3940MODULE_LICENSE("GPL");
3941MODULE_ALIAS_NETPROTO(PF_PACKET);