Loading...
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#include <linux/percpu.h>
92#ifdef CONFIG_INET
93#include <net/inet_common.h>
94#endif
95#include <linux/bpf.h>
96
97#include "internal.h"
98
99/*
100 Assumptions:
101 - if device has no dev->hard_header routine, it adds and removes ll header
102 inside itself. In this case ll header is invisible outside of device,
103 but higher levels still should reserve dev->hard_header_len.
104 Some devices are enough clever to reallocate skb, when header
105 will not fit to reserved space (tunnel), another ones are silly
106 (PPP).
107 - packet socket receives packets with pulled ll header,
108 so that SOCK_RAW should push it back.
109
110On receive:
111-----------
112
113Incoming, dev->hard_header!=NULL
114 mac_header -> ll header
115 data -> data
116
117Outgoing, dev->hard_header!=NULL
118 mac_header -> ll header
119 data -> ll header
120
121Incoming, dev->hard_header==NULL
122 mac_header -> UNKNOWN position. It is very likely, that it points to ll
123 header. PPP makes it, that is wrong, because introduce
124 assymetry between rx and tx paths.
125 data -> data
126
127Outgoing, dev->hard_header==NULL
128 mac_header -> data. ll header is still not built!
129 data -> data
130
131Resume
132 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
133
134
135On transmit:
136------------
137
138dev->hard_header != NULL
139 mac_header -> ll header
140 data -> ll header
141
142dev->hard_header == NULL (ll header is added by device, we cannot control it)
143 mac_header -> data
144 data -> data
145
146 We should set nh.raw on output to correct posistion,
147 packet classifier depends on it.
148 */
149
150/* Private packet socket structures. */
151
152/* identical to struct packet_mreq except it has
153 * a longer address field.
154 */
155struct packet_mreq_max {
156 int mr_ifindex;
157 unsigned short mr_type;
158 unsigned short mr_alen;
159 unsigned char mr_address[MAX_ADDR_LEN];
160};
161
162union tpacket_uhdr {
163 struct tpacket_hdr *h1;
164 struct tpacket2_hdr *h2;
165 struct tpacket3_hdr *h3;
166 void *raw;
167};
168
169static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
170 int closing, int tx_ring);
171
172#define V3_ALIGNMENT (8)
173
174#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175
176#define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178
179#define PGV_FROM_VMALLOC 1
180
181#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
182#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
183#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
184#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
185#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
186#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
187#define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
188
189struct packet_sock;
190static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
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_kbdq_core *,
199 struct tpacket_block_desc *);
200static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
201 struct packet_sock *);
202static void prb_retire_current_block(struct tpacket_kbdq_core *,
203 struct packet_sock *, unsigned int status);
204static int prb_queue_frozen(struct tpacket_kbdq_core *);
205static void prb_open_block(struct tpacket_kbdq_core *,
206 struct tpacket_block_desc *);
207static void prb_retire_rx_blk_timer_expired(unsigned long);
208static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
209static void prb_init_blk_timer(struct packet_sock *,
210 struct tpacket_kbdq_core *,
211 void (*func) (unsigned long));
212static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
213static void prb_clear_rxhash(struct tpacket_kbdq_core *,
214 struct tpacket3_hdr *);
215static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
216 struct tpacket3_hdr *);
217static void packet_flush_mclist(struct sock *sk);
218
219struct packet_skb_cb {
220 union {
221 struct sockaddr_pkt pkt;
222 union {
223 /* Trick: alias skb original length with
224 * ll.sll_family and ll.protocol in order
225 * to save room.
226 */
227 unsigned int origlen;
228 struct sockaddr_ll ll;
229 };
230 } sa;
231};
232
233#define vio_le() virtio_legacy_is_little_endian()
234
235#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
236
237#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
238#define GET_PBLOCK_DESC(x, bid) \
239 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
240#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
241 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
242#define GET_NEXT_PRB_BLK_NUM(x) \
243 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
244 ((x)->kactive_blk_num+1) : 0)
245
246static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
247static void __fanout_link(struct sock *sk, struct packet_sock *po);
248
249static int packet_direct_xmit(struct sk_buff *skb)
250{
251 struct net_device *dev = skb->dev;
252 netdev_features_t features;
253 struct netdev_queue *txq;
254 int ret = NETDEV_TX_BUSY;
255
256 if (unlikely(!netif_running(dev) ||
257 !netif_carrier_ok(dev)))
258 goto drop;
259
260 features = netif_skb_features(skb);
261 if (skb_needs_linearize(skb, features) &&
262 __skb_linearize(skb))
263 goto drop;
264
265 txq = skb_get_tx_queue(dev, skb);
266
267 local_bh_disable();
268
269 HARD_TX_LOCK(dev, txq, smp_processor_id());
270 if (!netif_xmit_frozen_or_drv_stopped(txq))
271 ret = netdev_start_xmit(skb, dev, txq, false);
272 HARD_TX_UNLOCK(dev, txq);
273
274 local_bh_enable();
275
276 if (!dev_xmit_complete(ret))
277 kfree_skb(skb);
278
279 return ret;
280drop:
281 atomic_long_inc(&dev->tx_dropped);
282 kfree_skb(skb);
283 return NET_XMIT_DROP;
284}
285
286static struct net_device *packet_cached_dev_get(struct packet_sock *po)
287{
288 struct net_device *dev;
289
290 rcu_read_lock();
291 dev = rcu_dereference(po->cached_dev);
292 if (likely(dev))
293 dev_hold(dev);
294 rcu_read_unlock();
295
296 return dev;
297}
298
299static void packet_cached_dev_assign(struct packet_sock *po,
300 struct net_device *dev)
301{
302 rcu_assign_pointer(po->cached_dev, dev);
303}
304
305static void packet_cached_dev_reset(struct packet_sock *po)
306{
307 RCU_INIT_POINTER(po->cached_dev, NULL);
308}
309
310static bool packet_use_direct_xmit(const struct packet_sock *po)
311{
312 return po->xmit == packet_direct_xmit;
313}
314
315static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
316{
317 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
318}
319
320static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
321{
322 const struct net_device_ops *ops = dev->netdev_ops;
323 u16 queue_index;
324
325 if (ops->ndo_select_queue) {
326 queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 __packet_pick_tx_queue);
328 queue_index = netdev_cap_txqueue(dev, queue_index);
329 } else {
330 queue_index = __packet_pick_tx_queue(dev, skb);
331 }
332
333 skb_set_queue_mapping(skb, queue_index);
334}
335
336/* register_prot_hook must be invoked with the po->bind_lock held,
337 * or from a context in which asynchronous accesses to the packet
338 * socket is not possible (packet_create()).
339 */
340static void register_prot_hook(struct sock *sk)
341{
342 struct packet_sock *po = pkt_sk(sk);
343
344 if (!po->running) {
345 if (po->fanout)
346 __fanout_link(sk, po);
347 else
348 dev_add_pack(&po->prot_hook);
349
350 sock_hold(sk);
351 po->running = 1;
352 }
353}
354
355/* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356 * held. If the sync parameter is true, we will temporarily drop
357 * the po->bind_lock and do a synchronize_net to make sure no
358 * asynchronous packet processing paths still refer to the elements
359 * of po->prot_hook. If the sync parameter is false, it is the
360 * callers responsibility to take care of this.
361 */
362static void __unregister_prot_hook(struct sock *sk, bool sync)
363{
364 struct packet_sock *po = pkt_sk(sk);
365
366 po->running = 0;
367
368 if (po->fanout)
369 __fanout_unlink(sk, po);
370 else
371 __dev_remove_pack(&po->prot_hook);
372
373 __sock_put(sk);
374
375 if (sync) {
376 spin_unlock(&po->bind_lock);
377 synchronize_net();
378 spin_lock(&po->bind_lock);
379 }
380}
381
382static void unregister_prot_hook(struct sock *sk, bool sync)
383{
384 struct packet_sock *po = pkt_sk(sk);
385
386 if (po->running)
387 __unregister_prot_hook(sk, sync);
388}
389
390static inline struct page * __pure pgv_to_page(void *addr)
391{
392 if (is_vmalloc_addr(addr))
393 return vmalloc_to_page(addr);
394 return virt_to_page(addr);
395}
396
397static void __packet_set_status(struct packet_sock *po, void *frame, int status)
398{
399 union tpacket_uhdr h;
400
401 h.raw = frame;
402 switch (po->tp_version) {
403 case TPACKET_V1:
404 h.h1->tp_status = status;
405 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
406 break;
407 case TPACKET_V2:
408 h.h2->tp_status = status;
409 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
410 break;
411 case TPACKET_V3:
412 default:
413 WARN(1, "TPACKET version not supported.\n");
414 BUG();
415 }
416
417 smp_wmb();
418}
419
420static int __packet_get_status(struct packet_sock *po, void *frame)
421{
422 union tpacket_uhdr h;
423
424 smp_rmb();
425
426 h.raw = frame;
427 switch (po->tp_version) {
428 case TPACKET_V1:
429 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
430 return h.h1->tp_status;
431 case TPACKET_V2:
432 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
433 return h.h2->tp_status;
434 case TPACKET_V3:
435 default:
436 WARN(1, "TPACKET version not supported.\n");
437 BUG();
438 return 0;
439 }
440}
441
442static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
443 unsigned int flags)
444{
445 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
446
447 if (shhwtstamps &&
448 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
449 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
450 return TP_STATUS_TS_RAW_HARDWARE;
451
452 if (ktime_to_timespec_cond(skb->tstamp, ts))
453 return TP_STATUS_TS_SOFTWARE;
454
455 return 0;
456}
457
458static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
459 struct sk_buff *skb)
460{
461 union tpacket_uhdr h;
462 struct timespec ts;
463 __u32 ts_status;
464
465 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
466 return 0;
467
468 h.raw = frame;
469 switch (po->tp_version) {
470 case TPACKET_V1:
471 h.h1->tp_sec = ts.tv_sec;
472 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
473 break;
474 case TPACKET_V2:
475 h.h2->tp_sec = ts.tv_sec;
476 h.h2->tp_nsec = ts.tv_nsec;
477 break;
478 case TPACKET_V3:
479 default:
480 WARN(1, "TPACKET version not supported.\n");
481 BUG();
482 }
483
484 /* one flush is safe, as both fields always lie on the same cacheline */
485 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
486 smp_wmb();
487
488 return ts_status;
489}
490
491static void *packet_lookup_frame(struct packet_sock *po,
492 struct packet_ring_buffer *rb,
493 unsigned int position,
494 int status)
495{
496 unsigned int pg_vec_pos, frame_offset;
497 union tpacket_uhdr h;
498
499 pg_vec_pos = position / rb->frames_per_block;
500 frame_offset = position % rb->frames_per_block;
501
502 h.raw = rb->pg_vec[pg_vec_pos].buffer +
503 (frame_offset * rb->frame_size);
504
505 if (status != __packet_get_status(po, h.raw))
506 return NULL;
507
508 return h.raw;
509}
510
511static void *packet_current_frame(struct packet_sock *po,
512 struct packet_ring_buffer *rb,
513 int status)
514{
515 return packet_lookup_frame(po, rb, rb->head, status);
516}
517
518static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
519{
520 del_timer_sync(&pkc->retire_blk_timer);
521}
522
523static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
524 struct sk_buff_head *rb_queue)
525{
526 struct tpacket_kbdq_core *pkc;
527
528 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
529
530 spin_lock_bh(&rb_queue->lock);
531 pkc->delete_blk_timer = 1;
532 spin_unlock_bh(&rb_queue->lock);
533
534 prb_del_retire_blk_timer(pkc);
535}
536
537static void prb_init_blk_timer(struct packet_sock *po,
538 struct tpacket_kbdq_core *pkc,
539 void (*func) (unsigned long))
540{
541 init_timer(&pkc->retire_blk_timer);
542 pkc->retire_blk_timer.data = (long)po;
543 pkc->retire_blk_timer.function = func;
544 pkc->retire_blk_timer.expires = jiffies;
545}
546
547static void prb_setup_retire_blk_timer(struct packet_sock *po)
548{
549 struct tpacket_kbdq_core *pkc;
550
551 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
552 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
553}
554
555static int prb_calc_retire_blk_tmo(struct packet_sock *po,
556 int blk_size_in_bytes)
557{
558 struct net_device *dev;
559 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
560 struct ethtool_link_ksettings ecmd;
561 int err;
562
563 rtnl_lock();
564 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
565 if (unlikely(!dev)) {
566 rtnl_unlock();
567 return DEFAULT_PRB_RETIRE_TOV;
568 }
569 err = __ethtool_get_link_ksettings(dev, &ecmd);
570 rtnl_unlock();
571 if (!err) {
572 /*
573 * If the link speed is so slow you don't really
574 * need to worry about perf anyways
575 */
576 if (ecmd.base.speed < SPEED_1000 ||
577 ecmd.base.speed == SPEED_UNKNOWN) {
578 return DEFAULT_PRB_RETIRE_TOV;
579 } else {
580 msec = 1;
581 div = ecmd.base.speed / 1000;
582 }
583 }
584
585 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
586
587 if (div)
588 mbits /= div;
589
590 tmo = mbits * msec;
591
592 if (div)
593 return tmo+1;
594 return tmo;
595}
596
597static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
598 union tpacket_req_u *req_u)
599{
600 p1->feature_req_word = req_u->req3.tp_feature_req_word;
601}
602
603static void init_prb_bdqc(struct packet_sock *po,
604 struct packet_ring_buffer *rb,
605 struct pgv *pg_vec,
606 union tpacket_req_u *req_u)
607{
608 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
609 struct tpacket_block_desc *pbd;
610
611 memset(p1, 0x0, sizeof(*p1));
612
613 p1->knxt_seq_num = 1;
614 p1->pkbdq = pg_vec;
615 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
616 p1->pkblk_start = pg_vec[0].buffer;
617 p1->kblk_size = req_u->req3.tp_block_size;
618 p1->knum_blocks = req_u->req3.tp_block_nr;
619 p1->hdrlen = po->tp_hdrlen;
620 p1->version = po->tp_version;
621 p1->last_kactive_blk_num = 0;
622 po->stats.stats3.tp_freeze_q_cnt = 0;
623 if (req_u->req3.tp_retire_blk_tov)
624 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
625 else
626 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
627 req_u->req3.tp_block_size);
628 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
629 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
630
631 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
632 prb_init_ft_ops(p1, req_u);
633 prb_setup_retire_blk_timer(po);
634 prb_open_block(p1, pbd);
635}
636
637/* Do NOT update the last_blk_num first.
638 * Assumes sk_buff_head lock is held.
639 */
640static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
641{
642 mod_timer(&pkc->retire_blk_timer,
643 jiffies + pkc->tov_in_jiffies);
644 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
645}
646
647/*
648 * Timer logic:
649 * 1) We refresh the timer only when we open a block.
650 * By doing this we don't waste cycles refreshing the timer
651 * on packet-by-packet basis.
652 *
653 * With a 1MB block-size, on a 1Gbps line, it will take
654 * i) ~8 ms to fill a block + ii) memcpy etc.
655 * In this cut we are not accounting for the memcpy time.
656 *
657 * So, if the user sets the 'tmo' to 10ms then the timer
658 * will never fire while the block is still getting filled
659 * (which is what we want). However, the user could choose
660 * to close a block early and that's fine.
661 *
662 * But when the timer does fire, we check whether or not to refresh it.
663 * Since the tmo granularity is in msecs, it is not too expensive
664 * to refresh the timer, lets say every '8' msecs.
665 * Either the user can set the 'tmo' or we can derive it based on
666 * a) line-speed and b) block-size.
667 * prb_calc_retire_blk_tmo() calculates the tmo.
668 *
669 */
670static void prb_retire_rx_blk_timer_expired(unsigned long data)
671{
672 struct packet_sock *po = (struct packet_sock *)data;
673 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
674 unsigned int frozen;
675 struct tpacket_block_desc *pbd;
676
677 spin_lock(&po->sk.sk_receive_queue.lock);
678
679 frozen = prb_queue_frozen(pkc);
680 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
681
682 if (unlikely(pkc->delete_blk_timer))
683 goto out;
684
685 /* We only need to plug the race when the block is partially filled.
686 * tpacket_rcv:
687 * lock(); increment BLOCK_NUM_PKTS; unlock()
688 * copy_bits() is in progress ...
689 * timer fires on other cpu:
690 * we can't retire the current block because copy_bits
691 * is in progress.
692 *
693 */
694 if (BLOCK_NUM_PKTS(pbd)) {
695 while (atomic_read(&pkc->blk_fill_in_prog)) {
696 /* Waiting for skb_copy_bits to finish... */
697 cpu_relax();
698 }
699 }
700
701 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
702 if (!frozen) {
703 if (!BLOCK_NUM_PKTS(pbd)) {
704 /* An empty block. Just refresh the timer. */
705 goto refresh_timer;
706 }
707 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
708 if (!prb_dispatch_next_block(pkc, po))
709 goto refresh_timer;
710 else
711 goto out;
712 } else {
713 /* Case 1. Queue was frozen because user-space was
714 * lagging behind.
715 */
716 if (prb_curr_blk_in_use(pkc, pbd)) {
717 /*
718 * Ok, user-space is still behind.
719 * So just refresh the timer.
720 */
721 goto refresh_timer;
722 } else {
723 /* Case 2. queue was frozen,user-space caught up,
724 * now the link went idle && the timer fired.
725 * We don't have a block to close.So we open this
726 * block and restart the timer.
727 * opening a block thaws the queue,restarts timer
728 * Thawing/timer-refresh is a side effect.
729 */
730 prb_open_block(pkc, pbd);
731 goto out;
732 }
733 }
734 }
735
736refresh_timer:
737 _prb_refresh_rx_retire_blk_timer(pkc);
738
739out:
740 spin_unlock(&po->sk.sk_receive_queue.lock);
741}
742
743static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
744 struct tpacket_block_desc *pbd1, __u32 status)
745{
746 /* Flush everything minus the block header */
747
748#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
749 u8 *start, *end;
750
751 start = (u8 *)pbd1;
752
753 /* Skip the block header(we know header WILL fit in 4K) */
754 start += PAGE_SIZE;
755
756 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
757 for (; start < end; start += PAGE_SIZE)
758 flush_dcache_page(pgv_to_page(start));
759
760 smp_wmb();
761#endif
762
763 /* Now update the block status. */
764
765 BLOCK_STATUS(pbd1) = status;
766
767 /* Flush the block header */
768
769#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
770 start = (u8 *)pbd1;
771 flush_dcache_page(pgv_to_page(start));
772
773 smp_wmb();
774#endif
775}
776
777/*
778 * Side effect:
779 *
780 * 1) flush the block
781 * 2) Increment active_blk_num
782 *
783 * Note:We DONT refresh the timer on purpose.
784 * Because almost always the next block will be opened.
785 */
786static void prb_close_block(struct tpacket_kbdq_core *pkc1,
787 struct tpacket_block_desc *pbd1,
788 struct packet_sock *po, unsigned int stat)
789{
790 __u32 status = TP_STATUS_USER | stat;
791
792 struct tpacket3_hdr *last_pkt;
793 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
794 struct sock *sk = &po->sk;
795
796 if (po->stats.stats3.tp_drops)
797 status |= TP_STATUS_LOSING;
798
799 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
800 last_pkt->tp_next_offset = 0;
801
802 /* Get the ts of the last pkt */
803 if (BLOCK_NUM_PKTS(pbd1)) {
804 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
805 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
806 } else {
807 /* Ok, we tmo'd - so get the current time.
808 *
809 * It shouldn't really happen as we don't close empty
810 * blocks. See prb_retire_rx_blk_timer_expired().
811 */
812 struct timespec ts;
813 getnstimeofday(&ts);
814 h1->ts_last_pkt.ts_sec = ts.tv_sec;
815 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
816 }
817
818 smp_wmb();
819
820 /* Flush the block */
821 prb_flush_block(pkc1, pbd1, status);
822
823 sk->sk_data_ready(sk);
824
825 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
826}
827
828static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
829{
830 pkc->reset_pending_on_curr_blk = 0;
831}
832
833/*
834 * Side effect of opening a block:
835 *
836 * 1) prb_queue is thawed.
837 * 2) retire_blk_timer is refreshed.
838 *
839 */
840static void prb_open_block(struct tpacket_kbdq_core *pkc1,
841 struct tpacket_block_desc *pbd1)
842{
843 struct timespec ts;
844 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
845
846 smp_rmb();
847
848 /* We could have just memset this but we will lose the
849 * flexibility of making the priv area sticky
850 */
851
852 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
853 BLOCK_NUM_PKTS(pbd1) = 0;
854 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
855
856 getnstimeofday(&ts);
857
858 h1->ts_first_pkt.ts_sec = ts.tv_sec;
859 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
860
861 pkc1->pkblk_start = (char *)pbd1;
862 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
863
864 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
866
867 pbd1->version = pkc1->version;
868 pkc1->prev = pkc1->nxt_offset;
869 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
870
871 prb_thaw_queue(pkc1);
872 _prb_refresh_rx_retire_blk_timer(pkc1);
873
874 smp_wmb();
875}
876
877/*
878 * Queue freeze logic:
879 * 1) Assume tp_block_nr = 8 blocks.
880 * 2) At time 't0', user opens Rx ring.
881 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
882 * 4) user-space is either sleeping or processing block '0'.
883 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
884 * it will close block-7,loop around and try to fill block '0'.
885 * call-flow:
886 * __packet_lookup_frame_in_block
887 * prb_retire_current_block()
888 * prb_dispatch_next_block()
889 * |->(BLOCK_STATUS == USER) evaluates to true
890 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
891 * 6) Now there are two cases:
892 * 6.1) Link goes idle right after the queue is frozen.
893 * But remember, the last open_block() refreshed the timer.
894 * When this timer expires,it will refresh itself so that we can
895 * re-open block-0 in near future.
896 * 6.2) Link is busy and keeps on receiving packets. This is a simple
897 * case and __packet_lookup_frame_in_block will check if block-0
898 * is free and can now be re-used.
899 */
900static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
901 struct packet_sock *po)
902{
903 pkc->reset_pending_on_curr_blk = 1;
904 po->stats.stats3.tp_freeze_q_cnt++;
905}
906
907#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
908
909/*
910 * If the next block is free then we will dispatch it
911 * and return a good offset.
912 * Else, we will freeze the queue.
913 * So, caller must check the return value.
914 */
915static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
916 struct packet_sock *po)
917{
918 struct tpacket_block_desc *pbd;
919
920 smp_rmb();
921
922 /* 1. Get current block num */
923 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
924
925 /* 2. If this block is currently in_use then freeze the queue */
926 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
927 prb_freeze_queue(pkc, po);
928 return NULL;
929 }
930
931 /*
932 * 3.
933 * open this block and return the offset where the first packet
934 * needs to get stored.
935 */
936 prb_open_block(pkc, pbd);
937 return (void *)pkc->nxt_offset;
938}
939
940static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
941 struct packet_sock *po, unsigned int status)
942{
943 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
944
945 /* retire/close the current block */
946 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
947 /*
948 * Plug the case where copy_bits() is in progress on
949 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
950 * have space to copy the pkt in the current block and
951 * called prb_retire_current_block()
952 *
953 * We don't need to worry about the TMO case because
954 * the timer-handler already handled this case.
955 */
956 if (!(status & TP_STATUS_BLK_TMO)) {
957 while (atomic_read(&pkc->blk_fill_in_prog)) {
958 /* Waiting for skb_copy_bits to finish... */
959 cpu_relax();
960 }
961 }
962 prb_close_block(pkc, pbd, po, status);
963 return;
964 }
965}
966
967static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
968 struct tpacket_block_desc *pbd)
969{
970 return TP_STATUS_USER & BLOCK_STATUS(pbd);
971}
972
973static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
974{
975 return pkc->reset_pending_on_curr_blk;
976}
977
978static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
979{
980 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
981 atomic_dec(&pkc->blk_fill_in_prog);
982}
983
984static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
985 struct tpacket3_hdr *ppd)
986{
987 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
988}
989
990static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
991 struct tpacket3_hdr *ppd)
992{
993 ppd->hv1.tp_rxhash = 0;
994}
995
996static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
997 struct tpacket3_hdr *ppd)
998{
999 if (skb_vlan_tag_present(pkc->skb)) {
1000 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1001 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1002 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1003 } else {
1004 ppd->hv1.tp_vlan_tci = 0;
1005 ppd->hv1.tp_vlan_tpid = 0;
1006 ppd->tp_status = TP_STATUS_AVAILABLE;
1007 }
1008}
1009
1010static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1011 struct tpacket3_hdr *ppd)
1012{
1013 ppd->hv1.tp_padding = 0;
1014 prb_fill_vlan_info(pkc, ppd);
1015
1016 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1017 prb_fill_rxhash(pkc, ppd);
1018 else
1019 prb_clear_rxhash(pkc, ppd);
1020}
1021
1022static void prb_fill_curr_block(char *curr,
1023 struct tpacket_kbdq_core *pkc,
1024 struct tpacket_block_desc *pbd,
1025 unsigned int len)
1026{
1027 struct tpacket3_hdr *ppd;
1028
1029 ppd = (struct tpacket3_hdr *)curr;
1030 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1031 pkc->prev = curr;
1032 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1033 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034 BLOCK_NUM_PKTS(pbd) += 1;
1035 atomic_inc(&pkc->blk_fill_in_prog);
1036 prb_run_all_ft_ops(pkc, ppd);
1037}
1038
1039/* Assumes caller has the sk->rx_queue.lock */
1040static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1041 struct sk_buff *skb,
1042 int status,
1043 unsigned int len
1044 )
1045{
1046 struct tpacket_kbdq_core *pkc;
1047 struct tpacket_block_desc *pbd;
1048 char *curr, *end;
1049
1050 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1051 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1052
1053 /* Queue is frozen when user space is lagging behind */
1054 if (prb_queue_frozen(pkc)) {
1055 /*
1056 * Check if that last block which caused the queue to freeze,
1057 * is still in_use by user-space.
1058 */
1059 if (prb_curr_blk_in_use(pkc, pbd)) {
1060 /* Can't record this packet */
1061 return NULL;
1062 } else {
1063 /*
1064 * Ok, the block was released by user-space.
1065 * Now let's open that block.
1066 * opening a block also thaws the queue.
1067 * Thawing is a side effect.
1068 */
1069 prb_open_block(pkc, pbd);
1070 }
1071 }
1072
1073 smp_mb();
1074 curr = pkc->nxt_offset;
1075 pkc->skb = skb;
1076 end = (char *)pbd + pkc->kblk_size;
1077
1078 /* first try the current block */
1079 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1080 prb_fill_curr_block(curr, pkc, pbd, len);
1081 return (void *)curr;
1082 }
1083
1084 /* Ok, close the current block */
1085 prb_retire_current_block(pkc, po, 0);
1086
1087 /* Now, try to dispatch the next block */
1088 curr = (char *)prb_dispatch_next_block(pkc, po);
1089 if (curr) {
1090 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1091 prb_fill_curr_block(curr, pkc, pbd, len);
1092 return (void *)curr;
1093 }
1094
1095 /*
1096 * No free blocks are available.user_space hasn't caught up yet.
1097 * Queue was just frozen and now this packet will get dropped.
1098 */
1099 return NULL;
1100}
1101
1102static void *packet_current_rx_frame(struct packet_sock *po,
1103 struct sk_buff *skb,
1104 int status, unsigned int len)
1105{
1106 char *curr = NULL;
1107 switch (po->tp_version) {
1108 case TPACKET_V1:
1109 case TPACKET_V2:
1110 curr = packet_lookup_frame(po, &po->rx_ring,
1111 po->rx_ring.head, status);
1112 return curr;
1113 case TPACKET_V3:
1114 return __packet_lookup_frame_in_block(po, skb, status, len);
1115 default:
1116 WARN(1, "TPACKET version not supported\n");
1117 BUG();
1118 return NULL;
1119 }
1120}
1121
1122static void *prb_lookup_block(struct packet_sock *po,
1123 struct packet_ring_buffer *rb,
1124 unsigned int idx,
1125 int status)
1126{
1127 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1128 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1129
1130 if (status != BLOCK_STATUS(pbd))
1131 return NULL;
1132 return pbd;
1133}
1134
1135static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1136{
1137 unsigned int prev;
1138 if (rb->prb_bdqc.kactive_blk_num)
1139 prev = rb->prb_bdqc.kactive_blk_num-1;
1140 else
1141 prev = rb->prb_bdqc.knum_blocks-1;
1142 return prev;
1143}
1144
1145/* Assumes caller has held the rx_queue.lock */
1146static void *__prb_previous_block(struct packet_sock *po,
1147 struct packet_ring_buffer *rb,
1148 int status)
1149{
1150 unsigned int previous = prb_previous_blk_num(rb);
1151 return prb_lookup_block(po, rb, previous, status);
1152}
1153
1154static void *packet_previous_rx_frame(struct packet_sock *po,
1155 struct packet_ring_buffer *rb,
1156 int status)
1157{
1158 if (po->tp_version <= TPACKET_V2)
1159 return packet_previous_frame(po, rb, status);
1160
1161 return __prb_previous_block(po, rb, status);
1162}
1163
1164static void packet_increment_rx_head(struct packet_sock *po,
1165 struct packet_ring_buffer *rb)
1166{
1167 switch (po->tp_version) {
1168 case TPACKET_V1:
1169 case TPACKET_V2:
1170 return packet_increment_head(rb);
1171 case TPACKET_V3:
1172 default:
1173 WARN(1, "TPACKET version not supported.\n");
1174 BUG();
1175 return;
1176 }
1177}
1178
1179static void *packet_previous_frame(struct packet_sock *po,
1180 struct packet_ring_buffer *rb,
1181 int status)
1182{
1183 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1184 return packet_lookup_frame(po, rb, previous, status);
1185}
1186
1187static void packet_increment_head(struct packet_ring_buffer *buff)
1188{
1189 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1190}
1191
1192static void packet_inc_pending(struct packet_ring_buffer *rb)
1193{
1194 this_cpu_inc(*rb->pending_refcnt);
1195}
1196
1197static void packet_dec_pending(struct packet_ring_buffer *rb)
1198{
1199 this_cpu_dec(*rb->pending_refcnt);
1200}
1201
1202static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1203{
1204 unsigned int refcnt = 0;
1205 int cpu;
1206
1207 /* We don't use pending refcount in rx_ring. */
1208 if (rb->pending_refcnt == NULL)
1209 return 0;
1210
1211 for_each_possible_cpu(cpu)
1212 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1213
1214 return refcnt;
1215}
1216
1217static int packet_alloc_pending(struct packet_sock *po)
1218{
1219 po->rx_ring.pending_refcnt = NULL;
1220
1221 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1222 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1223 return -ENOBUFS;
1224
1225 return 0;
1226}
1227
1228static void packet_free_pending(struct packet_sock *po)
1229{
1230 free_percpu(po->tx_ring.pending_refcnt);
1231}
1232
1233#define ROOM_POW_OFF 2
1234#define ROOM_NONE 0x0
1235#define ROOM_LOW 0x1
1236#define ROOM_NORMAL 0x2
1237
1238static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1239{
1240 int idx, len;
1241
1242 len = po->rx_ring.frame_max + 1;
1243 idx = po->rx_ring.head;
1244 if (pow_off)
1245 idx += len >> pow_off;
1246 if (idx >= len)
1247 idx -= len;
1248 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1249}
1250
1251static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1252{
1253 int idx, len;
1254
1255 len = po->rx_ring.prb_bdqc.knum_blocks;
1256 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1257 if (pow_off)
1258 idx += len >> pow_off;
1259 if (idx >= len)
1260 idx -= len;
1261 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1262}
1263
1264static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1265{
1266 struct sock *sk = &po->sk;
1267 int ret = ROOM_NONE;
1268
1269 if (po->prot_hook.func != tpacket_rcv) {
1270 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1271 - (skb ? skb->truesize : 0);
1272 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1273 return ROOM_NORMAL;
1274 else if (avail > 0)
1275 return ROOM_LOW;
1276 else
1277 return ROOM_NONE;
1278 }
1279
1280 if (po->tp_version == TPACKET_V3) {
1281 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1282 ret = ROOM_NORMAL;
1283 else if (__tpacket_v3_has_room(po, 0))
1284 ret = ROOM_LOW;
1285 } else {
1286 if (__tpacket_has_room(po, ROOM_POW_OFF))
1287 ret = ROOM_NORMAL;
1288 else if (__tpacket_has_room(po, 0))
1289 ret = ROOM_LOW;
1290 }
1291
1292 return ret;
1293}
1294
1295static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1296{
1297 int ret;
1298 bool has_room;
1299
1300 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1301 ret = __packet_rcv_has_room(po, skb);
1302 has_room = ret == ROOM_NORMAL;
1303 if (po->pressure == has_room)
1304 po->pressure = !has_room;
1305 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1306
1307 return ret;
1308}
1309
1310static void packet_sock_destruct(struct sock *sk)
1311{
1312 skb_queue_purge(&sk->sk_error_queue);
1313
1314 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1315 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1316
1317 if (!sock_flag(sk, SOCK_DEAD)) {
1318 pr_err("Attempt to release alive packet socket: %p\n", sk);
1319 return;
1320 }
1321
1322 sk_refcnt_debug_dec(sk);
1323}
1324
1325static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1326{
1327 u32 rxhash;
1328 int i, count = 0;
1329
1330 rxhash = skb_get_hash(skb);
1331 for (i = 0; i < ROLLOVER_HLEN; i++)
1332 if (po->rollover->history[i] == rxhash)
1333 count++;
1334
1335 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1336 return count > (ROLLOVER_HLEN >> 1);
1337}
1338
1339static unsigned int fanout_demux_hash(struct packet_fanout *f,
1340 struct sk_buff *skb,
1341 unsigned int num)
1342{
1343 return reciprocal_scale(skb_get_hash(skb), num);
1344}
1345
1346static unsigned int fanout_demux_lb(struct packet_fanout *f,
1347 struct sk_buff *skb,
1348 unsigned int num)
1349{
1350 unsigned int val = atomic_inc_return(&f->rr_cur);
1351
1352 return val % num;
1353}
1354
1355static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1356 struct sk_buff *skb,
1357 unsigned int num)
1358{
1359 return smp_processor_id() % num;
1360}
1361
1362static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1363 struct sk_buff *skb,
1364 unsigned int num)
1365{
1366 return prandom_u32_max(num);
1367}
1368
1369static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1370 struct sk_buff *skb,
1371 unsigned int idx, bool try_self,
1372 unsigned int num)
1373{
1374 struct packet_sock *po, *po_next, *po_skip = NULL;
1375 unsigned int i, j, room = ROOM_NONE;
1376
1377 po = pkt_sk(f->arr[idx]);
1378
1379 if (try_self) {
1380 room = packet_rcv_has_room(po, skb);
1381 if (room == ROOM_NORMAL ||
1382 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1383 return idx;
1384 po_skip = po;
1385 }
1386
1387 i = j = min_t(int, po->rollover->sock, num - 1);
1388 do {
1389 po_next = pkt_sk(f->arr[i]);
1390 if (po_next != po_skip && !po_next->pressure &&
1391 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1392 if (i != j)
1393 po->rollover->sock = i;
1394 atomic_long_inc(&po->rollover->num);
1395 if (room == ROOM_LOW)
1396 atomic_long_inc(&po->rollover->num_huge);
1397 return i;
1398 }
1399
1400 if (++i == num)
1401 i = 0;
1402 } while (i != j);
1403
1404 atomic_long_inc(&po->rollover->num_failed);
1405 return idx;
1406}
1407
1408static unsigned int fanout_demux_qm(struct packet_fanout *f,
1409 struct sk_buff *skb,
1410 unsigned int num)
1411{
1412 return skb_get_queue_mapping(skb) % num;
1413}
1414
1415static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1416 struct sk_buff *skb,
1417 unsigned int num)
1418{
1419 struct bpf_prog *prog;
1420 unsigned int ret = 0;
1421
1422 rcu_read_lock();
1423 prog = rcu_dereference(f->bpf_prog);
1424 if (prog)
1425 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1426 rcu_read_unlock();
1427
1428 return ret;
1429}
1430
1431static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1432{
1433 return f->flags & (flag >> 8);
1434}
1435
1436static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1437 struct packet_type *pt, struct net_device *orig_dev)
1438{
1439 struct packet_fanout *f = pt->af_packet_priv;
1440 unsigned int num = READ_ONCE(f->num_members);
1441 struct net *net = read_pnet(&f->net);
1442 struct packet_sock *po;
1443 unsigned int idx;
1444
1445 if (!net_eq(dev_net(dev), net) || !num) {
1446 kfree_skb(skb);
1447 return 0;
1448 }
1449
1450 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1451 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1452 if (!skb)
1453 return 0;
1454 }
1455 switch (f->type) {
1456 case PACKET_FANOUT_HASH:
1457 default:
1458 idx = fanout_demux_hash(f, skb, num);
1459 break;
1460 case PACKET_FANOUT_LB:
1461 idx = fanout_demux_lb(f, skb, num);
1462 break;
1463 case PACKET_FANOUT_CPU:
1464 idx = fanout_demux_cpu(f, skb, num);
1465 break;
1466 case PACKET_FANOUT_RND:
1467 idx = fanout_demux_rnd(f, skb, num);
1468 break;
1469 case PACKET_FANOUT_QM:
1470 idx = fanout_demux_qm(f, skb, num);
1471 break;
1472 case PACKET_FANOUT_ROLLOVER:
1473 idx = fanout_demux_rollover(f, skb, 0, false, num);
1474 break;
1475 case PACKET_FANOUT_CBPF:
1476 case PACKET_FANOUT_EBPF:
1477 idx = fanout_demux_bpf(f, skb, num);
1478 break;
1479 }
1480
1481 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1482 idx = fanout_demux_rollover(f, skb, idx, true, num);
1483
1484 po = pkt_sk(f->arr[idx]);
1485 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1486}
1487
1488DEFINE_MUTEX(fanout_mutex);
1489EXPORT_SYMBOL_GPL(fanout_mutex);
1490static LIST_HEAD(fanout_list);
1491
1492static void __fanout_link(struct sock *sk, struct packet_sock *po)
1493{
1494 struct packet_fanout *f = po->fanout;
1495
1496 spin_lock(&f->lock);
1497 f->arr[f->num_members] = sk;
1498 smp_wmb();
1499 f->num_members++;
1500 spin_unlock(&f->lock);
1501}
1502
1503static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1504{
1505 struct packet_fanout *f = po->fanout;
1506 int i;
1507
1508 spin_lock(&f->lock);
1509 for (i = 0; i < f->num_members; i++) {
1510 if (f->arr[i] == sk)
1511 break;
1512 }
1513 BUG_ON(i >= f->num_members);
1514 f->arr[i] = f->arr[f->num_members - 1];
1515 f->num_members--;
1516 spin_unlock(&f->lock);
1517}
1518
1519static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1520{
1521 if (sk->sk_family != PF_PACKET)
1522 return false;
1523
1524 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1525}
1526
1527static void fanout_init_data(struct packet_fanout *f)
1528{
1529 switch (f->type) {
1530 case PACKET_FANOUT_LB:
1531 atomic_set(&f->rr_cur, 0);
1532 break;
1533 case PACKET_FANOUT_CBPF:
1534 case PACKET_FANOUT_EBPF:
1535 RCU_INIT_POINTER(f->bpf_prog, NULL);
1536 break;
1537 }
1538}
1539
1540static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1541{
1542 struct bpf_prog *old;
1543
1544 spin_lock(&f->lock);
1545 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1546 rcu_assign_pointer(f->bpf_prog, new);
1547 spin_unlock(&f->lock);
1548
1549 if (old) {
1550 synchronize_net();
1551 bpf_prog_destroy(old);
1552 }
1553}
1554
1555static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1556 unsigned int len)
1557{
1558 struct bpf_prog *new;
1559 struct sock_fprog fprog;
1560 int ret;
1561
1562 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1563 return -EPERM;
1564 if (len != sizeof(fprog))
1565 return -EINVAL;
1566 if (copy_from_user(&fprog, data, len))
1567 return -EFAULT;
1568
1569 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1570 if (ret)
1571 return ret;
1572
1573 __fanout_set_data_bpf(po->fanout, new);
1574 return 0;
1575}
1576
1577static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1578 unsigned int len)
1579{
1580 struct bpf_prog *new;
1581 u32 fd;
1582
1583 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1584 return -EPERM;
1585 if (len != sizeof(fd))
1586 return -EINVAL;
1587 if (copy_from_user(&fd, data, len))
1588 return -EFAULT;
1589
1590 new = bpf_prog_get(fd);
1591 if (IS_ERR(new))
1592 return PTR_ERR(new);
1593 if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1594 bpf_prog_put(new);
1595 return -EINVAL;
1596 }
1597
1598 __fanout_set_data_bpf(po->fanout, new);
1599 return 0;
1600}
1601
1602static int fanout_set_data(struct packet_sock *po, char __user *data,
1603 unsigned int len)
1604{
1605 switch (po->fanout->type) {
1606 case PACKET_FANOUT_CBPF:
1607 return fanout_set_data_cbpf(po, data, len);
1608 case PACKET_FANOUT_EBPF:
1609 return fanout_set_data_ebpf(po, data, len);
1610 default:
1611 return -EINVAL;
1612 };
1613}
1614
1615static void fanout_release_data(struct packet_fanout *f)
1616{
1617 switch (f->type) {
1618 case PACKET_FANOUT_CBPF:
1619 case PACKET_FANOUT_EBPF:
1620 __fanout_set_data_bpf(f, NULL);
1621 };
1622}
1623
1624static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1625{
1626 struct packet_sock *po = pkt_sk(sk);
1627 struct packet_fanout *f, *match;
1628 u8 type = type_flags & 0xff;
1629 u8 flags = type_flags >> 8;
1630 int err;
1631
1632 switch (type) {
1633 case PACKET_FANOUT_ROLLOVER:
1634 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1635 return -EINVAL;
1636 case PACKET_FANOUT_HASH:
1637 case PACKET_FANOUT_LB:
1638 case PACKET_FANOUT_CPU:
1639 case PACKET_FANOUT_RND:
1640 case PACKET_FANOUT_QM:
1641 case PACKET_FANOUT_CBPF:
1642 case PACKET_FANOUT_EBPF:
1643 break;
1644 default:
1645 return -EINVAL;
1646 }
1647
1648 if (!po->running)
1649 return -EINVAL;
1650
1651 if (po->fanout)
1652 return -EALREADY;
1653
1654 if (type == PACKET_FANOUT_ROLLOVER ||
1655 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1656 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1657 if (!po->rollover)
1658 return -ENOMEM;
1659 atomic_long_set(&po->rollover->num, 0);
1660 atomic_long_set(&po->rollover->num_huge, 0);
1661 atomic_long_set(&po->rollover->num_failed, 0);
1662 }
1663
1664 mutex_lock(&fanout_mutex);
1665 match = NULL;
1666 list_for_each_entry(f, &fanout_list, list) {
1667 if (f->id == id &&
1668 read_pnet(&f->net) == sock_net(sk)) {
1669 match = f;
1670 break;
1671 }
1672 }
1673 err = -EINVAL;
1674 if (match && match->flags != flags)
1675 goto out;
1676 if (!match) {
1677 err = -ENOMEM;
1678 match = kzalloc(sizeof(*match), GFP_KERNEL);
1679 if (!match)
1680 goto out;
1681 write_pnet(&match->net, sock_net(sk));
1682 match->id = id;
1683 match->type = type;
1684 match->flags = flags;
1685 INIT_LIST_HEAD(&match->list);
1686 spin_lock_init(&match->lock);
1687 atomic_set(&match->sk_ref, 0);
1688 fanout_init_data(match);
1689 match->prot_hook.type = po->prot_hook.type;
1690 match->prot_hook.dev = po->prot_hook.dev;
1691 match->prot_hook.func = packet_rcv_fanout;
1692 match->prot_hook.af_packet_priv = match;
1693 match->prot_hook.id_match = match_fanout_group;
1694 dev_add_pack(&match->prot_hook);
1695 list_add(&match->list, &fanout_list);
1696 }
1697 err = -EINVAL;
1698 if (match->type == type &&
1699 match->prot_hook.type == po->prot_hook.type &&
1700 match->prot_hook.dev == po->prot_hook.dev) {
1701 err = -ENOSPC;
1702 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1703 __dev_remove_pack(&po->prot_hook);
1704 po->fanout = match;
1705 atomic_inc(&match->sk_ref);
1706 __fanout_link(sk, po);
1707 err = 0;
1708 }
1709 }
1710out:
1711 mutex_unlock(&fanout_mutex);
1712 if (err) {
1713 kfree(po->rollover);
1714 po->rollover = NULL;
1715 }
1716 return err;
1717}
1718
1719static void fanout_release(struct sock *sk)
1720{
1721 struct packet_sock *po = pkt_sk(sk);
1722 struct packet_fanout *f;
1723
1724 f = po->fanout;
1725 if (!f)
1726 return;
1727
1728 mutex_lock(&fanout_mutex);
1729 po->fanout = NULL;
1730
1731 if (atomic_dec_and_test(&f->sk_ref)) {
1732 list_del(&f->list);
1733 dev_remove_pack(&f->prot_hook);
1734 fanout_release_data(f);
1735 kfree(f);
1736 }
1737 mutex_unlock(&fanout_mutex);
1738
1739 if (po->rollover)
1740 kfree_rcu(po->rollover, rcu);
1741}
1742
1743static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1744 struct sk_buff *skb)
1745{
1746 /* Earlier code assumed this would be a VLAN pkt, double-check
1747 * this now that we have the actual packet in hand. We can only
1748 * do this check on Ethernet devices.
1749 */
1750 if (unlikely(dev->type != ARPHRD_ETHER))
1751 return false;
1752
1753 skb_reset_mac_header(skb);
1754 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1755}
1756
1757static const struct proto_ops packet_ops;
1758
1759static const struct proto_ops packet_ops_spkt;
1760
1761static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1762 struct packet_type *pt, struct net_device *orig_dev)
1763{
1764 struct sock *sk;
1765 struct sockaddr_pkt *spkt;
1766
1767 /*
1768 * When we registered the protocol we saved the socket in the data
1769 * field for just this event.
1770 */
1771
1772 sk = pt->af_packet_priv;
1773
1774 /*
1775 * Yank back the headers [hope the device set this
1776 * right or kerboom...]
1777 *
1778 * Incoming packets have ll header pulled,
1779 * push it back.
1780 *
1781 * For outgoing ones skb->data == skb_mac_header(skb)
1782 * so that this procedure is noop.
1783 */
1784
1785 if (skb->pkt_type == PACKET_LOOPBACK)
1786 goto out;
1787
1788 if (!net_eq(dev_net(dev), sock_net(sk)))
1789 goto out;
1790
1791 skb = skb_share_check(skb, GFP_ATOMIC);
1792 if (skb == NULL)
1793 goto oom;
1794
1795 /* drop any routing info */
1796 skb_dst_drop(skb);
1797
1798 /* drop conntrack reference */
1799 nf_reset(skb);
1800
1801 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1802
1803 skb_push(skb, skb->data - skb_mac_header(skb));
1804
1805 /*
1806 * The SOCK_PACKET socket receives _all_ frames.
1807 */
1808
1809 spkt->spkt_family = dev->type;
1810 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1811 spkt->spkt_protocol = skb->protocol;
1812
1813 /*
1814 * Charge the memory to the socket. This is done specifically
1815 * to prevent sockets using all the memory up.
1816 */
1817
1818 if (sock_queue_rcv_skb(sk, skb) == 0)
1819 return 0;
1820
1821out:
1822 kfree_skb(skb);
1823oom:
1824 return 0;
1825}
1826
1827
1828/*
1829 * Output a raw packet to a device layer. This bypasses all the other
1830 * protocol layers and you must therefore supply it with a complete frame
1831 */
1832
1833static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1834 size_t len)
1835{
1836 struct sock *sk = sock->sk;
1837 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1838 struct sk_buff *skb = NULL;
1839 struct net_device *dev;
1840 __be16 proto = 0;
1841 int err;
1842 int extra_len = 0;
1843
1844 /*
1845 * Get and verify the address.
1846 */
1847
1848 if (saddr) {
1849 if (msg->msg_namelen < sizeof(struct sockaddr))
1850 return -EINVAL;
1851 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1852 proto = saddr->spkt_protocol;
1853 } else
1854 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1855
1856 /*
1857 * Find the device first to size check it
1858 */
1859
1860 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1861retry:
1862 rcu_read_lock();
1863 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1864 err = -ENODEV;
1865 if (dev == NULL)
1866 goto out_unlock;
1867
1868 err = -ENETDOWN;
1869 if (!(dev->flags & IFF_UP))
1870 goto out_unlock;
1871
1872 /*
1873 * You may not queue a frame bigger than the mtu. This is the lowest level
1874 * raw protocol and you must do your own fragmentation at this level.
1875 */
1876
1877 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1878 if (!netif_supports_nofcs(dev)) {
1879 err = -EPROTONOSUPPORT;
1880 goto out_unlock;
1881 }
1882 extra_len = 4; /* We're doing our own CRC */
1883 }
1884
1885 err = -EMSGSIZE;
1886 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1887 goto out_unlock;
1888
1889 if (!skb) {
1890 size_t reserved = LL_RESERVED_SPACE(dev);
1891 int tlen = dev->needed_tailroom;
1892 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1893
1894 rcu_read_unlock();
1895 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1896 if (skb == NULL)
1897 return -ENOBUFS;
1898 /* FIXME: Save some space for broken drivers that write a hard
1899 * header at transmission time by themselves. PPP is the notable
1900 * one here. This should really be fixed at the driver level.
1901 */
1902 skb_reserve(skb, reserved);
1903 skb_reset_network_header(skb);
1904
1905 /* Try to align data part correctly */
1906 if (hhlen) {
1907 skb->data -= hhlen;
1908 skb->tail -= hhlen;
1909 if (len < hhlen)
1910 skb_reset_network_header(skb);
1911 }
1912 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1913 if (err)
1914 goto out_free;
1915 goto retry;
1916 }
1917
1918 if (!dev_validate_header(dev, skb->data, len)) {
1919 err = -EINVAL;
1920 goto out_unlock;
1921 }
1922 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1923 !packet_extra_vlan_len_allowed(dev, skb)) {
1924 err = -EMSGSIZE;
1925 goto out_unlock;
1926 }
1927
1928 skb->protocol = proto;
1929 skb->dev = dev;
1930 skb->priority = sk->sk_priority;
1931 skb->mark = sk->sk_mark;
1932
1933 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1934
1935 if (unlikely(extra_len == 4))
1936 skb->no_fcs = 1;
1937
1938 skb_probe_transport_header(skb, 0);
1939
1940 dev_queue_xmit(skb);
1941 rcu_read_unlock();
1942 return len;
1943
1944out_unlock:
1945 rcu_read_unlock();
1946out_free:
1947 kfree_skb(skb);
1948 return err;
1949}
1950
1951static unsigned int run_filter(struct sk_buff *skb,
1952 const struct sock *sk,
1953 unsigned int res)
1954{
1955 struct sk_filter *filter;
1956
1957 rcu_read_lock();
1958 filter = rcu_dereference(sk->sk_filter);
1959 if (filter != NULL)
1960 res = bpf_prog_run_clear_cb(filter->prog, skb);
1961 rcu_read_unlock();
1962
1963 return res;
1964}
1965
1966static int __packet_rcv_vnet(const struct sk_buff *skb,
1967 struct virtio_net_hdr *vnet_hdr)
1968{
1969 *vnet_hdr = (const struct virtio_net_hdr) { 0 };
1970
1971 if (skb_is_gso(skb)) {
1972 struct skb_shared_info *sinfo = skb_shinfo(skb);
1973
1974 /* This is a hint as to how much should be linear. */
1975 vnet_hdr->hdr_len =
1976 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
1977 vnet_hdr->gso_size =
1978 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
1979
1980 if (sinfo->gso_type & SKB_GSO_TCPV4)
1981 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1982 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1983 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1984 else if (sinfo->gso_type & SKB_GSO_UDP)
1985 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
1986 else if (sinfo->gso_type & SKB_GSO_FCOE)
1987 return -EINVAL;
1988 else
1989 BUG();
1990
1991 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1992 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1993 } else
1994 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
1995
1996 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1997 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1998 vnet_hdr->csum_start = __cpu_to_virtio16(vio_le(),
1999 skb_checksum_start_offset(skb));
2000 vnet_hdr->csum_offset = __cpu_to_virtio16(vio_le(),
2001 skb->csum_offset);
2002 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2003 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
2004 } /* else everything is zero */
2005
2006 return 0;
2007}
2008
2009static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2010 size_t *len)
2011{
2012 struct virtio_net_hdr vnet_hdr;
2013
2014 if (*len < sizeof(vnet_hdr))
2015 return -EINVAL;
2016 *len -= sizeof(vnet_hdr);
2017
2018 if (__packet_rcv_vnet(skb, &vnet_hdr))
2019 return -EINVAL;
2020
2021 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2022}
2023
2024/*
2025 * This function makes lazy skb cloning in hope that most of packets
2026 * are discarded by BPF.
2027 *
2028 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2029 * and skb->cb are mangled. It works because (and until) packets
2030 * falling here are owned by current CPU. Output packets are cloned
2031 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2032 * sequencially, so that if we return skb to original state on exit,
2033 * we will not harm anyone.
2034 */
2035
2036static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2037 struct packet_type *pt, struct net_device *orig_dev)
2038{
2039 struct sock *sk;
2040 struct sockaddr_ll *sll;
2041 struct packet_sock *po;
2042 u8 *skb_head = skb->data;
2043 int skb_len = skb->len;
2044 unsigned int snaplen, res;
2045
2046 if (skb->pkt_type == PACKET_LOOPBACK)
2047 goto drop;
2048
2049 sk = pt->af_packet_priv;
2050 po = pkt_sk(sk);
2051
2052 if (!net_eq(dev_net(dev), sock_net(sk)))
2053 goto drop;
2054
2055 skb->dev = dev;
2056
2057 if (dev->header_ops) {
2058 /* The device has an explicit notion of ll header,
2059 * exported to higher levels.
2060 *
2061 * Otherwise, the device hides details of its frame
2062 * structure, so that corresponding packet head is
2063 * never delivered to user.
2064 */
2065 if (sk->sk_type != SOCK_DGRAM)
2066 skb_push(skb, skb->data - skb_mac_header(skb));
2067 else if (skb->pkt_type == PACKET_OUTGOING) {
2068 /* Special case: outgoing packets have ll header at head */
2069 skb_pull(skb, skb_network_offset(skb));
2070 }
2071 }
2072
2073 snaplen = skb->len;
2074
2075 res = run_filter(skb, sk, snaplen);
2076 if (!res)
2077 goto drop_n_restore;
2078 if (snaplen > res)
2079 snaplen = res;
2080
2081 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2082 goto drop_n_acct;
2083
2084 if (skb_shared(skb)) {
2085 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2086 if (nskb == NULL)
2087 goto drop_n_acct;
2088
2089 if (skb_head != skb->data) {
2090 skb->data = skb_head;
2091 skb->len = skb_len;
2092 }
2093 consume_skb(skb);
2094 skb = nskb;
2095 }
2096
2097 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2098
2099 sll = &PACKET_SKB_CB(skb)->sa.ll;
2100 sll->sll_hatype = dev->type;
2101 sll->sll_pkttype = skb->pkt_type;
2102 if (unlikely(po->origdev))
2103 sll->sll_ifindex = orig_dev->ifindex;
2104 else
2105 sll->sll_ifindex = dev->ifindex;
2106
2107 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2108
2109 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2110 * Use their space for storing the original skb length.
2111 */
2112 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2113
2114 if (pskb_trim(skb, snaplen))
2115 goto drop_n_acct;
2116
2117 skb_set_owner_r(skb, sk);
2118 skb->dev = NULL;
2119 skb_dst_drop(skb);
2120
2121 /* drop conntrack reference */
2122 nf_reset(skb);
2123
2124 spin_lock(&sk->sk_receive_queue.lock);
2125 po->stats.stats1.tp_packets++;
2126 sock_skb_set_dropcount(sk, skb);
2127 __skb_queue_tail(&sk->sk_receive_queue, skb);
2128 spin_unlock(&sk->sk_receive_queue.lock);
2129 sk->sk_data_ready(sk);
2130 return 0;
2131
2132drop_n_acct:
2133 spin_lock(&sk->sk_receive_queue.lock);
2134 po->stats.stats1.tp_drops++;
2135 atomic_inc(&sk->sk_drops);
2136 spin_unlock(&sk->sk_receive_queue.lock);
2137
2138drop_n_restore:
2139 if (skb_head != skb->data && skb_shared(skb)) {
2140 skb->data = skb_head;
2141 skb->len = skb_len;
2142 }
2143drop:
2144 consume_skb(skb);
2145 return 0;
2146}
2147
2148static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2149 struct packet_type *pt, struct net_device *orig_dev)
2150{
2151 struct sock *sk;
2152 struct packet_sock *po;
2153 struct sockaddr_ll *sll;
2154 union tpacket_uhdr h;
2155 u8 *skb_head = skb->data;
2156 int skb_len = skb->len;
2157 unsigned int snaplen, res;
2158 unsigned long status = TP_STATUS_USER;
2159 unsigned short macoff, netoff, hdrlen;
2160 struct sk_buff *copy_skb = NULL;
2161 struct timespec ts;
2162 __u32 ts_status;
2163
2164 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2165 * We may add members to them until current aligned size without forcing
2166 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2167 */
2168 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2169 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2170
2171 if (skb->pkt_type == PACKET_LOOPBACK)
2172 goto drop;
2173
2174 sk = pt->af_packet_priv;
2175 po = pkt_sk(sk);
2176
2177 if (!net_eq(dev_net(dev), sock_net(sk)))
2178 goto drop;
2179
2180 if (dev->header_ops) {
2181 if (sk->sk_type != SOCK_DGRAM)
2182 skb_push(skb, skb->data - skb_mac_header(skb));
2183 else if (skb->pkt_type == PACKET_OUTGOING) {
2184 /* Special case: outgoing packets have ll header at head */
2185 skb_pull(skb, skb_network_offset(skb));
2186 }
2187 }
2188
2189 snaplen = skb->len;
2190
2191 res = run_filter(skb, sk, snaplen);
2192 if (!res)
2193 goto drop_n_restore;
2194
2195 if (skb->ip_summed == CHECKSUM_PARTIAL)
2196 status |= TP_STATUS_CSUMNOTREADY;
2197 else if (skb->pkt_type != PACKET_OUTGOING &&
2198 (skb->ip_summed == CHECKSUM_COMPLETE ||
2199 skb_csum_unnecessary(skb)))
2200 status |= TP_STATUS_CSUM_VALID;
2201
2202 if (snaplen > res)
2203 snaplen = res;
2204
2205 if (sk->sk_type == SOCK_DGRAM) {
2206 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2207 po->tp_reserve;
2208 } else {
2209 unsigned int maclen = skb_network_offset(skb);
2210 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2211 (maclen < 16 ? 16 : maclen)) +
2212 po->tp_reserve;
2213 if (po->has_vnet_hdr)
2214 netoff += sizeof(struct virtio_net_hdr);
2215 macoff = netoff - maclen;
2216 }
2217 if (po->tp_version <= TPACKET_V2) {
2218 if (macoff + snaplen > po->rx_ring.frame_size) {
2219 if (po->copy_thresh &&
2220 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2221 if (skb_shared(skb)) {
2222 copy_skb = skb_clone(skb, GFP_ATOMIC);
2223 } else {
2224 copy_skb = skb_get(skb);
2225 skb_head = skb->data;
2226 }
2227 if (copy_skb)
2228 skb_set_owner_r(copy_skb, sk);
2229 }
2230 snaplen = po->rx_ring.frame_size - macoff;
2231 if ((int)snaplen < 0)
2232 snaplen = 0;
2233 }
2234 } else if (unlikely(macoff + snaplen >
2235 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2236 u32 nval;
2237
2238 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2239 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2240 snaplen, nval, macoff);
2241 snaplen = nval;
2242 if (unlikely((int)snaplen < 0)) {
2243 snaplen = 0;
2244 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2245 }
2246 }
2247 spin_lock(&sk->sk_receive_queue.lock);
2248 h.raw = packet_current_rx_frame(po, skb,
2249 TP_STATUS_KERNEL, (macoff+snaplen));
2250 if (!h.raw)
2251 goto drop_n_account;
2252 if (po->tp_version <= TPACKET_V2) {
2253 packet_increment_rx_head(po, &po->rx_ring);
2254 /*
2255 * LOSING will be reported till you read the stats,
2256 * because it's COR - Clear On Read.
2257 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2258 * at packet level.
2259 */
2260 if (po->stats.stats1.tp_drops)
2261 status |= TP_STATUS_LOSING;
2262 }
2263 po->stats.stats1.tp_packets++;
2264 if (copy_skb) {
2265 status |= TP_STATUS_COPY;
2266 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2267 }
2268 spin_unlock(&sk->sk_receive_queue.lock);
2269
2270 if (po->has_vnet_hdr) {
2271 if (__packet_rcv_vnet(skb, h.raw + macoff -
2272 sizeof(struct virtio_net_hdr))) {
2273 spin_lock(&sk->sk_receive_queue.lock);
2274 goto drop_n_account;
2275 }
2276 }
2277
2278 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2279
2280 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2281 getnstimeofday(&ts);
2282
2283 status |= ts_status;
2284
2285 switch (po->tp_version) {
2286 case TPACKET_V1:
2287 h.h1->tp_len = skb->len;
2288 h.h1->tp_snaplen = snaplen;
2289 h.h1->tp_mac = macoff;
2290 h.h1->tp_net = netoff;
2291 h.h1->tp_sec = ts.tv_sec;
2292 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2293 hdrlen = sizeof(*h.h1);
2294 break;
2295 case TPACKET_V2:
2296 h.h2->tp_len = skb->len;
2297 h.h2->tp_snaplen = snaplen;
2298 h.h2->tp_mac = macoff;
2299 h.h2->tp_net = netoff;
2300 h.h2->tp_sec = ts.tv_sec;
2301 h.h2->tp_nsec = ts.tv_nsec;
2302 if (skb_vlan_tag_present(skb)) {
2303 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2304 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2305 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2306 } else {
2307 h.h2->tp_vlan_tci = 0;
2308 h.h2->tp_vlan_tpid = 0;
2309 }
2310 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2311 hdrlen = sizeof(*h.h2);
2312 break;
2313 case TPACKET_V3:
2314 /* tp_nxt_offset,vlan are already populated above.
2315 * So DONT clear those fields here
2316 */
2317 h.h3->tp_status |= status;
2318 h.h3->tp_len = skb->len;
2319 h.h3->tp_snaplen = snaplen;
2320 h.h3->tp_mac = macoff;
2321 h.h3->tp_net = netoff;
2322 h.h3->tp_sec = ts.tv_sec;
2323 h.h3->tp_nsec = ts.tv_nsec;
2324 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2325 hdrlen = sizeof(*h.h3);
2326 break;
2327 default:
2328 BUG();
2329 }
2330
2331 sll = h.raw + TPACKET_ALIGN(hdrlen);
2332 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2333 sll->sll_family = AF_PACKET;
2334 sll->sll_hatype = dev->type;
2335 sll->sll_protocol = skb->protocol;
2336 sll->sll_pkttype = skb->pkt_type;
2337 if (unlikely(po->origdev))
2338 sll->sll_ifindex = orig_dev->ifindex;
2339 else
2340 sll->sll_ifindex = dev->ifindex;
2341
2342 smp_mb();
2343
2344#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2345 if (po->tp_version <= TPACKET_V2) {
2346 u8 *start, *end;
2347
2348 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2349 macoff + snaplen);
2350
2351 for (start = h.raw; start < end; start += PAGE_SIZE)
2352 flush_dcache_page(pgv_to_page(start));
2353 }
2354 smp_wmb();
2355#endif
2356
2357 if (po->tp_version <= TPACKET_V2) {
2358 __packet_set_status(po, h.raw, status);
2359 sk->sk_data_ready(sk);
2360 } else {
2361 prb_clear_blk_fill_status(&po->rx_ring);
2362 }
2363
2364drop_n_restore:
2365 if (skb_head != skb->data && skb_shared(skb)) {
2366 skb->data = skb_head;
2367 skb->len = skb_len;
2368 }
2369drop:
2370 kfree_skb(skb);
2371 return 0;
2372
2373drop_n_account:
2374 po->stats.stats1.tp_drops++;
2375 spin_unlock(&sk->sk_receive_queue.lock);
2376
2377 sk->sk_data_ready(sk);
2378 kfree_skb(copy_skb);
2379 goto drop_n_restore;
2380}
2381
2382static void tpacket_destruct_skb(struct sk_buff *skb)
2383{
2384 struct packet_sock *po = pkt_sk(skb->sk);
2385
2386 if (likely(po->tx_ring.pg_vec)) {
2387 void *ph;
2388 __u32 ts;
2389
2390 ph = skb_shinfo(skb)->destructor_arg;
2391 packet_dec_pending(&po->tx_ring);
2392
2393 ts = __packet_set_timestamp(po, ph, skb);
2394 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2395 }
2396
2397 sock_wfree(skb);
2398}
2399
2400static void tpacket_set_protocol(const struct net_device *dev,
2401 struct sk_buff *skb)
2402{
2403 if (dev->type == ARPHRD_ETHER) {
2404 skb_reset_mac_header(skb);
2405 skb->protocol = eth_hdr(skb)->h_proto;
2406 }
2407}
2408
2409static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2410{
2411 unsigned short gso_type = 0;
2412
2413 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2414 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2415 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2416 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2417 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2418 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2419 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2420
2421 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2422 return -EINVAL;
2423
2424 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2425 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2426 case VIRTIO_NET_HDR_GSO_TCPV4:
2427 gso_type = SKB_GSO_TCPV4;
2428 break;
2429 case VIRTIO_NET_HDR_GSO_TCPV6:
2430 gso_type = SKB_GSO_TCPV6;
2431 break;
2432 case VIRTIO_NET_HDR_GSO_UDP:
2433 gso_type = SKB_GSO_UDP;
2434 break;
2435 default:
2436 return -EINVAL;
2437 }
2438
2439 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
2440 gso_type |= SKB_GSO_TCP_ECN;
2441
2442 if (vnet_hdr->gso_size == 0)
2443 return -EINVAL;
2444 }
2445
2446 vnet_hdr->gso_type = gso_type; /* changes type, temporary storage */
2447 return 0;
2448}
2449
2450static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2451 struct virtio_net_hdr *vnet_hdr)
2452{
2453 int n;
2454
2455 if (*len < sizeof(*vnet_hdr))
2456 return -EINVAL;
2457 *len -= sizeof(*vnet_hdr);
2458
2459 n = copy_from_iter(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter);
2460 if (n != sizeof(*vnet_hdr))
2461 return -EFAULT;
2462
2463 return __packet_snd_vnet_parse(vnet_hdr, *len);
2464}
2465
2466static int packet_snd_vnet_gso(struct sk_buff *skb,
2467 struct virtio_net_hdr *vnet_hdr)
2468{
2469 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2470 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start);
2471 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset);
2472
2473 if (!skb_partial_csum_set(skb, s, o))
2474 return -EINVAL;
2475 }
2476
2477 skb_shinfo(skb)->gso_size =
2478 __virtio16_to_cpu(vio_le(), vnet_hdr->gso_size);
2479 skb_shinfo(skb)->gso_type = vnet_hdr->gso_type;
2480
2481 /* Header must be checked, and gso_segs computed. */
2482 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2483 skb_shinfo(skb)->gso_segs = 0;
2484 return 0;
2485}
2486
2487static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2488 void *frame, struct net_device *dev, void *data, int tp_len,
2489 __be16 proto, unsigned char *addr, int hlen, int copylen)
2490{
2491 union tpacket_uhdr ph;
2492 int to_write, offset, len, nr_frags, len_max;
2493 struct socket *sock = po->sk.sk_socket;
2494 struct page *page;
2495 int err;
2496
2497 ph.raw = frame;
2498
2499 skb->protocol = proto;
2500 skb->dev = dev;
2501 skb->priority = po->sk.sk_priority;
2502 skb->mark = po->sk.sk_mark;
2503 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2504 skb_shinfo(skb)->destructor_arg = ph.raw;
2505
2506 skb_reserve(skb, hlen);
2507 skb_reset_network_header(skb);
2508
2509 to_write = tp_len;
2510
2511 if (sock->type == SOCK_DGRAM) {
2512 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2513 NULL, tp_len);
2514 if (unlikely(err < 0))
2515 return -EINVAL;
2516 } else if (copylen) {
2517 int hdrlen = min_t(int, copylen, tp_len);
2518
2519 skb_push(skb, dev->hard_header_len);
2520 skb_put(skb, copylen - dev->hard_header_len);
2521 err = skb_store_bits(skb, 0, data, hdrlen);
2522 if (unlikely(err))
2523 return err;
2524 if (!dev_validate_header(dev, skb->data, hdrlen))
2525 return -EINVAL;
2526 if (!skb->protocol)
2527 tpacket_set_protocol(dev, skb);
2528
2529 data += hdrlen;
2530 to_write -= hdrlen;
2531 }
2532
2533 offset = offset_in_page(data);
2534 len_max = PAGE_SIZE - offset;
2535 len = ((to_write > len_max) ? len_max : to_write);
2536
2537 skb->data_len = to_write;
2538 skb->len += to_write;
2539 skb->truesize += to_write;
2540 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2541
2542 while (likely(to_write)) {
2543 nr_frags = skb_shinfo(skb)->nr_frags;
2544
2545 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2546 pr_err("Packet exceed the number of skb frags(%lu)\n",
2547 MAX_SKB_FRAGS);
2548 return -EFAULT;
2549 }
2550
2551 page = pgv_to_page(data);
2552 data += len;
2553 flush_dcache_page(page);
2554 get_page(page);
2555 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2556 to_write -= len;
2557 offset = 0;
2558 len_max = PAGE_SIZE;
2559 len = ((to_write > len_max) ? len_max : to_write);
2560 }
2561
2562 skb_probe_transport_header(skb, 0);
2563
2564 return tp_len;
2565}
2566
2567static int tpacket_parse_header(struct packet_sock *po, void *frame,
2568 int size_max, void **data)
2569{
2570 union tpacket_uhdr ph;
2571 int tp_len, off;
2572
2573 ph.raw = frame;
2574
2575 switch (po->tp_version) {
2576 case TPACKET_V2:
2577 tp_len = ph.h2->tp_len;
2578 break;
2579 default:
2580 tp_len = ph.h1->tp_len;
2581 break;
2582 }
2583 if (unlikely(tp_len > size_max)) {
2584 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2585 return -EMSGSIZE;
2586 }
2587
2588 if (unlikely(po->tp_tx_has_off)) {
2589 int off_min, off_max;
2590
2591 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2592 off_max = po->tx_ring.frame_size - tp_len;
2593 if (po->sk.sk_type == SOCK_DGRAM) {
2594 switch (po->tp_version) {
2595 case TPACKET_V2:
2596 off = ph.h2->tp_net;
2597 break;
2598 default:
2599 off = ph.h1->tp_net;
2600 break;
2601 }
2602 } else {
2603 switch (po->tp_version) {
2604 case TPACKET_V2:
2605 off = ph.h2->tp_mac;
2606 break;
2607 default:
2608 off = ph.h1->tp_mac;
2609 break;
2610 }
2611 }
2612 if (unlikely((off < off_min) || (off_max < off)))
2613 return -EINVAL;
2614 } else {
2615 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2616 }
2617
2618 *data = frame + off;
2619 return tp_len;
2620}
2621
2622static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2623{
2624 struct sk_buff *skb;
2625 struct net_device *dev;
2626 struct virtio_net_hdr *vnet_hdr = NULL;
2627 __be16 proto;
2628 int err, reserve = 0;
2629 void *ph;
2630 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2631 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2632 int tp_len, size_max;
2633 unsigned char *addr;
2634 void *data;
2635 int len_sum = 0;
2636 int status = TP_STATUS_AVAILABLE;
2637 int hlen, tlen, copylen = 0;
2638
2639 mutex_lock(&po->pg_vec_lock);
2640
2641 if (likely(saddr == NULL)) {
2642 dev = packet_cached_dev_get(po);
2643 proto = po->num;
2644 addr = NULL;
2645 } else {
2646 err = -EINVAL;
2647 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2648 goto out;
2649 if (msg->msg_namelen < (saddr->sll_halen
2650 + offsetof(struct sockaddr_ll,
2651 sll_addr)))
2652 goto out;
2653 proto = saddr->sll_protocol;
2654 addr = saddr->sll_addr;
2655 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2656 }
2657
2658 err = -ENXIO;
2659 if (unlikely(dev == NULL))
2660 goto out;
2661 err = -ENETDOWN;
2662 if (unlikely(!(dev->flags & IFF_UP)))
2663 goto out_put;
2664
2665 if (po->sk.sk_socket->type == SOCK_RAW)
2666 reserve = dev->hard_header_len;
2667 size_max = po->tx_ring.frame_size
2668 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2669
2670 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2671 size_max = dev->mtu + reserve + VLAN_HLEN;
2672
2673 do {
2674 ph = packet_current_frame(po, &po->tx_ring,
2675 TP_STATUS_SEND_REQUEST);
2676 if (unlikely(ph == NULL)) {
2677 if (need_wait && need_resched())
2678 schedule();
2679 continue;
2680 }
2681
2682 skb = NULL;
2683 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2684 if (tp_len < 0)
2685 goto tpacket_error;
2686
2687 status = TP_STATUS_SEND_REQUEST;
2688 hlen = LL_RESERVED_SPACE(dev);
2689 tlen = dev->needed_tailroom;
2690 if (po->has_vnet_hdr) {
2691 vnet_hdr = data;
2692 data += sizeof(*vnet_hdr);
2693 tp_len -= sizeof(*vnet_hdr);
2694 if (tp_len < 0 ||
2695 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2696 tp_len = -EINVAL;
2697 goto tpacket_error;
2698 }
2699 copylen = __virtio16_to_cpu(vio_le(),
2700 vnet_hdr->hdr_len);
2701 }
2702 copylen = max_t(int, copylen, dev->hard_header_len);
2703 skb = sock_alloc_send_skb(&po->sk,
2704 hlen + tlen + sizeof(struct sockaddr_ll) +
2705 (copylen - dev->hard_header_len),
2706 !need_wait, &err);
2707
2708 if (unlikely(skb == NULL)) {
2709 /* we assume the socket was initially writeable ... */
2710 if (likely(len_sum > 0))
2711 err = len_sum;
2712 goto out_status;
2713 }
2714 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2715 addr, hlen, copylen);
2716 if (likely(tp_len >= 0) &&
2717 tp_len > dev->mtu + reserve &&
2718 !po->has_vnet_hdr &&
2719 !packet_extra_vlan_len_allowed(dev, skb))
2720 tp_len = -EMSGSIZE;
2721
2722 if (unlikely(tp_len < 0)) {
2723tpacket_error:
2724 if (po->tp_loss) {
2725 __packet_set_status(po, ph,
2726 TP_STATUS_AVAILABLE);
2727 packet_increment_head(&po->tx_ring);
2728 kfree_skb(skb);
2729 continue;
2730 } else {
2731 status = TP_STATUS_WRONG_FORMAT;
2732 err = tp_len;
2733 goto out_status;
2734 }
2735 }
2736
2737 if (po->has_vnet_hdr && packet_snd_vnet_gso(skb, vnet_hdr)) {
2738 tp_len = -EINVAL;
2739 goto tpacket_error;
2740 }
2741
2742 packet_pick_tx_queue(dev, skb);
2743
2744 skb->destructor = tpacket_destruct_skb;
2745 __packet_set_status(po, ph, TP_STATUS_SENDING);
2746 packet_inc_pending(&po->tx_ring);
2747
2748 status = TP_STATUS_SEND_REQUEST;
2749 err = po->xmit(skb);
2750 if (unlikely(err > 0)) {
2751 err = net_xmit_errno(err);
2752 if (err && __packet_get_status(po, ph) ==
2753 TP_STATUS_AVAILABLE) {
2754 /* skb was destructed already */
2755 skb = NULL;
2756 goto out_status;
2757 }
2758 /*
2759 * skb was dropped but not destructed yet;
2760 * let's treat it like congestion or err < 0
2761 */
2762 err = 0;
2763 }
2764 packet_increment_head(&po->tx_ring);
2765 len_sum += tp_len;
2766 } while (likely((ph != NULL) ||
2767 /* Note: packet_read_pending() might be slow if we have
2768 * to call it as it's per_cpu variable, but in fast-path
2769 * we already short-circuit the loop with the first
2770 * condition, and luckily don't have to go that path
2771 * anyway.
2772 */
2773 (need_wait && packet_read_pending(&po->tx_ring))));
2774
2775 err = len_sum;
2776 goto out_put;
2777
2778out_status:
2779 __packet_set_status(po, ph, status);
2780 kfree_skb(skb);
2781out_put:
2782 dev_put(dev);
2783out:
2784 mutex_unlock(&po->pg_vec_lock);
2785 return err;
2786}
2787
2788static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2789 size_t reserve, size_t len,
2790 size_t linear, int noblock,
2791 int *err)
2792{
2793 struct sk_buff *skb;
2794
2795 /* Under a page? Don't bother with paged skb. */
2796 if (prepad + len < PAGE_SIZE || !linear)
2797 linear = len;
2798
2799 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2800 err, 0);
2801 if (!skb)
2802 return NULL;
2803
2804 skb_reserve(skb, reserve);
2805 skb_put(skb, linear);
2806 skb->data_len = len - linear;
2807 skb->len += len - linear;
2808
2809 return skb;
2810}
2811
2812static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2813{
2814 struct sock *sk = sock->sk;
2815 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2816 struct sk_buff *skb;
2817 struct net_device *dev;
2818 __be16 proto;
2819 unsigned char *addr;
2820 int err, reserve = 0;
2821 struct sockcm_cookie sockc;
2822 struct virtio_net_hdr vnet_hdr = { 0 };
2823 int offset = 0;
2824 struct packet_sock *po = pkt_sk(sk);
2825 int hlen, tlen;
2826 int extra_len = 0;
2827
2828 /*
2829 * Get and verify the address.
2830 */
2831
2832 if (likely(saddr == NULL)) {
2833 dev = packet_cached_dev_get(po);
2834 proto = po->num;
2835 addr = NULL;
2836 } else {
2837 err = -EINVAL;
2838 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2839 goto out;
2840 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2841 goto out;
2842 proto = saddr->sll_protocol;
2843 addr = saddr->sll_addr;
2844 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2845 }
2846
2847 err = -ENXIO;
2848 if (unlikely(dev == NULL))
2849 goto out_unlock;
2850 err = -ENETDOWN;
2851 if (unlikely(!(dev->flags & IFF_UP)))
2852 goto out_unlock;
2853
2854 sockc.mark = sk->sk_mark;
2855 if (msg->msg_controllen) {
2856 err = sock_cmsg_send(sk, msg, &sockc);
2857 if (unlikely(err))
2858 goto out_unlock;
2859 }
2860
2861 if (sock->type == SOCK_RAW)
2862 reserve = dev->hard_header_len;
2863 if (po->has_vnet_hdr) {
2864 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2865 if (err)
2866 goto out_unlock;
2867 }
2868
2869 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2870 if (!netif_supports_nofcs(dev)) {
2871 err = -EPROTONOSUPPORT;
2872 goto out_unlock;
2873 }
2874 extra_len = 4; /* We're doing our own CRC */
2875 }
2876
2877 err = -EMSGSIZE;
2878 if (!vnet_hdr.gso_type &&
2879 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2880 goto out_unlock;
2881
2882 err = -ENOBUFS;
2883 hlen = LL_RESERVED_SPACE(dev);
2884 tlen = dev->needed_tailroom;
2885 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2886 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2887 msg->msg_flags & MSG_DONTWAIT, &err);
2888 if (skb == NULL)
2889 goto out_unlock;
2890
2891 skb_set_network_header(skb, reserve);
2892
2893 err = -EINVAL;
2894 if (sock->type == SOCK_DGRAM) {
2895 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2896 if (unlikely(offset < 0))
2897 goto out_free;
2898 }
2899
2900 /* Returns -EFAULT on error */
2901 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2902 if (err)
2903 goto out_free;
2904
2905 if (sock->type == SOCK_RAW &&
2906 !dev_validate_header(dev, skb->data, len)) {
2907 err = -EINVAL;
2908 goto out_free;
2909 }
2910
2911 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2912
2913 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2914 !packet_extra_vlan_len_allowed(dev, skb)) {
2915 err = -EMSGSIZE;
2916 goto out_free;
2917 }
2918
2919 skb->protocol = proto;
2920 skb->dev = dev;
2921 skb->priority = sk->sk_priority;
2922 skb->mark = sockc.mark;
2923
2924 packet_pick_tx_queue(dev, skb);
2925
2926 if (po->has_vnet_hdr) {
2927 err = packet_snd_vnet_gso(skb, &vnet_hdr);
2928 if (err)
2929 goto out_free;
2930 len += sizeof(vnet_hdr);
2931 }
2932
2933 skb_probe_transport_header(skb, reserve);
2934
2935 if (unlikely(extra_len == 4))
2936 skb->no_fcs = 1;
2937
2938 err = po->xmit(skb);
2939 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2940 goto out_unlock;
2941
2942 dev_put(dev);
2943
2944 return len;
2945
2946out_free:
2947 kfree_skb(skb);
2948out_unlock:
2949 if (dev)
2950 dev_put(dev);
2951out:
2952 return err;
2953}
2954
2955static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2956{
2957 struct sock *sk = sock->sk;
2958 struct packet_sock *po = pkt_sk(sk);
2959
2960 if (po->tx_ring.pg_vec)
2961 return tpacket_snd(po, msg);
2962 else
2963 return packet_snd(sock, msg, len);
2964}
2965
2966/*
2967 * Close a PACKET socket. This is fairly simple. We immediately go
2968 * to 'closed' state and remove our protocol entry in the device list.
2969 */
2970
2971static int packet_release(struct socket *sock)
2972{
2973 struct sock *sk = sock->sk;
2974 struct packet_sock *po;
2975 struct net *net;
2976 union tpacket_req_u req_u;
2977
2978 if (!sk)
2979 return 0;
2980
2981 net = sock_net(sk);
2982 po = pkt_sk(sk);
2983
2984 mutex_lock(&net->packet.sklist_lock);
2985 sk_del_node_init_rcu(sk);
2986 mutex_unlock(&net->packet.sklist_lock);
2987
2988 preempt_disable();
2989 sock_prot_inuse_add(net, sk->sk_prot, -1);
2990 preempt_enable();
2991
2992 spin_lock(&po->bind_lock);
2993 unregister_prot_hook(sk, false);
2994 packet_cached_dev_reset(po);
2995
2996 if (po->prot_hook.dev) {
2997 dev_put(po->prot_hook.dev);
2998 po->prot_hook.dev = NULL;
2999 }
3000 spin_unlock(&po->bind_lock);
3001
3002 packet_flush_mclist(sk);
3003
3004 if (po->rx_ring.pg_vec) {
3005 memset(&req_u, 0, sizeof(req_u));
3006 packet_set_ring(sk, &req_u, 1, 0);
3007 }
3008
3009 if (po->tx_ring.pg_vec) {
3010 memset(&req_u, 0, sizeof(req_u));
3011 packet_set_ring(sk, &req_u, 1, 1);
3012 }
3013
3014 fanout_release(sk);
3015
3016 synchronize_net();
3017 /*
3018 * Now the socket is dead. No more input will appear.
3019 */
3020 sock_orphan(sk);
3021 sock->sk = NULL;
3022
3023 /* Purge queues */
3024
3025 skb_queue_purge(&sk->sk_receive_queue);
3026 packet_free_pending(po);
3027 sk_refcnt_debug_release(sk);
3028
3029 sock_put(sk);
3030 return 0;
3031}
3032
3033/*
3034 * Attach a packet hook.
3035 */
3036
3037static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3038 __be16 proto)
3039{
3040 struct packet_sock *po = pkt_sk(sk);
3041 struct net_device *dev_curr;
3042 __be16 proto_curr;
3043 bool need_rehook;
3044 struct net_device *dev = NULL;
3045 int ret = 0;
3046 bool unlisted = false;
3047
3048 if (po->fanout)
3049 return -EINVAL;
3050
3051 lock_sock(sk);
3052 spin_lock(&po->bind_lock);
3053 rcu_read_lock();
3054
3055 if (name) {
3056 dev = dev_get_by_name_rcu(sock_net(sk), name);
3057 if (!dev) {
3058 ret = -ENODEV;
3059 goto out_unlock;
3060 }
3061 } else if (ifindex) {
3062 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3063 if (!dev) {
3064 ret = -ENODEV;
3065 goto out_unlock;
3066 }
3067 }
3068
3069 if (dev)
3070 dev_hold(dev);
3071
3072 proto_curr = po->prot_hook.type;
3073 dev_curr = po->prot_hook.dev;
3074
3075 need_rehook = proto_curr != proto || dev_curr != dev;
3076
3077 if (need_rehook) {
3078 if (po->running) {
3079 rcu_read_unlock();
3080 __unregister_prot_hook(sk, true);
3081 rcu_read_lock();
3082 dev_curr = po->prot_hook.dev;
3083 if (dev)
3084 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3085 dev->ifindex);
3086 }
3087
3088 po->num = proto;
3089 po->prot_hook.type = proto;
3090
3091 if (unlikely(unlisted)) {
3092 dev_put(dev);
3093 po->prot_hook.dev = NULL;
3094 po->ifindex = -1;
3095 packet_cached_dev_reset(po);
3096 } else {
3097 po->prot_hook.dev = dev;
3098 po->ifindex = dev ? dev->ifindex : 0;
3099 packet_cached_dev_assign(po, dev);
3100 }
3101 }
3102 if (dev_curr)
3103 dev_put(dev_curr);
3104
3105 if (proto == 0 || !need_rehook)
3106 goto out_unlock;
3107
3108 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3109 register_prot_hook(sk);
3110 } else {
3111 sk->sk_err = ENETDOWN;
3112 if (!sock_flag(sk, SOCK_DEAD))
3113 sk->sk_error_report(sk);
3114 }
3115
3116out_unlock:
3117 rcu_read_unlock();
3118 spin_unlock(&po->bind_lock);
3119 release_sock(sk);
3120 return ret;
3121}
3122
3123/*
3124 * Bind a packet socket to a device
3125 */
3126
3127static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3128 int addr_len)
3129{
3130 struct sock *sk = sock->sk;
3131 char name[15];
3132
3133 /*
3134 * Check legality
3135 */
3136
3137 if (addr_len != sizeof(struct sockaddr))
3138 return -EINVAL;
3139 strlcpy(name, uaddr->sa_data, sizeof(name));
3140
3141 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3142}
3143
3144static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3145{
3146 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3147 struct sock *sk = sock->sk;
3148
3149 /*
3150 * Check legality
3151 */
3152
3153 if (addr_len < sizeof(struct sockaddr_ll))
3154 return -EINVAL;
3155 if (sll->sll_family != AF_PACKET)
3156 return -EINVAL;
3157
3158 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3159 sll->sll_protocol ? : pkt_sk(sk)->num);
3160}
3161
3162static struct proto packet_proto = {
3163 .name = "PACKET",
3164 .owner = THIS_MODULE,
3165 .obj_size = sizeof(struct packet_sock),
3166};
3167
3168/*
3169 * Create a packet of type SOCK_PACKET.
3170 */
3171
3172static int packet_create(struct net *net, struct socket *sock, int protocol,
3173 int kern)
3174{
3175 struct sock *sk;
3176 struct packet_sock *po;
3177 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3178 int err;
3179
3180 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3181 return -EPERM;
3182 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3183 sock->type != SOCK_PACKET)
3184 return -ESOCKTNOSUPPORT;
3185
3186 sock->state = SS_UNCONNECTED;
3187
3188 err = -ENOBUFS;
3189 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3190 if (sk == NULL)
3191 goto out;
3192
3193 sock->ops = &packet_ops;
3194 if (sock->type == SOCK_PACKET)
3195 sock->ops = &packet_ops_spkt;
3196
3197 sock_init_data(sock, sk);
3198
3199 po = pkt_sk(sk);
3200 sk->sk_family = PF_PACKET;
3201 po->num = proto;
3202 po->xmit = dev_queue_xmit;
3203
3204 err = packet_alloc_pending(po);
3205 if (err)
3206 goto out2;
3207
3208 packet_cached_dev_reset(po);
3209
3210 sk->sk_destruct = packet_sock_destruct;
3211 sk_refcnt_debug_inc(sk);
3212
3213 /*
3214 * Attach a protocol block
3215 */
3216
3217 spin_lock_init(&po->bind_lock);
3218 mutex_init(&po->pg_vec_lock);
3219 po->rollover = NULL;
3220 po->prot_hook.func = packet_rcv;
3221
3222 if (sock->type == SOCK_PACKET)
3223 po->prot_hook.func = packet_rcv_spkt;
3224
3225 po->prot_hook.af_packet_priv = sk;
3226
3227 if (proto) {
3228 po->prot_hook.type = proto;
3229 register_prot_hook(sk);
3230 }
3231
3232 mutex_lock(&net->packet.sklist_lock);
3233 sk_add_node_rcu(sk, &net->packet.sklist);
3234 mutex_unlock(&net->packet.sklist_lock);
3235
3236 preempt_disable();
3237 sock_prot_inuse_add(net, &packet_proto, 1);
3238 preempt_enable();
3239
3240 return 0;
3241out2:
3242 sk_free(sk);
3243out:
3244 return err;
3245}
3246
3247/*
3248 * Pull a packet from our receive queue and hand it to the user.
3249 * If necessary we block.
3250 */
3251
3252static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3253 int flags)
3254{
3255 struct sock *sk = sock->sk;
3256 struct sk_buff *skb;
3257 int copied, err;
3258 int vnet_hdr_len = 0;
3259 unsigned int origlen = 0;
3260
3261 err = -EINVAL;
3262 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3263 goto out;
3264
3265#if 0
3266 /* What error should we return now? EUNATTACH? */
3267 if (pkt_sk(sk)->ifindex < 0)
3268 return -ENODEV;
3269#endif
3270
3271 if (flags & MSG_ERRQUEUE) {
3272 err = sock_recv_errqueue(sk, msg, len,
3273 SOL_PACKET, PACKET_TX_TIMESTAMP);
3274 goto out;
3275 }
3276
3277 /*
3278 * Call the generic datagram receiver. This handles all sorts
3279 * of horrible races and re-entrancy so we can forget about it
3280 * in the protocol layers.
3281 *
3282 * Now it will return ENETDOWN, if device have just gone down,
3283 * but then it will block.
3284 */
3285
3286 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3287
3288 /*
3289 * An error occurred so return it. Because skb_recv_datagram()
3290 * handles the blocking we don't see and worry about blocking
3291 * retries.
3292 */
3293
3294 if (skb == NULL)
3295 goto out;
3296
3297 if (pkt_sk(sk)->pressure)
3298 packet_rcv_has_room(pkt_sk(sk), NULL);
3299
3300 if (pkt_sk(sk)->has_vnet_hdr) {
3301 err = packet_rcv_vnet(msg, skb, &len);
3302 if (err)
3303 goto out_free;
3304 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3305 }
3306
3307 /* You lose any data beyond the buffer you gave. If it worries
3308 * a user program they can ask the device for its MTU
3309 * anyway.
3310 */
3311 copied = skb->len;
3312 if (copied > len) {
3313 copied = len;
3314 msg->msg_flags |= MSG_TRUNC;
3315 }
3316
3317 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3318 if (err)
3319 goto out_free;
3320
3321 if (sock->type != SOCK_PACKET) {
3322 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3323
3324 /* Original length was stored in sockaddr_ll fields */
3325 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3326 sll->sll_family = AF_PACKET;
3327 sll->sll_protocol = skb->protocol;
3328 }
3329
3330 sock_recv_ts_and_drops(msg, sk, skb);
3331
3332 if (msg->msg_name) {
3333 /* If the address length field is there to be filled
3334 * in, we fill it in now.
3335 */
3336 if (sock->type == SOCK_PACKET) {
3337 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3338 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3339 } else {
3340 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3341
3342 msg->msg_namelen = sll->sll_halen +
3343 offsetof(struct sockaddr_ll, sll_addr);
3344 }
3345 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3346 msg->msg_namelen);
3347 }
3348
3349 if (pkt_sk(sk)->auxdata) {
3350 struct tpacket_auxdata aux;
3351
3352 aux.tp_status = TP_STATUS_USER;
3353 if (skb->ip_summed == CHECKSUM_PARTIAL)
3354 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3355 else if (skb->pkt_type != PACKET_OUTGOING &&
3356 (skb->ip_summed == CHECKSUM_COMPLETE ||
3357 skb_csum_unnecessary(skb)))
3358 aux.tp_status |= TP_STATUS_CSUM_VALID;
3359
3360 aux.tp_len = origlen;
3361 aux.tp_snaplen = skb->len;
3362 aux.tp_mac = 0;
3363 aux.tp_net = skb_network_offset(skb);
3364 if (skb_vlan_tag_present(skb)) {
3365 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3366 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3367 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3368 } else {
3369 aux.tp_vlan_tci = 0;
3370 aux.tp_vlan_tpid = 0;
3371 }
3372 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3373 }
3374
3375 /*
3376 * Free or return the buffer as appropriate. Again this
3377 * hides all the races and re-entrancy issues from us.
3378 */
3379 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3380
3381out_free:
3382 skb_free_datagram(sk, skb);
3383out:
3384 return err;
3385}
3386
3387static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3388 int *uaddr_len, int peer)
3389{
3390 struct net_device *dev;
3391 struct sock *sk = sock->sk;
3392
3393 if (peer)
3394 return -EOPNOTSUPP;
3395
3396 uaddr->sa_family = AF_PACKET;
3397 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3398 rcu_read_lock();
3399 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3400 if (dev)
3401 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3402 rcu_read_unlock();
3403 *uaddr_len = sizeof(*uaddr);
3404
3405 return 0;
3406}
3407
3408static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3409 int *uaddr_len, int peer)
3410{
3411 struct net_device *dev;
3412 struct sock *sk = sock->sk;
3413 struct packet_sock *po = pkt_sk(sk);
3414 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3415
3416 if (peer)
3417 return -EOPNOTSUPP;
3418
3419 sll->sll_family = AF_PACKET;
3420 sll->sll_ifindex = po->ifindex;
3421 sll->sll_protocol = po->num;
3422 sll->sll_pkttype = 0;
3423 rcu_read_lock();
3424 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3425 if (dev) {
3426 sll->sll_hatype = dev->type;
3427 sll->sll_halen = dev->addr_len;
3428 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3429 } else {
3430 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3431 sll->sll_halen = 0;
3432 }
3433 rcu_read_unlock();
3434 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3435
3436 return 0;
3437}
3438
3439static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3440 int what)
3441{
3442 switch (i->type) {
3443 case PACKET_MR_MULTICAST:
3444 if (i->alen != dev->addr_len)
3445 return -EINVAL;
3446 if (what > 0)
3447 return dev_mc_add(dev, i->addr);
3448 else
3449 return dev_mc_del(dev, i->addr);
3450 break;
3451 case PACKET_MR_PROMISC:
3452 return dev_set_promiscuity(dev, what);
3453 case PACKET_MR_ALLMULTI:
3454 return dev_set_allmulti(dev, what);
3455 case PACKET_MR_UNICAST:
3456 if (i->alen != dev->addr_len)
3457 return -EINVAL;
3458 if (what > 0)
3459 return dev_uc_add(dev, i->addr);
3460 else
3461 return dev_uc_del(dev, i->addr);
3462 break;
3463 default:
3464 break;
3465 }
3466 return 0;
3467}
3468
3469static void packet_dev_mclist_delete(struct net_device *dev,
3470 struct packet_mclist **mlp)
3471{
3472 struct packet_mclist *ml;
3473
3474 while ((ml = *mlp) != NULL) {
3475 if (ml->ifindex == dev->ifindex) {
3476 packet_dev_mc(dev, ml, -1);
3477 *mlp = ml->next;
3478 kfree(ml);
3479 } else
3480 mlp = &ml->next;
3481 }
3482}
3483
3484static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3485{
3486 struct packet_sock *po = pkt_sk(sk);
3487 struct packet_mclist *ml, *i;
3488 struct net_device *dev;
3489 int err;
3490
3491 rtnl_lock();
3492
3493 err = -ENODEV;
3494 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3495 if (!dev)
3496 goto done;
3497
3498 err = -EINVAL;
3499 if (mreq->mr_alen > dev->addr_len)
3500 goto done;
3501
3502 err = -ENOBUFS;
3503 i = kmalloc(sizeof(*i), GFP_KERNEL);
3504 if (i == NULL)
3505 goto done;
3506
3507 err = 0;
3508 for (ml = po->mclist; ml; ml = ml->next) {
3509 if (ml->ifindex == mreq->mr_ifindex &&
3510 ml->type == mreq->mr_type &&
3511 ml->alen == mreq->mr_alen &&
3512 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3513 ml->count++;
3514 /* Free the new element ... */
3515 kfree(i);
3516 goto done;
3517 }
3518 }
3519
3520 i->type = mreq->mr_type;
3521 i->ifindex = mreq->mr_ifindex;
3522 i->alen = mreq->mr_alen;
3523 memcpy(i->addr, mreq->mr_address, i->alen);
3524 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3525 i->count = 1;
3526 i->next = po->mclist;
3527 po->mclist = i;
3528 err = packet_dev_mc(dev, i, 1);
3529 if (err) {
3530 po->mclist = i->next;
3531 kfree(i);
3532 }
3533
3534done:
3535 rtnl_unlock();
3536 return err;
3537}
3538
3539static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3540{
3541 struct packet_mclist *ml, **mlp;
3542
3543 rtnl_lock();
3544
3545 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3546 if (ml->ifindex == mreq->mr_ifindex &&
3547 ml->type == mreq->mr_type &&
3548 ml->alen == mreq->mr_alen &&
3549 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3550 if (--ml->count == 0) {
3551 struct net_device *dev;
3552 *mlp = ml->next;
3553 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3554 if (dev)
3555 packet_dev_mc(dev, ml, -1);
3556 kfree(ml);
3557 }
3558 break;
3559 }
3560 }
3561 rtnl_unlock();
3562 return 0;
3563}
3564
3565static void packet_flush_mclist(struct sock *sk)
3566{
3567 struct packet_sock *po = pkt_sk(sk);
3568 struct packet_mclist *ml;
3569
3570 if (!po->mclist)
3571 return;
3572
3573 rtnl_lock();
3574 while ((ml = po->mclist) != NULL) {
3575 struct net_device *dev;
3576
3577 po->mclist = ml->next;
3578 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3579 if (dev != NULL)
3580 packet_dev_mc(dev, ml, -1);
3581 kfree(ml);
3582 }
3583 rtnl_unlock();
3584}
3585
3586static int
3587packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3588{
3589 struct sock *sk = sock->sk;
3590 struct packet_sock *po = pkt_sk(sk);
3591 int ret;
3592
3593 if (level != SOL_PACKET)
3594 return -ENOPROTOOPT;
3595
3596 switch (optname) {
3597 case PACKET_ADD_MEMBERSHIP:
3598 case PACKET_DROP_MEMBERSHIP:
3599 {
3600 struct packet_mreq_max mreq;
3601 int len = optlen;
3602 memset(&mreq, 0, sizeof(mreq));
3603 if (len < sizeof(struct packet_mreq))
3604 return -EINVAL;
3605 if (len > sizeof(mreq))
3606 len = sizeof(mreq);
3607 if (copy_from_user(&mreq, optval, len))
3608 return -EFAULT;
3609 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3610 return -EINVAL;
3611 if (optname == PACKET_ADD_MEMBERSHIP)
3612 ret = packet_mc_add(sk, &mreq);
3613 else
3614 ret = packet_mc_drop(sk, &mreq);
3615 return ret;
3616 }
3617
3618 case PACKET_RX_RING:
3619 case PACKET_TX_RING:
3620 {
3621 union tpacket_req_u req_u;
3622 int len;
3623
3624 switch (po->tp_version) {
3625 case TPACKET_V1:
3626 case TPACKET_V2:
3627 len = sizeof(req_u.req);
3628 break;
3629 case TPACKET_V3:
3630 default:
3631 len = sizeof(req_u.req3);
3632 break;
3633 }
3634 if (optlen < len)
3635 return -EINVAL;
3636 if (copy_from_user(&req_u.req, optval, len))
3637 return -EFAULT;
3638 return packet_set_ring(sk, &req_u, 0,
3639 optname == PACKET_TX_RING);
3640 }
3641 case PACKET_COPY_THRESH:
3642 {
3643 int val;
3644
3645 if (optlen != sizeof(val))
3646 return -EINVAL;
3647 if (copy_from_user(&val, optval, sizeof(val)))
3648 return -EFAULT;
3649
3650 pkt_sk(sk)->copy_thresh = val;
3651 return 0;
3652 }
3653 case PACKET_VERSION:
3654 {
3655 int val;
3656
3657 if (optlen != sizeof(val))
3658 return -EINVAL;
3659 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3660 return -EBUSY;
3661 if (copy_from_user(&val, optval, sizeof(val)))
3662 return -EFAULT;
3663 switch (val) {
3664 case TPACKET_V1:
3665 case TPACKET_V2:
3666 case TPACKET_V3:
3667 po->tp_version = val;
3668 return 0;
3669 default:
3670 return -EINVAL;
3671 }
3672 }
3673 case PACKET_RESERVE:
3674 {
3675 unsigned int val;
3676
3677 if (optlen != sizeof(val))
3678 return -EINVAL;
3679 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3680 return -EBUSY;
3681 if (copy_from_user(&val, optval, sizeof(val)))
3682 return -EFAULT;
3683 po->tp_reserve = val;
3684 return 0;
3685 }
3686 case PACKET_LOSS:
3687 {
3688 unsigned int val;
3689
3690 if (optlen != sizeof(val))
3691 return -EINVAL;
3692 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3693 return -EBUSY;
3694 if (copy_from_user(&val, optval, sizeof(val)))
3695 return -EFAULT;
3696 po->tp_loss = !!val;
3697 return 0;
3698 }
3699 case PACKET_AUXDATA:
3700 {
3701 int val;
3702
3703 if (optlen < sizeof(val))
3704 return -EINVAL;
3705 if (copy_from_user(&val, optval, sizeof(val)))
3706 return -EFAULT;
3707
3708 po->auxdata = !!val;
3709 return 0;
3710 }
3711 case PACKET_ORIGDEV:
3712 {
3713 int val;
3714
3715 if (optlen < sizeof(val))
3716 return -EINVAL;
3717 if (copy_from_user(&val, optval, sizeof(val)))
3718 return -EFAULT;
3719
3720 po->origdev = !!val;
3721 return 0;
3722 }
3723 case PACKET_VNET_HDR:
3724 {
3725 int val;
3726
3727 if (sock->type != SOCK_RAW)
3728 return -EINVAL;
3729 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3730 return -EBUSY;
3731 if (optlen < sizeof(val))
3732 return -EINVAL;
3733 if (copy_from_user(&val, optval, sizeof(val)))
3734 return -EFAULT;
3735
3736 po->has_vnet_hdr = !!val;
3737 return 0;
3738 }
3739 case PACKET_TIMESTAMP:
3740 {
3741 int val;
3742
3743 if (optlen != sizeof(val))
3744 return -EINVAL;
3745 if (copy_from_user(&val, optval, sizeof(val)))
3746 return -EFAULT;
3747
3748 po->tp_tstamp = val;
3749 return 0;
3750 }
3751 case PACKET_FANOUT:
3752 {
3753 int val;
3754
3755 if (optlen != sizeof(val))
3756 return -EINVAL;
3757 if (copy_from_user(&val, optval, sizeof(val)))
3758 return -EFAULT;
3759
3760 return fanout_add(sk, val & 0xffff, val >> 16);
3761 }
3762 case PACKET_FANOUT_DATA:
3763 {
3764 if (!po->fanout)
3765 return -EINVAL;
3766
3767 return fanout_set_data(po, optval, optlen);
3768 }
3769 case PACKET_TX_HAS_OFF:
3770 {
3771 unsigned int val;
3772
3773 if (optlen != sizeof(val))
3774 return -EINVAL;
3775 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3776 return -EBUSY;
3777 if (copy_from_user(&val, optval, sizeof(val)))
3778 return -EFAULT;
3779 po->tp_tx_has_off = !!val;
3780 return 0;
3781 }
3782 case PACKET_QDISC_BYPASS:
3783 {
3784 int val;
3785
3786 if (optlen != sizeof(val))
3787 return -EINVAL;
3788 if (copy_from_user(&val, optval, sizeof(val)))
3789 return -EFAULT;
3790
3791 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3792 return 0;
3793 }
3794 default:
3795 return -ENOPROTOOPT;
3796 }
3797}
3798
3799static int packet_getsockopt(struct socket *sock, int level, int optname,
3800 char __user *optval, int __user *optlen)
3801{
3802 int len;
3803 int val, lv = sizeof(val);
3804 struct sock *sk = sock->sk;
3805 struct packet_sock *po = pkt_sk(sk);
3806 void *data = &val;
3807 union tpacket_stats_u st;
3808 struct tpacket_rollover_stats rstats;
3809
3810 if (level != SOL_PACKET)
3811 return -ENOPROTOOPT;
3812
3813 if (get_user(len, optlen))
3814 return -EFAULT;
3815
3816 if (len < 0)
3817 return -EINVAL;
3818
3819 switch (optname) {
3820 case PACKET_STATISTICS:
3821 spin_lock_bh(&sk->sk_receive_queue.lock);
3822 memcpy(&st, &po->stats, sizeof(st));
3823 memset(&po->stats, 0, sizeof(po->stats));
3824 spin_unlock_bh(&sk->sk_receive_queue.lock);
3825
3826 if (po->tp_version == TPACKET_V3) {
3827 lv = sizeof(struct tpacket_stats_v3);
3828 st.stats3.tp_packets += st.stats3.tp_drops;
3829 data = &st.stats3;
3830 } else {
3831 lv = sizeof(struct tpacket_stats);
3832 st.stats1.tp_packets += st.stats1.tp_drops;
3833 data = &st.stats1;
3834 }
3835
3836 break;
3837 case PACKET_AUXDATA:
3838 val = po->auxdata;
3839 break;
3840 case PACKET_ORIGDEV:
3841 val = po->origdev;
3842 break;
3843 case PACKET_VNET_HDR:
3844 val = po->has_vnet_hdr;
3845 break;
3846 case PACKET_VERSION:
3847 val = po->tp_version;
3848 break;
3849 case PACKET_HDRLEN:
3850 if (len > sizeof(int))
3851 len = sizeof(int);
3852 if (copy_from_user(&val, optval, len))
3853 return -EFAULT;
3854 switch (val) {
3855 case TPACKET_V1:
3856 val = sizeof(struct tpacket_hdr);
3857 break;
3858 case TPACKET_V2:
3859 val = sizeof(struct tpacket2_hdr);
3860 break;
3861 case TPACKET_V3:
3862 val = sizeof(struct tpacket3_hdr);
3863 break;
3864 default:
3865 return -EINVAL;
3866 }
3867 break;
3868 case PACKET_RESERVE:
3869 val = po->tp_reserve;
3870 break;
3871 case PACKET_LOSS:
3872 val = po->tp_loss;
3873 break;
3874 case PACKET_TIMESTAMP:
3875 val = po->tp_tstamp;
3876 break;
3877 case PACKET_FANOUT:
3878 val = (po->fanout ?
3879 ((u32)po->fanout->id |
3880 ((u32)po->fanout->type << 16) |
3881 ((u32)po->fanout->flags << 24)) :
3882 0);
3883 break;
3884 case PACKET_ROLLOVER_STATS:
3885 if (!po->rollover)
3886 return -EINVAL;
3887 rstats.tp_all = atomic_long_read(&po->rollover->num);
3888 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3889 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3890 data = &rstats;
3891 lv = sizeof(rstats);
3892 break;
3893 case PACKET_TX_HAS_OFF:
3894 val = po->tp_tx_has_off;
3895 break;
3896 case PACKET_QDISC_BYPASS:
3897 val = packet_use_direct_xmit(po);
3898 break;
3899 default:
3900 return -ENOPROTOOPT;
3901 }
3902
3903 if (len > lv)
3904 len = lv;
3905 if (put_user(len, optlen))
3906 return -EFAULT;
3907 if (copy_to_user(optval, data, len))
3908 return -EFAULT;
3909 return 0;
3910}
3911
3912
3913static int packet_notifier(struct notifier_block *this,
3914 unsigned long msg, void *ptr)
3915{
3916 struct sock *sk;
3917 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3918 struct net *net = dev_net(dev);
3919
3920 rcu_read_lock();
3921 sk_for_each_rcu(sk, &net->packet.sklist) {
3922 struct packet_sock *po = pkt_sk(sk);
3923
3924 switch (msg) {
3925 case NETDEV_UNREGISTER:
3926 if (po->mclist)
3927 packet_dev_mclist_delete(dev, &po->mclist);
3928 /* fallthrough */
3929
3930 case NETDEV_DOWN:
3931 if (dev->ifindex == po->ifindex) {
3932 spin_lock(&po->bind_lock);
3933 if (po->running) {
3934 __unregister_prot_hook(sk, false);
3935 sk->sk_err = ENETDOWN;
3936 if (!sock_flag(sk, SOCK_DEAD))
3937 sk->sk_error_report(sk);
3938 }
3939 if (msg == NETDEV_UNREGISTER) {
3940 packet_cached_dev_reset(po);
3941 po->ifindex = -1;
3942 if (po->prot_hook.dev)
3943 dev_put(po->prot_hook.dev);
3944 po->prot_hook.dev = NULL;
3945 }
3946 spin_unlock(&po->bind_lock);
3947 }
3948 break;
3949 case NETDEV_UP:
3950 if (dev->ifindex == po->ifindex) {
3951 spin_lock(&po->bind_lock);
3952 if (po->num)
3953 register_prot_hook(sk);
3954 spin_unlock(&po->bind_lock);
3955 }
3956 break;
3957 }
3958 }
3959 rcu_read_unlock();
3960 return NOTIFY_DONE;
3961}
3962
3963
3964static int packet_ioctl(struct socket *sock, unsigned int cmd,
3965 unsigned long arg)
3966{
3967 struct sock *sk = sock->sk;
3968
3969 switch (cmd) {
3970 case SIOCOUTQ:
3971 {
3972 int amount = sk_wmem_alloc_get(sk);
3973
3974 return put_user(amount, (int __user *)arg);
3975 }
3976 case SIOCINQ:
3977 {
3978 struct sk_buff *skb;
3979 int amount = 0;
3980
3981 spin_lock_bh(&sk->sk_receive_queue.lock);
3982 skb = skb_peek(&sk->sk_receive_queue);
3983 if (skb)
3984 amount = skb->len;
3985 spin_unlock_bh(&sk->sk_receive_queue.lock);
3986 return put_user(amount, (int __user *)arg);
3987 }
3988 case SIOCGSTAMP:
3989 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3990 case SIOCGSTAMPNS:
3991 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3992
3993#ifdef CONFIG_INET
3994 case SIOCADDRT:
3995 case SIOCDELRT:
3996 case SIOCDARP:
3997 case SIOCGARP:
3998 case SIOCSARP:
3999 case SIOCGIFADDR:
4000 case SIOCSIFADDR:
4001 case SIOCGIFBRDADDR:
4002 case SIOCSIFBRDADDR:
4003 case SIOCGIFNETMASK:
4004 case SIOCSIFNETMASK:
4005 case SIOCGIFDSTADDR:
4006 case SIOCSIFDSTADDR:
4007 case SIOCSIFFLAGS:
4008 return inet_dgram_ops.ioctl(sock, cmd, arg);
4009#endif
4010
4011 default:
4012 return -ENOIOCTLCMD;
4013 }
4014 return 0;
4015}
4016
4017static unsigned int packet_poll(struct file *file, struct socket *sock,
4018 poll_table *wait)
4019{
4020 struct sock *sk = sock->sk;
4021 struct packet_sock *po = pkt_sk(sk);
4022 unsigned int mask = datagram_poll(file, sock, wait);
4023
4024 spin_lock_bh(&sk->sk_receive_queue.lock);
4025 if (po->rx_ring.pg_vec) {
4026 if (!packet_previous_rx_frame(po, &po->rx_ring,
4027 TP_STATUS_KERNEL))
4028 mask |= POLLIN | POLLRDNORM;
4029 }
4030 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4031 po->pressure = 0;
4032 spin_unlock_bh(&sk->sk_receive_queue.lock);
4033 spin_lock_bh(&sk->sk_write_queue.lock);
4034 if (po->tx_ring.pg_vec) {
4035 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4036 mask |= POLLOUT | POLLWRNORM;
4037 }
4038 spin_unlock_bh(&sk->sk_write_queue.lock);
4039 return mask;
4040}
4041
4042
4043/* Dirty? Well, I still did not learn better way to account
4044 * for user mmaps.
4045 */
4046
4047static void packet_mm_open(struct vm_area_struct *vma)
4048{
4049 struct file *file = vma->vm_file;
4050 struct socket *sock = file->private_data;
4051 struct sock *sk = sock->sk;
4052
4053 if (sk)
4054 atomic_inc(&pkt_sk(sk)->mapped);
4055}
4056
4057static void packet_mm_close(struct vm_area_struct *vma)
4058{
4059 struct file *file = vma->vm_file;
4060 struct socket *sock = file->private_data;
4061 struct sock *sk = sock->sk;
4062
4063 if (sk)
4064 atomic_dec(&pkt_sk(sk)->mapped);
4065}
4066
4067static const struct vm_operations_struct packet_mmap_ops = {
4068 .open = packet_mm_open,
4069 .close = packet_mm_close,
4070};
4071
4072static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4073 unsigned int len)
4074{
4075 int i;
4076
4077 for (i = 0; i < len; i++) {
4078 if (likely(pg_vec[i].buffer)) {
4079 if (is_vmalloc_addr(pg_vec[i].buffer))
4080 vfree(pg_vec[i].buffer);
4081 else
4082 free_pages((unsigned long)pg_vec[i].buffer,
4083 order);
4084 pg_vec[i].buffer = NULL;
4085 }
4086 }
4087 kfree(pg_vec);
4088}
4089
4090static char *alloc_one_pg_vec_page(unsigned long order)
4091{
4092 char *buffer;
4093 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4094 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4095
4096 buffer = (char *) __get_free_pages(gfp_flags, order);
4097 if (buffer)
4098 return buffer;
4099
4100 /* __get_free_pages failed, fall back to vmalloc */
4101 buffer = vzalloc((1 << order) * PAGE_SIZE);
4102 if (buffer)
4103 return buffer;
4104
4105 /* vmalloc failed, lets dig into swap here */
4106 gfp_flags &= ~__GFP_NORETRY;
4107 buffer = (char *) __get_free_pages(gfp_flags, order);
4108 if (buffer)
4109 return buffer;
4110
4111 /* complete and utter failure */
4112 return NULL;
4113}
4114
4115static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4116{
4117 unsigned int block_nr = req->tp_block_nr;
4118 struct pgv *pg_vec;
4119 int i;
4120
4121 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4122 if (unlikely(!pg_vec))
4123 goto out;
4124
4125 for (i = 0; i < block_nr; i++) {
4126 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4127 if (unlikely(!pg_vec[i].buffer))
4128 goto out_free_pgvec;
4129 }
4130
4131out:
4132 return pg_vec;
4133
4134out_free_pgvec:
4135 free_pg_vec(pg_vec, order, block_nr);
4136 pg_vec = NULL;
4137 goto out;
4138}
4139
4140static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4141 int closing, int tx_ring)
4142{
4143 struct pgv *pg_vec = NULL;
4144 struct packet_sock *po = pkt_sk(sk);
4145 int was_running, order = 0;
4146 struct packet_ring_buffer *rb;
4147 struct sk_buff_head *rb_queue;
4148 __be16 num;
4149 int err = -EINVAL;
4150 /* Added to avoid minimal code churn */
4151 struct tpacket_req *req = &req_u->req;
4152
4153 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4154 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4155 net_warn_ratelimited("Tx-ring is not supported.\n");
4156 goto out;
4157 }
4158
4159 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4160 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4161
4162 err = -EBUSY;
4163 if (!closing) {
4164 if (atomic_read(&po->mapped))
4165 goto out;
4166 if (packet_read_pending(rb))
4167 goto out;
4168 }
4169
4170 if (req->tp_block_nr) {
4171 /* Sanity tests and some calculations */
4172 err = -EBUSY;
4173 if (unlikely(rb->pg_vec))
4174 goto out;
4175
4176 switch (po->tp_version) {
4177 case TPACKET_V1:
4178 po->tp_hdrlen = TPACKET_HDRLEN;
4179 break;
4180 case TPACKET_V2:
4181 po->tp_hdrlen = TPACKET2_HDRLEN;
4182 break;
4183 case TPACKET_V3:
4184 po->tp_hdrlen = TPACKET3_HDRLEN;
4185 break;
4186 }
4187
4188 err = -EINVAL;
4189 if (unlikely((int)req->tp_block_size <= 0))
4190 goto out;
4191 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4192 goto out;
4193 if (po->tp_version >= TPACKET_V3 &&
4194 (int)(req->tp_block_size -
4195 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4196 goto out;
4197 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4198 po->tp_reserve))
4199 goto out;
4200 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4201 goto out;
4202
4203 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4204 if (unlikely(rb->frames_per_block == 0))
4205 goto out;
4206 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4207 req->tp_frame_nr))
4208 goto out;
4209
4210 err = -ENOMEM;
4211 order = get_order(req->tp_block_size);
4212 pg_vec = alloc_pg_vec(req, order);
4213 if (unlikely(!pg_vec))
4214 goto out;
4215 switch (po->tp_version) {
4216 case TPACKET_V3:
4217 /* Transmit path is not supported. We checked
4218 * it above but just being paranoid
4219 */
4220 if (!tx_ring)
4221 init_prb_bdqc(po, rb, pg_vec, req_u);
4222 break;
4223 default:
4224 break;
4225 }
4226 }
4227 /* Done */
4228 else {
4229 err = -EINVAL;
4230 if (unlikely(req->tp_frame_nr))
4231 goto out;
4232 }
4233
4234 lock_sock(sk);
4235
4236 /* Detach socket from network */
4237 spin_lock(&po->bind_lock);
4238 was_running = po->running;
4239 num = po->num;
4240 if (was_running) {
4241 po->num = 0;
4242 __unregister_prot_hook(sk, false);
4243 }
4244 spin_unlock(&po->bind_lock);
4245
4246 synchronize_net();
4247
4248 err = -EBUSY;
4249 mutex_lock(&po->pg_vec_lock);
4250 if (closing || atomic_read(&po->mapped) == 0) {
4251 err = 0;
4252 spin_lock_bh(&rb_queue->lock);
4253 swap(rb->pg_vec, pg_vec);
4254 rb->frame_max = (req->tp_frame_nr - 1);
4255 rb->head = 0;
4256 rb->frame_size = req->tp_frame_size;
4257 spin_unlock_bh(&rb_queue->lock);
4258
4259 swap(rb->pg_vec_order, order);
4260 swap(rb->pg_vec_len, req->tp_block_nr);
4261
4262 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4263 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4264 tpacket_rcv : packet_rcv;
4265 skb_queue_purge(rb_queue);
4266 if (atomic_read(&po->mapped))
4267 pr_err("packet_mmap: vma is busy: %d\n",
4268 atomic_read(&po->mapped));
4269 }
4270 mutex_unlock(&po->pg_vec_lock);
4271
4272 spin_lock(&po->bind_lock);
4273 if (was_running) {
4274 po->num = num;
4275 register_prot_hook(sk);
4276 }
4277 spin_unlock(&po->bind_lock);
4278 if (closing && (po->tp_version > TPACKET_V2)) {
4279 /* Because we don't support block-based V3 on tx-ring */
4280 if (!tx_ring)
4281 prb_shutdown_retire_blk_timer(po, rb_queue);
4282 }
4283 release_sock(sk);
4284
4285 if (pg_vec)
4286 free_pg_vec(pg_vec, order, req->tp_block_nr);
4287out:
4288 return err;
4289}
4290
4291static int packet_mmap(struct file *file, struct socket *sock,
4292 struct vm_area_struct *vma)
4293{
4294 struct sock *sk = sock->sk;
4295 struct packet_sock *po = pkt_sk(sk);
4296 unsigned long size, expected_size;
4297 struct packet_ring_buffer *rb;
4298 unsigned long start;
4299 int err = -EINVAL;
4300 int i;
4301
4302 if (vma->vm_pgoff)
4303 return -EINVAL;
4304
4305 mutex_lock(&po->pg_vec_lock);
4306
4307 expected_size = 0;
4308 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4309 if (rb->pg_vec) {
4310 expected_size += rb->pg_vec_len
4311 * rb->pg_vec_pages
4312 * PAGE_SIZE;
4313 }
4314 }
4315
4316 if (expected_size == 0)
4317 goto out;
4318
4319 size = vma->vm_end - vma->vm_start;
4320 if (size != expected_size)
4321 goto out;
4322
4323 start = vma->vm_start;
4324 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4325 if (rb->pg_vec == NULL)
4326 continue;
4327
4328 for (i = 0; i < rb->pg_vec_len; i++) {
4329 struct page *page;
4330 void *kaddr = rb->pg_vec[i].buffer;
4331 int pg_num;
4332
4333 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4334 page = pgv_to_page(kaddr);
4335 err = vm_insert_page(vma, start, page);
4336 if (unlikely(err))
4337 goto out;
4338 start += PAGE_SIZE;
4339 kaddr += PAGE_SIZE;
4340 }
4341 }
4342 }
4343
4344 atomic_inc(&po->mapped);
4345 vma->vm_ops = &packet_mmap_ops;
4346 err = 0;
4347
4348out:
4349 mutex_unlock(&po->pg_vec_lock);
4350 return err;
4351}
4352
4353static const struct proto_ops packet_ops_spkt = {
4354 .family = PF_PACKET,
4355 .owner = THIS_MODULE,
4356 .release = packet_release,
4357 .bind = packet_bind_spkt,
4358 .connect = sock_no_connect,
4359 .socketpair = sock_no_socketpair,
4360 .accept = sock_no_accept,
4361 .getname = packet_getname_spkt,
4362 .poll = datagram_poll,
4363 .ioctl = packet_ioctl,
4364 .listen = sock_no_listen,
4365 .shutdown = sock_no_shutdown,
4366 .setsockopt = sock_no_setsockopt,
4367 .getsockopt = sock_no_getsockopt,
4368 .sendmsg = packet_sendmsg_spkt,
4369 .recvmsg = packet_recvmsg,
4370 .mmap = sock_no_mmap,
4371 .sendpage = sock_no_sendpage,
4372};
4373
4374static const struct proto_ops packet_ops = {
4375 .family = PF_PACKET,
4376 .owner = THIS_MODULE,
4377 .release = packet_release,
4378 .bind = packet_bind,
4379 .connect = sock_no_connect,
4380 .socketpair = sock_no_socketpair,
4381 .accept = sock_no_accept,
4382 .getname = packet_getname,
4383 .poll = packet_poll,
4384 .ioctl = packet_ioctl,
4385 .listen = sock_no_listen,
4386 .shutdown = sock_no_shutdown,
4387 .setsockopt = packet_setsockopt,
4388 .getsockopt = packet_getsockopt,
4389 .sendmsg = packet_sendmsg,
4390 .recvmsg = packet_recvmsg,
4391 .mmap = packet_mmap,
4392 .sendpage = sock_no_sendpage,
4393};
4394
4395static const struct net_proto_family packet_family_ops = {
4396 .family = PF_PACKET,
4397 .create = packet_create,
4398 .owner = THIS_MODULE,
4399};
4400
4401static struct notifier_block packet_netdev_notifier = {
4402 .notifier_call = packet_notifier,
4403};
4404
4405#ifdef CONFIG_PROC_FS
4406
4407static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4408 __acquires(RCU)
4409{
4410 struct net *net = seq_file_net(seq);
4411
4412 rcu_read_lock();
4413 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4414}
4415
4416static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4417{
4418 struct net *net = seq_file_net(seq);
4419 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4420}
4421
4422static void packet_seq_stop(struct seq_file *seq, void *v)
4423 __releases(RCU)
4424{
4425 rcu_read_unlock();
4426}
4427
4428static int packet_seq_show(struct seq_file *seq, void *v)
4429{
4430 if (v == SEQ_START_TOKEN)
4431 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4432 else {
4433 struct sock *s = sk_entry(v);
4434 const struct packet_sock *po = pkt_sk(s);
4435
4436 seq_printf(seq,
4437 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4438 s,
4439 atomic_read(&s->sk_refcnt),
4440 s->sk_type,
4441 ntohs(po->num),
4442 po->ifindex,
4443 po->running,
4444 atomic_read(&s->sk_rmem_alloc),
4445 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4446 sock_i_ino(s));
4447 }
4448
4449 return 0;
4450}
4451
4452static const struct seq_operations packet_seq_ops = {
4453 .start = packet_seq_start,
4454 .next = packet_seq_next,
4455 .stop = packet_seq_stop,
4456 .show = packet_seq_show,
4457};
4458
4459static int packet_seq_open(struct inode *inode, struct file *file)
4460{
4461 return seq_open_net(inode, file, &packet_seq_ops,
4462 sizeof(struct seq_net_private));
4463}
4464
4465static const struct file_operations packet_seq_fops = {
4466 .owner = THIS_MODULE,
4467 .open = packet_seq_open,
4468 .read = seq_read,
4469 .llseek = seq_lseek,
4470 .release = seq_release_net,
4471};
4472
4473#endif
4474
4475static int __net_init packet_net_init(struct net *net)
4476{
4477 mutex_init(&net->packet.sklist_lock);
4478 INIT_HLIST_HEAD(&net->packet.sklist);
4479
4480 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4481 return -ENOMEM;
4482
4483 return 0;
4484}
4485
4486static void __net_exit packet_net_exit(struct net *net)
4487{
4488 remove_proc_entry("packet", net->proc_net);
4489}
4490
4491static struct pernet_operations packet_net_ops = {
4492 .init = packet_net_init,
4493 .exit = packet_net_exit,
4494};
4495
4496
4497static void __exit packet_exit(void)
4498{
4499 unregister_netdevice_notifier(&packet_netdev_notifier);
4500 unregister_pernet_subsys(&packet_net_ops);
4501 sock_unregister(PF_PACKET);
4502 proto_unregister(&packet_proto);
4503}
4504
4505static int __init packet_init(void)
4506{
4507 int rc = proto_register(&packet_proto, 0);
4508
4509 if (rc != 0)
4510 goto out;
4511
4512 sock_register(&packet_family_ops);
4513 register_pernet_subsys(&packet_net_ops);
4514 register_netdevice_notifier(&packet_netdev_notifier);
4515out:
4516 return rc;
4517}
4518
4519module_init(packet_init);
4520module_exit(packet_exit);
4521MODULE_LICENSE("GPL");
4522MODULE_ALIAS_NETPROTO(PF_PACKET);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * PACKET - implements raw packet sockets.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 *
13 * Fixes:
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
42 * and packet_mreq.
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * layer.
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
47 */
48
49#include <linux/types.h>
50#include <linux/mm.h>
51#include <linux/capability.h>
52#include <linux/fcntl.h>
53#include <linux/socket.h>
54#include <linux/in.h>
55#include <linux/inet.h>
56#include <linux/netdevice.h>
57#include <linux/if_packet.h>
58#include <linux/wireless.h>
59#include <linux/kernel.h>
60#include <linux/kmod.h>
61#include <linux/slab.h>
62#include <linux/vmalloc.h>
63#include <net/net_namespace.h>
64#include <net/ip.h>
65#include <net/protocol.h>
66#include <linux/skbuff.h>
67#include <net/sock.h>
68#include <linux/errno.h>
69#include <linux/timer.h>
70#include <linux/uaccess.h>
71#include <asm/ioctls.h>
72#include <asm/page.h>
73#include <asm/cacheflush.h>
74#include <asm/io.h>
75#include <linux/proc_fs.h>
76#include <linux/seq_file.h>
77#include <linux/poll.h>
78#include <linux/module.h>
79#include <linux/init.h>
80#include <linux/mutex.h>
81#include <linux/if_vlan.h>
82#include <linux/virtio_net.h>
83#include <linux/errqueue.h>
84#include <linux/net_tstamp.h>
85#include <linux/percpu.h>
86#ifdef CONFIG_INET
87#include <net/inet_common.h>
88#endif
89#include <linux/bpf.h>
90#include <net/compat.h>
91
92#include "internal.h"
93
94/*
95 Assumptions:
96 - if device has no dev->hard_header routine, it adds and removes ll header
97 inside itself. In this case ll header is invisible outside of device,
98 but higher levels still should reserve dev->hard_header_len.
99 Some devices are enough clever to reallocate skb, when header
100 will not fit to reserved space (tunnel), another ones are silly
101 (PPP).
102 - packet socket receives packets with pulled ll header,
103 so that SOCK_RAW should push it back.
104
105On receive:
106-----------
107
108Incoming, dev->hard_header!=NULL
109 mac_header -> ll header
110 data -> data
111
112Outgoing, dev->hard_header!=NULL
113 mac_header -> ll header
114 data -> ll header
115
116Incoming, dev->hard_header==NULL
117 mac_header -> UNKNOWN position. It is very likely, that it points to ll
118 header. PPP makes it, that is wrong, because introduce
119 assymetry between rx and tx paths.
120 data -> data
121
122Outgoing, dev->hard_header==NULL
123 mac_header -> data. ll header is still not built!
124 data -> data
125
126Resume
127 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
128
129
130On transmit:
131------------
132
133dev->hard_header != NULL
134 mac_header -> ll header
135 data -> ll header
136
137dev->hard_header == NULL (ll header is added by device, we cannot control it)
138 mac_header -> data
139 data -> data
140
141 We should set nh.raw on output to correct posistion,
142 packet classifier depends on it.
143 */
144
145/* Private packet socket structures. */
146
147/* identical to struct packet_mreq except it has
148 * a longer address field.
149 */
150struct packet_mreq_max {
151 int mr_ifindex;
152 unsigned short mr_type;
153 unsigned short mr_alen;
154 unsigned char mr_address[MAX_ADDR_LEN];
155};
156
157union tpacket_uhdr {
158 struct tpacket_hdr *h1;
159 struct tpacket2_hdr *h2;
160 struct tpacket3_hdr *h3;
161 void *raw;
162};
163
164static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
165 int closing, int tx_ring);
166
167#define V3_ALIGNMENT (8)
168
169#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
170
171#define BLK_PLUS_PRIV(sz_of_priv) \
172 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
173
174#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
175#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
176#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
177#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
178#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
179#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
180#define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
181
182struct packet_sock;
183static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
184 struct packet_type *pt, struct net_device *orig_dev);
185
186static void *packet_previous_frame(struct packet_sock *po,
187 struct packet_ring_buffer *rb,
188 int status);
189static void packet_increment_head(struct packet_ring_buffer *buff);
190static int prb_curr_blk_in_use(struct tpacket_block_desc *);
191static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
192 struct packet_sock *);
193static void prb_retire_current_block(struct tpacket_kbdq_core *,
194 struct packet_sock *, unsigned int status);
195static int prb_queue_frozen(struct tpacket_kbdq_core *);
196static void prb_open_block(struct tpacket_kbdq_core *,
197 struct tpacket_block_desc *);
198static void prb_retire_rx_blk_timer_expired(struct timer_list *);
199static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
200static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
201static void prb_clear_rxhash(struct tpacket_kbdq_core *,
202 struct tpacket3_hdr *);
203static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
204 struct tpacket3_hdr *);
205static void packet_flush_mclist(struct sock *sk);
206static u16 packet_pick_tx_queue(struct sk_buff *skb);
207
208struct packet_skb_cb {
209 union {
210 struct sockaddr_pkt pkt;
211 union {
212 /* Trick: alias skb original length with
213 * ll.sll_family and ll.protocol in order
214 * to save room.
215 */
216 unsigned int origlen;
217 struct sockaddr_ll ll;
218 };
219 } sa;
220};
221
222#define vio_le() virtio_legacy_is_little_endian()
223
224#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
225
226#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
227#define GET_PBLOCK_DESC(x, bid) \
228 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
229#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
230 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
231#define GET_NEXT_PRB_BLK_NUM(x) \
232 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
233 ((x)->kactive_blk_num+1) : 0)
234
235static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
236static void __fanout_link(struct sock *sk, struct packet_sock *po);
237
238static int packet_direct_xmit(struct sk_buff *skb)
239{
240 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
241}
242
243static struct net_device *packet_cached_dev_get(struct packet_sock *po)
244{
245 struct net_device *dev;
246
247 rcu_read_lock();
248 dev = rcu_dereference(po->cached_dev);
249 if (likely(dev))
250 dev_hold(dev);
251 rcu_read_unlock();
252
253 return dev;
254}
255
256static void packet_cached_dev_assign(struct packet_sock *po,
257 struct net_device *dev)
258{
259 rcu_assign_pointer(po->cached_dev, dev);
260}
261
262static void packet_cached_dev_reset(struct packet_sock *po)
263{
264 RCU_INIT_POINTER(po->cached_dev, NULL);
265}
266
267static bool packet_use_direct_xmit(const struct packet_sock *po)
268{
269 return po->xmit == packet_direct_xmit;
270}
271
272static u16 packet_pick_tx_queue(struct sk_buff *skb)
273{
274 struct net_device *dev = skb->dev;
275 const struct net_device_ops *ops = dev->netdev_ops;
276 int cpu = raw_smp_processor_id();
277 u16 queue_index;
278
279#ifdef CONFIG_XPS
280 skb->sender_cpu = cpu + 1;
281#endif
282 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
283 if (ops->ndo_select_queue) {
284 queue_index = ops->ndo_select_queue(dev, skb, NULL);
285 queue_index = netdev_cap_txqueue(dev, queue_index);
286 } else {
287 queue_index = netdev_pick_tx(dev, skb, NULL);
288 }
289
290 return queue_index;
291}
292
293/* __register_prot_hook must be invoked through register_prot_hook
294 * or from a context in which asynchronous accesses to the packet
295 * socket is not possible (packet_create()).
296 */
297static void __register_prot_hook(struct sock *sk)
298{
299 struct packet_sock *po = pkt_sk(sk);
300
301 if (!po->running) {
302 if (po->fanout)
303 __fanout_link(sk, po);
304 else
305 dev_add_pack(&po->prot_hook);
306
307 sock_hold(sk);
308 po->running = 1;
309 }
310}
311
312static void register_prot_hook(struct sock *sk)
313{
314 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
315 __register_prot_hook(sk);
316}
317
318/* If the sync parameter is true, we will temporarily drop
319 * the po->bind_lock and do a synchronize_net to make sure no
320 * asynchronous packet processing paths still refer to the elements
321 * of po->prot_hook. If the sync parameter is false, it is the
322 * callers responsibility to take care of this.
323 */
324static void __unregister_prot_hook(struct sock *sk, bool sync)
325{
326 struct packet_sock *po = pkt_sk(sk);
327
328 lockdep_assert_held_once(&po->bind_lock);
329
330 po->running = 0;
331
332 if (po->fanout)
333 __fanout_unlink(sk, po);
334 else
335 __dev_remove_pack(&po->prot_hook);
336
337 __sock_put(sk);
338
339 if (sync) {
340 spin_unlock(&po->bind_lock);
341 synchronize_net();
342 spin_lock(&po->bind_lock);
343 }
344}
345
346static void unregister_prot_hook(struct sock *sk, bool sync)
347{
348 struct packet_sock *po = pkt_sk(sk);
349
350 if (po->running)
351 __unregister_prot_hook(sk, sync);
352}
353
354static inline struct page * __pure pgv_to_page(void *addr)
355{
356 if (is_vmalloc_addr(addr))
357 return vmalloc_to_page(addr);
358 return virt_to_page(addr);
359}
360
361static void __packet_set_status(struct packet_sock *po, void *frame, int status)
362{
363 union tpacket_uhdr h;
364
365 h.raw = frame;
366 switch (po->tp_version) {
367 case TPACKET_V1:
368 h.h1->tp_status = status;
369 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
370 break;
371 case TPACKET_V2:
372 h.h2->tp_status = status;
373 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
374 break;
375 case TPACKET_V3:
376 h.h3->tp_status = status;
377 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
378 break;
379 default:
380 WARN(1, "TPACKET version not supported.\n");
381 BUG();
382 }
383
384 smp_wmb();
385}
386
387static int __packet_get_status(const struct packet_sock *po, void *frame)
388{
389 union tpacket_uhdr h;
390
391 smp_rmb();
392
393 h.raw = frame;
394 switch (po->tp_version) {
395 case TPACKET_V1:
396 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
397 return h.h1->tp_status;
398 case TPACKET_V2:
399 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
400 return h.h2->tp_status;
401 case TPACKET_V3:
402 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
403 return h.h3->tp_status;
404 default:
405 WARN(1, "TPACKET version not supported.\n");
406 BUG();
407 return 0;
408 }
409}
410
411static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
412 unsigned int flags)
413{
414 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
415
416 if (shhwtstamps &&
417 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
418 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
419 return TP_STATUS_TS_RAW_HARDWARE;
420
421 if (ktime_to_timespec_cond(skb->tstamp, ts))
422 return TP_STATUS_TS_SOFTWARE;
423
424 return 0;
425}
426
427static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
428 struct sk_buff *skb)
429{
430 union tpacket_uhdr h;
431 struct timespec ts;
432 __u32 ts_status;
433
434 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
435 return 0;
436
437 h.raw = frame;
438 switch (po->tp_version) {
439 case TPACKET_V1:
440 h.h1->tp_sec = ts.tv_sec;
441 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
442 break;
443 case TPACKET_V2:
444 h.h2->tp_sec = ts.tv_sec;
445 h.h2->tp_nsec = ts.tv_nsec;
446 break;
447 case TPACKET_V3:
448 h.h3->tp_sec = ts.tv_sec;
449 h.h3->tp_nsec = ts.tv_nsec;
450 break;
451 default:
452 WARN(1, "TPACKET version not supported.\n");
453 BUG();
454 }
455
456 /* one flush is safe, as both fields always lie on the same cacheline */
457 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
458 smp_wmb();
459
460 return ts_status;
461}
462
463static void *packet_lookup_frame(const struct packet_sock *po,
464 const struct packet_ring_buffer *rb,
465 unsigned int position,
466 int status)
467{
468 unsigned int pg_vec_pos, frame_offset;
469 union tpacket_uhdr h;
470
471 pg_vec_pos = position / rb->frames_per_block;
472 frame_offset = position % rb->frames_per_block;
473
474 h.raw = rb->pg_vec[pg_vec_pos].buffer +
475 (frame_offset * rb->frame_size);
476
477 if (status != __packet_get_status(po, h.raw))
478 return NULL;
479
480 return h.raw;
481}
482
483static void *packet_current_frame(struct packet_sock *po,
484 struct packet_ring_buffer *rb,
485 int status)
486{
487 return packet_lookup_frame(po, rb, rb->head, status);
488}
489
490static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
491{
492 del_timer_sync(&pkc->retire_blk_timer);
493}
494
495static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
496 struct sk_buff_head *rb_queue)
497{
498 struct tpacket_kbdq_core *pkc;
499
500 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
501
502 spin_lock_bh(&rb_queue->lock);
503 pkc->delete_blk_timer = 1;
504 spin_unlock_bh(&rb_queue->lock);
505
506 prb_del_retire_blk_timer(pkc);
507}
508
509static void prb_setup_retire_blk_timer(struct packet_sock *po)
510{
511 struct tpacket_kbdq_core *pkc;
512
513 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
514 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
515 0);
516 pkc->retire_blk_timer.expires = jiffies;
517}
518
519static int prb_calc_retire_blk_tmo(struct packet_sock *po,
520 int blk_size_in_bytes)
521{
522 struct net_device *dev;
523 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
524 struct ethtool_link_ksettings ecmd;
525 int err;
526
527 rtnl_lock();
528 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
529 if (unlikely(!dev)) {
530 rtnl_unlock();
531 return DEFAULT_PRB_RETIRE_TOV;
532 }
533 err = __ethtool_get_link_ksettings(dev, &ecmd);
534 rtnl_unlock();
535 if (!err) {
536 /*
537 * If the link speed is so slow you don't really
538 * need to worry about perf anyways
539 */
540 if (ecmd.base.speed < SPEED_1000 ||
541 ecmd.base.speed == SPEED_UNKNOWN) {
542 return DEFAULT_PRB_RETIRE_TOV;
543 } else {
544 msec = 1;
545 div = ecmd.base.speed / 1000;
546 }
547 }
548
549 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
550
551 if (div)
552 mbits /= div;
553
554 tmo = mbits * msec;
555
556 if (div)
557 return tmo+1;
558 return tmo;
559}
560
561static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
562 union tpacket_req_u *req_u)
563{
564 p1->feature_req_word = req_u->req3.tp_feature_req_word;
565}
566
567static void init_prb_bdqc(struct packet_sock *po,
568 struct packet_ring_buffer *rb,
569 struct pgv *pg_vec,
570 union tpacket_req_u *req_u)
571{
572 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
573 struct tpacket_block_desc *pbd;
574
575 memset(p1, 0x0, sizeof(*p1));
576
577 p1->knxt_seq_num = 1;
578 p1->pkbdq = pg_vec;
579 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
580 p1->pkblk_start = pg_vec[0].buffer;
581 p1->kblk_size = req_u->req3.tp_block_size;
582 p1->knum_blocks = req_u->req3.tp_block_nr;
583 p1->hdrlen = po->tp_hdrlen;
584 p1->version = po->tp_version;
585 p1->last_kactive_blk_num = 0;
586 po->stats.stats3.tp_freeze_q_cnt = 0;
587 if (req_u->req3.tp_retire_blk_tov)
588 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
589 else
590 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
591 req_u->req3.tp_block_size);
592 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
593 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
594
595 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
596 prb_init_ft_ops(p1, req_u);
597 prb_setup_retire_blk_timer(po);
598 prb_open_block(p1, pbd);
599}
600
601/* Do NOT update the last_blk_num first.
602 * Assumes sk_buff_head lock is held.
603 */
604static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
605{
606 mod_timer(&pkc->retire_blk_timer,
607 jiffies + pkc->tov_in_jiffies);
608 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
609}
610
611/*
612 * Timer logic:
613 * 1) We refresh the timer only when we open a block.
614 * By doing this we don't waste cycles refreshing the timer
615 * on packet-by-packet basis.
616 *
617 * With a 1MB block-size, on a 1Gbps line, it will take
618 * i) ~8 ms to fill a block + ii) memcpy etc.
619 * In this cut we are not accounting for the memcpy time.
620 *
621 * So, if the user sets the 'tmo' to 10ms then the timer
622 * will never fire while the block is still getting filled
623 * (which is what we want). However, the user could choose
624 * to close a block early and that's fine.
625 *
626 * But when the timer does fire, we check whether or not to refresh it.
627 * Since the tmo granularity is in msecs, it is not too expensive
628 * to refresh the timer, lets say every '8' msecs.
629 * Either the user can set the 'tmo' or we can derive it based on
630 * a) line-speed and b) block-size.
631 * prb_calc_retire_blk_tmo() calculates the tmo.
632 *
633 */
634static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
635{
636 struct packet_sock *po =
637 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
638 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
639 unsigned int frozen;
640 struct tpacket_block_desc *pbd;
641
642 spin_lock(&po->sk.sk_receive_queue.lock);
643
644 frozen = prb_queue_frozen(pkc);
645 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
646
647 if (unlikely(pkc->delete_blk_timer))
648 goto out;
649
650 /* We only need to plug the race when the block is partially filled.
651 * tpacket_rcv:
652 * lock(); increment BLOCK_NUM_PKTS; unlock()
653 * copy_bits() is in progress ...
654 * timer fires on other cpu:
655 * we can't retire the current block because copy_bits
656 * is in progress.
657 *
658 */
659 if (BLOCK_NUM_PKTS(pbd)) {
660 while (atomic_read(&pkc->blk_fill_in_prog)) {
661 /* Waiting for skb_copy_bits to finish... */
662 cpu_relax();
663 }
664 }
665
666 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
667 if (!frozen) {
668 if (!BLOCK_NUM_PKTS(pbd)) {
669 /* An empty block. Just refresh the timer. */
670 goto refresh_timer;
671 }
672 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
673 if (!prb_dispatch_next_block(pkc, po))
674 goto refresh_timer;
675 else
676 goto out;
677 } else {
678 /* Case 1. Queue was frozen because user-space was
679 * lagging behind.
680 */
681 if (prb_curr_blk_in_use(pbd)) {
682 /*
683 * Ok, user-space is still behind.
684 * So just refresh the timer.
685 */
686 goto refresh_timer;
687 } else {
688 /* Case 2. queue was frozen,user-space caught up,
689 * now the link went idle && the timer fired.
690 * We don't have a block to close.So we open this
691 * block and restart the timer.
692 * opening a block thaws the queue,restarts timer
693 * Thawing/timer-refresh is a side effect.
694 */
695 prb_open_block(pkc, pbd);
696 goto out;
697 }
698 }
699 }
700
701refresh_timer:
702 _prb_refresh_rx_retire_blk_timer(pkc);
703
704out:
705 spin_unlock(&po->sk.sk_receive_queue.lock);
706}
707
708static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
709 struct tpacket_block_desc *pbd1, __u32 status)
710{
711 /* Flush everything minus the block header */
712
713#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
714 u8 *start, *end;
715
716 start = (u8 *)pbd1;
717
718 /* Skip the block header(we know header WILL fit in 4K) */
719 start += PAGE_SIZE;
720
721 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
722 for (; start < end; start += PAGE_SIZE)
723 flush_dcache_page(pgv_to_page(start));
724
725 smp_wmb();
726#endif
727
728 /* Now update the block status. */
729
730 BLOCK_STATUS(pbd1) = status;
731
732 /* Flush the block header */
733
734#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
735 start = (u8 *)pbd1;
736 flush_dcache_page(pgv_to_page(start));
737
738 smp_wmb();
739#endif
740}
741
742/*
743 * Side effect:
744 *
745 * 1) flush the block
746 * 2) Increment active_blk_num
747 *
748 * Note:We DONT refresh the timer on purpose.
749 * Because almost always the next block will be opened.
750 */
751static void prb_close_block(struct tpacket_kbdq_core *pkc1,
752 struct tpacket_block_desc *pbd1,
753 struct packet_sock *po, unsigned int stat)
754{
755 __u32 status = TP_STATUS_USER | stat;
756
757 struct tpacket3_hdr *last_pkt;
758 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
759 struct sock *sk = &po->sk;
760
761 if (atomic_read(&po->tp_drops))
762 status |= TP_STATUS_LOSING;
763
764 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
765 last_pkt->tp_next_offset = 0;
766
767 /* Get the ts of the last pkt */
768 if (BLOCK_NUM_PKTS(pbd1)) {
769 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
770 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
771 } else {
772 /* Ok, we tmo'd - so get the current time.
773 *
774 * It shouldn't really happen as we don't close empty
775 * blocks. See prb_retire_rx_blk_timer_expired().
776 */
777 struct timespec ts;
778 getnstimeofday(&ts);
779 h1->ts_last_pkt.ts_sec = ts.tv_sec;
780 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
781 }
782
783 smp_wmb();
784
785 /* Flush the block */
786 prb_flush_block(pkc1, pbd1, status);
787
788 sk->sk_data_ready(sk);
789
790 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
791}
792
793static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
794{
795 pkc->reset_pending_on_curr_blk = 0;
796}
797
798/*
799 * Side effect of opening a block:
800 *
801 * 1) prb_queue is thawed.
802 * 2) retire_blk_timer is refreshed.
803 *
804 */
805static void prb_open_block(struct tpacket_kbdq_core *pkc1,
806 struct tpacket_block_desc *pbd1)
807{
808 struct timespec ts;
809 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
810
811 smp_rmb();
812
813 /* We could have just memset this but we will lose the
814 * flexibility of making the priv area sticky
815 */
816
817 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
818 BLOCK_NUM_PKTS(pbd1) = 0;
819 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
820
821 getnstimeofday(&ts);
822
823 h1->ts_first_pkt.ts_sec = ts.tv_sec;
824 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
825
826 pkc1->pkblk_start = (char *)pbd1;
827 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
828
829 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
831
832 pbd1->version = pkc1->version;
833 pkc1->prev = pkc1->nxt_offset;
834 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
835
836 prb_thaw_queue(pkc1);
837 _prb_refresh_rx_retire_blk_timer(pkc1);
838
839 smp_wmb();
840}
841
842/*
843 * Queue freeze logic:
844 * 1) Assume tp_block_nr = 8 blocks.
845 * 2) At time 't0', user opens Rx ring.
846 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
847 * 4) user-space is either sleeping or processing block '0'.
848 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
849 * it will close block-7,loop around and try to fill block '0'.
850 * call-flow:
851 * __packet_lookup_frame_in_block
852 * prb_retire_current_block()
853 * prb_dispatch_next_block()
854 * |->(BLOCK_STATUS == USER) evaluates to true
855 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
856 * 6) Now there are two cases:
857 * 6.1) Link goes idle right after the queue is frozen.
858 * But remember, the last open_block() refreshed the timer.
859 * When this timer expires,it will refresh itself so that we can
860 * re-open block-0 in near future.
861 * 6.2) Link is busy and keeps on receiving packets. This is a simple
862 * case and __packet_lookup_frame_in_block will check if block-0
863 * is free and can now be re-used.
864 */
865static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
866 struct packet_sock *po)
867{
868 pkc->reset_pending_on_curr_blk = 1;
869 po->stats.stats3.tp_freeze_q_cnt++;
870}
871
872#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
873
874/*
875 * If the next block is free then we will dispatch it
876 * and return a good offset.
877 * Else, we will freeze the queue.
878 * So, caller must check the return value.
879 */
880static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
881 struct packet_sock *po)
882{
883 struct tpacket_block_desc *pbd;
884
885 smp_rmb();
886
887 /* 1. Get current block num */
888 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
889
890 /* 2. If this block is currently in_use then freeze the queue */
891 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
892 prb_freeze_queue(pkc, po);
893 return NULL;
894 }
895
896 /*
897 * 3.
898 * open this block and return the offset where the first packet
899 * needs to get stored.
900 */
901 prb_open_block(pkc, pbd);
902 return (void *)pkc->nxt_offset;
903}
904
905static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
906 struct packet_sock *po, unsigned int status)
907{
908 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
909
910 /* retire/close the current block */
911 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
912 /*
913 * Plug the case where copy_bits() is in progress on
914 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
915 * have space to copy the pkt in the current block and
916 * called prb_retire_current_block()
917 *
918 * We don't need to worry about the TMO case because
919 * the timer-handler already handled this case.
920 */
921 if (!(status & TP_STATUS_BLK_TMO)) {
922 while (atomic_read(&pkc->blk_fill_in_prog)) {
923 /* Waiting for skb_copy_bits to finish... */
924 cpu_relax();
925 }
926 }
927 prb_close_block(pkc, pbd, po, status);
928 return;
929 }
930}
931
932static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
933{
934 return TP_STATUS_USER & BLOCK_STATUS(pbd);
935}
936
937static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
938{
939 return pkc->reset_pending_on_curr_blk;
940}
941
942static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
943{
944 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
945 atomic_dec(&pkc->blk_fill_in_prog);
946}
947
948static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
949 struct tpacket3_hdr *ppd)
950{
951 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
952}
953
954static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
955 struct tpacket3_hdr *ppd)
956{
957 ppd->hv1.tp_rxhash = 0;
958}
959
960static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
961 struct tpacket3_hdr *ppd)
962{
963 if (skb_vlan_tag_present(pkc->skb)) {
964 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
965 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
966 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
967 } else {
968 ppd->hv1.tp_vlan_tci = 0;
969 ppd->hv1.tp_vlan_tpid = 0;
970 ppd->tp_status = TP_STATUS_AVAILABLE;
971 }
972}
973
974static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
975 struct tpacket3_hdr *ppd)
976{
977 ppd->hv1.tp_padding = 0;
978 prb_fill_vlan_info(pkc, ppd);
979
980 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
981 prb_fill_rxhash(pkc, ppd);
982 else
983 prb_clear_rxhash(pkc, ppd);
984}
985
986static void prb_fill_curr_block(char *curr,
987 struct tpacket_kbdq_core *pkc,
988 struct tpacket_block_desc *pbd,
989 unsigned int len)
990{
991 struct tpacket3_hdr *ppd;
992
993 ppd = (struct tpacket3_hdr *)curr;
994 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
995 pkc->prev = curr;
996 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
997 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
998 BLOCK_NUM_PKTS(pbd) += 1;
999 atomic_inc(&pkc->blk_fill_in_prog);
1000 prb_run_all_ft_ops(pkc, ppd);
1001}
1002
1003/* Assumes caller has the sk->rx_queue.lock */
1004static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1005 struct sk_buff *skb,
1006 unsigned int len
1007 )
1008{
1009 struct tpacket_kbdq_core *pkc;
1010 struct tpacket_block_desc *pbd;
1011 char *curr, *end;
1012
1013 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1014 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1015
1016 /* Queue is frozen when user space is lagging behind */
1017 if (prb_queue_frozen(pkc)) {
1018 /*
1019 * Check if that last block which caused the queue to freeze,
1020 * is still in_use by user-space.
1021 */
1022 if (prb_curr_blk_in_use(pbd)) {
1023 /* Can't record this packet */
1024 return NULL;
1025 } else {
1026 /*
1027 * Ok, the block was released by user-space.
1028 * Now let's open that block.
1029 * opening a block also thaws the queue.
1030 * Thawing is a side effect.
1031 */
1032 prb_open_block(pkc, pbd);
1033 }
1034 }
1035
1036 smp_mb();
1037 curr = pkc->nxt_offset;
1038 pkc->skb = skb;
1039 end = (char *)pbd + pkc->kblk_size;
1040
1041 /* first try the current block */
1042 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1043 prb_fill_curr_block(curr, pkc, pbd, len);
1044 return (void *)curr;
1045 }
1046
1047 /* Ok, close the current block */
1048 prb_retire_current_block(pkc, po, 0);
1049
1050 /* Now, try to dispatch the next block */
1051 curr = (char *)prb_dispatch_next_block(pkc, po);
1052 if (curr) {
1053 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1054 prb_fill_curr_block(curr, pkc, pbd, len);
1055 return (void *)curr;
1056 }
1057
1058 /*
1059 * No free blocks are available.user_space hasn't caught up yet.
1060 * Queue was just frozen and now this packet will get dropped.
1061 */
1062 return NULL;
1063}
1064
1065static void *packet_current_rx_frame(struct packet_sock *po,
1066 struct sk_buff *skb,
1067 int status, unsigned int len)
1068{
1069 char *curr = NULL;
1070 switch (po->tp_version) {
1071 case TPACKET_V1:
1072 case TPACKET_V2:
1073 curr = packet_lookup_frame(po, &po->rx_ring,
1074 po->rx_ring.head, status);
1075 return curr;
1076 case TPACKET_V3:
1077 return __packet_lookup_frame_in_block(po, skb, len);
1078 default:
1079 WARN(1, "TPACKET version not supported\n");
1080 BUG();
1081 return NULL;
1082 }
1083}
1084
1085static void *prb_lookup_block(const struct packet_sock *po,
1086 const struct packet_ring_buffer *rb,
1087 unsigned int idx,
1088 int status)
1089{
1090 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1091 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1092
1093 if (status != BLOCK_STATUS(pbd))
1094 return NULL;
1095 return pbd;
1096}
1097
1098static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1099{
1100 unsigned int prev;
1101 if (rb->prb_bdqc.kactive_blk_num)
1102 prev = rb->prb_bdqc.kactive_blk_num-1;
1103 else
1104 prev = rb->prb_bdqc.knum_blocks-1;
1105 return prev;
1106}
1107
1108/* Assumes caller has held the rx_queue.lock */
1109static void *__prb_previous_block(struct packet_sock *po,
1110 struct packet_ring_buffer *rb,
1111 int status)
1112{
1113 unsigned int previous = prb_previous_blk_num(rb);
1114 return prb_lookup_block(po, rb, previous, status);
1115}
1116
1117static void *packet_previous_rx_frame(struct packet_sock *po,
1118 struct packet_ring_buffer *rb,
1119 int status)
1120{
1121 if (po->tp_version <= TPACKET_V2)
1122 return packet_previous_frame(po, rb, status);
1123
1124 return __prb_previous_block(po, rb, status);
1125}
1126
1127static void packet_increment_rx_head(struct packet_sock *po,
1128 struct packet_ring_buffer *rb)
1129{
1130 switch (po->tp_version) {
1131 case TPACKET_V1:
1132 case TPACKET_V2:
1133 return packet_increment_head(rb);
1134 case TPACKET_V3:
1135 default:
1136 WARN(1, "TPACKET version not supported.\n");
1137 BUG();
1138 return;
1139 }
1140}
1141
1142static void *packet_previous_frame(struct packet_sock *po,
1143 struct packet_ring_buffer *rb,
1144 int status)
1145{
1146 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1147 return packet_lookup_frame(po, rb, previous, status);
1148}
1149
1150static void packet_increment_head(struct packet_ring_buffer *buff)
1151{
1152 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1153}
1154
1155static void packet_inc_pending(struct packet_ring_buffer *rb)
1156{
1157 this_cpu_inc(*rb->pending_refcnt);
1158}
1159
1160static void packet_dec_pending(struct packet_ring_buffer *rb)
1161{
1162 this_cpu_dec(*rb->pending_refcnt);
1163}
1164
1165static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1166{
1167 unsigned int refcnt = 0;
1168 int cpu;
1169
1170 /* We don't use pending refcount in rx_ring. */
1171 if (rb->pending_refcnt == NULL)
1172 return 0;
1173
1174 for_each_possible_cpu(cpu)
1175 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1176
1177 return refcnt;
1178}
1179
1180static int packet_alloc_pending(struct packet_sock *po)
1181{
1182 po->rx_ring.pending_refcnt = NULL;
1183
1184 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1185 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1186 return -ENOBUFS;
1187
1188 return 0;
1189}
1190
1191static void packet_free_pending(struct packet_sock *po)
1192{
1193 free_percpu(po->tx_ring.pending_refcnt);
1194}
1195
1196#define ROOM_POW_OFF 2
1197#define ROOM_NONE 0x0
1198#define ROOM_LOW 0x1
1199#define ROOM_NORMAL 0x2
1200
1201static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1202{
1203 int idx, len;
1204
1205 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1206 idx = READ_ONCE(po->rx_ring.head);
1207 if (pow_off)
1208 idx += len >> pow_off;
1209 if (idx >= len)
1210 idx -= len;
1211 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1212}
1213
1214static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1215{
1216 int idx, len;
1217
1218 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1219 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1220 if (pow_off)
1221 idx += len >> pow_off;
1222 if (idx >= len)
1223 idx -= len;
1224 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1225}
1226
1227static int __packet_rcv_has_room(const struct packet_sock *po,
1228 const struct sk_buff *skb)
1229{
1230 const struct sock *sk = &po->sk;
1231 int ret = ROOM_NONE;
1232
1233 if (po->prot_hook.func != tpacket_rcv) {
1234 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1235 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1236 - (skb ? skb->truesize : 0);
1237
1238 if (avail > (rcvbuf >> ROOM_POW_OFF))
1239 return ROOM_NORMAL;
1240 else if (avail > 0)
1241 return ROOM_LOW;
1242 else
1243 return ROOM_NONE;
1244 }
1245
1246 if (po->tp_version == TPACKET_V3) {
1247 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1248 ret = ROOM_NORMAL;
1249 else if (__tpacket_v3_has_room(po, 0))
1250 ret = ROOM_LOW;
1251 } else {
1252 if (__tpacket_has_room(po, ROOM_POW_OFF))
1253 ret = ROOM_NORMAL;
1254 else if (__tpacket_has_room(po, 0))
1255 ret = ROOM_LOW;
1256 }
1257
1258 return ret;
1259}
1260
1261static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1262{
1263 int pressure, ret;
1264
1265 ret = __packet_rcv_has_room(po, skb);
1266 pressure = ret != ROOM_NORMAL;
1267
1268 if (READ_ONCE(po->pressure) != pressure)
1269 WRITE_ONCE(po->pressure, pressure);
1270
1271 return ret;
1272}
1273
1274static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1275{
1276 if (READ_ONCE(po->pressure) &&
1277 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1278 WRITE_ONCE(po->pressure, 0);
1279}
1280
1281static void packet_sock_destruct(struct sock *sk)
1282{
1283 skb_queue_purge(&sk->sk_error_queue);
1284
1285 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1286 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1287
1288 if (!sock_flag(sk, SOCK_DEAD)) {
1289 pr_err("Attempt to release alive packet socket: %p\n", sk);
1290 return;
1291 }
1292
1293 sk_refcnt_debug_dec(sk);
1294}
1295
1296static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1297{
1298 u32 rxhash;
1299 int i, count = 0;
1300
1301 rxhash = skb_get_hash(skb);
1302 for (i = 0; i < ROLLOVER_HLEN; i++)
1303 if (po->rollover->history[i] == rxhash)
1304 count++;
1305
1306 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1307 return count > (ROLLOVER_HLEN >> 1);
1308}
1309
1310static unsigned int fanout_demux_hash(struct packet_fanout *f,
1311 struct sk_buff *skb,
1312 unsigned int num)
1313{
1314 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1315}
1316
1317static unsigned int fanout_demux_lb(struct packet_fanout *f,
1318 struct sk_buff *skb,
1319 unsigned int num)
1320{
1321 unsigned int val = atomic_inc_return(&f->rr_cur);
1322
1323 return val % num;
1324}
1325
1326static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1327 struct sk_buff *skb,
1328 unsigned int num)
1329{
1330 return smp_processor_id() % num;
1331}
1332
1333static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1334 struct sk_buff *skb,
1335 unsigned int num)
1336{
1337 return prandom_u32_max(num);
1338}
1339
1340static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1341 struct sk_buff *skb,
1342 unsigned int idx, bool try_self,
1343 unsigned int num)
1344{
1345 struct packet_sock *po, *po_next, *po_skip = NULL;
1346 unsigned int i, j, room = ROOM_NONE;
1347
1348 po = pkt_sk(f->arr[idx]);
1349
1350 if (try_self) {
1351 room = packet_rcv_has_room(po, skb);
1352 if (room == ROOM_NORMAL ||
1353 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1354 return idx;
1355 po_skip = po;
1356 }
1357
1358 i = j = min_t(int, po->rollover->sock, num - 1);
1359 do {
1360 po_next = pkt_sk(f->arr[i]);
1361 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1362 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1363 if (i != j)
1364 po->rollover->sock = i;
1365 atomic_long_inc(&po->rollover->num);
1366 if (room == ROOM_LOW)
1367 atomic_long_inc(&po->rollover->num_huge);
1368 return i;
1369 }
1370
1371 if (++i == num)
1372 i = 0;
1373 } while (i != j);
1374
1375 atomic_long_inc(&po->rollover->num_failed);
1376 return idx;
1377}
1378
1379static unsigned int fanout_demux_qm(struct packet_fanout *f,
1380 struct sk_buff *skb,
1381 unsigned int num)
1382{
1383 return skb_get_queue_mapping(skb) % num;
1384}
1385
1386static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1387 struct sk_buff *skb,
1388 unsigned int num)
1389{
1390 struct bpf_prog *prog;
1391 unsigned int ret = 0;
1392
1393 rcu_read_lock();
1394 prog = rcu_dereference(f->bpf_prog);
1395 if (prog)
1396 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1397 rcu_read_unlock();
1398
1399 return ret;
1400}
1401
1402static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1403{
1404 return f->flags & (flag >> 8);
1405}
1406
1407static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1408 struct packet_type *pt, struct net_device *orig_dev)
1409{
1410 struct packet_fanout *f = pt->af_packet_priv;
1411 unsigned int num = READ_ONCE(f->num_members);
1412 struct net *net = read_pnet(&f->net);
1413 struct packet_sock *po;
1414 unsigned int idx;
1415
1416 if (!net_eq(dev_net(dev), net) || !num) {
1417 kfree_skb(skb);
1418 return 0;
1419 }
1420
1421 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1422 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1423 if (!skb)
1424 return 0;
1425 }
1426 switch (f->type) {
1427 case PACKET_FANOUT_HASH:
1428 default:
1429 idx = fanout_demux_hash(f, skb, num);
1430 break;
1431 case PACKET_FANOUT_LB:
1432 idx = fanout_demux_lb(f, skb, num);
1433 break;
1434 case PACKET_FANOUT_CPU:
1435 idx = fanout_demux_cpu(f, skb, num);
1436 break;
1437 case PACKET_FANOUT_RND:
1438 idx = fanout_demux_rnd(f, skb, num);
1439 break;
1440 case PACKET_FANOUT_QM:
1441 idx = fanout_demux_qm(f, skb, num);
1442 break;
1443 case PACKET_FANOUT_ROLLOVER:
1444 idx = fanout_demux_rollover(f, skb, 0, false, num);
1445 break;
1446 case PACKET_FANOUT_CBPF:
1447 case PACKET_FANOUT_EBPF:
1448 idx = fanout_demux_bpf(f, skb, num);
1449 break;
1450 }
1451
1452 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1453 idx = fanout_demux_rollover(f, skb, idx, true, num);
1454
1455 po = pkt_sk(f->arr[idx]);
1456 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1457}
1458
1459DEFINE_MUTEX(fanout_mutex);
1460EXPORT_SYMBOL_GPL(fanout_mutex);
1461static LIST_HEAD(fanout_list);
1462static u16 fanout_next_id;
1463
1464static void __fanout_link(struct sock *sk, struct packet_sock *po)
1465{
1466 struct packet_fanout *f = po->fanout;
1467
1468 spin_lock(&f->lock);
1469 f->arr[f->num_members] = sk;
1470 smp_wmb();
1471 f->num_members++;
1472 if (f->num_members == 1)
1473 dev_add_pack(&f->prot_hook);
1474 spin_unlock(&f->lock);
1475}
1476
1477static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1478{
1479 struct packet_fanout *f = po->fanout;
1480 int i;
1481
1482 spin_lock(&f->lock);
1483 for (i = 0; i < f->num_members; i++) {
1484 if (f->arr[i] == sk)
1485 break;
1486 }
1487 BUG_ON(i >= f->num_members);
1488 f->arr[i] = f->arr[f->num_members - 1];
1489 f->num_members--;
1490 if (f->num_members == 0)
1491 __dev_remove_pack(&f->prot_hook);
1492 spin_unlock(&f->lock);
1493}
1494
1495static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1496{
1497 if (sk->sk_family != PF_PACKET)
1498 return false;
1499
1500 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1501}
1502
1503static void fanout_init_data(struct packet_fanout *f)
1504{
1505 switch (f->type) {
1506 case PACKET_FANOUT_LB:
1507 atomic_set(&f->rr_cur, 0);
1508 break;
1509 case PACKET_FANOUT_CBPF:
1510 case PACKET_FANOUT_EBPF:
1511 RCU_INIT_POINTER(f->bpf_prog, NULL);
1512 break;
1513 }
1514}
1515
1516static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1517{
1518 struct bpf_prog *old;
1519
1520 spin_lock(&f->lock);
1521 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1522 rcu_assign_pointer(f->bpf_prog, new);
1523 spin_unlock(&f->lock);
1524
1525 if (old) {
1526 synchronize_net();
1527 bpf_prog_destroy(old);
1528 }
1529}
1530
1531static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1532 unsigned int len)
1533{
1534 struct bpf_prog *new;
1535 struct sock_fprog fprog;
1536 int ret;
1537
1538 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1539 return -EPERM;
1540 if (len != sizeof(fprog))
1541 return -EINVAL;
1542 if (copy_from_user(&fprog, data, len))
1543 return -EFAULT;
1544
1545 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1546 if (ret)
1547 return ret;
1548
1549 __fanout_set_data_bpf(po->fanout, new);
1550 return 0;
1551}
1552
1553static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1554 unsigned int len)
1555{
1556 struct bpf_prog *new;
1557 u32 fd;
1558
1559 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1560 return -EPERM;
1561 if (len != sizeof(fd))
1562 return -EINVAL;
1563 if (copy_from_user(&fd, data, len))
1564 return -EFAULT;
1565
1566 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1567 if (IS_ERR(new))
1568 return PTR_ERR(new);
1569
1570 __fanout_set_data_bpf(po->fanout, new);
1571 return 0;
1572}
1573
1574static int fanout_set_data(struct packet_sock *po, char __user *data,
1575 unsigned int len)
1576{
1577 switch (po->fanout->type) {
1578 case PACKET_FANOUT_CBPF:
1579 return fanout_set_data_cbpf(po, data, len);
1580 case PACKET_FANOUT_EBPF:
1581 return fanout_set_data_ebpf(po, data, len);
1582 default:
1583 return -EINVAL;
1584 }
1585}
1586
1587static void fanout_release_data(struct packet_fanout *f)
1588{
1589 switch (f->type) {
1590 case PACKET_FANOUT_CBPF:
1591 case PACKET_FANOUT_EBPF:
1592 __fanout_set_data_bpf(f, NULL);
1593 }
1594}
1595
1596static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1597{
1598 struct packet_fanout *f;
1599
1600 list_for_each_entry(f, &fanout_list, list) {
1601 if (f->id == candidate_id &&
1602 read_pnet(&f->net) == sock_net(sk)) {
1603 return false;
1604 }
1605 }
1606 return true;
1607}
1608
1609static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1610{
1611 u16 id = fanout_next_id;
1612
1613 do {
1614 if (__fanout_id_is_free(sk, id)) {
1615 *new_id = id;
1616 fanout_next_id = id + 1;
1617 return true;
1618 }
1619
1620 id++;
1621 } while (id != fanout_next_id);
1622
1623 return false;
1624}
1625
1626static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1627{
1628 struct packet_rollover *rollover = NULL;
1629 struct packet_sock *po = pkt_sk(sk);
1630 struct packet_fanout *f, *match;
1631 u8 type = type_flags & 0xff;
1632 u8 flags = type_flags >> 8;
1633 int err;
1634
1635 switch (type) {
1636 case PACKET_FANOUT_ROLLOVER:
1637 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1638 return -EINVAL;
1639 case PACKET_FANOUT_HASH:
1640 case PACKET_FANOUT_LB:
1641 case PACKET_FANOUT_CPU:
1642 case PACKET_FANOUT_RND:
1643 case PACKET_FANOUT_QM:
1644 case PACKET_FANOUT_CBPF:
1645 case PACKET_FANOUT_EBPF:
1646 break;
1647 default:
1648 return -EINVAL;
1649 }
1650
1651 mutex_lock(&fanout_mutex);
1652
1653 err = -EALREADY;
1654 if (po->fanout)
1655 goto out;
1656
1657 if (type == PACKET_FANOUT_ROLLOVER ||
1658 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1659 err = -ENOMEM;
1660 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1661 if (!rollover)
1662 goto out;
1663 atomic_long_set(&rollover->num, 0);
1664 atomic_long_set(&rollover->num_huge, 0);
1665 atomic_long_set(&rollover->num_failed, 0);
1666 }
1667
1668 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1669 if (id != 0) {
1670 err = -EINVAL;
1671 goto out;
1672 }
1673 if (!fanout_find_new_id(sk, &id)) {
1674 err = -ENOMEM;
1675 goto out;
1676 }
1677 /* ephemeral flag for the first socket in the group: drop it */
1678 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1679 }
1680
1681 match = NULL;
1682 list_for_each_entry(f, &fanout_list, list) {
1683 if (f->id == id &&
1684 read_pnet(&f->net) == sock_net(sk)) {
1685 match = f;
1686 break;
1687 }
1688 }
1689 err = -EINVAL;
1690 if (match && match->flags != flags)
1691 goto out;
1692 if (!match) {
1693 err = -ENOMEM;
1694 match = kzalloc(sizeof(*match), GFP_KERNEL);
1695 if (!match)
1696 goto out;
1697 write_pnet(&match->net, sock_net(sk));
1698 match->id = id;
1699 match->type = type;
1700 match->flags = flags;
1701 INIT_LIST_HEAD(&match->list);
1702 spin_lock_init(&match->lock);
1703 refcount_set(&match->sk_ref, 0);
1704 fanout_init_data(match);
1705 match->prot_hook.type = po->prot_hook.type;
1706 match->prot_hook.dev = po->prot_hook.dev;
1707 match->prot_hook.func = packet_rcv_fanout;
1708 match->prot_hook.af_packet_priv = match;
1709 match->prot_hook.id_match = match_fanout_group;
1710 list_add(&match->list, &fanout_list);
1711 }
1712 err = -EINVAL;
1713
1714 spin_lock(&po->bind_lock);
1715 if (po->running &&
1716 match->type == type &&
1717 match->prot_hook.type == po->prot_hook.type &&
1718 match->prot_hook.dev == po->prot_hook.dev) {
1719 err = -ENOSPC;
1720 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1721 __dev_remove_pack(&po->prot_hook);
1722 po->fanout = match;
1723 po->rollover = rollover;
1724 rollover = NULL;
1725 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1726 __fanout_link(sk, po);
1727 err = 0;
1728 }
1729 }
1730 spin_unlock(&po->bind_lock);
1731
1732 if (err && !refcount_read(&match->sk_ref)) {
1733 list_del(&match->list);
1734 kfree(match);
1735 }
1736
1737out:
1738 kfree(rollover);
1739 mutex_unlock(&fanout_mutex);
1740 return err;
1741}
1742
1743/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1744 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1745 * It is the responsibility of the caller to call fanout_release_data() and
1746 * free the returned packet_fanout (after synchronize_net())
1747 */
1748static struct packet_fanout *fanout_release(struct sock *sk)
1749{
1750 struct packet_sock *po = pkt_sk(sk);
1751 struct packet_fanout *f;
1752
1753 mutex_lock(&fanout_mutex);
1754 f = po->fanout;
1755 if (f) {
1756 po->fanout = NULL;
1757
1758 if (refcount_dec_and_test(&f->sk_ref))
1759 list_del(&f->list);
1760 else
1761 f = NULL;
1762 }
1763 mutex_unlock(&fanout_mutex);
1764
1765 return f;
1766}
1767
1768static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1769 struct sk_buff *skb)
1770{
1771 /* Earlier code assumed this would be a VLAN pkt, double-check
1772 * this now that we have the actual packet in hand. We can only
1773 * do this check on Ethernet devices.
1774 */
1775 if (unlikely(dev->type != ARPHRD_ETHER))
1776 return false;
1777
1778 skb_reset_mac_header(skb);
1779 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1780}
1781
1782static const struct proto_ops packet_ops;
1783
1784static const struct proto_ops packet_ops_spkt;
1785
1786static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1787 struct packet_type *pt, struct net_device *orig_dev)
1788{
1789 struct sock *sk;
1790 struct sockaddr_pkt *spkt;
1791
1792 /*
1793 * When we registered the protocol we saved the socket in the data
1794 * field for just this event.
1795 */
1796
1797 sk = pt->af_packet_priv;
1798
1799 /*
1800 * Yank back the headers [hope the device set this
1801 * right or kerboom...]
1802 *
1803 * Incoming packets have ll header pulled,
1804 * push it back.
1805 *
1806 * For outgoing ones skb->data == skb_mac_header(skb)
1807 * so that this procedure is noop.
1808 */
1809
1810 if (skb->pkt_type == PACKET_LOOPBACK)
1811 goto out;
1812
1813 if (!net_eq(dev_net(dev), sock_net(sk)))
1814 goto out;
1815
1816 skb = skb_share_check(skb, GFP_ATOMIC);
1817 if (skb == NULL)
1818 goto oom;
1819
1820 /* drop any routing info */
1821 skb_dst_drop(skb);
1822
1823 /* drop conntrack reference */
1824 nf_reset_ct(skb);
1825
1826 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1827
1828 skb_push(skb, skb->data - skb_mac_header(skb));
1829
1830 /*
1831 * The SOCK_PACKET socket receives _all_ frames.
1832 */
1833
1834 spkt->spkt_family = dev->type;
1835 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1836 spkt->spkt_protocol = skb->protocol;
1837
1838 /*
1839 * Charge the memory to the socket. This is done specifically
1840 * to prevent sockets using all the memory up.
1841 */
1842
1843 if (sock_queue_rcv_skb(sk, skb) == 0)
1844 return 0;
1845
1846out:
1847 kfree_skb(skb);
1848oom:
1849 return 0;
1850}
1851
1852static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1853{
1854 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1855 sock->type == SOCK_RAW) {
1856 skb_reset_mac_header(skb);
1857 skb->protocol = dev_parse_header_protocol(skb);
1858 }
1859
1860 skb_probe_transport_header(skb);
1861}
1862
1863/*
1864 * Output a raw packet to a device layer. This bypasses all the other
1865 * protocol layers and you must therefore supply it with a complete frame
1866 */
1867
1868static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1869 size_t len)
1870{
1871 struct sock *sk = sock->sk;
1872 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1873 struct sk_buff *skb = NULL;
1874 struct net_device *dev;
1875 struct sockcm_cookie sockc;
1876 __be16 proto = 0;
1877 int err;
1878 int extra_len = 0;
1879
1880 /*
1881 * Get and verify the address.
1882 */
1883
1884 if (saddr) {
1885 if (msg->msg_namelen < sizeof(struct sockaddr))
1886 return -EINVAL;
1887 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1888 proto = saddr->spkt_protocol;
1889 } else
1890 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1891
1892 /*
1893 * Find the device first to size check it
1894 */
1895
1896 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1897retry:
1898 rcu_read_lock();
1899 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1900 err = -ENODEV;
1901 if (dev == NULL)
1902 goto out_unlock;
1903
1904 err = -ENETDOWN;
1905 if (!(dev->flags & IFF_UP))
1906 goto out_unlock;
1907
1908 /*
1909 * You may not queue a frame bigger than the mtu. This is the lowest level
1910 * raw protocol and you must do your own fragmentation at this level.
1911 */
1912
1913 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1914 if (!netif_supports_nofcs(dev)) {
1915 err = -EPROTONOSUPPORT;
1916 goto out_unlock;
1917 }
1918 extra_len = 4; /* We're doing our own CRC */
1919 }
1920
1921 err = -EMSGSIZE;
1922 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1923 goto out_unlock;
1924
1925 if (!skb) {
1926 size_t reserved = LL_RESERVED_SPACE(dev);
1927 int tlen = dev->needed_tailroom;
1928 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1929
1930 rcu_read_unlock();
1931 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1932 if (skb == NULL)
1933 return -ENOBUFS;
1934 /* FIXME: Save some space for broken drivers that write a hard
1935 * header at transmission time by themselves. PPP is the notable
1936 * one here. This should really be fixed at the driver level.
1937 */
1938 skb_reserve(skb, reserved);
1939 skb_reset_network_header(skb);
1940
1941 /* Try to align data part correctly */
1942 if (hhlen) {
1943 skb->data -= hhlen;
1944 skb->tail -= hhlen;
1945 if (len < hhlen)
1946 skb_reset_network_header(skb);
1947 }
1948 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1949 if (err)
1950 goto out_free;
1951 goto retry;
1952 }
1953
1954 if (!dev_validate_header(dev, skb->data, len)) {
1955 err = -EINVAL;
1956 goto out_unlock;
1957 }
1958 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1959 !packet_extra_vlan_len_allowed(dev, skb)) {
1960 err = -EMSGSIZE;
1961 goto out_unlock;
1962 }
1963
1964 sockcm_init(&sockc, sk);
1965 if (msg->msg_controllen) {
1966 err = sock_cmsg_send(sk, msg, &sockc);
1967 if (unlikely(err))
1968 goto out_unlock;
1969 }
1970
1971 skb->protocol = proto;
1972 skb->dev = dev;
1973 skb->priority = sk->sk_priority;
1974 skb->mark = sk->sk_mark;
1975 skb->tstamp = sockc.transmit_time;
1976
1977 skb_setup_tx_timestamp(skb, sockc.tsflags);
1978
1979 if (unlikely(extra_len == 4))
1980 skb->no_fcs = 1;
1981
1982 packet_parse_headers(skb, sock);
1983
1984 dev_queue_xmit(skb);
1985 rcu_read_unlock();
1986 return len;
1987
1988out_unlock:
1989 rcu_read_unlock();
1990out_free:
1991 kfree_skb(skb);
1992 return err;
1993}
1994
1995static unsigned int run_filter(struct sk_buff *skb,
1996 const struct sock *sk,
1997 unsigned int res)
1998{
1999 struct sk_filter *filter;
2000
2001 rcu_read_lock();
2002 filter = rcu_dereference(sk->sk_filter);
2003 if (filter != NULL)
2004 res = bpf_prog_run_clear_cb(filter->prog, skb);
2005 rcu_read_unlock();
2006
2007 return res;
2008}
2009
2010static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2011 size_t *len)
2012{
2013 struct virtio_net_hdr vnet_hdr;
2014
2015 if (*len < sizeof(vnet_hdr))
2016 return -EINVAL;
2017 *len -= sizeof(vnet_hdr);
2018
2019 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2020 return -EINVAL;
2021
2022 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2023}
2024
2025/*
2026 * This function makes lazy skb cloning in hope that most of packets
2027 * are discarded by BPF.
2028 *
2029 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2030 * and skb->cb are mangled. It works because (and until) packets
2031 * falling here are owned by current CPU. Output packets are cloned
2032 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2033 * sequencially, so that if we return skb to original state on exit,
2034 * we will not harm anyone.
2035 */
2036
2037static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2038 struct packet_type *pt, struct net_device *orig_dev)
2039{
2040 struct sock *sk;
2041 struct sockaddr_ll *sll;
2042 struct packet_sock *po;
2043 u8 *skb_head = skb->data;
2044 int skb_len = skb->len;
2045 unsigned int snaplen, res;
2046 bool is_drop_n_account = false;
2047
2048 if (skb->pkt_type == PACKET_LOOPBACK)
2049 goto drop;
2050
2051 sk = pt->af_packet_priv;
2052 po = pkt_sk(sk);
2053
2054 if (!net_eq(dev_net(dev), sock_net(sk)))
2055 goto drop;
2056
2057 skb->dev = dev;
2058
2059 if (dev->header_ops) {
2060 /* The device has an explicit notion of ll header,
2061 * exported to higher levels.
2062 *
2063 * Otherwise, the device hides details of its frame
2064 * structure, so that corresponding packet head is
2065 * never delivered to user.
2066 */
2067 if (sk->sk_type != SOCK_DGRAM)
2068 skb_push(skb, skb->data - skb_mac_header(skb));
2069 else if (skb->pkt_type == PACKET_OUTGOING) {
2070 /* Special case: outgoing packets have ll header at head */
2071 skb_pull(skb, skb_network_offset(skb));
2072 }
2073 }
2074
2075 snaplen = skb->len;
2076
2077 res = run_filter(skb, sk, snaplen);
2078 if (!res)
2079 goto drop_n_restore;
2080 if (snaplen > res)
2081 snaplen = res;
2082
2083 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2084 goto drop_n_acct;
2085
2086 if (skb_shared(skb)) {
2087 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2088 if (nskb == NULL)
2089 goto drop_n_acct;
2090
2091 if (skb_head != skb->data) {
2092 skb->data = skb_head;
2093 skb->len = skb_len;
2094 }
2095 consume_skb(skb);
2096 skb = nskb;
2097 }
2098
2099 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2100
2101 sll = &PACKET_SKB_CB(skb)->sa.ll;
2102 sll->sll_hatype = dev->type;
2103 sll->sll_pkttype = skb->pkt_type;
2104 if (unlikely(po->origdev))
2105 sll->sll_ifindex = orig_dev->ifindex;
2106 else
2107 sll->sll_ifindex = dev->ifindex;
2108
2109 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2110
2111 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2112 * Use their space for storing the original skb length.
2113 */
2114 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2115
2116 if (pskb_trim(skb, snaplen))
2117 goto drop_n_acct;
2118
2119 skb_set_owner_r(skb, sk);
2120 skb->dev = NULL;
2121 skb_dst_drop(skb);
2122
2123 /* drop conntrack reference */
2124 nf_reset_ct(skb);
2125
2126 spin_lock(&sk->sk_receive_queue.lock);
2127 po->stats.stats1.tp_packets++;
2128 sock_skb_set_dropcount(sk, skb);
2129 __skb_queue_tail(&sk->sk_receive_queue, skb);
2130 spin_unlock(&sk->sk_receive_queue.lock);
2131 sk->sk_data_ready(sk);
2132 return 0;
2133
2134drop_n_acct:
2135 is_drop_n_account = true;
2136 atomic_inc(&po->tp_drops);
2137 atomic_inc(&sk->sk_drops);
2138
2139drop_n_restore:
2140 if (skb_head != skb->data && skb_shared(skb)) {
2141 skb->data = skb_head;
2142 skb->len = skb_len;
2143 }
2144drop:
2145 if (!is_drop_n_account)
2146 consume_skb(skb);
2147 else
2148 kfree_skb(skb);
2149 return 0;
2150}
2151
2152static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2153 struct packet_type *pt, struct net_device *orig_dev)
2154{
2155 struct sock *sk;
2156 struct packet_sock *po;
2157 struct sockaddr_ll *sll;
2158 union tpacket_uhdr h;
2159 u8 *skb_head = skb->data;
2160 int skb_len = skb->len;
2161 unsigned int snaplen, res;
2162 unsigned long status = TP_STATUS_USER;
2163 unsigned short macoff, netoff, hdrlen;
2164 struct sk_buff *copy_skb = NULL;
2165 struct timespec ts;
2166 __u32 ts_status;
2167 bool is_drop_n_account = false;
2168 bool do_vnet = false;
2169
2170 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2171 * We may add members to them until current aligned size without forcing
2172 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2173 */
2174 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2175 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2176
2177 if (skb->pkt_type == PACKET_LOOPBACK)
2178 goto drop;
2179
2180 sk = pt->af_packet_priv;
2181 po = pkt_sk(sk);
2182
2183 if (!net_eq(dev_net(dev), sock_net(sk)))
2184 goto drop;
2185
2186 if (dev->header_ops) {
2187 if (sk->sk_type != SOCK_DGRAM)
2188 skb_push(skb, skb->data - skb_mac_header(skb));
2189 else if (skb->pkt_type == PACKET_OUTGOING) {
2190 /* Special case: outgoing packets have ll header at head */
2191 skb_pull(skb, skb_network_offset(skb));
2192 }
2193 }
2194
2195 snaplen = skb->len;
2196
2197 res = run_filter(skb, sk, snaplen);
2198 if (!res)
2199 goto drop_n_restore;
2200
2201 /* If we are flooded, just give up */
2202 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2203 atomic_inc(&po->tp_drops);
2204 goto drop_n_restore;
2205 }
2206
2207 if (skb->ip_summed == CHECKSUM_PARTIAL)
2208 status |= TP_STATUS_CSUMNOTREADY;
2209 else if (skb->pkt_type != PACKET_OUTGOING &&
2210 (skb->ip_summed == CHECKSUM_COMPLETE ||
2211 skb_csum_unnecessary(skb)))
2212 status |= TP_STATUS_CSUM_VALID;
2213
2214 if (snaplen > res)
2215 snaplen = res;
2216
2217 if (sk->sk_type == SOCK_DGRAM) {
2218 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2219 po->tp_reserve;
2220 } else {
2221 unsigned int maclen = skb_network_offset(skb);
2222 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2223 (maclen < 16 ? 16 : maclen)) +
2224 po->tp_reserve;
2225 if (po->has_vnet_hdr) {
2226 netoff += sizeof(struct virtio_net_hdr);
2227 do_vnet = true;
2228 }
2229 macoff = netoff - maclen;
2230 }
2231 if (po->tp_version <= TPACKET_V2) {
2232 if (macoff + snaplen > po->rx_ring.frame_size) {
2233 if (po->copy_thresh &&
2234 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2235 if (skb_shared(skb)) {
2236 copy_skb = skb_clone(skb, GFP_ATOMIC);
2237 } else {
2238 copy_skb = skb_get(skb);
2239 skb_head = skb->data;
2240 }
2241 if (copy_skb)
2242 skb_set_owner_r(copy_skb, sk);
2243 }
2244 snaplen = po->rx_ring.frame_size - macoff;
2245 if ((int)snaplen < 0) {
2246 snaplen = 0;
2247 do_vnet = false;
2248 }
2249 }
2250 } else if (unlikely(macoff + snaplen >
2251 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2252 u32 nval;
2253
2254 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2255 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2256 snaplen, nval, macoff);
2257 snaplen = nval;
2258 if (unlikely((int)snaplen < 0)) {
2259 snaplen = 0;
2260 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2261 do_vnet = false;
2262 }
2263 }
2264 spin_lock(&sk->sk_receive_queue.lock);
2265 h.raw = packet_current_rx_frame(po, skb,
2266 TP_STATUS_KERNEL, (macoff+snaplen));
2267 if (!h.raw)
2268 goto drop_n_account;
2269 if (po->tp_version <= TPACKET_V2) {
2270 packet_increment_rx_head(po, &po->rx_ring);
2271 /*
2272 * LOSING will be reported till you read the stats,
2273 * because it's COR - Clear On Read.
2274 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2275 * at packet level.
2276 */
2277 if (atomic_read(&po->tp_drops))
2278 status |= TP_STATUS_LOSING;
2279 }
2280
2281 if (do_vnet &&
2282 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2283 sizeof(struct virtio_net_hdr),
2284 vio_le(), true, 0))
2285 goto drop_n_account;
2286
2287 po->stats.stats1.tp_packets++;
2288 if (copy_skb) {
2289 status |= TP_STATUS_COPY;
2290 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2291 }
2292 spin_unlock(&sk->sk_receive_queue.lock);
2293
2294 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2295
2296 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2297 getnstimeofday(&ts);
2298
2299 status |= ts_status;
2300
2301 switch (po->tp_version) {
2302 case TPACKET_V1:
2303 h.h1->tp_len = skb->len;
2304 h.h1->tp_snaplen = snaplen;
2305 h.h1->tp_mac = macoff;
2306 h.h1->tp_net = netoff;
2307 h.h1->tp_sec = ts.tv_sec;
2308 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2309 hdrlen = sizeof(*h.h1);
2310 break;
2311 case TPACKET_V2:
2312 h.h2->tp_len = skb->len;
2313 h.h2->tp_snaplen = snaplen;
2314 h.h2->tp_mac = macoff;
2315 h.h2->tp_net = netoff;
2316 h.h2->tp_sec = ts.tv_sec;
2317 h.h2->tp_nsec = ts.tv_nsec;
2318 if (skb_vlan_tag_present(skb)) {
2319 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2320 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2321 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2322 } else {
2323 h.h2->tp_vlan_tci = 0;
2324 h.h2->tp_vlan_tpid = 0;
2325 }
2326 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2327 hdrlen = sizeof(*h.h2);
2328 break;
2329 case TPACKET_V3:
2330 /* tp_nxt_offset,vlan are already populated above.
2331 * So DONT clear those fields here
2332 */
2333 h.h3->tp_status |= status;
2334 h.h3->tp_len = skb->len;
2335 h.h3->tp_snaplen = snaplen;
2336 h.h3->tp_mac = macoff;
2337 h.h3->tp_net = netoff;
2338 h.h3->tp_sec = ts.tv_sec;
2339 h.h3->tp_nsec = ts.tv_nsec;
2340 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2341 hdrlen = sizeof(*h.h3);
2342 break;
2343 default:
2344 BUG();
2345 }
2346
2347 sll = h.raw + TPACKET_ALIGN(hdrlen);
2348 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2349 sll->sll_family = AF_PACKET;
2350 sll->sll_hatype = dev->type;
2351 sll->sll_protocol = skb->protocol;
2352 sll->sll_pkttype = skb->pkt_type;
2353 if (unlikely(po->origdev))
2354 sll->sll_ifindex = orig_dev->ifindex;
2355 else
2356 sll->sll_ifindex = dev->ifindex;
2357
2358 smp_mb();
2359
2360#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2361 if (po->tp_version <= TPACKET_V2) {
2362 u8 *start, *end;
2363
2364 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2365 macoff + snaplen);
2366
2367 for (start = h.raw; start < end; start += PAGE_SIZE)
2368 flush_dcache_page(pgv_to_page(start));
2369 }
2370 smp_wmb();
2371#endif
2372
2373 if (po->tp_version <= TPACKET_V2) {
2374 __packet_set_status(po, h.raw, status);
2375 sk->sk_data_ready(sk);
2376 } else {
2377 prb_clear_blk_fill_status(&po->rx_ring);
2378 }
2379
2380drop_n_restore:
2381 if (skb_head != skb->data && skb_shared(skb)) {
2382 skb->data = skb_head;
2383 skb->len = skb_len;
2384 }
2385drop:
2386 if (!is_drop_n_account)
2387 consume_skb(skb);
2388 else
2389 kfree_skb(skb);
2390 return 0;
2391
2392drop_n_account:
2393 spin_unlock(&sk->sk_receive_queue.lock);
2394 atomic_inc(&po->tp_drops);
2395 is_drop_n_account = true;
2396
2397 sk->sk_data_ready(sk);
2398 kfree_skb(copy_skb);
2399 goto drop_n_restore;
2400}
2401
2402static void tpacket_destruct_skb(struct sk_buff *skb)
2403{
2404 struct packet_sock *po = pkt_sk(skb->sk);
2405
2406 if (likely(po->tx_ring.pg_vec)) {
2407 void *ph;
2408 __u32 ts;
2409
2410 ph = skb_zcopy_get_nouarg(skb);
2411 packet_dec_pending(&po->tx_ring);
2412
2413 ts = __packet_set_timestamp(po, ph, skb);
2414 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2415
2416 if (!packet_read_pending(&po->tx_ring))
2417 complete(&po->skb_completion);
2418 }
2419
2420 sock_wfree(skb);
2421}
2422
2423static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2424{
2425 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2426 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2427 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2428 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2429 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2430 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2431 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2432
2433 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2434 return -EINVAL;
2435
2436 return 0;
2437}
2438
2439static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2440 struct virtio_net_hdr *vnet_hdr)
2441{
2442 if (*len < sizeof(*vnet_hdr))
2443 return -EINVAL;
2444 *len -= sizeof(*vnet_hdr);
2445
2446 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2447 return -EFAULT;
2448
2449 return __packet_snd_vnet_parse(vnet_hdr, *len);
2450}
2451
2452static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2453 void *frame, struct net_device *dev, void *data, int tp_len,
2454 __be16 proto, unsigned char *addr, int hlen, int copylen,
2455 const struct sockcm_cookie *sockc)
2456{
2457 union tpacket_uhdr ph;
2458 int to_write, offset, len, nr_frags, len_max;
2459 struct socket *sock = po->sk.sk_socket;
2460 struct page *page;
2461 int err;
2462
2463 ph.raw = frame;
2464
2465 skb->protocol = proto;
2466 skb->dev = dev;
2467 skb->priority = po->sk.sk_priority;
2468 skb->mark = po->sk.sk_mark;
2469 skb->tstamp = sockc->transmit_time;
2470 skb_setup_tx_timestamp(skb, sockc->tsflags);
2471 skb_zcopy_set_nouarg(skb, ph.raw);
2472
2473 skb_reserve(skb, hlen);
2474 skb_reset_network_header(skb);
2475
2476 to_write = tp_len;
2477
2478 if (sock->type == SOCK_DGRAM) {
2479 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2480 NULL, tp_len);
2481 if (unlikely(err < 0))
2482 return -EINVAL;
2483 } else if (copylen) {
2484 int hdrlen = min_t(int, copylen, tp_len);
2485
2486 skb_push(skb, dev->hard_header_len);
2487 skb_put(skb, copylen - dev->hard_header_len);
2488 err = skb_store_bits(skb, 0, data, hdrlen);
2489 if (unlikely(err))
2490 return err;
2491 if (!dev_validate_header(dev, skb->data, hdrlen))
2492 return -EINVAL;
2493
2494 data += hdrlen;
2495 to_write -= hdrlen;
2496 }
2497
2498 offset = offset_in_page(data);
2499 len_max = PAGE_SIZE - offset;
2500 len = ((to_write > len_max) ? len_max : to_write);
2501
2502 skb->data_len = to_write;
2503 skb->len += to_write;
2504 skb->truesize += to_write;
2505 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2506
2507 while (likely(to_write)) {
2508 nr_frags = skb_shinfo(skb)->nr_frags;
2509
2510 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2511 pr_err("Packet exceed the number of skb frags(%lu)\n",
2512 MAX_SKB_FRAGS);
2513 return -EFAULT;
2514 }
2515
2516 page = pgv_to_page(data);
2517 data += len;
2518 flush_dcache_page(page);
2519 get_page(page);
2520 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2521 to_write -= len;
2522 offset = 0;
2523 len_max = PAGE_SIZE;
2524 len = ((to_write > len_max) ? len_max : to_write);
2525 }
2526
2527 packet_parse_headers(skb, sock);
2528
2529 return tp_len;
2530}
2531
2532static int tpacket_parse_header(struct packet_sock *po, void *frame,
2533 int size_max, void **data)
2534{
2535 union tpacket_uhdr ph;
2536 int tp_len, off;
2537
2538 ph.raw = frame;
2539
2540 switch (po->tp_version) {
2541 case TPACKET_V3:
2542 if (ph.h3->tp_next_offset != 0) {
2543 pr_warn_once("variable sized slot not supported");
2544 return -EINVAL;
2545 }
2546 tp_len = ph.h3->tp_len;
2547 break;
2548 case TPACKET_V2:
2549 tp_len = ph.h2->tp_len;
2550 break;
2551 default:
2552 tp_len = ph.h1->tp_len;
2553 break;
2554 }
2555 if (unlikely(tp_len > size_max)) {
2556 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2557 return -EMSGSIZE;
2558 }
2559
2560 if (unlikely(po->tp_tx_has_off)) {
2561 int off_min, off_max;
2562
2563 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2564 off_max = po->tx_ring.frame_size - tp_len;
2565 if (po->sk.sk_type == SOCK_DGRAM) {
2566 switch (po->tp_version) {
2567 case TPACKET_V3:
2568 off = ph.h3->tp_net;
2569 break;
2570 case TPACKET_V2:
2571 off = ph.h2->tp_net;
2572 break;
2573 default:
2574 off = ph.h1->tp_net;
2575 break;
2576 }
2577 } else {
2578 switch (po->tp_version) {
2579 case TPACKET_V3:
2580 off = ph.h3->tp_mac;
2581 break;
2582 case TPACKET_V2:
2583 off = ph.h2->tp_mac;
2584 break;
2585 default:
2586 off = ph.h1->tp_mac;
2587 break;
2588 }
2589 }
2590 if (unlikely((off < off_min) || (off_max < off)))
2591 return -EINVAL;
2592 } else {
2593 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2594 }
2595
2596 *data = frame + off;
2597 return tp_len;
2598}
2599
2600static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2601{
2602 struct sk_buff *skb = NULL;
2603 struct net_device *dev;
2604 struct virtio_net_hdr *vnet_hdr = NULL;
2605 struct sockcm_cookie sockc;
2606 __be16 proto;
2607 int err, reserve = 0;
2608 void *ph;
2609 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2610 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2611 unsigned char *addr = NULL;
2612 int tp_len, size_max;
2613 void *data;
2614 int len_sum = 0;
2615 int status = TP_STATUS_AVAILABLE;
2616 int hlen, tlen, copylen = 0;
2617 long timeo = 0;
2618
2619 mutex_lock(&po->pg_vec_lock);
2620
2621 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2622 * we need to confirm it under protection of pg_vec_lock.
2623 */
2624 if (unlikely(!po->tx_ring.pg_vec)) {
2625 err = -EBUSY;
2626 goto out;
2627 }
2628 if (likely(saddr == NULL)) {
2629 dev = packet_cached_dev_get(po);
2630 proto = po->num;
2631 } else {
2632 err = -EINVAL;
2633 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2634 goto out;
2635 if (msg->msg_namelen < (saddr->sll_halen
2636 + offsetof(struct sockaddr_ll,
2637 sll_addr)))
2638 goto out;
2639 proto = saddr->sll_protocol;
2640 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2641 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2642 if (dev && msg->msg_namelen < dev->addr_len +
2643 offsetof(struct sockaddr_ll, sll_addr))
2644 goto out_put;
2645 addr = saddr->sll_addr;
2646 }
2647 }
2648
2649 err = -ENXIO;
2650 if (unlikely(dev == NULL))
2651 goto out;
2652 err = -ENETDOWN;
2653 if (unlikely(!(dev->flags & IFF_UP)))
2654 goto out_put;
2655
2656 sockcm_init(&sockc, &po->sk);
2657 if (msg->msg_controllen) {
2658 err = sock_cmsg_send(&po->sk, msg, &sockc);
2659 if (unlikely(err))
2660 goto out_put;
2661 }
2662
2663 if (po->sk.sk_socket->type == SOCK_RAW)
2664 reserve = dev->hard_header_len;
2665 size_max = po->tx_ring.frame_size
2666 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2667
2668 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2669 size_max = dev->mtu + reserve + VLAN_HLEN;
2670
2671 reinit_completion(&po->skb_completion);
2672
2673 do {
2674 ph = packet_current_frame(po, &po->tx_ring,
2675 TP_STATUS_SEND_REQUEST);
2676 if (unlikely(ph == NULL)) {
2677 if (need_wait && skb) {
2678 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2679 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2680 if (timeo <= 0) {
2681 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2682 goto out_put;
2683 }
2684 }
2685 /* check for additional frames */
2686 continue;
2687 }
2688
2689 skb = NULL;
2690 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2691 if (tp_len < 0)
2692 goto tpacket_error;
2693
2694 status = TP_STATUS_SEND_REQUEST;
2695 hlen = LL_RESERVED_SPACE(dev);
2696 tlen = dev->needed_tailroom;
2697 if (po->has_vnet_hdr) {
2698 vnet_hdr = data;
2699 data += sizeof(*vnet_hdr);
2700 tp_len -= sizeof(*vnet_hdr);
2701 if (tp_len < 0 ||
2702 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2703 tp_len = -EINVAL;
2704 goto tpacket_error;
2705 }
2706 copylen = __virtio16_to_cpu(vio_le(),
2707 vnet_hdr->hdr_len);
2708 }
2709 copylen = max_t(int, copylen, dev->hard_header_len);
2710 skb = sock_alloc_send_skb(&po->sk,
2711 hlen + tlen + sizeof(struct sockaddr_ll) +
2712 (copylen - dev->hard_header_len),
2713 !need_wait, &err);
2714
2715 if (unlikely(skb == NULL)) {
2716 /* we assume the socket was initially writeable ... */
2717 if (likely(len_sum > 0))
2718 err = len_sum;
2719 goto out_status;
2720 }
2721 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2722 addr, hlen, copylen, &sockc);
2723 if (likely(tp_len >= 0) &&
2724 tp_len > dev->mtu + reserve &&
2725 !po->has_vnet_hdr &&
2726 !packet_extra_vlan_len_allowed(dev, skb))
2727 tp_len = -EMSGSIZE;
2728
2729 if (unlikely(tp_len < 0)) {
2730tpacket_error:
2731 if (po->tp_loss) {
2732 __packet_set_status(po, ph,
2733 TP_STATUS_AVAILABLE);
2734 packet_increment_head(&po->tx_ring);
2735 kfree_skb(skb);
2736 continue;
2737 } else {
2738 status = TP_STATUS_WRONG_FORMAT;
2739 err = tp_len;
2740 goto out_status;
2741 }
2742 }
2743
2744 if (po->has_vnet_hdr) {
2745 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2746 tp_len = -EINVAL;
2747 goto tpacket_error;
2748 }
2749 virtio_net_hdr_set_proto(skb, vnet_hdr);
2750 }
2751
2752 skb->destructor = tpacket_destruct_skb;
2753 __packet_set_status(po, ph, TP_STATUS_SENDING);
2754 packet_inc_pending(&po->tx_ring);
2755
2756 status = TP_STATUS_SEND_REQUEST;
2757 err = po->xmit(skb);
2758 if (unlikely(err > 0)) {
2759 err = net_xmit_errno(err);
2760 if (err && __packet_get_status(po, ph) ==
2761 TP_STATUS_AVAILABLE) {
2762 /* skb was destructed already */
2763 skb = NULL;
2764 goto out_status;
2765 }
2766 /*
2767 * skb was dropped but not destructed yet;
2768 * let's treat it like congestion or err < 0
2769 */
2770 err = 0;
2771 }
2772 packet_increment_head(&po->tx_ring);
2773 len_sum += tp_len;
2774 } while (likely((ph != NULL) ||
2775 /* Note: packet_read_pending() might be slow if we have
2776 * to call it as it's per_cpu variable, but in fast-path
2777 * we already short-circuit the loop with the first
2778 * condition, and luckily don't have to go that path
2779 * anyway.
2780 */
2781 (need_wait && packet_read_pending(&po->tx_ring))));
2782
2783 err = len_sum;
2784 goto out_put;
2785
2786out_status:
2787 __packet_set_status(po, ph, status);
2788 kfree_skb(skb);
2789out_put:
2790 dev_put(dev);
2791out:
2792 mutex_unlock(&po->pg_vec_lock);
2793 return err;
2794}
2795
2796static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2797 size_t reserve, size_t len,
2798 size_t linear, int noblock,
2799 int *err)
2800{
2801 struct sk_buff *skb;
2802
2803 /* Under a page? Don't bother with paged skb. */
2804 if (prepad + len < PAGE_SIZE || !linear)
2805 linear = len;
2806
2807 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2808 err, 0);
2809 if (!skb)
2810 return NULL;
2811
2812 skb_reserve(skb, reserve);
2813 skb_put(skb, linear);
2814 skb->data_len = len - linear;
2815 skb->len += len - linear;
2816
2817 return skb;
2818}
2819
2820static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2821{
2822 struct sock *sk = sock->sk;
2823 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2824 struct sk_buff *skb;
2825 struct net_device *dev;
2826 __be16 proto;
2827 unsigned char *addr = NULL;
2828 int err, reserve = 0;
2829 struct sockcm_cookie sockc;
2830 struct virtio_net_hdr vnet_hdr = { 0 };
2831 int offset = 0;
2832 struct packet_sock *po = pkt_sk(sk);
2833 bool has_vnet_hdr = false;
2834 int hlen, tlen, linear;
2835 int extra_len = 0;
2836
2837 /*
2838 * Get and verify the address.
2839 */
2840
2841 if (likely(saddr == NULL)) {
2842 dev = packet_cached_dev_get(po);
2843 proto = po->num;
2844 } else {
2845 err = -EINVAL;
2846 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2847 goto out;
2848 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2849 goto out;
2850 proto = saddr->sll_protocol;
2851 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2852 if (sock->type == SOCK_DGRAM) {
2853 if (dev && msg->msg_namelen < dev->addr_len +
2854 offsetof(struct sockaddr_ll, sll_addr))
2855 goto out_unlock;
2856 addr = saddr->sll_addr;
2857 }
2858 }
2859
2860 err = -ENXIO;
2861 if (unlikely(dev == NULL))
2862 goto out_unlock;
2863 err = -ENETDOWN;
2864 if (unlikely(!(dev->flags & IFF_UP)))
2865 goto out_unlock;
2866
2867 sockcm_init(&sockc, sk);
2868 sockc.mark = sk->sk_mark;
2869 if (msg->msg_controllen) {
2870 err = sock_cmsg_send(sk, msg, &sockc);
2871 if (unlikely(err))
2872 goto out_unlock;
2873 }
2874
2875 if (sock->type == SOCK_RAW)
2876 reserve = dev->hard_header_len;
2877 if (po->has_vnet_hdr) {
2878 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2879 if (err)
2880 goto out_unlock;
2881 has_vnet_hdr = true;
2882 }
2883
2884 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2885 if (!netif_supports_nofcs(dev)) {
2886 err = -EPROTONOSUPPORT;
2887 goto out_unlock;
2888 }
2889 extra_len = 4; /* We're doing our own CRC */
2890 }
2891
2892 err = -EMSGSIZE;
2893 if (!vnet_hdr.gso_type &&
2894 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2895 goto out_unlock;
2896
2897 err = -ENOBUFS;
2898 hlen = LL_RESERVED_SPACE(dev);
2899 tlen = dev->needed_tailroom;
2900 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2901 linear = max(linear, min_t(int, len, dev->hard_header_len));
2902 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2903 msg->msg_flags & MSG_DONTWAIT, &err);
2904 if (skb == NULL)
2905 goto out_unlock;
2906
2907 skb_reset_network_header(skb);
2908
2909 err = -EINVAL;
2910 if (sock->type == SOCK_DGRAM) {
2911 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2912 if (unlikely(offset < 0))
2913 goto out_free;
2914 } else if (reserve) {
2915 skb_reserve(skb, -reserve);
2916 if (len < reserve + sizeof(struct ipv6hdr) &&
2917 dev->min_header_len != dev->hard_header_len)
2918 skb_reset_network_header(skb);
2919 }
2920
2921 /* Returns -EFAULT on error */
2922 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2923 if (err)
2924 goto out_free;
2925
2926 if (sock->type == SOCK_RAW &&
2927 !dev_validate_header(dev, skb->data, len)) {
2928 err = -EINVAL;
2929 goto out_free;
2930 }
2931
2932 skb_setup_tx_timestamp(skb, sockc.tsflags);
2933
2934 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2935 !packet_extra_vlan_len_allowed(dev, skb)) {
2936 err = -EMSGSIZE;
2937 goto out_free;
2938 }
2939
2940 skb->protocol = proto;
2941 skb->dev = dev;
2942 skb->priority = sk->sk_priority;
2943 skb->mark = sockc.mark;
2944 skb->tstamp = sockc.transmit_time;
2945
2946 if (has_vnet_hdr) {
2947 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2948 if (err)
2949 goto out_free;
2950 len += sizeof(vnet_hdr);
2951 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2952 }
2953
2954 packet_parse_headers(skb, sock);
2955
2956 if (unlikely(extra_len == 4))
2957 skb->no_fcs = 1;
2958
2959 err = po->xmit(skb);
2960 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2961 goto out_unlock;
2962
2963 dev_put(dev);
2964
2965 return len;
2966
2967out_free:
2968 kfree_skb(skb);
2969out_unlock:
2970 if (dev)
2971 dev_put(dev);
2972out:
2973 return err;
2974}
2975
2976static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2977{
2978 struct sock *sk = sock->sk;
2979 struct packet_sock *po = pkt_sk(sk);
2980
2981 if (po->tx_ring.pg_vec)
2982 return tpacket_snd(po, msg);
2983 else
2984 return packet_snd(sock, msg, len);
2985}
2986
2987/*
2988 * Close a PACKET socket. This is fairly simple. We immediately go
2989 * to 'closed' state and remove our protocol entry in the device list.
2990 */
2991
2992static int packet_release(struct socket *sock)
2993{
2994 struct sock *sk = sock->sk;
2995 struct packet_sock *po;
2996 struct packet_fanout *f;
2997 struct net *net;
2998 union tpacket_req_u req_u;
2999
3000 if (!sk)
3001 return 0;
3002
3003 net = sock_net(sk);
3004 po = pkt_sk(sk);
3005
3006 mutex_lock(&net->packet.sklist_lock);
3007 sk_del_node_init_rcu(sk);
3008 mutex_unlock(&net->packet.sklist_lock);
3009
3010 preempt_disable();
3011 sock_prot_inuse_add(net, sk->sk_prot, -1);
3012 preempt_enable();
3013
3014 spin_lock(&po->bind_lock);
3015 unregister_prot_hook(sk, false);
3016 packet_cached_dev_reset(po);
3017
3018 if (po->prot_hook.dev) {
3019 dev_put(po->prot_hook.dev);
3020 po->prot_hook.dev = NULL;
3021 }
3022 spin_unlock(&po->bind_lock);
3023
3024 packet_flush_mclist(sk);
3025
3026 lock_sock(sk);
3027 if (po->rx_ring.pg_vec) {
3028 memset(&req_u, 0, sizeof(req_u));
3029 packet_set_ring(sk, &req_u, 1, 0);
3030 }
3031
3032 if (po->tx_ring.pg_vec) {
3033 memset(&req_u, 0, sizeof(req_u));
3034 packet_set_ring(sk, &req_u, 1, 1);
3035 }
3036 release_sock(sk);
3037
3038 f = fanout_release(sk);
3039
3040 synchronize_net();
3041
3042 kfree(po->rollover);
3043 if (f) {
3044 fanout_release_data(f);
3045 kfree(f);
3046 }
3047 /*
3048 * Now the socket is dead. No more input will appear.
3049 */
3050 sock_orphan(sk);
3051 sock->sk = NULL;
3052
3053 /* Purge queues */
3054
3055 skb_queue_purge(&sk->sk_receive_queue);
3056 packet_free_pending(po);
3057 sk_refcnt_debug_release(sk);
3058
3059 sock_put(sk);
3060 return 0;
3061}
3062
3063/*
3064 * Attach a packet hook.
3065 */
3066
3067static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3068 __be16 proto)
3069{
3070 struct packet_sock *po = pkt_sk(sk);
3071 struct net_device *dev_curr;
3072 __be16 proto_curr;
3073 bool need_rehook;
3074 struct net_device *dev = NULL;
3075 int ret = 0;
3076 bool unlisted = false;
3077
3078 lock_sock(sk);
3079 spin_lock(&po->bind_lock);
3080 rcu_read_lock();
3081
3082 if (po->fanout) {
3083 ret = -EINVAL;
3084 goto out_unlock;
3085 }
3086
3087 if (name) {
3088 dev = dev_get_by_name_rcu(sock_net(sk), name);
3089 if (!dev) {
3090 ret = -ENODEV;
3091 goto out_unlock;
3092 }
3093 } else if (ifindex) {
3094 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3095 if (!dev) {
3096 ret = -ENODEV;
3097 goto out_unlock;
3098 }
3099 }
3100
3101 if (dev)
3102 dev_hold(dev);
3103
3104 proto_curr = po->prot_hook.type;
3105 dev_curr = po->prot_hook.dev;
3106
3107 need_rehook = proto_curr != proto || dev_curr != dev;
3108
3109 if (need_rehook) {
3110 if (po->running) {
3111 rcu_read_unlock();
3112 /* prevents packet_notifier() from calling
3113 * register_prot_hook()
3114 */
3115 po->num = 0;
3116 __unregister_prot_hook(sk, true);
3117 rcu_read_lock();
3118 dev_curr = po->prot_hook.dev;
3119 if (dev)
3120 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3121 dev->ifindex);
3122 }
3123
3124 BUG_ON(po->running);
3125 po->num = proto;
3126 po->prot_hook.type = proto;
3127
3128 if (unlikely(unlisted)) {
3129 dev_put(dev);
3130 po->prot_hook.dev = NULL;
3131 po->ifindex = -1;
3132 packet_cached_dev_reset(po);
3133 } else {
3134 po->prot_hook.dev = dev;
3135 po->ifindex = dev ? dev->ifindex : 0;
3136 packet_cached_dev_assign(po, dev);
3137 }
3138 }
3139 if (dev_curr)
3140 dev_put(dev_curr);
3141
3142 if (proto == 0 || !need_rehook)
3143 goto out_unlock;
3144
3145 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3146 register_prot_hook(sk);
3147 } else {
3148 sk->sk_err = ENETDOWN;
3149 if (!sock_flag(sk, SOCK_DEAD))
3150 sk->sk_error_report(sk);
3151 }
3152
3153out_unlock:
3154 rcu_read_unlock();
3155 spin_unlock(&po->bind_lock);
3156 release_sock(sk);
3157 return ret;
3158}
3159
3160/*
3161 * Bind a packet socket to a device
3162 */
3163
3164static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3165 int addr_len)
3166{
3167 struct sock *sk = sock->sk;
3168 char name[sizeof(uaddr->sa_data) + 1];
3169
3170 /*
3171 * Check legality
3172 */
3173
3174 if (addr_len != sizeof(struct sockaddr))
3175 return -EINVAL;
3176 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3177 * zero-terminated.
3178 */
3179 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3180 name[sizeof(uaddr->sa_data)] = 0;
3181
3182 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3183}
3184
3185static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3186{
3187 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3188 struct sock *sk = sock->sk;
3189
3190 /*
3191 * Check legality
3192 */
3193
3194 if (addr_len < sizeof(struct sockaddr_ll))
3195 return -EINVAL;
3196 if (sll->sll_family != AF_PACKET)
3197 return -EINVAL;
3198
3199 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3200 sll->sll_protocol ? : pkt_sk(sk)->num);
3201}
3202
3203static struct proto packet_proto = {
3204 .name = "PACKET",
3205 .owner = THIS_MODULE,
3206 .obj_size = sizeof(struct packet_sock),
3207};
3208
3209/*
3210 * Create a packet of type SOCK_PACKET.
3211 */
3212
3213static int packet_create(struct net *net, struct socket *sock, int protocol,
3214 int kern)
3215{
3216 struct sock *sk;
3217 struct packet_sock *po;
3218 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3219 int err;
3220
3221 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3222 return -EPERM;
3223 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3224 sock->type != SOCK_PACKET)
3225 return -ESOCKTNOSUPPORT;
3226
3227 sock->state = SS_UNCONNECTED;
3228
3229 err = -ENOBUFS;
3230 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3231 if (sk == NULL)
3232 goto out;
3233
3234 sock->ops = &packet_ops;
3235 if (sock->type == SOCK_PACKET)
3236 sock->ops = &packet_ops_spkt;
3237
3238 sock_init_data(sock, sk);
3239
3240 po = pkt_sk(sk);
3241 init_completion(&po->skb_completion);
3242 sk->sk_family = PF_PACKET;
3243 po->num = proto;
3244 po->xmit = dev_queue_xmit;
3245
3246 err = packet_alloc_pending(po);
3247 if (err)
3248 goto out2;
3249
3250 packet_cached_dev_reset(po);
3251
3252 sk->sk_destruct = packet_sock_destruct;
3253 sk_refcnt_debug_inc(sk);
3254
3255 /*
3256 * Attach a protocol block
3257 */
3258
3259 spin_lock_init(&po->bind_lock);
3260 mutex_init(&po->pg_vec_lock);
3261 po->rollover = NULL;
3262 po->prot_hook.func = packet_rcv;
3263
3264 if (sock->type == SOCK_PACKET)
3265 po->prot_hook.func = packet_rcv_spkt;
3266
3267 po->prot_hook.af_packet_priv = sk;
3268
3269 if (proto) {
3270 po->prot_hook.type = proto;
3271 __register_prot_hook(sk);
3272 }
3273
3274 mutex_lock(&net->packet.sklist_lock);
3275 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3276 mutex_unlock(&net->packet.sklist_lock);
3277
3278 preempt_disable();
3279 sock_prot_inuse_add(net, &packet_proto, 1);
3280 preempt_enable();
3281
3282 return 0;
3283out2:
3284 sk_free(sk);
3285out:
3286 return err;
3287}
3288
3289/*
3290 * Pull a packet from our receive queue and hand it to the user.
3291 * If necessary we block.
3292 */
3293
3294static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3295 int flags)
3296{
3297 struct sock *sk = sock->sk;
3298 struct sk_buff *skb;
3299 int copied, err;
3300 int vnet_hdr_len = 0;
3301 unsigned int origlen = 0;
3302
3303 err = -EINVAL;
3304 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3305 goto out;
3306
3307#if 0
3308 /* What error should we return now? EUNATTACH? */
3309 if (pkt_sk(sk)->ifindex < 0)
3310 return -ENODEV;
3311#endif
3312
3313 if (flags & MSG_ERRQUEUE) {
3314 err = sock_recv_errqueue(sk, msg, len,
3315 SOL_PACKET, PACKET_TX_TIMESTAMP);
3316 goto out;
3317 }
3318
3319 /*
3320 * Call the generic datagram receiver. This handles all sorts
3321 * of horrible races and re-entrancy so we can forget about it
3322 * in the protocol layers.
3323 *
3324 * Now it will return ENETDOWN, if device have just gone down,
3325 * but then it will block.
3326 */
3327
3328 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3329
3330 /*
3331 * An error occurred so return it. Because skb_recv_datagram()
3332 * handles the blocking we don't see and worry about blocking
3333 * retries.
3334 */
3335
3336 if (skb == NULL)
3337 goto out;
3338
3339 packet_rcv_try_clear_pressure(pkt_sk(sk));
3340
3341 if (pkt_sk(sk)->has_vnet_hdr) {
3342 err = packet_rcv_vnet(msg, skb, &len);
3343 if (err)
3344 goto out_free;
3345 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3346 }
3347
3348 /* You lose any data beyond the buffer you gave. If it worries
3349 * a user program they can ask the device for its MTU
3350 * anyway.
3351 */
3352 copied = skb->len;
3353 if (copied > len) {
3354 copied = len;
3355 msg->msg_flags |= MSG_TRUNC;
3356 }
3357
3358 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3359 if (err)
3360 goto out_free;
3361
3362 if (sock->type != SOCK_PACKET) {
3363 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3364
3365 /* Original length was stored in sockaddr_ll fields */
3366 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3367 sll->sll_family = AF_PACKET;
3368 sll->sll_protocol = skb->protocol;
3369 }
3370
3371 sock_recv_ts_and_drops(msg, sk, skb);
3372
3373 if (msg->msg_name) {
3374 int copy_len;
3375
3376 /* If the address length field is there to be filled
3377 * in, we fill it in now.
3378 */
3379 if (sock->type == SOCK_PACKET) {
3380 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3381 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3382 copy_len = msg->msg_namelen;
3383 } else {
3384 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3385
3386 msg->msg_namelen = sll->sll_halen +
3387 offsetof(struct sockaddr_ll, sll_addr);
3388 copy_len = msg->msg_namelen;
3389 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3390 memset(msg->msg_name +
3391 offsetof(struct sockaddr_ll, sll_addr),
3392 0, sizeof(sll->sll_addr));
3393 msg->msg_namelen = sizeof(struct sockaddr_ll);
3394 }
3395 }
3396 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3397 }
3398
3399 if (pkt_sk(sk)->auxdata) {
3400 struct tpacket_auxdata aux;
3401
3402 aux.tp_status = TP_STATUS_USER;
3403 if (skb->ip_summed == CHECKSUM_PARTIAL)
3404 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3405 else if (skb->pkt_type != PACKET_OUTGOING &&
3406 (skb->ip_summed == CHECKSUM_COMPLETE ||
3407 skb_csum_unnecessary(skb)))
3408 aux.tp_status |= TP_STATUS_CSUM_VALID;
3409
3410 aux.tp_len = origlen;
3411 aux.tp_snaplen = skb->len;
3412 aux.tp_mac = 0;
3413 aux.tp_net = skb_network_offset(skb);
3414 if (skb_vlan_tag_present(skb)) {
3415 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3416 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3417 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3418 } else {
3419 aux.tp_vlan_tci = 0;
3420 aux.tp_vlan_tpid = 0;
3421 }
3422 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3423 }
3424
3425 /*
3426 * Free or return the buffer as appropriate. Again this
3427 * hides all the races and re-entrancy issues from us.
3428 */
3429 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3430
3431out_free:
3432 skb_free_datagram(sk, skb);
3433out:
3434 return err;
3435}
3436
3437static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3438 int peer)
3439{
3440 struct net_device *dev;
3441 struct sock *sk = sock->sk;
3442
3443 if (peer)
3444 return -EOPNOTSUPP;
3445
3446 uaddr->sa_family = AF_PACKET;
3447 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3448 rcu_read_lock();
3449 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3450 if (dev)
3451 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3452 rcu_read_unlock();
3453
3454 return sizeof(*uaddr);
3455}
3456
3457static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3458 int peer)
3459{
3460 struct net_device *dev;
3461 struct sock *sk = sock->sk;
3462 struct packet_sock *po = pkt_sk(sk);
3463 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3464
3465 if (peer)
3466 return -EOPNOTSUPP;
3467
3468 sll->sll_family = AF_PACKET;
3469 sll->sll_ifindex = po->ifindex;
3470 sll->sll_protocol = po->num;
3471 sll->sll_pkttype = 0;
3472 rcu_read_lock();
3473 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3474 if (dev) {
3475 sll->sll_hatype = dev->type;
3476 sll->sll_halen = dev->addr_len;
3477 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3478 } else {
3479 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3480 sll->sll_halen = 0;
3481 }
3482 rcu_read_unlock();
3483
3484 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3485}
3486
3487static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3488 int what)
3489{
3490 switch (i->type) {
3491 case PACKET_MR_MULTICAST:
3492 if (i->alen != dev->addr_len)
3493 return -EINVAL;
3494 if (what > 0)
3495 return dev_mc_add(dev, i->addr);
3496 else
3497 return dev_mc_del(dev, i->addr);
3498 break;
3499 case PACKET_MR_PROMISC:
3500 return dev_set_promiscuity(dev, what);
3501 case PACKET_MR_ALLMULTI:
3502 return dev_set_allmulti(dev, what);
3503 case PACKET_MR_UNICAST:
3504 if (i->alen != dev->addr_len)
3505 return -EINVAL;
3506 if (what > 0)
3507 return dev_uc_add(dev, i->addr);
3508 else
3509 return dev_uc_del(dev, i->addr);
3510 break;
3511 default:
3512 break;
3513 }
3514 return 0;
3515}
3516
3517static void packet_dev_mclist_delete(struct net_device *dev,
3518 struct packet_mclist **mlp)
3519{
3520 struct packet_mclist *ml;
3521
3522 while ((ml = *mlp) != NULL) {
3523 if (ml->ifindex == dev->ifindex) {
3524 packet_dev_mc(dev, ml, -1);
3525 *mlp = ml->next;
3526 kfree(ml);
3527 } else
3528 mlp = &ml->next;
3529 }
3530}
3531
3532static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3533{
3534 struct packet_sock *po = pkt_sk(sk);
3535 struct packet_mclist *ml, *i;
3536 struct net_device *dev;
3537 int err;
3538
3539 rtnl_lock();
3540
3541 err = -ENODEV;
3542 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3543 if (!dev)
3544 goto done;
3545
3546 err = -EINVAL;
3547 if (mreq->mr_alen > dev->addr_len)
3548 goto done;
3549
3550 err = -ENOBUFS;
3551 i = kmalloc(sizeof(*i), GFP_KERNEL);
3552 if (i == NULL)
3553 goto done;
3554
3555 err = 0;
3556 for (ml = po->mclist; ml; ml = ml->next) {
3557 if (ml->ifindex == mreq->mr_ifindex &&
3558 ml->type == mreq->mr_type &&
3559 ml->alen == mreq->mr_alen &&
3560 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3561 ml->count++;
3562 /* Free the new element ... */
3563 kfree(i);
3564 goto done;
3565 }
3566 }
3567
3568 i->type = mreq->mr_type;
3569 i->ifindex = mreq->mr_ifindex;
3570 i->alen = mreq->mr_alen;
3571 memcpy(i->addr, mreq->mr_address, i->alen);
3572 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3573 i->count = 1;
3574 i->next = po->mclist;
3575 po->mclist = i;
3576 err = packet_dev_mc(dev, i, 1);
3577 if (err) {
3578 po->mclist = i->next;
3579 kfree(i);
3580 }
3581
3582done:
3583 rtnl_unlock();
3584 return err;
3585}
3586
3587static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3588{
3589 struct packet_mclist *ml, **mlp;
3590
3591 rtnl_lock();
3592
3593 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3594 if (ml->ifindex == mreq->mr_ifindex &&
3595 ml->type == mreq->mr_type &&
3596 ml->alen == mreq->mr_alen &&
3597 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3598 if (--ml->count == 0) {
3599 struct net_device *dev;
3600 *mlp = ml->next;
3601 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3602 if (dev)
3603 packet_dev_mc(dev, ml, -1);
3604 kfree(ml);
3605 }
3606 break;
3607 }
3608 }
3609 rtnl_unlock();
3610 return 0;
3611}
3612
3613static void packet_flush_mclist(struct sock *sk)
3614{
3615 struct packet_sock *po = pkt_sk(sk);
3616 struct packet_mclist *ml;
3617
3618 if (!po->mclist)
3619 return;
3620
3621 rtnl_lock();
3622 while ((ml = po->mclist) != NULL) {
3623 struct net_device *dev;
3624
3625 po->mclist = ml->next;
3626 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3627 if (dev != NULL)
3628 packet_dev_mc(dev, ml, -1);
3629 kfree(ml);
3630 }
3631 rtnl_unlock();
3632}
3633
3634static int
3635packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3636{
3637 struct sock *sk = sock->sk;
3638 struct packet_sock *po = pkt_sk(sk);
3639 int ret;
3640
3641 if (level != SOL_PACKET)
3642 return -ENOPROTOOPT;
3643
3644 switch (optname) {
3645 case PACKET_ADD_MEMBERSHIP:
3646 case PACKET_DROP_MEMBERSHIP:
3647 {
3648 struct packet_mreq_max mreq;
3649 int len = optlen;
3650 memset(&mreq, 0, sizeof(mreq));
3651 if (len < sizeof(struct packet_mreq))
3652 return -EINVAL;
3653 if (len > sizeof(mreq))
3654 len = sizeof(mreq);
3655 if (copy_from_user(&mreq, optval, len))
3656 return -EFAULT;
3657 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3658 return -EINVAL;
3659 if (optname == PACKET_ADD_MEMBERSHIP)
3660 ret = packet_mc_add(sk, &mreq);
3661 else
3662 ret = packet_mc_drop(sk, &mreq);
3663 return ret;
3664 }
3665
3666 case PACKET_RX_RING:
3667 case PACKET_TX_RING:
3668 {
3669 union tpacket_req_u req_u;
3670 int len;
3671
3672 lock_sock(sk);
3673 switch (po->tp_version) {
3674 case TPACKET_V1:
3675 case TPACKET_V2:
3676 len = sizeof(req_u.req);
3677 break;
3678 case TPACKET_V3:
3679 default:
3680 len = sizeof(req_u.req3);
3681 break;
3682 }
3683 if (optlen < len) {
3684 ret = -EINVAL;
3685 } else {
3686 if (copy_from_user(&req_u.req, optval, len))
3687 ret = -EFAULT;
3688 else
3689 ret = packet_set_ring(sk, &req_u, 0,
3690 optname == PACKET_TX_RING);
3691 }
3692 release_sock(sk);
3693 return ret;
3694 }
3695 case PACKET_COPY_THRESH:
3696 {
3697 int val;
3698
3699 if (optlen != sizeof(val))
3700 return -EINVAL;
3701 if (copy_from_user(&val, optval, sizeof(val)))
3702 return -EFAULT;
3703
3704 pkt_sk(sk)->copy_thresh = val;
3705 return 0;
3706 }
3707 case PACKET_VERSION:
3708 {
3709 int val;
3710
3711 if (optlen != sizeof(val))
3712 return -EINVAL;
3713 if (copy_from_user(&val, optval, sizeof(val)))
3714 return -EFAULT;
3715 switch (val) {
3716 case TPACKET_V1:
3717 case TPACKET_V2:
3718 case TPACKET_V3:
3719 break;
3720 default:
3721 return -EINVAL;
3722 }
3723 lock_sock(sk);
3724 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3725 ret = -EBUSY;
3726 } else {
3727 po->tp_version = val;
3728 ret = 0;
3729 }
3730 release_sock(sk);
3731 return ret;
3732 }
3733 case PACKET_RESERVE:
3734 {
3735 unsigned int val;
3736
3737 if (optlen != sizeof(val))
3738 return -EINVAL;
3739 if (copy_from_user(&val, optval, sizeof(val)))
3740 return -EFAULT;
3741 if (val > INT_MAX)
3742 return -EINVAL;
3743 lock_sock(sk);
3744 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3745 ret = -EBUSY;
3746 } else {
3747 po->tp_reserve = val;
3748 ret = 0;
3749 }
3750 release_sock(sk);
3751 return ret;
3752 }
3753 case PACKET_LOSS:
3754 {
3755 unsigned int val;
3756
3757 if (optlen != sizeof(val))
3758 return -EINVAL;
3759 if (copy_from_user(&val, optval, sizeof(val)))
3760 return -EFAULT;
3761
3762 lock_sock(sk);
3763 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3764 ret = -EBUSY;
3765 } else {
3766 po->tp_loss = !!val;
3767 ret = 0;
3768 }
3769 release_sock(sk);
3770 return ret;
3771 }
3772 case PACKET_AUXDATA:
3773 {
3774 int val;
3775
3776 if (optlen < sizeof(val))
3777 return -EINVAL;
3778 if (copy_from_user(&val, optval, sizeof(val)))
3779 return -EFAULT;
3780
3781 lock_sock(sk);
3782 po->auxdata = !!val;
3783 release_sock(sk);
3784 return 0;
3785 }
3786 case PACKET_ORIGDEV:
3787 {
3788 int val;
3789
3790 if (optlen < sizeof(val))
3791 return -EINVAL;
3792 if (copy_from_user(&val, optval, sizeof(val)))
3793 return -EFAULT;
3794
3795 lock_sock(sk);
3796 po->origdev = !!val;
3797 release_sock(sk);
3798 return 0;
3799 }
3800 case PACKET_VNET_HDR:
3801 {
3802 int val;
3803
3804 if (sock->type != SOCK_RAW)
3805 return -EINVAL;
3806 if (optlen < sizeof(val))
3807 return -EINVAL;
3808 if (copy_from_user(&val, optval, sizeof(val)))
3809 return -EFAULT;
3810
3811 lock_sock(sk);
3812 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3813 ret = -EBUSY;
3814 } else {
3815 po->has_vnet_hdr = !!val;
3816 ret = 0;
3817 }
3818 release_sock(sk);
3819 return ret;
3820 }
3821 case PACKET_TIMESTAMP:
3822 {
3823 int val;
3824
3825 if (optlen != sizeof(val))
3826 return -EINVAL;
3827 if (copy_from_user(&val, optval, sizeof(val)))
3828 return -EFAULT;
3829
3830 po->tp_tstamp = val;
3831 return 0;
3832 }
3833 case PACKET_FANOUT:
3834 {
3835 int val;
3836
3837 if (optlen != sizeof(val))
3838 return -EINVAL;
3839 if (copy_from_user(&val, optval, sizeof(val)))
3840 return -EFAULT;
3841
3842 return fanout_add(sk, val & 0xffff, val >> 16);
3843 }
3844 case PACKET_FANOUT_DATA:
3845 {
3846 if (!po->fanout)
3847 return -EINVAL;
3848
3849 return fanout_set_data(po, optval, optlen);
3850 }
3851 case PACKET_IGNORE_OUTGOING:
3852 {
3853 int val;
3854
3855 if (optlen != sizeof(val))
3856 return -EINVAL;
3857 if (copy_from_user(&val, optval, sizeof(val)))
3858 return -EFAULT;
3859 if (val < 0 || val > 1)
3860 return -EINVAL;
3861
3862 po->prot_hook.ignore_outgoing = !!val;
3863 return 0;
3864 }
3865 case PACKET_TX_HAS_OFF:
3866 {
3867 unsigned int val;
3868
3869 if (optlen != sizeof(val))
3870 return -EINVAL;
3871 if (copy_from_user(&val, optval, sizeof(val)))
3872 return -EFAULT;
3873
3874 lock_sock(sk);
3875 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3876 ret = -EBUSY;
3877 } else {
3878 po->tp_tx_has_off = !!val;
3879 ret = 0;
3880 }
3881 release_sock(sk);
3882 return 0;
3883 }
3884 case PACKET_QDISC_BYPASS:
3885 {
3886 int val;
3887
3888 if (optlen != sizeof(val))
3889 return -EINVAL;
3890 if (copy_from_user(&val, optval, sizeof(val)))
3891 return -EFAULT;
3892
3893 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3894 return 0;
3895 }
3896 default:
3897 return -ENOPROTOOPT;
3898 }
3899}
3900
3901static int packet_getsockopt(struct socket *sock, int level, int optname,
3902 char __user *optval, int __user *optlen)
3903{
3904 int len;
3905 int val, lv = sizeof(val);
3906 struct sock *sk = sock->sk;
3907 struct packet_sock *po = pkt_sk(sk);
3908 void *data = &val;
3909 union tpacket_stats_u st;
3910 struct tpacket_rollover_stats rstats;
3911 int drops;
3912
3913 if (level != SOL_PACKET)
3914 return -ENOPROTOOPT;
3915
3916 if (get_user(len, optlen))
3917 return -EFAULT;
3918
3919 if (len < 0)
3920 return -EINVAL;
3921
3922 switch (optname) {
3923 case PACKET_STATISTICS:
3924 spin_lock_bh(&sk->sk_receive_queue.lock);
3925 memcpy(&st, &po->stats, sizeof(st));
3926 memset(&po->stats, 0, sizeof(po->stats));
3927 spin_unlock_bh(&sk->sk_receive_queue.lock);
3928 drops = atomic_xchg(&po->tp_drops, 0);
3929
3930 if (po->tp_version == TPACKET_V3) {
3931 lv = sizeof(struct tpacket_stats_v3);
3932 st.stats3.tp_drops = drops;
3933 st.stats3.tp_packets += drops;
3934 data = &st.stats3;
3935 } else {
3936 lv = sizeof(struct tpacket_stats);
3937 st.stats1.tp_drops = drops;
3938 st.stats1.tp_packets += drops;
3939 data = &st.stats1;
3940 }
3941
3942 break;
3943 case PACKET_AUXDATA:
3944 val = po->auxdata;
3945 break;
3946 case PACKET_ORIGDEV:
3947 val = po->origdev;
3948 break;
3949 case PACKET_VNET_HDR:
3950 val = po->has_vnet_hdr;
3951 break;
3952 case PACKET_VERSION:
3953 val = po->tp_version;
3954 break;
3955 case PACKET_HDRLEN:
3956 if (len > sizeof(int))
3957 len = sizeof(int);
3958 if (len < sizeof(int))
3959 return -EINVAL;
3960 if (copy_from_user(&val, optval, len))
3961 return -EFAULT;
3962 switch (val) {
3963 case TPACKET_V1:
3964 val = sizeof(struct tpacket_hdr);
3965 break;
3966 case TPACKET_V2:
3967 val = sizeof(struct tpacket2_hdr);
3968 break;
3969 case TPACKET_V3:
3970 val = sizeof(struct tpacket3_hdr);
3971 break;
3972 default:
3973 return -EINVAL;
3974 }
3975 break;
3976 case PACKET_RESERVE:
3977 val = po->tp_reserve;
3978 break;
3979 case PACKET_LOSS:
3980 val = po->tp_loss;
3981 break;
3982 case PACKET_TIMESTAMP:
3983 val = po->tp_tstamp;
3984 break;
3985 case PACKET_FANOUT:
3986 val = (po->fanout ?
3987 ((u32)po->fanout->id |
3988 ((u32)po->fanout->type << 16) |
3989 ((u32)po->fanout->flags << 24)) :
3990 0);
3991 break;
3992 case PACKET_IGNORE_OUTGOING:
3993 val = po->prot_hook.ignore_outgoing;
3994 break;
3995 case PACKET_ROLLOVER_STATS:
3996 if (!po->rollover)
3997 return -EINVAL;
3998 rstats.tp_all = atomic_long_read(&po->rollover->num);
3999 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4000 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4001 data = &rstats;
4002 lv = sizeof(rstats);
4003 break;
4004 case PACKET_TX_HAS_OFF:
4005 val = po->tp_tx_has_off;
4006 break;
4007 case PACKET_QDISC_BYPASS:
4008 val = packet_use_direct_xmit(po);
4009 break;
4010 default:
4011 return -ENOPROTOOPT;
4012 }
4013
4014 if (len > lv)
4015 len = lv;
4016 if (put_user(len, optlen))
4017 return -EFAULT;
4018 if (copy_to_user(optval, data, len))
4019 return -EFAULT;
4020 return 0;
4021}
4022
4023
4024#ifdef CONFIG_COMPAT
4025static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
4026 char __user *optval, unsigned int optlen)
4027{
4028 struct packet_sock *po = pkt_sk(sock->sk);
4029
4030 if (level != SOL_PACKET)
4031 return -ENOPROTOOPT;
4032
4033 if (optname == PACKET_FANOUT_DATA &&
4034 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4035 optval = (char __user *)get_compat_bpf_fprog(optval);
4036 if (!optval)
4037 return -EFAULT;
4038 optlen = sizeof(struct sock_fprog);
4039 }
4040
4041 return packet_setsockopt(sock, level, optname, optval, optlen);
4042}
4043#endif
4044
4045static int packet_notifier(struct notifier_block *this,
4046 unsigned long msg, void *ptr)
4047{
4048 struct sock *sk;
4049 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4050 struct net *net = dev_net(dev);
4051
4052 rcu_read_lock();
4053 sk_for_each_rcu(sk, &net->packet.sklist) {
4054 struct packet_sock *po = pkt_sk(sk);
4055
4056 switch (msg) {
4057 case NETDEV_UNREGISTER:
4058 if (po->mclist)
4059 packet_dev_mclist_delete(dev, &po->mclist);
4060 /* fallthrough */
4061
4062 case NETDEV_DOWN:
4063 if (dev->ifindex == po->ifindex) {
4064 spin_lock(&po->bind_lock);
4065 if (po->running) {
4066 __unregister_prot_hook(sk, false);
4067 sk->sk_err = ENETDOWN;
4068 if (!sock_flag(sk, SOCK_DEAD))
4069 sk->sk_error_report(sk);
4070 }
4071 if (msg == NETDEV_UNREGISTER) {
4072 packet_cached_dev_reset(po);
4073 po->ifindex = -1;
4074 if (po->prot_hook.dev)
4075 dev_put(po->prot_hook.dev);
4076 po->prot_hook.dev = NULL;
4077 }
4078 spin_unlock(&po->bind_lock);
4079 }
4080 break;
4081 case NETDEV_UP:
4082 if (dev->ifindex == po->ifindex) {
4083 spin_lock(&po->bind_lock);
4084 if (po->num)
4085 register_prot_hook(sk);
4086 spin_unlock(&po->bind_lock);
4087 }
4088 break;
4089 }
4090 }
4091 rcu_read_unlock();
4092 return NOTIFY_DONE;
4093}
4094
4095
4096static int packet_ioctl(struct socket *sock, unsigned int cmd,
4097 unsigned long arg)
4098{
4099 struct sock *sk = sock->sk;
4100
4101 switch (cmd) {
4102 case SIOCOUTQ:
4103 {
4104 int amount = sk_wmem_alloc_get(sk);
4105
4106 return put_user(amount, (int __user *)arg);
4107 }
4108 case SIOCINQ:
4109 {
4110 struct sk_buff *skb;
4111 int amount = 0;
4112
4113 spin_lock_bh(&sk->sk_receive_queue.lock);
4114 skb = skb_peek(&sk->sk_receive_queue);
4115 if (skb)
4116 amount = skb->len;
4117 spin_unlock_bh(&sk->sk_receive_queue.lock);
4118 return put_user(amount, (int __user *)arg);
4119 }
4120#ifdef CONFIG_INET
4121 case SIOCADDRT:
4122 case SIOCDELRT:
4123 case SIOCDARP:
4124 case SIOCGARP:
4125 case SIOCSARP:
4126 case SIOCGIFADDR:
4127 case SIOCSIFADDR:
4128 case SIOCGIFBRDADDR:
4129 case SIOCSIFBRDADDR:
4130 case SIOCGIFNETMASK:
4131 case SIOCSIFNETMASK:
4132 case SIOCGIFDSTADDR:
4133 case SIOCSIFDSTADDR:
4134 case SIOCSIFFLAGS:
4135 return inet_dgram_ops.ioctl(sock, cmd, arg);
4136#endif
4137
4138 default:
4139 return -ENOIOCTLCMD;
4140 }
4141 return 0;
4142}
4143
4144static __poll_t packet_poll(struct file *file, struct socket *sock,
4145 poll_table *wait)
4146{
4147 struct sock *sk = sock->sk;
4148 struct packet_sock *po = pkt_sk(sk);
4149 __poll_t mask = datagram_poll(file, sock, wait);
4150
4151 spin_lock_bh(&sk->sk_receive_queue.lock);
4152 if (po->rx_ring.pg_vec) {
4153 if (!packet_previous_rx_frame(po, &po->rx_ring,
4154 TP_STATUS_KERNEL))
4155 mask |= EPOLLIN | EPOLLRDNORM;
4156 }
4157 packet_rcv_try_clear_pressure(po);
4158 spin_unlock_bh(&sk->sk_receive_queue.lock);
4159 spin_lock_bh(&sk->sk_write_queue.lock);
4160 if (po->tx_ring.pg_vec) {
4161 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4162 mask |= EPOLLOUT | EPOLLWRNORM;
4163 }
4164 spin_unlock_bh(&sk->sk_write_queue.lock);
4165 return mask;
4166}
4167
4168
4169/* Dirty? Well, I still did not learn better way to account
4170 * for user mmaps.
4171 */
4172
4173static void packet_mm_open(struct vm_area_struct *vma)
4174{
4175 struct file *file = vma->vm_file;
4176 struct socket *sock = file->private_data;
4177 struct sock *sk = sock->sk;
4178
4179 if (sk)
4180 atomic_inc(&pkt_sk(sk)->mapped);
4181}
4182
4183static void packet_mm_close(struct vm_area_struct *vma)
4184{
4185 struct file *file = vma->vm_file;
4186 struct socket *sock = file->private_data;
4187 struct sock *sk = sock->sk;
4188
4189 if (sk)
4190 atomic_dec(&pkt_sk(sk)->mapped);
4191}
4192
4193static const struct vm_operations_struct packet_mmap_ops = {
4194 .open = packet_mm_open,
4195 .close = packet_mm_close,
4196};
4197
4198static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4199 unsigned int len)
4200{
4201 int i;
4202
4203 for (i = 0; i < len; i++) {
4204 if (likely(pg_vec[i].buffer)) {
4205 if (is_vmalloc_addr(pg_vec[i].buffer))
4206 vfree(pg_vec[i].buffer);
4207 else
4208 free_pages((unsigned long)pg_vec[i].buffer,
4209 order);
4210 pg_vec[i].buffer = NULL;
4211 }
4212 }
4213 kfree(pg_vec);
4214}
4215
4216static char *alloc_one_pg_vec_page(unsigned long order)
4217{
4218 char *buffer;
4219 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4220 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4221
4222 buffer = (char *) __get_free_pages(gfp_flags, order);
4223 if (buffer)
4224 return buffer;
4225
4226 /* __get_free_pages failed, fall back to vmalloc */
4227 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4228 if (buffer)
4229 return buffer;
4230
4231 /* vmalloc failed, lets dig into swap here */
4232 gfp_flags &= ~__GFP_NORETRY;
4233 buffer = (char *) __get_free_pages(gfp_flags, order);
4234 if (buffer)
4235 return buffer;
4236
4237 /* complete and utter failure */
4238 return NULL;
4239}
4240
4241static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4242{
4243 unsigned int block_nr = req->tp_block_nr;
4244 struct pgv *pg_vec;
4245 int i;
4246
4247 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4248 if (unlikely(!pg_vec))
4249 goto out;
4250
4251 for (i = 0; i < block_nr; i++) {
4252 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4253 if (unlikely(!pg_vec[i].buffer))
4254 goto out_free_pgvec;
4255 }
4256
4257out:
4258 return pg_vec;
4259
4260out_free_pgvec:
4261 free_pg_vec(pg_vec, order, block_nr);
4262 pg_vec = NULL;
4263 goto out;
4264}
4265
4266static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4267 int closing, int tx_ring)
4268{
4269 struct pgv *pg_vec = NULL;
4270 struct packet_sock *po = pkt_sk(sk);
4271 int was_running, order = 0;
4272 struct packet_ring_buffer *rb;
4273 struct sk_buff_head *rb_queue;
4274 __be16 num;
4275 int err = -EINVAL;
4276 /* Added to avoid minimal code churn */
4277 struct tpacket_req *req = &req_u->req;
4278
4279 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4280 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4281
4282 err = -EBUSY;
4283 if (!closing) {
4284 if (atomic_read(&po->mapped))
4285 goto out;
4286 if (packet_read_pending(rb))
4287 goto out;
4288 }
4289
4290 if (req->tp_block_nr) {
4291 unsigned int min_frame_size;
4292
4293 /* Sanity tests and some calculations */
4294 err = -EBUSY;
4295 if (unlikely(rb->pg_vec))
4296 goto out;
4297
4298 switch (po->tp_version) {
4299 case TPACKET_V1:
4300 po->tp_hdrlen = TPACKET_HDRLEN;
4301 break;
4302 case TPACKET_V2:
4303 po->tp_hdrlen = TPACKET2_HDRLEN;
4304 break;
4305 case TPACKET_V3:
4306 po->tp_hdrlen = TPACKET3_HDRLEN;
4307 break;
4308 }
4309
4310 err = -EINVAL;
4311 if (unlikely((int)req->tp_block_size <= 0))
4312 goto out;
4313 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4314 goto out;
4315 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4316 if (po->tp_version >= TPACKET_V3 &&
4317 req->tp_block_size <
4318 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4319 goto out;
4320 if (unlikely(req->tp_frame_size < min_frame_size))
4321 goto out;
4322 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4323 goto out;
4324
4325 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4326 if (unlikely(rb->frames_per_block == 0))
4327 goto out;
4328 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4329 goto out;
4330 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4331 req->tp_frame_nr))
4332 goto out;
4333
4334 err = -ENOMEM;
4335 order = get_order(req->tp_block_size);
4336 pg_vec = alloc_pg_vec(req, order);
4337 if (unlikely(!pg_vec))
4338 goto out;
4339 switch (po->tp_version) {
4340 case TPACKET_V3:
4341 /* Block transmit is not supported yet */
4342 if (!tx_ring) {
4343 init_prb_bdqc(po, rb, pg_vec, req_u);
4344 } else {
4345 struct tpacket_req3 *req3 = &req_u->req3;
4346
4347 if (req3->tp_retire_blk_tov ||
4348 req3->tp_sizeof_priv ||
4349 req3->tp_feature_req_word) {
4350 err = -EINVAL;
4351 goto out_free_pg_vec;
4352 }
4353 }
4354 break;
4355 default:
4356 break;
4357 }
4358 }
4359 /* Done */
4360 else {
4361 err = -EINVAL;
4362 if (unlikely(req->tp_frame_nr))
4363 goto out;
4364 }
4365
4366
4367 /* Detach socket from network */
4368 spin_lock(&po->bind_lock);
4369 was_running = po->running;
4370 num = po->num;
4371 if (was_running) {
4372 po->num = 0;
4373 __unregister_prot_hook(sk, false);
4374 }
4375 spin_unlock(&po->bind_lock);
4376
4377 synchronize_net();
4378
4379 err = -EBUSY;
4380 mutex_lock(&po->pg_vec_lock);
4381 if (closing || atomic_read(&po->mapped) == 0) {
4382 err = 0;
4383 spin_lock_bh(&rb_queue->lock);
4384 swap(rb->pg_vec, pg_vec);
4385 rb->frame_max = (req->tp_frame_nr - 1);
4386 rb->head = 0;
4387 rb->frame_size = req->tp_frame_size;
4388 spin_unlock_bh(&rb_queue->lock);
4389
4390 swap(rb->pg_vec_order, order);
4391 swap(rb->pg_vec_len, req->tp_block_nr);
4392
4393 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4394 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4395 tpacket_rcv : packet_rcv;
4396 skb_queue_purge(rb_queue);
4397 if (atomic_read(&po->mapped))
4398 pr_err("packet_mmap: vma is busy: %d\n",
4399 atomic_read(&po->mapped));
4400 }
4401 mutex_unlock(&po->pg_vec_lock);
4402
4403 spin_lock(&po->bind_lock);
4404 if (was_running) {
4405 po->num = num;
4406 register_prot_hook(sk);
4407 }
4408 spin_unlock(&po->bind_lock);
4409 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4410 /* Because we don't support block-based V3 on tx-ring */
4411 if (!tx_ring)
4412 prb_shutdown_retire_blk_timer(po, rb_queue);
4413 }
4414
4415out_free_pg_vec:
4416 if (pg_vec)
4417 free_pg_vec(pg_vec, order, req->tp_block_nr);
4418out:
4419 return err;
4420}
4421
4422static int packet_mmap(struct file *file, struct socket *sock,
4423 struct vm_area_struct *vma)
4424{
4425 struct sock *sk = sock->sk;
4426 struct packet_sock *po = pkt_sk(sk);
4427 unsigned long size, expected_size;
4428 struct packet_ring_buffer *rb;
4429 unsigned long start;
4430 int err = -EINVAL;
4431 int i;
4432
4433 if (vma->vm_pgoff)
4434 return -EINVAL;
4435
4436 mutex_lock(&po->pg_vec_lock);
4437
4438 expected_size = 0;
4439 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4440 if (rb->pg_vec) {
4441 expected_size += rb->pg_vec_len
4442 * rb->pg_vec_pages
4443 * PAGE_SIZE;
4444 }
4445 }
4446
4447 if (expected_size == 0)
4448 goto out;
4449
4450 size = vma->vm_end - vma->vm_start;
4451 if (size != expected_size)
4452 goto out;
4453
4454 start = vma->vm_start;
4455 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4456 if (rb->pg_vec == NULL)
4457 continue;
4458
4459 for (i = 0; i < rb->pg_vec_len; i++) {
4460 struct page *page;
4461 void *kaddr = rb->pg_vec[i].buffer;
4462 int pg_num;
4463
4464 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4465 page = pgv_to_page(kaddr);
4466 err = vm_insert_page(vma, start, page);
4467 if (unlikely(err))
4468 goto out;
4469 start += PAGE_SIZE;
4470 kaddr += PAGE_SIZE;
4471 }
4472 }
4473 }
4474
4475 atomic_inc(&po->mapped);
4476 vma->vm_ops = &packet_mmap_ops;
4477 err = 0;
4478
4479out:
4480 mutex_unlock(&po->pg_vec_lock);
4481 return err;
4482}
4483
4484static const struct proto_ops packet_ops_spkt = {
4485 .family = PF_PACKET,
4486 .owner = THIS_MODULE,
4487 .release = packet_release,
4488 .bind = packet_bind_spkt,
4489 .connect = sock_no_connect,
4490 .socketpair = sock_no_socketpair,
4491 .accept = sock_no_accept,
4492 .getname = packet_getname_spkt,
4493 .poll = datagram_poll,
4494 .ioctl = packet_ioctl,
4495 .gettstamp = sock_gettstamp,
4496 .listen = sock_no_listen,
4497 .shutdown = sock_no_shutdown,
4498 .setsockopt = sock_no_setsockopt,
4499 .getsockopt = sock_no_getsockopt,
4500 .sendmsg = packet_sendmsg_spkt,
4501 .recvmsg = packet_recvmsg,
4502 .mmap = sock_no_mmap,
4503 .sendpage = sock_no_sendpage,
4504};
4505
4506static const struct proto_ops packet_ops = {
4507 .family = PF_PACKET,
4508 .owner = THIS_MODULE,
4509 .release = packet_release,
4510 .bind = packet_bind,
4511 .connect = sock_no_connect,
4512 .socketpair = sock_no_socketpair,
4513 .accept = sock_no_accept,
4514 .getname = packet_getname,
4515 .poll = packet_poll,
4516 .ioctl = packet_ioctl,
4517 .gettstamp = sock_gettstamp,
4518 .listen = sock_no_listen,
4519 .shutdown = sock_no_shutdown,
4520 .setsockopt = packet_setsockopt,
4521 .getsockopt = packet_getsockopt,
4522#ifdef CONFIG_COMPAT
4523 .compat_setsockopt = compat_packet_setsockopt,
4524#endif
4525 .sendmsg = packet_sendmsg,
4526 .recvmsg = packet_recvmsg,
4527 .mmap = packet_mmap,
4528 .sendpage = sock_no_sendpage,
4529};
4530
4531static const struct net_proto_family packet_family_ops = {
4532 .family = PF_PACKET,
4533 .create = packet_create,
4534 .owner = THIS_MODULE,
4535};
4536
4537static struct notifier_block packet_netdev_notifier = {
4538 .notifier_call = packet_notifier,
4539};
4540
4541#ifdef CONFIG_PROC_FS
4542
4543static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4544 __acquires(RCU)
4545{
4546 struct net *net = seq_file_net(seq);
4547
4548 rcu_read_lock();
4549 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4550}
4551
4552static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4553{
4554 struct net *net = seq_file_net(seq);
4555 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4556}
4557
4558static void packet_seq_stop(struct seq_file *seq, void *v)
4559 __releases(RCU)
4560{
4561 rcu_read_unlock();
4562}
4563
4564static int packet_seq_show(struct seq_file *seq, void *v)
4565{
4566 if (v == SEQ_START_TOKEN)
4567 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4568 else {
4569 struct sock *s = sk_entry(v);
4570 const struct packet_sock *po = pkt_sk(s);
4571
4572 seq_printf(seq,
4573 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4574 s,
4575 refcount_read(&s->sk_refcnt),
4576 s->sk_type,
4577 ntohs(po->num),
4578 po->ifindex,
4579 po->running,
4580 atomic_read(&s->sk_rmem_alloc),
4581 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4582 sock_i_ino(s));
4583 }
4584
4585 return 0;
4586}
4587
4588static const struct seq_operations packet_seq_ops = {
4589 .start = packet_seq_start,
4590 .next = packet_seq_next,
4591 .stop = packet_seq_stop,
4592 .show = packet_seq_show,
4593};
4594#endif
4595
4596static int __net_init packet_net_init(struct net *net)
4597{
4598 mutex_init(&net->packet.sklist_lock);
4599 INIT_HLIST_HEAD(&net->packet.sklist);
4600
4601 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4602 sizeof(struct seq_net_private)))
4603 return -ENOMEM;
4604
4605 return 0;
4606}
4607
4608static void __net_exit packet_net_exit(struct net *net)
4609{
4610 remove_proc_entry("packet", net->proc_net);
4611 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4612}
4613
4614static struct pernet_operations packet_net_ops = {
4615 .init = packet_net_init,
4616 .exit = packet_net_exit,
4617};
4618
4619
4620static void __exit packet_exit(void)
4621{
4622 unregister_netdevice_notifier(&packet_netdev_notifier);
4623 unregister_pernet_subsys(&packet_net_ops);
4624 sock_unregister(PF_PACKET);
4625 proto_unregister(&packet_proto);
4626}
4627
4628static int __init packet_init(void)
4629{
4630 int rc;
4631
4632 rc = proto_register(&packet_proto, 0);
4633 if (rc)
4634 goto out;
4635 rc = sock_register(&packet_family_ops);
4636 if (rc)
4637 goto out_proto;
4638 rc = register_pernet_subsys(&packet_net_ops);
4639 if (rc)
4640 goto out_sock;
4641 rc = register_netdevice_notifier(&packet_netdev_notifier);
4642 if (rc)
4643 goto out_pernet;
4644
4645 return 0;
4646
4647out_pernet:
4648 unregister_pernet_subsys(&packet_net_ops);
4649out_sock:
4650 sock_unregister(PF_PACKET);
4651out_proto:
4652 proto_unregister(&packet_proto);
4653out:
4654 return rc;
4655}
4656
4657module_init(packet_init);
4658module_exit(packet_exit);
4659MODULE_LICENSE("GPL");
4660MODULE_ALIAS_NETPROTO(PF_PACKET);