Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * PACKET - implements raw packet sockets.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 *
13 * Fixes:
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
42 * and packet_mreq.
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * layer.
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
47 */
48
49#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 timespec64 *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_timespec64_cond(shhwtstamps->hwtstamp, ts))
419 return TP_STATUS_TS_RAW_HARDWARE;
420
421 if (ktime_to_timespec64_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 timespec64 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 /*
439 * versions 1 through 3 overflow the timestamps in y2106, since they
440 * all store the seconds in a 32-bit unsigned integer.
441 * If we create a version 4, that should have a 64-bit timestamp,
442 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
443 * nanoseconds.
444 */
445 switch (po->tp_version) {
446 case TPACKET_V1:
447 h.h1->tp_sec = ts.tv_sec;
448 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
449 break;
450 case TPACKET_V2:
451 h.h2->tp_sec = ts.tv_sec;
452 h.h2->tp_nsec = ts.tv_nsec;
453 break;
454 case TPACKET_V3:
455 h.h3->tp_sec = ts.tv_sec;
456 h.h3->tp_nsec = ts.tv_nsec;
457 break;
458 default:
459 WARN(1, "TPACKET version not supported.\n");
460 BUG();
461 }
462
463 /* one flush is safe, as both fields always lie on the same cacheline */
464 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
465 smp_wmb();
466
467 return ts_status;
468}
469
470static void *packet_lookup_frame(const struct packet_sock *po,
471 const struct packet_ring_buffer *rb,
472 unsigned int position,
473 int status)
474{
475 unsigned int pg_vec_pos, frame_offset;
476 union tpacket_uhdr h;
477
478 pg_vec_pos = position / rb->frames_per_block;
479 frame_offset = position % rb->frames_per_block;
480
481 h.raw = rb->pg_vec[pg_vec_pos].buffer +
482 (frame_offset * rb->frame_size);
483
484 if (status != __packet_get_status(po, h.raw))
485 return NULL;
486
487 return h.raw;
488}
489
490static void *packet_current_frame(struct packet_sock *po,
491 struct packet_ring_buffer *rb,
492 int status)
493{
494 return packet_lookup_frame(po, rb, rb->head, status);
495}
496
497static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
498{
499 del_timer_sync(&pkc->retire_blk_timer);
500}
501
502static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
503 struct sk_buff_head *rb_queue)
504{
505 struct tpacket_kbdq_core *pkc;
506
507 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
508
509 spin_lock_bh(&rb_queue->lock);
510 pkc->delete_blk_timer = 1;
511 spin_unlock_bh(&rb_queue->lock);
512
513 prb_del_retire_blk_timer(pkc);
514}
515
516static void prb_setup_retire_blk_timer(struct packet_sock *po)
517{
518 struct tpacket_kbdq_core *pkc;
519
520 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
521 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
522 0);
523 pkc->retire_blk_timer.expires = jiffies;
524}
525
526static int prb_calc_retire_blk_tmo(struct packet_sock *po,
527 int blk_size_in_bytes)
528{
529 struct net_device *dev;
530 unsigned int mbits, div;
531 struct ethtool_link_ksettings ecmd;
532 int err;
533
534 rtnl_lock();
535 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
536 if (unlikely(!dev)) {
537 rtnl_unlock();
538 return DEFAULT_PRB_RETIRE_TOV;
539 }
540 err = __ethtool_get_link_ksettings(dev, &ecmd);
541 rtnl_unlock();
542 if (err)
543 return DEFAULT_PRB_RETIRE_TOV;
544
545 /* If the link speed is so slow you don't really
546 * need to worry about perf anyways
547 */
548 if (ecmd.base.speed < SPEED_1000 ||
549 ecmd.base.speed == SPEED_UNKNOWN)
550 return DEFAULT_PRB_RETIRE_TOV;
551
552 div = ecmd.base.speed / 1000;
553 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
554
555 if (div)
556 mbits /= div;
557
558 if (div)
559 return mbits + 1;
560 return mbits;
561}
562
563static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
564 union tpacket_req_u *req_u)
565{
566 p1->feature_req_word = req_u->req3.tp_feature_req_word;
567}
568
569static void init_prb_bdqc(struct packet_sock *po,
570 struct packet_ring_buffer *rb,
571 struct pgv *pg_vec,
572 union tpacket_req_u *req_u)
573{
574 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
575 struct tpacket_block_desc *pbd;
576
577 memset(p1, 0x0, sizeof(*p1));
578
579 p1->knxt_seq_num = 1;
580 p1->pkbdq = pg_vec;
581 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
582 p1->pkblk_start = pg_vec[0].buffer;
583 p1->kblk_size = req_u->req3.tp_block_size;
584 p1->knum_blocks = req_u->req3.tp_block_nr;
585 p1->hdrlen = po->tp_hdrlen;
586 p1->version = po->tp_version;
587 p1->last_kactive_blk_num = 0;
588 po->stats.stats3.tp_freeze_q_cnt = 0;
589 if (req_u->req3.tp_retire_blk_tov)
590 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
591 else
592 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
593 req_u->req3.tp_block_size);
594 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
595 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
596 rwlock_init(&p1->blk_fill_in_prog_lock);
597
598 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
599 prb_init_ft_ops(p1, req_u);
600 prb_setup_retire_blk_timer(po);
601 prb_open_block(p1, pbd);
602}
603
604/* Do NOT update the last_blk_num first.
605 * Assumes sk_buff_head lock is held.
606 */
607static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
608{
609 mod_timer(&pkc->retire_blk_timer,
610 jiffies + pkc->tov_in_jiffies);
611 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
612}
613
614/*
615 * Timer logic:
616 * 1) We refresh the timer only when we open a block.
617 * By doing this we don't waste cycles refreshing the timer
618 * on packet-by-packet basis.
619 *
620 * With a 1MB block-size, on a 1Gbps line, it will take
621 * i) ~8 ms to fill a block + ii) memcpy etc.
622 * In this cut we are not accounting for the memcpy time.
623 *
624 * So, if the user sets the 'tmo' to 10ms then the timer
625 * will never fire while the block is still getting filled
626 * (which is what we want). However, the user could choose
627 * to close a block early and that's fine.
628 *
629 * But when the timer does fire, we check whether or not to refresh it.
630 * Since the tmo granularity is in msecs, it is not too expensive
631 * to refresh the timer, lets say every '8' msecs.
632 * Either the user can set the 'tmo' or we can derive it based on
633 * a) line-speed and b) block-size.
634 * prb_calc_retire_blk_tmo() calculates the tmo.
635 *
636 */
637static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
638{
639 struct packet_sock *po =
640 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
641 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
642 unsigned int frozen;
643 struct tpacket_block_desc *pbd;
644
645 spin_lock(&po->sk.sk_receive_queue.lock);
646
647 frozen = prb_queue_frozen(pkc);
648 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
649
650 if (unlikely(pkc->delete_blk_timer))
651 goto out;
652
653 /* We only need to plug the race when the block is partially filled.
654 * tpacket_rcv:
655 * lock(); increment BLOCK_NUM_PKTS; unlock()
656 * copy_bits() is in progress ...
657 * timer fires on other cpu:
658 * we can't retire the current block because copy_bits
659 * is in progress.
660 *
661 */
662 if (BLOCK_NUM_PKTS(pbd)) {
663 /* Waiting for skb_copy_bits to finish... */
664 write_lock(&pkc->blk_fill_in_prog_lock);
665 write_unlock(&pkc->blk_fill_in_prog_lock);
666 }
667
668 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
669 if (!frozen) {
670 if (!BLOCK_NUM_PKTS(pbd)) {
671 /* An empty block. Just refresh the timer. */
672 goto refresh_timer;
673 }
674 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
675 if (!prb_dispatch_next_block(pkc, po))
676 goto refresh_timer;
677 else
678 goto out;
679 } else {
680 /* Case 1. Queue was frozen because user-space was
681 * lagging behind.
682 */
683 if (prb_curr_blk_in_use(pbd)) {
684 /*
685 * Ok, user-space is still behind.
686 * So just refresh the timer.
687 */
688 goto refresh_timer;
689 } else {
690 /* Case 2. queue was frozen,user-space caught up,
691 * now the link went idle && the timer fired.
692 * We don't have a block to close.So we open this
693 * block and restart the timer.
694 * opening a block thaws the queue,restarts timer
695 * Thawing/timer-refresh is a side effect.
696 */
697 prb_open_block(pkc, pbd);
698 goto out;
699 }
700 }
701 }
702
703refresh_timer:
704 _prb_refresh_rx_retire_blk_timer(pkc);
705
706out:
707 spin_unlock(&po->sk.sk_receive_queue.lock);
708}
709
710static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
711 struct tpacket_block_desc *pbd1, __u32 status)
712{
713 /* Flush everything minus the block header */
714
715#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
716 u8 *start, *end;
717
718 start = (u8 *)pbd1;
719
720 /* Skip the block header(we know header WILL fit in 4K) */
721 start += PAGE_SIZE;
722
723 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
724 for (; start < end; start += PAGE_SIZE)
725 flush_dcache_page(pgv_to_page(start));
726
727 smp_wmb();
728#endif
729
730 /* Now update the block status. */
731
732 BLOCK_STATUS(pbd1) = status;
733
734 /* Flush the block header */
735
736#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
737 start = (u8 *)pbd1;
738 flush_dcache_page(pgv_to_page(start));
739
740 smp_wmb();
741#endif
742}
743
744/*
745 * Side effect:
746 *
747 * 1) flush the block
748 * 2) Increment active_blk_num
749 *
750 * Note:We DONT refresh the timer on purpose.
751 * Because almost always the next block will be opened.
752 */
753static void prb_close_block(struct tpacket_kbdq_core *pkc1,
754 struct tpacket_block_desc *pbd1,
755 struct packet_sock *po, unsigned int stat)
756{
757 __u32 status = TP_STATUS_USER | stat;
758
759 struct tpacket3_hdr *last_pkt;
760 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
761 struct sock *sk = &po->sk;
762
763 if (atomic_read(&po->tp_drops))
764 status |= TP_STATUS_LOSING;
765
766 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
767 last_pkt->tp_next_offset = 0;
768
769 /* Get the ts of the last pkt */
770 if (BLOCK_NUM_PKTS(pbd1)) {
771 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
772 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
773 } else {
774 /* Ok, we tmo'd - so get the current time.
775 *
776 * It shouldn't really happen as we don't close empty
777 * blocks. See prb_retire_rx_blk_timer_expired().
778 */
779 struct timespec64 ts;
780 ktime_get_real_ts64(&ts);
781 h1->ts_last_pkt.ts_sec = ts.tv_sec;
782 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
783 }
784
785 smp_wmb();
786
787 /* Flush the block */
788 prb_flush_block(pkc1, pbd1, status);
789
790 sk->sk_data_ready(sk);
791
792 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
793}
794
795static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
796{
797 pkc->reset_pending_on_curr_blk = 0;
798}
799
800/*
801 * Side effect of opening a block:
802 *
803 * 1) prb_queue is thawed.
804 * 2) retire_blk_timer is refreshed.
805 *
806 */
807static void prb_open_block(struct tpacket_kbdq_core *pkc1,
808 struct tpacket_block_desc *pbd1)
809{
810 struct timespec64 ts;
811 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
812
813 smp_rmb();
814
815 /* We could have just memset this but we will lose the
816 * flexibility of making the priv area sticky
817 */
818
819 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
820 BLOCK_NUM_PKTS(pbd1) = 0;
821 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
822
823 ktime_get_real_ts64(&ts);
824
825 h1->ts_first_pkt.ts_sec = ts.tv_sec;
826 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
827
828 pkc1->pkblk_start = (char *)pbd1;
829 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830
831 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
832 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
833
834 pbd1->version = pkc1->version;
835 pkc1->prev = pkc1->nxt_offset;
836 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
837
838 prb_thaw_queue(pkc1);
839 _prb_refresh_rx_retire_blk_timer(pkc1);
840
841 smp_wmb();
842}
843
844/*
845 * Queue freeze logic:
846 * 1) Assume tp_block_nr = 8 blocks.
847 * 2) At time 't0', user opens Rx ring.
848 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
849 * 4) user-space is either sleeping or processing block '0'.
850 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
851 * it will close block-7,loop around and try to fill block '0'.
852 * call-flow:
853 * __packet_lookup_frame_in_block
854 * prb_retire_current_block()
855 * prb_dispatch_next_block()
856 * |->(BLOCK_STATUS == USER) evaluates to true
857 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
858 * 6) Now there are two cases:
859 * 6.1) Link goes idle right after the queue is frozen.
860 * But remember, the last open_block() refreshed the timer.
861 * When this timer expires,it will refresh itself so that we can
862 * re-open block-0 in near future.
863 * 6.2) Link is busy and keeps on receiving packets. This is a simple
864 * case and __packet_lookup_frame_in_block will check if block-0
865 * is free and can now be re-used.
866 */
867static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
868 struct packet_sock *po)
869{
870 pkc->reset_pending_on_curr_blk = 1;
871 po->stats.stats3.tp_freeze_q_cnt++;
872}
873
874#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
875
876/*
877 * If the next block is free then we will dispatch it
878 * and return a good offset.
879 * Else, we will freeze the queue.
880 * So, caller must check the return value.
881 */
882static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
883 struct packet_sock *po)
884{
885 struct tpacket_block_desc *pbd;
886
887 smp_rmb();
888
889 /* 1. Get current block num */
890 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
891
892 /* 2. If this block is currently in_use then freeze the queue */
893 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
894 prb_freeze_queue(pkc, po);
895 return NULL;
896 }
897
898 /*
899 * 3.
900 * open this block and return the offset where the first packet
901 * needs to get stored.
902 */
903 prb_open_block(pkc, pbd);
904 return (void *)pkc->nxt_offset;
905}
906
907static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
908 struct packet_sock *po, unsigned int status)
909{
910 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
911
912 /* retire/close the current block */
913 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
914 /*
915 * Plug the case where copy_bits() is in progress on
916 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
917 * have space to copy the pkt in the current block and
918 * called prb_retire_current_block()
919 *
920 * We don't need to worry about the TMO case because
921 * the timer-handler already handled this case.
922 */
923 if (!(status & TP_STATUS_BLK_TMO)) {
924 /* Waiting for skb_copy_bits to finish... */
925 write_lock(&pkc->blk_fill_in_prog_lock);
926 write_unlock(&pkc->blk_fill_in_prog_lock);
927 }
928 prb_close_block(pkc, pbd, po, status);
929 return;
930 }
931}
932
933static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
934{
935 return TP_STATUS_USER & BLOCK_STATUS(pbd);
936}
937
938static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
939{
940 return pkc->reset_pending_on_curr_blk;
941}
942
943static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
944 __releases(&pkc->blk_fill_in_prog_lock)
945{
946 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
947
948 read_unlock(&pkc->blk_fill_in_prog_lock);
949}
950
951static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
952 struct tpacket3_hdr *ppd)
953{
954 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
955}
956
957static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
958 struct tpacket3_hdr *ppd)
959{
960 ppd->hv1.tp_rxhash = 0;
961}
962
963static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
964 struct tpacket3_hdr *ppd)
965{
966 if (skb_vlan_tag_present(pkc->skb)) {
967 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
968 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
969 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
970 } else {
971 ppd->hv1.tp_vlan_tci = 0;
972 ppd->hv1.tp_vlan_tpid = 0;
973 ppd->tp_status = TP_STATUS_AVAILABLE;
974 }
975}
976
977static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
978 struct tpacket3_hdr *ppd)
979{
980 ppd->hv1.tp_padding = 0;
981 prb_fill_vlan_info(pkc, ppd);
982
983 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
984 prb_fill_rxhash(pkc, ppd);
985 else
986 prb_clear_rxhash(pkc, ppd);
987}
988
989static void prb_fill_curr_block(char *curr,
990 struct tpacket_kbdq_core *pkc,
991 struct tpacket_block_desc *pbd,
992 unsigned int len)
993 __acquires(&pkc->blk_fill_in_prog_lock)
994{
995 struct tpacket3_hdr *ppd;
996
997 ppd = (struct tpacket3_hdr *)curr;
998 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
999 pkc->prev = curr;
1000 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1001 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1002 BLOCK_NUM_PKTS(pbd) += 1;
1003 read_lock(&pkc->blk_fill_in_prog_lock);
1004 prb_run_all_ft_ops(pkc, ppd);
1005}
1006
1007/* Assumes caller has the sk->rx_queue.lock */
1008static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1009 struct sk_buff *skb,
1010 unsigned int len
1011 )
1012{
1013 struct tpacket_kbdq_core *pkc;
1014 struct tpacket_block_desc *pbd;
1015 char *curr, *end;
1016
1017 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1018 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1019
1020 /* Queue is frozen when user space is lagging behind */
1021 if (prb_queue_frozen(pkc)) {
1022 /*
1023 * Check if that last block which caused the queue to freeze,
1024 * is still in_use by user-space.
1025 */
1026 if (prb_curr_blk_in_use(pbd)) {
1027 /* Can't record this packet */
1028 return NULL;
1029 } else {
1030 /*
1031 * Ok, the block was released by user-space.
1032 * Now let's open that block.
1033 * opening a block also thaws the queue.
1034 * Thawing is a side effect.
1035 */
1036 prb_open_block(pkc, pbd);
1037 }
1038 }
1039
1040 smp_mb();
1041 curr = pkc->nxt_offset;
1042 pkc->skb = skb;
1043 end = (char *)pbd + pkc->kblk_size;
1044
1045 /* first try the current block */
1046 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1047 prb_fill_curr_block(curr, pkc, pbd, len);
1048 return (void *)curr;
1049 }
1050
1051 /* Ok, close the current block */
1052 prb_retire_current_block(pkc, po, 0);
1053
1054 /* Now, try to dispatch the next block */
1055 curr = (char *)prb_dispatch_next_block(pkc, po);
1056 if (curr) {
1057 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1058 prb_fill_curr_block(curr, pkc, pbd, len);
1059 return (void *)curr;
1060 }
1061
1062 /*
1063 * No free blocks are available.user_space hasn't caught up yet.
1064 * Queue was just frozen and now this packet will get dropped.
1065 */
1066 return NULL;
1067}
1068
1069static void *packet_current_rx_frame(struct packet_sock *po,
1070 struct sk_buff *skb,
1071 int status, unsigned int len)
1072{
1073 char *curr = NULL;
1074 switch (po->tp_version) {
1075 case TPACKET_V1:
1076 case TPACKET_V2:
1077 curr = packet_lookup_frame(po, &po->rx_ring,
1078 po->rx_ring.head, status);
1079 return curr;
1080 case TPACKET_V3:
1081 return __packet_lookup_frame_in_block(po, skb, len);
1082 default:
1083 WARN(1, "TPACKET version not supported\n");
1084 BUG();
1085 return NULL;
1086 }
1087}
1088
1089static void *prb_lookup_block(const struct packet_sock *po,
1090 const struct packet_ring_buffer *rb,
1091 unsigned int idx,
1092 int status)
1093{
1094 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1095 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1096
1097 if (status != BLOCK_STATUS(pbd))
1098 return NULL;
1099 return pbd;
1100}
1101
1102static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1103{
1104 unsigned int prev;
1105 if (rb->prb_bdqc.kactive_blk_num)
1106 prev = rb->prb_bdqc.kactive_blk_num-1;
1107 else
1108 prev = rb->prb_bdqc.knum_blocks-1;
1109 return prev;
1110}
1111
1112/* Assumes caller has held the rx_queue.lock */
1113static void *__prb_previous_block(struct packet_sock *po,
1114 struct packet_ring_buffer *rb,
1115 int status)
1116{
1117 unsigned int previous = prb_previous_blk_num(rb);
1118 return prb_lookup_block(po, rb, previous, status);
1119}
1120
1121static void *packet_previous_rx_frame(struct packet_sock *po,
1122 struct packet_ring_buffer *rb,
1123 int status)
1124{
1125 if (po->tp_version <= TPACKET_V2)
1126 return packet_previous_frame(po, rb, status);
1127
1128 return __prb_previous_block(po, rb, status);
1129}
1130
1131static void packet_increment_rx_head(struct packet_sock *po,
1132 struct packet_ring_buffer *rb)
1133{
1134 switch (po->tp_version) {
1135 case TPACKET_V1:
1136 case TPACKET_V2:
1137 return packet_increment_head(rb);
1138 case TPACKET_V3:
1139 default:
1140 WARN(1, "TPACKET version not supported.\n");
1141 BUG();
1142 return;
1143 }
1144}
1145
1146static void *packet_previous_frame(struct packet_sock *po,
1147 struct packet_ring_buffer *rb,
1148 int status)
1149{
1150 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1151 return packet_lookup_frame(po, rb, previous, status);
1152}
1153
1154static void packet_increment_head(struct packet_ring_buffer *buff)
1155{
1156 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1157}
1158
1159static void packet_inc_pending(struct packet_ring_buffer *rb)
1160{
1161 this_cpu_inc(*rb->pending_refcnt);
1162}
1163
1164static void packet_dec_pending(struct packet_ring_buffer *rb)
1165{
1166 this_cpu_dec(*rb->pending_refcnt);
1167}
1168
1169static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1170{
1171 unsigned int refcnt = 0;
1172 int cpu;
1173
1174 /* We don't use pending refcount in rx_ring. */
1175 if (rb->pending_refcnt == NULL)
1176 return 0;
1177
1178 for_each_possible_cpu(cpu)
1179 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1180
1181 return refcnt;
1182}
1183
1184static int packet_alloc_pending(struct packet_sock *po)
1185{
1186 po->rx_ring.pending_refcnt = NULL;
1187
1188 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1189 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1190 return -ENOBUFS;
1191
1192 return 0;
1193}
1194
1195static void packet_free_pending(struct packet_sock *po)
1196{
1197 free_percpu(po->tx_ring.pending_refcnt);
1198}
1199
1200#define ROOM_POW_OFF 2
1201#define ROOM_NONE 0x0
1202#define ROOM_LOW 0x1
1203#define ROOM_NORMAL 0x2
1204
1205static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1206{
1207 int idx, len;
1208
1209 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1210 idx = READ_ONCE(po->rx_ring.head);
1211 if (pow_off)
1212 idx += len >> pow_off;
1213 if (idx >= len)
1214 idx -= len;
1215 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1216}
1217
1218static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1219{
1220 int idx, len;
1221
1222 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1223 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1224 if (pow_off)
1225 idx += len >> pow_off;
1226 if (idx >= len)
1227 idx -= len;
1228 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1229}
1230
1231static int __packet_rcv_has_room(const struct packet_sock *po,
1232 const struct sk_buff *skb)
1233{
1234 const struct sock *sk = &po->sk;
1235 int ret = ROOM_NONE;
1236
1237 if (po->prot_hook.func != tpacket_rcv) {
1238 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1239 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1240 - (skb ? skb->truesize : 0);
1241
1242 if (avail > (rcvbuf >> ROOM_POW_OFF))
1243 return ROOM_NORMAL;
1244 else if (avail > 0)
1245 return ROOM_LOW;
1246 else
1247 return ROOM_NONE;
1248 }
1249
1250 if (po->tp_version == TPACKET_V3) {
1251 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1252 ret = ROOM_NORMAL;
1253 else if (__tpacket_v3_has_room(po, 0))
1254 ret = ROOM_LOW;
1255 } else {
1256 if (__tpacket_has_room(po, ROOM_POW_OFF))
1257 ret = ROOM_NORMAL;
1258 else if (__tpacket_has_room(po, 0))
1259 ret = ROOM_LOW;
1260 }
1261
1262 return ret;
1263}
1264
1265static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266{
1267 int pressure, ret;
1268
1269 ret = __packet_rcv_has_room(po, skb);
1270 pressure = ret != ROOM_NORMAL;
1271
1272 if (READ_ONCE(po->pressure) != pressure)
1273 WRITE_ONCE(po->pressure, pressure);
1274
1275 return ret;
1276}
1277
1278static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1279{
1280 if (READ_ONCE(po->pressure) &&
1281 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1282 WRITE_ONCE(po->pressure, 0);
1283}
1284
1285static void packet_sock_destruct(struct sock *sk)
1286{
1287 skb_queue_purge(&sk->sk_error_queue);
1288
1289 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1290 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1291
1292 if (!sock_flag(sk, SOCK_DEAD)) {
1293 pr_err("Attempt to release alive packet socket: %p\n", sk);
1294 return;
1295 }
1296
1297 sk_refcnt_debug_dec(sk);
1298}
1299
1300static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1301{
1302 u32 *history = po->rollover->history;
1303 u32 victim, rxhash;
1304 int i, count = 0;
1305
1306 rxhash = skb_get_hash(skb);
1307 for (i = 0; i < ROLLOVER_HLEN; i++)
1308 if (READ_ONCE(history[i]) == rxhash)
1309 count++;
1310
1311 victim = prandom_u32() % ROLLOVER_HLEN;
1312
1313 /* Avoid dirtying the cache line if possible */
1314 if (READ_ONCE(history[victim]) != rxhash)
1315 WRITE_ONCE(history[victim], rxhash);
1316
1317 return count > (ROLLOVER_HLEN >> 1);
1318}
1319
1320static unsigned int fanout_demux_hash(struct packet_fanout *f,
1321 struct sk_buff *skb,
1322 unsigned int num)
1323{
1324 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1325}
1326
1327static unsigned int fanout_demux_lb(struct packet_fanout *f,
1328 struct sk_buff *skb,
1329 unsigned int num)
1330{
1331 unsigned int val = atomic_inc_return(&f->rr_cur);
1332
1333 return val % num;
1334}
1335
1336static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1337 struct sk_buff *skb,
1338 unsigned int num)
1339{
1340 return smp_processor_id() % num;
1341}
1342
1343static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1344 struct sk_buff *skb,
1345 unsigned int num)
1346{
1347 return prandom_u32_max(num);
1348}
1349
1350static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1351 struct sk_buff *skb,
1352 unsigned int idx, bool try_self,
1353 unsigned int num)
1354{
1355 struct packet_sock *po, *po_next, *po_skip = NULL;
1356 unsigned int i, j, room = ROOM_NONE;
1357
1358 po = pkt_sk(f->arr[idx]);
1359
1360 if (try_self) {
1361 room = packet_rcv_has_room(po, skb);
1362 if (room == ROOM_NORMAL ||
1363 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1364 return idx;
1365 po_skip = po;
1366 }
1367
1368 i = j = min_t(int, po->rollover->sock, num - 1);
1369 do {
1370 po_next = pkt_sk(f->arr[i]);
1371 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1372 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1373 if (i != j)
1374 po->rollover->sock = i;
1375 atomic_long_inc(&po->rollover->num);
1376 if (room == ROOM_LOW)
1377 atomic_long_inc(&po->rollover->num_huge);
1378 return i;
1379 }
1380
1381 if (++i == num)
1382 i = 0;
1383 } while (i != j);
1384
1385 atomic_long_inc(&po->rollover->num_failed);
1386 return idx;
1387}
1388
1389static unsigned int fanout_demux_qm(struct packet_fanout *f,
1390 struct sk_buff *skb,
1391 unsigned int num)
1392{
1393 return skb_get_queue_mapping(skb) % num;
1394}
1395
1396static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1397 struct sk_buff *skb,
1398 unsigned int num)
1399{
1400 struct bpf_prog *prog;
1401 unsigned int ret = 0;
1402
1403 rcu_read_lock();
1404 prog = rcu_dereference(f->bpf_prog);
1405 if (prog)
1406 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1407 rcu_read_unlock();
1408
1409 return ret;
1410}
1411
1412static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1413{
1414 return f->flags & (flag >> 8);
1415}
1416
1417static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1418 struct packet_type *pt, struct net_device *orig_dev)
1419{
1420 struct packet_fanout *f = pt->af_packet_priv;
1421 unsigned int num = READ_ONCE(f->num_members);
1422 struct net *net = read_pnet(&f->net);
1423 struct packet_sock *po;
1424 unsigned int idx;
1425
1426 if (!net_eq(dev_net(dev), net) || !num) {
1427 kfree_skb(skb);
1428 return 0;
1429 }
1430
1431 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1432 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1433 if (!skb)
1434 return 0;
1435 }
1436 switch (f->type) {
1437 case PACKET_FANOUT_HASH:
1438 default:
1439 idx = fanout_demux_hash(f, skb, num);
1440 break;
1441 case PACKET_FANOUT_LB:
1442 idx = fanout_demux_lb(f, skb, num);
1443 break;
1444 case PACKET_FANOUT_CPU:
1445 idx = fanout_demux_cpu(f, skb, num);
1446 break;
1447 case PACKET_FANOUT_RND:
1448 idx = fanout_demux_rnd(f, skb, num);
1449 break;
1450 case PACKET_FANOUT_QM:
1451 idx = fanout_demux_qm(f, skb, num);
1452 break;
1453 case PACKET_FANOUT_ROLLOVER:
1454 idx = fanout_demux_rollover(f, skb, 0, false, num);
1455 break;
1456 case PACKET_FANOUT_CBPF:
1457 case PACKET_FANOUT_EBPF:
1458 idx = fanout_demux_bpf(f, skb, num);
1459 break;
1460 }
1461
1462 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1463 idx = fanout_demux_rollover(f, skb, idx, true, num);
1464
1465 po = pkt_sk(f->arr[idx]);
1466 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1467}
1468
1469DEFINE_MUTEX(fanout_mutex);
1470EXPORT_SYMBOL_GPL(fanout_mutex);
1471static LIST_HEAD(fanout_list);
1472static u16 fanout_next_id;
1473
1474static void __fanout_link(struct sock *sk, struct packet_sock *po)
1475{
1476 struct packet_fanout *f = po->fanout;
1477
1478 spin_lock(&f->lock);
1479 f->arr[f->num_members] = sk;
1480 smp_wmb();
1481 f->num_members++;
1482 if (f->num_members == 1)
1483 dev_add_pack(&f->prot_hook);
1484 spin_unlock(&f->lock);
1485}
1486
1487static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1488{
1489 struct packet_fanout *f = po->fanout;
1490 int i;
1491
1492 spin_lock(&f->lock);
1493 for (i = 0; i < f->num_members; i++) {
1494 if (f->arr[i] == sk)
1495 break;
1496 }
1497 BUG_ON(i >= f->num_members);
1498 f->arr[i] = f->arr[f->num_members - 1];
1499 f->num_members--;
1500 if (f->num_members == 0)
1501 __dev_remove_pack(&f->prot_hook);
1502 spin_unlock(&f->lock);
1503}
1504
1505static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1506{
1507 if (sk->sk_family != PF_PACKET)
1508 return false;
1509
1510 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1511}
1512
1513static void fanout_init_data(struct packet_fanout *f)
1514{
1515 switch (f->type) {
1516 case PACKET_FANOUT_LB:
1517 atomic_set(&f->rr_cur, 0);
1518 break;
1519 case PACKET_FANOUT_CBPF:
1520 case PACKET_FANOUT_EBPF:
1521 RCU_INIT_POINTER(f->bpf_prog, NULL);
1522 break;
1523 }
1524}
1525
1526static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1527{
1528 struct bpf_prog *old;
1529
1530 spin_lock(&f->lock);
1531 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1532 rcu_assign_pointer(f->bpf_prog, new);
1533 spin_unlock(&f->lock);
1534
1535 if (old) {
1536 synchronize_net();
1537 bpf_prog_destroy(old);
1538 }
1539}
1540
1541static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1542 unsigned int len)
1543{
1544 struct bpf_prog *new;
1545 struct sock_fprog fprog;
1546 int ret;
1547
1548 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1549 return -EPERM;
1550
1551 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1552 if (ret)
1553 return ret;
1554
1555 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1556 if (ret)
1557 return ret;
1558
1559 __fanout_set_data_bpf(po->fanout, new);
1560 return 0;
1561}
1562
1563static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1564 unsigned int len)
1565{
1566 struct bpf_prog *new;
1567 u32 fd;
1568
1569 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1570 return -EPERM;
1571 if (len != sizeof(fd))
1572 return -EINVAL;
1573 if (copy_from_sockptr(&fd, data, len))
1574 return -EFAULT;
1575
1576 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1577 if (IS_ERR(new))
1578 return PTR_ERR(new);
1579
1580 __fanout_set_data_bpf(po->fanout, new);
1581 return 0;
1582}
1583
1584static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1585 unsigned int len)
1586{
1587 switch (po->fanout->type) {
1588 case PACKET_FANOUT_CBPF:
1589 return fanout_set_data_cbpf(po, data, len);
1590 case PACKET_FANOUT_EBPF:
1591 return fanout_set_data_ebpf(po, data, len);
1592 default:
1593 return -EINVAL;
1594 }
1595}
1596
1597static void fanout_release_data(struct packet_fanout *f)
1598{
1599 switch (f->type) {
1600 case PACKET_FANOUT_CBPF:
1601 case PACKET_FANOUT_EBPF:
1602 __fanout_set_data_bpf(f, NULL);
1603 }
1604}
1605
1606static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1607{
1608 struct packet_fanout *f;
1609
1610 list_for_each_entry(f, &fanout_list, list) {
1611 if (f->id == candidate_id &&
1612 read_pnet(&f->net) == sock_net(sk)) {
1613 return false;
1614 }
1615 }
1616 return true;
1617}
1618
1619static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1620{
1621 u16 id = fanout_next_id;
1622
1623 do {
1624 if (__fanout_id_is_free(sk, id)) {
1625 *new_id = id;
1626 fanout_next_id = id + 1;
1627 return true;
1628 }
1629
1630 id++;
1631 } while (id != fanout_next_id);
1632
1633 return false;
1634}
1635
1636static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1637{
1638 struct packet_rollover *rollover = NULL;
1639 struct packet_sock *po = pkt_sk(sk);
1640 struct packet_fanout *f, *match;
1641 u8 type = type_flags & 0xff;
1642 u8 flags = type_flags >> 8;
1643 int err;
1644
1645 switch (type) {
1646 case PACKET_FANOUT_ROLLOVER:
1647 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1648 return -EINVAL;
1649 case PACKET_FANOUT_HASH:
1650 case PACKET_FANOUT_LB:
1651 case PACKET_FANOUT_CPU:
1652 case PACKET_FANOUT_RND:
1653 case PACKET_FANOUT_QM:
1654 case PACKET_FANOUT_CBPF:
1655 case PACKET_FANOUT_EBPF:
1656 break;
1657 default:
1658 return -EINVAL;
1659 }
1660
1661 mutex_lock(&fanout_mutex);
1662
1663 err = -EALREADY;
1664 if (po->fanout)
1665 goto out;
1666
1667 if (type == PACKET_FANOUT_ROLLOVER ||
1668 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1669 err = -ENOMEM;
1670 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1671 if (!rollover)
1672 goto out;
1673 atomic_long_set(&rollover->num, 0);
1674 atomic_long_set(&rollover->num_huge, 0);
1675 atomic_long_set(&rollover->num_failed, 0);
1676 }
1677
1678 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1679 if (id != 0) {
1680 err = -EINVAL;
1681 goto out;
1682 }
1683 if (!fanout_find_new_id(sk, &id)) {
1684 err = -ENOMEM;
1685 goto out;
1686 }
1687 /* ephemeral flag for the first socket in the group: drop it */
1688 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1689 }
1690
1691 match = NULL;
1692 list_for_each_entry(f, &fanout_list, list) {
1693 if (f->id == id &&
1694 read_pnet(&f->net) == sock_net(sk)) {
1695 match = f;
1696 break;
1697 }
1698 }
1699 err = -EINVAL;
1700 if (match && match->flags != flags)
1701 goto out;
1702 if (!match) {
1703 err = -ENOMEM;
1704 match = kzalloc(sizeof(*match), GFP_KERNEL);
1705 if (!match)
1706 goto out;
1707 write_pnet(&match->net, sock_net(sk));
1708 match->id = id;
1709 match->type = type;
1710 match->flags = flags;
1711 INIT_LIST_HEAD(&match->list);
1712 spin_lock_init(&match->lock);
1713 refcount_set(&match->sk_ref, 0);
1714 fanout_init_data(match);
1715 match->prot_hook.type = po->prot_hook.type;
1716 match->prot_hook.dev = po->prot_hook.dev;
1717 match->prot_hook.func = packet_rcv_fanout;
1718 match->prot_hook.af_packet_priv = match;
1719 match->prot_hook.id_match = match_fanout_group;
1720 list_add(&match->list, &fanout_list);
1721 }
1722 err = -EINVAL;
1723
1724 spin_lock(&po->bind_lock);
1725 if (po->running &&
1726 match->type == type &&
1727 match->prot_hook.type == po->prot_hook.type &&
1728 match->prot_hook.dev == po->prot_hook.dev) {
1729 err = -ENOSPC;
1730 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1731 __dev_remove_pack(&po->prot_hook);
1732 po->fanout = match;
1733 po->rollover = rollover;
1734 rollover = NULL;
1735 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1736 __fanout_link(sk, po);
1737 err = 0;
1738 }
1739 }
1740 spin_unlock(&po->bind_lock);
1741
1742 if (err && !refcount_read(&match->sk_ref)) {
1743 list_del(&match->list);
1744 kfree(match);
1745 }
1746
1747out:
1748 kfree(rollover);
1749 mutex_unlock(&fanout_mutex);
1750 return err;
1751}
1752
1753/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1754 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1755 * It is the responsibility of the caller to call fanout_release_data() and
1756 * free the returned packet_fanout (after synchronize_net())
1757 */
1758static struct packet_fanout *fanout_release(struct sock *sk)
1759{
1760 struct packet_sock *po = pkt_sk(sk);
1761 struct packet_fanout *f;
1762
1763 mutex_lock(&fanout_mutex);
1764 f = po->fanout;
1765 if (f) {
1766 po->fanout = NULL;
1767
1768 if (refcount_dec_and_test(&f->sk_ref))
1769 list_del(&f->list);
1770 else
1771 f = NULL;
1772 }
1773 mutex_unlock(&fanout_mutex);
1774
1775 return f;
1776}
1777
1778static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1779 struct sk_buff *skb)
1780{
1781 /* Earlier code assumed this would be a VLAN pkt, double-check
1782 * this now that we have the actual packet in hand. We can only
1783 * do this check on Ethernet devices.
1784 */
1785 if (unlikely(dev->type != ARPHRD_ETHER))
1786 return false;
1787
1788 skb_reset_mac_header(skb);
1789 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1790}
1791
1792static const struct proto_ops packet_ops;
1793
1794static const struct proto_ops packet_ops_spkt;
1795
1796static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1797 struct packet_type *pt, struct net_device *orig_dev)
1798{
1799 struct sock *sk;
1800 struct sockaddr_pkt *spkt;
1801
1802 /*
1803 * When we registered the protocol we saved the socket in the data
1804 * field for just this event.
1805 */
1806
1807 sk = pt->af_packet_priv;
1808
1809 /*
1810 * Yank back the headers [hope the device set this
1811 * right or kerboom...]
1812 *
1813 * Incoming packets have ll header pulled,
1814 * push it back.
1815 *
1816 * For outgoing ones skb->data == skb_mac_header(skb)
1817 * so that this procedure is noop.
1818 */
1819
1820 if (skb->pkt_type == PACKET_LOOPBACK)
1821 goto out;
1822
1823 if (!net_eq(dev_net(dev), sock_net(sk)))
1824 goto out;
1825
1826 skb = skb_share_check(skb, GFP_ATOMIC);
1827 if (skb == NULL)
1828 goto oom;
1829
1830 /* drop any routing info */
1831 skb_dst_drop(skb);
1832
1833 /* drop conntrack reference */
1834 nf_reset_ct(skb);
1835
1836 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1837
1838 skb_push(skb, skb->data - skb_mac_header(skb));
1839
1840 /*
1841 * The SOCK_PACKET socket receives _all_ frames.
1842 */
1843
1844 spkt->spkt_family = dev->type;
1845 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1846 spkt->spkt_protocol = skb->protocol;
1847
1848 /*
1849 * Charge the memory to the socket. This is done specifically
1850 * to prevent sockets using all the memory up.
1851 */
1852
1853 if (sock_queue_rcv_skb(sk, skb) == 0)
1854 return 0;
1855
1856out:
1857 kfree_skb(skb);
1858oom:
1859 return 0;
1860}
1861
1862static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1863{
1864 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1865 sock->type == SOCK_RAW) {
1866 skb_reset_mac_header(skb);
1867 skb->protocol = dev_parse_header_protocol(skb);
1868 }
1869
1870 skb_probe_transport_header(skb);
1871}
1872
1873/*
1874 * Output a raw packet to a device layer. This bypasses all the other
1875 * protocol layers and you must therefore supply it with a complete frame
1876 */
1877
1878static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1879 size_t len)
1880{
1881 struct sock *sk = sock->sk;
1882 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1883 struct sk_buff *skb = NULL;
1884 struct net_device *dev;
1885 struct sockcm_cookie sockc;
1886 __be16 proto = 0;
1887 int err;
1888 int extra_len = 0;
1889
1890 /*
1891 * Get and verify the address.
1892 */
1893
1894 if (saddr) {
1895 if (msg->msg_namelen < sizeof(struct sockaddr))
1896 return -EINVAL;
1897 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1898 proto = saddr->spkt_protocol;
1899 } else
1900 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1901
1902 /*
1903 * Find the device first to size check it
1904 */
1905
1906 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1907retry:
1908 rcu_read_lock();
1909 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1910 err = -ENODEV;
1911 if (dev == NULL)
1912 goto out_unlock;
1913
1914 err = -ENETDOWN;
1915 if (!(dev->flags & IFF_UP))
1916 goto out_unlock;
1917
1918 /*
1919 * You may not queue a frame bigger than the mtu. This is the lowest level
1920 * raw protocol and you must do your own fragmentation at this level.
1921 */
1922
1923 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1924 if (!netif_supports_nofcs(dev)) {
1925 err = -EPROTONOSUPPORT;
1926 goto out_unlock;
1927 }
1928 extra_len = 4; /* We're doing our own CRC */
1929 }
1930
1931 err = -EMSGSIZE;
1932 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1933 goto out_unlock;
1934
1935 if (!skb) {
1936 size_t reserved = LL_RESERVED_SPACE(dev);
1937 int tlen = dev->needed_tailroom;
1938 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1939
1940 rcu_read_unlock();
1941 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1942 if (skb == NULL)
1943 return -ENOBUFS;
1944 /* FIXME: Save some space for broken drivers that write a hard
1945 * header at transmission time by themselves. PPP is the notable
1946 * one here. This should really be fixed at the driver level.
1947 */
1948 skb_reserve(skb, reserved);
1949 skb_reset_network_header(skb);
1950
1951 /* Try to align data part correctly */
1952 if (hhlen) {
1953 skb->data -= hhlen;
1954 skb->tail -= hhlen;
1955 if (len < hhlen)
1956 skb_reset_network_header(skb);
1957 }
1958 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1959 if (err)
1960 goto out_free;
1961 goto retry;
1962 }
1963
1964 if (!dev_validate_header(dev, skb->data, len)) {
1965 err = -EINVAL;
1966 goto out_unlock;
1967 }
1968 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1969 !packet_extra_vlan_len_allowed(dev, skb)) {
1970 err = -EMSGSIZE;
1971 goto out_unlock;
1972 }
1973
1974 sockcm_init(&sockc, sk);
1975 if (msg->msg_controllen) {
1976 err = sock_cmsg_send(sk, msg, &sockc);
1977 if (unlikely(err))
1978 goto out_unlock;
1979 }
1980
1981 skb->protocol = proto;
1982 skb->dev = dev;
1983 skb->priority = sk->sk_priority;
1984 skb->mark = sk->sk_mark;
1985 skb->tstamp = sockc.transmit_time;
1986
1987 skb_setup_tx_timestamp(skb, sockc.tsflags);
1988
1989 if (unlikely(extra_len == 4))
1990 skb->no_fcs = 1;
1991
1992 packet_parse_headers(skb, sock);
1993
1994 dev_queue_xmit(skb);
1995 rcu_read_unlock();
1996 return len;
1997
1998out_unlock:
1999 rcu_read_unlock();
2000out_free:
2001 kfree_skb(skb);
2002 return err;
2003}
2004
2005static unsigned int run_filter(struct sk_buff *skb,
2006 const struct sock *sk,
2007 unsigned int res)
2008{
2009 struct sk_filter *filter;
2010
2011 rcu_read_lock();
2012 filter = rcu_dereference(sk->sk_filter);
2013 if (filter != NULL)
2014 res = bpf_prog_run_clear_cb(filter->prog, skb);
2015 rcu_read_unlock();
2016
2017 return res;
2018}
2019
2020static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2021 size_t *len)
2022{
2023 struct virtio_net_hdr vnet_hdr;
2024
2025 if (*len < sizeof(vnet_hdr))
2026 return -EINVAL;
2027 *len -= sizeof(vnet_hdr);
2028
2029 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2030 return -EINVAL;
2031
2032 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2033}
2034
2035/*
2036 * This function makes lazy skb cloning in hope that most of packets
2037 * are discarded by BPF.
2038 *
2039 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2040 * and skb->cb are mangled. It works because (and until) packets
2041 * falling here are owned by current CPU. Output packets are cloned
2042 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2043 * sequencially, so that if we return skb to original state on exit,
2044 * we will not harm anyone.
2045 */
2046
2047static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2048 struct packet_type *pt, struct net_device *orig_dev)
2049{
2050 struct sock *sk;
2051 struct sockaddr_ll *sll;
2052 struct packet_sock *po;
2053 u8 *skb_head = skb->data;
2054 int skb_len = skb->len;
2055 unsigned int snaplen, res;
2056 bool is_drop_n_account = false;
2057
2058 if (skb->pkt_type == PACKET_LOOPBACK)
2059 goto drop;
2060
2061 sk = pt->af_packet_priv;
2062 po = pkt_sk(sk);
2063
2064 if (!net_eq(dev_net(dev), sock_net(sk)))
2065 goto drop;
2066
2067 skb->dev = dev;
2068
2069 if (dev->header_ops) {
2070 /* The device has an explicit notion of ll header,
2071 * exported to higher levels.
2072 *
2073 * Otherwise, the device hides details of its frame
2074 * structure, so that corresponding packet head is
2075 * never delivered to user.
2076 */
2077 if (sk->sk_type != SOCK_DGRAM)
2078 skb_push(skb, skb->data - skb_mac_header(skb));
2079 else if (skb->pkt_type == PACKET_OUTGOING) {
2080 /* Special case: outgoing packets have ll header at head */
2081 skb_pull(skb, skb_network_offset(skb));
2082 }
2083 }
2084
2085 snaplen = skb->len;
2086
2087 res = run_filter(skb, sk, snaplen);
2088 if (!res)
2089 goto drop_n_restore;
2090 if (snaplen > res)
2091 snaplen = res;
2092
2093 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2094 goto drop_n_acct;
2095
2096 if (skb_shared(skb)) {
2097 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2098 if (nskb == NULL)
2099 goto drop_n_acct;
2100
2101 if (skb_head != skb->data) {
2102 skb->data = skb_head;
2103 skb->len = skb_len;
2104 }
2105 consume_skb(skb);
2106 skb = nskb;
2107 }
2108
2109 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2110
2111 sll = &PACKET_SKB_CB(skb)->sa.ll;
2112 sll->sll_hatype = dev->type;
2113 sll->sll_pkttype = skb->pkt_type;
2114 if (unlikely(po->origdev))
2115 sll->sll_ifindex = orig_dev->ifindex;
2116 else
2117 sll->sll_ifindex = dev->ifindex;
2118
2119 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2120
2121 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2122 * Use their space for storing the original skb length.
2123 */
2124 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2125
2126 if (pskb_trim(skb, snaplen))
2127 goto drop_n_acct;
2128
2129 skb_set_owner_r(skb, sk);
2130 skb->dev = NULL;
2131 skb_dst_drop(skb);
2132
2133 /* drop conntrack reference */
2134 nf_reset_ct(skb);
2135
2136 spin_lock(&sk->sk_receive_queue.lock);
2137 po->stats.stats1.tp_packets++;
2138 sock_skb_set_dropcount(sk, skb);
2139 __skb_queue_tail(&sk->sk_receive_queue, skb);
2140 spin_unlock(&sk->sk_receive_queue.lock);
2141 sk->sk_data_ready(sk);
2142 return 0;
2143
2144drop_n_acct:
2145 is_drop_n_account = true;
2146 atomic_inc(&po->tp_drops);
2147 atomic_inc(&sk->sk_drops);
2148
2149drop_n_restore:
2150 if (skb_head != skb->data && skb_shared(skb)) {
2151 skb->data = skb_head;
2152 skb->len = skb_len;
2153 }
2154drop:
2155 if (!is_drop_n_account)
2156 consume_skb(skb);
2157 else
2158 kfree_skb(skb);
2159 return 0;
2160}
2161
2162static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2163 struct packet_type *pt, struct net_device *orig_dev)
2164{
2165 struct sock *sk;
2166 struct packet_sock *po;
2167 struct sockaddr_ll *sll;
2168 union tpacket_uhdr h;
2169 u8 *skb_head = skb->data;
2170 int skb_len = skb->len;
2171 unsigned int snaplen, res;
2172 unsigned long status = TP_STATUS_USER;
2173 unsigned short macoff, hdrlen;
2174 unsigned int netoff;
2175 struct sk_buff *copy_skb = NULL;
2176 struct timespec64 ts;
2177 __u32 ts_status;
2178 bool is_drop_n_account = false;
2179 unsigned int slot_id = 0;
2180 bool do_vnet = false;
2181
2182 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2183 * We may add members to them until current aligned size without forcing
2184 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2185 */
2186 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2187 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2188
2189 if (skb->pkt_type == PACKET_LOOPBACK)
2190 goto drop;
2191
2192 sk = pt->af_packet_priv;
2193 po = pkt_sk(sk);
2194
2195 if (!net_eq(dev_net(dev), sock_net(sk)))
2196 goto drop;
2197
2198 if (dev->header_ops) {
2199 if (sk->sk_type != SOCK_DGRAM)
2200 skb_push(skb, skb->data - skb_mac_header(skb));
2201 else if (skb->pkt_type == PACKET_OUTGOING) {
2202 /* Special case: outgoing packets have ll header at head */
2203 skb_pull(skb, skb_network_offset(skb));
2204 }
2205 }
2206
2207 snaplen = skb->len;
2208
2209 res = run_filter(skb, sk, snaplen);
2210 if (!res)
2211 goto drop_n_restore;
2212
2213 /* If we are flooded, just give up */
2214 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2215 atomic_inc(&po->tp_drops);
2216 goto drop_n_restore;
2217 }
2218
2219 if (skb->ip_summed == CHECKSUM_PARTIAL)
2220 status |= TP_STATUS_CSUMNOTREADY;
2221 else if (skb->pkt_type != PACKET_OUTGOING &&
2222 (skb->ip_summed == CHECKSUM_COMPLETE ||
2223 skb_csum_unnecessary(skb)))
2224 status |= TP_STATUS_CSUM_VALID;
2225
2226 if (snaplen > res)
2227 snaplen = res;
2228
2229 if (sk->sk_type == SOCK_DGRAM) {
2230 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2231 po->tp_reserve;
2232 } else {
2233 unsigned int maclen = skb_network_offset(skb);
2234 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2235 (maclen < 16 ? 16 : maclen)) +
2236 po->tp_reserve;
2237 if (po->has_vnet_hdr) {
2238 netoff += sizeof(struct virtio_net_hdr);
2239 do_vnet = true;
2240 }
2241 macoff = netoff - maclen;
2242 }
2243 if (netoff > USHRT_MAX) {
2244 atomic_inc(&po->tp_drops);
2245 goto drop_n_restore;
2246 }
2247 if (po->tp_version <= TPACKET_V2) {
2248 if (macoff + snaplen > po->rx_ring.frame_size) {
2249 if (po->copy_thresh &&
2250 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2251 if (skb_shared(skb)) {
2252 copy_skb = skb_clone(skb, GFP_ATOMIC);
2253 } else {
2254 copy_skb = skb_get(skb);
2255 skb_head = skb->data;
2256 }
2257 if (copy_skb)
2258 skb_set_owner_r(copy_skb, sk);
2259 }
2260 snaplen = po->rx_ring.frame_size - macoff;
2261 if ((int)snaplen < 0) {
2262 snaplen = 0;
2263 do_vnet = false;
2264 }
2265 }
2266 } else if (unlikely(macoff + snaplen >
2267 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2268 u32 nval;
2269
2270 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2271 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2272 snaplen, nval, macoff);
2273 snaplen = nval;
2274 if (unlikely((int)snaplen < 0)) {
2275 snaplen = 0;
2276 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2277 do_vnet = false;
2278 }
2279 }
2280 spin_lock(&sk->sk_receive_queue.lock);
2281 h.raw = packet_current_rx_frame(po, skb,
2282 TP_STATUS_KERNEL, (macoff+snaplen));
2283 if (!h.raw)
2284 goto drop_n_account;
2285
2286 if (po->tp_version <= TPACKET_V2) {
2287 slot_id = po->rx_ring.head;
2288 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2289 goto drop_n_account;
2290 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2291 }
2292
2293 if (do_vnet &&
2294 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2295 sizeof(struct virtio_net_hdr),
2296 vio_le(), true, 0)) {
2297 if (po->tp_version == TPACKET_V3)
2298 prb_clear_blk_fill_status(&po->rx_ring);
2299 goto drop_n_account;
2300 }
2301
2302 if (po->tp_version <= TPACKET_V2) {
2303 packet_increment_rx_head(po, &po->rx_ring);
2304 /*
2305 * LOSING will be reported till you read the stats,
2306 * because it's COR - Clear On Read.
2307 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2308 * at packet level.
2309 */
2310 if (atomic_read(&po->tp_drops))
2311 status |= TP_STATUS_LOSING;
2312 }
2313
2314 po->stats.stats1.tp_packets++;
2315 if (copy_skb) {
2316 status |= TP_STATUS_COPY;
2317 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2318 }
2319 spin_unlock(&sk->sk_receive_queue.lock);
2320
2321 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2322
2323 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2324 ktime_get_real_ts64(&ts);
2325
2326 status |= ts_status;
2327
2328 switch (po->tp_version) {
2329 case TPACKET_V1:
2330 h.h1->tp_len = skb->len;
2331 h.h1->tp_snaplen = snaplen;
2332 h.h1->tp_mac = macoff;
2333 h.h1->tp_net = netoff;
2334 h.h1->tp_sec = ts.tv_sec;
2335 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2336 hdrlen = sizeof(*h.h1);
2337 break;
2338 case TPACKET_V2:
2339 h.h2->tp_len = skb->len;
2340 h.h2->tp_snaplen = snaplen;
2341 h.h2->tp_mac = macoff;
2342 h.h2->tp_net = netoff;
2343 h.h2->tp_sec = ts.tv_sec;
2344 h.h2->tp_nsec = ts.tv_nsec;
2345 if (skb_vlan_tag_present(skb)) {
2346 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2347 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2348 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2349 } else {
2350 h.h2->tp_vlan_tci = 0;
2351 h.h2->tp_vlan_tpid = 0;
2352 }
2353 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2354 hdrlen = sizeof(*h.h2);
2355 break;
2356 case TPACKET_V3:
2357 /* tp_nxt_offset,vlan are already populated above.
2358 * So DONT clear those fields here
2359 */
2360 h.h3->tp_status |= status;
2361 h.h3->tp_len = skb->len;
2362 h.h3->tp_snaplen = snaplen;
2363 h.h3->tp_mac = macoff;
2364 h.h3->tp_net = netoff;
2365 h.h3->tp_sec = ts.tv_sec;
2366 h.h3->tp_nsec = ts.tv_nsec;
2367 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2368 hdrlen = sizeof(*h.h3);
2369 break;
2370 default:
2371 BUG();
2372 }
2373
2374 sll = h.raw + TPACKET_ALIGN(hdrlen);
2375 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2376 sll->sll_family = AF_PACKET;
2377 sll->sll_hatype = dev->type;
2378 sll->sll_protocol = skb->protocol;
2379 sll->sll_pkttype = skb->pkt_type;
2380 if (unlikely(po->origdev))
2381 sll->sll_ifindex = orig_dev->ifindex;
2382 else
2383 sll->sll_ifindex = dev->ifindex;
2384
2385 smp_mb();
2386
2387#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2388 if (po->tp_version <= TPACKET_V2) {
2389 u8 *start, *end;
2390
2391 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2392 macoff + snaplen);
2393
2394 for (start = h.raw; start < end; start += PAGE_SIZE)
2395 flush_dcache_page(pgv_to_page(start));
2396 }
2397 smp_wmb();
2398#endif
2399
2400 if (po->tp_version <= TPACKET_V2) {
2401 spin_lock(&sk->sk_receive_queue.lock);
2402 __packet_set_status(po, h.raw, status);
2403 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2404 spin_unlock(&sk->sk_receive_queue.lock);
2405 sk->sk_data_ready(sk);
2406 } else if (po->tp_version == TPACKET_V3) {
2407 prb_clear_blk_fill_status(&po->rx_ring);
2408 }
2409
2410drop_n_restore:
2411 if (skb_head != skb->data && skb_shared(skb)) {
2412 skb->data = skb_head;
2413 skb->len = skb_len;
2414 }
2415drop:
2416 if (!is_drop_n_account)
2417 consume_skb(skb);
2418 else
2419 kfree_skb(skb);
2420 return 0;
2421
2422drop_n_account:
2423 spin_unlock(&sk->sk_receive_queue.lock);
2424 atomic_inc(&po->tp_drops);
2425 is_drop_n_account = true;
2426
2427 sk->sk_data_ready(sk);
2428 kfree_skb(copy_skb);
2429 goto drop_n_restore;
2430}
2431
2432static void tpacket_destruct_skb(struct sk_buff *skb)
2433{
2434 struct packet_sock *po = pkt_sk(skb->sk);
2435
2436 if (likely(po->tx_ring.pg_vec)) {
2437 void *ph;
2438 __u32 ts;
2439
2440 ph = skb_zcopy_get_nouarg(skb);
2441 packet_dec_pending(&po->tx_ring);
2442
2443 ts = __packet_set_timestamp(po, ph, skb);
2444 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2445
2446 if (!packet_read_pending(&po->tx_ring))
2447 complete(&po->skb_completion);
2448 }
2449
2450 sock_wfree(skb);
2451}
2452
2453static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2454{
2455 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2456 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2457 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2458 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2459 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2460 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2461 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2462
2463 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2464 return -EINVAL;
2465
2466 return 0;
2467}
2468
2469static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2470 struct virtio_net_hdr *vnet_hdr)
2471{
2472 if (*len < sizeof(*vnet_hdr))
2473 return -EINVAL;
2474 *len -= sizeof(*vnet_hdr);
2475
2476 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2477 return -EFAULT;
2478
2479 return __packet_snd_vnet_parse(vnet_hdr, *len);
2480}
2481
2482static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2483 void *frame, struct net_device *dev, void *data, int tp_len,
2484 __be16 proto, unsigned char *addr, int hlen, int copylen,
2485 const struct sockcm_cookie *sockc)
2486{
2487 union tpacket_uhdr ph;
2488 int to_write, offset, len, nr_frags, len_max;
2489 struct socket *sock = po->sk.sk_socket;
2490 struct page *page;
2491 int err;
2492
2493 ph.raw = frame;
2494
2495 skb->protocol = proto;
2496 skb->dev = dev;
2497 skb->priority = po->sk.sk_priority;
2498 skb->mark = po->sk.sk_mark;
2499 skb->tstamp = sockc->transmit_time;
2500 skb_setup_tx_timestamp(skb, sockc->tsflags);
2501 skb_zcopy_set_nouarg(skb, ph.raw);
2502
2503 skb_reserve(skb, hlen);
2504 skb_reset_network_header(skb);
2505
2506 to_write = tp_len;
2507
2508 if (sock->type == SOCK_DGRAM) {
2509 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2510 NULL, tp_len);
2511 if (unlikely(err < 0))
2512 return -EINVAL;
2513 } else if (copylen) {
2514 int hdrlen = min_t(int, copylen, tp_len);
2515
2516 skb_push(skb, dev->hard_header_len);
2517 skb_put(skb, copylen - dev->hard_header_len);
2518 err = skb_store_bits(skb, 0, data, hdrlen);
2519 if (unlikely(err))
2520 return err;
2521 if (!dev_validate_header(dev, skb->data, hdrlen))
2522 return -EINVAL;
2523
2524 data += hdrlen;
2525 to_write -= hdrlen;
2526 }
2527
2528 offset = offset_in_page(data);
2529 len_max = PAGE_SIZE - offset;
2530 len = ((to_write > len_max) ? len_max : to_write);
2531
2532 skb->data_len = to_write;
2533 skb->len += to_write;
2534 skb->truesize += to_write;
2535 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2536
2537 while (likely(to_write)) {
2538 nr_frags = skb_shinfo(skb)->nr_frags;
2539
2540 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2541 pr_err("Packet exceed the number of skb frags(%lu)\n",
2542 MAX_SKB_FRAGS);
2543 return -EFAULT;
2544 }
2545
2546 page = pgv_to_page(data);
2547 data += len;
2548 flush_dcache_page(page);
2549 get_page(page);
2550 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2551 to_write -= len;
2552 offset = 0;
2553 len_max = PAGE_SIZE;
2554 len = ((to_write > len_max) ? len_max : to_write);
2555 }
2556
2557 packet_parse_headers(skb, sock);
2558
2559 return tp_len;
2560}
2561
2562static int tpacket_parse_header(struct packet_sock *po, void *frame,
2563 int size_max, void **data)
2564{
2565 union tpacket_uhdr ph;
2566 int tp_len, off;
2567
2568 ph.raw = frame;
2569
2570 switch (po->tp_version) {
2571 case TPACKET_V3:
2572 if (ph.h3->tp_next_offset != 0) {
2573 pr_warn_once("variable sized slot not supported");
2574 return -EINVAL;
2575 }
2576 tp_len = ph.h3->tp_len;
2577 break;
2578 case TPACKET_V2:
2579 tp_len = ph.h2->tp_len;
2580 break;
2581 default:
2582 tp_len = ph.h1->tp_len;
2583 break;
2584 }
2585 if (unlikely(tp_len > size_max)) {
2586 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2587 return -EMSGSIZE;
2588 }
2589
2590 if (unlikely(po->tp_tx_has_off)) {
2591 int off_min, off_max;
2592
2593 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2594 off_max = po->tx_ring.frame_size - tp_len;
2595 if (po->sk.sk_type == SOCK_DGRAM) {
2596 switch (po->tp_version) {
2597 case TPACKET_V3:
2598 off = ph.h3->tp_net;
2599 break;
2600 case TPACKET_V2:
2601 off = ph.h2->tp_net;
2602 break;
2603 default:
2604 off = ph.h1->tp_net;
2605 break;
2606 }
2607 } else {
2608 switch (po->tp_version) {
2609 case TPACKET_V3:
2610 off = ph.h3->tp_mac;
2611 break;
2612 case TPACKET_V2:
2613 off = ph.h2->tp_mac;
2614 break;
2615 default:
2616 off = ph.h1->tp_mac;
2617 break;
2618 }
2619 }
2620 if (unlikely((off < off_min) || (off_max < off)))
2621 return -EINVAL;
2622 } else {
2623 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2624 }
2625
2626 *data = frame + off;
2627 return tp_len;
2628}
2629
2630static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2631{
2632 struct sk_buff *skb = NULL;
2633 struct net_device *dev;
2634 struct virtio_net_hdr *vnet_hdr = NULL;
2635 struct sockcm_cookie sockc;
2636 __be16 proto;
2637 int err, reserve = 0;
2638 void *ph;
2639 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2640 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2641 unsigned char *addr = NULL;
2642 int tp_len, size_max;
2643 void *data;
2644 int len_sum = 0;
2645 int status = TP_STATUS_AVAILABLE;
2646 int hlen, tlen, copylen = 0;
2647 long timeo = 0;
2648
2649 mutex_lock(&po->pg_vec_lock);
2650
2651 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2652 * we need to confirm it under protection of pg_vec_lock.
2653 */
2654 if (unlikely(!po->tx_ring.pg_vec)) {
2655 err = -EBUSY;
2656 goto out;
2657 }
2658 if (likely(saddr == NULL)) {
2659 dev = packet_cached_dev_get(po);
2660 proto = po->num;
2661 } else {
2662 err = -EINVAL;
2663 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2664 goto out;
2665 if (msg->msg_namelen < (saddr->sll_halen
2666 + offsetof(struct sockaddr_ll,
2667 sll_addr)))
2668 goto out;
2669 proto = saddr->sll_protocol;
2670 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2671 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2672 if (dev && msg->msg_namelen < dev->addr_len +
2673 offsetof(struct sockaddr_ll, sll_addr))
2674 goto out_put;
2675 addr = saddr->sll_addr;
2676 }
2677 }
2678
2679 err = -ENXIO;
2680 if (unlikely(dev == NULL))
2681 goto out;
2682 err = -ENETDOWN;
2683 if (unlikely(!(dev->flags & IFF_UP)))
2684 goto out_put;
2685
2686 sockcm_init(&sockc, &po->sk);
2687 if (msg->msg_controllen) {
2688 err = sock_cmsg_send(&po->sk, msg, &sockc);
2689 if (unlikely(err))
2690 goto out_put;
2691 }
2692
2693 if (po->sk.sk_socket->type == SOCK_RAW)
2694 reserve = dev->hard_header_len;
2695 size_max = po->tx_ring.frame_size
2696 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2697
2698 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2699 size_max = dev->mtu + reserve + VLAN_HLEN;
2700
2701 reinit_completion(&po->skb_completion);
2702
2703 do {
2704 ph = packet_current_frame(po, &po->tx_ring,
2705 TP_STATUS_SEND_REQUEST);
2706 if (unlikely(ph == NULL)) {
2707 if (need_wait && skb) {
2708 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2709 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2710 if (timeo <= 0) {
2711 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2712 goto out_put;
2713 }
2714 }
2715 /* check for additional frames */
2716 continue;
2717 }
2718
2719 skb = NULL;
2720 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2721 if (tp_len < 0)
2722 goto tpacket_error;
2723
2724 status = TP_STATUS_SEND_REQUEST;
2725 hlen = LL_RESERVED_SPACE(dev);
2726 tlen = dev->needed_tailroom;
2727 if (po->has_vnet_hdr) {
2728 vnet_hdr = data;
2729 data += sizeof(*vnet_hdr);
2730 tp_len -= sizeof(*vnet_hdr);
2731 if (tp_len < 0 ||
2732 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2733 tp_len = -EINVAL;
2734 goto tpacket_error;
2735 }
2736 copylen = __virtio16_to_cpu(vio_le(),
2737 vnet_hdr->hdr_len);
2738 }
2739 copylen = max_t(int, copylen, dev->hard_header_len);
2740 skb = sock_alloc_send_skb(&po->sk,
2741 hlen + tlen + sizeof(struct sockaddr_ll) +
2742 (copylen - dev->hard_header_len),
2743 !need_wait, &err);
2744
2745 if (unlikely(skb == NULL)) {
2746 /* we assume the socket was initially writeable ... */
2747 if (likely(len_sum > 0))
2748 err = len_sum;
2749 goto out_status;
2750 }
2751 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2752 addr, hlen, copylen, &sockc);
2753 if (likely(tp_len >= 0) &&
2754 tp_len > dev->mtu + reserve &&
2755 !po->has_vnet_hdr &&
2756 !packet_extra_vlan_len_allowed(dev, skb))
2757 tp_len = -EMSGSIZE;
2758
2759 if (unlikely(tp_len < 0)) {
2760tpacket_error:
2761 if (po->tp_loss) {
2762 __packet_set_status(po, ph,
2763 TP_STATUS_AVAILABLE);
2764 packet_increment_head(&po->tx_ring);
2765 kfree_skb(skb);
2766 continue;
2767 } else {
2768 status = TP_STATUS_WRONG_FORMAT;
2769 err = tp_len;
2770 goto out_status;
2771 }
2772 }
2773
2774 if (po->has_vnet_hdr) {
2775 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2776 tp_len = -EINVAL;
2777 goto tpacket_error;
2778 }
2779 virtio_net_hdr_set_proto(skb, vnet_hdr);
2780 }
2781
2782 skb->destructor = tpacket_destruct_skb;
2783 __packet_set_status(po, ph, TP_STATUS_SENDING);
2784 packet_inc_pending(&po->tx_ring);
2785
2786 status = TP_STATUS_SEND_REQUEST;
2787 err = po->xmit(skb);
2788 if (unlikely(err > 0)) {
2789 err = net_xmit_errno(err);
2790 if (err && __packet_get_status(po, ph) ==
2791 TP_STATUS_AVAILABLE) {
2792 /* skb was destructed already */
2793 skb = NULL;
2794 goto out_status;
2795 }
2796 /*
2797 * skb was dropped but not destructed yet;
2798 * let's treat it like congestion or err < 0
2799 */
2800 err = 0;
2801 }
2802 packet_increment_head(&po->tx_ring);
2803 len_sum += tp_len;
2804 } while (likely((ph != NULL) ||
2805 /* Note: packet_read_pending() might be slow if we have
2806 * to call it as it's per_cpu variable, but in fast-path
2807 * we already short-circuit the loop with the first
2808 * condition, and luckily don't have to go that path
2809 * anyway.
2810 */
2811 (need_wait && packet_read_pending(&po->tx_ring))));
2812
2813 err = len_sum;
2814 goto out_put;
2815
2816out_status:
2817 __packet_set_status(po, ph, status);
2818 kfree_skb(skb);
2819out_put:
2820 dev_put(dev);
2821out:
2822 mutex_unlock(&po->pg_vec_lock);
2823 return err;
2824}
2825
2826static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2827 size_t reserve, size_t len,
2828 size_t linear, int noblock,
2829 int *err)
2830{
2831 struct sk_buff *skb;
2832
2833 /* Under a page? Don't bother with paged skb. */
2834 if (prepad + len < PAGE_SIZE || !linear)
2835 linear = len;
2836
2837 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2838 err, 0);
2839 if (!skb)
2840 return NULL;
2841
2842 skb_reserve(skb, reserve);
2843 skb_put(skb, linear);
2844 skb->data_len = len - linear;
2845 skb->len += len - linear;
2846
2847 return skb;
2848}
2849
2850static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2851{
2852 struct sock *sk = sock->sk;
2853 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2854 struct sk_buff *skb;
2855 struct net_device *dev;
2856 __be16 proto;
2857 unsigned char *addr = NULL;
2858 int err, reserve = 0;
2859 struct sockcm_cookie sockc;
2860 struct virtio_net_hdr vnet_hdr = { 0 };
2861 int offset = 0;
2862 struct packet_sock *po = pkt_sk(sk);
2863 bool has_vnet_hdr = false;
2864 int hlen, tlen, linear;
2865 int extra_len = 0;
2866
2867 /*
2868 * Get and verify the address.
2869 */
2870
2871 if (likely(saddr == NULL)) {
2872 dev = packet_cached_dev_get(po);
2873 proto = po->num;
2874 } else {
2875 err = -EINVAL;
2876 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2877 goto out;
2878 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2879 goto out;
2880 proto = saddr->sll_protocol;
2881 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2882 if (sock->type == SOCK_DGRAM) {
2883 if (dev && msg->msg_namelen < dev->addr_len +
2884 offsetof(struct sockaddr_ll, sll_addr))
2885 goto out_unlock;
2886 addr = saddr->sll_addr;
2887 }
2888 }
2889
2890 err = -ENXIO;
2891 if (unlikely(dev == NULL))
2892 goto out_unlock;
2893 err = -ENETDOWN;
2894 if (unlikely(!(dev->flags & IFF_UP)))
2895 goto out_unlock;
2896
2897 sockcm_init(&sockc, sk);
2898 sockc.mark = sk->sk_mark;
2899 if (msg->msg_controllen) {
2900 err = sock_cmsg_send(sk, msg, &sockc);
2901 if (unlikely(err))
2902 goto out_unlock;
2903 }
2904
2905 if (sock->type == SOCK_RAW)
2906 reserve = dev->hard_header_len;
2907 if (po->has_vnet_hdr) {
2908 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2909 if (err)
2910 goto out_unlock;
2911 has_vnet_hdr = true;
2912 }
2913
2914 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2915 if (!netif_supports_nofcs(dev)) {
2916 err = -EPROTONOSUPPORT;
2917 goto out_unlock;
2918 }
2919 extra_len = 4; /* We're doing our own CRC */
2920 }
2921
2922 err = -EMSGSIZE;
2923 if (!vnet_hdr.gso_type &&
2924 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2925 goto out_unlock;
2926
2927 err = -ENOBUFS;
2928 hlen = LL_RESERVED_SPACE(dev);
2929 tlen = dev->needed_tailroom;
2930 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2931 linear = max(linear, min_t(int, len, dev->hard_header_len));
2932 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2933 msg->msg_flags & MSG_DONTWAIT, &err);
2934 if (skb == NULL)
2935 goto out_unlock;
2936
2937 skb_reset_network_header(skb);
2938
2939 err = -EINVAL;
2940 if (sock->type == SOCK_DGRAM) {
2941 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2942 if (unlikely(offset < 0))
2943 goto out_free;
2944 } else if (reserve) {
2945 skb_reserve(skb, -reserve);
2946 if (len < reserve + sizeof(struct ipv6hdr) &&
2947 dev->min_header_len != dev->hard_header_len)
2948 skb_reset_network_header(skb);
2949 }
2950
2951 /* Returns -EFAULT on error */
2952 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2953 if (err)
2954 goto out_free;
2955
2956 if (sock->type == SOCK_RAW &&
2957 !dev_validate_header(dev, skb->data, len)) {
2958 err = -EINVAL;
2959 goto out_free;
2960 }
2961
2962 skb_setup_tx_timestamp(skb, sockc.tsflags);
2963
2964 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2965 !packet_extra_vlan_len_allowed(dev, skb)) {
2966 err = -EMSGSIZE;
2967 goto out_free;
2968 }
2969
2970 skb->protocol = proto;
2971 skb->dev = dev;
2972 skb->priority = sk->sk_priority;
2973 skb->mark = sockc.mark;
2974 skb->tstamp = sockc.transmit_time;
2975
2976 if (has_vnet_hdr) {
2977 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2978 if (err)
2979 goto out_free;
2980 len += sizeof(vnet_hdr);
2981 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2982 }
2983
2984 packet_parse_headers(skb, sock);
2985
2986 if (unlikely(extra_len == 4))
2987 skb->no_fcs = 1;
2988
2989 err = po->xmit(skb);
2990 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2991 goto out_unlock;
2992
2993 dev_put(dev);
2994
2995 return len;
2996
2997out_free:
2998 kfree_skb(skb);
2999out_unlock:
3000 if (dev)
3001 dev_put(dev);
3002out:
3003 return err;
3004}
3005
3006static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3007{
3008 struct sock *sk = sock->sk;
3009 struct packet_sock *po = pkt_sk(sk);
3010
3011 if (po->tx_ring.pg_vec)
3012 return tpacket_snd(po, msg);
3013 else
3014 return packet_snd(sock, msg, len);
3015}
3016
3017/*
3018 * Close a PACKET socket. This is fairly simple. We immediately go
3019 * to 'closed' state and remove our protocol entry in the device list.
3020 */
3021
3022static int packet_release(struct socket *sock)
3023{
3024 struct sock *sk = sock->sk;
3025 struct packet_sock *po;
3026 struct packet_fanout *f;
3027 struct net *net;
3028 union tpacket_req_u req_u;
3029
3030 if (!sk)
3031 return 0;
3032
3033 net = sock_net(sk);
3034 po = pkt_sk(sk);
3035
3036 mutex_lock(&net->packet.sklist_lock);
3037 sk_del_node_init_rcu(sk);
3038 mutex_unlock(&net->packet.sklist_lock);
3039
3040 preempt_disable();
3041 sock_prot_inuse_add(net, sk->sk_prot, -1);
3042 preempt_enable();
3043
3044 spin_lock(&po->bind_lock);
3045 unregister_prot_hook(sk, false);
3046 packet_cached_dev_reset(po);
3047
3048 if (po->prot_hook.dev) {
3049 dev_put(po->prot_hook.dev);
3050 po->prot_hook.dev = NULL;
3051 }
3052 spin_unlock(&po->bind_lock);
3053
3054 packet_flush_mclist(sk);
3055
3056 lock_sock(sk);
3057 if (po->rx_ring.pg_vec) {
3058 memset(&req_u, 0, sizeof(req_u));
3059 packet_set_ring(sk, &req_u, 1, 0);
3060 }
3061
3062 if (po->tx_ring.pg_vec) {
3063 memset(&req_u, 0, sizeof(req_u));
3064 packet_set_ring(sk, &req_u, 1, 1);
3065 }
3066 release_sock(sk);
3067
3068 f = fanout_release(sk);
3069
3070 synchronize_net();
3071
3072 kfree(po->rollover);
3073 if (f) {
3074 fanout_release_data(f);
3075 kfree(f);
3076 }
3077 /*
3078 * Now the socket is dead. No more input will appear.
3079 */
3080 sock_orphan(sk);
3081 sock->sk = NULL;
3082
3083 /* Purge queues */
3084
3085 skb_queue_purge(&sk->sk_receive_queue);
3086 packet_free_pending(po);
3087 sk_refcnt_debug_release(sk);
3088
3089 sock_put(sk);
3090 return 0;
3091}
3092
3093/*
3094 * Attach a packet hook.
3095 */
3096
3097static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3098 __be16 proto)
3099{
3100 struct packet_sock *po = pkt_sk(sk);
3101 struct net_device *dev_curr;
3102 __be16 proto_curr;
3103 bool need_rehook;
3104 struct net_device *dev = NULL;
3105 int ret = 0;
3106 bool unlisted = false;
3107
3108 lock_sock(sk);
3109 spin_lock(&po->bind_lock);
3110 rcu_read_lock();
3111
3112 if (po->fanout) {
3113 ret = -EINVAL;
3114 goto out_unlock;
3115 }
3116
3117 if (name) {
3118 dev = dev_get_by_name_rcu(sock_net(sk), name);
3119 if (!dev) {
3120 ret = -ENODEV;
3121 goto out_unlock;
3122 }
3123 } else if (ifindex) {
3124 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3125 if (!dev) {
3126 ret = -ENODEV;
3127 goto out_unlock;
3128 }
3129 }
3130
3131 if (dev)
3132 dev_hold(dev);
3133
3134 proto_curr = po->prot_hook.type;
3135 dev_curr = po->prot_hook.dev;
3136
3137 need_rehook = proto_curr != proto || dev_curr != dev;
3138
3139 if (need_rehook) {
3140 if (po->running) {
3141 rcu_read_unlock();
3142 /* prevents packet_notifier() from calling
3143 * register_prot_hook()
3144 */
3145 po->num = 0;
3146 __unregister_prot_hook(sk, true);
3147 rcu_read_lock();
3148 dev_curr = po->prot_hook.dev;
3149 if (dev)
3150 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3151 dev->ifindex);
3152 }
3153
3154 BUG_ON(po->running);
3155 po->num = proto;
3156 po->prot_hook.type = proto;
3157
3158 if (unlikely(unlisted)) {
3159 dev_put(dev);
3160 po->prot_hook.dev = NULL;
3161 po->ifindex = -1;
3162 packet_cached_dev_reset(po);
3163 } else {
3164 po->prot_hook.dev = dev;
3165 po->ifindex = dev ? dev->ifindex : 0;
3166 packet_cached_dev_assign(po, dev);
3167 }
3168 }
3169 if (dev_curr)
3170 dev_put(dev_curr);
3171
3172 if (proto == 0 || !need_rehook)
3173 goto out_unlock;
3174
3175 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3176 register_prot_hook(sk);
3177 } else {
3178 sk->sk_err = ENETDOWN;
3179 if (!sock_flag(sk, SOCK_DEAD))
3180 sk->sk_error_report(sk);
3181 }
3182
3183out_unlock:
3184 rcu_read_unlock();
3185 spin_unlock(&po->bind_lock);
3186 release_sock(sk);
3187 return ret;
3188}
3189
3190/*
3191 * Bind a packet socket to a device
3192 */
3193
3194static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3195 int addr_len)
3196{
3197 struct sock *sk = sock->sk;
3198 char name[sizeof(uaddr->sa_data) + 1];
3199
3200 /*
3201 * Check legality
3202 */
3203
3204 if (addr_len != sizeof(struct sockaddr))
3205 return -EINVAL;
3206 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3207 * zero-terminated.
3208 */
3209 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3210 name[sizeof(uaddr->sa_data)] = 0;
3211
3212 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3213}
3214
3215static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3216{
3217 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3218 struct sock *sk = sock->sk;
3219
3220 /*
3221 * Check legality
3222 */
3223
3224 if (addr_len < sizeof(struct sockaddr_ll))
3225 return -EINVAL;
3226 if (sll->sll_family != AF_PACKET)
3227 return -EINVAL;
3228
3229 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3230 sll->sll_protocol ? : pkt_sk(sk)->num);
3231}
3232
3233static struct proto packet_proto = {
3234 .name = "PACKET",
3235 .owner = THIS_MODULE,
3236 .obj_size = sizeof(struct packet_sock),
3237};
3238
3239/*
3240 * Create a packet of type SOCK_PACKET.
3241 */
3242
3243static int packet_create(struct net *net, struct socket *sock, int protocol,
3244 int kern)
3245{
3246 struct sock *sk;
3247 struct packet_sock *po;
3248 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3249 int err;
3250
3251 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3252 return -EPERM;
3253 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3254 sock->type != SOCK_PACKET)
3255 return -ESOCKTNOSUPPORT;
3256
3257 sock->state = SS_UNCONNECTED;
3258
3259 err = -ENOBUFS;
3260 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3261 if (sk == NULL)
3262 goto out;
3263
3264 sock->ops = &packet_ops;
3265 if (sock->type == SOCK_PACKET)
3266 sock->ops = &packet_ops_spkt;
3267
3268 sock_init_data(sock, sk);
3269
3270 po = pkt_sk(sk);
3271 init_completion(&po->skb_completion);
3272 sk->sk_family = PF_PACKET;
3273 po->num = proto;
3274 po->xmit = dev_queue_xmit;
3275
3276 err = packet_alloc_pending(po);
3277 if (err)
3278 goto out2;
3279
3280 packet_cached_dev_reset(po);
3281
3282 sk->sk_destruct = packet_sock_destruct;
3283 sk_refcnt_debug_inc(sk);
3284
3285 /*
3286 * Attach a protocol block
3287 */
3288
3289 spin_lock_init(&po->bind_lock);
3290 mutex_init(&po->pg_vec_lock);
3291 po->rollover = NULL;
3292 po->prot_hook.func = packet_rcv;
3293
3294 if (sock->type == SOCK_PACKET)
3295 po->prot_hook.func = packet_rcv_spkt;
3296
3297 po->prot_hook.af_packet_priv = sk;
3298
3299 if (proto) {
3300 po->prot_hook.type = proto;
3301 __register_prot_hook(sk);
3302 }
3303
3304 mutex_lock(&net->packet.sklist_lock);
3305 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3306 mutex_unlock(&net->packet.sklist_lock);
3307
3308 preempt_disable();
3309 sock_prot_inuse_add(net, &packet_proto, 1);
3310 preempt_enable();
3311
3312 return 0;
3313out2:
3314 sk_free(sk);
3315out:
3316 return err;
3317}
3318
3319/*
3320 * Pull a packet from our receive queue and hand it to the user.
3321 * If necessary we block.
3322 */
3323
3324static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3325 int flags)
3326{
3327 struct sock *sk = sock->sk;
3328 struct sk_buff *skb;
3329 int copied, err;
3330 int vnet_hdr_len = 0;
3331 unsigned int origlen = 0;
3332
3333 err = -EINVAL;
3334 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3335 goto out;
3336
3337#if 0
3338 /* What error should we return now? EUNATTACH? */
3339 if (pkt_sk(sk)->ifindex < 0)
3340 return -ENODEV;
3341#endif
3342
3343 if (flags & MSG_ERRQUEUE) {
3344 err = sock_recv_errqueue(sk, msg, len,
3345 SOL_PACKET, PACKET_TX_TIMESTAMP);
3346 goto out;
3347 }
3348
3349 /*
3350 * Call the generic datagram receiver. This handles all sorts
3351 * of horrible races and re-entrancy so we can forget about it
3352 * in the protocol layers.
3353 *
3354 * Now it will return ENETDOWN, if device have just gone down,
3355 * but then it will block.
3356 */
3357
3358 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3359
3360 /*
3361 * An error occurred so return it. Because skb_recv_datagram()
3362 * handles the blocking we don't see and worry about blocking
3363 * retries.
3364 */
3365
3366 if (skb == NULL)
3367 goto out;
3368
3369 packet_rcv_try_clear_pressure(pkt_sk(sk));
3370
3371 if (pkt_sk(sk)->has_vnet_hdr) {
3372 err = packet_rcv_vnet(msg, skb, &len);
3373 if (err)
3374 goto out_free;
3375 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3376 }
3377
3378 /* You lose any data beyond the buffer you gave. If it worries
3379 * a user program they can ask the device for its MTU
3380 * anyway.
3381 */
3382 copied = skb->len;
3383 if (copied > len) {
3384 copied = len;
3385 msg->msg_flags |= MSG_TRUNC;
3386 }
3387
3388 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3389 if (err)
3390 goto out_free;
3391
3392 if (sock->type != SOCK_PACKET) {
3393 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3394
3395 /* Original length was stored in sockaddr_ll fields */
3396 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3397 sll->sll_family = AF_PACKET;
3398 sll->sll_protocol = skb->protocol;
3399 }
3400
3401 sock_recv_ts_and_drops(msg, sk, skb);
3402
3403 if (msg->msg_name) {
3404 int copy_len;
3405
3406 /* If the address length field is there to be filled
3407 * in, we fill it in now.
3408 */
3409 if (sock->type == SOCK_PACKET) {
3410 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3411 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3412 copy_len = msg->msg_namelen;
3413 } else {
3414 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3415
3416 msg->msg_namelen = sll->sll_halen +
3417 offsetof(struct sockaddr_ll, sll_addr);
3418 copy_len = msg->msg_namelen;
3419 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3420 memset(msg->msg_name +
3421 offsetof(struct sockaddr_ll, sll_addr),
3422 0, sizeof(sll->sll_addr));
3423 msg->msg_namelen = sizeof(struct sockaddr_ll);
3424 }
3425 }
3426 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3427 }
3428
3429 if (pkt_sk(sk)->auxdata) {
3430 struct tpacket_auxdata aux;
3431
3432 aux.tp_status = TP_STATUS_USER;
3433 if (skb->ip_summed == CHECKSUM_PARTIAL)
3434 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3435 else if (skb->pkt_type != PACKET_OUTGOING &&
3436 (skb->ip_summed == CHECKSUM_COMPLETE ||
3437 skb_csum_unnecessary(skb)))
3438 aux.tp_status |= TP_STATUS_CSUM_VALID;
3439
3440 aux.tp_len = origlen;
3441 aux.tp_snaplen = skb->len;
3442 aux.tp_mac = 0;
3443 aux.tp_net = skb_network_offset(skb);
3444 if (skb_vlan_tag_present(skb)) {
3445 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3446 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3447 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3448 } else {
3449 aux.tp_vlan_tci = 0;
3450 aux.tp_vlan_tpid = 0;
3451 }
3452 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3453 }
3454
3455 /*
3456 * Free or return the buffer as appropriate. Again this
3457 * hides all the races and re-entrancy issues from us.
3458 */
3459 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3460
3461out_free:
3462 skb_free_datagram(sk, skb);
3463out:
3464 return err;
3465}
3466
3467static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3468 int peer)
3469{
3470 struct net_device *dev;
3471 struct sock *sk = sock->sk;
3472
3473 if (peer)
3474 return -EOPNOTSUPP;
3475
3476 uaddr->sa_family = AF_PACKET;
3477 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3478 rcu_read_lock();
3479 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3480 if (dev)
3481 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3482 rcu_read_unlock();
3483
3484 return sizeof(*uaddr);
3485}
3486
3487static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3488 int peer)
3489{
3490 struct net_device *dev;
3491 struct sock *sk = sock->sk;
3492 struct packet_sock *po = pkt_sk(sk);
3493 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3494
3495 if (peer)
3496 return -EOPNOTSUPP;
3497
3498 sll->sll_family = AF_PACKET;
3499 sll->sll_ifindex = po->ifindex;
3500 sll->sll_protocol = po->num;
3501 sll->sll_pkttype = 0;
3502 rcu_read_lock();
3503 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3504 if (dev) {
3505 sll->sll_hatype = dev->type;
3506 sll->sll_halen = dev->addr_len;
3507 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3508 } else {
3509 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3510 sll->sll_halen = 0;
3511 }
3512 rcu_read_unlock();
3513
3514 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3515}
3516
3517static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3518 int what)
3519{
3520 switch (i->type) {
3521 case PACKET_MR_MULTICAST:
3522 if (i->alen != dev->addr_len)
3523 return -EINVAL;
3524 if (what > 0)
3525 return dev_mc_add(dev, i->addr);
3526 else
3527 return dev_mc_del(dev, i->addr);
3528 break;
3529 case PACKET_MR_PROMISC:
3530 return dev_set_promiscuity(dev, what);
3531 case PACKET_MR_ALLMULTI:
3532 return dev_set_allmulti(dev, what);
3533 case PACKET_MR_UNICAST:
3534 if (i->alen != dev->addr_len)
3535 return -EINVAL;
3536 if (what > 0)
3537 return dev_uc_add(dev, i->addr);
3538 else
3539 return dev_uc_del(dev, i->addr);
3540 break;
3541 default:
3542 break;
3543 }
3544 return 0;
3545}
3546
3547static void packet_dev_mclist_delete(struct net_device *dev,
3548 struct packet_mclist **mlp)
3549{
3550 struct packet_mclist *ml;
3551
3552 while ((ml = *mlp) != NULL) {
3553 if (ml->ifindex == dev->ifindex) {
3554 packet_dev_mc(dev, ml, -1);
3555 *mlp = ml->next;
3556 kfree(ml);
3557 } else
3558 mlp = &ml->next;
3559 }
3560}
3561
3562static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3563{
3564 struct packet_sock *po = pkt_sk(sk);
3565 struct packet_mclist *ml, *i;
3566 struct net_device *dev;
3567 int err;
3568
3569 rtnl_lock();
3570
3571 err = -ENODEV;
3572 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3573 if (!dev)
3574 goto done;
3575
3576 err = -EINVAL;
3577 if (mreq->mr_alen > dev->addr_len)
3578 goto done;
3579
3580 err = -ENOBUFS;
3581 i = kmalloc(sizeof(*i), GFP_KERNEL);
3582 if (i == NULL)
3583 goto done;
3584
3585 err = 0;
3586 for (ml = po->mclist; ml; ml = ml->next) {
3587 if (ml->ifindex == mreq->mr_ifindex &&
3588 ml->type == mreq->mr_type &&
3589 ml->alen == mreq->mr_alen &&
3590 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3591 ml->count++;
3592 /* Free the new element ... */
3593 kfree(i);
3594 goto done;
3595 }
3596 }
3597
3598 i->type = mreq->mr_type;
3599 i->ifindex = mreq->mr_ifindex;
3600 i->alen = mreq->mr_alen;
3601 memcpy(i->addr, mreq->mr_address, i->alen);
3602 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3603 i->count = 1;
3604 i->next = po->mclist;
3605 po->mclist = i;
3606 err = packet_dev_mc(dev, i, 1);
3607 if (err) {
3608 po->mclist = i->next;
3609 kfree(i);
3610 }
3611
3612done:
3613 rtnl_unlock();
3614 return err;
3615}
3616
3617static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3618{
3619 struct packet_mclist *ml, **mlp;
3620
3621 rtnl_lock();
3622
3623 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3624 if (ml->ifindex == mreq->mr_ifindex &&
3625 ml->type == mreq->mr_type &&
3626 ml->alen == mreq->mr_alen &&
3627 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3628 if (--ml->count == 0) {
3629 struct net_device *dev;
3630 *mlp = ml->next;
3631 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3632 if (dev)
3633 packet_dev_mc(dev, ml, -1);
3634 kfree(ml);
3635 }
3636 break;
3637 }
3638 }
3639 rtnl_unlock();
3640 return 0;
3641}
3642
3643static void packet_flush_mclist(struct sock *sk)
3644{
3645 struct packet_sock *po = pkt_sk(sk);
3646 struct packet_mclist *ml;
3647
3648 if (!po->mclist)
3649 return;
3650
3651 rtnl_lock();
3652 while ((ml = po->mclist) != NULL) {
3653 struct net_device *dev;
3654
3655 po->mclist = ml->next;
3656 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3657 if (dev != NULL)
3658 packet_dev_mc(dev, ml, -1);
3659 kfree(ml);
3660 }
3661 rtnl_unlock();
3662}
3663
3664static int
3665packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3666 unsigned int optlen)
3667{
3668 struct sock *sk = sock->sk;
3669 struct packet_sock *po = pkt_sk(sk);
3670 int ret;
3671
3672 if (level != SOL_PACKET)
3673 return -ENOPROTOOPT;
3674
3675 switch (optname) {
3676 case PACKET_ADD_MEMBERSHIP:
3677 case PACKET_DROP_MEMBERSHIP:
3678 {
3679 struct packet_mreq_max mreq;
3680 int len = optlen;
3681 memset(&mreq, 0, sizeof(mreq));
3682 if (len < sizeof(struct packet_mreq))
3683 return -EINVAL;
3684 if (len > sizeof(mreq))
3685 len = sizeof(mreq);
3686 if (copy_from_sockptr(&mreq, optval, len))
3687 return -EFAULT;
3688 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3689 return -EINVAL;
3690 if (optname == PACKET_ADD_MEMBERSHIP)
3691 ret = packet_mc_add(sk, &mreq);
3692 else
3693 ret = packet_mc_drop(sk, &mreq);
3694 return ret;
3695 }
3696
3697 case PACKET_RX_RING:
3698 case PACKET_TX_RING:
3699 {
3700 union tpacket_req_u req_u;
3701 int len;
3702
3703 lock_sock(sk);
3704 switch (po->tp_version) {
3705 case TPACKET_V1:
3706 case TPACKET_V2:
3707 len = sizeof(req_u.req);
3708 break;
3709 case TPACKET_V3:
3710 default:
3711 len = sizeof(req_u.req3);
3712 break;
3713 }
3714 if (optlen < len) {
3715 ret = -EINVAL;
3716 } else {
3717 if (copy_from_sockptr(&req_u.req, optval, len))
3718 ret = -EFAULT;
3719 else
3720 ret = packet_set_ring(sk, &req_u, 0,
3721 optname == PACKET_TX_RING);
3722 }
3723 release_sock(sk);
3724 return ret;
3725 }
3726 case PACKET_COPY_THRESH:
3727 {
3728 int val;
3729
3730 if (optlen != sizeof(val))
3731 return -EINVAL;
3732 if (copy_from_sockptr(&val, optval, sizeof(val)))
3733 return -EFAULT;
3734
3735 pkt_sk(sk)->copy_thresh = val;
3736 return 0;
3737 }
3738 case PACKET_VERSION:
3739 {
3740 int val;
3741
3742 if (optlen != sizeof(val))
3743 return -EINVAL;
3744 if (copy_from_sockptr(&val, optval, sizeof(val)))
3745 return -EFAULT;
3746 switch (val) {
3747 case TPACKET_V1:
3748 case TPACKET_V2:
3749 case TPACKET_V3:
3750 break;
3751 default:
3752 return -EINVAL;
3753 }
3754 lock_sock(sk);
3755 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3756 ret = -EBUSY;
3757 } else {
3758 po->tp_version = val;
3759 ret = 0;
3760 }
3761 release_sock(sk);
3762 return ret;
3763 }
3764 case PACKET_RESERVE:
3765 {
3766 unsigned int val;
3767
3768 if (optlen != sizeof(val))
3769 return -EINVAL;
3770 if (copy_from_sockptr(&val, optval, sizeof(val)))
3771 return -EFAULT;
3772 if (val > INT_MAX)
3773 return -EINVAL;
3774 lock_sock(sk);
3775 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3776 ret = -EBUSY;
3777 } else {
3778 po->tp_reserve = val;
3779 ret = 0;
3780 }
3781 release_sock(sk);
3782 return ret;
3783 }
3784 case PACKET_LOSS:
3785 {
3786 unsigned int val;
3787
3788 if (optlen != sizeof(val))
3789 return -EINVAL;
3790 if (copy_from_sockptr(&val, optval, sizeof(val)))
3791 return -EFAULT;
3792
3793 lock_sock(sk);
3794 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3795 ret = -EBUSY;
3796 } else {
3797 po->tp_loss = !!val;
3798 ret = 0;
3799 }
3800 release_sock(sk);
3801 return ret;
3802 }
3803 case PACKET_AUXDATA:
3804 {
3805 int val;
3806
3807 if (optlen < sizeof(val))
3808 return -EINVAL;
3809 if (copy_from_sockptr(&val, optval, sizeof(val)))
3810 return -EFAULT;
3811
3812 lock_sock(sk);
3813 po->auxdata = !!val;
3814 release_sock(sk);
3815 return 0;
3816 }
3817 case PACKET_ORIGDEV:
3818 {
3819 int val;
3820
3821 if (optlen < sizeof(val))
3822 return -EINVAL;
3823 if (copy_from_sockptr(&val, optval, sizeof(val)))
3824 return -EFAULT;
3825
3826 lock_sock(sk);
3827 po->origdev = !!val;
3828 release_sock(sk);
3829 return 0;
3830 }
3831 case PACKET_VNET_HDR:
3832 {
3833 int val;
3834
3835 if (sock->type != SOCK_RAW)
3836 return -EINVAL;
3837 if (optlen < sizeof(val))
3838 return -EINVAL;
3839 if (copy_from_sockptr(&val, optval, sizeof(val)))
3840 return -EFAULT;
3841
3842 lock_sock(sk);
3843 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3844 ret = -EBUSY;
3845 } else {
3846 po->has_vnet_hdr = !!val;
3847 ret = 0;
3848 }
3849 release_sock(sk);
3850 return ret;
3851 }
3852 case PACKET_TIMESTAMP:
3853 {
3854 int val;
3855
3856 if (optlen != sizeof(val))
3857 return -EINVAL;
3858 if (copy_from_sockptr(&val, optval, sizeof(val)))
3859 return -EFAULT;
3860
3861 po->tp_tstamp = val;
3862 return 0;
3863 }
3864 case PACKET_FANOUT:
3865 {
3866 int val;
3867
3868 if (optlen != sizeof(val))
3869 return -EINVAL;
3870 if (copy_from_sockptr(&val, optval, sizeof(val)))
3871 return -EFAULT;
3872
3873 return fanout_add(sk, val & 0xffff, val >> 16);
3874 }
3875 case PACKET_FANOUT_DATA:
3876 {
3877 if (!po->fanout)
3878 return -EINVAL;
3879
3880 return fanout_set_data(po, optval, optlen);
3881 }
3882 case PACKET_IGNORE_OUTGOING:
3883 {
3884 int val;
3885
3886 if (optlen != sizeof(val))
3887 return -EINVAL;
3888 if (copy_from_sockptr(&val, optval, sizeof(val)))
3889 return -EFAULT;
3890 if (val < 0 || val > 1)
3891 return -EINVAL;
3892
3893 po->prot_hook.ignore_outgoing = !!val;
3894 return 0;
3895 }
3896 case PACKET_TX_HAS_OFF:
3897 {
3898 unsigned int val;
3899
3900 if (optlen != sizeof(val))
3901 return -EINVAL;
3902 if (copy_from_sockptr(&val, optval, sizeof(val)))
3903 return -EFAULT;
3904
3905 lock_sock(sk);
3906 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3907 ret = -EBUSY;
3908 } else {
3909 po->tp_tx_has_off = !!val;
3910 ret = 0;
3911 }
3912 release_sock(sk);
3913 return 0;
3914 }
3915 case PACKET_QDISC_BYPASS:
3916 {
3917 int val;
3918
3919 if (optlen != sizeof(val))
3920 return -EINVAL;
3921 if (copy_from_sockptr(&val, optval, sizeof(val)))
3922 return -EFAULT;
3923
3924 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3925 return 0;
3926 }
3927 default:
3928 return -ENOPROTOOPT;
3929 }
3930}
3931
3932static int packet_getsockopt(struct socket *sock, int level, int optname,
3933 char __user *optval, int __user *optlen)
3934{
3935 int len;
3936 int val, lv = sizeof(val);
3937 struct sock *sk = sock->sk;
3938 struct packet_sock *po = pkt_sk(sk);
3939 void *data = &val;
3940 union tpacket_stats_u st;
3941 struct tpacket_rollover_stats rstats;
3942 int drops;
3943
3944 if (level != SOL_PACKET)
3945 return -ENOPROTOOPT;
3946
3947 if (get_user(len, optlen))
3948 return -EFAULT;
3949
3950 if (len < 0)
3951 return -EINVAL;
3952
3953 switch (optname) {
3954 case PACKET_STATISTICS:
3955 spin_lock_bh(&sk->sk_receive_queue.lock);
3956 memcpy(&st, &po->stats, sizeof(st));
3957 memset(&po->stats, 0, sizeof(po->stats));
3958 spin_unlock_bh(&sk->sk_receive_queue.lock);
3959 drops = atomic_xchg(&po->tp_drops, 0);
3960
3961 if (po->tp_version == TPACKET_V3) {
3962 lv = sizeof(struct tpacket_stats_v3);
3963 st.stats3.tp_drops = drops;
3964 st.stats3.tp_packets += drops;
3965 data = &st.stats3;
3966 } else {
3967 lv = sizeof(struct tpacket_stats);
3968 st.stats1.tp_drops = drops;
3969 st.stats1.tp_packets += drops;
3970 data = &st.stats1;
3971 }
3972
3973 break;
3974 case PACKET_AUXDATA:
3975 val = po->auxdata;
3976 break;
3977 case PACKET_ORIGDEV:
3978 val = po->origdev;
3979 break;
3980 case PACKET_VNET_HDR:
3981 val = po->has_vnet_hdr;
3982 break;
3983 case PACKET_VERSION:
3984 val = po->tp_version;
3985 break;
3986 case PACKET_HDRLEN:
3987 if (len > sizeof(int))
3988 len = sizeof(int);
3989 if (len < sizeof(int))
3990 return -EINVAL;
3991 if (copy_from_user(&val, optval, len))
3992 return -EFAULT;
3993 switch (val) {
3994 case TPACKET_V1:
3995 val = sizeof(struct tpacket_hdr);
3996 break;
3997 case TPACKET_V2:
3998 val = sizeof(struct tpacket2_hdr);
3999 break;
4000 case TPACKET_V3:
4001 val = sizeof(struct tpacket3_hdr);
4002 break;
4003 default:
4004 return -EINVAL;
4005 }
4006 break;
4007 case PACKET_RESERVE:
4008 val = po->tp_reserve;
4009 break;
4010 case PACKET_LOSS:
4011 val = po->tp_loss;
4012 break;
4013 case PACKET_TIMESTAMP:
4014 val = po->tp_tstamp;
4015 break;
4016 case PACKET_FANOUT:
4017 val = (po->fanout ?
4018 ((u32)po->fanout->id |
4019 ((u32)po->fanout->type << 16) |
4020 ((u32)po->fanout->flags << 24)) :
4021 0);
4022 break;
4023 case PACKET_IGNORE_OUTGOING:
4024 val = po->prot_hook.ignore_outgoing;
4025 break;
4026 case PACKET_ROLLOVER_STATS:
4027 if (!po->rollover)
4028 return -EINVAL;
4029 rstats.tp_all = atomic_long_read(&po->rollover->num);
4030 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4031 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4032 data = &rstats;
4033 lv = sizeof(rstats);
4034 break;
4035 case PACKET_TX_HAS_OFF:
4036 val = po->tp_tx_has_off;
4037 break;
4038 case PACKET_QDISC_BYPASS:
4039 val = packet_use_direct_xmit(po);
4040 break;
4041 default:
4042 return -ENOPROTOOPT;
4043 }
4044
4045 if (len > lv)
4046 len = lv;
4047 if (put_user(len, optlen))
4048 return -EFAULT;
4049 if (copy_to_user(optval, data, len))
4050 return -EFAULT;
4051 return 0;
4052}
4053
4054static int packet_notifier(struct notifier_block *this,
4055 unsigned long msg, void *ptr)
4056{
4057 struct sock *sk;
4058 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4059 struct net *net = dev_net(dev);
4060
4061 rcu_read_lock();
4062 sk_for_each_rcu(sk, &net->packet.sklist) {
4063 struct packet_sock *po = pkt_sk(sk);
4064
4065 switch (msg) {
4066 case NETDEV_UNREGISTER:
4067 if (po->mclist)
4068 packet_dev_mclist_delete(dev, &po->mclist);
4069 fallthrough;
4070
4071 case NETDEV_DOWN:
4072 if (dev->ifindex == po->ifindex) {
4073 spin_lock(&po->bind_lock);
4074 if (po->running) {
4075 __unregister_prot_hook(sk, false);
4076 sk->sk_err = ENETDOWN;
4077 if (!sock_flag(sk, SOCK_DEAD))
4078 sk->sk_error_report(sk);
4079 }
4080 if (msg == NETDEV_UNREGISTER) {
4081 packet_cached_dev_reset(po);
4082 po->ifindex = -1;
4083 if (po->prot_hook.dev)
4084 dev_put(po->prot_hook.dev);
4085 po->prot_hook.dev = NULL;
4086 }
4087 spin_unlock(&po->bind_lock);
4088 }
4089 break;
4090 case NETDEV_UP:
4091 if (dev->ifindex == po->ifindex) {
4092 spin_lock(&po->bind_lock);
4093 if (po->num)
4094 register_prot_hook(sk);
4095 spin_unlock(&po->bind_lock);
4096 }
4097 break;
4098 }
4099 }
4100 rcu_read_unlock();
4101 return NOTIFY_DONE;
4102}
4103
4104
4105static int packet_ioctl(struct socket *sock, unsigned int cmd,
4106 unsigned long arg)
4107{
4108 struct sock *sk = sock->sk;
4109
4110 switch (cmd) {
4111 case SIOCOUTQ:
4112 {
4113 int amount = sk_wmem_alloc_get(sk);
4114
4115 return put_user(amount, (int __user *)arg);
4116 }
4117 case SIOCINQ:
4118 {
4119 struct sk_buff *skb;
4120 int amount = 0;
4121
4122 spin_lock_bh(&sk->sk_receive_queue.lock);
4123 skb = skb_peek(&sk->sk_receive_queue);
4124 if (skb)
4125 amount = skb->len;
4126 spin_unlock_bh(&sk->sk_receive_queue.lock);
4127 return put_user(amount, (int __user *)arg);
4128 }
4129#ifdef CONFIG_INET
4130 case SIOCADDRT:
4131 case SIOCDELRT:
4132 case SIOCDARP:
4133 case SIOCGARP:
4134 case SIOCSARP:
4135 case SIOCGIFADDR:
4136 case SIOCSIFADDR:
4137 case SIOCGIFBRDADDR:
4138 case SIOCSIFBRDADDR:
4139 case SIOCGIFNETMASK:
4140 case SIOCSIFNETMASK:
4141 case SIOCGIFDSTADDR:
4142 case SIOCSIFDSTADDR:
4143 case SIOCSIFFLAGS:
4144 return inet_dgram_ops.ioctl(sock, cmd, arg);
4145#endif
4146
4147 default:
4148 return -ENOIOCTLCMD;
4149 }
4150 return 0;
4151}
4152
4153static __poll_t packet_poll(struct file *file, struct socket *sock,
4154 poll_table *wait)
4155{
4156 struct sock *sk = sock->sk;
4157 struct packet_sock *po = pkt_sk(sk);
4158 __poll_t mask = datagram_poll(file, sock, wait);
4159
4160 spin_lock_bh(&sk->sk_receive_queue.lock);
4161 if (po->rx_ring.pg_vec) {
4162 if (!packet_previous_rx_frame(po, &po->rx_ring,
4163 TP_STATUS_KERNEL))
4164 mask |= EPOLLIN | EPOLLRDNORM;
4165 }
4166 packet_rcv_try_clear_pressure(po);
4167 spin_unlock_bh(&sk->sk_receive_queue.lock);
4168 spin_lock_bh(&sk->sk_write_queue.lock);
4169 if (po->tx_ring.pg_vec) {
4170 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4171 mask |= EPOLLOUT | EPOLLWRNORM;
4172 }
4173 spin_unlock_bh(&sk->sk_write_queue.lock);
4174 return mask;
4175}
4176
4177
4178/* Dirty? Well, I still did not learn better way to account
4179 * for user mmaps.
4180 */
4181
4182static void packet_mm_open(struct vm_area_struct *vma)
4183{
4184 struct file *file = vma->vm_file;
4185 struct socket *sock = file->private_data;
4186 struct sock *sk = sock->sk;
4187
4188 if (sk)
4189 atomic_inc(&pkt_sk(sk)->mapped);
4190}
4191
4192static void packet_mm_close(struct vm_area_struct *vma)
4193{
4194 struct file *file = vma->vm_file;
4195 struct socket *sock = file->private_data;
4196 struct sock *sk = sock->sk;
4197
4198 if (sk)
4199 atomic_dec(&pkt_sk(sk)->mapped);
4200}
4201
4202static const struct vm_operations_struct packet_mmap_ops = {
4203 .open = packet_mm_open,
4204 .close = packet_mm_close,
4205};
4206
4207static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4208 unsigned int len)
4209{
4210 int i;
4211
4212 for (i = 0; i < len; i++) {
4213 if (likely(pg_vec[i].buffer)) {
4214 if (is_vmalloc_addr(pg_vec[i].buffer))
4215 vfree(pg_vec[i].buffer);
4216 else
4217 free_pages((unsigned long)pg_vec[i].buffer,
4218 order);
4219 pg_vec[i].buffer = NULL;
4220 }
4221 }
4222 kfree(pg_vec);
4223}
4224
4225static char *alloc_one_pg_vec_page(unsigned long order)
4226{
4227 char *buffer;
4228 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4229 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4230
4231 buffer = (char *) __get_free_pages(gfp_flags, order);
4232 if (buffer)
4233 return buffer;
4234
4235 /* __get_free_pages failed, fall back to vmalloc */
4236 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4237 if (buffer)
4238 return buffer;
4239
4240 /* vmalloc failed, lets dig into swap here */
4241 gfp_flags &= ~__GFP_NORETRY;
4242 buffer = (char *) __get_free_pages(gfp_flags, order);
4243 if (buffer)
4244 return buffer;
4245
4246 /* complete and utter failure */
4247 return NULL;
4248}
4249
4250static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4251{
4252 unsigned int block_nr = req->tp_block_nr;
4253 struct pgv *pg_vec;
4254 int i;
4255
4256 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4257 if (unlikely(!pg_vec))
4258 goto out;
4259
4260 for (i = 0; i < block_nr; i++) {
4261 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4262 if (unlikely(!pg_vec[i].buffer))
4263 goto out_free_pgvec;
4264 }
4265
4266out:
4267 return pg_vec;
4268
4269out_free_pgvec:
4270 free_pg_vec(pg_vec, order, block_nr);
4271 pg_vec = NULL;
4272 goto out;
4273}
4274
4275static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4276 int closing, int tx_ring)
4277{
4278 struct pgv *pg_vec = NULL;
4279 struct packet_sock *po = pkt_sk(sk);
4280 unsigned long *rx_owner_map = NULL;
4281 int was_running, order = 0;
4282 struct packet_ring_buffer *rb;
4283 struct sk_buff_head *rb_queue;
4284 __be16 num;
4285 int err;
4286 /* Added to avoid minimal code churn */
4287 struct tpacket_req *req = &req_u->req;
4288
4289 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4290 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4291
4292 err = -EBUSY;
4293 if (!closing) {
4294 if (atomic_read(&po->mapped))
4295 goto out;
4296 if (packet_read_pending(rb))
4297 goto out;
4298 }
4299
4300 if (req->tp_block_nr) {
4301 unsigned int min_frame_size;
4302
4303 /* Sanity tests and some calculations */
4304 err = -EBUSY;
4305 if (unlikely(rb->pg_vec))
4306 goto out;
4307
4308 switch (po->tp_version) {
4309 case TPACKET_V1:
4310 po->tp_hdrlen = TPACKET_HDRLEN;
4311 break;
4312 case TPACKET_V2:
4313 po->tp_hdrlen = TPACKET2_HDRLEN;
4314 break;
4315 case TPACKET_V3:
4316 po->tp_hdrlen = TPACKET3_HDRLEN;
4317 break;
4318 }
4319
4320 err = -EINVAL;
4321 if (unlikely((int)req->tp_block_size <= 0))
4322 goto out;
4323 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4324 goto out;
4325 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4326 if (po->tp_version >= TPACKET_V3 &&
4327 req->tp_block_size <
4328 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4329 goto out;
4330 if (unlikely(req->tp_frame_size < min_frame_size))
4331 goto out;
4332 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4333 goto out;
4334
4335 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4336 if (unlikely(rb->frames_per_block == 0))
4337 goto out;
4338 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4339 goto out;
4340 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4341 req->tp_frame_nr))
4342 goto out;
4343
4344 err = -ENOMEM;
4345 order = get_order(req->tp_block_size);
4346 pg_vec = alloc_pg_vec(req, order);
4347 if (unlikely(!pg_vec))
4348 goto out;
4349 switch (po->tp_version) {
4350 case TPACKET_V3:
4351 /* Block transmit is not supported yet */
4352 if (!tx_ring) {
4353 init_prb_bdqc(po, rb, pg_vec, req_u);
4354 } else {
4355 struct tpacket_req3 *req3 = &req_u->req3;
4356
4357 if (req3->tp_retire_blk_tov ||
4358 req3->tp_sizeof_priv ||
4359 req3->tp_feature_req_word) {
4360 err = -EINVAL;
4361 goto out_free_pg_vec;
4362 }
4363 }
4364 break;
4365 default:
4366 if (!tx_ring) {
4367 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4368 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4369 if (!rx_owner_map)
4370 goto out_free_pg_vec;
4371 }
4372 break;
4373 }
4374 }
4375 /* Done */
4376 else {
4377 err = -EINVAL;
4378 if (unlikely(req->tp_frame_nr))
4379 goto out;
4380 }
4381
4382
4383 /* Detach socket from network */
4384 spin_lock(&po->bind_lock);
4385 was_running = po->running;
4386 num = po->num;
4387 if (was_running) {
4388 po->num = 0;
4389 __unregister_prot_hook(sk, false);
4390 }
4391 spin_unlock(&po->bind_lock);
4392
4393 synchronize_net();
4394
4395 err = -EBUSY;
4396 mutex_lock(&po->pg_vec_lock);
4397 if (closing || atomic_read(&po->mapped) == 0) {
4398 err = 0;
4399 spin_lock_bh(&rb_queue->lock);
4400 swap(rb->pg_vec, pg_vec);
4401 if (po->tp_version <= TPACKET_V2)
4402 swap(rb->rx_owner_map, rx_owner_map);
4403 rb->frame_max = (req->tp_frame_nr - 1);
4404 rb->head = 0;
4405 rb->frame_size = req->tp_frame_size;
4406 spin_unlock_bh(&rb_queue->lock);
4407
4408 swap(rb->pg_vec_order, order);
4409 swap(rb->pg_vec_len, req->tp_block_nr);
4410
4411 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4412 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4413 tpacket_rcv : packet_rcv;
4414 skb_queue_purge(rb_queue);
4415 if (atomic_read(&po->mapped))
4416 pr_err("packet_mmap: vma is busy: %d\n",
4417 atomic_read(&po->mapped));
4418 }
4419 mutex_unlock(&po->pg_vec_lock);
4420
4421 spin_lock(&po->bind_lock);
4422 if (was_running) {
4423 po->num = num;
4424 register_prot_hook(sk);
4425 }
4426 spin_unlock(&po->bind_lock);
4427 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4428 /* Because we don't support block-based V3 on tx-ring */
4429 if (!tx_ring)
4430 prb_shutdown_retire_blk_timer(po, rb_queue);
4431 }
4432
4433out_free_pg_vec:
4434 bitmap_free(rx_owner_map);
4435 if (pg_vec)
4436 free_pg_vec(pg_vec, order, req->tp_block_nr);
4437out:
4438 return err;
4439}
4440
4441static int packet_mmap(struct file *file, struct socket *sock,
4442 struct vm_area_struct *vma)
4443{
4444 struct sock *sk = sock->sk;
4445 struct packet_sock *po = pkt_sk(sk);
4446 unsigned long size, expected_size;
4447 struct packet_ring_buffer *rb;
4448 unsigned long start;
4449 int err = -EINVAL;
4450 int i;
4451
4452 if (vma->vm_pgoff)
4453 return -EINVAL;
4454
4455 mutex_lock(&po->pg_vec_lock);
4456
4457 expected_size = 0;
4458 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4459 if (rb->pg_vec) {
4460 expected_size += rb->pg_vec_len
4461 * rb->pg_vec_pages
4462 * PAGE_SIZE;
4463 }
4464 }
4465
4466 if (expected_size == 0)
4467 goto out;
4468
4469 size = vma->vm_end - vma->vm_start;
4470 if (size != expected_size)
4471 goto out;
4472
4473 start = vma->vm_start;
4474 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4475 if (rb->pg_vec == NULL)
4476 continue;
4477
4478 for (i = 0; i < rb->pg_vec_len; i++) {
4479 struct page *page;
4480 void *kaddr = rb->pg_vec[i].buffer;
4481 int pg_num;
4482
4483 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4484 page = pgv_to_page(kaddr);
4485 err = vm_insert_page(vma, start, page);
4486 if (unlikely(err))
4487 goto out;
4488 start += PAGE_SIZE;
4489 kaddr += PAGE_SIZE;
4490 }
4491 }
4492 }
4493
4494 atomic_inc(&po->mapped);
4495 vma->vm_ops = &packet_mmap_ops;
4496 err = 0;
4497
4498out:
4499 mutex_unlock(&po->pg_vec_lock);
4500 return err;
4501}
4502
4503static const struct proto_ops packet_ops_spkt = {
4504 .family = PF_PACKET,
4505 .owner = THIS_MODULE,
4506 .release = packet_release,
4507 .bind = packet_bind_spkt,
4508 .connect = sock_no_connect,
4509 .socketpair = sock_no_socketpair,
4510 .accept = sock_no_accept,
4511 .getname = packet_getname_spkt,
4512 .poll = datagram_poll,
4513 .ioctl = packet_ioctl,
4514 .gettstamp = sock_gettstamp,
4515 .listen = sock_no_listen,
4516 .shutdown = sock_no_shutdown,
4517 .sendmsg = packet_sendmsg_spkt,
4518 .recvmsg = packet_recvmsg,
4519 .mmap = sock_no_mmap,
4520 .sendpage = sock_no_sendpage,
4521};
4522
4523static const struct proto_ops packet_ops = {
4524 .family = PF_PACKET,
4525 .owner = THIS_MODULE,
4526 .release = packet_release,
4527 .bind = packet_bind,
4528 .connect = sock_no_connect,
4529 .socketpair = sock_no_socketpair,
4530 .accept = sock_no_accept,
4531 .getname = packet_getname,
4532 .poll = packet_poll,
4533 .ioctl = packet_ioctl,
4534 .gettstamp = sock_gettstamp,
4535 .listen = sock_no_listen,
4536 .shutdown = sock_no_shutdown,
4537 .setsockopt = packet_setsockopt,
4538 .getsockopt = packet_getsockopt,
4539 .sendmsg = packet_sendmsg,
4540 .recvmsg = packet_recvmsg,
4541 .mmap = packet_mmap,
4542 .sendpage = sock_no_sendpage,
4543};
4544
4545static const struct net_proto_family packet_family_ops = {
4546 .family = PF_PACKET,
4547 .create = packet_create,
4548 .owner = THIS_MODULE,
4549};
4550
4551static struct notifier_block packet_netdev_notifier = {
4552 .notifier_call = packet_notifier,
4553};
4554
4555#ifdef CONFIG_PROC_FS
4556
4557static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4558 __acquires(RCU)
4559{
4560 struct net *net = seq_file_net(seq);
4561
4562 rcu_read_lock();
4563 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4564}
4565
4566static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4567{
4568 struct net *net = seq_file_net(seq);
4569 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4570}
4571
4572static void packet_seq_stop(struct seq_file *seq, void *v)
4573 __releases(RCU)
4574{
4575 rcu_read_unlock();
4576}
4577
4578static int packet_seq_show(struct seq_file *seq, void *v)
4579{
4580 if (v == SEQ_START_TOKEN)
4581 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4582 else {
4583 struct sock *s = sk_entry(v);
4584 const struct packet_sock *po = pkt_sk(s);
4585
4586 seq_printf(seq,
4587 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4588 s,
4589 refcount_read(&s->sk_refcnt),
4590 s->sk_type,
4591 ntohs(po->num),
4592 po->ifindex,
4593 po->running,
4594 atomic_read(&s->sk_rmem_alloc),
4595 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4596 sock_i_ino(s));
4597 }
4598
4599 return 0;
4600}
4601
4602static const struct seq_operations packet_seq_ops = {
4603 .start = packet_seq_start,
4604 .next = packet_seq_next,
4605 .stop = packet_seq_stop,
4606 .show = packet_seq_show,
4607};
4608#endif
4609
4610static int __net_init packet_net_init(struct net *net)
4611{
4612 mutex_init(&net->packet.sklist_lock);
4613 INIT_HLIST_HEAD(&net->packet.sklist);
4614
4615 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4616 sizeof(struct seq_net_private)))
4617 return -ENOMEM;
4618
4619 return 0;
4620}
4621
4622static void __net_exit packet_net_exit(struct net *net)
4623{
4624 remove_proc_entry("packet", net->proc_net);
4625 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4626}
4627
4628static struct pernet_operations packet_net_ops = {
4629 .init = packet_net_init,
4630 .exit = packet_net_exit,
4631};
4632
4633
4634static void __exit packet_exit(void)
4635{
4636 unregister_netdevice_notifier(&packet_netdev_notifier);
4637 unregister_pernet_subsys(&packet_net_ops);
4638 sock_unregister(PF_PACKET);
4639 proto_unregister(&packet_proto);
4640}
4641
4642static int __init packet_init(void)
4643{
4644 int rc;
4645
4646 rc = proto_register(&packet_proto, 0);
4647 if (rc)
4648 goto out;
4649 rc = sock_register(&packet_family_ops);
4650 if (rc)
4651 goto out_proto;
4652 rc = register_pernet_subsys(&packet_net_ops);
4653 if (rc)
4654 goto out_sock;
4655 rc = register_netdevice_notifier(&packet_netdev_notifier);
4656 if (rc)
4657 goto out_pernet;
4658
4659 return 0;
4660
4661out_pernet:
4662 unregister_pernet_subsys(&packet_net_ops);
4663out_sock:
4664 sock_unregister(PF_PACKET);
4665out_proto:
4666 proto_unregister(&packet_proto);
4667out:
4668 return rc;
4669}
4670
4671module_init(packet_init);
4672module_exit(packet_exit);
4673MODULE_LICENSE("GPL");
4674MODULE_ALIAS_NETPROTO(PF_PACKET);
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PACKET - implements raw packet sockets.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 *
12 * Fixes:
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
41 * and packet_mreq.
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
44 * layer.
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
46 *
47 *
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
52 *
53 */
54
55#include <linux/types.h>
56#include <linux/mm.h>
57#include <linux/capability.h>
58#include <linux/fcntl.h>
59#include <linux/socket.h>
60#include <linux/in.h>
61#include <linux/inet.h>
62#include <linux/netdevice.h>
63#include <linux/if_packet.h>
64#include <linux/wireless.h>
65#include <linux/kernel.h>
66#include <linux/kmod.h>
67#include <linux/slab.h>
68#include <linux/vmalloc.h>
69#include <net/net_namespace.h>
70#include <net/ip.h>
71#include <net/protocol.h>
72#include <linux/skbuff.h>
73#include <net/sock.h>
74#include <linux/errno.h>
75#include <linux/timer.h>
76#include <asm/uaccess.h>
77#include <asm/ioctls.h>
78#include <asm/page.h>
79#include <asm/cacheflush.h>
80#include <asm/io.h>
81#include <linux/proc_fs.h>
82#include <linux/seq_file.h>
83#include <linux/poll.h>
84#include <linux/module.h>
85#include <linux/init.h>
86#include <linux/mutex.h>
87#include <linux/if_vlan.h>
88#include <linux/virtio_net.h>
89#include <linux/errqueue.h>
90#include <linux/net_tstamp.h>
91
92#ifdef CONFIG_INET
93#include <net/inet_common.h>
94#endif
95
96/*
97 Assumptions:
98 - if device has no dev->hard_header routine, it adds and removes ll header
99 inside itself. In this case ll header is invisible outside of device,
100 but higher levels still should reserve dev->hard_header_len.
101 Some devices are enough clever to reallocate skb, when header
102 will not fit to reserved space (tunnel), another ones are silly
103 (PPP).
104 - packet socket receives packets with pulled ll header,
105 so that SOCK_RAW should push it back.
106
107On receive:
108-----------
109
110Incoming, dev->hard_header!=NULL
111 mac_header -> ll header
112 data -> data
113
114Outgoing, dev->hard_header!=NULL
115 mac_header -> ll header
116 data -> ll header
117
118Incoming, dev->hard_header==NULL
119 mac_header -> UNKNOWN position. It is very likely, that it points to ll
120 header. PPP makes it, that is wrong, because introduce
121 assymetry between rx and tx paths.
122 data -> data
123
124Outgoing, dev->hard_header==NULL
125 mac_header -> data. ll header is still not built!
126 data -> data
127
128Resume
129 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
130
131
132On transmit:
133------------
134
135dev->hard_header != NULL
136 mac_header -> ll header
137 data -> ll header
138
139dev->hard_header == NULL (ll header is added by device, we cannot control it)
140 mac_header -> data
141 data -> data
142
143 We should set nh.raw on output to correct posistion,
144 packet classifier depends on it.
145 */
146
147/* Private packet socket structures. */
148
149struct packet_mclist {
150 struct packet_mclist *next;
151 int ifindex;
152 int count;
153 unsigned short type;
154 unsigned short alen;
155 unsigned char addr[MAX_ADDR_LEN];
156};
157/* identical to struct packet_mreq except it has
158 * a longer address field.
159 */
160struct packet_mreq_max {
161 int mr_ifindex;
162 unsigned short mr_type;
163 unsigned short mr_alen;
164 unsigned char mr_address[MAX_ADDR_LEN];
165};
166
167static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
168 int closing, int tx_ring);
169
170
171#define V3_ALIGNMENT (8)
172
173#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
174
175#define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
177
178/* kbdq - kernel block descriptor queue */
179struct tpacket_kbdq_core {
180 struct pgv *pkbdq;
181 unsigned int feature_req_word;
182 unsigned int hdrlen;
183 unsigned char reset_pending_on_curr_blk;
184 unsigned char delete_blk_timer;
185 unsigned short kactive_blk_num;
186 unsigned short blk_sizeof_priv;
187
188 /* last_kactive_blk_num:
189 * trick to see if user-space has caught up
190 * in order to avoid refreshing timer when every single pkt arrives.
191 */
192 unsigned short last_kactive_blk_num;
193
194 char *pkblk_start;
195 char *pkblk_end;
196 int kblk_size;
197 unsigned int knum_blocks;
198 uint64_t knxt_seq_num;
199 char *prev;
200 char *nxt_offset;
201 struct sk_buff *skb;
202
203 atomic_t blk_fill_in_prog;
204
205 /* Default is set to 8ms */
206#define DEFAULT_PRB_RETIRE_TOV (8)
207
208 unsigned short retire_blk_tov;
209 unsigned short version;
210 unsigned long tov_in_jiffies;
211
212 /* timer to retire an outstanding block */
213 struct timer_list retire_blk_timer;
214};
215
216#define PGV_FROM_VMALLOC 1
217struct pgv {
218 char *buffer;
219};
220
221struct packet_ring_buffer {
222 struct pgv *pg_vec;
223 unsigned int head;
224 unsigned int frames_per_block;
225 unsigned int frame_size;
226 unsigned int frame_max;
227
228 unsigned int pg_vec_order;
229 unsigned int pg_vec_pages;
230 unsigned int pg_vec_len;
231
232 struct tpacket_kbdq_core prb_bdqc;
233 atomic_t pending;
234};
235
236#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
237#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
238#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
239#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
240#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
241#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
242#define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
243
244struct packet_sock;
245static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
246
247static void *packet_previous_frame(struct packet_sock *po,
248 struct packet_ring_buffer *rb,
249 int status);
250static void packet_increment_head(struct packet_ring_buffer *buff);
251static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
252 struct tpacket_block_desc *);
253static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
254 struct packet_sock *);
255static void prb_retire_current_block(struct tpacket_kbdq_core *,
256 struct packet_sock *, unsigned int status);
257static int prb_queue_frozen(struct tpacket_kbdq_core *);
258static void prb_open_block(struct tpacket_kbdq_core *,
259 struct tpacket_block_desc *);
260static void prb_retire_rx_blk_timer_expired(unsigned long);
261static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
262static void prb_init_blk_timer(struct packet_sock *,
263 struct tpacket_kbdq_core *,
264 void (*func) (unsigned long));
265static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
266static void prb_clear_rxhash(struct tpacket_kbdq_core *,
267 struct tpacket3_hdr *);
268static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
269 struct tpacket3_hdr *);
270static void packet_flush_mclist(struct sock *sk);
271
272struct packet_fanout;
273struct packet_sock {
274 /* struct sock has to be the first member of packet_sock */
275 struct sock sk;
276 struct packet_fanout *fanout;
277 struct tpacket_stats stats;
278 union tpacket_stats_u stats_u;
279 struct packet_ring_buffer rx_ring;
280 struct packet_ring_buffer tx_ring;
281 int copy_thresh;
282 spinlock_t bind_lock;
283 struct mutex pg_vec_lock;
284 unsigned int running:1, /* prot_hook is attached*/
285 auxdata:1,
286 origdev:1,
287 has_vnet_hdr:1;
288 int ifindex; /* bound device */
289 __be16 num;
290 struct packet_mclist *mclist;
291 atomic_t mapped;
292 enum tpacket_versions tp_version;
293 unsigned int tp_hdrlen;
294 unsigned int tp_reserve;
295 unsigned int tp_loss:1;
296 unsigned int tp_tstamp;
297 struct packet_type prot_hook ____cacheline_aligned_in_smp;
298};
299
300#define PACKET_FANOUT_MAX 256
301
302struct packet_fanout {
303#ifdef CONFIG_NET_NS
304 struct net *net;
305#endif
306 unsigned int num_members;
307 u16 id;
308 u8 type;
309 u8 defrag;
310 atomic_t rr_cur;
311 struct list_head list;
312 struct sock *arr[PACKET_FANOUT_MAX];
313 spinlock_t lock;
314 atomic_t sk_ref;
315 struct packet_type prot_hook ____cacheline_aligned_in_smp;
316};
317
318struct packet_skb_cb {
319 unsigned int origlen;
320 union {
321 struct sockaddr_pkt pkt;
322 struct sockaddr_ll ll;
323 } sa;
324};
325
326#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
327
328#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
329#define GET_PBLOCK_DESC(x, bid) \
330 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
331#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
332 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
333#define GET_NEXT_PRB_BLK_NUM(x) \
334 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
335 ((x)->kactive_blk_num+1) : 0)
336
337static struct packet_sock *pkt_sk(struct sock *sk)
338{
339 return (struct packet_sock *)sk;
340}
341
342static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
343static void __fanout_link(struct sock *sk, struct packet_sock *po);
344
345/* register_prot_hook must be invoked with the po->bind_lock held,
346 * or from a context in which asynchronous accesses to the packet
347 * socket is not possible (packet_create()).
348 */
349static void register_prot_hook(struct sock *sk)
350{
351 struct packet_sock *po = pkt_sk(sk);
352 if (!po->running) {
353 if (po->fanout)
354 __fanout_link(sk, po);
355 else
356 dev_add_pack(&po->prot_hook);
357 sock_hold(sk);
358 po->running = 1;
359 }
360}
361
362/* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
363 * held. If the sync parameter is true, we will temporarily drop
364 * the po->bind_lock and do a synchronize_net to make sure no
365 * asynchronous packet processing paths still refer to the elements
366 * of po->prot_hook. If the sync parameter is false, it is the
367 * callers responsibility to take care of this.
368 */
369static void __unregister_prot_hook(struct sock *sk, bool sync)
370{
371 struct packet_sock *po = pkt_sk(sk);
372
373 po->running = 0;
374 if (po->fanout)
375 __fanout_unlink(sk, po);
376 else
377 __dev_remove_pack(&po->prot_hook);
378 __sock_put(sk);
379
380 if (sync) {
381 spin_unlock(&po->bind_lock);
382 synchronize_net();
383 spin_lock(&po->bind_lock);
384 }
385}
386
387static void unregister_prot_hook(struct sock *sk, bool sync)
388{
389 struct packet_sock *po = pkt_sk(sk);
390
391 if (po->running)
392 __unregister_prot_hook(sk, sync);
393}
394
395static inline __pure struct page *pgv_to_page(void *addr)
396{
397 if (is_vmalloc_addr(addr))
398 return vmalloc_to_page(addr);
399 return virt_to_page(addr);
400}
401
402static void __packet_set_status(struct packet_sock *po, void *frame, int status)
403{
404 union {
405 struct tpacket_hdr *h1;
406 struct tpacket2_hdr *h2;
407 void *raw;
408 } h;
409
410 h.raw = frame;
411 switch (po->tp_version) {
412 case TPACKET_V1:
413 h.h1->tp_status = status;
414 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
415 break;
416 case TPACKET_V2:
417 h.h2->tp_status = status;
418 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
419 break;
420 case TPACKET_V3:
421 default:
422 WARN(1, "TPACKET version not supported.\n");
423 BUG();
424 }
425
426 smp_wmb();
427}
428
429static int __packet_get_status(struct packet_sock *po, void *frame)
430{
431 union {
432 struct tpacket_hdr *h1;
433 struct tpacket2_hdr *h2;
434 void *raw;
435 } h;
436
437 smp_rmb();
438
439 h.raw = frame;
440 switch (po->tp_version) {
441 case TPACKET_V1:
442 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
443 return h.h1->tp_status;
444 case TPACKET_V2:
445 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
446 return h.h2->tp_status;
447 case TPACKET_V3:
448 default:
449 WARN(1, "TPACKET version not supported.\n");
450 BUG();
451 return 0;
452 }
453}
454
455static void *packet_lookup_frame(struct packet_sock *po,
456 struct packet_ring_buffer *rb,
457 unsigned int position,
458 int status)
459{
460 unsigned int pg_vec_pos, frame_offset;
461 union {
462 struct tpacket_hdr *h1;
463 struct tpacket2_hdr *h2;
464 void *raw;
465 } h;
466
467 pg_vec_pos = position / rb->frames_per_block;
468 frame_offset = position % rb->frames_per_block;
469
470 h.raw = rb->pg_vec[pg_vec_pos].buffer +
471 (frame_offset * rb->frame_size);
472
473 if (status != __packet_get_status(po, h.raw))
474 return NULL;
475
476 return h.raw;
477}
478
479static void *packet_current_frame(struct packet_sock *po,
480 struct packet_ring_buffer *rb,
481 int status)
482{
483 return packet_lookup_frame(po, rb, rb->head, status);
484}
485
486static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
487{
488 del_timer_sync(&pkc->retire_blk_timer);
489}
490
491static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
492 int tx_ring,
493 struct sk_buff_head *rb_queue)
494{
495 struct tpacket_kbdq_core *pkc;
496
497 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
498
499 spin_lock(&rb_queue->lock);
500 pkc->delete_blk_timer = 1;
501 spin_unlock(&rb_queue->lock);
502
503 prb_del_retire_blk_timer(pkc);
504}
505
506static void prb_init_blk_timer(struct packet_sock *po,
507 struct tpacket_kbdq_core *pkc,
508 void (*func) (unsigned long))
509{
510 init_timer(&pkc->retire_blk_timer);
511 pkc->retire_blk_timer.data = (long)po;
512 pkc->retire_blk_timer.function = func;
513 pkc->retire_blk_timer.expires = jiffies;
514}
515
516static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
517{
518 struct tpacket_kbdq_core *pkc;
519
520 if (tx_ring)
521 BUG();
522
523 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
524 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
525}
526
527static int prb_calc_retire_blk_tmo(struct packet_sock *po,
528 int blk_size_in_bytes)
529{
530 struct net_device *dev;
531 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
532 struct ethtool_cmd ecmd;
533 int err;
534
535 rtnl_lock();
536 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
537 if (unlikely(!dev)) {
538 rtnl_unlock();
539 return DEFAULT_PRB_RETIRE_TOV;
540 }
541 err = __ethtool_get_settings(dev, &ecmd);
542 rtnl_unlock();
543 if (!err) {
544 switch (ecmd.speed) {
545 case SPEED_10000:
546 msec = 1;
547 div = 10000/1000;
548 break;
549 case SPEED_1000:
550 msec = 1;
551 div = 1000/1000;
552 break;
553 /*
554 * If the link speed is so slow you don't really
555 * need to worry about perf anyways
556 */
557 case SPEED_100:
558 case SPEED_10:
559 default:
560 return DEFAULT_PRB_RETIRE_TOV;
561 }
562 }
563
564 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
565
566 if (div)
567 mbits /= div;
568
569 tmo = mbits * msec;
570
571 if (div)
572 return tmo+1;
573 return tmo;
574}
575
576static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
577 union tpacket_req_u *req_u)
578{
579 p1->feature_req_word = req_u->req3.tp_feature_req_word;
580}
581
582static void init_prb_bdqc(struct packet_sock *po,
583 struct packet_ring_buffer *rb,
584 struct pgv *pg_vec,
585 union tpacket_req_u *req_u, int tx_ring)
586{
587 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
588 struct tpacket_block_desc *pbd;
589
590 memset(p1, 0x0, sizeof(*p1));
591
592 p1->knxt_seq_num = 1;
593 p1->pkbdq = pg_vec;
594 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
595 p1->pkblk_start = (char *)pg_vec[0].buffer;
596 p1->kblk_size = req_u->req3.tp_block_size;
597 p1->knum_blocks = req_u->req3.tp_block_nr;
598 p1->hdrlen = po->tp_hdrlen;
599 p1->version = po->tp_version;
600 p1->last_kactive_blk_num = 0;
601 po->stats_u.stats3.tp_freeze_q_cnt = 0;
602 if (req_u->req3.tp_retire_blk_tov)
603 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
604 else
605 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
606 req_u->req3.tp_block_size);
607 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
608 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
609
610 prb_init_ft_ops(p1, req_u);
611 prb_setup_retire_blk_timer(po, tx_ring);
612 prb_open_block(p1, pbd);
613}
614
615/* Do NOT update the last_blk_num first.
616 * Assumes sk_buff_head lock is held.
617 */
618static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
619{
620 mod_timer(&pkc->retire_blk_timer,
621 jiffies + pkc->tov_in_jiffies);
622 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
623}
624
625/*
626 * Timer logic:
627 * 1) We refresh the timer only when we open a block.
628 * By doing this we don't waste cycles refreshing the timer
629 * on packet-by-packet basis.
630 *
631 * With a 1MB block-size, on a 1Gbps line, it will take
632 * i) ~8 ms to fill a block + ii) memcpy etc.
633 * In this cut we are not accounting for the memcpy time.
634 *
635 * So, if the user sets the 'tmo' to 10ms then the timer
636 * will never fire while the block is still getting filled
637 * (which is what we want). However, the user could choose
638 * to close a block early and that's fine.
639 *
640 * But when the timer does fire, we check whether or not to refresh it.
641 * Since the tmo granularity is in msecs, it is not too expensive
642 * to refresh the timer, lets say every '8' msecs.
643 * Either the user can set the 'tmo' or we can derive it based on
644 * a) line-speed and b) block-size.
645 * prb_calc_retire_blk_tmo() calculates the tmo.
646 *
647 */
648static void prb_retire_rx_blk_timer_expired(unsigned long data)
649{
650 struct packet_sock *po = (struct packet_sock *)data;
651 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
652 unsigned int frozen;
653 struct tpacket_block_desc *pbd;
654
655 spin_lock(&po->sk.sk_receive_queue.lock);
656
657 frozen = prb_queue_frozen(pkc);
658 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
659
660 if (unlikely(pkc->delete_blk_timer))
661 goto out;
662
663 /* We only need to plug the race when the block is partially filled.
664 * tpacket_rcv:
665 * lock(); increment BLOCK_NUM_PKTS; unlock()
666 * copy_bits() is in progress ...
667 * timer fires on other cpu:
668 * we can't retire the current block because copy_bits
669 * is in progress.
670 *
671 */
672 if (BLOCK_NUM_PKTS(pbd)) {
673 while (atomic_read(&pkc->blk_fill_in_prog)) {
674 /* Waiting for skb_copy_bits to finish... */
675 cpu_relax();
676 }
677 }
678
679 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
680 if (!frozen) {
681 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
682 if (!prb_dispatch_next_block(pkc, po))
683 goto refresh_timer;
684 else
685 goto out;
686 } else {
687 /* Case 1. Queue was frozen because user-space was
688 * lagging behind.
689 */
690 if (prb_curr_blk_in_use(pkc, pbd)) {
691 /*
692 * Ok, user-space is still behind.
693 * So just refresh the timer.
694 */
695 goto refresh_timer;
696 } else {
697 /* Case 2. queue was frozen,user-space caught up,
698 * now the link went idle && the timer fired.
699 * We don't have a block to close.So we open this
700 * block and restart the timer.
701 * opening a block thaws the queue,restarts timer
702 * Thawing/timer-refresh is a side effect.
703 */
704 prb_open_block(pkc, pbd);
705 goto out;
706 }
707 }
708 }
709
710refresh_timer:
711 _prb_refresh_rx_retire_blk_timer(pkc);
712
713out:
714 spin_unlock(&po->sk.sk_receive_queue.lock);
715}
716
717static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
718 struct tpacket_block_desc *pbd1, __u32 status)
719{
720 /* Flush everything minus the block header */
721
722#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
723 u8 *start, *end;
724
725 start = (u8 *)pbd1;
726
727 /* Skip the block header(we know header WILL fit in 4K) */
728 start += PAGE_SIZE;
729
730 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
731 for (; start < end; start += PAGE_SIZE)
732 flush_dcache_page(pgv_to_page(start));
733
734 smp_wmb();
735#endif
736
737 /* Now update the block status. */
738
739 BLOCK_STATUS(pbd1) = status;
740
741 /* Flush the block header */
742
743#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
744 start = (u8 *)pbd1;
745 flush_dcache_page(pgv_to_page(start));
746
747 smp_wmb();
748#endif
749}
750
751/*
752 * Side effect:
753 *
754 * 1) flush the block
755 * 2) Increment active_blk_num
756 *
757 * Note:We DONT refresh the timer on purpose.
758 * Because almost always the next block will be opened.
759 */
760static void prb_close_block(struct tpacket_kbdq_core *pkc1,
761 struct tpacket_block_desc *pbd1,
762 struct packet_sock *po, unsigned int stat)
763{
764 __u32 status = TP_STATUS_USER | stat;
765
766 struct tpacket3_hdr *last_pkt;
767 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
768
769 if (po->stats.tp_drops)
770 status |= TP_STATUS_LOSING;
771
772 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
773 last_pkt->tp_next_offset = 0;
774
775 /* Get the ts of the last pkt */
776 if (BLOCK_NUM_PKTS(pbd1)) {
777 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
778 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
779 } else {
780 /* Ok, we tmo'd - so get the current time */
781 struct timespec ts;
782 getnstimeofday(&ts);
783 h1->ts_last_pkt.ts_sec = ts.tv_sec;
784 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
785 }
786
787 smp_wmb();
788
789 /* Flush the block */
790 prb_flush_block(pkc1, pbd1, status);
791
792 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
793}
794
795static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
796{
797 pkc->reset_pending_on_curr_blk = 0;
798}
799
800/*
801 * Side effect of opening a block:
802 *
803 * 1) prb_queue is thawed.
804 * 2) retire_blk_timer is refreshed.
805 *
806 */
807static void prb_open_block(struct tpacket_kbdq_core *pkc1,
808 struct tpacket_block_desc *pbd1)
809{
810 struct timespec ts;
811 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
812
813 smp_rmb();
814
815 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
816
817 /* We could have just memset this but we will lose the
818 * flexibility of making the priv area sticky
819 */
820 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
821 BLOCK_NUM_PKTS(pbd1) = 0;
822 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
823 getnstimeofday(&ts);
824 h1->ts_first_pkt.ts_sec = ts.tv_sec;
825 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
826 pkc1->pkblk_start = (char *)pbd1;
827 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
828 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
829 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
831 pbd1->version = pkc1->version;
832 pkc1->prev = pkc1->nxt_offset;
833 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
834 prb_thaw_queue(pkc1);
835 _prb_refresh_rx_retire_blk_timer(pkc1);
836
837 smp_wmb();
838
839 return;
840 }
841
842 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
843 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
844 dump_stack();
845 BUG();
846}
847
848/*
849 * Queue freeze logic:
850 * 1) Assume tp_block_nr = 8 blocks.
851 * 2) At time 't0', user opens Rx ring.
852 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
853 * 4) user-space is either sleeping or processing block '0'.
854 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
855 * it will close block-7,loop around and try to fill block '0'.
856 * call-flow:
857 * __packet_lookup_frame_in_block
858 * prb_retire_current_block()
859 * prb_dispatch_next_block()
860 * |->(BLOCK_STATUS == USER) evaluates to true
861 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
862 * 6) Now there are two cases:
863 * 6.1) Link goes idle right after the queue is frozen.
864 * But remember, the last open_block() refreshed the timer.
865 * When this timer expires,it will refresh itself so that we can
866 * re-open block-0 in near future.
867 * 6.2) Link is busy and keeps on receiving packets. This is a simple
868 * case and __packet_lookup_frame_in_block will check if block-0
869 * is free and can now be re-used.
870 */
871static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
872 struct packet_sock *po)
873{
874 pkc->reset_pending_on_curr_blk = 1;
875 po->stats_u.stats3.tp_freeze_q_cnt++;
876}
877
878#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
879
880/*
881 * If the next block is free then we will dispatch it
882 * and return a good offset.
883 * Else, we will freeze the queue.
884 * So, caller must check the return value.
885 */
886static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
887 struct packet_sock *po)
888{
889 struct tpacket_block_desc *pbd;
890
891 smp_rmb();
892
893 /* 1. Get current block num */
894 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
895
896 /* 2. If this block is currently in_use then freeze the queue */
897 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
898 prb_freeze_queue(pkc, po);
899 return NULL;
900 }
901
902 /*
903 * 3.
904 * open this block and return the offset where the first packet
905 * needs to get stored.
906 */
907 prb_open_block(pkc, pbd);
908 return (void *)pkc->nxt_offset;
909}
910
911static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
912 struct packet_sock *po, unsigned int status)
913{
914 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
915
916 /* retire/close the current block */
917 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
918 /*
919 * Plug the case where copy_bits() is in progress on
920 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
921 * have space to copy the pkt in the current block and
922 * called prb_retire_current_block()
923 *
924 * We don't need to worry about the TMO case because
925 * the timer-handler already handled this case.
926 */
927 if (!(status & TP_STATUS_BLK_TMO)) {
928 while (atomic_read(&pkc->blk_fill_in_prog)) {
929 /* Waiting for skb_copy_bits to finish... */
930 cpu_relax();
931 }
932 }
933 prb_close_block(pkc, pbd, po, status);
934 return;
935 }
936
937 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
938 dump_stack();
939 BUG();
940}
941
942static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
943 struct tpacket_block_desc *pbd)
944{
945 return TP_STATUS_USER & BLOCK_STATUS(pbd);
946}
947
948static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
949{
950 return pkc->reset_pending_on_curr_blk;
951}
952
953static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
954{
955 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
956 atomic_dec(&pkc->blk_fill_in_prog);
957}
958
959static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
960 struct tpacket3_hdr *ppd)
961{
962 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
963}
964
965static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
966 struct tpacket3_hdr *ppd)
967{
968 ppd->hv1.tp_rxhash = 0;
969}
970
971static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
972 struct tpacket3_hdr *ppd)
973{
974 if (vlan_tx_tag_present(pkc->skb)) {
975 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
976 ppd->tp_status = TP_STATUS_VLAN_VALID;
977 } else {
978 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
979 }
980}
981
982static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
983 struct tpacket3_hdr *ppd)
984{
985 prb_fill_vlan_info(pkc, ppd);
986
987 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
988 prb_fill_rxhash(pkc, ppd);
989 else
990 prb_clear_rxhash(pkc, ppd);
991}
992
993static void prb_fill_curr_block(char *curr,
994 struct tpacket_kbdq_core *pkc,
995 struct tpacket_block_desc *pbd,
996 unsigned int len)
997{
998 struct tpacket3_hdr *ppd;
999
1000 ppd = (struct tpacket3_hdr *)curr;
1001 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1002 pkc->prev = curr;
1003 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1004 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 BLOCK_NUM_PKTS(pbd) += 1;
1006 atomic_inc(&pkc->blk_fill_in_prog);
1007 prb_run_all_ft_ops(pkc, ppd);
1008}
1009
1010/* Assumes caller has the sk->rx_queue.lock */
1011static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1012 struct sk_buff *skb,
1013 int status,
1014 unsigned int len
1015 )
1016{
1017 struct tpacket_kbdq_core *pkc;
1018 struct tpacket_block_desc *pbd;
1019 char *curr, *end;
1020
1021 pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1022 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1023
1024 /* Queue is frozen when user space is lagging behind */
1025 if (prb_queue_frozen(pkc)) {
1026 /*
1027 * Check if that last block which caused the queue to freeze,
1028 * is still in_use by user-space.
1029 */
1030 if (prb_curr_blk_in_use(pkc, pbd)) {
1031 /* Can't record this packet */
1032 return NULL;
1033 } else {
1034 /*
1035 * Ok, the block was released by user-space.
1036 * Now let's open that block.
1037 * opening a block also thaws the queue.
1038 * Thawing is a side effect.
1039 */
1040 prb_open_block(pkc, pbd);
1041 }
1042 }
1043
1044 smp_mb();
1045 curr = pkc->nxt_offset;
1046 pkc->skb = skb;
1047 end = (char *) ((char *)pbd + pkc->kblk_size);
1048
1049 /* first try the current block */
1050 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1051 prb_fill_curr_block(curr, pkc, pbd, len);
1052 return (void *)curr;
1053 }
1054
1055 /* Ok, close the current block */
1056 prb_retire_current_block(pkc, po, 0);
1057
1058 /* Now, try to dispatch the next block */
1059 curr = (char *)prb_dispatch_next_block(pkc, po);
1060 if (curr) {
1061 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1062 prb_fill_curr_block(curr, pkc, pbd, len);
1063 return (void *)curr;
1064 }
1065
1066 /*
1067 * No free blocks are available.user_space hasn't caught up yet.
1068 * Queue was just frozen and now this packet will get dropped.
1069 */
1070 return NULL;
1071}
1072
1073static void *packet_current_rx_frame(struct packet_sock *po,
1074 struct sk_buff *skb,
1075 int status, unsigned int len)
1076{
1077 char *curr = NULL;
1078 switch (po->tp_version) {
1079 case TPACKET_V1:
1080 case TPACKET_V2:
1081 curr = packet_lookup_frame(po, &po->rx_ring,
1082 po->rx_ring.head, status);
1083 return curr;
1084 case TPACKET_V3:
1085 return __packet_lookup_frame_in_block(po, skb, status, len);
1086 default:
1087 WARN(1, "TPACKET version not supported\n");
1088 BUG();
1089 return 0;
1090 }
1091}
1092
1093static void *prb_lookup_block(struct packet_sock *po,
1094 struct packet_ring_buffer *rb,
1095 unsigned int previous,
1096 int status)
1097{
1098 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1099 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1100
1101 if (status != BLOCK_STATUS(pbd))
1102 return NULL;
1103 return pbd;
1104}
1105
1106static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1107{
1108 unsigned int prev;
1109 if (rb->prb_bdqc.kactive_blk_num)
1110 prev = rb->prb_bdqc.kactive_blk_num-1;
1111 else
1112 prev = rb->prb_bdqc.knum_blocks-1;
1113 return prev;
1114}
1115
1116/* Assumes caller has held the rx_queue.lock */
1117static void *__prb_previous_block(struct packet_sock *po,
1118 struct packet_ring_buffer *rb,
1119 int status)
1120{
1121 unsigned int previous = prb_previous_blk_num(rb);
1122 return prb_lookup_block(po, rb, previous, status);
1123}
1124
1125static void *packet_previous_rx_frame(struct packet_sock *po,
1126 struct packet_ring_buffer *rb,
1127 int status)
1128{
1129 if (po->tp_version <= TPACKET_V2)
1130 return packet_previous_frame(po, rb, status);
1131
1132 return __prb_previous_block(po, rb, status);
1133}
1134
1135static void packet_increment_rx_head(struct packet_sock *po,
1136 struct packet_ring_buffer *rb)
1137{
1138 switch (po->tp_version) {
1139 case TPACKET_V1:
1140 case TPACKET_V2:
1141 return packet_increment_head(rb);
1142 case TPACKET_V3:
1143 default:
1144 WARN(1, "TPACKET version not supported.\n");
1145 BUG();
1146 return;
1147 }
1148}
1149
1150static void *packet_previous_frame(struct packet_sock *po,
1151 struct packet_ring_buffer *rb,
1152 int status)
1153{
1154 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1155 return packet_lookup_frame(po, rb, previous, status);
1156}
1157
1158static void packet_increment_head(struct packet_ring_buffer *buff)
1159{
1160 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1161}
1162
1163static void packet_sock_destruct(struct sock *sk)
1164{
1165 skb_queue_purge(&sk->sk_error_queue);
1166
1167 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1168 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1169
1170 if (!sock_flag(sk, SOCK_DEAD)) {
1171 pr_err("Attempt to release alive packet socket: %p\n", sk);
1172 return;
1173 }
1174
1175 sk_refcnt_debug_dec(sk);
1176}
1177
1178static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1179{
1180 int x = atomic_read(&f->rr_cur) + 1;
1181
1182 if (x >= num)
1183 x = 0;
1184
1185 return x;
1186}
1187
1188static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1189{
1190 u32 idx, hash = skb->rxhash;
1191
1192 idx = ((u64)hash * num) >> 32;
1193
1194 return f->arr[idx];
1195}
1196
1197static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1198{
1199 int cur, old;
1200
1201 cur = atomic_read(&f->rr_cur);
1202 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1203 fanout_rr_next(f, num))) != cur)
1204 cur = old;
1205 return f->arr[cur];
1206}
1207
1208static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1209{
1210 unsigned int cpu = smp_processor_id();
1211
1212 return f->arr[cpu % num];
1213}
1214
1215static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1216 struct packet_type *pt, struct net_device *orig_dev)
1217{
1218 struct packet_fanout *f = pt->af_packet_priv;
1219 unsigned int num = f->num_members;
1220 struct packet_sock *po;
1221 struct sock *sk;
1222
1223 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1224 !num) {
1225 kfree_skb(skb);
1226 return 0;
1227 }
1228
1229 switch (f->type) {
1230 case PACKET_FANOUT_HASH:
1231 default:
1232 if (f->defrag) {
1233 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1234 if (!skb)
1235 return 0;
1236 }
1237 skb_get_rxhash(skb);
1238 sk = fanout_demux_hash(f, skb, num);
1239 break;
1240 case PACKET_FANOUT_LB:
1241 sk = fanout_demux_lb(f, skb, num);
1242 break;
1243 case PACKET_FANOUT_CPU:
1244 sk = fanout_demux_cpu(f, skb, num);
1245 break;
1246 }
1247
1248 po = pkt_sk(sk);
1249
1250 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1251}
1252
1253static DEFINE_MUTEX(fanout_mutex);
1254static LIST_HEAD(fanout_list);
1255
1256static void __fanout_link(struct sock *sk, struct packet_sock *po)
1257{
1258 struct packet_fanout *f = po->fanout;
1259
1260 spin_lock(&f->lock);
1261 f->arr[f->num_members] = sk;
1262 smp_wmb();
1263 f->num_members++;
1264 spin_unlock(&f->lock);
1265}
1266
1267static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1268{
1269 struct packet_fanout *f = po->fanout;
1270 int i;
1271
1272 spin_lock(&f->lock);
1273 for (i = 0; i < f->num_members; i++) {
1274 if (f->arr[i] == sk)
1275 break;
1276 }
1277 BUG_ON(i >= f->num_members);
1278 f->arr[i] = f->arr[f->num_members - 1];
1279 f->num_members--;
1280 spin_unlock(&f->lock);
1281}
1282
1283bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1284{
1285 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1286 return true;
1287
1288 return false;
1289}
1290
1291static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1292{
1293 struct packet_sock *po = pkt_sk(sk);
1294 struct packet_fanout *f, *match;
1295 u8 type = type_flags & 0xff;
1296 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1297 int err;
1298
1299 switch (type) {
1300 case PACKET_FANOUT_HASH:
1301 case PACKET_FANOUT_LB:
1302 case PACKET_FANOUT_CPU:
1303 break;
1304 default:
1305 return -EINVAL;
1306 }
1307
1308 if (!po->running)
1309 return -EINVAL;
1310
1311 if (po->fanout)
1312 return -EALREADY;
1313
1314 mutex_lock(&fanout_mutex);
1315 match = NULL;
1316 list_for_each_entry(f, &fanout_list, list) {
1317 if (f->id == id &&
1318 read_pnet(&f->net) == sock_net(sk)) {
1319 match = f;
1320 break;
1321 }
1322 }
1323 err = -EINVAL;
1324 if (match && match->defrag != defrag)
1325 goto out;
1326 if (!match) {
1327 err = -ENOMEM;
1328 match = kzalloc(sizeof(*match), GFP_KERNEL);
1329 if (!match)
1330 goto out;
1331 write_pnet(&match->net, sock_net(sk));
1332 match->id = id;
1333 match->type = type;
1334 match->defrag = defrag;
1335 atomic_set(&match->rr_cur, 0);
1336 INIT_LIST_HEAD(&match->list);
1337 spin_lock_init(&match->lock);
1338 atomic_set(&match->sk_ref, 0);
1339 match->prot_hook.type = po->prot_hook.type;
1340 match->prot_hook.dev = po->prot_hook.dev;
1341 match->prot_hook.func = packet_rcv_fanout;
1342 match->prot_hook.af_packet_priv = match;
1343 match->prot_hook.id_match = match_fanout_group;
1344 dev_add_pack(&match->prot_hook);
1345 list_add(&match->list, &fanout_list);
1346 }
1347 err = -EINVAL;
1348 if (match->type == type &&
1349 match->prot_hook.type == po->prot_hook.type &&
1350 match->prot_hook.dev == po->prot_hook.dev) {
1351 err = -ENOSPC;
1352 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1353 __dev_remove_pack(&po->prot_hook);
1354 po->fanout = match;
1355 atomic_inc(&match->sk_ref);
1356 __fanout_link(sk, po);
1357 err = 0;
1358 }
1359 }
1360out:
1361 mutex_unlock(&fanout_mutex);
1362 return err;
1363}
1364
1365static void fanout_release(struct sock *sk)
1366{
1367 struct packet_sock *po = pkt_sk(sk);
1368 struct packet_fanout *f;
1369
1370 f = po->fanout;
1371 if (!f)
1372 return;
1373
1374 po->fanout = NULL;
1375
1376 mutex_lock(&fanout_mutex);
1377 if (atomic_dec_and_test(&f->sk_ref)) {
1378 list_del(&f->list);
1379 dev_remove_pack(&f->prot_hook);
1380 kfree(f);
1381 }
1382 mutex_unlock(&fanout_mutex);
1383}
1384
1385static const struct proto_ops packet_ops;
1386
1387static const struct proto_ops packet_ops_spkt;
1388
1389static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1390 struct packet_type *pt, struct net_device *orig_dev)
1391{
1392 struct sock *sk;
1393 struct sockaddr_pkt *spkt;
1394
1395 /*
1396 * When we registered the protocol we saved the socket in the data
1397 * field for just this event.
1398 */
1399
1400 sk = pt->af_packet_priv;
1401
1402 /*
1403 * Yank back the headers [hope the device set this
1404 * right or kerboom...]
1405 *
1406 * Incoming packets have ll header pulled,
1407 * push it back.
1408 *
1409 * For outgoing ones skb->data == skb_mac_header(skb)
1410 * so that this procedure is noop.
1411 */
1412
1413 if (skb->pkt_type == PACKET_LOOPBACK)
1414 goto out;
1415
1416 if (!net_eq(dev_net(dev), sock_net(sk)))
1417 goto out;
1418
1419 skb = skb_share_check(skb, GFP_ATOMIC);
1420 if (skb == NULL)
1421 goto oom;
1422
1423 /* drop any routing info */
1424 skb_dst_drop(skb);
1425
1426 /* drop conntrack reference */
1427 nf_reset(skb);
1428
1429 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1430
1431 skb_push(skb, skb->data - skb_mac_header(skb));
1432
1433 /*
1434 * The SOCK_PACKET socket receives _all_ frames.
1435 */
1436
1437 spkt->spkt_family = dev->type;
1438 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1439 spkt->spkt_protocol = skb->protocol;
1440
1441 /*
1442 * Charge the memory to the socket. This is done specifically
1443 * to prevent sockets using all the memory up.
1444 */
1445
1446 if (sock_queue_rcv_skb(sk, skb) == 0)
1447 return 0;
1448
1449out:
1450 kfree_skb(skb);
1451oom:
1452 return 0;
1453}
1454
1455
1456/*
1457 * Output a raw packet to a device layer. This bypasses all the other
1458 * protocol layers and you must therefore supply it with a complete frame
1459 */
1460
1461static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1462 struct msghdr *msg, size_t len)
1463{
1464 struct sock *sk = sock->sk;
1465 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1466 struct sk_buff *skb = NULL;
1467 struct net_device *dev;
1468 __be16 proto = 0;
1469 int err;
1470 int extra_len = 0;
1471
1472 /*
1473 * Get and verify the address.
1474 */
1475
1476 if (saddr) {
1477 if (msg->msg_namelen < sizeof(struct sockaddr))
1478 return -EINVAL;
1479 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1480 proto = saddr->spkt_protocol;
1481 } else
1482 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1483
1484 /*
1485 * Find the device first to size check it
1486 */
1487
1488 saddr->spkt_device[13] = 0;
1489retry:
1490 rcu_read_lock();
1491 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1492 err = -ENODEV;
1493 if (dev == NULL)
1494 goto out_unlock;
1495
1496 err = -ENETDOWN;
1497 if (!(dev->flags & IFF_UP))
1498 goto out_unlock;
1499
1500 /*
1501 * You may not queue a frame bigger than the mtu. This is the lowest level
1502 * raw protocol and you must do your own fragmentation at this level.
1503 */
1504
1505 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1506 if (!netif_supports_nofcs(dev)) {
1507 err = -EPROTONOSUPPORT;
1508 goto out_unlock;
1509 }
1510 extra_len = 4; /* We're doing our own CRC */
1511 }
1512
1513 err = -EMSGSIZE;
1514 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1515 goto out_unlock;
1516
1517 if (!skb) {
1518 size_t reserved = LL_RESERVED_SPACE(dev);
1519 int tlen = dev->needed_tailroom;
1520 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1521
1522 rcu_read_unlock();
1523 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1524 if (skb == NULL)
1525 return -ENOBUFS;
1526 /* FIXME: Save some space for broken drivers that write a hard
1527 * header at transmission time by themselves. PPP is the notable
1528 * one here. This should really be fixed at the driver level.
1529 */
1530 skb_reserve(skb, reserved);
1531 skb_reset_network_header(skb);
1532
1533 /* Try to align data part correctly */
1534 if (hhlen) {
1535 skb->data -= hhlen;
1536 skb->tail -= hhlen;
1537 if (len < hhlen)
1538 skb_reset_network_header(skb);
1539 }
1540 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1541 if (err)
1542 goto out_free;
1543 goto retry;
1544 }
1545
1546 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1547 /* Earlier code assumed this would be a VLAN pkt,
1548 * double-check this now that we have the actual
1549 * packet in hand.
1550 */
1551 struct ethhdr *ehdr;
1552 skb_reset_mac_header(skb);
1553 ehdr = eth_hdr(skb);
1554 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1555 err = -EMSGSIZE;
1556 goto out_unlock;
1557 }
1558 }
1559
1560 skb->protocol = proto;
1561 skb->dev = dev;
1562 skb->priority = sk->sk_priority;
1563 skb->mark = sk->sk_mark;
1564 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1565 if (err < 0)
1566 goto out_unlock;
1567
1568 if (unlikely(extra_len == 4))
1569 skb->no_fcs = 1;
1570
1571 dev_queue_xmit(skb);
1572 rcu_read_unlock();
1573 return len;
1574
1575out_unlock:
1576 rcu_read_unlock();
1577out_free:
1578 kfree_skb(skb);
1579 return err;
1580}
1581
1582static unsigned int run_filter(const struct sk_buff *skb,
1583 const struct sock *sk,
1584 unsigned int res)
1585{
1586 struct sk_filter *filter;
1587
1588 rcu_read_lock();
1589 filter = rcu_dereference(sk->sk_filter);
1590 if (filter != NULL)
1591 res = SK_RUN_FILTER(filter, skb);
1592 rcu_read_unlock();
1593
1594 return res;
1595}
1596
1597/*
1598 * This function makes lazy skb cloning in hope that most of packets
1599 * are discarded by BPF.
1600 *
1601 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1602 * and skb->cb are mangled. It works because (and until) packets
1603 * falling here are owned by current CPU. Output packets are cloned
1604 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1605 * sequencially, so that if we return skb to original state on exit,
1606 * we will not harm anyone.
1607 */
1608
1609static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1610 struct packet_type *pt, struct net_device *orig_dev)
1611{
1612 struct sock *sk;
1613 struct sockaddr_ll *sll;
1614 struct packet_sock *po;
1615 u8 *skb_head = skb->data;
1616 int skb_len = skb->len;
1617 unsigned int snaplen, res;
1618
1619 if (skb->pkt_type == PACKET_LOOPBACK)
1620 goto drop;
1621
1622 sk = pt->af_packet_priv;
1623 po = pkt_sk(sk);
1624
1625 if (!net_eq(dev_net(dev), sock_net(sk)))
1626 goto drop;
1627
1628 skb->dev = dev;
1629
1630 if (dev->header_ops) {
1631 /* The device has an explicit notion of ll header,
1632 * exported to higher levels.
1633 *
1634 * Otherwise, the device hides details of its frame
1635 * structure, so that corresponding packet head is
1636 * never delivered to user.
1637 */
1638 if (sk->sk_type != SOCK_DGRAM)
1639 skb_push(skb, skb->data - skb_mac_header(skb));
1640 else if (skb->pkt_type == PACKET_OUTGOING) {
1641 /* Special case: outgoing packets have ll header at head */
1642 skb_pull(skb, skb_network_offset(skb));
1643 }
1644 }
1645
1646 snaplen = skb->len;
1647
1648 res = run_filter(skb, sk, snaplen);
1649 if (!res)
1650 goto drop_n_restore;
1651 if (snaplen > res)
1652 snaplen = res;
1653
1654 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1655 goto drop_n_acct;
1656
1657 if (skb_shared(skb)) {
1658 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1659 if (nskb == NULL)
1660 goto drop_n_acct;
1661
1662 if (skb_head != skb->data) {
1663 skb->data = skb_head;
1664 skb->len = skb_len;
1665 }
1666 consume_skb(skb);
1667 skb = nskb;
1668 }
1669
1670 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1671 sizeof(skb->cb));
1672
1673 sll = &PACKET_SKB_CB(skb)->sa.ll;
1674 sll->sll_family = AF_PACKET;
1675 sll->sll_hatype = dev->type;
1676 sll->sll_protocol = skb->protocol;
1677 sll->sll_pkttype = skb->pkt_type;
1678 if (unlikely(po->origdev))
1679 sll->sll_ifindex = orig_dev->ifindex;
1680 else
1681 sll->sll_ifindex = dev->ifindex;
1682
1683 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1684
1685 PACKET_SKB_CB(skb)->origlen = skb->len;
1686
1687 if (pskb_trim(skb, snaplen))
1688 goto drop_n_acct;
1689
1690 skb_set_owner_r(skb, sk);
1691 skb->dev = NULL;
1692 skb_dst_drop(skb);
1693
1694 /* drop conntrack reference */
1695 nf_reset(skb);
1696
1697 spin_lock(&sk->sk_receive_queue.lock);
1698 po->stats.tp_packets++;
1699 skb->dropcount = atomic_read(&sk->sk_drops);
1700 __skb_queue_tail(&sk->sk_receive_queue, skb);
1701 spin_unlock(&sk->sk_receive_queue.lock);
1702 sk->sk_data_ready(sk, skb->len);
1703 return 0;
1704
1705drop_n_acct:
1706 spin_lock(&sk->sk_receive_queue.lock);
1707 po->stats.tp_drops++;
1708 atomic_inc(&sk->sk_drops);
1709 spin_unlock(&sk->sk_receive_queue.lock);
1710
1711drop_n_restore:
1712 if (skb_head != skb->data && skb_shared(skb)) {
1713 skb->data = skb_head;
1714 skb->len = skb_len;
1715 }
1716drop:
1717 consume_skb(skb);
1718 return 0;
1719}
1720
1721static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1722 struct packet_type *pt, struct net_device *orig_dev)
1723{
1724 struct sock *sk;
1725 struct packet_sock *po;
1726 struct sockaddr_ll *sll;
1727 union {
1728 struct tpacket_hdr *h1;
1729 struct tpacket2_hdr *h2;
1730 struct tpacket3_hdr *h3;
1731 void *raw;
1732 } h;
1733 u8 *skb_head = skb->data;
1734 int skb_len = skb->len;
1735 unsigned int snaplen, res;
1736 unsigned long status = TP_STATUS_USER;
1737 unsigned short macoff, netoff, hdrlen;
1738 struct sk_buff *copy_skb = NULL;
1739 struct timeval tv;
1740 struct timespec ts;
1741 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1742
1743 if (skb->pkt_type == PACKET_LOOPBACK)
1744 goto drop;
1745
1746 sk = pt->af_packet_priv;
1747 po = pkt_sk(sk);
1748
1749 if (!net_eq(dev_net(dev), sock_net(sk)))
1750 goto drop;
1751
1752 if (dev->header_ops) {
1753 if (sk->sk_type != SOCK_DGRAM)
1754 skb_push(skb, skb->data - skb_mac_header(skb));
1755 else if (skb->pkt_type == PACKET_OUTGOING) {
1756 /* Special case: outgoing packets have ll header at head */
1757 skb_pull(skb, skb_network_offset(skb));
1758 }
1759 }
1760
1761 if (skb->ip_summed == CHECKSUM_PARTIAL)
1762 status |= TP_STATUS_CSUMNOTREADY;
1763
1764 snaplen = skb->len;
1765
1766 res = run_filter(skb, sk, snaplen);
1767 if (!res)
1768 goto drop_n_restore;
1769 if (snaplen > res)
1770 snaplen = res;
1771
1772 if (sk->sk_type == SOCK_DGRAM) {
1773 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1774 po->tp_reserve;
1775 } else {
1776 unsigned int maclen = skb_network_offset(skb);
1777 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1778 (maclen < 16 ? 16 : maclen)) +
1779 po->tp_reserve;
1780 macoff = netoff - maclen;
1781 }
1782 if (po->tp_version <= TPACKET_V2) {
1783 if (macoff + snaplen > po->rx_ring.frame_size) {
1784 if (po->copy_thresh &&
1785 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1786 if (skb_shared(skb)) {
1787 copy_skb = skb_clone(skb, GFP_ATOMIC);
1788 } else {
1789 copy_skb = skb_get(skb);
1790 skb_head = skb->data;
1791 }
1792 if (copy_skb)
1793 skb_set_owner_r(copy_skb, sk);
1794 }
1795 snaplen = po->rx_ring.frame_size - macoff;
1796 if ((int)snaplen < 0)
1797 snaplen = 0;
1798 }
1799 }
1800 spin_lock(&sk->sk_receive_queue.lock);
1801 h.raw = packet_current_rx_frame(po, skb,
1802 TP_STATUS_KERNEL, (macoff+snaplen));
1803 if (!h.raw)
1804 goto ring_is_full;
1805 if (po->tp_version <= TPACKET_V2) {
1806 packet_increment_rx_head(po, &po->rx_ring);
1807 /*
1808 * LOSING will be reported till you read the stats,
1809 * because it's COR - Clear On Read.
1810 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1811 * at packet level.
1812 */
1813 if (po->stats.tp_drops)
1814 status |= TP_STATUS_LOSING;
1815 }
1816 po->stats.tp_packets++;
1817 if (copy_skb) {
1818 status |= TP_STATUS_COPY;
1819 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1820 }
1821 spin_unlock(&sk->sk_receive_queue.lock);
1822
1823 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1824
1825 switch (po->tp_version) {
1826 case TPACKET_V1:
1827 h.h1->tp_len = skb->len;
1828 h.h1->tp_snaplen = snaplen;
1829 h.h1->tp_mac = macoff;
1830 h.h1->tp_net = netoff;
1831 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1832 && shhwtstamps->syststamp.tv64)
1833 tv = ktime_to_timeval(shhwtstamps->syststamp);
1834 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1835 && shhwtstamps->hwtstamp.tv64)
1836 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1837 else if (skb->tstamp.tv64)
1838 tv = ktime_to_timeval(skb->tstamp);
1839 else
1840 do_gettimeofday(&tv);
1841 h.h1->tp_sec = tv.tv_sec;
1842 h.h1->tp_usec = tv.tv_usec;
1843 hdrlen = sizeof(*h.h1);
1844 break;
1845 case TPACKET_V2:
1846 h.h2->tp_len = skb->len;
1847 h.h2->tp_snaplen = snaplen;
1848 h.h2->tp_mac = macoff;
1849 h.h2->tp_net = netoff;
1850 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1851 && shhwtstamps->syststamp.tv64)
1852 ts = ktime_to_timespec(shhwtstamps->syststamp);
1853 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1854 && shhwtstamps->hwtstamp.tv64)
1855 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1856 else if (skb->tstamp.tv64)
1857 ts = ktime_to_timespec(skb->tstamp);
1858 else
1859 getnstimeofday(&ts);
1860 h.h2->tp_sec = ts.tv_sec;
1861 h.h2->tp_nsec = ts.tv_nsec;
1862 if (vlan_tx_tag_present(skb)) {
1863 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1864 status |= TP_STATUS_VLAN_VALID;
1865 } else {
1866 h.h2->tp_vlan_tci = 0;
1867 }
1868 h.h2->tp_padding = 0;
1869 hdrlen = sizeof(*h.h2);
1870 break;
1871 case TPACKET_V3:
1872 /* tp_nxt_offset,vlan are already populated above.
1873 * So DONT clear those fields here
1874 */
1875 h.h3->tp_status |= status;
1876 h.h3->tp_len = skb->len;
1877 h.h3->tp_snaplen = snaplen;
1878 h.h3->tp_mac = macoff;
1879 h.h3->tp_net = netoff;
1880 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1881 && shhwtstamps->syststamp.tv64)
1882 ts = ktime_to_timespec(shhwtstamps->syststamp);
1883 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1884 && shhwtstamps->hwtstamp.tv64)
1885 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1886 else if (skb->tstamp.tv64)
1887 ts = ktime_to_timespec(skb->tstamp);
1888 else
1889 getnstimeofday(&ts);
1890 h.h3->tp_sec = ts.tv_sec;
1891 h.h3->tp_nsec = ts.tv_nsec;
1892 hdrlen = sizeof(*h.h3);
1893 break;
1894 default:
1895 BUG();
1896 }
1897
1898 sll = h.raw + TPACKET_ALIGN(hdrlen);
1899 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1900 sll->sll_family = AF_PACKET;
1901 sll->sll_hatype = dev->type;
1902 sll->sll_protocol = skb->protocol;
1903 sll->sll_pkttype = skb->pkt_type;
1904 if (unlikely(po->origdev))
1905 sll->sll_ifindex = orig_dev->ifindex;
1906 else
1907 sll->sll_ifindex = dev->ifindex;
1908
1909 smp_mb();
1910#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1911 {
1912 u8 *start, *end;
1913
1914 if (po->tp_version <= TPACKET_V2) {
1915 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1916 + macoff + snaplen);
1917 for (start = h.raw; start < end; start += PAGE_SIZE)
1918 flush_dcache_page(pgv_to_page(start));
1919 }
1920 smp_wmb();
1921 }
1922#endif
1923 if (po->tp_version <= TPACKET_V2)
1924 __packet_set_status(po, h.raw, status);
1925 else
1926 prb_clear_blk_fill_status(&po->rx_ring);
1927
1928 sk->sk_data_ready(sk, 0);
1929
1930drop_n_restore:
1931 if (skb_head != skb->data && skb_shared(skb)) {
1932 skb->data = skb_head;
1933 skb->len = skb_len;
1934 }
1935drop:
1936 kfree_skb(skb);
1937 return 0;
1938
1939ring_is_full:
1940 po->stats.tp_drops++;
1941 spin_unlock(&sk->sk_receive_queue.lock);
1942
1943 sk->sk_data_ready(sk, 0);
1944 kfree_skb(copy_skb);
1945 goto drop_n_restore;
1946}
1947
1948static void tpacket_destruct_skb(struct sk_buff *skb)
1949{
1950 struct packet_sock *po = pkt_sk(skb->sk);
1951 void *ph;
1952
1953 if (likely(po->tx_ring.pg_vec)) {
1954 ph = skb_shinfo(skb)->destructor_arg;
1955 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1956 atomic_dec(&po->tx_ring.pending);
1957 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1958 }
1959
1960 sock_wfree(skb);
1961}
1962
1963static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1964 void *frame, struct net_device *dev, int size_max,
1965 __be16 proto, unsigned char *addr, int hlen)
1966{
1967 union {
1968 struct tpacket_hdr *h1;
1969 struct tpacket2_hdr *h2;
1970 void *raw;
1971 } ph;
1972 int to_write, offset, len, tp_len, nr_frags, len_max;
1973 struct socket *sock = po->sk.sk_socket;
1974 struct page *page;
1975 void *data;
1976 int err;
1977
1978 ph.raw = frame;
1979
1980 skb->protocol = proto;
1981 skb->dev = dev;
1982 skb->priority = po->sk.sk_priority;
1983 skb->mark = po->sk.sk_mark;
1984 skb_shinfo(skb)->destructor_arg = ph.raw;
1985
1986 switch (po->tp_version) {
1987 case TPACKET_V2:
1988 tp_len = ph.h2->tp_len;
1989 break;
1990 default:
1991 tp_len = ph.h1->tp_len;
1992 break;
1993 }
1994 if (unlikely(tp_len > size_max)) {
1995 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1996 return -EMSGSIZE;
1997 }
1998
1999 skb_reserve(skb, hlen);
2000 skb_reset_network_header(skb);
2001
2002 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2003 to_write = tp_len;
2004
2005 if (sock->type == SOCK_DGRAM) {
2006 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2007 NULL, tp_len);
2008 if (unlikely(err < 0))
2009 return -EINVAL;
2010 } else if (dev->hard_header_len) {
2011 /* net device doesn't like empty head */
2012 if (unlikely(tp_len <= dev->hard_header_len)) {
2013 pr_err("packet size is too short (%d < %d)\n",
2014 tp_len, dev->hard_header_len);
2015 return -EINVAL;
2016 }
2017
2018 skb_push(skb, dev->hard_header_len);
2019 err = skb_store_bits(skb, 0, data,
2020 dev->hard_header_len);
2021 if (unlikely(err))
2022 return err;
2023
2024 data += dev->hard_header_len;
2025 to_write -= dev->hard_header_len;
2026 }
2027
2028 err = -EFAULT;
2029 offset = offset_in_page(data);
2030 len_max = PAGE_SIZE - offset;
2031 len = ((to_write > len_max) ? len_max : to_write);
2032
2033 skb->data_len = to_write;
2034 skb->len += to_write;
2035 skb->truesize += to_write;
2036 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2037
2038 while (likely(to_write)) {
2039 nr_frags = skb_shinfo(skb)->nr_frags;
2040
2041 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2042 pr_err("Packet exceed the number of skb frags(%lu)\n",
2043 MAX_SKB_FRAGS);
2044 return -EFAULT;
2045 }
2046
2047 page = pgv_to_page(data);
2048 data += len;
2049 flush_dcache_page(page);
2050 get_page(page);
2051 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2052 to_write -= len;
2053 offset = 0;
2054 len_max = PAGE_SIZE;
2055 len = ((to_write > len_max) ? len_max : to_write);
2056 }
2057
2058 return tp_len;
2059}
2060
2061static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2062{
2063 struct sk_buff *skb;
2064 struct net_device *dev;
2065 __be16 proto;
2066 bool need_rls_dev = false;
2067 int err, reserve = 0;
2068 void *ph;
2069 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2070 int tp_len, size_max;
2071 unsigned char *addr;
2072 int len_sum = 0;
2073 int status = 0;
2074 int hlen, tlen;
2075
2076 mutex_lock(&po->pg_vec_lock);
2077
2078 err = -EBUSY;
2079 if (saddr == NULL) {
2080 dev = po->prot_hook.dev;
2081 proto = po->num;
2082 addr = NULL;
2083 } else {
2084 err = -EINVAL;
2085 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2086 goto out;
2087 if (msg->msg_namelen < (saddr->sll_halen
2088 + offsetof(struct sockaddr_ll,
2089 sll_addr)))
2090 goto out;
2091 proto = saddr->sll_protocol;
2092 addr = saddr->sll_addr;
2093 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2094 need_rls_dev = true;
2095 }
2096
2097 err = -ENXIO;
2098 if (unlikely(dev == NULL))
2099 goto out;
2100
2101 reserve = dev->hard_header_len;
2102
2103 err = -ENETDOWN;
2104 if (unlikely(!(dev->flags & IFF_UP)))
2105 goto out_put;
2106
2107 size_max = po->tx_ring.frame_size
2108 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2109
2110 if (size_max > dev->mtu + reserve)
2111 size_max = dev->mtu + reserve;
2112
2113 do {
2114 ph = packet_current_frame(po, &po->tx_ring,
2115 TP_STATUS_SEND_REQUEST);
2116
2117 if (unlikely(ph == NULL)) {
2118 schedule();
2119 continue;
2120 }
2121
2122 status = TP_STATUS_SEND_REQUEST;
2123 hlen = LL_RESERVED_SPACE(dev);
2124 tlen = dev->needed_tailroom;
2125 skb = sock_alloc_send_skb(&po->sk,
2126 hlen + tlen + sizeof(struct sockaddr_ll),
2127 0, &err);
2128
2129 if (unlikely(skb == NULL))
2130 goto out_status;
2131
2132 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2133 addr, hlen);
2134
2135 if (unlikely(tp_len < 0)) {
2136 if (po->tp_loss) {
2137 __packet_set_status(po, ph,
2138 TP_STATUS_AVAILABLE);
2139 packet_increment_head(&po->tx_ring);
2140 kfree_skb(skb);
2141 continue;
2142 } else {
2143 status = TP_STATUS_WRONG_FORMAT;
2144 err = tp_len;
2145 goto out_status;
2146 }
2147 }
2148
2149 skb->destructor = tpacket_destruct_skb;
2150 __packet_set_status(po, ph, TP_STATUS_SENDING);
2151 atomic_inc(&po->tx_ring.pending);
2152
2153 status = TP_STATUS_SEND_REQUEST;
2154 err = dev_queue_xmit(skb);
2155 if (unlikely(err > 0)) {
2156 err = net_xmit_errno(err);
2157 if (err && __packet_get_status(po, ph) ==
2158 TP_STATUS_AVAILABLE) {
2159 /* skb was destructed already */
2160 skb = NULL;
2161 goto out_status;
2162 }
2163 /*
2164 * skb was dropped but not destructed yet;
2165 * let's treat it like congestion or err < 0
2166 */
2167 err = 0;
2168 }
2169 packet_increment_head(&po->tx_ring);
2170 len_sum += tp_len;
2171 } while (likely((ph != NULL) ||
2172 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2173 (atomic_read(&po->tx_ring.pending))))
2174 );
2175
2176 err = len_sum;
2177 goto out_put;
2178
2179out_status:
2180 __packet_set_status(po, ph, status);
2181 kfree_skb(skb);
2182out_put:
2183 if (need_rls_dev)
2184 dev_put(dev);
2185out:
2186 mutex_unlock(&po->pg_vec_lock);
2187 return err;
2188}
2189
2190static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2191 size_t reserve, size_t len,
2192 size_t linear, int noblock,
2193 int *err)
2194{
2195 struct sk_buff *skb;
2196
2197 /* Under a page? Don't bother with paged skb. */
2198 if (prepad + len < PAGE_SIZE || !linear)
2199 linear = len;
2200
2201 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2202 err);
2203 if (!skb)
2204 return NULL;
2205
2206 skb_reserve(skb, reserve);
2207 skb_put(skb, linear);
2208 skb->data_len = len - linear;
2209 skb->len += len - linear;
2210
2211 return skb;
2212}
2213
2214static int packet_snd(struct socket *sock,
2215 struct msghdr *msg, size_t len)
2216{
2217 struct sock *sk = sock->sk;
2218 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2219 struct sk_buff *skb;
2220 struct net_device *dev;
2221 __be16 proto;
2222 bool need_rls_dev = false;
2223 unsigned char *addr;
2224 int err, reserve = 0;
2225 struct virtio_net_hdr vnet_hdr = { 0 };
2226 int offset = 0;
2227 int vnet_hdr_len;
2228 struct packet_sock *po = pkt_sk(sk);
2229 unsigned short gso_type = 0;
2230 int hlen, tlen;
2231 int extra_len = 0;
2232
2233 /*
2234 * Get and verify the address.
2235 */
2236
2237 if (saddr == NULL) {
2238 dev = po->prot_hook.dev;
2239 proto = po->num;
2240 addr = NULL;
2241 } else {
2242 err = -EINVAL;
2243 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2244 goto out;
2245 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2246 goto out;
2247 proto = saddr->sll_protocol;
2248 addr = saddr->sll_addr;
2249 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2250 need_rls_dev = true;
2251 }
2252
2253 err = -ENXIO;
2254 if (dev == NULL)
2255 goto out_unlock;
2256 if (sock->type == SOCK_RAW)
2257 reserve = dev->hard_header_len;
2258
2259 err = -ENETDOWN;
2260 if (!(dev->flags & IFF_UP))
2261 goto out_unlock;
2262
2263 if (po->has_vnet_hdr) {
2264 vnet_hdr_len = sizeof(vnet_hdr);
2265
2266 err = -EINVAL;
2267 if (len < vnet_hdr_len)
2268 goto out_unlock;
2269
2270 len -= vnet_hdr_len;
2271
2272 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2273 vnet_hdr_len);
2274 if (err < 0)
2275 goto out_unlock;
2276
2277 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2278 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2279 vnet_hdr.hdr_len))
2280 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2281 vnet_hdr.csum_offset + 2;
2282
2283 err = -EINVAL;
2284 if (vnet_hdr.hdr_len > len)
2285 goto out_unlock;
2286
2287 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2288 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2289 case VIRTIO_NET_HDR_GSO_TCPV4:
2290 gso_type = SKB_GSO_TCPV4;
2291 break;
2292 case VIRTIO_NET_HDR_GSO_TCPV6:
2293 gso_type = SKB_GSO_TCPV6;
2294 break;
2295 case VIRTIO_NET_HDR_GSO_UDP:
2296 gso_type = SKB_GSO_UDP;
2297 break;
2298 default:
2299 goto out_unlock;
2300 }
2301
2302 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2303 gso_type |= SKB_GSO_TCP_ECN;
2304
2305 if (vnet_hdr.gso_size == 0)
2306 goto out_unlock;
2307
2308 }
2309 }
2310
2311 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2312 if (!netif_supports_nofcs(dev)) {
2313 err = -EPROTONOSUPPORT;
2314 goto out_unlock;
2315 }
2316 extra_len = 4; /* We're doing our own CRC */
2317 }
2318
2319 err = -EMSGSIZE;
2320 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2321 goto out_unlock;
2322
2323 err = -ENOBUFS;
2324 hlen = LL_RESERVED_SPACE(dev);
2325 tlen = dev->needed_tailroom;
2326 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2327 msg->msg_flags & MSG_DONTWAIT, &err);
2328 if (skb == NULL)
2329 goto out_unlock;
2330
2331 skb_set_network_header(skb, reserve);
2332
2333 err = -EINVAL;
2334 if (sock->type == SOCK_DGRAM &&
2335 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2336 goto out_free;
2337
2338 /* Returns -EFAULT on error */
2339 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2340 if (err)
2341 goto out_free;
2342 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2343 if (err < 0)
2344 goto out_free;
2345
2346 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2347 /* Earlier code assumed this would be a VLAN pkt,
2348 * double-check this now that we have the actual
2349 * packet in hand.
2350 */
2351 struct ethhdr *ehdr;
2352 skb_reset_mac_header(skb);
2353 ehdr = eth_hdr(skb);
2354 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2355 err = -EMSGSIZE;
2356 goto out_free;
2357 }
2358 }
2359
2360 skb->protocol = proto;
2361 skb->dev = dev;
2362 skb->priority = sk->sk_priority;
2363 skb->mark = sk->sk_mark;
2364
2365 if (po->has_vnet_hdr) {
2366 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2367 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2368 vnet_hdr.csum_offset)) {
2369 err = -EINVAL;
2370 goto out_free;
2371 }
2372 }
2373
2374 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2375 skb_shinfo(skb)->gso_type = gso_type;
2376
2377 /* Header must be checked, and gso_segs computed. */
2378 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2379 skb_shinfo(skb)->gso_segs = 0;
2380
2381 len += vnet_hdr_len;
2382 }
2383
2384 if (unlikely(extra_len == 4))
2385 skb->no_fcs = 1;
2386
2387 /*
2388 * Now send it
2389 */
2390
2391 err = dev_queue_xmit(skb);
2392 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2393 goto out_unlock;
2394
2395 if (need_rls_dev)
2396 dev_put(dev);
2397
2398 return len;
2399
2400out_free:
2401 kfree_skb(skb);
2402out_unlock:
2403 if (dev && need_rls_dev)
2404 dev_put(dev);
2405out:
2406 return err;
2407}
2408
2409static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2410 struct msghdr *msg, size_t len)
2411{
2412 struct sock *sk = sock->sk;
2413 struct packet_sock *po = pkt_sk(sk);
2414 if (po->tx_ring.pg_vec)
2415 return tpacket_snd(po, msg);
2416 else
2417 return packet_snd(sock, msg, len);
2418}
2419
2420/*
2421 * Close a PACKET socket. This is fairly simple. We immediately go
2422 * to 'closed' state and remove our protocol entry in the device list.
2423 */
2424
2425static int packet_release(struct socket *sock)
2426{
2427 struct sock *sk = sock->sk;
2428 struct packet_sock *po;
2429 struct net *net;
2430 union tpacket_req_u req_u;
2431
2432 if (!sk)
2433 return 0;
2434
2435 net = sock_net(sk);
2436 po = pkt_sk(sk);
2437
2438 spin_lock_bh(&net->packet.sklist_lock);
2439 sk_del_node_init_rcu(sk);
2440 sock_prot_inuse_add(net, sk->sk_prot, -1);
2441 spin_unlock_bh(&net->packet.sklist_lock);
2442
2443 spin_lock(&po->bind_lock);
2444 unregister_prot_hook(sk, false);
2445 if (po->prot_hook.dev) {
2446 dev_put(po->prot_hook.dev);
2447 po->prot_hook.dev = NULL;
2448 }
2449 spin_unlock(&po->bind_lock);
2450
2451 packet_flush_mclist(sk);
2452
2453 memset(&req_u, 0, sizeof(req_u));
2454
2455 if (po->rx_ring.pg_vec)
2456 packet_set_ring(sk, &req_u, 1, 0);
2457
2458 if (po->tx_ring.pg_vec)
2459 packet_set_ring(sk, &req_u, 1, 1);
2460
2461 fanout_release(sk);
2462
2463 synchronize_net();
2464 /*
2465 * Now the socket is dead. No more input will appear.
2466 */
2467 sock_orphan(sk);
2468 sock->sk = NULL;
2469
2470 /* Purge queues */
2471
2472 skb_queue_purge(&sk->sk_receive_queue);
2473 sk_refcnt_debug_release(sk);
2474
2475 sock_put(sk);
2476 return 0;
2477}
2478
2479/*
2480 * Attach a packet hook.
2481 */
2482
2483static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2484{
2485 struct packet_sock *po = pkt_sk(sk);
2486
2487 if (po->fanout) {
2488 if (dev)
2489 dev_put(dev);
2490
2491 return -EINVAL;
2492 }
2493
2494 lock_sock(sk);
2495
2496 spin_lock(&po->bind_lock);
2497 unregister_prot_hook(sk, true);
2498 po->num = protocol;
2499 po->prot_hook.type = protocol;
2500 if (po->prot_hook.dev)
2501 dev_put(po->prot_hook.dev);
2502 po->prot_hook.dev = dev;
2503
2504 po->ifindex = dev ? dev->ifindex : 0;
2505
2506 if (protocol == 0)
2507 goto out_unlock;
2508
2509 if (!dev || (dev->flags & IFF_UP)) {
2510 register_prot_hook(sk);
2511 } else {
2512 sk->sk_err = ENETDOWN;
2513 if (!sock_flag(sk, SOCK_DEAD))
2514 sk->sk_error_report(sk);
2515 }
2516
2517out_unlock:
2518 spin_unlock(&po->bind_lock);
2519 release_sock(sk);
2520 return 0;
2521}
2522
2523/*
2524 * Bind a packet socket to a device
2525 */
2526
2527static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2528 int addr_len)
2529{
2530 struct sock *sk = sock->sk;
2531 char name[15];
2532 struct net_device *dev;
2533 int err = -ENODEV;
2534
2535 /*
2536 * Check legality
2537 */
2538
2539 if (addr_len != sizeof(struct sockaddr))
2540 return -EINVAL;
2541 strlcpy(name, uaddr->sa_data, sizeof(name));
2542
2543 dev = dev_get_by_name(sock_net(sk), name);
2544 if (dev)
2545 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2546 return err;
2547}
2548
2549static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2550{
2551 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2552 struct sock *sk = sock->sk;
2553 struct net_device *dev = NULL;
2554 int err;
2555
2556
2557 /*
2558 * Check legality
2559 */
2560
2561 if (addr_len < sizeof(struct sockaddr_ll))
2562 return -EINVAL;
2563 if (sll->sll_family != AF_PACKET)
2564 return -EINVAL;
2565
2566 if (sll->sll_ifindex) {
2567 err = -ENODEV;
2568 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2569 if (dev == NULL)
2570 goto out;
2571 }
2572 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2573
2574out:
2575 return err;
2576}
2577
2578static struct proto packet_proto = {
2579 .name = "PACKET",
2580 .owner = THIS_MODULE,
2581 .obj_size = sizeof(struct packet_sock),
2582};
2583
2584/*
2585 * Create a packet of type SOCK_PACKET.
2586 */
2587
2588static int packet_create(struct net *net, struct socket *sock, int protocol,
2589 int kern)
2590{
2591 struct sock *sk;
2592 struct packet_sock *po;
2593 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2594 int err;
2595
2596 if (!capable(CAP_NET_RAW))
2597 return -EPERM;
2598 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2599 sock->type != SOCK_PACKET)
2600 return -ESOCKTNOSUPPORT;
2601
2602 sock->state = SS_UNCONNECTED;
2603
2604 err = -ENOBUFS;
2605 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2606 if (sk == NULL)
2607 goto out;
2608
2609 sock->ops = &packet_ops;
2610 if (sock->type == SOCK_PACKET)
2611 sock->ops = &packet_ops_spkt;
2612
2613 sock_init_data(sock, sk);
2614
2615 po = pkt_sk(sk);
2616 sk->sk_family = PF_PACKET;
2617 po->num = proto;
2618
2619 sk->sk_destruct = packet_sock_destruct;
2620 sk_refcnt_debug_inc(sk);
2621
2622 /*
2623 * Attach a protocol block
2624 */
2625
2626 spin_lock_init(&po->bind_lock);
2627 mutex_init(&po->pg_vec_lock);
2628 po->prot_hook.func = packet_rcv;
2629
2630 if (sock->type == SOCK_PACKET)
2631 po->prot_hook.func = packet_rcv_spkt;
2632
2633 po->prot_hook.af_packet_priv = sk;
2634
2635 if (proto) {
2636 po->prot_hook.type = proto;
2637 register_prot_hook(sk);
2638 }
2639
2640 spin_lock_bh(&net->packet.sklist_lock);
2641 sk_add_node_rcu(sk, &net->packet.sklist);
2642 sock_prot_inuse_add(net, &packet_proto, 1);
2643 spin_unlock_bh(&net->packet.sklist_lock);
2644
2645 return 0;
2646out:
2647 return err;
2648}
2649
2650static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2651{
2652 struct sock_exterr_skb *serr;
2653 struct sk_buff *skb, *skb2;
2654 int copied, err;
2655
2656 err = -EAGAIN;
2657 skb = skb_dequeue(&sk->sk_error_queue);
2658 if (skb == NULL)
2659 goto out;
2660
2661 copied = skb->len;
2662 if (copied > len) {
2663 msg->msg_flags |= MSG_TRUNC;
2664 copied = len;
2665 }
2666 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2667 if (err)
2668 goto out_free_skb;
2669
2670 sock_recv_timestamp(msg, sk, skb);
2671
2672 serr = SKB_EXT_ERR(skb);
2673 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2674 sizeof(serr->ee), &serr->ee);
2675
2676 msg->msg_flags |= MSG_ERRQUEUE;
2677 err = copied;
2678
2679 /* Reset and regenerate socket error */
2680 spin_lock_bh(&sk->sk_error_queue.lock);
2681 sk->sk_err = 0;
2682 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2683 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2684 spin_unlock_bh(&sk->sk_error_queue.lock);
2685 sk->sk_error_report(sk);
2686 } else
2687 spin_unlock_bh(&sk->sk_error_queue.lock);
2688
2689out_free_skb:
2690 kfree_skb(skb);
2691out:
2692 return err;
2693}
2694
2695/*
2696 * Pull a packet from our receive queue and hand it to the user.
2697 * If necessary we block.
2698 */
2699
2700static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2701 struct msghdr *msg, size_t len, int flags)
2702{
2703 struct sock *sk = sock->sk;
2704 struct sk_buff *skb;
2705 int copied, err;
2706 struct sockaddr_ll *sll;
2707 int vnet_hdr_len = 0;
2708
2709 err = -EINVAL;
2710 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2711 goto out;
2712
2713#if 0
2714 /* What error should we return now? EUNATTACH? */
2715 if (pkt_sk(sk)->ifindex < 0)
2716 return -ENODEV;
2717#endif
2718
2719 if (flags & MSG_ERRQUEUE) {
2720 err = packet_recv_error(sk, msg, len);
2721 goto out;
2722 }
2723
2724 /*
2725 * Call the generic datagram receiver. This handles all sorts
2726 * of horrible races and re-entrancy so we can forget about it
2727 * in the protocol layers.
2728 *
2729 * Now it will return ENETDOWN, if device have just gone down,
2730 * but then it will block.
2731 */
2732
2733 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2734
2735 /*
2736 * An error occurred so return it. Because skb_recv_datagram()
2737 * handles the blocking we don't see and worry about blocking
2738 * retries.
2739 */
2740
2741 if (skb == NULL)
2742 goto out;
2743
2744 if (pkt_sk(sk)->has_vnet_hdr) {
2745 struct virtio_net_hdr vnet_hdr = { 0 };
2746
2747 err = -EINVAL;
2748 vnet_hdr_len = sizeof(vnet_hdr);
2749 if (len < vnet_hdr_len)
2750 goto out_free;
2751
2752 len -= vnet_hdr_len;
2753
2754 if (skb_is_gso(skb)) {
2755 struct skb_shared_info *sinfo = skb_shinfo(skb);
2756
2757 /* This is a hint as to how much should be linear. */
2758 vnet_hdr.hdr_len = skb_headlen(skb);
2759 vnet_hdr.gso_size = sinfo->gso_size;
2760 if (sinfo->gso_type & SKB_GSO_TCPV4)
2761 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2762 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2763 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2764 else if (sinfo->gso_type & SKB_GSO_UDP)
2765 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2766 else if (sinfo->gso_type & SKB_GSO_FCOE)
2767 goto out_free;
2768 else
2769 BUG();
2770 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2771 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2772 } else
2773 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2774
2775 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2776 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2777 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2778 vnet_hdr.csum_offset = skb->csum_offset;
2779 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2780 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2781 } /* else everything is zero */
2782
2783 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2784 vnet_hdr_len);
2785 if (err < 0)
2786 goto out_free;
2787 }
2788
2789 /*
2790 * If the address length field is there to be filled in, we fill
2791 * it in now.
2792 */
2793
2794 sll = &PACKET_SKB_CB(skb)->sa.ll;
2795 if (sock->type == SOCK_PACKET)
2796 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2797 else
2798 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2799
2800 /*
2801 * You lose any data beyond the buffer you gave. If it worries a
2802 * user program they can ask the device for its MTU anyway.
2803 */
2804
2805 copied = skb->len;
2806 if (copied > len) {
2807 copied = len;
2808 msg->msg_flags |= MSG_TRUNC;
2809 }
2810
2811 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2812 if (err)
2813 goto out_free;
2814
2815 sock_recv_ts_and_drops(msg, sk, skb);
2816
2817 if (msg->msg_name)
2818 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2819 msg->msg_namelen);
2820
2821 if (pkt_sk(sk)->auxdata) {
2822 struct tpacket_auxdata aux;
2823
2824 aux.tp_status = TP_STATUS_USER;
2825 if (skb->ip_summed == CHECKSUM_PARTIAL)
2826 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2827 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2828 aux.tp_snaplen = skb->len;
2829 aux.tp_mac = 0;
2830 aux.tp_net = skb_network_offset(skb);
2831 if (vlan_tx_tag_present(skb)) {
2832 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2833 aux.tp_status |= TP_STATUS_VLAN_VALID;
2834 } else {
2835 aux.tp_vlan_tci = 0;
2836 }
2837 aux.tp_padding = 0;
2838 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2839 }
2840
2841 /*
2842 * Free or return the buffer as appropriate. Again this
2843 * hides all the races and re-entrancy issues from us.
2844 */
2845 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2846
2847out_free:
2848 skb_free_datagram(sk, skb);
2849out:
2850 return err;
2851}
2852
2853static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2854 int *uaddr_len, int peer)
2855{
2856 struct net_device *dev;
2857 struct sock *sk = sock->sk;
2858
2859 if (peer)
2860 return -EOPNOTSUPP;
2861
2862 uaddr->sa_family = AF_PACKET;
2863 rcu_read_lock();
2864 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2865 if (dev)
2866 strncpy(uaddr->sa_data, dev->name, 14);
2867 else
2868 memset(uaddr->sa_data, 0, 14);
2869 rcu_read_unlock();
2870 *uaddr_len = sizeof(*uaddr);
2871
2872 return 0;
2873}
2874
2875static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2876 int *uaddr_len, int peer)
2877{
2878 struct net_device *dev;
2879 struct sock *sk = sock->sk;
2880 struct packet_sock *po = pkt_sk(sk);
2881 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2882
2883 if (peer)
2884 return -EOPNOTSUPP;
2885
2886 sll->sll_family = AF_PACKET;
2887 sll->sll_ifindex = po->ifindex;
2888 sll->sll_protocol = po->num;
2889 sll->sll_pkttype = 0;
2890 rcu_read_lock();
2891 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2892 if (dev) {
2893 sll->sll_hatype = dev->type;
2894 sll->sll_halen = dev->addr_len;
2895 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2896 } else {
2897 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2898 sll->sll_halen = 0;
2899 }
2900 rcu_read_unlock();
2901 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2902
2903 return 0;
2904}
2905
2906static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2907 int what)
2908{
2909 switch (i->type) {
2910 case PACKET_MR_MULTICAST:
2911 if (i->alen != dev->addr_len)
2912 return -EINVAL;
2913 if (what > 0)
2914 return dev_mc_add(dev, i->addr);
2915 else
2916 return dev_mc_del(dev, i->addr);
2917 break;
2918 case PACKET_MR_PROMISC:
2919 return dev_set_promiscuity(dev, what);
2920 break;
2921 case PACKET_MR_ALLMULTI:
2922 return dev_set_allmulti(dev, what);
2923 break;
2924 case PACKET_MR_UNICAST:
2925 if (i->alen != dev->addr_len)
2926 return -EINVAL;
2927 if (what > 0)
2928 return dev_uc_add(dev, i->addr);
2929 else
2930 return dev_uc_del(dev, i->addr);
2931 break;
2932 default:
2933 break;
2934 }
2935 return 0;
2936}
2937
2938static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2939{
2940 for ( ; i; i = i->next) {
2941 if (i->ifindex == dev->ifindex)
2942 packet_dev_mc(dev, i, what);
2943 }
2944}
2945
2946static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2947{
2948 struct packet_sock *po = pkt_sk(sk);
2949 struct packet_mclist *ml, *i;
2950 struct net_device *dev;
2951 int err;
2952
2953 rtnl_lock();
2954
2955 err = -ENODEV;
2956 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2957 if (!dev)
2958 goto done;
2959
2960 err = -EINVAL;
2961 if (mreq->mr_alen > dev->addr_len)
2962 goto done;
2963
2964 err = -ENOBUFS;
2965 i = kmalloc(sizeof(*i), GFP_KERNEL);
2966 if (i == NULL)
2967 goto done;
2968
2969 err = 0;
2970 for (ml = po->mclist; ml; ml = ml->next) {
2971 if (ml->ifindex == mreq->mr_ifindex &&
2972 ml->type == mreq->mr_type &&
2973 ml->alen == mreq->mr_alen &&
2974 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2975 ml->count++;
2976 /* Free the new element ... */
2977 kfree(i);
2978 goto done;
2979 }
2980 }
2981
2982 i->type = mreq->mr_type;
2983 i->ifindex = mreq->mr_ifindex;
2984 i->alen = mreq->mr_alen;
2985 memcpy(i->addr, mreq->mr_address, i->alen);
2986 i->count = 1;
2987 i->next = po->mclist;
2988 po->mclist = i;
2989 err = packet_dev_mc(dev, i, 1);
2990 if (err) {
2991 po->mclist = i->next;
2992 kfree(i);
2993 }
2994
2995done:
2996 rtnl_unlock();
2997 return err;
2998}
2999
3000static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3001{
3002 struct packet_mclist *ml, **mlp;
3003
3004 rtnl_lock();
3005
3006 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3007 if (ml->ifindex == mreq->mr_ifindex &&
3008 ml->type == mreq->mr_type &&
3009 ml->alen == mreq->mr_alen &&
3010 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3011 if (--ml->count == 0) {
3012 struct net_device *dev;
3013 *mlp = ml->next;
3014 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3015 if (dev)
3016 packet_dev_mc(dev, ml, -1);
3017 kfree(ml);
3018 }
3019 rtnl_unlock();
3020 return 0;
3021 }
3022 }
3023 rtnl_unlock();
3024 return -EADDRNOTAVAIL;
3025}
3026
3027static void packet_flush_mclist(struct sock *sk)
3028{
3029 struct packet_sock *po = pkt_sk(sk);
3030 struct packet_mclist *ml;
3031
3032 if (!po->mclist)
3033 return;
3034
3035 rtnl_lock();
3036 while ((ml = po->mclist) != NULL) {
3037 struct net_device *dev;
3038
3039 po->mclist = ml->next;
3040 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3041 if (dev != NULL)
3042 packet_dev_mc(dev, ml, -1);
3043 kfree(ml);
3044 }
3045 rtnl_unlock();
3046}
3047
3048static int
3049packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3050{
3051 struct sock *sk = sock->sk;
3052 struct packet_sock *po = pkt_sk(sk);
3053 int ret;
3054
3055 if (level != SOL_PACKET)
3056 return -ENOPROTOOPT;
3057
3058 switch (optname) {
3059 case PACKET_ADD_MEMBERSHIP:
3060 case PACKET_DROP_MEMBERSHIP:
3061 {
3062 struct packet_mreq_max mreq;
3063 int len = optlen;
3064 memset(&mreq, 0, sizeof(mreq));
3065 if (len < sizeof(struct packet_mreq))
3066 return -EINVAL;
3067 if (len > sizeof(mreq))
3068 len = sizeof(mreq);
3069 if (copy_from_user(&mreq, optval, len))
3070 return -EFAULT;
3071 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3072 return -EINVAL;
3073 if (optname == PACKET_ADD_MEMBERSHIP)
3074 ret = packet_mc_add(sk, &mreq);
3075 else
3076 ret = packet_mc_drop(sk, &mreq);
3077 return ret;
3078 }
3079
3080 case PACKET_RX_RING:
3081 case PACKET_TX_RING:
3082 {
3083 union tpacket_req_u req_u;
3084 int len;
3085
3086 switch (po->tp_version) {
3087 case TPACKET_V1:
3088 case TPACKET_V2:
3089 len = sizeof(req_u.req);
3090 break;
3091 case TPACKET_V3:
3092 default:
3093 len = sizeof(req_u.req3);
3094 break;
3095 }
3096 if (optlen < len)
3097 return -EINVAL;
3098 if (pkt_sk(sk)->has_vnet_hdr)
3099 return -EINVAL;
3100 if (copy_from_user(&req_u.req, optval, len))
3101 return -EFAULT;
3102 return packet_set_ring(sk, &req_u, 0,
3103 optname == PACKET_TX_RING);
3104 }
3105 case PACKET_COPY_THRESH:
3106 {
3107 int val;
3108
3109 if (optlen != sizeof(val))
3110 return -EINVAL;
3111 if (copy_from_user(&val, optval, sizeof(val)))
3112 return -EFAULT;
3113
3114 pkt_sk(sk)->copy_thresh = val;
3115 return 0;
3116 }
3117 case PACKET_VERSION:
3118 {
3119 int val;
3120
3121 if (optlen != sizeof(val))
3122 return -EINVAL;
3123 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3124 return -EBUSY;
3125 if (copy_from_user(&val, optval, sizeof(val)))
3126 return -EFAULT;
3127 switch (val) {
3128 case TPACKET_V1:
3129 case TPACKET_V2:
3130 case TPACKET_V3:
3131 po->tp_version = val;
3132 return 0;
3133 default:
3134 return -EINVAL;
3135 }
3136 }
3137 case PACKET_RESERVE:
3138 {
3139 unsigned int val;
3140
3141 if (optlen != sizeof(val))
3142 return -EINVAL;
3143 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3144 return -EBUSY;
3145 if (copy_from_user(&val, optval, sizeof(val)))
3146 return -EFAULT;
3147 po->tp_reserve = val;
3148 return 0;
3149 }
3150 case PACKET_LOSS:
3151 {
3152 unsigned int val;
3153
3154 if (optlen != sizeof(val))
3155 return -EINVAL;
3156 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3157 return -EBUSY;
3158 if (copy_from_user(&val, optval, sizeof(val)))
3159 return -EFAULT;
3160 po->tp_loss = !!val;
3161 return 0;
3162 }
3163 case PACKET_AUXDATA:
3164 {
3165 int val;
3166
3167 if (optlen < sizeof(val))
3168 return -EINVAL;
3169 if (copy_from_user(&val, optval, sizeof(val)))
3170 return -EFAULT;
3171
3172 po->auxdata = !!val;
3173 return 0;
3174 }
3175 case PACKET_ORIGDEV:
3176 {
3177 int val;
3178
3179 if (optlen < sizeof(val))
3180 return -EINVAL;
3181 if (copy_from_user(&val, optval, sizeof(val)))
3182 return -EFAULT;
3183
3184 po->origdev = !!val;
3185 return 0;
3186 }
3187 case PACKET_VNET_HDR:
3188 {
3189 int val;
3190
3191 if (sock->type != SOCK_RAW)
3192 return -EINVAL;
3193 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3194 return -EBUSY;
3195 if (optlen < sizeof(val))
3196 return -EINVAL;
3197 if (copy_from_user(&val, optval, sizeof(val)))
3198 return -EFAULT;
3199
3200 po->has_vnet_hdr = !!val;
3201 return 0;
3202 }
3203 case PACKET_TIMESTAMP:
3204 {
3205 int val;
3206
3207 if (optlen != sizeof(val))
3208 return -EINVAL;
3209 if (copy_from_user(&val, optval, sizeof(val)))
3210 return -EFAULT;
3211
3212 po->tp_tstamp = val;
3213 return 0;
3214 }
3215 case PACKET_FANOUT:
3216 {
3217 int val;
3218
3219 if (optlen != sizeof(val))
3220 return -EINVAL;
3221 if (copy_from_user(&val, optval, sizeof(val)))
3222 return -EFAULT;
3223
3224 return fanout_add(sk, val & 0xffff, val >> 16);
3225 }
3226 default:
3227 return -ENOPROTOOPT;
3228 }
3229}
3230
3231static int packet_getsockopt(struct socket *sock, int level, int optname,
3232 char __user *optval, int __user *optlen)
3233{
3234 int len;
3235 int val, lv = sizeof(val);
3236 struct sock *sk = sock->sk;
3237 struct packet_sock *po = pkt_sk(sk);
3238 void *data = &val;
3239 struct tpacket_stats st;
3240 union tpacket_stats_u st_u;
3241
3242 if (level != SOL_PACKET)
3243 return -ENOPROTOOPT;
3244
3245 if (get_user(len, optlen))
3246 return -EFAULT;
3247
3248 if (len < 0)
3249 return -EINVAL;
3250
3251 switch (optname) {
3252 case PACKET_STATISTICS:
3253 spin_lock_bh(&sk->sk_receive_queue.lock);
3254 if (po->tp_version == TPACKET_V3) {
3255 lv = sizeof(struct tpacket_stats_v3);
3256 memcpy(&st_u.stats3, &po->stats,
3257 sizeof(struct tpacket_stats));
3258 st_u.stats3.tp_freeze_q_cnt =
3259 po->stats_u.stats3.tp_freeze_q_cnt;
3260 st_u.stats3.tp_packets += po->stats.tp_drops;
3261 data = &st_u.stats3;
3262 } else {
3263 lv = sizeof(struct tpacket_stats);
3264 st = po->stats;
3265 st.tp_packets += st.tp_drops;
3266 data = &st;
3267 }
3268 memset(&po->stats, 0, sizeof(st));
3269 spin_unlock_bh(&sk->sk_receive_queue.lock);
3270 break;
3271 case PACKET_AUXDATA:
3272 val = po->auxdata;
3273 break;
3274 case PACKET_ORIGDEV:
3275 val = po->origdev;
3276 break;
3277 case PACKET_VNET_HDR:
3278 val = po->has_vnet_hdr;
3279 break;
3280 case PACKET_VERSION:
3281 val = po->tp_version;
3282 break;
3283 case PACKET_HDRLEN:
3284 if (len > sizeof(int))
3285 len = sizeof(int);
3286 if (copy_from_user(&val, optval, len))
3287 return -EFAULT;
3288 switch (val) {
3289 case TPACKET_V1:
3290 val = sizeof(struct tpacket_hdr);
3291 break;
3292 case TPACKET_V2:
3293 val = sizeof(struct tpacket2_hdr);
3294 break;
3295 case TPACKET_V3:
3296 val = sizeof(struct tpacket3_hdr);
3297 break;
3298 default:
3299 return -EINVAL;
3300 }
3301 break;
3302 case PACKET_RESERVE:
3303 val = po->tp_reserve;
3304 break;
3305 case PACKET_LOSS:
3306 val = po->tp_loss;
3307 break;
3308 case PACKET_TIMESTAMP:
3309 val = po->tp_tstamp;
3310 break;
3311 case PACKET_FANOUT:
3312 val = (po->fanout ?
3313 ((u32)po->fanout->id |
3314 ((u32)po->fanout->type << 16)) :
3315 0);
3316 break;
3317 default:
3318 return -ENOPROTOOPT;
3319 }
3320
3321 if (len > lv)
3322 len = lv;
3323 if (put_user(len, optlen))
3324 return -EFAULT;
3325 if (copy_to_user(optval, data, len))
3326 return -EFAULT;
3327 return 0;
3328}
3329
3330
3331static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3332{
3333 struct sock *sk;
3334 struct hlist_node *node;
3335 struct net_device *dev = data;
3336 struct net *net = dev_net(dev);
3337
3338 rcu_read_lock();
3339 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3340 struct packet_sock *po = pkt_sk(sk);
3341
3342 switch (msg) {
3343 case NETDEV_UNREGISTER:
3344 if (po->mclist)
3345 packet_dev_mclist(dev, po->mclist, -1);
3346 /* fallthrough */
3347
3348 case NETDEV_DOWN:
3349 if (dev->ifindex == po->ifindex) {
3350 spin_lock(&po->bind_lock);
3351 if (po->running) {
3352 __unregister_prot_hook(sk, false);
3353 sk->sk_err = ENETDOWN;
3354 if (!sock_flag(sk, SOCK_DEAD))
3355 sk->sk_error_report(sk);
3356 }
3357 if (msg == NETDEV_UNREGISTER) {
3358 po->ifindex = -1;
3359 if (po->prot_hook.dev)
3360 dev_put(po->prot_hook.dev);
3361 po->prot_hook.dev = NULL;
3362 }
3363 spin_unlock(&po->bind_lock);
3364 }
3365 break;
3366 case NETDEV_UP:
3367 if (dev->ifindex == po->ifindex) {
3368 spin_lock(&po->bind_lock);
3369 if (po->num)
3370 register_prot_hook(sk);
3371 spin_unlock(&po->bind_lock);
3372 }
3373 break;
3374 }
3375 }
3376 rcu_read_unlock();
3377 return NOTIFY_DONE;
3378}
3379
3380
3381static int packet_ioctl(struct socket *sock, unsigned int cmd,
3382 unsigned long arg)
3383{
3384 struct sock *sk = sock->sk;
3385
3386 switch (cmd) {
3387 case SIOCOUTQ:
3388 {
3389 int amount = sk_wmem_alloc_get(sk);
3390
3391 return put_user(amount, (int __user *)arg);
3392 }
3393 case SIOCINQ:
3394 {
3395 struct sk_buff *skb;
3396 int amount = 0;
3397
3398 spin_lock_bh(&sk->sk_receive_queue.lock);
3399 skb = skb_peek(&sk->sk_receive_queue);
3400 if (skb)
3401 amount = skb->len;
3402 spin_unlock_bh(&sk->sk_receive_queue.lock);
3403 return put_user(amount, (int __user *)arg);
3404 }
3405 case SIOCGSTAMP:
3406 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3407 case SIOCGSTAMPNS:
3408 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3409
3410#ifdef CONFIG_INET
3411 case SIOCADDRT:
3412 case SIOCDELRT:
3413 case SIOCDARP:
3414 case SIOCGARP:
3415 case SIOCSARP:
3416 case SIOCGIFADDR:
3417 case SIOCSIFADDR:
3418 case SIOCGIFBRDADDR:
3419 case SIOCSIFBRDADDR:
3420 case SIOCGIFNETMASK:
3421 case SIOCSIFNETMASK:
3422 case SIOCGIFDSTADDR:
3423 case SIOCSIFDSTADDR:
3424 case SIOCSIFFLAGS:
3425 return inet_dgram_ops.ioctl(sock, cmd, arg);
3426#endif
3427
3428 default:
3429 return -ENOIOCTLCMD;
3430 }
3431 return 0;
3432}
3433
3434static unsigned int packet_poll(struct file *file, struct socket *sock,
3435 poll_table *wait)
3436{
3437 struct sock *sk = sock->sk;
3438 struct packet_sock *po = pkt_sk(sk);
3439 unsigned int mask = datagram_poll(file, sock, wait);
3440
3441 spin_lock_bh(&sk->sk_receive_queue.lock);
3442 if (po->rx_ring.pg_vec) {
3443 if (!packet_previous_rx_frame(po, &po->rx_ring,
3444 TP_STATUS_KERNEL))
3445 mask |= POLLIN | POLLRDNORM;
3446 }
3447 spin_unlock_bh(&sk->sk_receive_queue.lock);
3448 spin_lock_bh(&sk->sk_write_queue.lock);
3449 if (po->tx_ring.pg_vec) {
3450 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3451 mask |= POLLOUT | POLLWRNORM;
3452 }
3453 spin_unlock_bh(&sk->sk_write_queue.lock);
3454 return mask;
3455}
3456
3457
3458/* Dirty? Well, I still did not learn better way to account
3459 * for user mmaps.
3460 */
3461
3462static void packet_mm_open(struct vm_area_struct *vma)
3463{
3464 struct file *file = vma->vm_file;
3465 struct socket *sock = file->private_data;
3466 struct sock *sk = sock->sk;
3467
3468 if (sk)
3469 atomic_inc(&pkt_sk(sk)->mapped);
3470}
3471
3472static void packet_mm_close(struct vm_area_struct *vma)
3473{
3474 struct file *file = vma->vm_file;
3475 struct socket *sock = file->private_data;
3476 struct sock *sk = sock->sk;
3477
3478 if (sk)
3479 atomic_dec(&pkt_sk(sk)->mapped);
3480}
3481
3482static const struct vm_operations_struct packet_mmap_ops = {
3483 .open = packet_mm_open,
3484 .close = packet_mm_close,
3485};
3486
3487static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3488 unsigned int len)
3489{
3490 int i;
3491
3492 for (i = 0; i < len; i++) {
3493 if (likely(pg_vec[i].buffer)) {
3494 if (is_vmalloc_addr(pg_vec[i].buffer))
3495 vfree(pg_vec[i].buffer);
3496 else
3497 free_pages((unsigned long)pg_vec[i].buffer,
3498 order);
3499 pg_vec[i].buffer = NULL;
3500 }
3501 }
3502 kfree(pg_vec);
3503}
3504
3505static char *alloc_one_pg_vec_page(unsigned long order)
3506{
3507 char *buffer = NULL;
3508 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3509 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3510
3511 buffer = (char *) __get_free_pages(gfp_flags, order);
3512
3513 if (buffer)
3514 return buffer;
3515
3516 /*
3517 * __get_free_pages failed, fall back to vmalloc
3518 */
3519 buffer = vzalloc((1 << order) * PAGE_SIZE);
3520
3521 if (buffer)
3522 return buffer;
3523
3524 /*
3525 * vmalloc failed, lets dig into swap here
3526 */
3527 gfp_flags &= ~__GFP_NORETRY;
3528 buffer = (char *)__get_free_pages(gfp_flags, order);
3529 if (buffer)
3530 return buffer;
3531
3532 /*
3533 * complete and utter failure
3534 */
3535 return NULL;
3536}
3537
3538static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3539{
3540 unsigned int block_nr = req->tp_block_nr;
3541 struct pgv *pg_vec;
3542 int i;
3543
3544 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3545 if (unlikely(!pg_vec))
3546 goto out;
3547
3548 for (i = 0; i < block_nr; i++) {
3549 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3550 if (unlikely(!pg_vec[i].buffer))
3551 goto out_free_pgvec;
3552 }
3553
3554out:
3555 return pg_vec;
3556
3557out_free_pgvec:
3558 free_pg_vec(pg_vec, order, block_nr);
3559 pg_vec = NULL;
3560 goto out;
3561}
3562
3563static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3564 int closing, int tx_ring)
3565{
3566 struct pgv *pg_vec = NULL;
3567 struct packet_sock *po = pkt_sk(sk);
3568 int was_running, order = 0;
3569 struct packet_ring_buffer *rb;
3570 struct sk_buff_head *rb_queue;
3571 __be16 num;
3572 int err = -EINVAL;
3573 /* Added to avoid minimal code churn */
3574 struct tpacket_req *req = &req_u->req;
3575
3576 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3577 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3578 WARN(1, "Tx-ring is not supported.\n");
3579 goto out;
3580 }
3581
3582 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3583 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3584
3585 err = -EBUSY;
3586 if (!closing) {
3587 if (atomic_read(&po->mapped))
3588 goto out;
3589 if (atomic_read(&rb->pending))
3590 goto out;
3591 }
3592
3593 if (req->tp_block_nr) {
3594 /* Sanity tests and some calculations */
3595 err = -EBUSY;
3596 if (unlikely(rb->pg_vec))
3597 goto out;
3598
3599 switch (po->tp_version) {
3600 case TPACKET_V1:
3601 po->tp_hdrlen = TPACKET_HDRLEN;
3602 break;
3603 case TPACKET_V2:
3604 po->tp_hdrlen = TPACKET2_HDRLEN;
3605 break;
3606 case TPACKET_V3:
3607 po->tp_hdrlen = TPACKET3_HDRLEN;
3608 break;
3609 }
3610
3611 err = -EINVAL;
3612 if (unlikely((int)req->tp_block_size <= 0))
3613 goto out;
3614 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3615 goto out;
3616 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3617 po->tp_reserve))
3618 goto out;
3619 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3620 goto out;
3621
3622 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3623 if (unlikely(rb->frames_per_block <= 0))
3624 goto out;
3625 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3626 req->tp_frame_nr))
3627 goto out;
3628
3629 err = -ENOMEM;
3630 order = get_order(req->tp_block_size);
3631 pg_vec = alloc_pg_vec(req, order);
3632 if (unlikely(!pg_vec))
3633 goto out;
3634 switch (po->tp_version) {
3635 case TPACKET_V3:
3636 /* Transmit path is not supported. We checked
3637 * it above but just being paranoid
3638 */
3639 if (!tx_ring)
3640 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3641 break;
3642 default:
3643 break;
3644 }
3645 }
3646 /* Done */
3647 else {
3648 err = -EINVAL;
3649 if (unlikely(req->tp_frame_nr))
3650 goto out;
3651 }
3652
3653 lock_sock(sk);
3654
3655 /* Detach socket from network */
3656 spin_lock(&po->bind_lock);
3657 was_running = po->running;
3658 num = po->num;
3659 if (was_running) {
3660 po->num = 0;
3661 __unregister_prot_hook(sk, false);
3662 }
3663 spin_unlock(&po->bind_lock);
3664
3665 synchronize_net();
3666
3667 err = -EBUSY;
3668 mutex_lock(&po->pg_vec_lock);
3669 if (closing || atomic_read(&po->mapped) == 0) {
3670 err = 0;
3671 spin_lock_bh(&rb_queue->lock);
3672 swap(rb->pg_vec, pg_vec);
3673 rb->frame_max = (req->tp_frame_nr - 1);
3674 rb->head = 0;
3675 rb->frame_size = req->tp_frame_size;
3676 spin_unlock_bh(&rb_queue->lock);
3677
3678 swap(rb->pg_vec_order, order);
3679 swap(rb->pg_vec_len, req->tp_block_nr);
3680
3681 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3682 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3683 tpacket_rcv : packet_rcv;
3684 skb_queue_purge(rb_queue);
3685 if (atomic_read(&po->mapped))
3686 pr_err("packet_mmap: vma is busy: %d\n",
3687 atomic_read(&po->mapped));
3688 }
3689 mutex_unlock(&po->pg_vec_lock);
3690
3691 spin_lock(&po->bind_lock);
3692 if (was_running) {
3693 po->num = num;
3694 register_prot_hook(sk);
3695 }
3696 spin_unlock(&po->bind_lock);
3697 if (closing && (po->tp_version > TPACKET_V2)) {
3698 /* Because we don't support block-based V3 on tx-ring */
3699 if (!tx_ring)
3700 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3701 }
3702 release_sock(sk);
3703
3704 if (pg_vec)
3705 free_pg_vec(pg_vec, order, req->tp_block_nr);
3706out:
3707 return err;
3708}
3709
3710static int packet_mmap(struct file *file, struct socket *sock,
3711 struct vm_area_struct *vma)
3712{
3713 struct sock *sk = sock->sk;
3714 struct packet_sock *po = pkt_sk(sk);
3715 unsigned long size, expected_size;
3716 struct packet_ring_buffer *rb;
3717 unsigned long start;
3718 int err = -EINVAL;
3719 int i;
3720
3721 if (vma->vm_pgoff)
3722 return -EINVAL;
3723
3724 mutex_lock(&po->pg_vec_lock);
3725
3726 expected_size = 0;
3727 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3728 if (rb->pg_vec) {
3729 expected_size += rb->pg_vec_len
3730 * rb->pg_vec_pages
3731 * PAGE_SIZE;
3732 }
3733 }
3734
3735 if (expected_size == 0)
3736 goto out;
3737
3738 size = vma->vm_end - vma->vm_start;
3739 if (size != expected_size)
3740 goto out;
3741
3742 start = vma->vm_start;
3743 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3744 if (rb->pg_vec == NULL)
3745 continue;
3746
3747 for (i = 0; i < rb->pg_vec_len; i++) {
3748 struct page *page;
3749 void *kaddr = rb->pg_vec[i].buffer;
3750 int pg_num;
3751
3752 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3753 page = pgv_to_page(kaddr);
3754 err = vm_insert_page(vma, start, page);
3755 if (unlikely(err))
3756 goto out;
3757 start += PAGE_SIZE;
3758 kaddr += PAGE_SIZE;
3759 }
3760 }
3761 }
3762
3763 atomic_inc(&po->mapped);
3764 vma->vm_ops = &packet_mmap_ops;
3765 err = 0;
3766
3767out:
3768 mutex_unlock(&po->pg_vec_lock);
3769 return err;
3770}
3771
3772static const struct proto_ops packet_ops_spkt = {
3773 .family = PF_PACKET,
3774 .owner = THIS_MODULE,
3775 .release = packet_release,
3776 .bind = packet_bind_spkt,
3777 .connect = sock_no_connect,
3778 .socketpair = sock_no_socketpair,
3779 .accept = sock_no_accept,
3780 .getname = packet_getname_spkt,
3781 .poll = datagram_poll,
3782 .ioctl = packet_ioctl,
3783 .listen = sock_no_listen,
3784 .shutdown = sock_no_shutdown,
3785 .setsockopt = sock_no_setsockopt,
3786 .getsockopt = sock_no_getsockopt,
3787 .sendmsg = packet_sendmsg_spkt,
3788 .recvmsg = packet_recvmsg,
3789 .mmap = sock_no_mmap,
3790 .sendpage = sock_no_sendpage,
3791};
3792
3793static const struct proto_ops packet_ops = {
3794 .family = PF_PACKET,
3795 .owner = THIS_MODULE,
3796 .release = packet_release,
3797 .bind = packet_bind,
3798 .connect = sock_no_connect,
3799 .socketpair = sock_no_socketpair,
3800 .accept = sock_no_accept,
3801 .getname = packet_getname,
3802 .poll = packet_poll,
3803 .ioctl = packet_ioctl,
3804 .listen = sock_no_listen,
3805 .shutdown = sock_no_shutdown,
3806 .setsockopt = packet_setsockopt,
3807 .getsockopt = packet_getsockopt,
3808 .sendmsg = packet_sendmsg,
3809 .recvmsg = packet_recvmsg,
3810 .mmap = packet_mmap,
3811 .sendpage = sock_no_sendpage,
3812};
3813
3814static const struct net_proto_family packet_family_ops = {
3815 .family = PF_PACKET,
3816 .create = packet_create,
3817 .owner = THIS_MODULE,
3818};
3819
3820static struct notifier_block packet_netdev_notifier = {
3821 .notifier_call = packet_notifier,
3822};
3823
3824#ifdef CONFIG_PROC_FS
3825
3826static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3827 __acquires(RCU)
3828{
3829 struct net *net = seq_file_net(seq);
3830
3831 rcu_read_lock();
3832 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3833}
3834
3835static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3836{
3837 struct net *net = seq_file_net(seq);
3838 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3839}
3840
3841static void packet_seq_stop(struct seq_file *seq, void *v)
3842 __releases(RCU)
3843{
3844 rcu_read_unlock();
3845}
3846
3847static int packet_seq_show(struct seq_file *seq, void *v)
3848{
3849 if (v == SEQ_START_TOKEN)
3850 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3851 else {
3852 struct sock *s = sk_entry(v);
3853 const struct packet_sock *po = pkt_sk(s);
3854
3855 seq_printf(seq,
3856 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3857 s,
3858 atomic_read(&s->sk_refcnt),
3859 s->sk_type,
3860 ntohs(po->num),
3861 po->ifindex,
3862 po->running,
3863 atomic_read(&s->sk_rmem_alloc),
3864 sock_i_uid(s),
3865 sock_i_ino(s));
3866 }
3867
3868 return 0;
3869}
3870
3871static const struct seq_operations packet_seq_ops = {
3872 .start = packet_seq_start,
3873 .next = packet_seq_next,
3874 .stop = packet_seq_stop,
3875 .show = packet_seq_show,
3876};
3877
3878static int packet_seq_open(struct inode *inode, struct file *file)
3879{
3880 return seq_open_net(inode, file, &packet_seq_ops,
3881 sizeof(struct seq_net_private));
3882}
3883
3884static const struct file_operations packet_seq_fops = {
3885 .owner = THIS_MODULE,
3886 .open = packet_seq_open,
3887 .read = seq_read,
3888 .llseek = seq_lseek,
3889 .release = seq_release_net,
3890};
3891
3892#endif
3893
3894static int __net_init packet_net_init(struct net *net)
3895{
3896 spin_lock_init(&net->packet.sklist_lock);
3897 INIT_HLIST_HEAD(&net->packet.sklist);
3898
3899 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3900 return -ENOMEM;
3901
3902 return 0;
3903}
3904
3905static void __net_exit packet_net_exit(struct net *net)
3906{
3907 proc_net_remove(net, "packet");
3908}
3909
3910static struct pernet_operations packet_net_ops = {
3911 .init = packet_net_init,
3912 .exit = packet_net_exit,
3913};
3914
3915
3916static void __exit packet_exit(void)
3917{
3918 unregister_netdevice_notifier(&packet_netdev_notifier);
3919 unregister_pernet_subsys(&packet_net_ops);
3920 sock_unregister(PF_PACKET);
3921 proto_unregister(&packet_proto);
3922}
3923
3924static int __init packet_init(void)
3925{
3926 int rc = proto_register(&packet_proto, 0);
3927
3928 if (rc != 0)
3929 goto out;
3930
3931 sock_register(&packet_family_ops);
3932 register_pernet_subsys(&packet_net_ops);
3933 register_netdevice_notifier(&packet_netdev_notifier);
3934out:
3935 return rc;
3936}
3937
3938module_init(packet_init);
3939module_exit(packet_exit);
3940MODULE_LICENSE("GPL");
3941MODULE_ALIAS_NETPROTO(PF_PACKET);