projects / deliverable / linux.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
net: call dev_queue_xmit_nit() after skb_dst_drop()
[deliverable/linux.git] / net / packet / af_packet.c
... / ...
CommitLineData
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 *
44 * This program is free software; you can redistribute it and/or
45 * modify it under the terms of the GNU General Public License
46 * as published by the Free Software Foundation; either version
47 * 2 of the License, or (at your option) any later version.
48 *
49 */
50
51#include <linux/types.h>
52#include <linux/mm.h>
53#include <linux/capability.h>
54#include <linux/fcntl.h>
55#include <linux/socket.h>
56#include <linux/in.h>
57#include <linux/inet.h>
58#include <linux/netdevice.h>
59#include <linux/if_packet.h>
60#include <linux/wireless.h>
61#include <linux/kernel.h>
62#include <linux/kmod.h>
63#include <linux/slab.h>
64#include <linux/vmalloc.h>
65#include <net/net_namespace.h>
66#include <net/ip.h>
67#include <net/protocol.h>
68#include <linux/skbuff.h>
69#include <net/sock.h>
70#include <linux/errno.h>
71#include <linux/timer.h>
72#include <asm/system.h>
73#include <asm/uaccess.h>
74#include <asm/ioctls.h>
75#include <asm/page.h>
76#include <asm/cacheflush.h>
77#include <asm/io.h>
78#include <linux/proc_fs.h>
79#include <linux/seq_file.h>
80#include <linux/poll.h>
81#include <linux/module.h>
82#include <linux/init.h>
83#include <linux/mutex.h>
84#include <linux/if_vlan.h>
85#include <linux/virtio_net.h>
86#include <linux/errqueue.h>
87#include <linux/net_tstamp.h>
88
89#ifdef CONFIG_INET
90#include <net/inet_common.h>
91#endif
92
93/*
94 Assumptions:
95 - if device has no dev->hard_header routine, it adds and removes ll header
96 inside itself. In this case ll header is invisible outside of device,
97 but higher levels still should reserve dev->hard_header_len.
98 Some devices are enough clever to reallocate skb, when header
99 will not fit to reserved space (tunnel), another ones are silly
100 (PPP).
101 - packet socket receives packets with pulled ll header,
102 so that SOCK_RAW should push it back.
103
104On receive:
105-----------
106
107Incoming, dev->hard_header!=NULL
108 mac_header -> ll header
109 data -> data
110
111Outgoing, dev->hard_header!=NULL
112 mac_header -> ll header
113 data -> ll header
114
115Incoming, dev->hard_header==NULL
116 mac_header -> UNKNOWN position. It is very likely, that it points to ll
117 header. PPP makes it, that is wrong, because introduce
118 assymetry between rx and tx paths.
119 data -> data
120
121Outgoing, dev->hard_header==NULL
122 mac_header -> data. ll header is still not built!
123 data -> data
124
125Resume
126 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
127
128
129On transmit:
130------------
131
132dev->hard_header != NULL
133 mac_header -> ll header
134 data -> ll header
135
136dev->hard_header == NULL (ll header is added by device, we cannot control it)
137 mac_header -> data
138 data -> data
139
140 We should set nh.raw on output to correct posistion,
141 packet classifier depends on it.
142 */
143
144/* Private packet socket structures. */
145
146struct packet_mclist {
147 struct packet_mclist *next;
148 int ifindex;
149 int count;
150 unsigned short type;
151 unsigned short alen;
152 unsigned char addr[MAX_ADDR_LEN];
153};
154/* identical to struct packet_mreq except it has
155 * a longer address field.
156 */
157struct packet_mreq_max {
158 int mr_ifindex;
159 unsigned short mr_type;
160 unsigned short mr_alen;
161 unsigned char mr_address[MAX_ADDR_LEN];
162};
163
164static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
165 int closing, int tx_ring);
166
167#define PGV_FROM_VMALLOC 1
168struct pgv {
169 char *buffer;
170};
171
172struct packet_ring_buffer {
173 struct pgv *pg_vec;
174 unsigned int head;
175 unsigned int frames_per_block;
176 unsigned int frame_size;
177 unsigned int frame_max;
178
179 unsigned int pg_vec_order;
180 unsigned int pg_vec_pages;
181 unsigned int pg_vec_len;
182
183 atomic_t pending;
184};
185
186struct packet_sock;
187static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
188
189static void packet_flush_mclist(struct sock *sk);
190
191struct packet_sock {
192 /* struct sock has to be the first member of packet_sock */
193 struct sock sk;
194 struct tpacket_stats stats;
195 struct packet_ring_buffer rx_ring;
196 struct packet_ring_buffer tx_ring;
197 int copy_thresh;
198 spinlock_t bind_lock;
199 struct mutex pg_vec_lock;
200 unsigned int running:1, /* prot_hook is attached*/
201 auxdata:1,
202 origdev:1,
203 has_vnet_hdr:1;
204 int ifindex; /* bound device */
205 __be16 num;
206 struct packet_mclist *mclist;
207 atomic_t mapped;
208 enum tpacket_versions tp_version;
209 unsigned int tp_hdrlen;
210 unsigned int tp_reserve;
211 unsigned int tp_loss:1;
212 unsigned int tp_tstamp;
213 struct packet_type prot_hook ____cacheline_aligned_in_smp;
214};
215
216struct packet_skb_cb {
217 unsigned int origlen;
218 union {
219 struct sockaddr_pkt pkt;
220 struct sockaddr_ll ll;
221 } sa;
222};
223
224#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
225
226static inline struct page *pgv_to_page(void *addr)
227{
228 if (is_vmalloc_addr(addr))
229 return vmalloc_to_page(addr);
230 return virt_to_page(addr);
231}
232
233static void __packet_set_status(struct packet_sock *po, void *frame, int status)
234{
235 union {
236 struct tpacket_hdr *h1;
237 struct tpacket2_hdr *h2;
238 void *raw;
239 } h;
240
241 h.raw = frame;
242 switch (po->tp_version) {
243 case TPACKET_V1:
244 h.h1->tp_status = status;
245 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
246 break;
247 case TPACKET_V2:
248 h.h2->tp_status = status;
249 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
250 break;
251 default:
252 pr_err("TPACKET version not supported\n");
253 BUG();
254 }
255
256 smp_wmb();
257}
258
259static int __packet_get_status(struct packet_sock *po, void *frame)
260{
261 union {
262 struct tpacket_hdr *h1;
263 struct tpacket2_hdr *h2;
264 void *raw;
265 } h;
266
267 smp_rmb();
268
269 h.raw = frame;
270 switch (po->tp_version) {
271 case TPACKET_V1:
272 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
273 return h.h1->tp_status;
274 case TPACKET_V2:
275 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
276 return h.h2->tp_status;
277 default:
278 pr_err("TPACKET version not supported\n");
279 BUG();
280 return 0;
281 }
282}
283
284static void *packet_lookup_frame(struct packet_sock *po,
285 struct packet_ring_buffer *rb,
286 unsigned int position,
287 int status)
288{
289 unsigned int pg_vec_pos, frame_offset;
290 union {
291 struct tpacket_hdr *h1;
292 struct tpacket2_hdr *h2;
293 void *raw;
294 } h;
295
296 pg_vec_pos = position / rb->frames_per_block;
297 frame_offset = position % rb->frames_per_block;
298
299 h.raw = rb->pg_vec[pg_vec_pos].buffer +
300 (frame_offset * rb->frame_size);
301
302 if (status != __packet_get_status(po, h.raw))
303 return NULL;
304
305 return h.raw;
306}
307
308static inline void *packet_current_frame(struct packet_sock *po,
309 struct packet_ring_buffer *rb,
310 int status)
311{
312 return packet_lookup_frame(po, rb, rb->head, status);
313}
314
315static inline void *packet_previous_frame(struct packet_sock *po,
316 struct packet_ring_buffer *rb,
317 int status)
318{
319 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
320 return packet_lookup_frame(po, rb, previous, status);
321}
322
323static inline void packet_increment_head(struct packet_ring_buffer *buff)
324{
325 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
326}
327
328static inline struct packet_sock *pkt_sk(struct sock *sk)
329{
330 return (struct packet_sock *)sk;
331}
332
333static void packet_sock_destruct(struct sock *sk)
334{
335 skb_queue_purge(&sk->sk_error_queue);
336
337 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
338 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
339
340 if (!sock_flag(sk, SOCK_DEAD)) {
341 pr_err("Attempt to release alive packet socket: %p\n", sk);
342 return;
343 }
344
345 sk_refcnt_debug_dec(sk);
346}
347
348
349static const struct proto_ops packet_ops;
350
351static const struct proto_ops packet_ops_spkt;
352
353static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
354 struct packet_type *pt, struct net_device *orig_dev)
355{
356 struct sock *sk;
357 struct sockaddr_pkt *spkt;
358
359 /*
360 * When we registered the protocol we saved the socket in the data
361 * field for just this event.
362 */
363
364 sk = pt->af_packet_priv;
365
366 /*
367 * Yank back the headers [hope the device set this
368 * right or kerboom...]
369 *
370 * Incoming packets have ll header pulled,
371 * push it back.
372 *
373 * For outgoing ones skb->data == skb_mac_header(skb)
374 * so that this procedure is noop.
375 */
376
377 if (skb->pkt_type == PACKET_LOOPBACK)
378 goto out;
379
380 if (!net_eq(dev_net(dev), sock_net(sk)))
381 goto out;
382
383 skb = skb_share_check(skb, GFP_ATOMIC);
384 if (skb == NULL)
385 goto oom;
386
387 /* drop any routing info */
388 skb_dst_drop(skb);
389
390 /* drop conntrack reference */
391 nf_reset(skb);
392
393 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
394
395 skb_push(skb, skb->data - skb_mac_header(skb));
396
397 /*
398 * The SOCK_PACKET socket receives _all_ frames.
399 */
400
401 spkt->spkt_family = dev->type;
402 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
403 spkt->spkt_protocol = skb->protocol;
404
405 /*
406 * Charge the memory to the socket. This is done specifically
407 * to prevent sockets using all the memory up.
408 */
409
410 if (sock_queue_rcv_skb(sk, skb) == 0)
411 return 0;
412
413out:
414 kfree_skb(skb);
415oom:
416 return 0;
417}
418
419
420/*
421 * Output a raw packet to a device layer. This bypasses all the other
422 * protocol layers and you must therefore supply it with a complete frame
423 */
424
425static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
426 struct msghdr *msg, size_t len)
427{
428 struct sock *sk = sock->sk;
429 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
430 struct sk_buff *skb = NULL;
431 struct net_device *dev;
432 __be16 proto = 0;
433 int err;
434
435 /*
436 * Get and verify the address.
437 */
438
439 if (saddr) {
440 if (msg->msg_namelen < sizeof(struct sockaddr))
441 return -EINVAL;
442 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
443 proto = saddr->spkt_protocol;
444 } else
445 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
446
447 /*
448 * Find the device first to size check it
449 */
450
451 saddr->spkt_device[13] = 0;
452retry:
453 rcu_read_lock();
454 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
455 err = -ENODEV;
456 if (dev == NULL)
457 goto out_unlock;
458
459 err = -ENETDOWN;
460 if (!(dev->flags & IFF_UP))
461 goto out_unlock;
462
463 /*
464 * You may not queue a frame bigger than the mtu. This is the lowest level
465 * raw protocol and you must do your own fragmentation at this level.
466 */
467
468 err = -EMSGSIZE;
469 if (len > dev->mtu + dev->hard_header_len)
470 goto out_unlock;
471
472 if (!skb) {
473 size_t reserved = LL_RESERVED_SPACE(dev);
474 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
475
476 rcu_read_unlock();
477 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
478 if (skb == NULL)
479 return -ENOBUFS;
480 /* FIXME: Save some space for broken drivers that write a hard
481 * header at transmission time by themselves. PPP is the notable
482 * one here. This should really be fixed at the driver level.
483 */
484 skb_reserve(skb, reserved);
485 skb_reset_network_header(skb);
486
487 /* Try to align data part correctly */
488 if (hhlen) {
489 skb->data -= hhlen;
490 skb->tail -= hhlen;
491 if (len < hhlen)
492 skb_reset_network_header(skb);
493 }
494 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
495 if (err)
496 goto out_free;
497 goto retry;
498 }
499
500
501 skb->protocol = proto;
502 skb->dev = dev;
503 skb->priority = sk->sk_priority;
504 skb->mark = sk->sk_mark;
505 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
506 if (err < 0)
507 goto out_unlock;
508
509 dev_queue_xmit(skb);
510 rcu_read_unlock();
511 return len;
512
513out_unlock:
514 rcu_read_unlock();
515out_free:
516 kfree_skb(skb);
517 return err;
518}
519
520static inline unsigned int run_filter(const struct sk_buff *skb,
521 const struct sock *sk,
522 unsigned int res)
523{
524 struct sk_filter *filter;
525
526 rcu_read_lock_bh();
527 filter = rcu_dereference_bh(sk->sk_filter);
528 if (filter != NULL)
529 res = sk_run_filter(skb, filter->insns);
530 rcu_read_unlock_bh();
531
532 return res;
533}
534
535/*
536 * This function makes lazy skb cloning in hope that most of packets
537 * are discarded by BPF.
538 *
539 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
540 * and skb->cb are mangled. It works because (and until) packets
541 * falling here are owned by current CPU. Output packets are cloned
542 * by dev_queue_xmit_nit(), input packets are processed by net_bh
543 * sequencially, so that if we return skb to original state on exit,
544 * we will not harm anyone.
545 */
546
547static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
548 struct packet_type *pt, struct net_device *orig_dev)
549{
550 struct sock *sk;
551 struct sockaddr_ll *sll;
552 struct packet_sock *po;
553 u8 *skb_head = skb->data;
554 int skb_len = skb->len;
555 unsigned int snaplen, res;
556
557 if (skb->pkt_type == PACKET_LOOPBACK)
558 goto drop;
559
560 sk = pt->af_packet_priv;
561 po = pkt_sk(sk);
562
563 if (!net_eq(dev_net(dev), sock_net(sk)))
564 goto drop;
565
566 skb->dev = dev;
567
568 if (dev->header_ops) {
569 /* The device has an explicit notion of ll header,
570 * exported to higher levels.
571 *
572 * Otherwise, the device hides details of its frame
573 * structure, so that corresponding packet head is
574 * never delivered to user.
575 */
576 if (sk->sk_type != SOCK_DGRAM)
577 skb_push(skb, skb->data - skb_mac_header(skb));
578 else if (skb->pkt_type == PACKET_OUTGOING) {
579 /* Special case: outgoing packets have ll header at head */
580 skb_pull(skb, skb_network_offset(skb));
581 }
582 }
583
584 snaplen = skb->len;
585
586 res = run_filter(skb, sk, snaplen);
587 if (!res)
588 goto drop_n_restore;
589 if (snaplen > res)
590 snaplen = res;
591
592 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
593 (unsigned)sk->sk_rcvbuf)
594 goto drop_n_acct;
595
596 if (skb_shared(skb)) {
597 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
598 if (nskb == NULL)
599 goto drop_n_acct;
600
601 if (skb_head != skb->data) {
602 skb->data = skb_head;
603 skb->len = skb_len;
604 }
605 kfree_skb(skb);
606 skb = nskb;
607 }
608
609 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
610 sizeof(skb->cb));
611
612 sll = &PACKET_SKB_CB(skb)->sa.ll;
613 sll->sll_family = AF_PACKET;
614 sll->sll_hatype = dev->type;
615 sll->sll_protocol = skb->protocol;
616 sll->sll_pkttype = skb->pkt_type;
617 if (unlikely(po->origdev))
618 sll->sll_ifindex = orig_dev->ifindex;
619 else
620 sll->sll_ifindex = dev->ifindex;
621
622 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
623
624 PACKET_SKB_CB(skb)->origlen = skb->len;
625
626 if (pskb_trim(skb, snaplen))
627 goto drop_n_acct;
628
629 skb_set_owner_r(skb, sk);
630 skb->dev = NULL;
631 skb_dst_drop(skb);
632
633 /* drop conntrack reference */
634 nf_reset(skb);
635
636 spin_lock(&sk->sk_receive_queue.lock);
637 po->stats.tp_packets++;
638 skb->dropcount = atomic_read(&sk->sk_drops);
639 __skb_queue_tail(&sk->sk_receive_queue, skb);
640 spin_unlock(&sk->sk_receive_queue.lock);
641 sk->sk_data_ready(sk, skb->len);
642 return 0;
643
644drop_n_acct:
645 po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
646
647drop_n_restore:
648 if (skb_head != skb->data && skb_shared(skb)) {
649 skb->data = skb_head;
650 skb->len = skb_len;
651 }
652drop:
653 consume_skb(skb);
654 return 0;
655}
656
657static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
658 struct packet_type *pt, struct net_device *orig_dev)
659{
660 struct sock *sk;
661 struct packet_sock *po;
662 struct sockaddr_ll *sll;
663 union {
664 struct tpacket_hdr *h1;
665 struct tpacket2_hdr *h2;
666 void *raw;
667 } h;
668 u8 *skb_head = skb->data;
669 int skb_len = skb->len;
670 unsigned int snaplen, res;
671 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
672 unsigned short macoff, netoff, hdrlen;
673 struct sk_buff *copy_skb = NULL;
674 struct timeval tv;
675 struct timespec ts;
676 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
677
678 if (skb->pkt_type == PACKET_LOOPBACK)
679 goto drop;
680
681 sk = pt->af_packet_priv;
682 po = pkt_sk(sk);
683
684 if (!net_eq(dev_net(dev), sock_net(sk)))
685 goto drop;
686
687 if (dev->header_ops) {
688 if (sk->sk_type != SOCK_DGRAM)
689 skb_push(skb, skb->data - skb_mac_header(skb));
690 else if (skb->pkt_type == PACKET_OUTGOING) {
691 /* Special case: outgoing packets have ll header at head */
692 skb_pull(skb, skb_network_offset(skb));
693 }
694 }
695
696 if (skb->ip_summed == CHECKSUM_PARTIAL)
697 status |= TP_STATUS_CSUMNOTREADY;
698
699 snaplen = skb->len;
700
701 res = run_filter(skb, sk, snaplen);
702 if (!res)
703 goto drop_n_restore;
704 if (snaplen > res)
705 snaplen = res;
706
707 if (sk->sk_type == SOCK_DGRAM) {
708 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
709 po->tp_reserve;
710 } else {
711 unsigned maclen = skb_network_offset(skb);
712 netoff = TPACKET_ALIGN(po->tp_hdrlen +
713 (maclen < 16 ? 16 : maclen)) +
714 po->tp_reserve;
715 macoff = netoff - maclen;
716 }
717
718 if (macoff + snaplen > po->rx_ring.frame_size) {
719 if (po->copy_thresh &&
720 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
721 (unsigned)sk->sk_rcvbuf) {
722 if (skb_shared(skb)) {
723 copy_skb = skb_clone(skb, GFP_ATOMIC);
724 } else {
725 copy_skb = skb_get(skb);
726 skb_head = skb->data;
727 }
728 if (copy_skb)
729 skb_set_owner_r(copy_skb, sk);
730 }
731 snaplen = po->rx_ring.frame_size - macoff;
732 if ((int)snaplen < 0)
733 snaplen = 0;
734 }
735
736 spin_lock(&sk->sk_receive_queue.lock);
737 h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
738 if (!h.raw)
739 goto ring_is_full;
740 packet_increment_head(&po->rx_ring);
741 po->stats.tp_packets++;
742 if (copy_skb) {
743 status |= TP_STATUS_COPY;
744 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
745 }
746 if (!po->stats.tp_drops)
747 status &= ~TP_STATUS_LOSING;
748 spin_unlock(&sk->sk_receive_queue.lock);
749
750 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
751
752 switch (po->tp_version) {
753 case TPACKET_V1:
754 h.h1->tp_len = skb->len;
755 h.h1->tp_snaplen = snaplen;
756 h.h1->tp_mac = macoff;
757 h.h1->tp_net = netoff;
758 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
759 && shhwtstamps->syststamp.tv64)
760 tv = ktime_to_timeval(shhwtstamps->syststamp);
761 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
762 && shhwtstamps->hwtstamp.tv64)
763 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
764 else if (skb->tstamp.tv64)
765 tv = ktime_to_timeval(skb->tstamp);
766 else
767 do_gettimeofday(&tv);
768 h.h1->tp_sec = tv.tv_sec;
769 h.h1->tp_usec = tv.tv_usec;
770 hdrlen = sizeof(*h.h1);
771 break;
772 case TPACKET_V2:
773 h.h2->tp_len = skb->len;
774 h.h2->tp_snaplen = snaplen;
775 h.h2->tp_mac = macoff;
776 h.h2->tp_net = netoff;
777 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
778 && shhwtstamps->syststamp.tv64)
779 ts = ktime_to_timespec(shhwtstamps->syststamp);
780 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
781 && shhwtstamps->hwtstamp.tv64)
782 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
783 else if (skb->tstamp.tv64)
784 ts = ktime_to_timespec(skb->tstamp);
785 else
786 getnstimeofday(&ts);
787 h.h2->tp_sec = ts.tv_sec;
788 h.h2->tp_nsec = ts.tv_nsec;
789 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
790 hdrlen = sizeof(*h.h2);
791 break;
792 default:
793 BUG();
794 }
795
796 sll = h.raw + TPACKET_ALIGN(hdrlen);
797 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
798 sll->sll_family = AF_PACKET;
799 sll->sll_hatype = dev->type;
800 sll->sll_protocol = skb->protocol;
801 sll->sll_pkttype = skb->pkt_type;
802 if (unlikely(po->origdev))
803 sll->sll_ifindex = orig_dev->ifindex;
804 else
805 sll->sll_ifindex = dev->ifindex;
806
807 __packet_set_status(po, h.raw, status);
808 smp_mb();
809 {
810 u8 *start, *end;
811
812 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw + macoff + snaplen);
813 for (start = h.raw; start < end; start += PAGE_SIZE)
814 flush_dcache_page(pgv_to_page(start));
815 }
816
817 sk->sk_data_ready(sk, 0);
818
819drop_n_restore:
820 if (skb_head != skb->data && skb_shared(skb)) {
821 skb->data = skb_head;
822 skb->len = skb_len;
823 }
824drop:
825 kfree_skb(skb);
826 return 0;
827
828ring_is_full:
829 po->stats.tp_drops++;
830 spin_unlock(&sk->sk_receive_queue.lock);
831
832 sk->sk_data_ready(sk, 0);
833 kfree_skb(copy_skb);
834 goto drop_n_restore;
835}
836
837static void tpacket_destruct_skb(struct sk_buff *skb)
838{
839 struct packet_sock *po = pkt_sk(skb->sk);
840 void *ph;
841
842 BUG_ON(skb == NULL);
843
844 if (likely(po->tx_ring.pg_vec)) {
845 ph = skb_shinfo(skb)->destructor_arg;
846 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
847 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
848 atomic_dec(&po->tx_ring.pending);
849 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
850 }
851
852 sock_wfree(skb);
853}
854
855static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
856 void *frame, struct net_device *dev, int size_max,
857 __be16 proto, unsigned char *addr)
858{
859 union {
860 struct tpacket_hdr *h1;
861 struct tpacket2_hdr *h2;
862 void *raw;
863 } ph;
864 int to_write, offset, len, tp_len, nr_frags, len_max;
865 struct socket *sock = po->sk.sk_socket;
866 struct page *page;
867 void *data;
868 int err;
869
870 ph.raw = frame;
871
872 skb->protocol = proto;
873 skb->dev = dev;
874 skb->priority = po->sk.sk_priority;
875 skb->mark = po->sk.sk_mark;
876 skb_shinfo(skb)->destructor_arg = ph.raw;
877
878 switch (po->tp_version) {
879 case TPACKET_V2:
880 tp_len = ph.h2->tp_len;
881 break;
882 default:
883 tp_len = ph.h1->tp_len;
884 break;
885 }
886 if (unlikely(tp_len > size_max)) {
887 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
888 return -EMSGSIZE;
889 }
890
891 skb_reserve(skb, LL_RESERVED_SPACE(dev));
892 skb_reset_network_header(skb);
893
894 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
895 to_write = tp_len;
896
897 if (sock->type == SOCK_DGRAM) {
898 err = dev_hard_header(skb, dev, ntohs(proto), addr,
899 NULL, tp_len);
900 if (unlikely(err < 0))
901 return -EINVAL;
902 } else if (dev->hard_header_len) {
903 /* net device doesn't like empty head */
904 if (unlikely(tp_len <= dev->hard_header_len)) {
905 pr_err("packet size is too short (%d < %d)\n",
906 tp_len, dev->hard_header_len);
907 return -EINVAL;
908 }
909
910 skb_push(skb, dev->hard_header_len);
911 err = skb_store_bits(skb, 0, data,
912 dev->hard_header_len);
913 if (unlikely(err))
914 return err;
915
916 data += dev->hard_header_len;
917 to_write -= dev->hard_header_len;
918 }
919
920 err = -EFAULT;
921 offset = offset_in_page(data);
922 len_max = PAGE_SIZE - offset;
923 len = ((to_write > len_max) ? len_max : to_write);
924
925 skb->data_len = to_write;
926 skb->len += to_write;
927 skb->truesize += to_write;
928 atomic_add(to_write, &po->sk.sk_wmem_alloc);
929
930 while (likely(to_write)) {
931 nr_frags = skb_shinfo(skb)->nr_frags;
932
933 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
934 pr_err("Packet exceed the number of skb frags(%lu)\n",
935 MAX_SKB_FRAGS);
936 return -EFAULT;
937 }
938
939 page = pgv_to_page(data);
940 data += len;
941 flush_dcache_page(page);
942 get_page(page);
943 skb_fill_page_desc(skb, nr_frags, page, offset, len);
944 to_write -= len;
945 offset = 0;
946 len_max = PAGE_SIZE;
947 len = ((to_write > len_max) ? len_max : to_write);
948 }
949
950 return tp_len;
951}
952
953static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
954{
955 struct socket *sock;
956 struct sk_buff *skb;
957 struct net_device *dev;
958 __be16 proto;
959 int ifindex, err, reserve = 0;
960 void *ph;
961 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
962 int tp_len, size_max;
963 unsigned char *addr;
964 int len_sum = 0;
965 int status = 0;
966
967 sock = po->sk.sk_socket;
968
969 mutex_lock(&po->pg_vec_lock);
970
971 err = -EBUSY;
972 if (saddr == NULL) {
973 ifindex = po->ifindex;
974 proto = po->num;
975 addr = NULL;
976 } else {
977 err = -EINVAL;
978 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
979 goto out;
980 if (msg->msg_namelen < (saddr->sll_halen
981 + offsetof(struct sockaddr_ll,
982 sll_addr)))
983 goto out;
984 ifindex = saddr->sll_ifindex;
985 proto = saddr->sll_protocol;
986 addr = saddr->sll_addr;
987 }
988
989 dev = dev_get_by_index(sock_net(&po->sk), ifindex);
990 err = -ENXIO;
991 if (unlikely(dev == NULL))
992 goto out;
993
994 reserve = dev->hard_header_len;
995
996 err = -ENETDOWN;
997 if (unlikely(!(dev->flags & IFF_UP)))
998 goto out_put;
999
1000 size_max = po->tx_ring.frame_size
1001 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1002
1003 if (size_max > dev->mtu + reserve)
1004 size_max = dev->mtu + reserve;
1005
1006 do {
1007 ph = packet_current_frame(po, &po->tx_ring,
1008 TP_STATUS_SEND_REQUEST);
1009
1010 if (unlikely(ph == NULL)) {
1011 schedule();
1012 continue;
1013 }
1014
1015 status = TP_STATUS_SEND_REQUEST;
1016 skb = sock_alloc_send_skb(&po->sk,
1017 LL_ALLOCATED_SPACE(dev)
1018 + sizeof(struct sockaddr_ll),
1019 0, &err);
1020
1021 if (unlikely(skb == NULL))
1022 goto out_status;
1023
1024 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1025 addr);
1026
1027 if (unlikely(tp_len < 0)) {
1028 if (po->tp_loss) {
1029 __packet_set_status(po, ph,
1030 TP_STATUS_AVAILABLE);
1031 packet_increment_head(&po->tx_ring);
1032 kfree_skb(skb);
1033 continue;
1034 } else {
1035 status = TP_STATUS_WRONG_FORMAT;
1036 err = tp_len;
1037 goto out_status;
1038 }
1039 }
1040
1041 skb->destructor = tpacket_destruct_skb;
1042 __packet_set_status(po, ph, TP_STATUS_SENDING);
1043 atomic_inc(&po->tx_ring.pending);
1044
1045 status = TP_STATUS_SEND_REQUEST;
1046 err = dev_queue_xmit(skb);
1047 if (unlikely(err > 0)) {
1048 err = net_xmit_errno(err);
1049 if (err && __packet_get_status(po, ph) ==
1050 TP_STATUS_AVAILABLE) {
1051 /* skb was destructed already */
1052 skb = NULL;
1053 goto out_status;
1054 }
1055 /*
1056 * skb was dropped but not destructed yet;
1057 * let's treat it like congestion or err < 0
1058 */
1059 err = 0;
1060 }
1061 packet_increment_head(&po->tx_ring);
1062 len_sum += tp_len;
1063 } while (likely((ph != NULL) ||
1064 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1065 (atomic_read(&po->tx_ring.pending))))
1066 );
1067
1068 err = len_sum;
1069 goto out_put;
1070
1071out_status:
1072 __packet_set_status(po, ph, status);
1073 kfree_skb(skb);
1074out_put:
1075 dev_put(dev);
1076out:
1077 mutex_unlock(&po->pg_vec_lock);
1078 return err;
1079}
1080
1081static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1082 size_t reserve, size_t len,
1083 size_t linear, int noblock,
1084 int *err)
1085{
1086 struct sk_buff *skb;
1087
1088 /* Under a page? Don't bother with paged skb. */
1089 if (prepad + len < PAGE_SIZE || !linear)
1090 linear = len;
1091
1092 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1093 err);
1094 if (!skb)
1095 return NULL;
1096
1097 skb_reserve(skb, reserve);
1098 skb_put(skb, linear);
1099 skb->data_len = len - linear;
1100 skb->len += len - linear;
1101
1102 return skb;
1103}
1104
1105static int packet_snd(struct socket *sock,
1106 struct msghdr *msg, size_t len)
1107{
1108 struct sock *sk = sock->sk;
1109 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1110 struct sk_buff *skb;
1111 struct net_device *dev;
1112 __be16 proto;
1113 unsigned char *addr;
1114 int ifindex, err, reserve = 0;
1115 struct virtio_net_hdr vnet_hdr = { 0 };
1116 int offset = 0;
1117 int vnet_hdr_len;
1118 struct packet_sock *po = pkt_sk(sk);
1119 unsigned short gso_type = 0;
1120
1121 /*
1122 * Get and verify the address.
1123 */
1124
1125 if (saddr == NULL) {
1126 ifindex = po->ifindex;
1127 proto = po->num;
1128 addr = NULL;
1129 } else {
1130 err = -EINVAL;
1131 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1132 goto out;
1133 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1134 goto out;
1135 ifindex = saddr->sll_ifindex;
1136 proto = saddr->sll_protocol;
1137 addr = saddr->sll_addr;
1138 }
1139
1140
1141 dev = dev_get_by_index(sock_net(sk), ifindex);
1142 err = -ENXIO;
1143 if (dev == NULL)
1144 goto out_unlock;
1145 if (sock->type == SOCK_RAW)
1146 reserve = dev->hard_header_len;
1147
1148 err = -ENETDOWN;
1149 if (!(dev->flags & IFF_UP))
1150 goto out_unlock;
1151
1152 if (po->has_vnet_hdr) {
1153 vnet_hdr_len = sizeof(vnet_hdr);
1154
1155 err = -EINVAL;
1156 if (len < vnet_hdr_len)
1157 goto out_unlock;
1158
1159 len -= vnet_hdr_len;
1160
1161 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1162 vnet_hdr_len);
1163 if (err < 0)
1164 goto out_unlock;
1165
1166 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1167 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1168 vnet_hdr.hdr_len))
1169 vnet_hdr.hdr_len = vnet_hdr.csum_start +
1170 vnet_hdr.csum_offset + 2;
1171
1172 err = -EINVAL;
1173 if (vnet_hdr.hdr_len > len)
1174 goto out_unlock;
1175
1176 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1177 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1178 case VIRTIO_NET_HDR_GSO_TCPV4:
1179 gso_type = SKB_GSO_TCPV4;
1180 break;
1181 case VIRTIO_NET_HDR_GSO_TCPV6:
1182 gso_type = SKB_GSO_TCPV6;
1183 break;
1184 case VIRTIO_NET_HDR_GSO_UDP:
1185 gso_type = SKB_GSO_UDP;
1186 break;
1187 default:
1188 goto out_unlock;
1189 }
1190
1191 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1192 gso_type |= SKB_GSO_TCP_ECN;
1193
1194 if (vnet_hdr.gso_size == 0)
1195 goto out_unlock;
1196
1197 }
1198 }
1199
1200 err = -EMSGSIZE;
1201 if (!gso_type && (len > dev->mtu+reserve))
1202 goto out_unlock;
1203
1204 err = -ENOBUFS;
1205 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1206 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1207 msg->msg_flags & MSG_DONTWAIT, &err);
1208 if (skb == NULL)
1209 goto out_unlock;
1210
1211 skb_set_network_header(skb, reserve);
1212
1213 err = -EINVAL;
1214 if (sock->type == SOCK_DGRAM &&
1215 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1216 goto out_free;
1217
1218 /* Returns -EFAULT on error */
1219 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1220 if (err)
1221 goto out_free;
1222 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1223 if (err < 0)
1224 goto out_free;
1225
1226 skb->protocol = proto;
1227 skb->dev = dev;
1228 skb->priority = sk->sk_priority;
1229 skb->mark = sk->sk_mark;
1230
1231 if (po->has_vnet_hdr) {
1232 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1233 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1234 vnet_hdr.csum_offset)) {
1235 err = -EINVAL;
1236 goto out_free;
1237 }
1238 }
1239
1240 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1241 skb_shinfo(skb)->gso_type = gso_type;
1242
1243 /* Header must be checked, and gso_segs computed. */
1244 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1245 skb_shinfo(skb)->gso_segs = 0;
1246
1247 len += vnet_hdr_len;
1248 }
1249
1250 /*
1251 * Now send it
1252 */
1253
1254 err = dev_queue_xmit(skb);
1255 if (err > 0 && (err = net_xmit_errno(err)) != 0)
1256 goto out_unlock;
1257
1258 dev_put(dev);
1259
1260 return len;
1261
1262out_free:
1263 kfree_skb(skb);
1264out_unlock:
1265 if (dev)
1266 dev_put(dev);
1267out:
1268 return err;
1269}
1270
1271static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1272 struct msghdr *msg, size_t len)
1273{
1274 struct sock *sk = sock->sk;
1275 struct packet_sock *po = pkt_sk(sk);
1276 if (po->tx_ring.pg_vec)
1277 return tpacket_snd(po, msg);
1278 else
1279 return packet_snd(sock, msg, len);
1280}
1281
1282/*
1283 * Close a PACKET socket. This is fairly simple. We immediately go
1284 * to 'closed' state and remove our protocol entry in the device list.
1285 */
1286
1287static int packet_release(struct socket *sock)
1288{
1289 struct sock *sk = sock->sk;
1290 struct packet_sock *po;
1291 struct net *net;
1292 struct tpacket_req req;
1293
1294 if (!sk)
1295 return 0;
1296
1297 net = sock_net(sk);
1298 po = pkt_sk(sk);
1299
1300 spin_lock_bh(&net->packet.sklist_lock);
1301 sk_del_node_init_rcu(sk);
1302 sock_prot_inuse_add(net, sk->sk_prot, -1);
1303 spin_unlock_bh(&net->packet.sklist_lock);
1304
1305 spin_lock(&po->bind_lock);
1306 if (po->running) {
1307 /*
1308 * Remove from protocol table
1309 */
1310 po->running = 0;
1311 po->num = 0;
1312 __dev_remove_pack(&po->prot_hook);
1313 __sock_put(sk);
1314 }
1315 spin_unlock(&po->bind_lock);
1316
1317 packet_flush_mclist(sk);
1318
1319 memset(&req, 0, sizeof(req));
1320
1321 if (po->rx_ring.pg_vec)
1322 packet_set_ring(sk, &req, 1, 0);
1323
1324 if (po->tx_ring.pg_vec)
1325 packet_set_ring(sk, &req, 1, 1);
1326
1327 synchronize_net();
1328 /*
1329 * Now the socket is dead. No more input will appear.
1330 */
1331 sock_orphan(sk);
1332 sock->sk = NULL;
1333
1334 /* Purge queues */
1335
1336 skb_queue_purge(&sk->sk_receive_queue);
1337 sk_refcnt_debug_release(sk);
1338
1339 sock_put(sk);
1340 return 0;
1341}
1342
1343/*
1344 * Attach a packet hook.
1345 */
1346
1347static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1348{
1349 struct packet_sock *po = pkt_sk(sk);
1350 /*
1351 * Detach an existing hook if present.
1352 */
1353
1354 lock_sock(sk);
1355
1356 spin_lock(&po->bind_lock);
1357 if (po->running) {
1358 __sock_put(sk);
1359 po->running = 0;
1360 po->num = 0;
1361 spin_unlock(&po->bind_lock);
1362 dev_remove_pack(&po->prot_hook);
1363 spin_lock(&po->bind_lock);
1364 }
1365
1366 po->num = protocol;
1367 po->prot_hook.type = protocol;
1368 po->prot_hook.dev = dev;
1369
1370 po->ifindex = dev ? dev->ifindex : 0;
1371
1372 if (protocol == 0)
1373 goto out_unlock;
1374
1375 if (!dev || (dev->flags & IFF_UP)) {
1376 dev_add_pack(&po->prot_hook);
1377 sock_hold(sk);
1378 po->running = 1;
1379 } else {
1380 sk->sk_err = ENETDOWN;
1381 if (!sock_flag(sk, SOCK_DEAD))
1382 sk->sk_error_report(sk);
1383 }
1384
1385out_unlock:
1386 spin_unlock(&po->bind_lock);
1387 release_sock(sk);
1388 return 0;
1389}
1390
1391/*
1392 * Bind a packet socket to a device
1393 */
1394
1395static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1396 int addr_len)
1397{
1398 struct sock *sk = sock->sk;
1399 char name[15];
1400 struct net_device *dev;
1401 int err = -ENODEV;
1402
1403 /*
1404 * Check legality
1405 */
1406
1407 if (addr_len != sizeof(struct sockaddr))
1408 return -EINVAL;
1409 strlcpy(name, uaddr->sa_data, sizeof(name));
1410
1411 dev = dev_get_by_name(sock_net(sk), name);
1412 if (dev) {
1413 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1414 dev_put(dev);
1415 }
1416 return err;
1417}
1418
1419static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1420{
1421 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1422 struct sock *sk = sock->sk;
1423 struct net_device *dev = NULL;
1424 int err;
1425
1426
1427 /*
1428 * Check legality
1429 */
1430
1431 if (addr_len < sizeof(struct sockaddr_ll))
1432 return -EINVAL;
1433 if (sll->sll_family != AF_PACKET)
1434 return -EINVAL;
1435
1436 if (sll->sll_ifindex) {
1437 err = -ENODEV;
1438 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1439 if (dev == NULL)
1440 goto out;
1441 }
1442 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1443 if (dev)
1444 dev_put(dev);
1445
1446out:
1447 return err;
1448}
1449
1450static struct proto packet_proto = {
1451 .name = "PACKET",
1452 .owner = THIS_MODULE,
1453 .obj_size = sizeof(struct packet_sock),
1454};
1455
1456/*
1457 * Create a packet of type SOCK_PACKET.
1458 */
1459
1460static int packet_create(struct net *net, struct socket *sock, int protocol,
1461 int kern)
1462{
1463 struct sock *sk;
1464 struct packet_sock *po;
1465 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1466 int err;
1467
1468 if (!capable(CAP_NET_RAW))
1469 return -EPERM;
1470 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1471 sock->type != SOCK_PACKET)
1472 return -ESOCKTNOSUPPORT;
1473
1474 sock->state = SS_UNCONNECTED;
1475
1476 err = -ENOBUFS;
1477 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1478 if (sk == NULL)
1479 goto out;
1480
1481 sock->ops = &packet_ops;
1482 if (sock->type == SOCK_PACKET)
1483 sock->ops = &packet_ops_spkt;
1484
1485 sock_init_data(sock, sk);
1486
1487 po = pkt_sk(sk);
1488 sk->sk_family = PF_PACKET;
1489 po->num = proto;
1490
1491 sk->sk_destruct = packet_sock_destruct;
1492 sk_refcnt_debug_inc(sk);
1493
1494 /*
1495 * Attach a protocol block
1496 */
1497
1498 spin_lock_init(&po->bind_lock);
1499 mutex_init(&po->pg_vec_lock);
1500 po->prot_hook.func = packet_rcv;
1501
1502 if (sock->type == SOCK_PACKET)
1503 po->prot_hook.func = packet_rcv_spkt;
1504
1505 po->prot_hook.af_packet_priv = sk;
1506
1507 if (proto) {
1508 po->prot_hook.type = proto;
1509 dev_add_pack(&po->prot_hook);
1510 sock_hold(sk);
1511 po->running = 1;
1512 }
1513
1514 spin_lock_bh(&net->packet.sklist_lock);
1515 sk_add_node_rcu(sk, &net->packet.sklist);
1516 sock_prot_inuse_add(net, &packet_proto, 1);
1517 spin_unlock_bh(&net->packet.sklist_lock);
1518
1519 return 0;
1520out:
1521 return err;
1522}
1523
1524static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1525{
1526 struct sock_exterr_skb *serr;
1527 struct sk_buff *skb, *skb2;
1528 int copied, err;
1529
1530 err = -EAGAIN;
1531 skb = skb_dequeue(&sk->sk_error_queue);
1532 if (skb == NULL)
1533 goto out;
1534
1535 copied = skb->len;
1536 if (copied > len) {
1537 msg->msg_flags |= MSG_TRUNC;
1538 copied = len;
1539 }
1540 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1541 if (err)
1542 goto out_free_skb;
1543
1544 sock_recv_timestamp(msg, sk, skb);
1545
1546 serr = SKB_EXT_ERR(skb);
1547 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
1548 sizeof(serr->ee), &serr->ee);
1549
1550 msg->msg_flags |= MSG_ERRQUEUE;
1551 err = copied;
1552
1553 /* Reset and regenerate socket error */
1554 spin_lock_bh(&sk->sk_error_queue.lock);
1555 sk->sk_err = 0;
1556 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
1557 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
1558 spin_unlock_bh(&sk->sk_error_queue.lock);
1559 sk->sk_error_report(sk);
1560 } else
1561 spin_unlock_bh(&sk->sk_error_queue.lock);
1562
1563out_free_skb:
1564 kfree_skb(skb);
1565out:
1566 return err;
1567}
1568
1569/*
1570 * Pull a packet from our receive queue and hand it to the user.
1571 * If necessary we block.
1572 */
1573
1574static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1575 struct msghdr *msg, size_t len, int flags)
1576{
1577 struct sock *sk = sock->sk;
1578 struct sk_buff *skb;
1579 int copied, err;
1580 struct sockaddr_ll *sll;
1581 int vnet_hdr_len = 0;
1582
1583 err = -EINVAL;
1584 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1585 goto out;
1586
1587#if 0
1588 /* What error should we return now? EUNATTACH? */
1589 if (pkt_sk(sk)->ifindex < 0)
1590 return -ENODEV;
1591#endif
1592
1593 if (flags & MSG_ERRQUEUE) {
1594 err = packet_recv_error(sk, msg, len);
1595 goto out;
1596 }
1597
1598 /*
1599 * Call the generic datagram receiver. This handles all sorts
1600 * of horrible races and re-entrancy so we can forget about it
1601 * in the protocol layers.
1602 *
1603 * Now it will return ENETDOWN, if device have just gone down,
1604 * but then it will block.
1605 */
1606
1607 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1608
1609 /*
1610 * An error occurred so return it. Because skb_recv_datagram()
1611 * handles the blocking we don't see and worry about blocking
1612 * retries.
1613 */
1614
1615 if (skb == NULL)
1616 goto out;
1617
1618 if (pkt_sk(sk)->has_vnet_hdr) {
1619 struct virtio_net_hdr vnet_hdr = { 0 };
1620
1621 err = -EINVAL;
1622 vnet_hdr_len = sizeof(vnet_hdr);
1623 if (len < vnet_hdr_len)
1624 goto out_free;
1625
1626 len -= vnet_hdr_len;
1627
1628 if (skb_is_gso(skb)) {
1629 struct skb_shared_info *sinfo = skb_shinfo(skb);
1630
1631 /* This is a hint as to how much should be linear. */
1632 vnet_hdr.hdr_len = skb_headlen(skb);
1633 vnet_hdr.gso_size = sinfo->gso_size;
1634 if (sinfo->gso_type & SKB_GSO_TCPV4)
1635 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1636 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1637 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1638 else if (sinfo->gso_type & SKB_GSO_UDP)
1639 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1640 else if (sinfo->gso_type & SKB_GSO_FCOE)
1641 goto out_free;
1642 else
1643 BUG();
1644 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1645 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1646 } else
1647 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1648
1649 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1650 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1651 vnet_hdr.csum_start = skb->csum_start -
1652 skb_headroom(skb);
1653 vnet_hdr.csum_offset = skb->csum_offset;
1654 } /* else everything is zero */
1655
1656 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1657 vnet_hdr_len);
1658 if (err < 0)
1659 goto out_free;
1660 }
1661
1662 /*
1663 * If the address length field is there to be filled in, we fill
1664 * it in now.
1665 */
1666
1667 sll = &PACKET_SKB_CB(skb)->sa.ll;
1668 if (sock->type == SOCK_PACKET)
1669 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1670 else
1671 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1672
1673 /*
1674 * You lose any data beyond the buffer you gave. If it worries a
1675 * user program they can ask the device for its MTU anyway.
1676 */
1677
1678 copied = skb->len;
1679 if (copied > len) {
1680 copied = len;
1681 msg->msg_flags |= MSG_TRUNC;
1682 }
1683
1684 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1685 if (err)
1686 goto out_free;
1687
1688 sock_recv_ts_and_drops(msg, sk, skb);
1689
1690 if (msg->msg_name)
1691 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1692 msg->msg_namelen);
1693
1694 if (pkt_sk(sk)->auxdata) {
1695 struct tpacket_auxdata aux;
1696
1697 aux.tp_status = TP_STATUS_USER;
1698 if (skb->ip_summed == CHECKSUM_PARTIAL)
1699 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1700 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1701 aux.tp_snaplen = skb->len;
1702 aux.tp_mac = 0;
1703 aux.tp_net = skb_network_offset(skb);
1704 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
1705
1706 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1707 }
1708
1709 /*
1710 * Free or return the buffer as appropriate. Again this
1711 * hides all the races and re-entrancy issues from us.
1712 */
1713 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1714
1715out_free:
1716 skb_free_datagram(sk, skb);
1717out:
1718 return err;
1719}
1720
1721static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1722 int *uaddr_len, int peer)
1723{
1724 struct net_device *dev;
1725 struct sock *sk = sock->sk;
1726
1727 if (peer)
1728 return -EOPNOTSUPP;
1729
1730 uaddr->sa_family = AF_PACKET;
1731 rcu_read_lock();
1732 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
1733 if (dev)
1734 strncpy(uaddr->sa_data, dev->name, 14);
1735 else
1736 memset(uaddr->sa_data, 0, 14);
1737 rcu_read_unlock();
1738 *uaddr_len = sizeof(*uaddr);
1739
1740 return 0;
1741}
1742
1743static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1744 int *uaddr_len, int peer)
1745{
1746 struct net_device *dev;
1747 struct sock *sk = sock->sk;
1748 struct packet_sock *po = pkt_sk(sk);
1749 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1750
1751 if (peer)
1752 return -EOPNOTSUPP;
1753
1754 sll->sll_family = AF_PACKET;
1755 sll->sll_ifindex = po->ifindex;
1756 sll->sll_protocol = po->num;
1757 sll->sll_pkttype = 0;
1758 rcu_read_lock();
1759 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1760 if (dev) {
1761 sll->sll_hatype = dev->type;
1762 sll->sll_halen = dev->addr_len;
1763 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1764 } else {
1765 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1766 sll->sll_halen = 0;
1767 }
1768 rcu_read_unlock();
1769 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1770
1771 return 0;
1772}
1773
1774static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1775 int what)
1776{
1777 switch (i->type) {
1778 case PACKET_MR_MULTICAST:
1779 if (i->alen != dev->addr_len)
1780 return -EINVAL;
1781 if (what > 0)
1782 return dev_mc_add(dev, i->addr);
1783 else
1784 return dev_mc_del(dev, i->addr);
1785 break;
1786 case PACKET_MR_PROMISC:
1787 return dev_set_promiscuity(dev, what);
1788 break;
1789 case PACKET_MR_ALLMULTI:
1790 return dev_set_allmulti(dev, what);
1791 break;
1792 case PACKET_MR_UNICAST:
1793 if (i->alen != dev->addr_len)
1794 return -EINVAL;
1795 if (what > 0)
1796 return dev_uc_add(dev, i->addr);
1797 else
1798 return dev_uc_del(dev, i->addr);
1799 break;
1800 default:
1801 break;
1802 }
1803 return 0;
1804}
1805
1806static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1807{
1808 for ( ; i; i = i->next) {
1809 if (i->ifindex == dev->ifindex)
1810 packet_dev_mc(dev, i, what);
1811 }
1812}
1813
1814static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1815{
1816 struct packet_sock *po = pkt_sk(sk);
1817 struct packet_mclist *ml, *i;
1818 struct net_device *dev;
1819 int err;
1820
1821 rtnl_lock();
1822
1823 err = -ENODEV;
1824 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1825 if (!dev)
1826 goto done;
1827
1828 err = -EINVAL;
1829 if (mreq->mr_alen > dev->addr_len)
1830 goto done;
1831
1832 err = -ENOBUFS;
1833 i = kmalloc(sizeof(*i), GFP_KERNEL);
1834 if (i == NULL)
1835 goto done;
1836
1837 err = 0;
1838 for (ml = po->mclist; ml; ml = ml->next) {
1839 if (ml->ifindex == mreq->mr_ifindex &&
1840 ml->type == mreq->mr_type &&
1841 ml->alen == mreq->mr_alen &&
1842 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1843 ml->count++;
1844 /* Free the new element ... */
1845 kfree(i);
1846 goto done;
1847 }
1848 }
1849
1850 i->type = mreq->mr_type;
1851 i->ifindex = mreq->mr_ifindex;
1852 i->alen = mreq->mr_alen;
1853 memcpy(i->addr, mreq->mr_address, i->alen);
1854 i->count = 1;
1855 i->next = po->mclist;
1856 po->mclist = i;
1857 err = packet_dev_mc(dev, i, 1);
1858 if (err) {
1859 po->mclist = i->next;
1860 kfree(i);
1861 }
1862
1863done:
1864 rtnl_unlock();
1865 return err;
1866}
1867
1868static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1869{
1870 struct packet_mclist *ml, **mlp;
1871
1872 rtnl_lock();
1873
1874 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1875 if (ml->ifindex == mreq->mr_ifindex &&
1876 ml->type == mreq->mr_type &&
1877 ml->alen == mreq->mr_alen &&
1878 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1879 if (--ml->count == 0) {
1880 struct net_device *dev;
1881 *mlp = ml->next;
1882 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1883 if (dev)
1884 packet_dev_mc(dev, ml, -1);
1885 kfree(ml);
1886 }
1887 rtnl_unlock();
1888 return 0;
1889 }
1890 }
1891 rtnl_unlock();
1892 return -EADDRNOTAVAIL;
1893}
1894
1895static void packet_flush_mclist(struct sock *sk)
1896{
1897 struct packet_sock *po = pkt_sk(sk);
1898 struct packet_mclist *ml;
1899
1900 if (!po->mclist)
1901 return;
1902
1903 rtnl_lock();
1904 while ((ml = po->mclist) != NULL) {
1905 struct net_device *dev;
1906
1907 po->mclist = ml->next;
1908 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1909 if (dev != NULL)
1910 packet_dev_mc(dev, ml, -1);
1911 kfree(ml);
1912 }
1913 rtnl_unlock();
1914}
1915
1916static int
1917packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1918{
1919 struct sock *sk = sock->sk;
1920 struct packet_sock *po = pkt_sk(sk);
1921 int ret;
1922
1923 if (level != SOL_PACKET)
1924 return -ENOPROTOOPT;
1925
1926 switch (optname) {
1927 case PACKET_ADD_MEMBERSHIP:
1928 case PACKET_DROP_MEMBERSHIP:
1929 {
1930 struct packet_mreq_max mreq;
1931 int len = optlen;
1932 memset(&mreq, 0, sizeof(mreq));
1933 if (len < sizeof(struct packet_mreq))
1934 return -EINVAL;
1935 if (len > sizeof(mreq))
1936 len = sizeof(mreq);
1937 if (copy_from_user(&mreq, optval, len))
1938 return -EFAULT;
1939 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1940 return -EINVAL;
1941 if (optname == PACKET_ADD_MEMBERSHIP)
1942 ret = packet_mc_add(sk, &mreq);
1943 else
1944 ret = packet_mc_drop(sk, &mreq);
1945 return ret;
1946 }
1947
1948 case PACKET_RX_RING:
1949 case PACKET_TX_RING:
1950 {
1951 struct tpacket_req req;
1952
1953 if (optlen < sizeof(req))
1954 return -EINVAL;
1955 if (pkt_sk(sk)->has_vnet_hdr)
1956 return -EINVAL;
1957 if (copy_from_user(&req, optval, sizeof(req)))
1958 return -EFAULT;
1959 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1960 }
1961 case PACKET_COPY_THRESH:
1962 {
1963 int val;
1964
1965 if (optlen != sizeof(val))
1966 return -EINVAL;
1967 if (copy_from_user(&val, optval, sizeof(val)))
1968 return -EFAULT;
1969
1970 pkt_sk(sk)->copy_thresh = val;
1971 return 0;
1972 }
1973 case PACKET_VERSION:
1974 {
1975 int val;
1976
1977 if (optlen != sizeof(val))
1978 return -EINVAL;
1979 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1980 return -EBUSY;
1981 if (copy_from_user(&val, optval, sizeof(val)))
1982 return -EFAULT;
1983 switch (val) {
1984 case TPACKET_V1:
1985 case TPACKET_V2:
1986 po->tp_version = val;
1987 return 0;
1988 default:
1989 return -EINVAL;
1990 }
1991 }
1992 case PACKET_RESERVE:
1993 {
1994 unsigned int val;
1995
1996 if (optlen != sizeof(val))
1997 return -EINVAL;
1998 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1999 return -EBUSY;
2000 if (copy_from_user(&val, optval, sizeof(val)))
2001 return -EFAULT;
2002 po->tp_reserve = val;
2003 return 0;
2004 }
2005 case PACKET_LOSS:
2006 {
2007 unsigned int val;
2008
2009 if (optlen != sizeof(val))
2010 return -EINVAL;
2011 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2012 return -EBUSY;
2013 if (copy_from_user(&val, optval, sizeof(val)))
2014 return -EFAULT;
2015 po->tp_loss = !!val;
2016 return 0;
2017 }
2018 case PACKET_AUXDATA:
2019 {
2020 int val;
2021
2022 if (optlen < sizeof(val))
2023 return -EINVAL;
2024 if (copy_from_user(&val, optval, sizeof(val)))
2025 return -EFAULT;
2026
2027 po->auxdata = !!val;
2028 return 0;
2029 }
2030 case PACKET_ORIGDEV:
2031 {
2032 int val;
2033
2034 if (optlen < sizeof(val))
2035 return -EINVAL;
2036 if (copy_from_user(&val, optval, sizeof(val)))
2037 return -EFAULT;
2038
2039 po->origdev = !!val;
2040 return 0;
2041 }
2042 case PACKET_VNET_HDR:
2043 {
2044 int val;
2045
2046 if (sock->type != SOCK_RAW)
2047 return -EINVAL;
2048 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2049 return -EBUSY;
2050 if (optlen < sizeof(val))
2051 return -EINVAL;
2052 if (copy_from_user(&val, optval, sizeof(val)))
2053 return -EFAULT;
2054
2055 po->has_vnet_hdr = !!val;
2056 return 0;
2057 }
2058 case PACKET_TIMESTAMP:
2059 {
2060 int val;
2061
2062 if (optlen != sizeof(val))
2063 return -EINVAL;
2064 if (copy_from_user(&val, optval, sizeof(val)))
2065 return -EFAULT;
2066
2067 po->tp_tstamp = val;
2068 return 0;
2069 }
2070 default:
2071 return -ENOPROTOOPT;
2072 }
2073}
2074
2075static int packet_getsockopt(struct socket *sock, int level, int optname,
2076 char __user *optval, int __user *optlen)
2077{
2078 int len;
2079 int val;
2080 struct sock *sk = sock->sk;
2081 struct packet_sock *po = pkt_sk(sk);
2082 void *data;
2083 struct tpacket_stats st;
2084
2085 if (level != SOL_PACKET)
2086 return -ENOPROTOOPT;
2087
2088 if (get_user(len, optlen))
2089 return -EFAULT;
2090
2091 if (len < 0)
2092 return -EINVAL;
2093
2094 switch (optname) {
2095 case PACKET_STATISTICS:
2096 if (len > sizeof(struct tpacket_stats))
2097 len = sizeof(struct tpacket_stats);
2098 spin_lock_bh(&sk->sk_receive_queue.lock);
2099 st = po->stats;
2100 memset(&po->stats, 0, sizeof(st));
2101 spin_unlock_bh(&sk->sk_receive_queue.lock);
2102 st.tp_packets += st.tp_drops;
2103
2104 data = &st;
2105 break;
2106 case PACKET_AUXDATA:
2107 if (len > sizeof(int))
2108 len = sizeof(int);
2109 val = po->auxdata;
2110
2111 data = &val;
2112 break;
2113 case PACKET_ORIGDEV:
2114 if (len > sizeof(int))
2115 len = sizeof(int);
2116 val = po->origdev;
2117
2118 data = &val;
2119 break;
2120 case PACKET_VNET_HDR:
2121 if (len > sizeof(int))
2122 len = sizeof(int);
2123 val = po->has_vnet_hdr;
2124
2125 data = &val;
2126 break;
2127 case PACKET_VERSION:
2128 if (len > sizeof(int))
2129 len = sizeof(int);
2130 val = po->tp_version;
2131 data = &val;
2132 break;
2133 case PACKET_HDRLEN:
2134 if (len > sizeof(int))
2135 len = sizeof(int);
2136 if (copy_from_user(&val, optval, len))
2137 return -EFAULT;
2138 switch (val) {
2139 case TPACKET_V1:
2140 val = sizeof(struct tpacket_hdr);
2141 break;
2142 case TPACKET_V2:
2143 val = sizeof(struct tpacket2_hdr);
2144 break;
2145 default:
2146 return -EINVAL;
2147 }
2148 data = &val;
2149 break;
2150 case PACKET_RESERVE:
2151 if (len > sizeof(unsigned int))
2152 len = sizeof(unsigned int);
2153 val = po->tp_reserve;
2154 data = &val;
2155 break;
2156 case PACKET_LOSS:
2157 if (len > sizeof(unsigned int))
2158 len = sizeof(unsigned int);
2159 val = po->tp_loss;
2160 data = &val;
2161 break;
2162 case PACKET_TIMESTAMP:
2163 if (len > sizeof(int))
2164 len = sizeof(int);
2165 val = po->tp_tstamp;
2166 data = &val;
2167 break;
2168 default:
2169 return -ENOPROTOOPT;
2170 }
2171
2172 if (put_user(len, optlen))
2173 return -EFAULT;
2174 if (copy_to_user(optval, data, len))
2175 return -EFAULT;
2176 return 0;
2177}
2178
2179
2180static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2181{
2182 struct sock *sk;
2183 struct hlist_node *node;
2184 struct net_device *dev = data;
2185 struct net *net = dev_net(dev);
2186
2187 rcu_read_lock();
2188 sk_for_each_rcu(sk, node, &net->packet.sklist) {
2189 struct packet_sock *po = pkt_sk(sk);
2190
2191 switch (msg) {
2192 case NETDEV_UNREGISTER:
2193 if (po->mclist)
2194 packet_dev_mclist(dev, po->mclist, -1);
2195 /* fallthrough */
2196
2197 case NETDEV_DOWN:
2198 if (dev->ifindex == po->ifindex) {
2199 spin_lock(&po->bind_lock);
2200 if (po->running) {
2201 __dev_remove_pack(&po->prot_hook);
2202 __sock_put(sk);
2203 po->running = 0;
2204 sk->sk_err = ENETDOWN;
2205 if (!sock_flag(sk, SOCK_DEAD))
2206 sk->sk_error_report(sk);
2207 }
2208 if (msg == NETDEV_UNREGISTER) {
2209 po->ifindex = -1;
2210 po->prot_hook.dev = NULL;
2211 }
2212 spin_unlock(&po->bind_lock);
2213 }
2214 break;
2215 case NETDEV_UP:
2216 if (dev->ifindex == po->ifindex) {
2217 spin_lock(&po->bind_lock);
2218 if (po->num && !po->running) {
2219 dev_add_pack(&po->prot_hook);
2220 sock_hold(sk);
2221 po->running = 1;
2222 }
2223 spin_unlock(&po->bind_lock);
2224 }
2225 break;
2226 }
2227 }
2228 rcu_read_unlock();
2229 return NOTIFY_DONE;
2230}
2231
2232
2233static int packet_ioctl(struct socket *sock, unsigned int cmd,
2234 unsigned long arg)
2235{
2236 struct sock *sk = sock->sk;
2237
2238 switch (cmd) {
2239 case SIOCOUTQ:
2240 {
2241 int amount = sk_wmem_alloc_get(sk);
2242
2243 return put_user(amount, (int __user *)arg);
2244 }
2245 case SIOCINQ:
2246 {
2247 struct sk_buff *skb;
2248 int amount = 0;
2249
2250 spin_lock_bh(&sk->sk_receive_queue.lock);
2251 skb = skb_peek(&sk->sk_receive_queue);
2252 if (skb)
2253 amount = skb->len;
2254 spin_unlock_bh(&sk->sk_receive_queue.lock);
2255 return put_user(amount, (int __user *)arg);
2256 }
2257 case SIOCGSTAMP:
2258 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2259 case SIOCGSTAMPNS:
2260 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2261
2262#ifdef CONFIG_INET
2263 case SIOCADDRT:
2264 case SIOCDELRT:
2265 case SIOCDARP:
2266 case SIOCGARP:
2267 case SIOCSARP:
2268 case SIOCGIFADDR:
2269 case SIOCSIFADDR:
2270 case SIOCGIFBRDADDR:
2271 case SIOCSIFBRDADDR:
2272 case SIOCGIFNETMASK:
2273 case SIOCSIFNETMASK:
2274 case SIOCGIFDSTADDR:
2275 case SIOCSIFDSTADDR:
2276 case SIOCSIFFLAGS:
2277 return inet_dgram_ops.ioctl(sock, cmd, arg);
2278#endif
2279
2280 default:
2281 return -ENOIOCTLCMD;
2282 }
2283 return 0;
2284}
2285
2286static unsigned int packet_poll(struct file *file, struct socket *sock,
2287 poll_table *wait)
2288{
2289 struct sock *sk = sock->sk;
2290 struct packet_sock *po = pkt_sk(sk);
2291 unsigned int mask = datagram_poll(file, sock, wait);
2292
2293 spin_lock_bh(&sk->sk_receive_queue.lock);
2294 if (po->rx_ring.pg_vec) {
2295 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2296 mask |= POLLIN | POLLRDNORM;
2297 }
2298 spin_unlock_bh(&sk->sk_receive_queue.lock);
2299 spin_lock_bh(&sk->sk_write_queue.lock);
2300 if (po->tx_ring.pg_vec) {
2301 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2302 mask |= POLLOUT | POLLWRNORM;
2303 }
2304 spin_unlock_bh(&sk->sk_write_queue.lock);
2305 return mask;
2306}
2307
2308
2309/* Dirty? Well, I still did not learn better way to account
2310 * for user mmaps.
2311 */
2312
2313static void packet_mm_open(struct vm_area_struct *vma)
2314{
2315 struct file *file = vma->vm_file;
2316 struct socket *sock = file->private_data;
2317 struct sock *sk = sock->sk;
2318
2319 if (sk)
2320 atomic_inc(&pkt_sk(sk)->mapped);
2321}
2322
2323static void packet_mm_close(struct vm_area_struct *vma)
2324{
2325 struct file *file = vma->vm_file;
2326 struct socket *sock = file->private_data;
2327 struct sock *sk = sock->sk;
2328
2329 if (sk)
2330 atomic_dec(&pkt_sk(sk)->mapped);
2331}
2332
2333static const struct vm_operations_struct packet_mmap_ops = {
2334 .open = packet_mm_open,
2335 .close = packet_mm_close,
2336};
2337
2338static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
2339 unsigned int len)
2340{
2341 int i;
2342
2343 for (i = 0; i < len; i++) {
2344 if (likely(pg_vec[i].buffer)) {
2345 if (is_vmalloc_addr(pg_vec[i].buffer))
2346 vfree(pg_vec[i].buffer);
2347 else
2348 free_pages((unsigned long)pg_vec[i].buffer,
2349 order);
2350 pg_vec[i].buffer = NULL;
2351 }
2352 }
2353 kfree(pg_vec);
2354}
2355
2356static inline char *alloc_one_pg_vec_page(unsigned long order)
2357{
2358 char *buffer = NULL;
2359 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
2360 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
2361
2362 buffer = (char *) __get_free_pages(gfp_flags, order);
2363
2364 if (buffer)
2365 return buffer;
2366
2367 /*
2368 * __get_free_pages failed, fall back to vmalloc
2369 */
2370 buffer = vzalloc((1 << order) * PAGE_SIZE);
2371
2372 if (buffer)
2373 return buffer;
2374
2375 /*
2376 * vmalloc failed, lets dig into swap here
2377 */
2378 gfp_flags &= ~__GFP_NORETRY;
2379 buffer = (char *)__get_free_pages(gfp_flags, order);
2380 if (buffer)
2381 return buffer;
2382
2383 /*
2384 * complete and utter failure
2385 */
2386 return NULL;
2387}
2388
2389static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
2390{
2391 unsigned int block_nr = req->tp_block_nr;
2392 struct pgv *pg_vec;
2393 int i;
2394
2395 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
2396 if (unlikely(!pg_vec))
2397 goto out;
2398
2399 for (i = 0; i < block_nr; i++) {
2400 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
2401 if (unlikely(!pg_vec[i].buffer))
2402 goto out_free_pgvec;
2403 }
2404
2405out:
2406 return pg_vec;
2407
2408out_free_pgvec:
2409 free_pg_vec(pg_vec, order, block_nr);
2410 kfree(pg_vec);
2411 pg_vec = NULL;
2412 goto out;
2413}
2414
2415static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2416 int closing, int tx_ring)
2417{
2418 struct pgv *pg_vec = NULL;
2419 struct packet_sock *po = pkt_sk(sk);
2420 int was_running, order = 0;
2421 struct packet_ring_buffer *rb;
2422 struct sk_buff_head *rb_queue;
2423 __be16 num;
2424 int err;
2425
2426 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2427 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2428
2429 err = -EBUSY;
2430 if (!closing) {
2431 if (atomic_read(&po->mapped))
2432 goto out;
2433 if (atomic_read(&rb->pending))
2434 goto out;
2435 }
2436
2437 if (req->tp_block_nr) {
2438 /* Sanity tests and some calculations */
2439 err = -EBUSY;
2440 if (unlikely(rb->pg_vec))
2441 goto out;
2442
2443 switch (po->tp_version) {
2444 case TPACKET_V1:
2445 po->tp_hdrlen = TPACKET_HDRLEN;
2446 break;
2447 case TPACKET_V2:
2448 po->tp_hdrlen = TPACKET2_HDRLEN;
2449 break;
2450 }
2451
2452 err = -EINVAL;
2453 if (unlikely((int)req->tp_block_size <= 0))
2454 goto out;
2455 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2456 goto out;
2457 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2458 po->tp_reserve))
2459 goto out;
2460 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2461 goto out;
2462
2463 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2464 if (unlikely(rb->frames_per_block <= 0))
2465 goto out;
2466 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2467 req->tp_frame_nr))
2468 goto out;
2469
2470 err = -ENOMEM;
2471 order = get_order(req->tp_block_size);
2472 pg_vec = alloc_pg_vec(req, order);
2473 if (unlikely(!pg_vec))
2474 goto out;
2475 }
2476 /* Done */
2477 else {
2478 err = -EINVAL;
2479 if (unlikely(req->tp_frame_nr))
2480 goto out;
2481 }
2482
2483 lock_sock(sk);
2484
2485 /* Detach socket from network */
2486 spin_lock(&po->bind_lock);
2487 was_running = po->running;
2488 num = po->num;
2489 if (was_running) {
2490 __dev_remove_pack(&po->prot_hook);
2491 po->num = 0;
2492 po->running = 0;
2493 __sock_put(sk);
2494 }
2495 spin_unlock(&po->bind_lock);
2496
2497 synchronize_net();
2498
2499 err = -EBUSY;
2500 mutex_lock(&po->pg_vec_lock);
2501 if (closing || atomic_read(&po->mapped) == 0) {
2502 err = 0;
2503#define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
2504 spin_lock_bh(&rb_queue->lock);
2505 pg_vec = XC(rb->pg_vec, pg_vec);
2506 rb->frame_max = (req->tp_frame_nr - 1);
2507 rb->head = 0;
2508 rb->frame_size = req->tp_frame_size;
2509 spin_unlock_bh(&rb_queue->lock);
2510
2511 order = XC(rb->pg_vec_order, order);
2512 req->tp_block_nr = XC(rb->pg_vec_len, req->tp_block_nr);
2513
2514 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2515 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2516 tpacket_rcv : packet_rcv;
2517 skb_queue_purge(rb_queue);
2518#undef XC
2519 if (atomic_read(&po->mapped))
2520 pr_err("packet_mmap: vma is busy: %d\n",
2521 atomic_read(&po->mapped));
2522 }
2523 mutex_unlock(&po->pg_vec_lock);
2524
2525 spin_lock(&po->bind_lock);
2526 if (was_running && !po->running) {
2527 sock_hold(sk);
2528 po->running = 1;
2529 po->num = num;
2530 dev_add_pack(&po->prot_hook);
2531 }
2532 spin_unlock(&po->bind_lock);
2533
2534 release_sock(sk);
2535
2536 if (pg_vec)
2537 free_pg_vec(pg_vec, order, req->tp_block_nr);
2538out:
2539 return err;
2540}
2541
2542static int packet_mmap(struct file *file, struct socket *sock,
2543 struct vm_area_struct *vma)
2544{
2545 struct sock *sk = sock->sk;
2546 struct packet_sock *po = pkt_sk(sk);
2547 unsigned long size, expected_size;
2548 struct packet_ring_buffer *rb;
2549 unsigned long start;
2550 int err = -EINVAL;
2551 int i;
2552
2553 if (vma->vm_pgoff)
2554 return -EINVAL;
2555
2556 mutex_lock(&po->pg_vec_lock);
2557
2558 expected_size = 0;
2559 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2560 if (rb->pg_vec) {
2561 expected_size += rb->pg_vec_len
2562 * rb->pg_vec_pages
2563 * PAGE_SIZE;
2564 }
2565 }
2566
2567 if (expected_size == 0)
2568 goto out;
2569
2570 size = vma->vm_end - vma->vm_start;
2571 if (size != expected_size)
2572 goto out;
2573
2574 start = vma->vm_start;
2575 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2576 if (rb->pg_vec == NULL)
2577 continue;
2578
2579 for (i = 0; i < rb->pg_vec_len; i++) {
2580 struct page *page;
2581 void *kaddr = rb->pg_vec[i].buffer;
2582 int pg_num;
2583
2584 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
2585 page = pgv_to_page(kaddr);
2586 err = vm_insert_page(vma, start, page);
2587 if (unlikely(err))
2588 goto out;
2589 start += PAGE_SIZE;
2590 kaddr += PAGE_SIZE;
2591 }
2592 }
2593 }
2594
2595 atomic_inc(&po->mapped);
2596 vma->vm_ops = &packet_mmap_ops;
2597 err = 0;
2598
2599out:
2600 mutex_unlock(&po->pg_vec_lock);
2601 return err;
2602}
2603
2604static const struct proto_ops packet_ops_spkt = {
2605 .family = PF_PACKET,
2606 .owner = THIS_MODULE,
2607 .release = packet_release,
2608 .bind = packet_bind_spkt,
2609 .connect = sock_no_connect,
2610 .socketpair = sock_no_socketpair,
2611 .accept = sock_no_accept,
2612 .getname = packet_getname_spkt,
2613 .poll = datagram_poll,
2614 .ioctl = packet_ioctl,
2615 .listen = sock_no_listen,
2616 .shutdown = sock_no_shutdown,
2617 .setsockopt = sock_no_setsockopt,
2618 .getsockopt = sock_no_getsockopt,
2619 .sendmsg = packet_sendmsg_spkt,
2620 .recvmsg = packet_recvmsg,
2621 .mmap = sock_no_mmap,
2622 .sendpage = sock_no_sendpage,
2623};
2624
2625static const struct proto_ops packet_ops = {
2626 .family = PF_PACKET,
2627 .owner = THIS_MODULE,
2628 .release = packet_release,
2629 .bind = packet_bind,
2630 .connect = sock_no_connect,
2631 .socketpair = sock_no_socketpair,
2632 .accept = sock_no_accept,
2633 .getname = packet_getname,
2634 .poll = packet_poll,
2635 .ioctl = packet_ioctl,
2636 .listen = sock_no_listen,
2637 .shutdown = sock_no_shutdown,
2638 .setsockopt = packet_setsockopt,
2639 .getsockopt = packet_getsockopt,
2640 .sendmsg = packet_sendmsg,
2641 .recvmsg = packet_recvmsg,
2642 .mmap = packet_mmap,
2643 .sendpage = sock_no_sendpage,
2644};
2645
2646static const struct net_proto_family packet_family_ops = {
2647 .family = PF_PACKET,
2648 .create = packet_create,
2649 .owner = THIS_MODULE,
2650};
2651
2652static struct notifier_block packet_netdev_notifier = {
2653 .notifier_call = packet_notifier,
2654};
2655
2656#ifdef CONFIG_PROC_FS
2657
2658static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2659 __acquires(RCU)
2660{
2661 struct net *net = seq_file_net(seq);
2662
2663 rcu_read_lock();
2664 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
2665}
2666
2667static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2668{
2669 struct net *net = seq_file_net(seq);
2670 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
2671}
2672
2673static void packet_seq_stop(struct seq_file *seq, void *v)
2674 __releases(RCU)
2675{
2676 rcu_read_unlock();
2677}
2678
2679static int packet_seq_show(struct seq_file *seq, void *v)
2680{
2681 if (v == SEQ_START_TOKEN)
2682 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
2683 else {
2684 struct sock *s = sk_entry(v);
2685 const struct packet_sock *po = pkt_sk(s);
2686
2687 seq_printf(seq,
2688 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
2689 s,
2690 atomic_read(&s->sk_refcnt),
2691 s->sk_type,
2692 ntohs(po->num),
2693 po->ifindex,
2694 po->running,
2695 atomic_read(&s->sk_rmem_alloc),
2696 sock_i_uid(s),
2697 sock_i_ino(s));
2698 }
2699
2700 return 0;
2701}
2702
2703static const struct seq_operations packet_seq_ops = {
2704 .start = packet_seq_start,
2705 .next = packet_seq_next,
2706 .stop = packet_seq_stop,
2707 .show = packet_seq_show,
2708};
2709
2710static int packet_seq_open(struct inode *inode, struct file *file)
2711{
2712 return seq_open_net(inode, file, &packet_seq_ops,
2713 sizeof(struct seq_net_private));
2714}
2715
2716static const struct file_operations packet_seq_fops = {
2717 .owner = THIS_MODULE,
2718 .open = packet_seq_open,
2719 .read = seq_read,
2720 .llseek = seq_lseek,
2721 .release = seq_release_net,
2722};
2723
2724#endif
2725
2726static int __net_init packet_net_init(struct net *net)
2727{
2728 spin_lock_init(&net->packet.sklist_lock);
2729 INIT_HLIST_HEAD(&net->packet.sklist);
2730
2731 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2732 return -ENOMEM;
2733
2734 return 0;
2735}
2736
2737static void __net_exit packet_net_exit(struct net *net)
2738{
2739 proc_net_remove(net, "packet");
2740}
2741
2742static struct pernet_operations packet_net_ops = {
2743 .init = packet_net_init,
2744 .exit = packet_net_exit,
2745};
2746
2747
2748static void __exit packet_exit(void)
2749{
2750 unregister_netdevice_notifier(&packet_netdev_notifier);
2751 unregister_pernet_subsys(&packet_net_ops);
2752 sock_unregister(PF_PACKET);
2753 proto_unregister(&packet_proto);
2754}
2755
2756static int __init packet_init(void)
2757{
2758 int rc = proto_register(&packet_proto, 0);
2759
2760 if (rc != 0)
2761 goto out;
2762
2763 sock_register(&packet_family_ops);
2764 register_pernet_subsys(&packet_net_ops);
2765 register_netdevice_notifier(&packet_netdev_notifier);
2766out:
2767 return rc;
2768}
2769
2770module_init(packet_init);
2771module_exit(packet_exit);
2772MODULE_LICENSE("GPL");
2773MODULE_ALIAS_NETPROTO(PF_PACKET);
Linux kernel - Deliverables
This page took 0.033581 seconds and 5 git commands to generate.