?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ocfs2: Make the left masklogs compat.
[taoma-kernel.git] / net / packet / af_packet.c
blob91cb1d71f018d8b6b77c029dca1b99cb2855ef21
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
104 On receive:
105 -----------
106
107 Incoming, dev->hard_header!=NULL
108 mac_header -> ll header
109 data -> data
110
111 Outgoing, dev->hard_header!=NULL
112 mac_header -> ll header
113 data -> ll header
114
115 Incoming, 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
121 Outgoing, dev->hard_header==NULL
122 mac_header -> data. ll header is still not built!
123 data -> data
124
125 Resume
126 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
127
128
129 On transmit:
130 ------------
131
132 dev->hard_header != NULL
133 mac_header -> ll header
134 data -> ll header
135
136 dev->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
146 struct 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 */
157 struct 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
164 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
165 int closing, int tx_ring);
166
167 #define PGV_FROM_VMALLOC 1
168 struct pgv {
169 char *buffer;
170 };
171
172 struct 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
186 struct packet_sock;
187 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
188
189 static void packet_flush_mclist(struct sock *sk);
190
191 struct 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
216 struct 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
226 static inline __pure 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
233 static 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
259 static 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
284 static 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
308 static 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
315 static 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
323 static 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
328 static inline struct packet_sock *pkt_sk(struct sock *sk)
329 {
330 return (struct packet_sock *)sk;
331 }
332
333 static 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
349 static const struct proto_ops packet_ops;
350
351 static const struct proto_ops packet_ops_spkt;
352
353 static 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
413 out:
414 kfree_skb(skb);
415 oom:
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
425 static 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;
452 retry:
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
513 out_unlock:
514 rcu_read_unlock();
515 out_free:
516 kfree_skb(skb);
517 return err;
518 }
519
520 static 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
547 static 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
644 drop_n_acct:
645 po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
646
647 drop_n_restore:
648 if (skb_head != skb->data && skb_shared(skb)) {
649 skb->data = skb_head;
650 skb->len = skb_len;
651 }
652 drop:
653 consume_skb(skb);
654 return 0;
655 }
656
657 static 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 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
810 {
811 u8 *start, *end;
812
813 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw + macoff + snaplen);
814 for (start = h.raw; start < end; start += PAGE_SIZE)
815 flush_dcache_page(pgv_to_page(start));
816 }
817 #endif
818
819 sk->sk_data_ready(sk, 0);
820
821 drop_n_restore:
822 if (skb_head != skb->data && skb_shared(skb)) {
823 skb->data = skb_head;
824 skb->len = skb_len;
825 }
826 drop:
827 kfree_skb(skb);
828 return 0;
829
830 ring_is_full:
831 po->stats.tp_drops++;
832 spin_unlock(&sk->sk_receive_queue.lock);
833
834 sk->sk_data_ready(sk, 0);
835 kfree_skb(copy_skb);
836 goto drop_n_restore;
837 }
838
839 static void tpacket_destruct_skb(struct sk_buff *skb)
840 {
841 struct packet_sock *po = pkt_sk(skb->sk);
842 void *ph;
843
844 BUG_ON(skb == NULL);
845
846 if (likely(po->tx_ring.pg_vec)) {
847 ph = skb_shinfo(skb)->destructor_arg;
848 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
849 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
850 atomic_dec(&po->tx_ring.pending);
851 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
852 }
853
854 sock_wfree(skb);
855 }
856
857 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
858 void *frame, struct net_device *dev, int size_max,
859 __be16 proto, unsigned char *addr)
860 {
861 union {
862 struct tpacket_hdr *h1;
863 struct tpacket2_hdr *h2;
864 void *raw;
865 } ph;
866 int to_write, offset, len, tp_len, nr_frags, len_max;
867 struct socket *sock = po->sk.sk_socket;
868 struct page *page;
869 void *data;
870 int err;
871
872 ph.raw = frame;
873
874 skb->protocol = proto;
875 skb->dev = dev;
876 skb->priority = po->sk.sk_priority;
877 skb->mark = po->sk.sk_mark;
878 skb_shinfo(skb)->destructor_arg = ph.raw;
879
880 switch (po->tp_version) {
881 case TPACKET_V2:
882 tp_len = ph.h2->tp_len;
883 break;
884 default:
885 tp_len = ph.h1->tp_len;
886 break;
887 }
888 if (unlikely(tp_len > size_max)) {
889 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
890 return -EMSGSIZE;
891 }
892
893 skb_reserve(skb, LL_RESERVED_SPACE(dev));
894 skb_reset_network_header(skb);
895
896 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
897 to_write = tp_len;
898
899 if (sock->type == SOCK_DGRAM) {
900 err = dev_hard_header(skb, dev, ntohs(proto), addr,
901 NULL, tp_len);
902 if (unlikely(err < 0))
903 return -EINVAL;
904 } else if (dev->hard_header_len) {
905 /* net device doesn't like empty head */
906 if (unlikely(tp_len <= dev->hard_header_len)) {
907 pr_err("packet size is too short (%d < %d)\n",
908 tp_len, dev->hard_header_len);
909 return -EINVAL;
910 }
911
912 skb_push(skb, dev->hard_header_len);
913 err = skb_store_bits(skb, 0, data,
914 dev->hard_header_len);
915 if (unlikely(err))
916 return err;
917
918 data += dev->hard_header_len;
919 to_write -= dev->hard_header_len;
920 }
921
922 err = -EFAULT;
923 offset = offset_in_page(data);
924 len_max = PAGE_SIZE - offset;
925 len = ((to_write > len_max) ? len_max : to_write);
926
927 skb->data_len = to_write;
928 skb->len += to_write;
929 skb->truesize += to_write;
930 atomic_add(to_write, &po->sk.sk_wmem_alloc);
931
932 while (likely(to_write)) {
933 nr_frags = skb_shinfo(skb)->nr_frags;
934
935 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
936 pr_err("Packet exceed the number of skb frags(%lu)\n",
937 MAX_SKB_FRAGS);
938 return -EFAULT;
939 }
940
941 page = pgv_to_page(data);
942 data += len;
943 flush_dcache_page(page);
944 get_page(page);
945 skb_fill_page_desc(skb, nr_frags, page, offset, len);
946 to_write -= len;
947 offset = 0;
948 len_max = PAGE_SIZE;
949 len = ((to_write > len_max) ? len_max : to_write);
950 }
951
952 return tp_len;
953 }
954
955 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
956 {
957 struct socket *sock;
958 struct sk_buff *skb;
959 struct net_device *dev;
960 __be16 proto;
961 int ifindex, err, reserve = 0;
962 void *ph;
963 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
964 int tp_len, size_max;
965 unsigned char *addr;
966 int len_sum = 0;
967 int status = 0;
968
969 sock = po->sk.sk_socket;
970
971 mutex_lock(&po->pg_vec_lock);
972
973 err = -EBUSY;
974 if (saddr == NULL) {
975 ifindex = po->ifindex;
976 proto = po->num;
977 addr = NULL;
978 } else {
979 err = -EINVAL;
980 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
981 goto out;
982 if (msg->msg_namelen < (saddr->sll_halen
983 + offsetof(struct sockaddr_ll,
984 sll_addr)))
985 goto out;
986 ifindex = saddr->sll_ifindex;
987 proto = saddr->sll_protocol;
988 addr = saddr->sll_addr;
989 }
990
991 dev = dev_get_by_index(sock_net(&po->sk), ifindex);
992 err = -ENXIO;
993 if (unlikely(dev == NULL))
994 goto out;
995
996 reserve = dev->hard_header_len;
997
998 err = -ENETDOWN;
999 if (unlikely(!(dev->flags & IFF_UP)))
1000 goto out_put;
1001
1002 size_max = po->tx_ring.frame_size
1003 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1004
1005 if (size_max > dev->mtu + reserve)
1006 size_max = dev->mtu + reserve;
1007
1008 do {
1009 ph = packet_current_frame(po, &po->tx_ring,
1010 TP_STATUS_SEND_REQUEST);
1011
1012 if (unlikely(ph == NULL)) {
1013 schedule();
1014 continue;
1015 }
1016
1017 status = TP_STATUS_SEND_REQUEST;
1018 skb = sock_alloc_send_skb(&po->sk,
1019 LL_ALLOCATED_SPACE(dev)
1020 + sizeof(struct sockaddr_ll),
1021 0, &err);
1022
1023 if (unlikely(skb == NULL))
1024 goto out_status;
1025
1026 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1027 addr);
1028
1029 if (unlikely(tp_len < 0)) {
1030 if (po->tp_loss) {
1031 __packet_set_status(po, ph,
1032 TP_STATUS_AVAILABLE);
1033 packet_increment_head(&po->tx_ring);
1034 kfree_skb(skb);
1035 continue;
1036 } else {
1037 status = TP_STATUS_WRONG_FORMAT;
1038 err = tp_len;
1039 goto out_status;
1040 }
1041 }
1042
1043 skb->destructor = tpacket_destruct_skb;
1044 __packet_set_status(po, ph, TP_STATUS_SENDING);
1045 atomic_inc(&po->tx_ring.pending);
1046
1047 status = TP_STATUS_SEND_REQUEST;
1048 err = dev_queue_xmit(skb);
1049 if (unlikely(err > 0)) {
1050 err = net_xmit_errno(err);
1051 if (err && __packet_get_status(po, ph) ==
1052 TP_STATUS_AVAILABLE) {
1053 /* skb was destructed already */
1054 skb = NULL;
1055 goto out_status;
1056 }
1057 /*
1058 * skb was dropped but not destructed yet;
1059 * let's treat it like congestion or err < 0
1060 */
1061 err = 0;
1062 }
1063 packet_increment_head(&po->tx_ring);
1064 len_sum += tp_len;
1065 } while (likely((ph != NULL) ||
1066 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1067 (atomic_read(&po->tx_ring.pending))))
1068 );
1069
1070 err = len_sum;
1071 goto out_put;
1072
1073 out_status:
1074 __packet_set_status(po, ph, status);
1075 kfree_skb(skb);
1076 out_put:
1077 dev_put(dev);
1078 out:
1079 mutex_unlock(&po->pg_vec_lock);
1080 return err;
1081 }
1082
1083 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1084 size_t reserve, size_t len,
1085 size_t linear, int noblock,
1086 int *err)
1087 {
1088 struct sk_buff *skb;
1089
1090 /* Under a page? Don't bother with paged skb. */
1091 if (prepad + len < PAGE_SIZE || !linear)
1092 linear = len;
1093
1094 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1095 err);
1096 if (!skb)
1097 return NULL;
1098
1099 skb_reserve(skb, reserve);
1100 skb_put(skb, linear);
1101 skb->data_len = len - linear;
1102 skb->len += len - linear;
1103
1104 return skb;
1105 }
1106
1107 static int packet_snd(struct socket *sock,
1108 struct msghdr *msg, size_t len)
1109 {
1110 struct sock *sk = sock->sk;
1111 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1112 struct sk_buff *skb;
1113 struct net_device *dev;
1114 __be16 proto;
1115 unsigned char *addr;
1116 int ifindex, err, reserve = 0;
1117 struct virtio_net_hdr vnet_hdr = { 0 };
1118 int offset = 0;
1119 int vnet_hdr_len;
1120 struct packet_sock *po = pkt_sk(sk);
1121 unsigned short gso_type = 0;
1122
1123 /*
1124 * Get and verify the address.
1125 */
1126
1127 if (saddr == NULL) {
1128 ifindex = po->ifindex;
1129 proto = po->num;
1130 addr = NULL;
1131 } else {
1132 err = -EINVAL;
1133 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1134 goto out;
1135 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1136 goto out;
1137 ifindex = saddr->sll_ifindex;
1138 proto = saddr->sll_protocol;
1139 addr = saddr->sll_addr;
1140 }
1141
1142
1143 dev = dev_get_by_index(sock_net(sk), ifindex);
1144 err = -ENXIO;
1145 if (dev == NULL)
1146 goto out_unlock;
1147 if (sock->type == SOCK_RAW)
1148 reserve = dev->hard_header_len;
1149
1150 err = -ENETDOWN;
1151 if (!(dev->flags & IFF_UP))
1152 goto out_unlock;
1153
1154 if (po->has_vnet_hdr) {
1155 vnet_hdr_len = sizeof(vnet_hdr);
1156
1157 err = -EINVAL;
1158 if (len < vnet_hdr_len)
1159 goto out_unlock;
1160
1161 len -= vnet_hdr_len;
1162
1163 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1164 vnet_hdr_len);
1165 if (err < 0)
1166 goto out_unlock;
1167
1168 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1169 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1170 vnet_hdr.hdr_len))
1171 vnet_hdr.hdr_len = vnet_hdr.csum_start +
1172 vnet_hdr.csum_offset + 2;
1173
1174 err = -EINVAL;
1175 if (vnet_hdr.hdr_len > len)
1176 goto out_unlock;
1177
1178 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1179 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1180 case VIRTIO_NET_HDR_GSO_TCPV4:
1181 gso_type = SKB_GSO_TCPV4;
1182 break;
1183 case VIRTIO_NET_HDR_GSO_TCPV6:
1184 gso_type = SKB_GSO_TCPV6;
1185 break;
1186 case VIRTIO_NET_HDR_GSO_UDP:
1187 gso_type = SKB_GSO_UDP;
1188 break;
1189 default:
1190 goto out_unlock;
1191 }
1192
1193 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1194 gso_type |= SKB_GSO_TCP_ECN;
1195
1196 if (vnet_hdr.gso_size == 0)
1197 goto out_unlock;
1198
1199 }
1200 }
1201
1202 err = -EMSGSIZE;
1203 if (!gso_type && (len > dev->mtu+reserve))
1204 goto out_unlock;
1205
1206 err = -ENOBUFS;
1207 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1208 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1209 msg->msg_flags & MSG_DONTWAIT, &err);
1210 if (skb == NULL)
1211 goto out_unlock;
1212
1213 skb_set_network_header(skb, reserve);
1214
1215 err = -EINVAL;
1216 if (sock->type == SOCK_DGRAM &&
1217 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1218 goto out_free;
1219
1220 /* Returns -EFAULT on error */
1221 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1222 if (err)
1223 goto out_free;
1224 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1225 if (err < 0)
1226 goto out_free;
1227
1228 skb->protocol = proto;
1229 skb->dev = dev;
1230 skb->priority = sk->sk_priority;
1231 skb->mark = sk->sk_mark;
1232
1233 if (po->has_vnet_hdr) {
1234 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1235 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1236 vnet_hdr.csum_offset)) {
1237 err = -EINVAL;
1238 goto out_free;
1239 }
1240 }
1241
1242 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1243 skb_shinfo(skb)->gso_type = gso_type;
1244
1245 /* Header must be checked, and gso_segs computed. */
1246 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1247 skb_shinfo(skb)->gso_segs = 0;
1248
1249 len += vnet_hdr_len;
1250 }
1251
1252 /*
1253 * Now send it
1254 */
1255
1256 err = dev_queue_xmit(skb);
1257 if (err > 0 && (err = net_xmit_errno(err)) != 0)
1258 goto out_unlock;
1259
1260 dev_put(dev);
1261
1262 return len;
1263
1264 out_free:
1265 kfree_skb(skb);
1266 out_unlock:
1267 if (dev)
1268 dev_put(dev);
1269 out:
1270 return err;
1271 }
1272
1273 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1274 struct msghdr *msg, size_t len)
1275 {
1276 struct sock *sk = sock->sk;
1277 struct packet_sock *po = pkt_sk(sk);
1278 if (po->tx_ring.pg_vec)
1279 return tpacket_snd(po, msg);
1280 else
1281 return packet_snd(sock, msg, len);
1282 }
1283
1284 /*
1285 * Close a PACKET socket. This is fairly simple. We immediately go
1286 * to 'closed' state and remove our protocol entry in the device list.
1287 */
1288
1289 static int packet_release(struct socket *sock)
1290 {
1291 struct sock *sk = sock->sk;
1292 struct packet_sock *po;
1293 struct net *net;
1294 struct tpacket_req req;
1295
1296 if (!sk)
1297 return 0;
1298
1299 net = sock_net(sk);
1300 po = pkt_sk(sk);
1301
1302 spin_lock_bh(&net->packet.sklist_lock);
1303 sk_del_node_init_rcu(sk);
1304 sock_prot_inuse_add(net, sk->sk_prot, -1);
1305 spin_unlock_bh(&net->packet.sklist_lock);
1306
1307 spin_lock(&po->bind_lock);
1308 if (po->running) {
1309 /*
1310 * Remove from protocol table
1311 */
1312 po->running = 0;
1313 po->num = 0;
1314 __dev_remove_pack(&po->prot_hook);
1315 __sock_put(sk);
1316 }
1317 spin_unlock(&po->bind_lock);
1318
1319 packet_flush_mclist(sk);
1320
1321 memset(&req, 0, sizeof(req));
1322
1323 if (po->rx_ring.pg_vec)
1324 packet_set_ring(sk, &req, 1, 0);
1325
1326 if (po->tx_ring.pg_vec)
1327 packet_set_ring(sk, &req, 1, 1);
1328
1329 synchronize_net();
1330 /*
1331 * Now the socket is dead. No more input will appear.
1332 */
1333 sock_orphan(sk);
1334 sock->sk = NULL;
1335
1336 /* Purge queues */
1337
1338 skb_queue_purge(&sk->sk_receive_queue);
1339 sk_refcnt_debug_release(sk);
1340
1341 sock_put(sk);
1342 return 0;
1343 }
1344
1345 /*
1346 * Attach a packet hook.
1347 */
1348
1349 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1350 {
1351 struct packet_sock *po = pkt_sk(sk);
1352 /*
1353 * Detach an existing hook if present.
1354 */
1355
1356 lock_sock(sk);
1357
1358 spin_lock(&po->bind_lock);
1359 if (po->running) {
1360 __sock_put(sk);
1361 po->running = 0;
1362 po->num = 0;
1363 spin_unlock(&po->bind_lock);
1364 dev_remove_pack(&po->prot_hook);
1365 spin_lock(&po->bind_lock);
1366 }
1367
1368 po->num = protocol;
1369 po->prot_hook.type = protocol;
1370 po->prot_hook.dev = dev;
1371
1372 po->ifindex = dev ? dev->ifindex : 0;
1373
1374 if (protocol == 0)
1375 goto out_unlock;
1376
1377 if (!dev || (dev->flags & IFF_UP)) {
1378 dev_add_pack(&po->prot_hook);
1379 sock_hold(sk);
1380 po->running = 1;
1381 } else {
1382 sk->sk_err = ENETDOWN;
1383 if (!sock_flag(sk, SOCK_DEAD))
1384 sk->sk_error_report(sk);
1385 }
1386
1387 out_unlock:
1388 spin_unlock(&po->bind_lock);
1389 release_sock(sk);
1390 return 0;
1391 }
1392
1393 /*
1394 * Bind a packet socket to a device
1395 */
1396
1397 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1398 int addr_len)
1399 {
1400 struct sock *sk = sock->sk;
1401 char name[15];
1402 struct net_device *dev;
1403 int err = -ENODEV;
1404
1405 /*
1406 * Check legality
1407 */
1408
1409 if (addr_len != sizeof(struct sockaddr))
1410 return -EINVAL;
1411 strlcpy(name, uaddr->sa_data, sizeof(name));
1412
1413 dev = dev_get_by_name(sock_net(sk), name);
1414 if (dev) {
1415 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1416 dev_put(dev);
1417 }
1418 return err;
1419 }
1420
1421 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1422 {
1423 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1424 struct sock *sk = sock->sk;
1425 struct net_device *dev = NULL;
1426 int err;
1427
1428
1429 /*
1430 * Check legality
1431 */
1432
1433 if (addr_len < sizeof(struct sockaddr_ll))
1434 return -EINVAL;
1435 if (sll->sll_family != AF_PACKET)
1436 return -EINVAL;
1437
1438 if (sll->sll_ifindex) {
1439 err = -ENODEV;
1440 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1441 if (dev == NULL)
1442 goto out;
1443 }
1444 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1445 if (dev)
1446 dev_put(dev);
1447
1448 out:
1449 return err;
1450 }
1451
1452 static struct proto packet_proto = {
1453 .name = "PACKET",
1454 .owner = THIS_MODULE,
1455 .obj_size = sizeof(struct packet_sock),
1456 };
1457
1458 /*
1459 * Create a packet of type SOCK_PACKET.
1460 */
1461
1462 static int packet_create(struct net *net, struct socket *sock, int protocol,
1463 int kern)
1464 {
1465 struct sock *sk;
1466 struct packet_sock *po;
1467 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1468 int err;
1469
1470 if (!capable(CAP_NET_RAW))
1471 return -EPERM;
1472 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1473 sock->type != SOCK_PACKET)
1474 return -ESOCKTNOSUPPORT;
1475
1476 sock->state = SS_UNCONNECTED;
1477
1478 err = -ENOBUFS;
1479 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1480 if (sk == NULL)
1481 goto out;
1482
1483 sock->ops = &packet_ops;
1484 if (sock->type == SOCK_PACKET)
1485 sock->ops = &packet_ops_spkt;
1486
1487 sock_init_data(sock, sk);
1488
1489 po = pkt_sk(sk);
1490 sk->sk_family = PF_PACKET;
1491 po->num = proto;
1492
1493 sk->sk_destruct = packet_sock_destruct;
1494 sk_refcnt_debug_inc(sk);
1495
1496 /*
1497 * Attach a protocol block
1498 */
1499
1500 spin_lock_init(&po->bind_lock);
1501 mutex_init(&po->pg_vec_lock);
1502 po->prot_hook.func = packet_rcv;
1503
1504 if (sock->type == SOCK_PACKET)
1505 po->prot_hook.func = packet_rcv_spkt;
1506
1507 po->prot_hook.af_packet_priv = sk;
1508
1509 if (proto) {
1510 po->prot_hook.type = proto;
1511 dev_add_pack(&po->prot_hook);
1512 sock_hold(sk);
1513 po->running = 1;
1514 }
1515
1516 spin_lock_bh(&net->packet.sklist_lock);
1517 sk_add_node_rcu(sk, &net->packet.sklist);
1518 sock_prot_inuse_add(net, &packet_proto, 1);
1519 spin_unlock_bh(&net->packet.sklist_lock);
1520
1521 return 0;
1522 out:
1523 return err;
1524 }
1525
1526 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1527 {
1528 struct sock_exterr_skb *serr;
1529 struct sk_buff *skb, *skb2;
1530 int copied, err;
1531
1532 err = -EAGAIN;
1533 skb = skb_dequeue(&sk->sk_error_queue);
1534 if (skb == NULL)
1535 goto out;
1536
1537 copied = skb->len;
1538 if (copied > len) {
1539 msg->msg_flags |= MSG_TRUNC;
1540 copied = len;
1541 }
1542 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1543 if (err)
1544 goto out_free_skb;
1545
1546 sock_recv_timestamp(msg, sk, skb);
1547
1548 serr = SKB_EXT_ERR(skb);
1549 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
1550 sizeof(serr->ee), &serr->ee);
1551
1552 msg->msg_flags |= MSG_ERRQUEUE;
1553 err = copied;
1554
1555 /* Reset and regenerate socket error */
1556 spin_lock_bh(&sk->sk_error_queue.lock);
1557 sk->sk_err = 0;
1558 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
1559 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
1560 spin_unlock_bh(&sk->sk_error_queue.lock);
1561 sk->sk_error_report(sk);
1562 } else
1563 spin_unlock_bh(&sk->sk_error_queue.lock);
1564
1565 out_free_skb:
1566 kfree_skb(skb);
1567 out:
1568 return err;
1569 }
1570
1571 /*
1572 * Pull a packet from our receive queue and hand it to the user.
1573 * If necessary we block.
1574 */
1575
1576 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1577 struct msghdr *msg, size_t len, int flags)
1578 {
1579 struct sock *sk = sock->sk;
1580 struct sk_buff *skb;
1581 int copied, err;
1582 struct sockaddr_ll *sll;
1583 int vnet_hdr_len = 0;
1584
1585 err = -EINVAL;
1586 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1587 goto out;
1588
1589 #if 0
1590 /* What error should we return now? EUNATTACH? */
1591 if (pkt_sk(sk)->ifindex < 0)
1592 return -ENODEV;
1593 #endif
1594
1595 if (flags & MSG_ERRQUEUE) {
1596 err = packet_recv_error(sk, msg, len);
1597 goto out;
1598 }
1599
1600 /*
1601 * Call the generic datagram receiver. This handles all sorts
1602 * of horrible races and re-entrancy so we can forget about it
1603 * in the protocol layers.
1604 *
1605 * Now it will return ENETDOWN, if device have just gone down,
1606 * but then it will block.
1607 */
1608
1609 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1610
1611 /*
1612 * An error occurred so return it. Because skb_recv_datagram()
1613 * handles the blocking we don't see and worry about blocking
1614 * retries.
1615 */
1616
1617 if (skb == NULL)
1618 goto out;
1619
1620 if (pkt_sk(sk)->has_vnet_hdr) {
1621 struct virtio_net_hdr vnet_hdr = { 0 };
1622
1623 err = -EINVAL;
1624 vnet_hdr_len = sizeof(vnet_hdr);
1625 if (len < vnet_hdr_len)
1626 goto out_free;
1627
1628 len -= vnet_hdr_len;
1629
1630 if (skb_is_gso(skb)) {
1631 struct skb_shared_info *sinfo = skb_shinfo(skb);
1632
1633 /* This is a hint as to how much should be linear. */
1634 vnet_hdr.hdr_len = skb_headlen(skb);
1635 vnet_hdr.gso_size = sinfo->gso_size;
1636 if (sinfo->gso_type & SKB_GSO_TCPV4)
1637 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1638 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1639 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1640 else if (sinfo->gso_type & SKB_GSO_UDP)
1641 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1642 else if (sinfo->gso_type & SKB_GSO_FCOE)
1643 goto out_free;
1644 else
1645 BUG();
1646 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1647 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1648 } else
1649 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1650
1651 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1652 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1653 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
1654 vnet_hdr.csum_offset = skb->csum_offset;
1655 } /* else everything is zero */
1656
1657 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1658 vnet_hdr_len);
1659 if (err < 0)
1660 goto out_free;
1661 }
1662
1663 /*
1664 * If the address length field is there to be filled in, we fill
1665 * it in now.
1666 */
1667
1668 sll = &PACKET_SKB_CB(skb)->sa.ll;
1669 if (sock->type == SOCK_PACKET)
1670 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1671 else
1672 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1673
1674 /*
1675 * You lose any data beyond the buffer you gave. If it worries a
1676 * user program they can ask the device for its MTU anyway.
1677 */
1678
1679 copied = skb->len;
1680 if (copied > len) {
1681 copied = len;
1682 msg->msg_flags |= MSG_TRUNC;
1683 }
1684
1685 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1686 if (err)
1687 goto out_free;
1688
1689 sock_recv_ts_and_drops(msg, sk, skb);
1690
1691 if (msg->msg_name)
1692 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1693 msg->msg_namelen);
1694
1695 if (pkt_sk(sk)->auxdata) {
1696 struct tpacket_auxdata aux;
1697
1698 aux.tp_status = TP_STATUS_USER;
1699 if (skb->ip_summed == CHECKSUM_PARTIAL)
1700 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1701 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1702 aux.tp_snaplen = skb->len;
1703 aux.tp_mac = 0;
1704 aux.tp_net = skb_network_offset(skb);
1705 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
1706
1707 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1708 }
1709
1710 /*
1711 * Free or return the buffer as appropriate. Again this
1712 * hides all the races and re-entrancy issues from us.
1713 */
1714 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1715
1716 out_free:
1717 skb_free_datagram(sk, skb);
1718 out:
1719 return err;
1720 }
1721
1722 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1723 int *uaddr_len, int peer)
1724 {
1725 struct net_device *dev;
1726 struct sock *sk = sock->sk;
1727
1728 if (peer)
1729 return -EOPNOTSUPP;
1730
1731 uaddr->sa_family = AF_PACKET;
1732 rcu_read_lock();
1733 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
1734 if (dev)
1735 strncpy(uaddr->sa_data, dev->name, 14);
1736 else
1737 memset(uaddr->sa_data, 0, 14);
1738 rcu_read_unlock();
1739 *uaddr_len = sizeof(*uaddr);
1740
1741 return 0;
1742 }
1743
1744 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1745 int *uaddr_len, int peer)
1746 {
1747 struct net_device *dev;
1748 struct sock *sk = sock->sk;
1749 struct packet_sock *po = pkt_sk(sk);
1750 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1751
1752 if (peer)
1753 return -EOPNOTSUPP;
1754
1755 sll->sll_family = AF_PACKET;
1756 sll->sll_ifindex = po->ifindex;
1757 sll->sll_protocol = po->num;
1758 sll->sll_pkttype = 0;
1759 rcu_read_lock();
1760 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1761 if (dev) {
1762 sll->sll_hatype = dev->type;
1763 sll->sll_halen = dev->addr_len;
1764 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1765 } else {
1766 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1767 sll->sll_halen = 0;
1768 }
1769 rcu_read_unlock();
1770 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1771
1772 return 0;
1773 }
1774
1775 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1776 int what)
1777 {
1778 switch (i->type) {
1779 case PACKET_MR_MULTICAST:
1780 if (i->alen != dev->addr_len)
1781 return -EINVAL;
1782 if (what > 0)
1783 return dev_mc_add(dev, i->addr);
1784 else
1785 return dev_mc_del(dev, i->addr);
1786 break;
1787 case PACKET_MR_PROMISC:
1788 return dev_set_promiscuity(dev, what);
1789 break;
1790 case PACKET_MR_ALLMULTI:
1791 return dev_set_allmulti(dev, what);
1792 break;
1793 case PACKET_MR_UNICAST:
1794 if (i->alen != dev->addr_len)
1795 return -EINVAL;
1796 if (what > 0)
1797 return dev_uc_add(dev, i->addr);
1798 else
1799 return dev_uc_del(dev, i->addr);
1800 break;
1801 default:
1802 break;
1803 }
1804 return 0;
1805 }
1806
1807 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1808 {
1809 for ( ; i; i = i->next) {
1810 if (i->ifindex == dev->ifindex)
1811 packet_dev_mc(dev, i, what);
1812 }
1813 }
1814
1815 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1816 {
1817 struct packet_sock *po = pkt_sk(sk);
1818 struct packet_mclist *ml, *i;
1819 struct net_device *dev;
1820 int err;
1821
1822 rtnl_lock();
1823
1824 err = -ENODEV;
1825 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1826 if (!dev)
1827 goto done;
1828
1829 err = -EINVAL;
1830 if (mreq->mr_alen > dev->addr_len)
1831 goto done;
1832
1833 err = -ENOBUFS;
1834 i = kmalloc(sizeof(*i), GFP_KERNEL);
1835 if (i == NULL)
1836 goto done;
1837
1838 err = 0;
1839 for (ml = po->mclist; ml; ml = ml->next) {
1840 if (ml->ifindex == mreq->mr_ifindex &&
1841 ml->type == mreq->mr_type &&
1842 ml->alen == mreq->mr_alen &&
1843 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1844 ml->count++;
1845 /* Free the new element ... */
1846 kfree(i);
1847 goto done;
1848 }
1849 }
1850
1851 i->type = mreq->mr_type;
1852 i->ifindex = mreq->mr_ifindex;
1853 i->alen = mreq->mr_alen;
1854 memcpy(i->addr, mreq->mr_address, i->alen);
1855 i->count = 1;
1856 i->next = po->mclist;
1857 po->mclist = i;
1858 err = packet_dev_mc(dev, i, 1);
1859 if (err) {
1860 po->mclist = i->next;
1861 kfree(i);
1862 }
1863
1864 done:
1865 rtnl_unlock();
1866 return err;
1867 }
1868
1869 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1870 {
1871 struct packet_mclist *ml, **mlp;
1872
1873 rtnl_lock();
1874
1875 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1876 if (ml->ifindex == mreq->mr_ifindex &&
1877 ml->type == mreq->mr_type &&
1878 ml->alen == mreq->mr_alen &&
1879 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1880 if (--ml->count == 0) {
1881 struct net_device *dev;
1882 *mlp = ml->next;
1883 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1884 if (dev)
1885 packet_dev_mc(dev, ml, -1);
1886 kfree(ml);
1887 }
1888 rtnl_unlock();
1889 return 0;
1890 }
1891 }
1892 rtnl_unlock();
1893 return -EADDRNOTAVAIL;
1894 }
1895
1896 static void packet_flush_mclist(struct sock *sk)
1897 {
1898 struct packet_sock *po = pkt_sk(sk);
1899 struct packet_mclist *ml;
1900
1901 if (!po->mclist)
1902 return;
1903
1904 rtnl_lock();
1905 while ((ml = po->mclist) != NULL) {
1906 struct net_device *dev;
1907
1908 po->mclist = ml->next;
1909 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1910 if (dev != NULL)
1911 packet_dev_mc(dev, ml, -1);
1912 kfree(ml);
1913 }
1914 rtnl_unlock();
1915 }
1916
1917 static int
1918 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1919 {
1920 struct sock *sk = sock->sk;
1921 struct packet_sock *po = pkt_sk(sk);
1922 int ret;
1923
1924 if (level != SOL_PACKET)
1925 return -ENOPROTOOPT;
1926
1927 switch (optname) {
1928 case PACKET_ADD_MEMBERSHIP:
1929 case PACKET_DROP_MEMBERSHIP:
1930 {
1931 struct packet_mreq_max mreq;
1932 int len = optlen;
1933 memset(&mreq, 0, sizeof(mreq));
1934 if (len < sizeof(struct packet_mreq))
1935 return -EINVAL;
1936 if (len > sizeof(mreq))
1937 len = sizeof(mreq);
1938 if (copy_from_user(&mreq, optval, len))
1939 return -EFAULT;
1940 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1941 return -EINVAL;
1942 if (optname == PACKET_ADD_MEMBERSHIP)
1943 ret = packet_mc_add(sk, &mreq);
1944 else
1945 ret = packet_mc_drop(sk, &mreq);
1946 return ret;
1947 }
1948
1949 case PACKET_RX_RING:
1950 case PACKET_TX_RING:
1951 {
1952 struct tpacket_req req;
1953
1954 if (optlen < sizeof(req))
1955 return -EINVAL;
1956 if (pkt_sk(sk)->has_vnet_hdr)
1957 return -EINVAL;
1958 if (copy_from_user(&req, optval, sizeof(req)))
1959 return -EFAULT;
1960 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1961 }
1962 case PACKET_COPY_THRESH:
1963 {
1964 int val;
1965
1966 if (optlen != sizeof(val))
1967 return -EINVAL;
1968 if (copy_from_user(&val, optval, sizeof(val)))
1969 return -EFAULT;
1970
1971 pkt_sk(sk)->copy_thresh = val;
1972 return 0;
1973 }
1974 case PACKET_VERSION:
1975 {
1976 int val;
1977
1978 if (optlen != sizeof(val))
1979 return -EINVAL;
1980 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1981 return -EBUSY;
1982 if (copy_from_user(&val, optval, sizeof(val)))
1983 return -EFAULT;
1984 switch (val) {
1985 case TPACKET_V1:
1986 case TPACKET_V2:
1987 po->tp_version = val;
1988 return 0;
1989 default:
1990 return -EINVAL;
1991 }
1992 }
1993 case PACKET_RESERVE:
1994 {
1995 unsigned int val;
1996
1997 if (optlen != sizeof(val))
1998 return -EINVAL;
1999 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2000 return -EBUSY;
2001 if (copy_from_user(&val, optval, sizeof(val)))
2002 return -EFAULT;
2003 po->tp_reserve = val;
2004 return 0;
2005 }
2006 case PACKET_LOSS:
2007 {
2008 unsigned int val;
2009
2010 if (optlen != sizeof(val))
2011 return -EINVAL;
2012 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2013 return -EBUSY;
2014 if (copy_from_user(&val, optval, sizeof(val)))
2015 return -EFAULT;
2016 po->tp_loss = !!val;
2017 return 0;
2018 }
2019 case PACKET_AUXDATA:
2020 {
2021 int val;
2022
2023 if (optlen < sizeof(val))
2024 return -EINVAL;
2025 if (copy_from_user(&val, optval, sizeof(val)))
2026 return -EFAULT;
2027
2028 po->auxdata = !!val;
2029 return 0;
2030 }
2031 case PACKET_ORIGDEV:
2032 {
2033 int val;
2034
2035 if (optlen < sizeof(val))
2036 return -EINVAL;
2037 if (copy_from_user(&val, optval, sizeof(val)))
2038 return -EFAULT;
2039
2040 po->origdev = !!val;
2041 return 0;
2042 }
2043 case PACKET_VNET_HDR:
2044 {
2045 int val;
2046
2047 if (sock->type != SOCK_RAW)
2048 return -EINVAL;
2049 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2050 return -EBUSY;
2051 if (optlen < sizeof(val))
2052 return -EINVAL;
2053 if (copy_from_user(&val, optval, sizeof(val)))
2054 return -EFAULT;
2055
2056 po->has_vnet_hdr = !!val;
2057 return 0;
2058 }
2059 case PACKET_TIMESTAMP:
2060 {
2061 int val;
2062
2063 if (optlen != sizeof(val))
2064 return -EINVAL;
2065 if (copy_from_user(&val, optval, sizeof(val)))
2066 return -EFAULT;
2067
2068 po->tp_tstamp = val;
2069 return 0;
2070 }
2071 default:
2072 return -ENOPROTOOPT;
2073 }
2074 }
2075
2076 static int packet_getsockopt(struct socket *sock, int level, int optname,
2077 char __user *optval, int __user *optlen)
2078 {
2079 int len;
2080 int val;
2081 struct sock *sk = sock->sk;
2082 struct packet_sock *po = pkt_sk(sk);
2083 void *data;
2084 struct tpacket_stats st;
2085
2086 if (level != SOL_PACKET)
2087 return -ENOPROTOOPT;
2088
2089 if (get_user(len, optlen))
2090 return -EFAULT;
2091
2092 if (len < 0)
2093 return -EINVAL;
2094
2095 switch (optname) {
2096 case PACKET_STATISTICS:
2097 if (len > sizeof(struct tpacket_stats))
2098 len = sizeof(struct tpacket_stats);
2099 spin_lock_bh(&sk->sk_receive_queue.lock);
2100 st = po->stats;
2101 memset(&po->stats, 0, sizeof(st));
2102 spin_unlock_bh(&sk->sk_receive_queue.lock);
2103 st.tp_packets += st.tp_drops;
2104
2105 data = &st;
2106 break;
2107 case PACKET_AUXDATA:
2108 if (len > sizeof(int))
2109 len = sizeof(int);
2110 val = po->auxdata;
2111
2112 data = &val;
2113 break;
2114 case PACKET_ORIGDEV:
2115 if (len > sizeof(int))
2116 len = sizeof(int);
2117 val = po->origdev;
2118
2119 data = &val;
2120 break;
2121 case PACKET_VNET_HDR:
2122 if (len > sizeof(int))
2123 len = sizeof(int);
2124 val = po->has_vnet_hdr;
2125
2126 data = &val;
2127 break;
2128 case PACKET_VERSION:
2129 if (len > sizeof(int))
2130 len = sizeof(int);
2131 val = po->tp_version;
2132 data = &val;
2133 break;
2134 case PACKET_HDRLEN:
2135 if (len > sizeof(int))
2136 len = sizeof(int);
2137 if (copy_from_user(&val, optval, len))
2138 return -EFAULT;
2139 switch (val) {
2140 case TPACKET_V1:
2141 val = sizeof(struct tpacket_hdr);
2142 break;
2143 case TPACKET_V2:
2144 val = sizeof(struct tpacket2_hdr);
2145 break;
2146 default:
2147 return -EINVAL;
2148 }
2149 data = &val;
2150 break;
2151 case PACKET_RESERVE:
2152 if (len > sizeof(unsigned int))
2153 len = sizeof(unsigned int);
2154 val = po->tp_reserve;
2155 data = &val;
2156 break;
2157 case PACKET_LOSS:
2158 if (len > sizeof(unsigned int))
2159 len = sizeof(unsigned int);
2160 val = po->tp_loss;
2161 data = &val;
2162 break;
2163 case PACKET_TIMESTAMP:
2164 if (len > sizeof(int))
2165 len = sizeof(int);
2166 val = po->tp_tstamp;
2167 data = &val;
2168 break;
2169 default:
2170 return -ENOPROTOOPT;
2171 }
2172
2173 if (put_user(len, optlen))
2174 return -EFAULT;
2175 if (copy_to_user(optval, data, len))
2176 return -EFAULT;
2177 return 0;
2178 }
2179
2180
2181 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2182 {
2183 struct sock *sk;
2184 struct hlist_node *node;
2185 struct net_device *dev = data;
2186 struct net *net = dev_net(dev);
2187
2188 rcu_read_lock();
2189 sk_for_each_rcu(sk, node, &net->packet.sklist) {
2190 struct packet_sock *po = pkt_sk(sk);
2191
2192 switch (msg) {
2193 case NETDEV_UNREGISTER:
2194 if (po->mclist)
2195 packet_dev_mclist(dev, po->mclist, -1);
2196 /* fallthrough */
2197
2198 case NETDEV_DOWN:
2199 if (dev->ifindex == po->ifindex) {
2200 spin_lock(&po->bind_lock);
2201 if (po->running) {
2202 __dev_remove_pack(&po->prot_hook);
2203 __sock_put(sk);
2204 po->running = 0;
2205 sk->sk_err = ENETDOWN;
2206 if (!sock_flag(sk, SOCK_DEAD))
2207 sk->sk_error_report(sk);
2208 }
2209 if (msg == NETDEV_UNREGISTER) {
2210 po->ifindex = -1;
2211 po->prot_hook.dev = NULL;
2212 }
2213 spin_unlock(&po->bind_lock);
2214 }
2215 break;
2216 case NETDEV_UP:
2217 if (dev->ifindex == po->ifindex) {
2218 spin_lock(&po->bind_lock);
2219 if (po->num && !po->running) {
2220 dev_add_pack(&po->prot_hook);
2221 sock_hold(sk);
2222 po->running = 1;
2223 }
2224 spin_unlock(&po->bind_lock);
2225 }
2226 break;
2227 }
2228 }
2229 rcu_read_unlock();
2230 return NOTIFY_DONE;
2231 }
2232
2233
2234 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2235 unsigned long arg)
2236 {
2237 struct sock *sk = sock->sk;
2238
2239 switch (cmd) {
2240 case SIOCOUTQ:
2241 {
2242 int amount = sk_wmem_alloc_get(sk);
2243
2244 return put_user(amount, (int __user *)arg);
2245 }
2246 case SIOCINQ:
2247 {
2248 struct sk_buff *skb;
2249 int amount = 0;
2250
2251 spin_lock_bh(&sk->sk_receive_queue.lock);
2252 skb = skb_peek(&sk->sk_receive_queue);
2253 if (skb)
2254 amount = skb->len;
2255 spin_unlock_bh(&sk->sk_receive_queue.lock);
2256 return put_user(amount, (int __user *)arg);
2257 }
2258 case SIOCGSTAMP:
2259 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2260 case SIOCGSTAMPNS:
2261 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2262
2263 #ifdef CONFIG_INET
2264 case SIOCADDRT:
2265 case SIOCDELRT:
2266 case SIOCDARP:
2267 case SIOCGARP:
2268 case SIOCSARP:
2269 case SIOCGIFADDR:
2270 case SIOCSIFADDR:
2271 case SIOCGIFBRDADDR:
2272 case SIOCSIFBRDADDR:
2273 case SIOCGIFNETMASK:
2274 case SIOCSIFNETMASK:
2275 case SIOCGIFDSTADDR:
2276 case SIOCSIFDSTADDR:
2277 case SIOCSIFFLAGS:
2278 return inet_dgram_ops.ioctl(sock, cmd, arg);
2279 #endif
2280
2281 default:
2282 return -ENOIOCTLCMD;
2283 }
2284 return 0;
2285 }
2286
2287 static unsigned int packet_poll(struct file *file, struct socket *sock,
2288 poll_table *wait)
2289 {
2290 struct sock *sk = sock->sk;
2291 struct packet_sock *po = pkt_sk(sk);
2292 unsigned int mask = datagram_poll(file, sock, wait);
2293
2294 spin_lock_bh(&sk->sk_receive_queue.lock);
2295 if (po->rx_ring.pg_vec) {
2296 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2297 mask |= POLLIN | POLLRDNORM;
2298 }
2299 spin_unlock_bh(&sk->sk_receive_queue.lock);
2300 spin_lock_bh(&sk->sk_write_queue.lock);
2301 if (po->tx_ring.pg_vec) {
2302 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2303 mask |= POLLOUT | POLLWRNORM;
2304 }
2305 spin_unlock_bh(&sk->sk_write_queue.lock);
2306 return mask;
2307 }
2308
2309
2310 /* Dirty? Well, I still did not learn better way to account
2311 * for user mmaps.
2312 */
2313
2314 static void packet_mm_open(struct vm_area_struct *vma)
2315 {
2316 struct file *file = vma->vm_file;
2317 struct socket *sock = file->private_data;
2318 struct sock *sk = sock->sk;
2319
2320 if (sk)
2321 atomic_inc(&pkt_sk(sk)->mapped);
2322 }
2323
2324 static void packet_mm_close(struct vm_area_struct *vma)
2325 {
2326 struct file *file = vma->vm_file;
2327 struct socket *sock = file->private_data;
2328 struct sock *sk = sock->sk;
2329
2330 if (sk)
2331 atomic_dec(&pkt_sk(sk)->mapped);
2332 }
2333
2334 static const struct vm_operations_struct packet_mmap_ops = {
2335 .open = packet_mm_open,
2336 .close = packet_mm_close,
2337 };
2338
2339 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
2340 unsigned int len)
2341 {
2342 int i;
2343
2344 for (i = 0; i < len; i++) {
2345 if (likely(pg_vec[i].buffer)) {
2346 if (is_vmalloc_addr(pg_vec[i].buffer))
2347 vfree(pg_vec[i].buffer);
2348 else
2349 free_pages((unsigned long)pg_vec[i].buffer,
2350 order);
2351 pg_vec[i].buffer = NULL;
2352 }
2353 }
2354 kfree(pg_vec);
2355 }
2356
2357 static inline char *alloc_one_pg_vec_page(unsigned long order)
2358 {
2359 char *buffer = NULL;
2360 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
2361 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
2362
2363 buffer = (char *) __get_free_pages(gfp_flags, order);
2364
2365 if (buffer)
2366 return buffer;
2367
2368 /*
2369 * __get_free_pages failed, fall back to vmalloc
2370 */
2371 buffer = vzalloc((1 << order) * PAGE_SIZE);
2372
2373 if (buffer)
2374 return buffer;
2375
2376 /*
2377 * vmalloc failed, lets dig into swap here
2378 */
2379 gfp_flags &= ~__GFP_NORETRY;
2380 buffer = (char *)__get_free_pages(gfp_flags, order);
2381 if (buffer)
2382 return buffer;
2383
2384 /*
2385 * complete and utter failure
2386 */
2387 return NULL;
2388 }
2389
2390 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
2391 {
2392 unsigned int block_nr = req->tp_block_nr;
2393 struct pgv *pg_vec;
2394 int i;
2395
2396 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
2397 if (unlikely(!pg_vec))
2398 goto out;
2399
2400 for (i = 0; i < block_nr; i++) {
2401 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
2402 if (unlikely(!pg_vec[i].buffer))
2403 goto out_free_pgvec;
2404 }
2405
2406 out:
2407 return pg_vec;
2408
2409 out_free_pgvec:
2410 free_pg_vec(pg_vec, order, block_nr);
2411 pg_vec = NULL;
2412 goto out;
2413 }
2414
2415 static 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 spin_lock_bh(&rb_queue->lock);
2504 swap(rb->pg_vec, pg_vec);
2505 rb->frame_max = (req->tp_frame_nr - 1);
2506 rb->head = 0;
2507 rb->frame_size = req->tp_frame_size;
2508 spin_unlock_bh(&rb_queue->lock);
2509
2510 swap(rb->pg_vec_order, order);
2511 swap(rb->pg_vec_len, req->tp_block_nr);
2512
2513 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2514 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2515 tpacket_rcv : packet_rcv;
2516 skb_queue_purge(rb_queue);
2517 if (atomic_read(&po->mapped))
2518 pr_err("packet_mmap: vma is busy: %d\n",
2519 atomic_read(&po->mapped));
2520 }
2521 mutex_unlock(&po->pg_vec_lock);
2522
2523 spin_lock(&po->bind_lock);
2524 if (was_running && !po->running) {
2525 sock_hold(sk);
2526 po->running = 1;
2527 po->num = num;
2528 dev_add_pack(&po->prot_hook);
2529 }
2530 spin_unlock(&po->bind_lock);
2531
2532 release_sock(sk);
2533
2534 if (pg_vec)
2535 free_pg_vec(pg_vec, order, req->tp_block_nr);
2536 out:
2537 return err;
2538 }
2539
2540 static int packet_mmap(struct file *file, struct socket *sock,
2541 struct vm_area_struct *vma)
2542 {
2543 struct sock *sk = sock->sk;
2544 struct packet_sock *po = pkt_sk(sk);
2545 unsigned long size, expected_size;
2546 struct packet_ring_buffer *rb;
2547 unsigned long start;
2548 int err = -EINVAL;
2549 int i;
2550
2551 if (vma->vm_pgoff)
2552 return -EINVAL;
2553
2554 mutex_lock(&po->pg_vec_lock);
2555
2556 expected_size = 0;
2557 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2558 if (rb->pg_vec) {
2559 expected_size += rb->pg_vec_len
2560 * rb->pg_vec_pages
2561 * PAGE_SIZE;
2562 }
2563 }
2564
2565 if (expected_size == 0)
2566 goto out;
2567
2568 size = vma->vm_end - vma->vm_start;
2569 if (size != expected_size)
2570 goto out;
2571
2572 start = vma->vm_start;
2573 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2574 if (rb->pg_vec == NULL)
2575 continue;
2576
2577 for (i = 0; i < rb->pg_vec_len; i++) {
2578 struct page *page;
2579 void *kaddr = rb->pg_vec[i].buffer;
2580 int pg_num;
2581
2582 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
2583 page = pgv_to_page(kaddr);
2584 err = vm_insert_page(vma, start, page);
2585 if (unlikely(err))
2586 goto out;
2587 start += PAGE_SIZE;
2588 kaddr += PAGE_SIZE;
2589 }
2590 }
2591 }
2592
2593 atomic_inc(&po->mapped);
2594 vma->vm_ops = &packet_mmap_ops;
2595 err = 0;
2596
2597 out:
2598 mutex_unlock(&po->pg_vec_lock);
2599 return err;
2600 }
2601
2602 static const struct proto_ops packet_ops_spkt = {
2603 .family = PF_PACKET,
2604 .owner = THIS_MODULE,
2605 .release = packet_release,
2606 .bind = packet_bind_spkt,
2607 .connect = sock_no_connect,
2608 .socketpair = sock_no_socketpair,
2609 .accept = sock_no_accept,
2610 .getname = packet_getname_spkt,
2611 .poll = datagram_poll,
2612 .ioctl = packet_ioctl,
2613 .listen = sock_no_listen,
2614 .shutdown = sock_no_shutdown,
2615 .setsockopt = sock_no_setsockopt,
2616 .getsockopt = sock_no_getsockopt,
2617 .sendmsg = packet_sendmsg_spkt,
2618 .recvmsg = packet_recvmsg,
2619 .mmap = sock_no_mmap,
2620 .sendpage = sock_no_sendpage,
2621 };
2622
2623 static const struct proto_ops packet_ops = {
2624 .family = PF_PACKET,
2625 .owner = THIS_MODULE,
2626 .release = packet_release,
2627 .bind = packet_bind,
2628 .connect = sock_no_connect,
2629 .socketpair = sock_no_socketpair,
2630 .accept = sock_no_accept,
2631 .getname = packet_getname,
2632 .poll = packet_poll,
2633 .ioctl = packet_ioctl,
2634 .listen = sock_no_listen,
2635 .shutdown = sock_no_shutdown,
2636 .setsockopt = packet_setsockopt,
2637 .getsockopt = packet_getsockopt,
2638 .sendmsg = packet_sendmsg,
2639 .recvmsg = packet_recvmsg,
2640 .mmap = packet_mmap,
2641 .sendpage = sock_no_sendpage,
2642 };
2643
2644 static const struct net_proto_family packet_family_ops = {
2645 .family = PF_PACKET,
2646 .create = packet_create,
2647 .owner = THIS_MODULE,
2648 };
2649
2650 static struct notifier_block packet_netdev_notifier = {
2651 .notifier_call = packet_notifier,
2652 };
2653
2654 #ifdef CONFIG_PROC_FS
2655
2656 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2657 __acquires(RCU)
2658 {
2659 struct net *net = seq_file_net(seq);
2660
2661 rcu_read_lock();
2662 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
2663 }
2664
2665 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2666 {
2667 struct net *net = seq_file_net(seq);
2668 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
2669 }
2670
2671 static void packet_seq_stop(struct seq_file *seq, void *v)
2672 __releases(RCU)
2673 {
2674 rcu_read_unlock();
2675 }
2676
2677 static int packet_seq_show(struct seq_file *seq, void *v)
2678 {
2679 if (v == SEQ_START_TOKEN)
2680 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
2681 else {
2682 struct sock *s = sk_entry(v);
2683 const struct packet_sock *po = pkt_sk(s);
2684
2685 seq_printf(seq,
2686 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
2687 s,
2688 atomic_read(&s->sk_refcnt),
2689 s->sk_type,
2690 ntohs(po->num),
2691 po->ifindex,
2692 po->running,
2693 atomic_read(&s->sk_rmem_alloc),
2694 sock_i_uid(s),
2695 sock_i_ino(s));
2696 }
2697
2698 return 0;
2699 }
2700
2701 static const struct seq_operations packet_seq_ops = {
2702 .start = packet_seq_start,
2703 .next = packet_seq_next,
2704 .stop = packet_seq_stop,
2705 .show = packet_seq_show,
2706 };
2707
2708 static int packet_seq_open(struct inode *inode, struct file *file)
2709 {
2710 return seq_open_net(inode, file, &packet_seq_ops,
2711 sizeof(struct seq_net_private));
2712 }
2713
2714 static const struct file_operations packet_seq_fops = {
2715 .owner = THIS_MODULE,
2716 .open = packet_seq_open,
2717 .read = seq_read,
2718 .llseek = seq_lseek,
2719 .release = seq_release_net,
2720 };
2721
2722 #endif
2723
2724 static int __net_init packet_net_init(struct net *net)
2725 {
2726 spin_lock_init(&net->packet.sklist_lock);
2727 INIT_HLIST_HEAD(&net->packet.sklist);
2728
2729 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2730 return -ENOMEM;
2731
2732 return 0;
2733 }
2734
2735 static void __net_exit packet_net_exit(struct net *net)
2736 {
2737 proc_net_remove(net, "packet");
2738 }
2739
2740 static struct pernet_operations packet_net_ops = {
2741 .init = packet_net_init,
2742 .exit = packet_net_exit,
2743 };
2744
2745
2746 static void __exit packet_exit(void)
2747 {
2748 unregister_netdevice_notifier(&packet_netdev_notifier);
2749 unregister_pernet_subsys(&packet_net_ops);
2750 sock_unregister(PF_PACKET);
2751 proto_unregister(&packet_proto);
2752 }
2753
2754 static int __init packet_init(void)
2755 {
2756 int rc = proto_register(&packet_proto, 0);
2757
2758 if (rc != 0)
2759 goto out;
2760
2761 sock_register(&packet_family_ops);
2762 register_pernet_subsys(&packet_net_ops);
2763 register_netdevice_notifier(&packet_netdev_notifier);
2764 out:
2765 return rc;
2766 }
2767
2768 module_init(packet_init);
2769 module_exit(packet_exit);
2770 MODULE_LICENSE("GPL");
2771 MODULE_ALIAS_NETPROTO(PF_PACKET);
The kernel tree for Tao Ma.