?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
af_packet: prevent information leak
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / packet / af_packet.c
blob04098c2801648593b97954d22c69209216240995
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 struct pgv {
168 char *buffer;
169 };
170
171 struct packet_ring_buffer {
172 struct pgv *pg_vec;
173 unsigned int head;
174 unsigned int frames_per_block;
175 unsigned int frame_size;
176 unsigned int frame_max;
177
178 unsigned int pg_vec_order;
179 unsigned int pg_vec_pages;
180 unsigned int pg_vec_len;
181
182 atomic_t pending;
183 };
184
185 struct packet_sock;
186 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
187
188 static void packet_flush_mclist(struct sock *sk);
189
190 struct packet_sock {
191 /* struct sock has to be the first member of packet_sock */
192 struct sock sk;
193 struct tpacket_stats stats;
194 struct packet_ring_buffer rx_ring;
195 struct packet_ring_buffer tx_ring;
196 int copy_thresh;
197 spinlock_t bind_lock;
198 struct mutex pg_vec_lock;
199 unsigned int running:1, /* prot_hook is attached*/
200 auxdata:1,
201 origdev:1,
202 has_vnet_hdr:1;
203 int ifindex; /* bound device */
204 __be16 num;
205 struct packet_mclist *mclist;
206 atomic_t mapped;
207 enum tpacket_versions tp_version;
208 unsigned int tp_hdrlen;
209 unsigned int tp_reserve;
210 unsigned int tp_loss:1;
211 unsigned int tp_tstamp;
212 struct packet_type prot_hook ____cacheline_aligned_in_smp;
213 };
214
215 struct packet_skb_cb {
216 unsigned int origlen;
217 union {
218 struct sockaddr_pkt pkt;
219 struct sockaddr_ll ll;
220 } sa;
221 };
222
223 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
224
225 static inline __pure struct page *pgv_to_page(void *addr)
226 {
227 if (is_vmalloc_addr(addr))
228 return vmalloc_to_page(addr);
229 return virt_to_page(addr);
230 }
231
232 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
233 {
234 union {
235 struct tpacket_hdr *h1;
236 struct tpacket2_hdr *h2;
237 void *raw;
238 } h;
239
240 h.raw = frame;
241 switch (po->tp_version) {
242 case TPACKET_V1:
243 h.h1->tp_status = status;
244 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
245 break;
246 case TPACKET_V2:
247 h.h2->tp_status = status;
248 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
249 break;
250 default:
251 pr_err("TPACKET version not supported\n");
252 BUG();
253 }
254
255 smp_wmb();
256 }
257
258 static int __packet_get_status(struct packet_sock *po, void *frame)
259 {
260 union {
261 struct tpacket_hdr *h1;
262 struct tpacket2_hdr *h2;
263 void *raw;
264 } h;
265
266 smp_rmb();
267
268 h.raw = frame;
269 switch (po->tp_version) {
270 case TPACKET_V1:
271 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
272 return h.h1->tp_status;
273 case TPACKET_V2:
274 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
275 return h.h2->tp_status;
276 default:
277 pr_err("TPACKET version not supported\n");
278 BUG();
279 return 0;
280 }
281 }
282
283 static void *packet_lookup_frame(struct packet_sock *po,
284 struct packet_ring_buffer *rb,
285 unsigned int position,
286 int status)
287 {
288 unsigned int pg_vec_pos, frame_offset;
289 union {
290 struct tpacket_hdr *h1;
291 struct tpacket2_hdr *h2;
292 void *raw;
293 } h;
294
295 pg_vec_pos = position / rb->frames_per_block;
296 frame_offset = position % rb->frames_per_block;
297
298 h.raw = rb->pg_vec[pg_vec_pos].buffer +
299 (frame_offset * rb->frame_size);
300
301 if (status != __packet_get_status(po, h.raw))
302 return NULL;
303
304 return h.raw;
305 }
306
307 static inline void *packet_current_frame(struct packet_sock *po,
308 struct packet_ring_buffer *rb,
309 int status)
310 {
311 return packet_lookup_frame(po, rb, rb->head, status);
312 }
313
314 static inline void *packet_previous_frame(struct packet_sock *po,
315 struct packet_ring_buffer *rb,
316 int status)
317 {
318 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
319 return packet_lookup_frame(po, rb, previous, status);
320 }
321
322 static inline void packet_increment_head(struct packet_ring_buffer *buff)
323 {
324 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
325 }
326
327 static inline struct packet_sock *pkt_sk(struct sock *sk)
328 {
329 return (struct packet_sock *)sk;
330 }
331
332 static void packet_sock_destruct(struct sock *sk)
333 {
334 skb_queue_purge(&sk->sk_error_queue);
335
336 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
337 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
338
339 if (!sock_flag(sk, SOCK_DEAD)) {
340 pr_err("Attempt to release alive packet socket: %p\n", sk);
341 return;
342 }
343
344 sk_refcnt_debug_dec(sk);
345 }
346
347
348 static const struct proto_ops packet_ops;
349
350 static const struct proto_ops packet_ops_spkt;
351
352 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
353 struct packet_type *pt, struct net_device *orig_dev)
354 {
355 struct sock *sk;
356 struct sockaddr_pkt *spkt;
357
358 /*
359 * When we registered the protocol we saved the socket in the data
360 * field for just this event.
361 */
362
363 sk = pt->af_packet_priv;
364
365 /*
366 * Yank back the headers [hope the device set this
367 * right or kerboom...]
368 *
369 * Incoming packets have ll header pulled,
370 * push it back.
371 *
372 * For outgoing ones skb->data == skb_mac_header(skb)
373 * so that this procedure is noop.
374 */
375
376 if (skb->pkt_type == PACKET_LOOPBACK)
377 goto out;
378
379 if (!net_eq(dev_net(dev), sock_net(sk)))
380 goto out;
381
382 skb = skb_share_check(skb, GFP_ATOMIC);
383 if (skb == NULL)
384 goto oom;
385
386 /* drop any routing info */
387 skb_dst_drop(skb);
388
389 /* drop conntrack reference */
390 nf_reset(skb);
391
392 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
393
394 skb_push(skb, skb->data - skb_mac_header(skb));
395
396 /*
397 * The SOCK_PACKET socket receives _all_ frames.
398 */
399
400 spkt->spkt_family = dev->type;
401 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
402 spkt->spkt_protocol = skb->protocol;
403
404 /*
405 * Charge the memory to the socket. This is done specifically
406 * to prevent sockets using all the memory up.
407 */
408
409 if (sock_queue_rcv_skb(sk, skb) == 0)
410 return 0;
411
412 out:
413 kfree_skb(skb);
414 oom:
415 return 0;
416 }
417
418
419 /*
420 * Output a raw packet to a device layer. This bypasses all the other
421 * protocol layers and you must therefore supply it with a complete frame
422 */
423
424 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
425 struct msghdr *msg, size_t len)
426 {
427 struct sock *sk = sock->sk;
428 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
429 struct sk_buff *skb = NULL;
430 struct net_device *dev;
431 __be16 proto = 0;
432 int err;
433
434 /*
435 * Get and verify the address.
436 */
437
438 if (saddr) {
439 if (msg->msg_namelen < sizeof(struct sockaddr))
440 return -EINVAL;
441 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
442 proto = saddr->spkt_protocol;
443 } else
444 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
445
446 /*
447 * Find the device first to size check it
448 */
449
450 saddr->spkt_device[13] = 0;
451 retry:
452 rcu_read_lock();
453 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
454 err = -ENODEV;
455 if (dev == NULL)
456 goto out_unlock;
457
458 err = -ENETDOWN;
459 if (!(dev->flags & IFF_UP))
460 goto out_unlock;
461
462 /*
463 * You may not queue a frame bigger than the mtu. This is the lowest level
464 * raw protocol and you must do your own fragmentation at this level.
465 */
466
467 err = -EMSGSIZE;
468 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
469 goto out_unlock;
470
471 if (!skb) {
472 size_t reserved = LL_RESERVED_SPACE(dev);
473 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
474
475 rcu_read_unlock();
476 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
477 if (skb == NULL)
478 return -ENOBUFS;
479 /* FIXME: Save some space for broken drivers that write a hard
480 * header at transmission time by themselves. PPP is the notable
481 * one here. This should really be fixed at the driver level.
482 */
483 skb_reserve(skb, reserved);
484 skb_reset_network_header(skb);
485
486 /* Try to align data part correctly */
487 if (hhlen) {
488 skb->data -= hhlen;
489 skb->tail -= hhlen;
490 if (len < hhlen)
491 skb_reset_network_header(skb);
492 }
493 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
494 if (err)
495 goto out_free;
496 goto retry;
497 }
498
499 if (len > (dev->mtu + dev->hard_header_len)) {
500 /* Earlier code assumed this would be a VLAN pkt,
501 * double-check this now that we have the actual
502 * packet in hand.
503 */
504 struct ethhdr *ehdr;
505 skb_reset_mac_header(skb);
506 ehdr = eth_hdr(skb);
507 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
508 err = -EMSGSIZE;
509 goto out_unlock;
510 }
511 }
512
513 skb->protocol = proto;
514 skb->dev = dev;
515 skb->priority = sk->sk_priority;
516 skb->mark = sk->sk_mark;
517 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
518 if (err < 0)
519 goto out_unlock;
520
521 dev_queue_xmit(skb);
522 rcu_read_unlock();
523 return len;
524
525 out_unlock:
526 rcu_read_unlock();
527 out_free:
528 kfree_skb(skb);
529 return err;
530 }
531
532 static inline unsigned int run_filter(const struct sk_buff *skb,
533 const struct sock *sk,
534 unsigned int res)
535 {
536 struct sk_filter *filter;
537
538 rcu_read_lock();
539 filter = rcu_dereference(sk->sk_filter);
540 if (filter != NULL)
541 res = sk_run_filter(skb, filter->insns);
542 rcu_read_unlock();
543
544 return res;
545 }
546
547 /*
548 * This function makes lazy skb cloning in hope that most of packets
549 * are discarded by BPF.
550 *
551 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
552 * and skb->cb are mangled. It works because (and until) packets
553 * falling here are owned by current CPU. Output packets are cloned
554 * by dev_queue_xmit_nit(), input packets are processed by net_bh
555 * sequencially, so that if we return skb to original state on exit,
556 * we will not harm anyone.
557 */
558
559 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
560 struct packet_type *pt, struct net_device *orig_dev)
561 {
562 struct sock *sk;
563 struct sockaddr_ll *sll;
564 struct packet_sock *po;
565 u8 *skb_head = skb->data;
566 int skb_len = skb->len;
567 unsigned int snaplen, res;
568
569 if (skb->pkt_type == PACKET_LOOPBACK)
570 goto drop;
571
572 sk = pt->af_packet_priv;
573 po = pkt_sk(sk);
574
575 if (!net_eq(dev_net(dev), sock_net(sk)))
576 goto drop;
577
578 skb->dev = dev;
579
580 if (dev->header_ops) {
581 /* The device has an explicit notion of ll header,
582 * exported to higher levels.
583 *
584 * Otherwise, the device hides details of its frame
585 * structure, so that corresponding packet head is
586 * never delivered to user.
587 */
588 if (sk->sk_type != SOCK_DGRAM)
589 skb_push(skb, skb->data - skb_mac_header(skb));
590 else if (skb->pkt_type == PACKET_OUTGOING) {
591 /* Special case: outgoing packets have ll header at head */
592 skb_pull(skb, skb_network_offset(skb));
593 }
594 }
595
596 snaplen = skb->len;
597
598 res = run_filter(skb, sk, snaplen);
599 if (!res)
600 goto drop_n_restore;
601 if (snaplen > res)
602 snaplen = res;
603
604 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
605 (unsigned)sk->sk_rcvbuf)
606 goto drop_n_acct;
607
608 if (skb_shared(skb)) {
609 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
610 if (nskb == NULL)
611 goto drop_n_acct;
612
613 if (skb_head != skb->data) {
614 skb->data = skb_head;
615 skb->len = skb_len;
616 }
617 kfree_skb(skb);
618 skb = nskb;
619 }
620
621 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
622 sizeof(skb->cb));
623
624 sll = &PACKET_SKB_CB(skb)->sa.ll;
625 sll->sll_family = AF_PACKET;
626 sll->sll_hatype = dev->type;
627 sll->sll_protocol = skb->protocol;
628 sll->sll_pkttype = skb->pkt_type;
629 if (unlikely(po->origdev))
630 sll->sll_ifindex = orig_dev->ifindex;
631 else
632 sll->sll_ifindex = dev->ifindex;
633
634 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
635
636 PACKET_SKB_CB(skb)->origlen = skb->len;
637
638 if (pskb_trim(skb, snaplen))
639 goto drop_n_acct;
640
641 skb_set_owner_r(skb, sk);
642 skb->dev = NULL;
643 skb_dst_drop(skb);
644
645 /* drop conntrack reference */
646 nf_reset(skb);
647
648 spin_lock(&sk->sk_receive_queue.lock);
649 po->stats.tp_packets++;
650 skb->dropcount = atomic_read(&sk->sk_drops);
651 __skb_queue_tail(&sk->sk_receive_queue, skb);
652 spin_unlock(&sk->sk_receive_queue.lock);
653 sk->sk_data_ready(sk, skb->len);
654 return 0;
655
656 drop_n_acct:
657 po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
658
659 drop_n_restore:
660 if (skb_head != skb->data && skb_shared(skb)) {
661 skb->data = skb_head;
662 skb->len = skb_len;
663 }
664 drop:
665 consume_skb(skb);
666 return 0;
667 }
668
669 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
670 struct packet_type *pt, struct net_device *orig_dev)
671 {
672 struct sock *sk;
673 struct packet_sock *po;
674 struct sockaddr_ll *sll;
675 union {
676 struct tpacket_hdr *h1;
677 struct tpacket2_hdr *h2;
678 void *raw;
679 } h;
680 u8 *skb_head = skb->data;
681 int skb_len = skb->len;
682 unsigned int snaplen, res;
683 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
684 unsigned short macoff, netoff, hdrlen;
685 struct sk_buff *copy_skb = NULL;
686 struct timeval tv;
687 struct timespec ts;
688 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
689
690 if (skb->pkt_type == PACKET_LOOPBACK)
691 goto drop;
692
693 sk = pt->af_packet_priv;
694 po = pkt_sk(sk);
695
696 if (!net_eq(dev_net(dev), sock_net(sk)))
697 goto drop;
698
699 if (dev->header_ops) {
700 if (sk->sk_type != SOCK_DGRAM)
701 skb_push(skb, skb->data - skb_mac_header(skb));
702 else if (skb->pkt_type == PACKET_OUTGOING) {
703 /* Special case: outgoing packets have ll header at head */
704 skb_pull(skb, skb_network_offset(skb));
705 }
706 }
707
708 if (skb->ip_summed == CHECKSUM_PARTIAL)
709 status |= TP_STATUS_CSUMNOTREADY;
710
711 snaplen = skb->len;
712
713 res = run_filter(skb, sk, snaplen);
714 if (!res)
715 goto drop_n_restore;
716 if (snaplen > res)
717 snaplen = res;
718
719 if (sk->sk_type == SOCK_DGRAM) {
720 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
721 po->tp_reserve;
722 } else {
723 unsigned maclen = skb_network_offset(skb);
724 netoff = TPACKET_ALIGN(po->tp_hdrlen +
725 (maclen < 16 ? 16 : maclen)) +
726 po->tp_reserve;
727 macoff = netoff - maclen;
728 }
729
730 if (macoff + snaplen > po->rx_ring.frame_size) {
731 if (po->copy_thresh &&
732 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
733 (unsigned)sk->sk_rcvbuf) {
734 if (skb_shared(skb)) {
735 copy_skb = skb_clone(skb, GFP_ATOMIC);
736 } else {
737 copy_skb = skb_get(skb);
738 skb_head = skb->data;
739 }
740 if (copy_skb)
741 skb_set_owner_r(copy_skb, sk);
742 }
743 snaplen = po->rx_ring.frame_size - macoff;
744 if ((int)snaplen < 0)
745 snaplen = 0;
746 }
747
748 spin_lock(&sk->sk_receive_queue.lock);
749 h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
750 if (!h.raw)
751 goto ring_is_full;
752 packet_increment_head(&po->rx_ring);
753 po->stats.tp_packets++;
754 if (copy_skb) {
755 status |= TP_STATUS_COPY;
756 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
757 }
758 if (!po->stats.tp_drops)
759 status &= ~TP_STATUS_LOSING;
760 spin_unlock(&sk->sk_receive_queue.lock);
761
762 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
763
764 switch (po->tp_version) {
765 case TPACKET_V1:
766 h.h1->tp_len = skb->len;
767 h.h1->tp_snaplen = snaplen;
768 h.h1->tp_mac = macoff;
769 h.h1->tp_net = netoff;
770 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
771 && shhwtstamps->syststamp.tv64)
772 tv = ktime_to_timeval(shhwtstamps->syststamp);
773 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
774 && shhwtstamps->hwtstamp.tv64)
775 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
776 else if (skb->tstamp.tv64)
777 tv = ktime_to_timeval(skb->tstamp);
778 else
779 do_gettimeofday(&tv);
780 h.h1->tp_sec = tv.tv_sec;
781 h.h1->tp_usec = tv.tv_usec;
782 hdrlen = sizeof(*h.h1);
783 break;
784 case TPACKET_V2:
785 h.h2->tp_len = skb->len;
786 h.h2->tp_snaplen = snaplen;
787 h.h2->tp_mac = macoff;
788 h.h2->tp_net = netoff;
789 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
790 && shhwtstamps->syststamp.tv64)
791 ts = ktime_to_timespec(shhwtstamps->syststamp);
792 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
793 && shhwtstamps->hwtstamp.tv64)
794 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
795 else if (skb->tstamp.tv64)
796 ts = ktime_to_timespec(skb->tstamp);
797 else
798 getnstimeofday(&ts);
799 h.h2->tp_sec = ts.tv_sec;
800 h.h2->tp_nsec = ts.tv_nsec;
801 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
802 h.h2->tp_padding = 0;
803 hdrlen = sizeof(*h.h2);
804 break;
805 default:
806 BUG();
807 }
808
809 sll = h.raw + TPACKET_ALIGN(hdrlen);
810 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
811 sll->sll_family = AF_PACKET;
812 sll->sll_hatype = dev->type;
813 sll->sll_protocol = skb->protocol;
814 sll->sll_pkttype = skb->pkt_type;
815 if (unlikely(po->origdev))
816 sll->sll_ifindex = orig_dev->ifindex;
817 else
818 sll->sll_ifindex = dev->ifindex;
819
820 __packet_set_status(po, h.raw, status);
821 smp_mb();
822 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
823 {
824 u8 *start, *end;
825
826 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw + macoff + snaplen);
827 for (start = h.raw; start < end; start += PAGE_SIZE)
828 flush_dcache_page(pgv_to_page(start));
829 }
830 #endif
831
832 sk->sk_data_ready(sk, 0);
833
834 drop_n_restore:
835 if (skb_head != skb->data && skb_shared(skb)) {
836 skb->data = skb_head;
837 skb->len = skb_len;
838 }
839 drop:
840 kfree_skb(skb);
841 return 0;
842
843 ring_is_full:
844 po->stats.tp_drops++;
845 spin_unlock(&sk->sk_receive_queue.lock);
846
847 sk->sk_data_ready(sk, 0);
848 kfree_skb(copy_skb);
849 goto drop_n_restore;
850 }
851
852 static void tpacket_destruct_skb(struct sk_buff *skb)
853 {
854 struct packet_sock *po = pkt_sk(skb->sk);
855 void *ph;
856
857 BUG_ON(skb == NULL);
858
859 if (likely(po->tx_ring.pg_vec)) {
860 ph = skb_shinfo(skb)->destructor_arg;
861 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
862 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
863 atomic_dec(&po->tx_ring.pending);
864 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
865 }
866
867 sock_wfree(skb);
868 }
869
870 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
871 void *frame, struct net_device *dev, int size_max,
872 __be16 proto, unsigned char *addr)
873 {
874 union {
875 struct tpacket_hdr *h1;
876 struct tpacket2_hdr *h2;
877 void *raw;
878 } ph;
879 int to_write, offset, len, tp_len, nr_frags, len_max;
880 struct socket *sock = po->sk.sk_socket;
881 struct page *page;
882 void *data;
883 int err;
884
885 ph.raw = frame;
886
887 skb->protocol = proto;
888 skb->dev = dev;
889 skb->priority = po->sk.sk_priority;
890 skb->mark = po->sk.sk_mark;
891 skb_shinfo(skb)->destructor_arg = ph.raw;
892
893 switch (po->tp_version) {
894 case TPACKET_V2:
895 tp_len = ph.h2->tp_len;
896 break;
897 default:
898 tp_len = ph.h1->tp_len;
899 break;
900 }
901 if (unlikely(tp_len > size_max)) {
902 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
903 return -EMSGSIZE;
904 }
905
906 skb_reserve(skb, LL_RESERVED_SPACE(dev));
907 skb_reset_network_header(skb);
908
909 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
910 to_write = tp_len;
911
912 if (sock->type == SOCK_DGRAM) {
913 err = dev_hard_header(skb, dev, ntohs(proto), addr,
914 NULL, tp_len);
915 if (unlikely(err < 0))
916 return -EINVAL;
917 } else if (dev->hard_header_len) {
918 /* net device doesn't like empty head */
919 if (unlikely(tp_len <= dev->hard_header_len)) {
920 pr_err("packet size is too short (%d < %d)\n",
921 tp_len, dev->hard_header_len);
922 return -EINVAL;
923 }
924
925 skb_push(skb, dev->hard_header_len);
926 err = skb_store_bits(skb, 0, data,
927 dev->hard_header_len);
928 if (unlikely(err))
929 return err;
930
931 data += dev->hard_header_len;
932 to_write -= dev->hard_header_len;
933 }
934
935 err = -EFAULT;
936 offset = offset_in_page(data);
937 len_max = PAGE_SIZE - offset;
938 len = ((to_write > len_max) ? len_max : to_write);
939
940 skb->data_len = to_write;
941 skb->len += to_write;
942 skb->truesize += to_write;
943 atomic_add(to_write, &po->sk.sk_wmem_alloc);
944
945 while (likely(to_write)) {
946 nr_frags = skb_shinfo(skb)->nr_frags;
947
948 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
949 pr_err("Packet exceed the number of skb frags(%lu)\n",
950 MAX_SKB_FRAGS);
951 return -EFAULT;
952 }
953
954 page = pgv_to_page(data);
955 data += len;
956 flush_dcache_page(page);
957 get_page(page);
958 skb_fill_page_desc(skb, nr_frags, page, offset, len);
959 to_write -= len;
960 offset = 0;
961 len_max = PAGE_SIZE;
962 len = ((to_write > len_max) ? len_max : to_write);
963 }
964
965 return tp_len;
966 }
967
968 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
969 {
970 struct sk_buff *skb;
971 struct net_device *dev;
972 __be16 proto;
973 int ifindex, err, reserve = 0;
974 void *ph;
975 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
976 int tp_len, size_max;
977 unsigned char *addr;
978 int len_sum = 0;
979 int status = 0;
980
981 mutex_lock(&po->pg_vec_lock);
982
983 err = -EBUSY;
984 if (saddr == NULL) {
985 ifindex = po->ifindex;
986 proto = po->num;
987 addr = NULL;
988 } else {
989 err = -EINVAL;
990 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
991 goto out;
992 if (msg->msg_namelen < (saddr->sll_halen
993 + offsetof(struct sockaddr_ll,
994 sll_addr)))
995 goto out;
996 ifindex = saddr->sll_ifindex;
997 proto = saddr->sll_protocol;
998 addr = saddr->sll_addr;
999 }
1000
1001 dev = dev_get_by_index(sock_net(&po->sk), ifindex);
1002 err = -ENXIO;
1003 if (unlikely(dev == NULL))
1004 goto out;
1005
1006 reserve = dev->hard_header_len;
1007
1008 err = -ENETDOWN;
1009 if (unlikely(!(dev->flags & IFF_UP)))
1010 goto out_put;
1011
1012 size_max = po->tx_ring.frame_size
1013 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1014
1015 if (size_max > dev->mtu + reserve)
1016 size_max = dev->mtu + reserve;
1017
1018 do {
1019 ph = packet_current_frame(po, &po->tx_ring,
1020 TP_STATUS_SEND_REQUEST);
1021
1022 if (unlikely(ph == NULL)) {
1023 schedule();
1024 continue;
1025 }
1026
1027 status = TP_STATUS_SEND_REQUEST;
1028 skb = sock_alloc_send_skb(&po->sk,
1029 LL_ALLOCATED_SPACE(dev)
1030 + sizeof(struct sockaddr_ll),
1031 0, &err);
1032
1033 if (unlikely(skb == NULL))
1034 goto out_status;
1035
1036 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1037 addr);
1038
1039 if (unlikely(tp_len < 0)) {
1040 if (po->tp_loss) {
1041 __packet_set_status(po, ph,
1042 TP_STATUS_AVAILABLE);
1043 packet_increment_head(&po->tx_ring);
1044 kfree_skb(skb);
1045 continue;
1046 } else {
1047 status = TP_STATUS_WRONG_FORMAT;
1048 err = tp_len;
1049 goto out_status;
1050 }
1051 }
1052
1053 skb->destructor = tpacket_destruct_skb;
1054 __packet_set_status(po, ph, TP_STATUS_SENDING);
1055 atomic_inc(&po->tx_ring.pending);
1056
1057 status = TP_STATUS_SEND_REQUEST;
1058 err = dev_queue_xmit(skb);
1059 if (unlikely(err > 0)) {
1060 err = net_xmit_errno(err);
1061 if (err && __packet_get_status(po, ph) ==
1062 TP_STATUS_AVAILABLE) {
1063 /* skb was destructed already */
1064 skb = NULL;
1065 goto out_status;
1066 }
1067 /*
1068 * skb was dropped but not destructed yet;
1069 * let's treat it like congestion or err < 0
1070 */
1071 err = 0;
1072 }
1073 packet_increment_head(&po->tx_ring);
1074 len_sum += tp_len;
1075 } while (likely((ph != NULL) ||
1076 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1077 (atomic_read(&po->tx_ring.pending))))
1078 );
1079
1080 err = len_sum;
1081 goto out_put;
1082
1083 out_status:
1084 __packet_set_status(po, ph, status);
1085 kfree_skb(skb);
1086 out_put:
1087 dev_put(dev);
1088 out:
1089 mutex_unlock(&po->pg_vec_lock);
1090 return err;
1091 }
1092
1093 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1094 size_t reserve, size_t len,
1095 size_t linear, int noblock,
1096 int *err)
1097 {
1098 struct sk_buff *skb;
1099
1100 /* Under a page? Don't bother with paged skb. */
1101 if (prepad + len < PAGE_SIZE || !linear)
1102 linear = len;
1103
1104 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1105 err);
1106 if (!skb)
1107 return NULL;
1108
1109 skb_reserve(skb, reserve);
1110 skb_put(skb, linear);
1111 skb->data_len = len - linear;
1112 skb->len += len - linear;
1113
1114 return skb;
1115 }
1116
1117 static int packet_snd(struct socket *sock,
1118 struct msghdr *msg, size_t len)
1119 {
1120 struct sock *sk = sock->sk;
1121 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1122 struct sk_buff *skb;
1123 struct net_device *dev;
1124 __be16 proto;
1125 unsigned char *addr;
1126 int ifindex, err, reserve = 0;
1127 struct virtio_net_hdr vnet_hdr = { 0 };
1128 int offset = 0;
1129 int vnet_hdr_len;
1130 struct packet_sock *po = pkt_sk(sk);
1131 unsigned short gso_type = 0;
1132
1133 /*
1134 * Get and verify the address.
1135 */
1136
1137 if (saddr == NULL) {
1138 ifindex = po->ifindex;
1139 proto = po->num;
1140 addr = NULL;
1141 } else {
1142 err = -EINVAL;
1143 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1144 goto out;
1145 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1146 goto out;
1147 ifindex = saddr->sll_ifindex;
1148 proto = saddr->sll_protocol;
1149 addr = saddr->sll_addr;
1150 }
1151
1152
1153 dev = dev_get_by_index(sock_net(sk), ifindex);
1154 err = -ENXIO;
1155 if (dev == NULL)
1156 goto out_unlock;
1157 if (sock->type == SOCK_RAW)
1158 reserve = dev->hard_header_len;
1159
1160 err = -ENETDOWN;
1161 if (!(dev->flags & IFF_UP))
1162 goto out_unlock;
1163
1164 if (po->has_vnet_hdr) {
1165 vnet_hdr_len = sizeof(vnet_hdr);
1166
1167 err = -EINVAL;
1168 if (len < vnet_hdr_len)
1169 goto out_unlock;
1170
1171 len -= vnet_hdr_len;
1172
1173 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1174 vnet_hdr_len);
1175 if (err < 0)
1176 goto out_unlock;
1177
1178 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1179 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1180 vnet_hdr.hdr_len))
1181 vnet_hdr.hdr_len = vnet_hdr.csum_start +
1182 vnet_hdr.csum_offset + 2;
1183
1184 err = -EINVAL;
1185 if (vnet_hdr.hdr_len > len)
1186 goto out_unlock;
1187
1188 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1189 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1190 case VIRTIO_NET_HDR_GSO_TCPV4:
1191 gso_type = SKB_GSO_TCPV4;
1192 break;
1193 case VIRTIO_NET_HDR_GSO_TCPV6:
1194 gso_type = SKB_GSO_TCPV6;
1195 break;
1196 case VIRTIO_NET_HDR_GSO_UDP:
1197 gso_type = SKB_GSO_UDP;
1198 break;
1199 default:
1200 goto out_unlock;
1201 }
1202
1203 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1204 gso_type |= SKB_GSO_TCP_ECN;
1205
1206 if (vnet_hdr.gso_size == 0)
1207 goto out_unlock;
1208
1209 }
1210 }
1211
1212 err = -EMSGSIZE;
1213 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
1214 goto out_unlock;
1215
1216 err = -ENOBUFS;
1217 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1218 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1219 msg->msg_flags & MSG_DONTWAIT, &err);
1220 if (skb == NULL)
1221 goto out_unlock;
1222
1223 skb_set_network_header(skb, reserve);
1224
1225 err = -EINVAL;
1226 if (sock->type == SOCK_DGRAM &&
1227 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1228 goto out_free;
1229
1230 /* Returns -EFAULT on error */
1231 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1232 if (err)
1233 goto out_free;
1234 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1235 if (err < 0)
1236 goto out_free;
1237
1238 if (!gso_type && (len > dev->mtu + reserve)) {
1239 /* Earlier code assumed this would be a VLAN pkt,
1240 * double-check this now that we have the actual
1241 * packet in hand.
1242 */
1243 struct ethhdr *ehdr;
1244 skb_reset_mac_header(skb);
1245 ehdr = eth_hdr(skb);
1246 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1247 err = -EMSGSIZE;
1248 goto out_free;
1249 }
1250 }
1251
1252 skb->protocol = proto;
1253 skb->dev = dev;
1254 skb->priority = sk->sk_priority;
1255 skb->mark = sk->sk_mark;
1256
1257 if (po->has_vnet_hdr) {
1258 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1259 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1260 vnet_hdr.csum_offset)) {
1261 err = -EINVAL;
1262 goto out_free;
1263 }
1264 }
1265
1266 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1267 skb_shinfo(skb)->gso_type = gso_type;
1268
1269 /* Header must be checked, and gso_segs computed. */
1270 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1271 skb_shinfo(skb)->gso_segs = 0;
1272
1273 len += vnet_hdr_len;
1274 }
1275
1276 /*
1277 * Now send it
1278 */
1279
1280 err = dev_queue_xmit(skb);
1281 if (err > 0 && (err = net_xmit_errno(err)) != 0)
1282 goto out_unlock;
1283
1284 dev_put(dev);
1285
1286 return len;
1287
1288 out_free:
1289 kfree_skb(skb);
1290 out_unlock:
1291 if (dev)
1292 dev_put(dev);
1293 out:
1294 return err;
1295 }
1296
1297 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1298 struct msghdr *msg, size_t len)
1299 {
1300 struct sock *sk = sock->sk;
1301 struct packet_sock *po = pkt_sk(sk);
1302 if (po->tx_ring.pg_vec)
1303 return tpacket_snd(po, msg);
1304 else
1305 return packet_snd(sock, msg, len);
1306 }
1307
1308 /*
1309 * Close a PACKET socket. This is fairly simple. We immediately go
1310 * to 'closed' state and remove our protocol entry in the device list.
1311 */
1312
1313 static int packet_release(struct socket *sock)
1314 {
1315 struct sock *sk = sock->sk;
1316 struct packet_sock *po;
1317 struct net *net;
1318 struct tpacket_req req;
1319
1320 if (!sk)
1321 return 0;
1322
1323 net = sock_net(sk);
1324 po = pkt_sk(sk);
1325
1326 spin_lock_bh(&net->packet.sklist_lock);
1327 sk_del_node_init_rcu(sk);
1328 sock_prot_inuse_add(net, sk->sk_prot, -1);
1329 spin_unlock_bh(&net->packet.sklist_lock);
1330
1331 spin_lock(&po->bind_lock);
1332 if (po->running) {
1333 /*
1334 * Remove from protocol table
1335 */
1336 po->running = 0;
1337 po->num = 0;
1338 __dev_remove_pack(&po->prot_hook);
1339 __sock_put(sk);
1340 }
1341 spin_unlock(&po->bind_lock);
1342
1343 packet_flush_mclist(sk);
1344
1345 memset(&req, 0, sizeof(req));
1346
1347 if (po->rx_ring.pg_vec)
1348 packet_set_ring(sk, &req, 1, 0);
1349
1350 if (po->tx_ring.pg_vec)
1351 packet_set_ring(sk, &req, 1, 1);
1352
1353 synchronize_net();
1354 /*
1355 * Now the socket is dead. No more input will appear.
1356 */
1357 sock_orphan(sk);
1358 sock->sk = NULL;
1359
1360 /* Purge queues */
1361
1362 skb_queue_purge(&sk->sk_receive_queue);
1363 sk_refcnt_debug_release(sk);
1364
1365 sock_put(sk);
1366 return 0;
1367 }
1368
1369 /*
1370 * Attach a packet hook.
1371 */
1372
1373 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1374 {
1375 struct packet_sock *po = pkt_sk(sk);
1376 /*
1377 * Detach an existing hook if present.
1378 */
1379
1380 lock_sock(sk);
1381
1382 spin_lock(&po->bind_lock);
1383 if (po->running) {
1384 __sock_put(sk);
1385 po->running = 0;
1386 po->num = 0;
1387 spin_unlock(&po->bind_lock);
1388 dev_remove_pack(&po->prot_hook);
1389 spin_lock(&po->bind_lock);
1390 }
1391
1392 po->num = protocol;
1393 po->prot_hook.type = protocol;
1394 po->prot_hook.dev = dev;
1395
1396 po->ifindex = dev ? dev->ifindex : 0;
1397
1398 if (protocol == 0)
1399 goto out_unlock;
1400
1401 if (!dev || (dev->flags & IFF_UP)) {
1402 dev_add_pack(&po->prot_hook);
1403 sock_hold(sk);
1404 po->running = 1;
1405 } else {
1406 sk->sk_err = ENETDOWN;
1407 if (!sock_flag(sk, SOCK_DEAD))
1408 sk->sk_error_report(sk);
1409 }
1410
1411 out_unlock:
1412 spin_unlock(&po->bind_lock);
1413 release_sock(sk);
1414 return 0;
1415 }
1416
1417 /*
1418 * Bind a packet socket to a device
1419 */
1420
1421 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1422 int addr_len)
1423 {
1424 struct sock *sk = sock->sk;
1425 char name[15];
1426 struct net_device *dev;
1427 int err = -ENODEV;
1428
1429 /*
1430 * Check legality
1431 */
1432
1433 if (addr_len != sizeof(struct sockaddr))
1434 return -EINVAL;
1435 strlcpy(name, uaddr->sa_data, sizeof(name));
1436
1437 dev = dev_get_by_name(sock_net(sk), name);
1438 if (dev) {
1439 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1440 dev_put(dev);
1441 }
1442 return err;
1443 }
1444
1445 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1446 {
1447 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1448 struct sock *sk = sock->sk;
1449 struct net_device *dev = NULL;
1450 int err;
1451
1452
1453 /*
1454 * Check legality
1455 */
1456
1457 if (addr_len < sizeof(struct sockaddr_ll))
1458 return -EINVAL;
1459 if (sll->sll_family != AF_PACKET)
1460 return -EINVAL;
1461
1462 if (sll->sll_ifindex) {
1463 err = -ENODEV;
1464 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1465 if (dev == NULL)
1466 goto out;
1467 }
1468 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1469 if (dev)
1470 dev_put(dev);
1471
1472 out:
1473 return err;
1474 }
1475
1476 static struct proto packet_proto = {
1477 .name = "PACKET",
1478 .owner = THIS_MODULE,
1479 .obj_size = sizeof(struct packet_sock),
1480 };
1481
1482 /*
1483 * Create a packet of type SOCK_PACKET.
1484 */
1485
1486 static int packet_create(struct net *net, struct socket *sock, int protocol,
1487 int kern)
1488 {
1489 struct sock *sk;
1490 struct packet_sock *po;
1491 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1492 int err;
1493
1494 if (!capable(CAP_NET_RAW))
1495 return -EPERM;
1496 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1497 sock->type != SOCK_PACKET)
1498 return -ESOCKTNOSUPPORT;
1499
1500 sock->state = SS_UNCONNECTED;
1501
1502 err = -ENOBUFS;
1503 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1504 if (sk == NULL)
1505 goto out;
1506
1507 sock->ops = &packet_ops;
1508 if (sock->type == SOCK_PACKET)
1509 sock->ops = &packet_ops_spkt;
1510
1511 sock_init_data(sock, sk);
1512
1513 po = pkt_sk(sk);
1514 sk->sk_family = PF_PACKET;
1515 po->num = proto;
1516
1517 sk->sk_destruct = packet_sock_destruct;
1518 sk_refcnt_debug_inc(sk);
1519
1520 /*
1521 * Attach a protocol block
1522 */
1523
1524 spin_lock_init(&po->bind_lock);
1525 mutex_init(&po->pg_vec_lock);
1526 po->prot_hook.func = packet_rcv;
1527
1528 if (sock->type == SOCK_PACKET)
1529 po->prot_hook.func = packet_rcv_spkt;
1530
1531 po->prot_hook.af_packet_priv = sk;
1532
1533 if (proto) {
1534 po->prot_hook.type = proto;
1535 dev_add_pack(&po->prot_hook);
1536 sock_hold(sk);
1537 po->running = 1;
1538 }
1539
1540 spin_lock_bh(&net->packet.sklist_lock);
1541 sk_add_node_rcu(sk, &net->packet.sklist);
1542 sock_prot_inuse_add(net, &packet_proto, 1);
1543 spin_unlock_bh(&net->packet.sklist_lock);
1544
1545 return 0;
1546 out:
1547 return err;
1548 }
1549
1550 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1551 {
1552 struct sock_exterr_skb *serr;
1553 struct sk_buff *skb, *skb2;
1554 int copied, err;
1555
1556 err = -EAGAIN;
1557 skb = skb_dequeue(&sk->sk_error_queue);
1558 if (skb == NULL)
1559 goto out;
1560
1561 copied = skb->len;
1562 if (copied > len) {
1563 msg->msg_flags |= MSG_TRUNC;
1564 copied = len;
1565 }
1566 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1567 if (err)
1568 goto out_free_skb;
1569
1570 sock_recv_timestamp(msg, sk, skb);
1571
1572 serr = SKB_EXT_ERR(skb);
1573 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
1574 sizeof(serr->ee), &serr->ee);
1575
1576 msg->msg_flags |= MSG_ERRQUEUE;
1577 err = copied;
1578
1579 /* Reset and regenerate socket error */
1580 spin_lock_bh(&sk->sk_error_queue.lock);
1581 sk->sk_err = 0;
1582 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
1583 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
1584 spin_unlock_bh(&sk->sk_error_queue.lock);
1585 sk->sk_error_report(sk);
1586 } else
1587 spin_unlock_bh(&sk->sk_error_queue.lock);
1588
1589 out_free_skb:
1590 kfree_skb(skb);
1591 out:
1592 return err;
1593 }
1594
1595 /*
1596 * Pull a packet from our receive queue and hand it to the user.
1597 * If necessary we block.
1598 */
1599
1600 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1601 struct msghdr *msg, size_t len, int flags)
1602 {
1603 struct sock *sk = sock->sk;
1604 struct sk_buff *skb;
1605 int copied, err;
1606 struct sockaddr_ll *sll;
1607 int vnet_hdr_len = 0;
1608
1609 err = -EINVAL;
1610 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1611 goto out;
1612
1613 #if 0
1614 /* What error should we return now? EUNATTACH? */
1615 if (pkt_sk(sk)->ifindex < 0)
1616 return -ENODEV;
1617 #endif
1618
1619 if (flags & MSG_ERRQUEUE) {
1620 err = packet_recv_error(sk, msg, len);
1621 goto out;
1622 }
1623
1624 /*
1625 * Call the generic datagram receiver. This handles all sorts
1626 * of horrible races and re-entrancy so we can forget about it
1627 * in the protocol layers.
1628 *
1629 * Now it will return ENETDOWN, if device have just gone down,
1630 * but then it will block.
1631 */
1632
1633 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1634
1635 /*
1636 * An error occurred so return it. Because skb_recv_datagram()
1637 * handles the blocking we don't see and worry about blocking
1638 * retries.
1639 */
1640
1641 if (skb == NULL)
1642 goto out;
1643
1644 if (pkt_sk(sk)->has_vnet_hdr) {
1645 struct virtio_net_hdr vnet_hdr = { 0 };
1646
1647 err = -EINVAL;
1648 vnet_hdr_len = sizeof(vnet_hdr);
1649 if (len < vnet_hdr_len)
1650 goto out_free;
1651
1652 len -= vnet_hdr_len;
1653
1654 if (skb_is_gso(skb)) {
1655 struct skb_shared_info *sinfo = skb_shinfo(skb);
1656
1657 /* This is a hint as to how much should be linear. */
1658 vnet_hdr.hdr_len = skb_headlen(skb);
1659 vnet_hdr.gso_size = sinfo->gso_size;
1660 if (sinfo->gso_type & SKB_GSO_TCPV4)
1661 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1662 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1663 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1664 else if (sinfo->gso_type & SKB_GSO_UDP)
1665 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1666 else if (sinfo->gso_type & SKB_GSO_FCOE)
1667 goto out_free;
1668 else
1669 BUG();
1670 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1671 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1672 } else
1673 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1674
1675 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1676 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1677 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
1678 vnet_hdr.csum_offset = skb->csum_offset;
1679 } /* else everything is zero */
1680
1681 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1682 vnet_hdr_len);
1683 if (err < 0)
1684 goto out_free;
1685 }
1686
1687 /*
1688 * If the address length field is there to be filled in, we fill
1689 * it in now.
1690 */
1691
1692 sll = &PACKET_SKB_CB(skb)->sa.ll;
1693 if (sock->type == SOCK_PACKET)
1694 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1695 else
1696 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1697
1698 /*
1699 * You lose any data beyond the buffer you gave. If it worries a
1700 * user program they can ask the device for its MTU anyway.
1701 */
1702
1703 copied = skb->len;
1704 if (copied > len) {
1705 copied = len;
1706 msg->msg_flags |= MSG_TRUNC;
1707 }
1708
1709 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1710 if (err)
1711 goto out_free;
1712
1713 sock_recv_ts_and_drops(msg, sk, skb);
1714
1715 if (msg->msg_name)
1716 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1717 msg->msg_namelen);
1718
1719 if (pkt_sk(sk)->auxdata) {
1720 struct tpacket_auxdata aux;
1721
1722 aux.tp_status = TP_STATUS_USER;
1723 if (skb->ip_summed == CHECKSUM_PARTIAL)
1724 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1725 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1726 aux.tp_snaplen = skb->len;
1727 aux.tp_mac = 0;
1728 aux.tp_net = skb_network_offset(skb);
1729 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
1730
1731 aux.tp_padding = 0;
1732 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1733 }
1734
1735 /*
1736 * Free or return the buffer as appropriate. Again this
1737 * hides all the races and re-entrancy issues from us.
1738 */
1739 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1740
1741 out_free:
1742 skb_free_datagram(sk, skb);
1743 out:
1744 return err;
1745 }
1746
1747 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1748 int *uaddr_len, int peer)
1749 {
1750 struct net_device *dev;
1751 struct sock *sk = sock->sk;
1752
1753 if (peer)
1754 return -EOPNOTSUPP;
1755
1756 uaddr->sa_family = AF_PACKET;
1757 rcu_read_lock();
1758 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
1759 if (dev)
1760 strncpy(uaddr->sa_data, dev->name, 14);
1761 else
1762 memset(uaddr->sa_data, 0, 14);
1763 rcu_read_unlock();
1764 *uaddr_len = sizeof(*uaddr);
1765
1766 return 0;
1767 }
1768
1769 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1770 int *uaddr_len, int peer)
1771 {
1772 struct net_device *dev;
1773 struct sock *sk = sock->sk;
1774 struct packet_sock *po = pkt_sk(sk);
1775 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1776
1777 if (peer)
1778 return -EOPNOTSUPP;
1779
1780 sll->sll_family = AF_PACKET;
1781 sll->sll_ifindex = po->ifindex;
1782 sll->sll_protocol = po->num;
1783 sll->sll_pkttype = 0;
1784 rcu_read_lock();
1785 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1786 if (dev) {
1787 sll->sll_hatype = dev->type;
1788 sll->sll_halen = dev->addr_len;
1789 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1790 } else {
1791 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1792 sll->sll_halen = 0;
1793 }
1794 rcu_read_unlock();
1795 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1796
1797 return 0;
1798 }
1799
1800 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1801 int what)
1802 {
1803 switch (i->type) {
1804 case PACKET_MR_MULTICAST:
1805 if (i->alen != dev->addr_len)
1806 return -EINVAL;
1807 if (what > 0)
1808 return dev_mc_add(dev, i->addr);
1809 else
1810 return dev_mc_del(dev, i->addr);
1811 break;
1812 case PACKET_MR_PROMISC:
1813 return dev_set_promiscuity(dev, what);
1814 break;
1815 case PACKET_MR_ALLMULTI:
1816 return dev_set_allmulti(dev, what);
1817 break;
1818 case PACKET_MR_UNICAST:
1819 if (i->alen != dev->addr_len)
1820 return -EINVAL;
1821 if (what > 0)
1822 return dev_uc_add(dev, i->addr);
1823 else
1824 return dev_uc_del(dev, i->addr);
1825 break;
1826 default:
1827 break;
1828 }
1829 return 0;
1830 }
1831
1832 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1833 {
1834 for ( ; i; i = i->next) {
1835 if (i->ifindex == dev->ifindex)
1836 packet_dev_mc(dev, i, what);
1837 }
1838 }
1839
1840 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1841 {
1842 struct packet_sock *po = pkt_sk(sk);
1843 struct packet_mclist *ml, *i;
1844 struct net_device *dev;
1845 int err;
1846
1847 rtnl_lock();
1848
1849 err = -ENODEV;
1850 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1851 if (!dev)
1852 goto done;
1853
1854 err = -EINVAL;
1855 if (mreq->mr_alen > dev->addr_len)
1856 goto done;
1857
1858 err = -ENOBUFS;
1859 i = kmalloc(sizeof(*i), GFP_KERNEL);
1860 if (i == NULL)
1861 goto done;
1862
1863 err = 0;
1864 for (ml = po->mclist; ml; ml = ml->next) {
1865 if (ml->ifindex == mreq->mr_ifindex &&
1866 ml->type == mreq->mr_type &&
1867 ml->alen == mreq->mr_alen &&
1868 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1869 ml->count++;
1870 /* Free the new element ... */
1871 kfree(i);
1872 goto done;
1873 }
1874 }
1875
1876 i->type = mreq->mr_type;
1877 i->ifindex = mreq->mr_ifindex;
1878 i->alen = mreq->mr_alen;
1879 memcpy(i->addr, mreq->mr_address, i->alen);
1880 i->count = 1;
1881 i->next = po->mclist;
1882 po->mclist = i;
1883 err = packet_dev_mc(dev, i, 1);
1884 if (err) {
1885 po->mclist = i->next;
1886 kfree(i);
1887 }
1888
1889 done:
1890 rtnl_unlock();
1891 return err;
1892 }
1893
1894 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1895 {
1896 struct packet_mclist *ml, **mlp;
1897
1898 rtnl_lock();
1899
1900 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1901 if (ml->ifindex == mreq->mr_ifindex &&
1902 ml->type == mreq->mr_type &&
1903 ml->alen == mreq->mr_alen &&
1904 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1905 if (--ml->count == 0) {
1906 struct net_device *dev;
1907 *mlp = ml->next;
1908 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1909 if (dev)
1910 packet_dev_mc(dev, ml, -1);
1911 kfree(ml);
1912 }
1913 rtnl_unlock();
1914 return 0;
1915 }
1916 }
1917 rtnl_unlock();
1918 return -EADDRNOTAVAIL;
1919 }
1920
1921 static void packet_flush_mclist(struct sock *sk)
1922 {
1923 struct packet_sock *po = pkt_sk(sk);
1924 struct packet_mclist *ml;
1925
1926 if (!po->mclist)
1927 return;
1928
1929 rtnl_lock();
1930 while ((ml = po->mclist) != NULL) {
1931 struct net_device *dev;
1932
1933 po->mclist = ml->next;
1934 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1935 if (dev != NULL)
1936 packet_dev_mc(dev, ml, -1);
1937 kfree(ml);
1938 }
1939 rtnl_unlock();
1940 }
1941
1942 static int
1943 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1944 {
1945 struct sock *sk = sock->sk;
1946 struct packet_sock *po = pkt_sk(sk);
1947 int ret;
1948
1949 if (level != SOL_PACKET)
1950 return -ENOPROTOOPT;
1951
1952 switch (optname) {
1953 case PACKET_ADD_MEMBERSHIP:
1954 case PACKET_DROP_MEMBERSHIP:
1955 {
1956 struct packet_mreq_max mreq;
1957 int len = optlen;
1958 memset(&mreq, 0, sizeof(mreq));
1959 if (len < sizeof(struct packet_mreq))
1960 return -EINVAL;
1961 if (len > sizeof(mreq))
1962 len = sizeof(mreq);
1963 if (copy_from_user(&mreq, optval, len))
1964 return -EFAULT;
1965 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1966 return -EINVAL;
1967 if (optname == PACKET_ADD_MEMBERSHIP)
1968 ret = packet_mc_add(sk, &mreq);
1969 else
1970 ret = packet_mc_drop(sk, &mreq);
1971 return ret;
1972 }
1973
1974 case PACKET_RX_RING:
1975 case PACKET_TX_RING:
1976 {
1977 struct tpacket_req req;
1978
1979 if (optlen < sizeof(req))
1980 return -EINVAL;
1981 if (pkt_sk(sk)->has_vnet_hdr)
1982 return -EINVAL;
1983 if (copy_from_user(&req, optval, sizeof(req)))
1984 return -EFAULT;
1985 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1986 }
1987 case PACKET_COPY_THRESH:
1988 {
1989 int val;
1990
1991 if (optlen != sizeof(val))
1992 return -EINVAL;
1993 if (copy_from_user(&val, optval, sizeof(val)))
1994 return -EFAULT;
1995
1996 pkt_sk(sk)->copy_thresh = val;
1997 return 0;
1998 }
1999 case PACKET_VERSION:
2000 {
2001 int val;
2002
2003 if (optlen != sizeof(val))
2004 return -EINVAL;
2005 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2006 return -EBUSY;
2007 if (copy_from_user(&val, optval, sizeof(val)))
2008 return -EFAULT;
2009 switch (val) {
2010 case TPACKET_V1:
2011 case TPACKET_V2:
2012 po->tp_version = val;
2013 return 0;
2014 default:
2015 return -EINVAL;
2016 }
2017 }
2018 case PACKET_RESERVE:
2019 {
2020 unsigned int val;
2021
2022 if (optlen != sizeof(val))
2023 return -EINVAL;
2024 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2025 return -EBUSY;
2026 if (copy_from_user(&val, optval, sizeof(val)))
2027 return -EFAULT;
2028 po->tp_reserve = val;
2029 return 0;
2030 }
2031 case PACKET_LOSS:
2032 {
2033 unsigned int val;
2034
2035 if (optlen != sizeof(val))
2036 return -EINVAL;
2037 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2038 return -EBUSY;
2039 if (copy_from_user(&val, optval, sizeof(val)))
2040 return -EFAULT;
2041 po->tp_loss = !!val;
2042 return 0;
2043 }
2044 case PACKET_AUXDATA:
2045 {
2046 int val;
2047
2048 if (optlen < sizeof(val))
2049 return -EINVAL;
2050 if (copy_from_user(&val, optval, sizeof(val)))
2051 return -EFAULT;
2052
2053 po->auxdata = !!val;
2054 return 0;
2055 }
2056 case PACKET_ORIGDEV:
2057 {
2058 int val;
2059
2060 if (optlen < sizeof(val))
2061 return -EINVAL;
2062 if (copy_from_user(&val, optval, sizeof(val)))
2063 return -EFAULT;
2064
2065 po->origdev = !!val;
2066 return 0;
2067 }
2068 case PACKET_VNET_HDR:
2069 {
2070 int val;
2071
2072 if (sock->type != SOCK_RAW)
2073 return -EINVAL;
2074 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2075 return -EBUSY;
2076 if (optlen < sizeof(val))
2077 return -EINVAL;
2078 if (copy_from_user(&val, optval, sizeof(val)))
2079 return -EFAULT;
2080
2081 po->has_vnet_hdr = !!val;
2082 return 0;
2083 }
2084 case PACKET_TIMESTAMP:
2085 {
2086 int val;
2087
2088 if (optlen != sizeof(val))
2089 return -EINVAL;
2090 if (copy_from_user(&val, optval, sizeof(val)))
2091 return -EFAULT;
2092
2093 po->tp_tstamp = val;
2094 return 0;
2095 }
2096 default:
2097 return -ENOPROTOOPT;
2098 }
2099 }
2100
2101 static int packet_getsockopt(struct socket *sock, int level, int optname,
2102 char __user *optval, int __user *optlen)
2103 {
2104 int len;
2105 int val;
2106 struct sock *sk = sock->sk;
2107 struct packet_sock *po = pkt_sk(sk);
2108 void *data;
2109 struct tpacket_stats st;
2110
2111 if (level != SOL_PACKET)
2112 return -ENOPROTOOPT;
2113
2114 if (get_user(len, optlen))
2115 return -EFAULT;
2116
2117 if (len < 0)
2118 return -EINVAL;
2119
2120 switch (optname) {
2121 case PACKET_STATISTICS:
2122 if (len > sizeof(struct tpacket_stats))
2123 len = sizeof(struct tpacket_stats);
2124 spin_lock_bh(&sk->sk_receive_queue.lock);
2125 st = po->stats;
2126 memset(&po->stats, 0, sizeof(st));
2127 spin_unlock_bh(&sk->sk_receive_queue.lock);
2128 st.tp_packets += st.tp_drops;
2129
2130 data = &st;
2131 break;
2132 case PACKET_AUXDATA:
2133 if (len > sizeof(int))
2134 len = sizeof(int);
2135 val = po->auxdata;
2136
2137 data = &val;
2138 break;
2139 case PACKET_ORIGDEV:
2140 if (len > sizeof(int))
2141 len = sizeof(int);
2142 val = po->origdev;
2143
2144 data = &val;
2145 break;
2146 case PACKET_VNET_HDR:
2147 if (len > sizeof(int))
2148 len = sizeof(int);
2149 val = po->has_vnet_hdr;
2150
2151 data = &val;
2152 break;
2153 case PACKET_VERSION:
2154 if (len > sizeof(int))
2155 len = sizeof(int);
2156 val = po->tp_version;
2157 data = &val;
2158 break;
2159 case PACKET_HDRLEN:
2160 if (len > sizeof(int))
2161 len = sizeof(int);
2162 if (copy_from_user(&val, optval, len))
2163 return -EFAULT;
2164 switch (val) {
2165 case TPACKET_V1:
2166 val = sizeof(struct tpacket_hdr);
2167 break;
2168 case TPACKET_V2:
2169 val = sizeof(struct tpacket2_hdr);
2170 break;
2171 default:
2172 return -EINVAL;
2173 }
2174 data = &val;
2175 break;
2176 case PACKET_RESERVE:
2177 if (len > sizeof(unsigned int))
2178 len = sizeof(unsigned int);
2179 val = po->tp_reserve;
2180 data = &val;
2181 break;
2182 case PACKET_LOSS:
2183 if (len > sizeof(unsigned int))
2184 len = sizeof(unsigned int);
2185 val = po->tp_loss;
2186 data = &val;
2187 break;
2188 case PACKET_TIMESTAMP:
2189 if (len > sizeof(int))
2190 len = sizeof(int);
2191 val = po->tp_tstamp;
2192 data = &val;
2193 break;
2194 default:
2195 return -ENOPROTOOPT;
2196 }
2197
2198 if (put_user(len, optlen))
2199 return -EFAULT;
2200 if (copy_to_user(optval, data, len))
2201 return -EFAULT;
2202 return 0;
2203 }
2204
2205
2206 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2207 {
2208 struct sock *sk;
2209 struct hlist_node *node;
2210 struct net_device *dev = data;
2211 struct net *net = dev_net(dev);
2212
2213 rcu_read_lock();
2214 sk_for_each_rcu(sk, node, &net->packet.sklist) {
2215 struct packet_sock *po = pkt_sk(sk);
2216
2217 switch (msg) {
2218 case NETDEV_UNREGISTER:
2219 if (po->mclist)
2220 packet_dev_mclist(dev, po->mclist, -1);
2221 /* fallthrough */
2222
2223 case NETDEV_DOWN:
2224 if (dev->ifindex == po->ifindex) {
2225 spin_lock(&po->bind_lock);
2226 if (po->running) {
2227 __dev_remove_pack(&po->prot_hook);
2228 __sock_put(sk);
2229 po->running = 0;
2230 sk->sk_err = ENETDOWN;
2231 if (!sock_flag(sk, SOCK_DEAD))
2232 sk->sk_error_report(sk);
2233 }
2234 if (msg == NETDEV_UNREGISTER) {
2235 po->ifindex = -1;
2236 po->prot_hook.dev = NULL;
2237 }
2238 spin_unlock(&po->bind_lock);
2239 }
2240 break;
2241 case NETDEV_UP:
2242 if (dev->ifindex == po->ifindex) {
2243 spin_lock(&po->bind_lock);
2244 if (po->num && !po->running) {
2245 dev_add_pack(&po->prot_hook);
2246 sock_hold(sk);
2247 po->running = 1;
2248 }
2249 spin_unlock(&po->bind_lock);
2250 }
2251 break;
2252 }
2253 }
2254 rcu_read_unlock();
2255 return NOTIFY_DONE;
2256 }
2257
2258
2259 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2260 unsigned long arg)
2261 {
2262 struct sock *sk = sock->sk;
2263
2264 switch (cmd) {
2265 case SIOCOUTQ:
2266 {
2267 int amount = sk_wmem_alloc_get(sk);
2268
2269 return put_user(amount, (int __user *)arg);
2270 }
2271 case SIOCINQ:
2272 {
2273 struct sk_buff *skb;
2274 int amount = 0;
2275
2276 spin_lock_bh(&sk->sk_receive_queue.lock);
2277 skb = skb_peek(&sk->sk_receive_queue);
2278 if (skb)
2279 amount = skb->len;
2280 spin_unlock_bh(&sk->sk_receive_queue.lock);
2281 return put_user(amount, (int __user *)arg);
2282 }
2283 case SIOCGSTAMP:
2284 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2285 case SIOCGSTAMPNS:
2286 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2287
2288 #ifdef CONFIG_INET
2289 case SIOCADDRT:
2290 case SIOCDELRT:
2291 case SIOCDARP:
2292 case SIOCGARP:
2293 case SIOCSARP:
2294 case SIOCGIFADDR:
2295 case SIOCSIFADDR:
2296 case SIOCGIFBRDADDR:
2297 case SIOCSIFBRDADDR:
2298 case SIOCGIFNETMASK:
2299 case SIOCSIFNETMASK:
2300 case SIOCGIFDSTADDR:
2301 case SIOCSIFDSTADDR:
2302 case SIOCSIFFLAGS:
2303 return inet_dgram_ops.ioctl(sock, cmd, arg);
2304 #endif
2305
2306 default:
2307 return -ENOIOCTLCMD;
2308 }
2309 return 0;
2310 }
2311
2312 static unsigned int packet_poll(struct file *file, struct socket *sock,
2313 poll_table *wait)
2314 {
2315 struct sock *sk = sock->sk;
2316 struct packet_sock *po = pkt_sk(sk);
2317 unsigned int mask = datagram_poll(file, sock, wait);
2318
2319 spin_lock_bh(&sk->sk_receive_queue.lock);
2320 if (po->rx_ring.pg_vec) {
2321 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2322 mask |= POLLIN | POLLRDNORM;
2323 }
2324 spin_unlock_bh(&sk->sk_receive_queue.lock);
2325 spin_lock_bh(&sk->sk_write_queue.lock);
2326 if (po->tx_ring.pg_vec) {
2327 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2328 mask |= POLLOUT | POLLWRNORM;
2329 }
2330 spin_unlock_bh(&sk->sk_write_queue.lock);
2331 return mask;
2332 }
2333
2334
2335 /* Dirty? Well, I still did not learn better way to account
2336 * for user mmaps.
2337 */
2338
2339 static void packet_mm_open(struct vm_area_struct *vma)
2340 {
2341 struct file *file = vma->vm_file;
2342 struct socket *sock = file->private_data;
2343 struct sock *sk = sock->sk;
2344
2345 if (sk)
2346 atomic_inc(&pkt_sk(sk)->mapped);
2347 }
2348
2349 static void packet_mm_close(struct vm_area_struct *vma)
2350 {
2351 struct file *file = vma->vm_file;
2352 struct socket *sock = file->private_data;
2353 struct sock *sk = sock->sk;
2354
2355 if (sk)
2356 atomic_dec(&pkt_sk(sk)->mapped);
2357 }
2358
2359 static const struct vm_operations_struct packet_mmap_ops = {
2360 .open = packet_mm_open,
2361 .close = packet_mm_close,
2362 };
2363
2364 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
2365 unsigned int len)
2366 {
2367 int i;
2368
2369 for (i = 0; i < len; i++) {
2370 if (likely(pg_vec[i].buffer)) {
2371 if (is_vmalloc_addr(pg_vec[i].buffer))
2372 vfree(pg_vec[i].buffer);
2373 else
2374 free_pages((unsigned long)pg_vec[i].buffer,
2375 order);
2376 pg_vec[i].buffer = NULL;
2377 }
2378 }
2379 kfree(pg_vec);
2380 }
2381
2382 static inline char *alloc_one_pg_vec_page(unsigned long order)
2383 {
2384 char *buffer = NULL;
2385 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
2386 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
2387
2388 buffer = (char *) __get_free_pages(gfp_flags, order);
2389
2390 if (buffer)
2391 return buffer;
2392
2393 /*
2394 * __get_free_pages failed, fall back to vmalloc
2395 */
2396 buffer = vzalloc((1 << order) * PAGE_SIZE);
2397
2398 if (buffer)
2399 return buffer;
2400
2401 /*
2402 * vmalloc failed, lets dig into swap here
2403 */
2404 gfp_flags &= ~__GFP_NORETRY;
2405 buffer = (char *)__get_free_pages(gfp_flags, order);
2406 if (buffer)
2407 return buffer;
2408
2409 /*
2410 * complete and utter failure
2411 */
2412 return NULL;
2413 }
2414
2415 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
2416 {
2417 unsigned int block_nr = req->tp_block_nr;
2418 struct pgv *pg_vec;
2419 int i;
2420
2421 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
2422 if (unlikely(!pg_vec))
2423 goto out;
2424
2425 for (i = 0; i < block_nr; i++) {
2426 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
2427 if (unlikely(!pg_vec[i].buffer))
2428 goto out_free_pgvec;
2429 }
2430
2431 out:
2432 return pg_vec;
2433
2434 out_free_pgvec:
2435 free_pg_vec(pg_vec, order, block_nr);
2436 pg_vec = NULL;
2437 goto out;
2438 }
2439
2440 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2441 int closing, int tx_ring)
2442 {
2443 struct pgv *pg_vec = NULL;
2444 struct packet_sock *po = pkt_sk(sk);
2445 int was_running, order = 0;
2446 struct packet_ring_buffer *rb;
2447 struct sk_buff_head *rb_queue;
2448 __be16 num;
2449 int err;
2450
2451 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2452 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2453
2454 err = -EBUSY;
2455 if (!closing) {
2456 if (atomic_read(&po->mapped))
2457 goto out;
2458 if (atomic_read(&rb->pending))
2459 goto out;
2460 }
2461
2462 if (req->tp_block_nr) {
2463 /* Sanity tests and some calculations */
2464 err = -EBUSY;
2465 if (unlikely(rb->pg_vec))
2466 goto out;
2467
2468 switch (po->tp_version) {
2469 case TPACKET_V1:
2470 po->tp_hdrlen = TPACKET_HDRLEN;
2471 break;
2472 case TPACKET_V2:
2473 po->tp_hdrlen = TPACKET2_HDRLEN;
2474 break;
2475 }
2476
2477 err = -EINVAL;
2478 if (unlikely((int)req->tp_block_size <= 0))
2479 goto out;
2480 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2481 goto out;
2482 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2483 po->tp_reserve))
2484 goto out;
2485 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2486 goto out;
2487
2488 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2489 if (unlikely(rb->frames_per_block <= 0))
2490 goto out;
2491 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2492 req->tp_frame_nr))
2493 goto out;
2494
2495 err = -ENOMEM;
2496 order = get_order(req->tp_block_size);
2497 pg_vec = alloc_pg_vec(req, order);
2498 if (unlikely(!pg_vec))
2499 goto out;
2500 }
2501 /* Done */
2502 else {
2503 err = -EINVAL;
2504 if (unlikely(req->tp_frame_nr))
2505 goto out;
2506 }
2507
2508 lock_sock(sk);
2509
2510 /* Detach socket from network */
2511 spin_lock(&po->bind_lock);
2512 was_running = po->running;
2513 num = po->num;
2514 if (was_running) {
2515 __dev_remove_pack(&po->prot_hook);
2516 po->num = 0;
2517 po->running = 0;
2518 __sock_put(sk);
2519 }
2520 spin_unlock(&po->bind_lock);
2521
2522 synchronize_net();
2523
2524 err = -EBUSY;
2525 mutex_lock(&po->pg_vec_lock);
2526 if (closing || atomic_read(&po->mapped) == 0) {
2527 err = 0;
2528 spin_lock_bh(&rb_queue->lock);
2529 swap(rb->pg_vec, pg_vec);
2530 rb->frame_max = (req->tp_frame_nr - 1);
2531 rb->head = 0;
2532 rb->frame_size = req->tp_frame_size;
2533 spin_unlock_bh(&rb_queue->lock);
2534
2535 swap(rb->pg_vec_order, order);
2536 swap(rb->pg_vec_len, req->tp_block_nr);
2537
2538 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2539 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2540 tpacket_rcv : packet_rcv;
2541 skb_queue_purge(rb_queue);
2542 if (atomic_read(&po->mapped))
2543 pr_err("packet_mmap: vma is busy: %d\n",
2544 atomic_read(&po->mapped));
2545 }
2546 mutex_unlock(&po->pg_vec_lock);
2547
2548 spin_lock(&po->bind_lock);
2549 if (was_running && !po->running) {
2550 sock_hold(sk);
2551 po->running = 1;
2552 po->num = num;
2553 dev_add_pack(&po->prot_hook);
2554 }
2555 spin_unlock(&po->bind_lock);
2556
2557 release_sock(sk);
2558
2559 if (pg_vec)
2560 free_pg_vec(pg_vec, order, req->tp_block_nr);
2561 out:
2562 return err;
2563 }
2564
2565 static int packet_mmap(struct file *file, struct socket *sock,
2566 struct vm_area_struct *vma)
2567 {
2568 struct sock *sk = sock->sk;
2569 struct packet_sock *po = pkt_sk(sk);
2570 unsigned long size, expected_size;
2571 struct packet_ring_buffer *rb;
2572 unsigned long start;
2573 int err = -EINVAL;
2574 int i;
2575
2576 if (vma->vm_pgoff)
2577 return -EINVAL;
2578
2579 mutex_lock(&po->pg_vec_lock);
2580
2581 expected_size = 0;
2582 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2583 if (rb->pg_vec) {
2584 expected_size += rb->pg_vec_len
2585 * rb->pg_vec_pages
2586 * PAGE_SIZE;
2587 }
2588 }
2589
2590 if (expected_size == 0)
2591 goto out;
2592
2593 size = vma->vm_end - vma->vm_start;
2594 if (size != expected_size)
2595 goto out;
2596
2597 start = vma->vm_start;
2598 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2599 if (rb->pg_vec == NULL)
2600 continue;
2601
2602 for (i = 0; i < rb->pg_vec_len; i++) {
2603 struct page *page;
2604 void *kaddr = rb->pg_vec[i].buffer;
2605 int pg_num;
2606
2607 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
2608 page = pgv_to_page(kaddr);
2609 err = vm_insert_page(vma, start, page);
2610 if (unlikely(err))
2611 goto out;
2612 start += PAGE_SIZE;
2613 kaddr += PAGE_SIZE;
2614 }
2615 }
2616 }
2617
2618 atomic_inc(&po->mapped);
2619 vma->vm_ops = &packet_mmap_ops;
2620 err = 0;
2621
2622 out:
2623 mutex_unlock(&po->pg_vec_lock);
2624 return err;
2625 }
2626
2627 static const struct proto_ops packet_ops_spkt = {
2628 .family = PF_PACKET,
2629 .owner = THIS_MODULE,
2630 .release = packet_release,
2631 .bind = packet_bind_spkt,
2632 .connect = sock_no_connect,
2633 .socketpair = sock_no_socketpair,
2634 .accept = sock_no_accept,
2635 .getname = packet_getname_spkt,
2636 .poll = datagram_poll,
2637 .ioctl = packet_ioctl,
2638 .listen = sock_no_listen,
2639 .shutdown = sock_no_shutdown,
2640 .setsockopt = sock_no_setsockopt,
2641 .getsockopt = sock_no_getsockopt,
2642 .sendmsg = packet_sendmsg_spkt,
2643 .recvmsg = packet_recvmsg,
2644 .mmap = sock_no_mmap,
2645 .sendpage = sock_no_sendpage,
2646 };
2647
2648 static const struct proto_ops packet_ops = {
2649 .family = PF_PACKET,
2650 .owner = THIS_MODULE,
2651 .release = packet_release,
2652 .bind = packet_bind,
2653 .connect = sock_no_connect,
2654 .socketpair = sock_no_socketpair,
2655 .accept = sock_no_accept,
2656 .getname = packet_getname,
2657 .poll = packet_poll,
2658 .ioctl = packet_ioctl,
2659 .listen = sock_no_listen,
2660 .shutdown = sock_no_shutdown,
2661 .setsockopt = packet_setsockopt,
2662 .getsockopt = packet_getsockopt,
2663 .sendmsg = packet_sendmsg,
2664 .recvmsg = packet_recvmsg,
2665 .mmap = packet_mmap,
2666 .sendpage = sock_no_sendpage,
2667 };
2668
2669 static const struct net_proto_family packet_family_ops = {
2670 .family = PF_PACKET,
2671 .create = packet_create,
2672 .owner = THIS_MODULE,
2673 };
2674
2675 static struct notifier_block packet_netdev_notifier = {
2676 .notifier_call = packet_notifier,
2677 };
2678
2679 #ifdef CONFIG_PROC_FS
2680
2681 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2682 __acquires(RCU)
2683 {
2684 struct net *net = seq_file_net(seq);
2685
2686 rcu_read_lock();
2687 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
2688 }
2689
2690 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2691 {
2692 struct net *net = seq_file_net(seq);
2693 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
2694 }
2695
2696 static void packet_seq_stop(struct seq_file *seq, void *v)
2697 __releases(RCU)
2698 {
2699 rcu_read_unlock();
2700 }
2701
2702 static int packet_seq_show(struct seq_file *seq, void *v)
2703 {
2704 if (v == SEQ_START_TOKEN)
2705 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
2706 else {
2707 struct sock *s = sk_entry(v);
2708 const struct packet_sock *po = pkt_sk(s);
2709
2710 seq_printf(seq,
2711 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
2712 s,
2713 atomic_read(&s->sk_refcnt),
2714 s->sk_type,
2715 ntohs(po->num),
2716 po->ifindex,
2717 po->running,
2718 atomic_read(&s->sk_rmem_alloc),
2719 sock_i_uid(s),
2720 sock_i_ino(s));
2721 }
2722
2723 return 0;
2724 }
2725
2726 static const struct seq_operations packet_seq_ops = {
2727 .start = packet_seq_start,
2728 .next = packet_seq_next,
2729 .stop = packet_seq_stop,
2730 .show = packet_seq_show,
2731 };
2732
2733 static int packet_seq_open(struct inode *inode, struct file *file)
2734 {
2735 return seq_open_net(inode, file, &packet_seq_ops,
2736 sizeof(struct seq_net_private));
2737 }
2738
2739 static const struct file_operations packet_seq_fops = {
2740 .owner = THIS_MODULE,
2741 .open = packet_seq_open,
2742 .read = seq_read,
2743 .llseek = seq_lseek,
2744 .release = seq_release_net,
2745 };
2746
2747 #endif
2748
2749 static int __net_init packet_net_init(struct net *net)
2750 {
2751 spin_lock_init(&net->packet.sklist_lock);
2752 INIT_HLIST_HEAD(&net->packet.sklist);
2753
2754 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2755 return -ENOMEM;
2756
2757 return 0;
2758 }
2759
2760 static void __net_exit packet_net_exit(struct net *net)
2761 {
2762 proc_net_remove(net, "packet");
2763 }
2764
2765 static struct pernet_operations packet_net_ops = {
2766 .init = packet_net_init,
2767 .exit = packet_net_exit,
2768 };
2769
2770
2771 static void __exit packet_exit(void)
2772 {
2773 unregister_netdevice_notifier(&packet_netdev_notifier);
2774 unregister_pernet_subsys(&packet_net_ops);
2775 sock_unregister(PF_PACKET);
2776 proto_unregister(&packet_proto);
2777 }
2778
2779 static int __init packet_init(void)
2780 {
2781 int rc = proto_register(&packet_proto, 0);
2782
2783 if (rc != 0)
2784 goto out;
2785
2786 sock_register(&packet_family_ops);
2787 register_pernet_subsys(&packet_net_ops);
2788 register_netdevice_notifier(&packet_netdev_notifier);
2789 out:
2790 return rc;
2791 }
2792
2793 module_init(packet_init);
2794 module_exit(packet_exit);
2795 MODULE_LICENSE("GPL");
2796 MODULE_ALIAS_NETPROTO(PF_PACKET);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree