?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
packet: Add helpers to register/unregister ->prot_hook
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / packet / af_packet.c
blobbb281bf330aaf30342fde4829fe2d710e1aad507
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 struct packet_sock *pkt_sk(struct sock *sk)
226 {
227 return (struct packet_sock *)sk;
228 }
229
230 /* register_prot_hook must be invoked with the po->bind_lock held,
231 * or from a context in which asynchronous accesses to the packet
232 * socket is not possible (packet_create()).
233 */
234 static void register_prot_hook(struct sock *sk)
235 {
236 struct packet_sock *po = pkt_sk(sk);
237 if (!po->running) {
238 dev_add_pack(&po->prot_hook);
239 sock_hold(sk);
240 po->running = 1;
241 }
242 }
243
244 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
245 * held. If the sync parameter is true, we will temporarily drop
246 * the po->bind_lock and do a synchronize_net to make sure no
247 * asynchronous packet processing paths still refer to the elements
248 * of po->prot_hook. If the sync parameter is false, it is the
249 * callers responsibility to take care of this.
250 */
251 static void __unregister_prot_hook(struct sock *sk, bool sync)
252 {
253 struct packet_sock *po = pkt_sk(sk);
254
255 po->running = 0;
256 __dev_remove_pack(&po->prot_hook);
257 __sock_put(sk);
258
259 if (sync) {
260 spin_unlock(&po->bind_lock);
261 synchronize_net();
262 spin_lock(&po->bind_lock);
263 }
264 }
265
266 static void unregister_prot_hook(struct sock *sk, bool sync)
267 {
268 struct packet_sock *po = pkt_sk(sk);
269
270 if (po->running)
271 __unregister_prot_hook(sk, sync);
272 }
273
274 static inline __pure struct page *pgv_to_page(void *addr)
275 {
276 if (is_vmalloc_addr(addr))
277 return vmalloc_to_page(addr);
278 return virt_to_page(addr);
279 }
280
281 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
282 {
283 union {
284 struct tpacket_hdr *h1;
285 struct tpacket2_hdr *h2;
286 void *raw;
287 } h;
288
289 h.raw = frame;
290 switch (po->tp_version) {
291 case TPACKET_V1:
292 h.h1->tp_status = status;
293 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
294 break;
295 case TPACKET_V2:
296 h.h2->tp_status = status;
297 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
298 break;
299 default:
300 pr_err("TPACKET version not supported\n");
301 BUG();
302 }
303
304 smp_wmb();
305 }
306
307 static int __packet_get_status(struct packet_sock *po, void *frame)
308 {
309 union {
310 struct tpacket_hdr *h1;
311 struct tpacket2_hdr *h2;
312 void *raw;
313 } h;
314
315 smp_rmb();
316
317 h.raw = frame;
318 switch (po->tp_version) {
319 case TPACKET_V1:
320 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
321 return h.h1->tp_status;
322 case TPACKET_V2:
323 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
324 return h.h2->tp_status;
325 default:
326 pr_err("TPACKET version not supported\n");
327 BUG();
328 return 0;
329 }
330 }
331
332 static void *packet_lookup_frame(struct packet_sock *po,
333 struct packet_ring_buffer *rb,
334 unsigned int position,
335 int status)
336 {
337 unsigned int pg_vec_pos, frame_offset;
338 union {
339 struct tpacket_hdr *h1;
340 struct tpacket2_hdr *h2;
341 void *raw;
342 } h;
343
344 pg_vec_pos = position / rb->frames_per_block;
345 frame_offset = position % rb->frames_per_block;
346
347 h.raw = rb->pg_vec[pg_vec_pos].buffer +
348 (frame_offset * rb->frame_size);
349
350 if (status != __packet_get_status(po, h.raw))
351 return NULL;
352
353 return h.raw;
354 }
355
356 static inline void *packet_current_frame(struct packet_sock *po,
357 struct packet_ring_buffer *rb,
358 int status)
359 {
360 return packet_lookup_frame(po, rb, rb->head, status);
361 }
362
363 static inline void *packet_previous_frame(struct packet_sock *po,
364 struct packet_ring_buffer *rb,
365 int status)
366 {
367 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
368 return packet_lookup_frame(po, rb, previous, status);
369 }
370
371 static inline void packet_increment_head(struct packet_ring_buffer *buff)
372 {
373 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
374 }
375
376 static void packet_sock_destruct(struct sock *sk)
377 {
378 skb_queue_purge(&sk->sk_error_queue);
379
380 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
381 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
382
383 if (!sock_flag(sk, SOCK_DEAD)) {
384 pr_err("Attempt to release alive packet socket: %p\n", sk);
385 return;
386 }
387
388 sk_refcnt_debug_dec(sk);
389 }
390
391
392 static const struct proto_ops packet_ops;
393
394 static const struct proto_ops packet_ops_spkt;
395
396 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
397 struct packet_type *pt, struct net_device *orig_dev)
398 {
399 struct sock *sk;
400 struct sockaddr_pkt *spkt;
401
402 /*
403 * When we registered the protocol we saved the socket in the data
404 * field for just this event.
405 */
406
407 sk = pt->af_packet_priv;
408
409 /*
410 * Yank back the headers [hope the device set this
411 * right or kerboom...]
412 *
413 * Incoming packets have ll header pulled,
414 * push it back.
415 *
416 * For outgoing ones skb->data == skb_mac_header(skb)
417 * so that this procedure is noop.
418 */
419
420 if (skb->pkt_type == PACKET_LOOPBACK)
421 goto out;
422
423 if (!net_eq(dev_net(dev), sock_net(sk)))
424 goto out;
425
426 skb = skb_share_check(skb, GFP_ATOMIC);
427 if (skb == NULL)
428 goto oom;
429
430 /* drop any routing info */
431 skb_dst_drop(skb);
432
433 /* drop conntrack reference */
434 nf_reset(skb);
435
436 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
437
438 skb_push(skb, skb->data - skb_mac_header(skb));
439
440 /*
441 * The SOCK_PACKET socket receives _all_ frames.
442 */
443
444 spkt->spkt_family = dev->type;
445 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
446 spkt->spkt_protocol = skb->protocol;
447
448 /*
449 * Charge the memory to the socket. This is done specifically
450 * to prevent sockets using all the memory up.
451 */
452
453 if (sock_queue_rcv_skb(sk, skb) == 0)
454 return 0;
455
456 out:
457 kfree_skb(skb);
458 oom:
459 return 0;
460 }
461
462
463 /*
464 * Output a raw packet to a device layer. This bypasses all the other
465 * protocol layers and you must therefore supply it with a complete frame
466 */
467
468 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
469 struct msghdr *msg, size_t len)
470 {
471 struct sock *sk = sock->sk;
472 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
473 struct sk_buff *skb = NULL;
474 struct net_device *dev;
475 __be16 proto = 0;
476 int err;
477
478 /*
479 * Get and verify the address.
480 */
481
482 if (saddr) {
483 if (msg->msg_namelen < sizeof(struct sockaddr))
484 return -EINVAL;
485 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
486 proto = saddr->spkt_protocol;
487 } else
488 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
489
490 /*
491 * Find the device first to size check it
492 */
493
494 saddr->spkt_device[13] = 0;
495 retry:
496 rcu_read_lock();
497 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
498 err = -ENODEV;
499 if (dev == NULL)
500 goto out_unlock;
501
502 err = -ENETDOWN;
503 if (!(dev->flags & IFF_UP))
504 goto out_unlock;
505
506 /*
507 * You may not queue a frame bigger than the mtu. This is the lowest level
508 * raw protocol and you must do your own fragmentation at this level.
509 */
510
511 err = -EMSGSIZE;
512 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
513 goto out_unlock;
514
515 if (!skb) {
516 size_t reserved = LL_RESERVED_SPACE(dev);
517 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
518
519 rcu_read_unlock();
520 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
521 if (skb == NULL)
522 return -ENOBUFS;
523 /* FIXME: Save some space for broken drivers that write a hard
524 * header at transmission time by themselves. PPP is the notable
525 * one here. This should really be fixed at the driver level.
526 */
527 skb_reserve(skb, reserved);
528 skb_reset_network_header(skb);
529
530 /* Try to align data part correctly */
531 if (hhlen) {
532 skb->data -= hhlen;
533 skb->tail -= hhlen;
534 if (len < hhlen)
535 skb_reset_network_header(skb);
536 }
537 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
538 if (err)
539 goto out_free;
540 goto retry;
541 }
542
543 if (len > (dev->mtu + dev->hard_header_len)) {
544 /* Earlier code assumed this would be a VLAN pkt,
545 * double-check this now that we have the actual
546 * packet in hand.
547 */
548 struct ethhdr *ehdr;
549 skb_reset_mac_header(skb);
550 ehdr = eth_hdr(skb);
551 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
552 err = -EMSGSIZE;
553 goto out_unlock;
554 }
555 }
556
557 skb->protocol = proto;
558 skb->dev = dev;
559 skb->priority = sk->sk_priority;
560 skb->mark = sk->sk_mark;
561 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
562 if (err < 0)
563 goto out_unlock;
564
565 dev_queue_xmit(skb);
566 rcu_read_unlock();
567 return len;
568
569 out_unlock:
570 rcu_read_unlock();
571 out_free:
572 kfree_skb(skb);
573 return err;
574 }
575
576 static inline unsigned int run_filter(const struct sk_buff *skb,
577 const struct sock *sk,
578 unsigned int res)
579 {
580 struct sk_filter *filter;
581
582 rcu_read_lock();
583 filter = rcu_dereference(sk->sk_filter);
584 if (filter != NULL)
585 res = SK_RUN_FILTER(filter, skb);
586 rcu_read_unlock();
587
588 return res;
589 }
590
591 /*
592 * This function makes lazy skb cloning in hope that most of packets
593 * are discarded by BPF.
594 *
595 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
596 * and skb->cb are mangled. It works because (and until) packets
597 * falling here are owned by current CPU. Output packets are cloned
598 * by dev_queue_xmit_nit(), input packets are processed by net_bh
599 * sequencially, so that if we return skb to original state on exit,
600 * we will not harm anyone.
601 */
602
603 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
604 struct packet_type *pt, struct net_device *orig_dev)
605 {
606 struct sock *sk;
607 struct sockaddr_ll *sll;
608 struct packet_sock *po;
609 u8 *skb_head = skb->data;
610 int skb_len = skb->len;
611 unsigned int snaplen, res;
612
613 if (skb->pkt_type == PACKET_LOOPBACK)
614 goto drop;
615
616 sk = pt->af_packet_priv;
617 po = pkt_sk(sk);
618
619 if (!net_eq(dev_net(dev), sock_net(sk)))
620 goto drop;
621
622 skb->dev = dev;
623
624 if (dev->header_ops) {
625 /* The device has an explicit notion of ll header,
626 * exported to higher levels.
627 *
628 * Otherwise, the device hides details of its frame
629 * structure, so that corresponding packet head is
630 * never delivered to user.
631 */
632 if (sk->sk_type != SOCK_DGRAM)
633 skb_push(skb, skb->data - skb_mac_header(skb));
634 else if (skb->pkt_type == PACKET_OUTGOING) {
635 /* Special case: outgoing packets have ll header at head */
636 skb_pull(skb, skb_network_offset(skb));
637 }
638 }
639
640 snaplen = skb->len;
641
642 res = run_filter(skb, sk, snaplen);
643 if (!res)
644 goto drop_n_restore;
645 if (snaplen > res)
646 snaplen = res;
647
648 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
649 (unsigned)sk->sk_rcvbuf)
650 goto drop_n_acct;
651
652 if (skb_shared(skb)) {
653 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
654 if (nskb == NULL)
655 goto drop_n_acct;
656
657 if (skb_head != skb->data) {
658 skb->data = skb_head;
659 skb->len = skb_len;
660 }
661 kfree_skb(skb);
662 skb = nskb;
663 }
664
665 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
666 sizeof(skb->cb));
667
668 sll = &PACKET_SKB_CB(skb)->sa.ll;
669 sll->sll_family = AF_PACKET;
670 sll->sll_hatype = dev->type;
671 sll->sll_protocol = skb->protocol;
672 sll->sll_pkttype = skb->pkt_type;
673 if (unlikely(po->origdev))
674 sll->sll_ifindex = orig_dev->ifindex;
675 else
676 sll->sll_ifindex = dev->ifindex;
677
678 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
679
680 PACKET_SKB_CB(skb)->origlen = skb->len;
681
682 if (pskb_trim(skb, snaplen))
683 goto drop_n_acct;
684
685 skb_set_owner_r(skb, sk);
686 skb->dev = NULL;
687 skb_dst_drop(skb);
688
689 /* drop conntrack reference */
690 nf_reset(skb);
691
692 spin_lock(&sk->sk_receive_queue.lock);
693 po->stats.tp_packets++;
694 skb->dropcount = atomic_read(&sk->sk_drops);
695 __skb_queue_tail(&sk->sk_receive_queue, skb);
696 spin_unlock(&sk->sk_receive_queue.lock);
697 sk->sk_data_ready(sk, skb->len);
698 return 0;
699
700 drop_n_acct:
701 po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
702
703 drop_n_restore:
704 if (skb_head != skb->data && skb_shared(skb)) {
705 skb->data = skb_head;
706 skb->len = skb_len;
707 }
708 drop:
709 consume_skb(skb);
710 return 0;
711 }
712
713 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
714 struct packet_type *pt, struct net_device *orig_dev)
715 {
716 struct sock *sk;
717 struct packet_sock *po;
718 struct sockaddr_ll *sll;
719 union {
720 struct tpacket_hdr *h1;
721 struct tpacket2_hdr *h2;
722 void *raw;
723 } h;
724 u8 *skb_head = skb->data;
725 int skb_len = skb->len;
726 unsigned int snaplen, res;
727 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
728 unsigned short macoff, netoff, hdrlen;
729 struct sk_buff *copy_skb = NULL;
730 struct timeval tv;
731 struct timespec ts;
732 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
733
734 if (skb->pkt_type == PACKET_LOOPBACK)
735 goto drop;
736
737 sk = pt->af_packet_priv;
738 po = pkt_sk(sk);
739
740 if (!net_eq(dev_net(dev), sock_net(sk)))
741 goto drop;
742
743 if (dev->header_ops) {
744 if (sk->sk_type != SOCK_DGRAM)
745 skb_push(skb, skb->data - skb_mac_header(skb));
746 else if (skb->pkt_type == PACKET_OUTGOING) {
747 /* Special case: outgoing packets have ll header at head */
748 skb_pull(skb, skb_network_offset(skb));
749 }
750 }
751
752 if (skb->ip_summed == CHECKSUM_PARTIAL)
753 status |= TP_STATUS_CSUMNOTREADY;
754
755 snaplen = skb->len;
756
757 res = run_filter(skb, sk, snaplen);
758 if (!res)
759 goto drop_n_restore;
760 if (snaplen > res)
761 snaplen = res;
762
763 if (sk->sk_type == SOCK_DGRAM) {
764 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
765 po->tp_reserve;
766 } else {
767 unsigned maclen = skb_network_offset(skb);
768 netoff = TPACKET_ALIGN(po->tp_hdrlen +
769 (maclen < 16 ? 16 : maclen)) +
770 po->tp_reserve;
771 macoff = netoff - maclen;
772 }
773
774 if (macoff + snaplen > po->rx_ring.frame_size) {
775 if (po->copy_thresh &&
776 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
777 (unsigned)sk->sk_rcvbuf) {
778 if (skb_shared(skb)) {
779 copy_skb = skb_clone(skb, GFP_ATOMIC);
780 } else {
781 copy_skb = skb_get(skb);
782 skb_head = skb->data;
783 }
784 if (copy_skb)
785 skb_set_owner_r(copy_skb, sk);
786 }
787 snaplen = po->rx_ring.frame_size - macoff;
788 if ((int)snaplen < 0)
789 snaplen = 0;
790 }
791
792 spin_lock(&sk->sk_receive_queue.lock);
793 h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
794 if (!h.raw)
795 goto ring_is_full;
796 packet_increment_head(&po->rx_ring);
797 po->stats.tp_packets++;
798 if (copy_skb) {
799 status |= TP_STATUS_COPY;
800 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
801 }
802 if (!po->stats.tp_drops)
803 status &= ~TP_STATUS_LOSING;
804 spin_unlock(&sk->sk_receive_queue.lock);
805
806 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
807
808 switch (po->tp_version) {
809 case TPACKET_V1:
810 h.h1->tp_len = skb->len;
811 h.h1->tp_snaplen = snaplen;
812 h.h1->tp_mac = macoff;
813 h.h1->tp_net = netoff;
814 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
815 && shhwtstamps->syststamp.tv64)
816 tv = ktime_to_timeval(shhwtstamps->syststamp);
817 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
818 && shhwtstamps->hwtstamp.tv64)
819 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
820 else if (skb->tstamp.tv64)
821 tv = ktime_to_timeval(skb->tstamp);
822 else
823 do_gettimeofday(&tv);
824 h.h1->tp_sec = tv.tv_sec;
825 h.h1->tp_usec = tv.tv_usec;
826 hdrlen = sizeof(*h.h1);
827 break;
828 case TPACKET_V2:
829 h.h2->tp_len = skb->len;
830 h.h2->tp_snaplen = snaplen;
831 h.h2->tp_mac = macoff;
832 h.h2->tp_net = netoff;
833 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
834 && shhwtstamps->syststamp.tv64)
835 ts = ktime_to_timespec(shhwtstamps->syststamp);
836 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
837 && shhwtstamps->hwtstamp.tv64)
838 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
839 else if (skb->tstamp.tv64)
840 ts = ktime_to_timespec(skb->tstamp);
841 else
842 getnstimeofday(&ts);
843 h.h2->tp_sec = ts.tv_sec;
844 h.h2->tp_nsec = ts.tv_nsec;
845 if (vlan_tx_tag_present(skb)) {
846 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
847 status |= TP_STATUS_VLAN_VALID;
848 } else {
849 h.h2->tp_vlan_tci = 0;
850 }
851 h.h2->tp_padding = 0;
852 hdrlen = sizeof(*h.h2);
853 break;
854 default:
855 BUG();
856 }
857
858 sll = h.raw + TPACKET_ALIGN(hdrlen);
859 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
860 sll->sll_family = AF_PACKET;
861 sll->sll_hatype = dev->type;
862 sll->sll_protocol = skb->protocol;
863 sll->sll_pkttype = skb->pkt_type;
864 if (unlikely(po->origdev))
865 sll->sll_ifindex = orig_dev->ifindex;
866 else
867 sll->sll_ifindex = dev->ifindex;
868
869 __packet_set_status(po, h.raw, status);
870 smp_mb();
871 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
872 {
873 u8 *start, *end;
874
875 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw + macoff + snaplen);
876 for (start = h.raw; start < end; start += PAGE_SIZE)
877 flush_dcache_page(pgv_to_page(start));
878 }
879 #endif
880
881 sk->sk_data_ready(sk, 0);
882
883 drop_n_restore:
884 if (skb_head != skb->data && skb_shared(skb)) {
885 skb->data = skb_head;
886 skb->len = skb_len;
887 }
888 drop:
889 kfree_skb(skb);
890 return 0;
891
892 ring_is_full:
893 po->stats.tp_drops++;
894 spin_unlock(&sk->sk_receive_queue.lock);
895
896 sk->sk_data_ready(sk, 0);
897 kfree_skb(copy_skb);
898 goto drop_n_restore;
899 }
900
901 static void tpacket_destruct_skb(struct sk_buff *skb)
902 {
903 struct packet_sock *po = pkt_sk(skb->sk);
904 void *ph;
905
906 BUG_ON(skb == NULL);
907
908 if (likely(po->tx_ring.pg_vec)) {
909 ph = skb_shinfo(skb)->destructor_arg;
910 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
911 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
912 atomic_dec(&po->tx_ring.pending);
913 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
914 }
915
916 sock_wfree(skb);
917 }
918
919 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
920 void *frame, struct net_device *dev, int size_max,
921 __be16 proto, unsigned char *addr)
922 {
923 union {
924 struct tpacket_hdr *h1;
925 struct tpacket2_hdr *h2;
926 void *raw;
927 } ph;
928 int to_write, offset, len, tp_len, nr_frags, len_max;
929 struct socket *sock = po->sk.sk_socket;
930 struct page *page;
931 void *data;
932 int err;
933
934 ph.raw = frame;
935
936 skb->protocol = proto;
937 skb->dev = dev;
938 skb->priority = po->sk.sk_priority;
939 skb->mark = po->sk.sk_mark;
940 skb_shinfo(skb)->destructor_arg = ph.raw;
941
942 switch (po->tp_version) {
943 case TPACKET_V2:
944 tp_len = ph.h2->tp_len;
945 break;
946 default:
947 tp_len = ph.h1->tp_len;
948 break;
949 }
950 if (unlikely(tp_len > size_max)) {
951 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
952 return -EMSGSIZE;
953 }
954
955 skb_reserve(skb, LL_RESERVED_SPACE(dev));
956 skb_reset_network_header(skb);
957
958 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
959 to_write = tp_len;
960
961 if (sock->type == SOCK_DGRAM) {
962 err = dev_hard_header(skb, dev, ntohs(proto), addr,
963 NULL, tp_len);
964 if (unlikely(err < 0))
965 return -EINVAL;
966 } else if (dev->hard_header_len) {
967 /* net device doesn't like empty head */
968 if (unlikely(tp_len <= dev->hard_header_len)) {
969 pr_err("packet size is too short (%d < %d)\n",
970 tp_len, dev->hard_header_len);
971 return -EINVAL;
972 }
973
974 skb_push(skb, dev->hard_header_len);
975 err = skb_store_bits(skb, 0, data,
976 dev->hard_header_len);
977 if (unlikely(err))
978 return err;
979
980 data += dev->hard_header_len;
981 to_write -= dev->hard_header_len;
982 }
983
984 err = -EFAULT;
985 offset = offset_in_page(data);
986 len_max = PAGE_SIZE - offset;
987 len = ((to_write > len_max) ? len_max : to_write);
988
989 skb->data_len = to_write;
990 skb->len += to_write;
991 skb->truesize += to_write;
992 atomic_add(to_write, &po->sk.sk_wmem_alloc);
993
994 while (likely(to_write)) {
995 nr_frags = skb_shinfo(skb)->nr_frags;
996
997 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
998 pr_err("Packet exceed the number of skb frags(%lu)\n",
999 MAX_SKB_FRAGS);
1000 return -EFAULT;
1001 }
1002
1003 page = pgv_to_page(data);
1004 data += len;
1005 flush_dcache_page(page);
1006 get_page(page);
1007 skb_fill_page_desc(skb, nr_frags, page, offset, len);
1008 to_write -= len;
1009 offset = 0;
1010 len_max = PAGE_SIZE;
1011 len = ((to_write > len_max) ? len_max : to_write);
1012 }
1013
1014 return tp_len;
1015 }
1016
1017 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
1018 {
1019 struct sk_buff *skb;
1020 struct net_device *dev;
1021 __be16 proto;
1022 bool need_rls_dev = false;
1023 int err, reserve = 0;
1024 void *ph;
1025 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1026 int tp_len, size_max;
1027 unsigned char *addr;
1028 int len_sum = 0;
1029 int status = 0;
1030
1031 mutex_lock(&po->pg_vec_lock);
1032
1033 err = -EBUSY;
1034 if (saddr == NULL) {
1035 dev = po->prot_hook.dev;
1036 proto = po->num;
1037 addr = NULL;
1038 } else {
1039 err = -EINVAL;
1040 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1041 goto out;
1042 if (msg->msg_namelen < (saddr->sll_halen
1043 + offsetof(struct sockaddr_ll,
1044 sll_addr)))
1045 goto out;
1046 proto = saddr->sll_protocol;
1047 addr = saddr->sll_addr;
1048 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
1049 need_rls_dev = true;
1050 }
1051
1052 err = -ENXIO;
1053 if (unlikely(dev == NULL))
1054 goto out;
1055
1056 reserve = dev->hard_header_len;
1057
1058 err = -ENETDOWN;
1059 if (unlikely(!(dev->flags & IFF_UP)))
1060 goto out_put;
1061
1062 size_max = po->tx_ring.frame_size
1063 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1064
1065 if (size_max > dev->mtu + reserve)
1066 size_max = dev->mtu + reserve;
1067
1068 do {
1069 ph = packet_current_frame(po, &po->tx_ring,
1070 TP_STATUS_SEND_REQUEST);
1071
1072 if (unlikely(ph == NULL)) {
1073 schedule();
1074 continue;
1075 }
1076
1077 status = TP_STATUS_SEND_REQUEST;
1078 skb = sock_alloc_send_skb(&po->sk,
1079 LL_ALLOCATED_SPACE(dev)
1080 + sizeof(struct sockaddr_ll),
1081 0, &err);
1082
1083 if (unlikely(skb == NULL))
1084 goto out_status;
1085
1086 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1087 addr);
1088
1089 if (unlikely(tp_len < 0)) {
1090 if (po->tp_loss) {
1091 __packet_set_status(po, ph,
1092 TP_STATUS_AVAILABLE);
1093 packet_increment_head(&po->tx_ring);
1094 kfree_skb(skb);
1095 continue;
1096 } else {
1097 status = TP_STATUS_WRONG_FORMAT;
1098 err = tp_len;
1099 goto out_status;
1100 }
1101 }
1102
1103 skb->destructor = tpacket_destruct_skb;
1104 __packet_set_status(po, ph, TP_STATUS_SENDING);
1105 atomic_inc(&po->tx_ring.pending);
1106
1107 status = TP_STATUS_SEND_REQUEST;
1108 err = dev_queue_xmit(skb);
1109 if (unlikely(err > 0)) {
1110 err = net_xmit_errno(err);
1111 if (err && __packet_get_status(po, ph) ==
1112 TP_STATUS_AVAILABLE) {
1113 /* skb was destructed already */
1114 skb = NULL;
1115 goto out_status;
1116 }
1117 /*
1118 * skb was dropped but not destructed yet;
1119 * let's treat it like congestion or err < 0
1120 */
1121 err = 0;
1122 }
1123 packet_increment_head(&po->tx_ring);
1124 len_sum += tp_len;
1125 } while (likely((ph != NULL) ||
1126 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1127 (atomic_read(&po->tx_ring.pending))))
1128 );
1129
1130 err = len_sum;
1131 goto out_put;
1132
1133 out_status:
1134 __packet_set_status(po, ph, status);
1135 kfree_skb(skb);
1136 out_put:
1137 if (need_rls_dev)
1138 dev_put(dev);
1139 out:
1140 mutex_unlock(&po->pg_vec_lock);
1141 return err;
1142 }
1143
1144 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1145 size_t reserve, size_t len,
1146 size_t linear, int noblock,
1147 int *err)
1148 {
1149 struct sk_buff *skb;
1150
1151 /* Under a page? Don't bother with paged skb. */
1152 if (prepad + len < PAGE_SIZE || !linear)
1153 linear = len;
1154
1155 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1156 err);
1157 if (!skb)
1158 return NULL;
1159
1160 skb_reserve(skb, reserve);
1161 skb_put(skb, linear);
1162 skb->data_len = len - linear;
1163 skb->len += len - linear;
1164
1165 return skb;
1166 }
1167
1168 static int packet_snd(struct socket *sock,
1169 struct msghdr *msg, size_t len)
1170 {
1171 struct sock *sk = sock->sk;
1172 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1173 struct sk_buff *skb;
1174 struct net_device *dev;
1175 __be16 proto;
1176 bool need_rls_dev = false;
1177 unsigned char *addr;
1178 int err, reserve = 0;
1179 struct virtio_net_hdr vnet_hdr = { 0 };
1180 int offset = 0;
1181 int vnet_hdr_len;
1182 struct packet_sock *po = pkt_sk(sk);
1183 unsigned short gso_type = 0;
1184
1185 /*
1186 * Get and verify the address.
1187 */
1188
1189 if (saddr == NULL) {
1190 dev = po->prot_hook.dev;
1191 proto = po->num;
1192 addr = NULL;
1193 } else {
1194 err = -EINVAL;
1195 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1196 goto out;
1197 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1198 goto out;
1199 proto = saddr->sll_protocol;
1200 addr = saddr->sll_addr;
1201 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
1202 need_rls_dev = true;
1203 }
1204
1205 err = -ENXIO;
1206 if (dev == NULL)
1207 goto out_unlock;
1208 if (sock->type == SOCK_RAW)
1209 reserve = dev->hard_header_len;
1210
1211 err = -ENETDOWN;
1212 if (!(dev->flags & IFF_UP))
1213 goto out_unlock;
1214
1215 if (po->has_vnet_hdr) {
1216 vnet_hdr_len = sizeof(vnet_hdr);
1217
1218 err = -EINVAL;
1219 if (len < vnet_hdr_len)
1220 goto out_unlock;
1221
1222 len -= vnet_hdr_len;
1223
1224 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1225 vnet_hdr_len);
1226 if (err < 0)
1227 goto out_unlock;
1228
1229 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1230 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1231 vnet_hdr.hdr_len))
1232 vnet_hdr.hdr_len = vnet_hdr.csum_start +
1233 vnet_hdr.csum_offset + 2;
1234
1235 err = -EINVAL;
1236 if (vnet_hdr.hdr_len > len)
1237 goto out_unlock;
1238
1239 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1240 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1241 case VIRTIO_NET_HDR_GSO_TCPV4:
1242 gso_type = SKB_GSO_TCPV4;
1243 break;
1244 case VIRTIO_NET_HDR_GSO_TCPV6:
1245 gso_type = SKB_GSO_TCPV6;
1246 break;
1247 case VIRTIO_NET_HDR_GSO_UDP:
1248 gso_type = SKB_GSO_UDP;
1249 break;
1250 default:
1251 goto out_unlock;
1252 }
1253
1254 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1255 gso_type |= SKB_GSO_TCP_ECN;
1256
1257 if (vnet_hdr.gso_size == 0)
1258 goto out_unlock;
1259
1260 }
1261 }
1262
1263 err = -EMSGSIZE;
1264 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
1265 goto out_unlock;
1266
1267 err = -ENOBUFS;
1268 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1269 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1270 msg->msg_flags & MSG_DONTWAIT, &err);
1271 if (skb == NULL)
1272 goto out_unlock;
1273
1274 skb_set_network_header(skb, reserve);
1275
1276 err = -EINVAL;
1277 if (sock->type == SOCK_DGRAM &&
1278 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1279 goto out_free;
1280
1281 /* Returns -EFAULT on error */
1282 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1283 if (err)
1284 goto out_free;
1285 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1286 if (err < 0)
1287 goto out_free;
1288
1289 if (!gso_type && (len > dev->mtu + reserve)) {
1290 /* Earlier code assumed this would be a VLAN pkt,
1291 * double-check this now that we have the actual
1292 * packet in hand.
1293 */
1294 struct ethhdr *ehdr;
1295 skb_reset_mac_header(skb);
1296 ehdr = eth_hdr(skb);
1297 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1298 err = -EMSGSIZE;
1299 goto out_free;
1300 }
1301 }
1302
1303 skb->protocol = proto;
1304 skb->dev = dev;
1305 skb->priority = sk->sk_priority;
1306 skb->mark = sk->sk_mark;
1307
1308 if (po->has_vnet_hdr) {
1309 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1310 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1311 vnet_hdr.csum_offset)) {
1312 err = -EINVAL;
1313 goto out_free;
1314 }
1315 }
1316
1317 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1318 skb_shinfo(skb)->gso_type = gso_type;
1319
1320 /* Header must be checked, and gso_segs computed. */
1321 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1322 skb_shinfo(skb)->gso_segs = 0;
1323
1324 len += vnet_hdr_len;
1325 }
1326
1327 /*
1328 * Now send it
1329 */
1330
1331 err = dev_queue_xmit(skb);
1332 if (err > 0 && (err = net_xmit_errno(err)) != 0)
1333 goto out_unlock;
1334
1335 if (need_rls_dev)
1336 dev_put(dev);
1337
1338 return len;
1339
1340 out_free:
1341 kfree_skb(skb);
1342 out_unlock:
1343 if (dev && need_rls_dev)
1344 dev_put(dev);
1345 out:
1346 return err;
1347 }
1348
1349 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1350 struct msghdr *msg, size_t len)
1351 {
1352 struct sock *sk = sock->sk;
1353 struct packet_sock *po = pkt_sk(sk);
1354 if (po->tx_ring.pg_vec)
1355 return tpacket_snd(po, msg);
1356 else
1357 return packet_snd(sock, msg, len);
1358 }
1359
1360 /*
1361 * Close a PACKET socket. This is fairly simple. We immediately go
1362 * to 'closed' state and remove our protocol entry in the device list.
1363 */
1364
1365 static int packet_release(struct socket *sock)
1366 {
1367 struct sock *sk = sock->sk;
1368 struct packet_sock *po;
1369 struct net *net;
1370 struct tpacket_req req;
1371
1372 if (!sk)
1373 return 0;
1374
1375 net = sock_net(sk);
1376 po = pkt_sk(sk);
1377
1378 spin_lock_bh(&net->packet.sklist_lock);
1379 sk_del_node_init_rcu(sk);
1380 sock_prot_inuse_add(net, sk->sk_prot, -1);
1381 spin_unlock_bh(&net->packet.sklist_lock);
1382
1383 spin_lock(&po->bind_lock);
1384 unregister_prot_hook(sk, false);
1385 if (po->prot_hook.dev) {
1386 dev_put(po->prot_hook.dev);
1387 po->prot_hook.dev = NULL;
1388 }
1389 spin_unlock(&po->bind_lock);
1390
1391 packet_flush_mclist(sk);
1392
1393 memset(&req, 0, sizeof(req));
1394
1395 if (po->rx_ring.pg_vec)
1396 packet_set_ring(sk, &req, 1, 0);
1397
1398 if (po->tx_ring.pg_vec)
1399 packet_set_ring(sk, &req, 1, 1);
1400
1401 synchronize_net();
1402 /*
1403 * Now the socket is dead. No more input will appear.
1404 */
1405 sock_orphan(sk);
1406 sock->sk = NULL;
1407
1408 /* Purge queues */
1409
1410 skb_queue_purge(&sk->sk_receive_queue);
1411 sk_refcnt_debug_release(sk);
1412
1413 sock_put(sk);
1414 return 0;
1415 }
1416
1417 /*
1418 * Attach a packet hook.
1419 */
1420
1421 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1422 {
1423 struct packet_sock *po = pkt_sk(sk);
1424 /*
1425 * Detach an existing hook if present.
1426 */
1427
1428 lock_sock(sk);
1429
1430 spin_lock(&po->bind_lock);
1431 unregister_prot_hook(sk, true);
1432 po->num = protocol;
1433 po->prot_hook.type = protocol;
1434 if (po->prot_hook.dev)
1435 dev_put(po->prot_hook.dev);
1436 po->prot_hook.dev = dev;
1437
1438 po->ifindex = dev ? dev->ifindex : 0;
1439
1440 if (protocol == 0)
1441 goto out_unlock;
1442
1443 if (!dev || (dev->flags & IFF_UP)) {
1444 register_prot_hook(sk);
1445 } else {
1446 sk->sk_err = ENETDOWN;
1447 if (!sock_flag(sk, SOCK_DEAD))
1448 sk->sk_error_report(sk);
1449 }
1450
1451 out_unlock:
1452 spin_unlock(&po->bind_lock);
1453 release_sock(sk);
1454 return 0;
1455 }
1456
1457 /*
1458 * Bind a packet socket to a device
1459 */
1460
1461 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1462 int addr_len)
1463 {
1464 struct sock *sk = sock->sk;
1465 char name[15];
1466 struct net_device *dev;
1467 int err = -ENODEV;
1468
1469 /*
1470 * Check legality
1471 */
1472
1473 if (addr_len != sizeof(struct sockaddr))
1474 return -EINVAL;
1475 strlcpy(name, uaddr->sa_data, sizeof(name));
1476
1477 dev = dev_get_by_name(sock_net(sk), name);
1478 if (dev)
1479 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1480 return err;
1481 }
1482
1483 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1484 {
1485 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1486 struct sock *sk = sock->sk;
1487 struct net_device *dev = NULL;
1488 int err;
1489
1490
1491 /*
1492 * Check legality
1493 */
1494
1495 if (addr_len < sizeof(struct sockaddr_ll))
1496 return -EINVAL;
1497 if (sll->sll_family != AF_PACKET)
1498 return -EINVAL;
1499
1500 if (sll->sll_ifindex) {
1501 err = -ENODEV;
1502 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1503 if (dev == NULL)
1504 goto out;
1505 }
1506 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1507
1508 out:
1509 return err;
1510 }
1511
1512 static struct proto packet_proto = {
1513 .name = "PACKET",
1514 .owner = THIS_MODULE,
1515 .obj_size = sizeof(struct packet_sock),
1516 };
1517
1518 /*
1519 * Create a packet of type SOCK_PACKET.
1520 */
1521
1522 static int packet_create(struct net *net, struct socket *sock, int protocol,
1523 int kern)
1524 {
1525 struct sock *sk;
1526 struct packet_sock *po;
1527 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1528 int err;
1529
1530 if (!capable(CAP_NET_RAW))
1531 return -EPERM;
1532 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1533 sock->type != SOCK_PACKET)
1534 return -ESOCKTNOSUPPORT;
1535
1536 sock->state = SS_UNCONNECTED;
1537
1538 err = -ENOBUFS;
1539 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1540 if (sk == NULL)
1541 goto out;
1542
1543 sock->ops = &packet_ops;
1544 if (sock->type == SOCK_PACKET)
1545 sock->ops = &packet_ops_spkt;
1546
1547 sock_init_data(sock, sk);
1548
1549 po = pkt_sk(sk);
1550 sk->sk_family = PF_PACKET;
1551 po->num = proto;
1552
1553 sk->sk_destruct = packet_sock_destruct;
1554 sk_refcnt_debug_inc(sk);
1555
1556 /*
1557 * Attach a protocol block
1558 */
1559
1560 spin_lock_init(&po->bind_lock);
1561 mutex_init(&po->pg_vec_lock);
1562 po->prot_hook.func = packet_rcv;
1563
1564 if (sock->type == SOCK_PACKET)
1565 po->prot_hook.func = packet_rcv_spkt;
1566
1567 po->prot_hook.af_packet_priv = sk;
1568
1569 if (proto) {
1570 po->prot_hook.type = proto;
1571 register_prot_hook(sk);
1572 }
1573
1574 spin_lock_bh(&net->packet.sklist_lock);
1575 sk_add_node_rcu(sk, &net->packet.sklist);
1576 sock_prot_inuse_add(net, &packet_proto, 1);
1577 spin_unlock_bh(&net->packet.sklist_lock);
1578
1579 return 0;
1580 out:
1581 return err;
1582 }
1583
1584 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1585 {
1586 struct sock_exterr_skb *serr;
1587 struct sk_buff *skb, *skb2;
1588 int copied, err;
1589
1590 err = -EAGAIN;
1591 skb = skb_dequeue(&sk->sk_error_queue);
1592 if (skb == NULL)
1593 goto out;
1594
1595 copied = skb->len;
1596 if (copied > len) {
1597 msg->msg_flags |= MSG_TRUNC;
1598 copied = len;
1599 }
1600 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1601 if (err)
1602 goto out_free_skb;
1603
1604 sock_recv_timestamp(msg, sk, skb);
1605
1606 serr = SKB_EXT_ERR(skb);
1607 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
1608 sizeof(serr->ee), &serr->ee);
1609
1610 msg->msg_flags |= MSG_ERRQUEUE;
1611 err = copied;
1612
1613 /* Reset and regenerate socket error */
1614 spin_lock_bh(&sk->sk_error_queue.lock);
1615 sk->sk_err = 0;
1616 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
1617 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
1618 spin_unlock_bh(&sk->sk_error_queue.lock);
1619 sk->sk_error_report(sk);
1620 } else
1621 spin_unlock_bh(&sk->sk_error_queue.lock);
1622
1623 out_free_skb:
1624 kfree_skb(skb);
1625 out:
1626 return err;
1627 }
1628
1629 /*
1630 * Pull a packet from our receive queue and hand it to the user.
1631 * If necessary we block.
1632 */
1633
1634 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1635 struct msghdr *msg, size_t len, int flags)
1636 {
1637 struct sock *sk = sock->sk;
1638 struct sk_buff *skb;
1639 int copied, err;
1640 struct sockaddr_ll *sll;
1641 int vnet_hdr_len = 0;
1642
1643 err = -EINVAL;
1644 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1645 goto out;
1646
1647 #if 0
1648 /* What error should we return now? EUNATTACH? */
1649 if (pkt_sk(sk)->ifindex < 0)
1650 return -ENODEV;
1651 #endif
1652
1653 if (flags & MSG_ERRQUEUE) {
1654 err = packet_recv_error(sk, msg, len);
1655 goto out;
1656 }
1657
1658 /*
1659 * Call the generic datagram receiver. This handles all sorts
1660 * of horrible races and re-entrancy so we can forget about it
1661 * in the protocol layers.
1662 *
1663 * Now it will return ENETDOWN, if device have just gone down,
1664 * but then it will block.
1665 */
1666
1667 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1668
1669 /*
1670 * An error occurred so return it. Because skb_recv_datagram()
1671 * handles the blocking we don't see and worry about blocking
1672 * retries.
1673 */
1674
1675 if (skb == NULL)
1676 goto out;
1677
1678 if (pkt_sk(sk)->has_vnet_hdr) {
1679 struct virtio_net_hdr vnet_hdr = { 0 };
1680
1681 err = -EINVAL;
1682 vnet_hdr_len = sizeof(vnet_hdr);
1683 if (len < vnet_hdr_len)
1684 goto out_free;
1685
1686 len -= vnet_hdr_len;
1687
1688 if (skb_is_gso(skb)) {
1689 struct skb_shared_info *sinfo = skb_shinfo(skb);
1690
1691 /* This is a hint as to how much should be linear. */
1692 vnet_hdr.hdr_len = skb_headlen(skb);
1693 vnet_hdr.gso_size = sinfo->gso_size;
1694 if (sinfo->gso_type & SKB_GSO_TCPV4)
1695 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1696 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1697 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1698 else if (sinfo->gso_type & SKB_GSO_UDP)
1699 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1700 else if (sinfo->gso_type & SKB_GSO_FCOE)
1701 goto out_free;
1702 else
1703 BUG();
1704 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1705 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1706 } else
1707 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1708
1709 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1710 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1711 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
1712 vnet_hdr.csum_offset = skb->csum_offset;
1713 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1714 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1715 } /* else everything is zero */
1716
1717 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1718 vnet_hdr_len);
1719 if (err < 0)
1720 goto out_free;
1721 }
1722
1723 /*
1724 * If the address length field is there to be filled in, we fill
1725 * it in now.
1726 */
1727
1728 sll = &PACKET_SKB_CB(skb)->sa.ll;
1729 if (sock->type == SOCK_PACKET)
1730 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1731 else
1732 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1733
1734 /*
1735 * You lose any data beyond the buffer you gave. If it worries a
1736 * user program they can ask the device for its MTU anyway.
1737 */
1738
1739 copied = skb->len;
1740 if (copied > len) {
1741 copied = len;
1742 msg->msg_flags |= MSG_TRUNC;
1743 }
1744
1745 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1746 if (err)
1747 goto out_free;
1748
1749 sock_recv_ts_and_drops(msg, sk, skb);
1750
1751 if (msg->msg_name)
1752 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1753 msg->msg_namelen);
1754
1755 if (pkt_sk(sk)->auxdata) {
1756 struct tpacket_auxdata aux;
1757
1758 aux.tp_status = TP_STATUS_USER;
1759 if (skb->ip_summed == CHECKSUM_PARTIAL)
1760 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1761 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1762 aux.tp_snaplen = skb->len;
1763 aux.tp_mac = 0;
1764 aux.tp_net = skb_network_offset(skb);
1765 if (vlan_tx_tag_present(skb)) {
1766 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
1767 aux.tp_status |= TP_STATUS_VLAN_VALID;
1768 } else {
1769 aux.tp_vlan_tci = 0;
1770 }
1771 aux.tp_padding = 0;
1772 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1773 }
1774
1775 /*
1776 * Free or return the buffer as appropriate. Again this
1777 * hides all the races and re-entrancy issues from us.
1778 */
1779 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1780
1781 out_free:
1782 skb_free_datagram(sk, skb);
1783 out:
1784 return err;
1785 }
1786
1787 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1788 int *uaddr_len, int peer)
1789 {
1790 struct net_device *dev;
1791 struct sock *sk = sock->sk;
1792
1793 if (peer)
1794 return -EOPNOTSUPP;
1795
1796 uaddr->sa_family = AF_PACKET;
1797 rcu_read_lock();
1798 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
1799 if (dev)
1800 strncpy(uaddr->sa_data, dev->name, 14);
1801 else
1802 memset(uaddr->sa_data, 0, 14);
1803 rcu_read_unlock();
1804 *uaddr_len = sizeof(*uaddr);
1805
1806 return 0;
1807 }
1808
1809 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1810 int *uaddr_len, int peer)
1811 {
1812 struct net_device *dev;
1813 struct sock *sk = sock->sk;
1814 struct packet_sock *po = pkt_sk(sk);
1815 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1816
1817 if (peer)
1818 return -EOPNOTSUPP;
1819
1820 sll->sll_family = AF_PACKET;
1821 sll->sll_ifindex = po->ifindex;
1822 sll->sll_protocol = po->num;
1823 sll->sll_pkttype = 0;
1824 rcu_read_lock();
1825 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1826 if (dev) {
1827 sll->sll_hatype = dev->type;
1828 sll->sll_halen = dev->addr_len;
1829 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1830 } else {
1831 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1832 sll->sll_halen = 0;
1833 }
1834 rcu_read_unlock();
1835 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1836
1837 return 0;
1838 }
1839
1840 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1841 int what)
1842 {
1843 switch (i->type) {
1844 case PACKET_MR_MULTICAST:
1845 if (i->alen != dev->addr_len)
1846 return -EINVAL;
1847 if (what > 0)
1848 return dev_mc_add(dev, i->addr);
1849 else
1850 return dev_mc_del(dev, i->addr);
1851 break;
1852 case PACKET_MR_PROMISC:
1853 return dev_set_promiscuity(dev, what);
1854 break;
1855 case PACKET_MR_ALLMULTI:
1856 return dev_set_allmulti(dev, what);
1857 break;
1858 case PACKET_MR_UNICAST:
1859 if (i->alen != dev->addr_len)
1860 return -EINVAL;
1861 if (what > 0)
1862 return dev_uc_add(dev, i->addr);
1863 else
1864 return dev_uc_del(dev, i->addr);
1865 break;
1866 default:
1867 break;
1868 }
1869 return 0;
1870 }
1871
1872 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1873 {
1874 for ( ; i; i = i->next) {
1875 if (i->ifindex == dev->ifindex)
1876 packet_dev_mc(dev, i, what);
1877 }
1878 }
1879
1880 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1881 {
1882 struct packet_sock *po = pkt_sk(sk);
1883 struct packet_mclist *ml, *i;
1884 struct net_device *dev;
1885 int err;
1886
1887 rtnl_lock();
1888
1889 err = -ENODEV;
1890 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1891 if (!dev)
1892 goto done;
1893
1894 err = -EINVAL;
1895 if (mreq->mr_alen > dev->addr_len)
1896 goto done;
1897
1898 err = -ENOBUFS;
1899 i = kmalloc(sizeof(*i), GFP_KERNEL);
1900 if (i == NULL)
1901 goto done;
1902
1903 err = 0;
1904 for (ml = po->mclist; ml; ml = ml->next) {
1905 if (ml->ifindex == mreq->mr_ifindex &&
1906 ml->type == mreq->mr_type &&
1907 ml->alen == mreq->mr_alen &&
1908 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1909 ml->count++;
1910 /* Free the new element ... */
1911 kfree(i);
1912 goto done;
1913 }
1914 }
1915
1916 i->type = mreq->mr_type;
1917 i->ifindex = mreq->mr_ifindex;
1918 i->alen = mreq->mr_alen;
1919 memcpy(i->addr, mreq->mr_address, i->alen);
1920 i->count = 1;
1921 i->next = po->mclist;
1922 po->mclist = i;
1923 err = packet_dev_mc(dev, i, 1);
1924 if (err) {
1925 po->mclist = i->next;
1926 kfree(i);
1927 }
1928
1929 done:
1930 rtnl_unlock();
1931 return err;
1932 }
1933
1934 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1935 {
1936 struct packet_mclist *ml, **mlp;
1937
1938 rtnl_lock();
1939
1940 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1941 if (ml->ifindex == mreq->mr_ifindex &&
1942 ml->type == mreq->mr_type &&
1943 ml->alen == mreq->mr_alen &&
1944 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1945 if (--ml->count == 0) {
1946 struct net_device *dev;
1947 *mlp = ml->next;
1948 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1949 if (dev)
1950 packet_dev_mc(dev, ml, -1);
1951 kfree(ml);
1952 }
1953 rtnl_unlock();
1954 return 0;
1955 }
1956 }
1957 rtnl_unlock();
1958 return -EADDRNOTAVAIL;
1959 }
1960
1961 static void packet_flush_mclist(struct sock *sk)
1962 {
1963 struct packet_sock *po = pkt_sk(sk);
1964 struct packet_mclist *ml;
1965
1966 if (!po->mclist)
1967 return;
1968
1969 rtnl_lock();
1970 while ((ml = po->mclist) != NULL) {
1971 struct net_device *dev;
1972
1973 po->mclist = ml->next;
1974 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1975 if (dev != NULL)
1976 packet_dev_mc(dev, ml, -1);
1977 kfree(ml);
1978 }
1979 rtnl_unlock();
1980 }
1981
1982 static int
1983 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1984 {
1985 struct sock *sk = sock->sk;
1986 struct packet_sock *po = pkt_sk(sk);
1987 int ret;
1988
1989 if (level != SOL_PACKET)
1990 return -ENOPROTOOPT;
1991
1992 switch (optname) {
1993 case PACKET_ADD_MEMBERSHIP:
1994 case PACKET_DROP_MEMBERSHIP:
1995 {
1996 struct packet_mreq_max mreq;
1997 int len = optlen;
1998 memset(&mreq, 0, sizeof(mreq));
1999 if (len < sizeof(struct packet_mreq))
2000 return -EINVAL;
2001 if (len > sizeof(mreq))
2002 len = sizeof(mreq);
2003 if (copy_from_user(&mreq, optval, len))
2004 return -EFAULT;
2005 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
2006 return -EINVAL;
2007 if (optname == PACKET_ADD_MEMBERSHIP)
2008 ret = packet_mc_add(sk, &mreq);
2009 else
2010 ret = packet_mc_drop(sk, &mreq);
2011 return ret;
2012 }
2013
2014 case PACKET_RX_RING:
2015 case PACKET_TX_RING:
2016 {
2017 struct tpacket_req req;
2018
2019 if (optlen < sizeof(req))
2020 return -EINVAL;
2021 if (pkt_sk(sk)->has_vnet_hdr)
2022 return -EINVAL;
2023 if (copy_from_user(&req, optval, sizeof(req)))
2024 return -EFAULT;
2025 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
2026 }
2027 case PACKET_COPY_THRESH:
2028 {
2029 int val;
2030
2031 if (optlen != sizeof(val))
2032 return -EINVAL;
2033 if (copy_from_user(&val, optval, sizeof(val)))
2034 return -EFAULT;
2035
2036 pkt_sk(sk)->copy_thresh = val;
2037 return 0;
2038 }
2039 case PACKET_VERSION:
2040 {
2041 int val;
2042
2043 if (optlen != sizeof(val))
2044 return -EINVAL;
2045 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2046 return -EBUSY;
2047 if (copy_from_user(&val, optval, sizeof(val)))
2048 return -EFAULT;
2049 switch (val) {
2050 case TPACKET_V1:
2051 case TPACKET_V2:
2052 po->tp_version = val;
2053 return 0;
2054 default:
2055 return -EINVAL;
2056 }
2057 }
2058 case PACKET_RESERVE:
2059 {
2060 unsigned int val;
2061
2062 if (optlen != sizeof(val))
2063 return -EINVAL;
2064 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2065 return -EBUSY;
2066 if (copy_from_user(&val, optval, sizeof(val)))
2067 return -EFAULT;
2068 po->tp_reserve = val;
2069 return 0;
2070 }
2071 case PACKET_LOSS:
2072 {
2073 unsigned int val;
2074
2075 if (optlen != sizeof(val))
2076 return -EINVAL;
2077 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2078 return -EBUSY;
2079 if (copy_from_user(&val, optval, sizeof(val)))
2080 return -EFAULT;
2081 po->tp_loss = !!val;
2082 return 0;
2083 }
2084 case PACKET_AUXDATA:
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->auxdata = !!val;
2094 return 0;
2095 }
2096 case PACKET_ORIGDEV:
2097 {
2098 int val;
2099
2100 if (optlen < sizeof(val))
2101 return -EINVAL;
2102 if (copy_from_user(&val, optval, sizeof(val)))
2103 return -EFAULT;
2104
2105 po->origdev = !!val;
2106 return 0;
2107 }
2108 case PACKET_VNET_HDR:
2109 {
2110 int val;
2111
2112 if (sock->type != SOCK_RAW)
2113 return -EINVAL;
2114 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2115 return -EBUSY;
2116 if (optlen < sizeof(val))
2117 return -EINVAL;
2118 if (copy_from_user(&val, optval, sizeof(val)))
2119 return -EFAULT;
2120
2121 po->has_vnet_hdr = !!val;
2122 return 0;
2123 }
2124 case PACKET_TIMESTAMP:
2125 {
2126 int val;
2127
2128 if (optlen != sizeof(val))
2129 return -EINVAL;
2130 if (copy_from_user(&val, optval, sizeof(val)))
2131 return -EFAULT;
2132
2133 po->tp_tstamp = val;
2134 return 0;
2135 }
2136 default:
2137 return -ENOPROTOOPT;
2138 }
2139 }
2140
2141 static int packet_getsockopt(struct socket *sock, int level, int optname,
2142 char __user *optval, int __user *optlen)
2143 {
2144 int len;
2145 int val;
2146 struct sock *sk = sock->sk;
2147 struct packet_sock *po = pkt_sk(sk);
2148 void *data;
2149 struct tpacket_stats st;
2150
2151 if (level != SOL_PACKET)
2152 return -ENOPROTOOPT;
2153
2154 if (get_user(len, optlen))
2155 return -EFAULT;
2156
2157 if (len < 0)
2158 return -EINVAL;
2159
2160 switch (optname) {
2161 case PACKET_STATISTICS:
2162 if (len > sizeof(struct tpacket_stats))
2163 len = sizeof(struct tpacket_stats);
2164 spin_lock_bh(&sk->sk_receive_queue.lock);
2165 st = po->stats;
2166 memset(&po->stats, 0, sizeof(st));
2167 spin_unlock_bh(&sk->sk_receive_queue.lock);
2168 st.tp_packets += st.tp_drops;
2169
2170 data = &st;
2171 break;
2172 case PACKET_AUXDATA:
2173 if (len > sizeof(int))
2174 len = sizeof(int);
2175 val = po->auxdata;
2176
2177 data = &val;
2178 break;
2179 case PACKET_ORIGDEV:
2180 if (len > sizeof(int))
2181 len = sizeof(int);
2182 val = po->origdev;
2183
2184 data = &val;
2185 break;
2186 case PACKET_VNET_HDR:
2187 if (len > sizeof(int))
2188 len = sizeof(int);
2189 val = po->has_vnet_hdr;
2190
2191 data = &val;
2192 break;
2193 case PACKET_VERSION:
2194 if (len > sizeof(int))
2195 len = sizeof(int);
2196 val = po->tp_version;
2197 data = &val;
2198 break;
2199 case PACKET_HDRLEN:
2200 if (len > sizeof(int))
2201 len = sizeof(int);
2202 if (copy_from_user(&val, optval, len))
2203 return -EFAULT;
2204 switch (val) {
2205 case TPACKET_V1:
2206 val = sizeof(struct tpacket_hdr);
2207 break;
2208 case TPACKET_V2:
2209 val = sizeof(struct tpacket2_hdr);
2210 break;
2211 default:
2212 return -EINVAL;
2213 }
2214 data = &val;
2215 break;
2216 case PACKET_RESERVE:
2217 if (len > sizeof(unsigned int))
2218 len = sizeof(unsigned int);
2219 val = po->tp_reserve;
2220 data = &val;
2221 break;
2222 case PACKET_LOSS:
2223 if (len > sizeof(unsigned int))
2224 len = sizeof(unsigned int);
2225 val = po->tp_loss;
2226 data = &val;
2227 break;
2228 case PACKET_TIMESTAMP:
2229 if (len > sizeof(int))
2230 len = sizeof(int);
2231 val = po->tp_tstamp;
2232 data = &val;
2233 break;
2234 default:
2235 return -ENOPROTOOPT;
2236 }
2237
2238 if (put_user(len, optlen))
2239 return -EFAULT;
2240 if (copy_to_user(optval, data, len))
2241 return -EFAULT;
2242 return 0;
2243 }
2244
2245
2246 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2247 {
2248 struct sock *sk;
2249 struct hlist_node *node;
2250 struct net_device *dev = data;
2251 struct net *net = dev_net(dev);
2252
2253 rcu_read_lock();
2254 sk_for_each_rcu(sk, node, &net->packet.sklist) {
2255 struct packet_sock *po = pkt_sk(sk);
2256
2257 switch (msg) {
2258 case NETDEV_UNREGISTER:
2259 if (po->mclist)
2260 packet_dev_mclist(dev, po->mclist, -1);
2261 /* fallthrough */
2262
2263 case NETDEV_DOWN:
2264 if (dev->ifindex == po->ifindex) {
2265 spin_lock(&po->bind_lock);
2266 if (po->running) {
2267 __unregister_prot_hook(sk, false);
2268 sk->sk_err = ENETDOWN;
2269 if (!sock_flag(sk, SOCK_DEAD))
2270 sk->sk_error_report(sk);
2271 }
2272 if (msg == NETDEV_UNREGISTER) {
2273 po->ifindex = -1;
2274 if (po->prot_hook.dev)
2275 dev_put(po->prot_hook.dev);
2276 po->prot_hook.dev = NULL;
2277 }
2278 spin_unlock(&po->bind_lock);
2279 }
2280 break;
2281 case NETDEV_UP:
2282 if (dev->ifindex == po->ifindex) {
2283 spin_lock(&po->bind_lock);
2284 if (po->num)
2285 register_prot_hook(sk);
2286 spin_unlock(&po->bind_lock);
2287 }
2288 break;
2289 }
2290 }
2291 rcu_read_unlock();
2292 return NOTIFY_DONE;
2293 }
2294
2295
2296 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2297 unsigned long arg)
2298 {
2299 struct sock *sk = sock->sk;
2300
2301 switch (cmd) {
2302 case SIOCOUTQ:
2303 {
2304 int amount = sk_wmem_alloc_get(sk);
2305
2306 return put_user(amount, (int __user *)arg);
2307 }
2308 case SIOCINQ:
2309 {
2310 struct sk_buff *skb;
2311 int amount = 0;
2312
2313 spin_lock_bh(&sk->sk_receive_queue.lock);
2314 skb = skb_peek(&sk->sk_receive_queue);
2315 if (skb)
2316 amount = skb->len;
2317 spin_unlock_bh(&sk->sk_receive_queue.lock);
2318 return put_user(amount, (int __user *)arg);
2319 }
2320 case SIOCGSTAMP:
2321 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2322 case SIOCGSTAMPNS:
2323 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2324
2325 #ifdef CONFIG_INET
2326 case SIOCADDRT:
2327 case SIOCDELRT:
2328 case SIOCDARP:
2329 case SIOCGARP:
2330 case SIOCSARP:
2331 case SIOCGIFADDR:
2332 case SIOCSIFADDR:
2333 case SIOCGIFBRDADDR:
2334 case SIOCSIFBRDADDR:
2335 case SIOCGIFNETMASK:
2336 case SIOCSIFNETMASK:
2337 case SIOCGIFDSTADDR:
2338 case SIOCSIFDSTADDR:
2339 case SIOCSIFFLAGS:
2340 return inet_dgram_ops.ioctl(sock, cmd, arg);
2341 #endif
2342
2343 default:
2344 return -ENOIOCTLCMD;
2345 }
2346 return 0;
2347 }
2348
2349 static unsigned int packet_poll(struct file *file, struct socket *sock,
2350 poll_table *wait)
2351 {
2352 struct sock *sk = sock->sk;
2353 struct packet_sock *po = pkt_sk(sk);
2354 unsigned int mask = datagram_poll(file, sock, wait);
2355
2356 spin_lock_bh(&sk->sk_receive_queue.lock);
2357 if (po->rx_ring.pg_vec) {
2358 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2359 mask |= POLLIN | POLLRDNORM;
2360 }
2361 spin_unlock_bh(&sk->sk_receive_queue.lock);
2362 spin_lock_bh(&sk->sk_write_queue.lock);
2363 if (po->tx_ring.pg_vec) {
2364 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2365 mask |= POLLOUT | POLLWRNORM;
2366 }
2367 spin_unlock_bh(&sk->sk_write_queue.lock);
2368 return mask;
2369 }
2370
2371
2372 /* Dirty? Well, I still did not learn better way to account
2373 * for user mmaps.
2374 */
2375
2376 static void packet_mm_open(struct vm_area_struct *vma)
2377 {
2378 struct file *file = vma->vm_file;
2379 struct socket *sock = file->private_data;
2380 struct sock *sk = sock->sk;
2381
2382 if (sk)
2383 atomic_inc(&pkt_sk(sk)->mapped);
2384 }
2385
2386 static void packet_mm_close(struct vm_area_struct *vma)
2387 {
2388 struct file *file = vma->vm_file;
2389 struct socket *sock = file->private_data;
2390 struct sock *sk = sock->sk;
2391
2392 if (sk)
2393 atomic_dec(&pkt_sk(sk)->mapped);
2394 }
2395
2396 static const struct vm_operations_struct packet_mmap_ops = {
2397 .open = packet_mm_open,
2398 .close = packet_mm_close,
2399 };
2400
2401 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
2402 unsigned int len)
2403 {
2404 int i;
2405
2406 for (i = 0; i < len; i++) {
2407 if (likely(pg_vec[i].buffer)) {
2408 if (is_vmalloc_addr(pg_vec[i].buffer))
2409 vfree(pg_vec[i].buffer);
2410 else
2411 free_pages((unsigned long)pg_vec[i].buffer,
2412 order);
2413 pg_vec[i].buffer = NULL;
2414 }
2415 }
2416 kfree(pg_vec);
2417 }
2418
2419 static inline char *alloc_one_pg_vec_page(unsigned long order)
2420 {
2421 char *buffer = NULL;
2422 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
2423 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
2424
2425 buffer = (char *) __get_free_pages(gfp_flags, order);
2426
2427 if (buffer)
2428 return buffer;
2429
2430 /*
2431 * __get_free_pages failed, fall back to vmalloc
2432 */
2433 buffer = vzalloc((1 << order) * PAGE_SIZE);
2434
2435 if (buffer)
2436 return buffer;
2437
2438 /*
2439 * vmalloc failed, lets dig into swap here
2440 */
2441 gfp_flags &= ~__GFP_NORETRY;
2442 buffer = (char *)__get_free_pages(gfp_flags, order);
2443 if (buffer)
2444 return buffer;
2445
2446 /*
2447 * complete and utter failure
2448 */
2449 return NULL;
2450 }
2451
2452 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
2453 {
2454 unsigned int block_nr = req->tp_block_nr;
2455 struct pgv *pg_vec;
2456 int i;
2457
2458 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
2459 if (unlikely(!pg_vec))
2460 goto out;
2461
2462 for (i = 0; i < block_nr; i++) {
2463 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
2464 if (unlikely(!pg_vec[i].buffer))
2465 goto out_free_pgvec;
2466 }
2467
2468 out:
2469 return pg_vec;
2470
2471 out_free_pgvec:
2472 free_pg_vec(pg_vec, order, block_nr);
2473 pg_vec = NULL;
2474 goto out;
2475 }
2476
2477 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2478 int closing, int tx_ring)
2479 {
2480 struct pgv *pg_vec = NULL;
2481 struct packet_sock *po = pkt_sk(sk);
2482 int was_running, order = 0;
2483 struct packet_ring_buffer *rb;
2484 struct sk_buff_head *rb_queue;
2485 __be16 num;
2486 int err;
2487
2488 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2489 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2490
2491 err = -EBUSY;
2492 if (!closing) {
2493 if (atomic_read(&po->mapped))
2494 goto out;
2495 if (atomic_read(&rb->pending))
2496 goto out;
2497 }
2498
2499 if (req->tp_block_nr) {
2500 /* Sanity tests and some calculations */
2501 err = -EBUSY;
2502 if (unlikely(rb->pg_vec))
2503 goto out;
2504
2505 switch (po->tp_version) {
2506 case TPACKET_V1:
2507 po->tp_hdrlen = TPACKET_HDRLEN;
2508 break;
2509 case TPACKET_V2:
2510 po->tp_hdrlen = TPACKET2_HDRLEN;
2511 break;
2512 }
2513
2514 err = -EINVAL;
2515 if (unlikely((int)req->tp_block_size <= 0))
2516 goto out;
2517 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2518 goto out;
2519 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2520 po->tp_reserve))
2521 goto out;
2522 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2523 goto out;
2524
2525 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2526 if (unlikely(rb->frames_per_block <= 0))
2527 goto out;
2528 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2529 req->tp_frame_nr))
2530 goto out;
2531
2532 err = -ENOMEM;
2533 order = get_order(req->tp_block_size);
2534 pg_vec = alloc_pg_vec(req, order);
2535 if (unlikely(!pg_vec))
2536 goto out;
2537 }
2538 /* Done */
2539 else {
2540 err = -EINVAL;
2541 if (unlikely(req->tp_frame_nr))
2542 goto out;
2543 }
2544
2545 lock_sock(sk);
2546
2547 /* Detach socket from network */
2548 spin_lock(&po->bind_lock);
2549 was_running = po->running;
2550 num = po->num;
2551 if (was_running) {
2552 po->num = 0;
2553 __unregister_prot_hook(sk, false);
2554 }
2555 spin_unlock(&po->bind_lock);
2556
2557 synchronize_net();
2558
2559 err = -EBUSY;
2560 mutex_lock(&po->pg_vec_lock);
2561 if (closing || atomic_read(&po->mapped) == 0) {
2562 err = 0;
2563 spin_lock_bh(&rb_queue->lock);
2564 swap(rb->pg_vec, pg_vec);
2565 rb->frame_max = (req->tp_frame_nr - 1);
2566 rb->head = 0;
2567 rb->frame_size = req->tp_frame_size;
2568 spin_unlock_bh(&rb_queue->lock);
2569
2570 swap(rb->pg_vec_order, order);
2571 swap(rb->pg_vec_len, req->tp_block_nr);
2572
2573 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2574 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2575 tpacket_rcv : packet_rcv;
2576 skb_queue_purge(rb_queue);
2577 if (atomic_read(&po->mapped))
2578 pr_err("packet_mmap: vma is busy: %d\n",
2579 atomic_read(&po->mapped));
2580 }
2581 mutex_unlock(&po->pg_vec_lock);
2582
2583 spin_lock(&po->bind_lock);
2584 if (was_running) {
2585 po->num = num;
2586 register_prot_hook(sk);
2587 }
2588 spin_unlock(&po->bind_lock);
2589
2590 release_sock(sk);
2591
2592 if (pg_vec)
2593 free_pg_vec(pg_vec, order, req->tp_block_nr);
2594 out:
2595 return err;
2596 }
2597
2598 static int packet_mmap(struct file *file, struct socket *sock,
2599 struct vm_area_struct *vma)
2600 {
2601 struct sock *sk = sock->sk;
2602 struct packet_sock *po = pkt_sk(sk);
2603 unsigned long size, expected_size;
2604 struct packet_ring_buffer *rb;
2605 unsigned long start;
2606 int err = -EINVAL;
2607 int i;
2608
2609 if (vma->vm_pgoff)
2610 return -EINVAL;
2611
2612 mutex_lock(&po->pg_vec_lock);
2613
2614 expected_size = 0;
2615 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2616 if (rb->pg_vec) {
2617 expected_size += rb->pg_vec_len
2618 * rb->pg_vec_pages
2619 * PAGE_SIZE;
2620 }
2621 }
2622
2623 if (expected_size == 0)
2624 goto out;
2625
2626 size = vma->vm_end - vma->vm_start;
2627 if (size != expected_size)
2628 goto out;
2629
2630 start = vma->vm_start;
2631 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2632 if (rb->pg_vec == NULL)
2633 continue;
2634
2635 for (i = 0; i < rb->pg_vec_len; i++) {
2636 struct page *page;
2637 void *kaddr = rb->pg_vec[i].buffer;
2638 int pg_num;
2639
2640 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
2641 page = pgv_to_page(kaddr);
2642 err = vm_insert_page(vma, start, page);
2643 if (unlikely(err))
2644 goto out;
2645 start += PAGE_SIZE;
2646 kaddr += PAGE_SIZE;
2647 }
2648 }
2649 }
2650
2651 atomic_inc(&po->mapped);
2652 vma->vm_ops = &packet_mmap_ops;
2653 err = 0;
2654
2655 out:
2656 mutex_unlock(&po->pg_vec_lock);
2657 return err;
2658 }
2659
2660 static const struct proto_ops packet_ops_spkt = {
2661 .family = PF_PACKET,
2662 .owner = THIS_MODULE,
2663 .release = packet_release,
2664 .bind = packet_bind_spkt,
2665 .connect = sock_no_connect,
2666 .socketpair = sock_no_socketpair,
2667 .accept = sock_no_accept,
2668 .getname = packet_getname_spkt,
2669 .poll = datagram_poll,
2670 .ioctl = packet_ioctl,
2671 .listen = sock_no_listen,
2672 .shutdown = sock_no_shutdown,
2673 .setsockopt = sock_no_setsockopt,
2674 .getsockopt = sock_no_getsockopt,
2675 .sendmsg = packet_sendmsg_spkt,
2676 .recvmsg = packet_recvmsg,
2677 .mmap = sock_no_mmap,
2678 .sendpage = sock_no_sendpage,
2679 };
2680
2681 static const struct proto_ops packet_ops = {
2682 .family = PF_PACKET,
2683 .owner = THIS_MODULE,
2684 .release = packet_release,
2685 .bind = packet_bind,
2686 .connect = sock_no_connect,
2687 .socketpair = sock_no_socketpair,
2688 .accept = sock_no_accept,
2689 .getname = packet_getname,
2690 .poll = packet_poll,
2691 .ioctl = packet_ioctl,
2692 .listen = sock_no_listen,
2693 .shutdown = sock_no_shutdown,
2694 .setsockopt = packet_setsockopt,
2695 .getsockopt = packet_getsockopt,
2696 .sendmsg = packet_sendmsg,
2697 .recvmsg = packet_recvmsg,
2698 .mmap = packet_mmap,
2699 .sendpage = sock_no_sendpage,
2700 };
2701
2702 static const struct net_proto_family packet_family_ops = {
2703 .family = PF_PACKET,
2704 .create = packet_create,
2705 .owner = THIS_MODULE,
2706 };
2707
2708 static struct notifier_block packet_netdev_notifier = {
2709 .notifier_call = packet_notifier,
2710 };
2711
2712 #ifdef CONFIG_PROC_FS
2713
2714 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2715 __acquires(RCU)
2716 {
2717 struct net *net = seq_file_net(seq);
2718
2719 rcu_read_lock();
2720 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
2721 }
2722
2723 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2724 {
2725 struct net *net = seq_file_net(seq);
2726 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
2727 }
2728
2729 static void packet_seq_stop(struct seq_file *seq, void *v)
2730 __releases(RCU)
2731 {
2732 rcu_read_unlock();
2733 }
2734
2735 static int packet_seq_show(struct seq_file *seq, void *v)
2736 {
2737 if (v == SEQ_START_TOKEN)
2738 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
2739 else {
2740 struct sock *s = sk_entry(v);
2741 const struct packet_sock *po = pkt_sk(s);
2742
2743 seq_printf(seq,
2744 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
2745 s,
2746 atomic_read(&s->sk_refcnt),
2747 s->sk_type,
2748 ntohs(po->num),
2749 po->ifindex,
2750 po->running,
2751 atomic_read(&s->sk_rmem_alloc),
2752 sock_i_uid(s),
2753 sock_i_ino(s));
2754 }
2755
2756 return 0;
2757 }
2758
2759 static const struct seq_operations packet_seq_ops = {
2760 .start = packet_seq_start,
2761 .next = packet_seq_next,
2762 .stop = packet_seq_stop,
2763 .show = packet_seq_show,
2764 };
2765
2766 static int packet_seq_open(struct inode *inode, struct file *file)
2767 {
2768 return seq_open_net(inode, file, &packet_seq_ops,
2769 sizeof(struct seq_net_private));
2770 }
2771
2772 static const struct file_operations packet_seq_fops = {
2773 .owner = THIS_MODULE,
2774 .open = packet_seq_open,
2775 .read = seq_read,
2776 .llseek = seq_lseek,
2777 .release = seq_release_net,
2778 };
2779
2780 #endif
2781
2782 static int __net_init packet_net_init(struct net *net)
2783 {
2784 spin_lock_init(&net->packet.sklist_lock);
2785 INIT_HLIST_HEAD(&net->packet.sklist);
2786
2787 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2788 return -ENOMEM;
2789
2790 return 0;
2791 }
2792
2793 static void __net_exit packet_net_exit(struct net *net)
2794 {
2795 proc_net_remove(net, "packet");
2796 }
2797
2798 static struct pernet_operations packet_net_ops = {
2799 .init = packet_net_init,
2800 .exit = packet_net_exit,
2801 };
2802
2803
2804 static void __exit packet_exit(void)
2805 {
2806 unregister_netdevice_notifier(&packet_netdev_notifier);
2807 unregister_pernet_subsys(&packet_net_ops);
2808 sock_unregister(PF_PACKET);
2809 proto_unregister(&packet_proto);
2810 }
2811
2812 static int __init packet_init(void)
2813 {
2814 int rc = proto_register(&packet_proto, 0);
2815
2816 if (rc != 0)
2817 goto out;
2818
2819 sock_register(&packet_family_ops);
2820 register_pernet_subsys(&packet_net_ops);
2821 register_netdevice_notifier(&packet_netdev_notifier);
2822 out:
2823 return rc;
2824 }
2825
2826 module_init(packet_init);
2827 module_exit(packet_exit);
2828 MODULE_LICENSE("GPL");
2829 MODULE_ALIAS_NETPROTO(PF_PACKET);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree