?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
relay: prevent integer overflow in relay_open()
[linux-2.6.git] / net / packet / af_packet.c
blob2dbb32b988c4d476406e212f0ff5632677e0501b
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PACKET - implements raw packet sockets.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 *
12 * Fixes:
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
41 * and packet_mreq.
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
44 * layer.
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
46 *
47 *
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
52 *
53 */
54
55 #include <linux/types.h>
56 #include <linux/mm.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
60 #include <linux/in.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
70 #include <net/ip.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/system.h>
77 #include <asm/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <asm/page.h>
80 #include <asm/cacheflush.h>
81 #include <asm/io.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84 #include <linux/poll.h>
85 #include <linux/module.h>
86 #include <linux/init.h>
87 #include <linux/mutex.h>
88 #include <linux/if_vlan.h>
89 #include <linux/virtio_net.h>
90 #include <linux/errqueue.h>
91 #include <linux/net_tstamp.h>
92
93 #ifdef CONFIG_INET
94 #include <net/inet_common.h>
95 #endif
96
97 /*
98 Assumptions:
99 - if device has no dev->hard_header routine, it adds and removes ll header
100 inside itself. In this case ll header is invisible outside of device,
101 but higher levels still should reserve dev->hard_header_len.
102 Some devices are enough clever to reallocate skb, when header
103 will not fit to reserved space (tunnel), another ones are silly
104 (PPP).
105 - packet socket receives packets with pulled ll header,
106 so that SOCK_RAW should push it back.
107
108 On receive:
109 -----------
110
111 Incoming, dev->hard_header!=NULL
112 mac_header -> ll header
113 data -> data
114
115 Outgoing, dev->hard_header!=NULL
116 mac_header -> ll header
117 data -> ll header
118
119 Incoming, dev->hard_header==NULL
120 mac_header -> UNKNOWN position. It is very likely, that it points to ll
121 header. PPP makes it, that is wrong, because introduce
122 assymetry between rx and tx paths.
123 data -> data
124
125 Outgoing, dev->hard_header==NULL
126 mac_header -> data. ll header is still not built!
127 data -> data
128
129 Resume
130 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
131
132
133 On transmit:
134 ------------
135
136 dev->hard_header != NULL
137 mac_header -> ll header
138 data -> ll header
139
140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
141 mac_header -> data
142 data -> data
143
144 We should set nh.raw on output to correct posistion,
145 packet classifier depends on it.
146 */
147
148 /* Private packet socket structures. */
149
150 struct packet_mclist {
151 struct packet_mclist *next;
152 int ifindex;
153 int count;
154 unsigned short type;
155 unsigned short alen;
156 unsigned char addr[MAX_ADDR_LEN];
157 };
158 /* identical to struct packet_mreq except it has
159 * a longer address field.
160 */
161 struct packet_mreq_max {
162 int mr_ifindex;
163 unsigned short mr_type;
164 unsigned short mr_alen;
165 unsigned char mr_address[MAX_ADDR_LEN];
166 };
167
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
170
171
172 #define V3_ALIGNMENT (8)
173
174 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178
179 /* kbdq - kernel block descriptor queue */
180 struct tpacket_kbdq_core {
181 struct pgv *pkbdq;
182 unsigned int feature_req_word;
183 unsigned int hdrlen;
184 unsigned char reset_pending_on_curr_blk;
185 unsigned char delete_blk_timer;
186 unsigned short kactive_blk_num;
187 unsigned short blk_sizeof_priv;
188
189 /* last_kactive_blk_num:
190 * trick to see if user-space has caught up
191 * in order to avoid refreshing timer when every single pkt arrives.
192 */
193 unsigned short last_kactive_blk_num;
194
195 char *pkblk_start;
196 char *pkblk_end;
197 int kblk_size;
198 unsigned int knum_blocks;
199 uint64_t knxt_seq_num;
200 char *prev;
201 char *nxt_offset;
202 struct sk_buff *skb;
203
204 atomic_t blk_fill_in_prog;
205
206 /* Default is set to 8ms */
207 #define DEFAULT_PRB_RETIRE_TOV (8)
208
209 unsigned short retire_blk_tov;
210 unsigned short version;
211 unsigned long tov_in_jiffies;
212
213 /* timer to retire an outstanding block */
214 struct timer_list retire_blk_timer;
215 };
216
217 #define PGV_FROM_VMALLOC 1
218 struct pgv {
219 char *buffer;
220 };
221
222 struct packet_ring_buffer {
223 struct pgv *pg_vec;
224 unsigned int head;
225 unsigned int frames_per_block;
226 unsigned int frame_size;
227 unsigned int frame_max;
228
229 unsigned int pg_vec_order;
230 unsigned int pg_vec_pages;
231 unsigned int pg_vec_len;
232
233 struct tpacket_kbdq_core prb_bdqc;
234 atomic_t pending;
235 };
236
237 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
238 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
239 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
240 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
241 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
242 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
243 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
244
245 struct packet_sock;
246 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
247
248 static void *packet_previous_frame(struct packet_sock *po,
249 struct packet_ring_buffer *rb,
250 int status);
251 static void packet_increment_head(struct packet_ring_buffer *buff);
252 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
253 struct tpacket_block_desc *);
254 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
255 struct packet_sock *);
256 static void prb_retire_current_block(struct tpacket_kbdq_core *,
257 struct packet_sock *, unsigned int status);
258 static int prb_queue_frozen(struct tpacket_kbdq_core *);
259 static void prb_open_block(struct tpacket_kbdq_core *,
260 struct tpacket_block_desc *);
261 static void prb_retire_rx_blk_timer_expired(unsigned long);
262 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
263 static void prb_init_blk_timer(struct packet_sock *,
264 struct tpacket_kbdq_core *,
265 void (*func) (unsigned long));
266 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
267 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
268 struct tpacket3_hdr *);
269 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
270 struct tpacket3_hdr *);
271 static void packet_flush_mclist(struct sock *sk);
272
273 struct packet_fanout;
274 struct packet_sock {
275 /* struct sock has to be the first member of packet_sock */
276 struct sock sk;
277 struct packet_fanout *fanout;
278 struct tpacket_stats stats;
279 union tpacket_stats_u stats_u;
280 struct packet_ring_buffer rx_ring;
281 struct packet_ring_buffer tx_ring;
282 int copy_thresh;
283 spinlock_t bind_lock;
284 struct mutex pg_vec_lock;
285 unsigned int running:1, /* prot_hook is attached*/
286 auxdata:1,
287 origdev:1,
288 has_vnet_hdr:1;
289 int ifindex; /* bound device */
290 __be16 num;
291 struct packet_mclist *mclist;
292 atomic_t mapped;
293 enum tpacket_versions tp_version;
294 unsigned int tp_hdrlen;
295 unsigned int tp_reserve;
296 unsigned int tp_loss:1;
297 unsigned int tp_tstamp;
298 struct packet_type prot_hook ____cacheline_aligned_in_smp;
299 };
300
301 #define PACKET_FANOUT_MAX 256
302
303 struct packet_fanout {
304 #ifdef CONFIG_NET_NS
305 struct net *net;
306 #endif
307 unsigned int num_members;
308 u16 id;
309 u8 type;
310 u8 defrag;
311 atomic_t rr_cur;
312 struct list_head list;
313 struct sock *arr[PACKET_FANOUT_MAX];
314 spinlock_t lock;
315 atomic_t sk_ref;
316 struct packet_type prot_hook ____cacheline_aligned_in_smp;
317 };
318
319 struct packet_skb_cb {
320 unsigned int origlen;
321 union {
322 struct sockaddr_pkt pkt;
323 struct sockaddr_ll ll;
324 } sa;
325 };
326
327 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
328
329 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
330 #define GET_PBLOCK_DESC(x, bid) \
331 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
332 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
333 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
334 #define GET_NEXT_PRB_BLK_NUM(x) \
335 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
336 ((x)->kactive_blk_num+1) : 0)
337
338 static struct packet_sock *pkt_sk(struct sock *sk)
339 {
340 return (struct packet_sock *)sk;
341 }
342
343 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
344 static void __fanout_link(struct sock *sk, struct packet_sock *po);
345
346 /* register_prot_hook must be invoked with the po->bind_lock held,
347 * or from a context in which asynchronous accesses to the packet
348 * socket is not possible (packet_create()).
349 */
350 static void register_prot_hook(struct sock *sk)
351 {
352 struct packet_sock *po = pkt_sk(sk);
353 if (!po->running) {
354 if (po->fanout)
355 __fanout_link(sk, po);
356 else
357 dev_add_pack(&po->prot_hook);
358 sock_hold(sk);
359 po->running = 1;
360 }
361 }
362
363 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
364 * held. If the sync parameter is true, we will temporarily drop
365 * the po->bind_lock and do a synchronize_net to make sure no
366 * asynchronous packet processing paths still refer to the elements
367 * of po->prot_hook. If the sync parameter is false, it is the
368 * callers responsibility to take care of this.
369 */
370 static void __unregister_prot_hook(struct sock *sk, bool sync)
371 {
372 struct packet_sock *po = pkt_sk(sk);
373
374 po->running = 0;
375 if (po->fanout)
376 __fanout_unlink(sk, po);
377 else
378 __dev_remove_pack(&po->prot_hook);
379 __sock_put(sk);
380
381 if (sync) {
382 spin_unlock(&po->bind_lock);
383 synchronize_net();
384 spin_lock(&po->bind_lock);
385 }
386 }
387
388 static void unregister_prot_hook(struct sock *sk, bool sync)
389 {
390 struct packet_sock *po = pkt_sk(sk);
391
392 if (po->running)
393 __unregister_prot_hook(sk, sync);
394 }
395
396 static inline __pure struct page *pgv_to_page(void *addr)
397 {
398 if (is_vmalloc_addr(addr))
399 return vmalloc_to_page(addr);
400 return virt_to_page(addr);
401 }
402
403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
404 {
405 union {
406 struct tpacket_hdr *h1;
407 struct tpacket2_hdr *h2;
408 void *raw;
409 } h;
410
411 h.raw = frame;
412 switch (po->tp_version) {
413 case TPACKET_V1:
414 h.h1->tp_status = status;
415 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
416 break;
417 case TPACKET_V2:
418 h.h2->tp_status = status;
419 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
420 break;
421 case TPACKET_V3:
422 default:
423 WARN(1, "TPACKET version not supported.\n");
424 BUG();
425 }
426
427 smp_wmb();
428 }
429
430 static int __packet_get_status(struct packet_sock *po, void *frame)
431 {
432 union {
433 struct tpacket_hdr *h1;
434 struct tpacket2_hdr *h2;
435 void *raw;
436 } h;
437
438 smp_rmb();
439
440 h.raw = frame;
441 switch (po->tp_version) {
442 case TPACKET_V1:
443 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
444 return h.h1->tp_status;
445 case TPACKET_V2:
446 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
447 return h.h2->tp_status;
448 case TPACKET_V3:
449 default:
450 WARN(1, "TPACKET version not supported.\n");
451 BUG();
452 return 0;
453 }
454 }
455
456 static void *packet_lookup_frame(struct packet_sock *po,
457 struct packet_ring_buffer *rb,
458 unsigned int position,
459 int status)
460 {
461 unsigned int pg_vec_pos, frame_offset;
462 union {
463 struct tpacket_hdr *h1;
464 struct tpacket2_hdr *h2;
465 void *raw;
466 } h;
467
468 pg_vec_pos = position / rb->frames_per_block;
469 frame_offset = position % rb->frames_per_block;
470
471 h.raw = rb->pg_vec[pg_vec_pos].buffer +
472 (frame_offset * rb->frame_size);
473
474 if (status != __packet_get_status(po, h.raw))
475 return NULL;
476
477 return h.raw;
478 }
479
480 static void *packet_current_frame(struct packet_sock *po,
481 struct packet_ring_buffer *rb,
482 int status)
483 {
484 return packet_lookup_frame(po, rb, rb->head, status);
485 }
486
487 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
488 {
489 del_timer_sync(&pkc->retire_blk_timer);
490 }
491
492 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
493 int tx_ring,
494 struct sk_buff_head *rb_queue)
495 {
496 struct tpacket_kbdq_core *pkc;
497
498 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
499
500 spin_lock(&rb_queue->lock);
501 pkc->delete_blk_timer = 1;
502 spin_unlock(&rb_queue->lock);
503
504 prb_del_retire_blk_timer(pkc);
505 }
506
507 static void prb_init_blk_timer(struct packet_sock *po,
508 struct tpacket_kbdq_core *pkc,
509 void (*func) (unsigned long))
510 {
511 init_timer(&pkc->retire_blk_timer);
512 pkc->retire_blk_timer.data = (long)po;
513 pkc->retire_blk_timer.function = func;
514 pkc->retire_blk_timer.expires = jiffies;
515 }
516
517 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
518 {
519 struct tpacket_kbdq_core *pkc;
520
521 if (tx_ring)
522 BUG();
523
524 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
525 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
526 }
527
528 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
529 int blk_size_in_bytes)
530 {
531 struct net_device *dev;
532 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
533 struct ethtool_cmd ecmd;
534 int err;
535
536 rtnl_lock();
537 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
538 if (unlikely(!dev)) {
539 rtnl_unlock();
540 return DEFAULT_PRB_RETIRE_TOV;
541 }
542 err = __ethtool_get_settings(dev, &ecmd);
543 rtnl_unlock();
544 if (!err) {
545 switch (ecmd.speed) {
546 case SPEED_10000:
547 msec = 1;
548 div = 10000/1000;
549 break;
550 case SPEED_1000:
551 msec = 1;
552 div = 1000/1000;
553 break;
554 /*
555 * If the link speed is so slow you don't really
556 * need to worry about perf anyways
557 */
558 case SPEED_100:
559 case SPEED_10:
560 default:
561 return DEFAULT_PRB_RETIRE_TOV;
562 }
563 }
564
565 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
566
567 if (div)
568 mbits /= div;
569
570 tmo = mbits * msec;
571
572 if (div)
573 return tmo+1;
574 return tmo;
575 }
576
577 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
578 union tpacket_req_u *req_u)
579 {
580 p1->feature_req_word = req_u->req3.tp_feature_req_word;
581 }
582
583 static void init_prb_bdqc(struct packet_sock *po,
584 struct packet_ring_buffer *rb,
585 struct pgv *pg_vec,
586 union tpacket_req_u *req_u, int tx_ring)
587 {
588 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
589 struct tpacket_block_desc *pbd;
590
591 memset(p1, 0x0, sizeof(*p1));
592
593 p1->knxt_seq_num = 1;
594 p1->pkbdq = pg_vec;
595 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
596 p1->pkblk_start = (char *)pg_vec[0].buffer;
597 p1->kblk_size = req_u->req3.tp_block_size;
598 p1->knum_blocks = req_u->req3.tp_block_nr;
599 p1->hdrlen = po->tp_hdrlen;
600 p1->version = po->tp_version;
601 p1->last_kactive_blk_num = 0;
602 po->stats_u.stats3.tp_freeze_q_cnt = 0;
603 if (req_u->req3.tp_retire_blk_tov)
604 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
605 else
606 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
607 req_u->req3.tp_block_size);
608 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
609 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
610
611 prb_init_ft_ops(p1, req_u);
612 prb_setup_retire_blk_timer(po, tx_ring);
613 prb_open_block(p1, pbd);
614 }
615
616 /* Do NOT update the last_blk_num first.
617 * Assumes sk_buff_head lock is held.
618 */
619 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
620 {
621 mod_timer(&pkc->retire_blk_timer,
622 jiffies + pkc->tov_in_jiffies);
623 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
624 }
625
626 /*
627 * Timer logic:
628 * 1) We refresh the timer only when we open a block.
629 * By doing this we don't waste cycles refreshing the timer
630 * on packet-by-packet basis.
631 *
632 * With a 1MB block-size, on a 1Gbps line, it will take
633 * i) ~8 ms to fill a block + ii) memcpy etc.
634 * In this cut we are not accounting for the memcpy time.
635 *
636 * So, if the user sets the 'tmo' to 10ms then the timer
637 * will never fire while the block is still getting filled
638 * (which is what we want). However, the user could choose
639 * to close a block early and that's fine.
640 *
641 * But when the timer does fire, we check whether or not to refresh it.
642 * Since the tmo granularity is in msecs, it is not too expensive
643 * to refresh the timer, lets say every '8' msecs.
644 * Either the user can set the 'tmo' or we can derive it based on
645 * a) line-speed and b) block-size.
646 * prb_calc_retire_blk_tmo() calculates the tmo.
647 *
648 */
649 static void prb_retire_rx_blk_timer_expired(unsigned long data)
650 {
651 struct packet_sock *po = (struct packet_sock *)data;
652 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
653 unsigned int frozen;
654 struct tpacket_block_desc *pbd;
655
656 spin_lock(&po->sk.sk_receive_queue.lock);
657
658 frozen = prb_queue_frozen(pkc);
659 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
660
661 if (unlikely(pkc->delete_blk_timer))
662 goto out;
663
664 /* We only need to plug the race when the block is partially filled.
665 * tpacket_rcv:
666 * lock(); increment BLOCK_NUM_PKTS; unlock()
667 * copy_bits() is in progress ...
668 * timer fires on other cpu:
669 * we can't retire the current block because copy_bits
670 * is in progress.
671 *
672 */
673 if (BLOCK_NUM_PKTS(pbd)) {
674 while (atomic_read(&pkc->blk_fill_in_prog)) {
675 /* Waiting for skb_copy_bits to finish... */
676 cpu_relax();
677 }
678 }
679
680 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
681 if (!frozen) {
682 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
683 if (!prb_dispatch_next_block(pkc, po))
684 goto refresh_timer;
685 else
686 goto out;
687 } else {
688 /* Case 1. Queue was frozen because user-space was
689 * lagging behind.
690 */
691 if (prb_curr_blk_in_use(pkc, pbd)) {
692 /*
693 * Ok, user-space is still behind.
694 * So just refresh the timer.
695 */
696 goto refresh_timer;
697 } else {
698 /* Case 2. queue was frozen,user-space caught up,
699 * now the link went idle && the timer fired.
700 * We don't have a block to close.So we open this
701 * block and restart the timer.
702 * opening a block thaws the queue,restarts timer
703 * Thawing/timer-refresh is a side effect.
704 */
705 prb_open_block(pkc, pbd);
706 goto out;
707 }
708 }
709 }
710
711 refresh_timer:
712 _prb_refresh_rx_retire_blk_timer(pkc);
713
714 out:
715 spin_unlock(&po->sk.sk_receive_queue.lock);
716 }
717
718 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
719 struct tpacket_block_desc *pbd1, __u32 status)
720 {
721 /* Flush everything minus the block header */
722
723 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
724 u8 *start, *end;
725
726 start = (u8 *)pbd1;
727
728 /* Skip the block header(we know header WILL fit in 4K) */
729 start += PAGE_SIZE;
730
731 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
732 for (; start < end; start += PAGE_SIZE)
733 flush_dcache_page(pgv_to_page(start));
734
735 smp_wmb();
736 #endif
737
738 /* Now update the block status. */
739
740 BLOCK_STATUS(pbd1) = status;
741
742 /* Flush the block header */
743
744 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
745 start = (u8 *)pbd1;
746 flush_dcache_page(pgv_to_page(start));
747
748 smp_wmb();
749 #endif
750 }
751
752 /*
753 * Side effect:
754 *
755 * 1) flush the block
756 * 2) Increment active_blk_num
757 *
758 * Note:We DONT refresh the timer on purpose.
759 * Because almost always the next block will be opened.
760 */
761 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
762 struct tpacket_block_desc *pbd1,
763 struct packet_sock *po, unsigned int stat)
764 {
765 __u32 status = TP_STATUS_USER | stat;
766
767 struct tpacket3_hdr *last_pkt;
768 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
769
770 if (po->stats.tp_drops)
771 status |= TP_STATUS_LOSING;
772
773 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
774 last_pkt->tp_next_offset = 0;
775
776 /* Get the ts of the last pkt */
777 if (BLOCK_NUM_PKTS(pbd1)) {
778 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
779 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
780 } else {
781 /* Ok, we tmo'd - so get the current time */
782 struct timespec ts;
783 getnstimeofday(&ts);
784 h1->ts_last_pkt.ts_sec = ts.tv_sec;
785 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
786 }
787
788 smp_wmb();
789
790 /* Flush the block */
791 prb_flush_block(pkc1, pbd1, status);
792
793 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
794 }
795
796 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
797 {
798 pkc->reset_pending_on_curr_blk = 0;
799 }
800
801 /*
802 * Side effect of opening a block:
803 *
804 * 1) prb_queue is thawed.
805 * 2) retire_blk_timer is refreshed.
806 *
807 */
808 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
809 struct tpacket_block_desc *pbd1)
810 {
811 struct timespec ts;
812 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
813
814 smp_rmb();
815
816 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
817
818 /* We could have just memset this but we will lose the
819 * flexibility of making the priv area sticky
820 */
821 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
822 BLOCK_NUM_PKTS(pbd1) = 0;
823 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
824 getnstimeofday(&ts);
825 h1->ts_first_pkt.ts_sec = ts.tv_sec;
826 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
827 pkc1->pkblk_start = (char *)pbd1;
828 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
829 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
830 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
831 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
832 pbd1->version = pkc1->version;
833 pkc1->prev = pkc1->nxt_offset;
834 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
835 prb_thaw_queue(pkc1);
836 _prb_refresh_rx_retire_blk_timer(pkc1);
837
838 smp_wmb();
839
840 return;
841 }
842
843 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
844 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
845 dump_stack();
846 BUG();
847 }
848
849 /*
850 * Queue freeze logic:
851 * 1) Assume tp_block_nr = 8 blocks.
852 * 2) At time 't0', user opens Rx ring.
853 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854 * 4) user-space is either sleeping or processing block '0'.
855 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856 * it will close block-7,loop around and try to fill block '0'.
857 * call-flow:
858 * __packet_lookup_frame_in_block
859 * prb_retire_current_block()
860 * prb_dispatch_next_block()
861 * |->(BLOCK_STATUS == USER) evaluates to true
862 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
863 * 6) Now there are two cases:
864 * 6.1) Link goes idle right after the queue is frozen.
865 * But remember, the last open_block() refreshed the timer.
866 * When this timer expires,it will refresh itself so that we can
867 * re-open block-0 in near future.
868 * 6.2) Link is busy and keeps on receiving packets. This is a simple
869 * case and __packet_lookup_frame_in_block will check if block-0
870 * is free and can now be re-used.
871 */
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873 struct packet_sock *po)
874 {
875 pkc->reset_pending_on_curr_blk = 1;
876 po->stats_u.stats3.tp_freeze_q_cnt++;
877 }
878
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
880
881 /*
882 * If the next block is free then we will dispatch it
883 * and return a good offset.
884 * Else, we will freeze the queue.
885 * So, caller must check the return value.
886 */
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888 struct packet_sock *po)
889 {
890 struct tpacket_block_desc *pbd;
891
892 smp_rmb();
893
894 /* 1. Get current block num */
895 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
896
897 /* 2. If this block is currently in_use then freeze the queue */
898 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899 prb_freeze_queue(pkc, po);
900 return NULL;
901 }
902
903 /*
904 * 3.
905 * open this block and return the offset where the first packet
906 * needs to get stored.
907 */
908 prb_open_block(pkc, pbd);
909 return (void *)pkc->nxt_offset;
910 }
911
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913 struct packet_sock *po, unsigned int status)
914 {
915 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
916
917 /* retire/close the current block */
918 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
919 /*
920 * Plug the case where copy_bits() is in progress on
921 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922 * have space to copy the pkt in the current block and
923 * called prb_retire_current_block()
924 *
925 * We don't need to worry about the TMO case because
926 * the timer-handler already handled this case.
927 */
928 if (!(status & TP_STATUS_BLK_TMO)) {
929 while (atomic_read(&pkc->blk_fill_in_prog)) {
930 /* Waiting for skb_copy_bits to finish... */
931 cpu_relax();
932 }
933 }
934 prb_close_block(pkc, pbd, po, status);
935 return;
936 }
937
938 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
939 dump_stack();
940 BUG();
941 }
942
943 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
944 struct tpacket_block_desc *pbd)
945 {
946 return TP_STATUS_USER & BLOCK_STATUS(pbd);
947 }
948
949 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
950 {
951 return pkc->reset_pending_on_curr_blk;
952 }
953
954 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
955 {
956 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
957 atomic_dec(&pkc->blk_fill_in_prog);
958 }
959
960 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
961 struct tpacket3_hdr *ppd)
962 {
963 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
964 }
965
966 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
967 struct tpacket3_hdr *ppd)
968 {
969 ppd->hv1.tp_rxhash = 0;
970 }
971
972 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
973 struct tpacket3_hdr *ppd)
974 {
975 if (vlan_tx_tag_present(pkc->skb)) {
976 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
977 ppd->tp_status = TP_STATUS_VLAN_VALID;
978 } else {
979 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
980 }
981 }
982
983 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
984 struct tpacket3_hdr *ppd)
985 {
986 prb_fill_vlan_info(pkc, ppd);
987
988 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
989 prb_fill_rxhash(pkc, ppd);
990 else
991 prb_clear_rxhash(pkc, ppd);
992 }
993
994 static void prb_fill_curr_block(char *curr,
995 struct tpacket_kbdq_core *pkc,
996 struct tpacket_block_desc *pbd,
997 unsigned int len)
998 {
999 struct tpacket3_hdr *ppd;
1000
1001 ppd = (struct tpacket3_hdr *)curr;
1002 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1003 pkc->prev = curr;
1004 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006 BLOCK_NUM_PKTS(pbd) += 1;
1007 atomic_inc(&pkc->blk_fill_in_prog);
1008 prb_run_all_ft_ops(pkc, ppd);
1009 }
1010
1011 /* Assumes caller has the sk->rx_queue.lock */
1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013 struct sk_buff *skb,
1014 int status,
1015 unsigned int len
1016 )
1017 {
1018 struct tpacket_kbdq_core *pkc;
1019 struct tpacket_block_desc *pbd;
1020 char *curr, *end;
1021
1022 pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1023 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024
1025 /* Queue is frozen when user space is lagging behind */
1026 if (prb_queue_frozen(pkc)) {
1027 /*
1028 * Check if that last block which caused the queue to freeze,
1029 * is still in_use by user-space.
1030 */
1031 if (prb_curr_blk_in_use(pkc, pbd)) {
1032 /* Can't record this packet */
1033 return NULL;
1034 } else {
1035 /*
1036 * Ok, the block was released by user-space.
1037 * Now let's open that block.
1038 * opening a block also thaws the queue.
1039 * Thawing is a side effect.
1040 */
1041 prb_open_block(pkc, pbd);
1042 }
1043 }
1044
1045 smp_mb();
1046 curr = pkc->nxt_offset;
1047 pkc->skb = skb;
1048 end = (char *) ((char *)pbd + pkc->kblk_size);
1049
1050 /* first try the current block */
1051 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052 prb_fill_curr_block(curr, pkc, pbd, len);
1053 return (void *)curr;
1054 }
1055
1056 /* Ok, close the current block */
1057 prb_retire_current_block(pkc, po, 0);
1058
1059 /* Now, try to dispatch the next block */
1060 curr = (char *)prb_dispatch_next_block(pkc, po);
1061 if (curr) {
1062 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063 prb_fill_curr_block(curr, pkc, pbd, len);
1064 return (void *)curr;
1065 }
1066
1067 /*
1068 * No free blocks are available.user_space hasn't caught up yet.
1069 * Queue was just frozen and now this packet will get dropped.
1070 */
1071 return NULL;
1072 }
1073
1074 static void *packet_current_rx_frame(struct packet_sock *po,
1075 struct sk_buff *skb,
1076 int status, unsigned int len)
1077 {
1078 char *curr = NULL;
1079 switch (po->tp_version) {
1080 case TPACKET_V1:
1081 case TPACKET_V2:
1082 curr = packet_lookup_frame(po, &po->rx_ring,
1083 po->rx_ring.head, status);
1084 return curr;
1085 case TPACKET_V3:
1086 return __packet_lookup_frame_in_block(po, skb, status, len);
1087 default:
1088 WARN(1, "TPACKET version not supported\n");
1089 BUG();
1090 return 0;
1091 }
1092 }
1093
1094 static void *prb_lookup_block(struct packet_sock *po,
1095 struct packet_ring_buffer *rb,
1096 unsigned int previous,
1097 int status)
1098 {
1099 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1100 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1101
1102 if (status != BLOCK_STATUS(pbd))
1103 return NULL;
1104 return pbd;
1105 }
1106
1107 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1108 {
1109 unsigned int prev;
1110 if (rb->prb_bdqc.kactive_blk_num)
1111 prev = rb->prb_bdqc.kactive_blk_num-1;
1112 else
1113 prev = rb->prb_bdqc.knum_blocks-1;
1114 return prev;
1115 }
1116
1117 /* Assumes caller has held the rx_queue.lock */
1118 static void *__prb_previous_block(struct packet_sock *po,
1119 struct packet_ring_buffer *rb,
1120 int status)
1121 {
1122 unsigned int previous = prb_previous_blk_num(rb);
1123 return prb_lookup_block(po, rb, previous, status);
1124 }
1125
1126 static void *packet_previous_rx_frame(struct packet_sock *po,
1127 struct packet_ring_buffer *rb,
1128 int status)
1129 {
1130 if (po->tp_version <= TPACKET_V2)
1131 return packet_previous_frame(po, rb, status);
1132
1133 return __prb_previous_block(po, rb, status);
1134 }
1135
1136 static void packet_increment_rx_head(struct packet_sock *po,
1137 struct packet_ring_buffer *rb)
1138 {
1139 switch (po->tp_version) {
1140 case TPACKET_V1:
1141 case TPACKET_V2:
1142 return packet_increment_head(rb);
1143 case TPACKET_V3:
1144 default:
1145 WARN(1, "TPACKET version not supported.\n");
1146 BUG();
1147 return;
1148 }
1149 }
1150
1151 static void *packet_previous_frame(struct packet_sock *po,
1152 struct packet_ring_buffer *rb,
1153 int status)
1154 {
1155 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156 return packet_lookup_frame(po, rb, previous, status);
1157 }
1158
1159 static void packet_increment_head(struct packet_ring_buffer *buff)
1160 {
1161 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1162 }
1163
1164 static void packet_sock_destruct(struct sock *sk)
1165 {
1166 skb_queue_purge(&sk->sk_error_queue);
1167
1168 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1169 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1170
1171 if (!sock_flag(sk, SOCK_DEAD)) {
1172 pr_err("Attempt to release alive packet socket: %p\n", sk);
1173 return;
1174 }
1175
1176 sk_refcnt_debug_dec(sk);
1177 }
1178
1179 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1180 {
1181 int x = atomic_read(&f->rr_cur) + 1;
1182
1183 if (x >= num)
1184 x = 0;
1185
1186 return x;
1187 }
1188
1189 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1190 {
1191 u32 idx, hash = skb->rxhash;
1192
1193 idx = ((u64)hash * num) >> 32;
1194
1195 return f->arr[idx];
1196 }
1197
1198 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1199 {
1200 int cur, old;
1201
1202 cur = atomic_read(&f->rr_cur);
1203 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1204 fanout_rr_next(f, num))) != cur)
1205 cur = old;
1206 return f->arr[cur];
1207 }
1208
1209 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1210 {
1211 unsigned int cpu = smp_processor_id();
1212
1213 return f->arr[cpu % num];
1214 }
1215
1216 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1217 struct packet_type *pt, struct net_device *orig_dev)
1218 {
1219 struct packet_fanout *f = pt->af_packet_priv;
1220 unsigned int num = f->num_members;
1221 struct packet_sock *po;
1222 struct sock *sk;
1223
1224 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1225 !num) {
1226 kfree_skb(skb);
1227 return 0;
1228 }
1229
1230 switch (f->type) {
1231 case PACKET_FANOUT_HASH:
1232 default:
1233 if (f->defrag) {
1234 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1235 if (!skb)
1236 return 0;
1237 }
1238 skb_get_rxhash(skb);
1239 sk = fanout_demux_hash(f, skb, num);
1240 break;
1241 case PACKET_FANOUT_LB:
1242 sk = fanout_demux_lb(f, skb, num);
1243 break;
1244 case PACKET_FANOUT_CPU:
1245 sk = fanout_demux_cpu(f, skb, num);
1246 break;
1247 }
1248
1249 po = pkt_sk(sk);
1250
1251 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1252 }
1253
1254 static DEFINE_MUTEX(fanout_mutex);
1255 static LIST_HEAD(fanout_list);
1256
1257 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1258 {
1259 struct packet_fanout *f = po->fanout;
1260
1261 spin_lock(&f->lock);
1262 f->arr[f->num_members] = sk;
1263 smp_wmb();
1264 f->num_members++;
1265 spin_unlock(&f->lock);
1266 }
1267
1268 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1269 {
1270 struct packet_fanout *f = po->fanout;
1271 int i;
1272
1273 spin_lock(&f->lock);
1274 for (i = 0; i < f->num_members; i++) {
1275 if (f->arr[i] == sk)
1276 break;
1277 }
1278 BUG_ON(i >= f->num_members);
1279 f->arr[i] = f->arr[f->num_members - 1];
1280 f->num_members--;
1281 spin_unlock(&f->lock);
1282 }
1283
1284 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1285 {
1286 struct packet_sock *po = pkt_sk(sk);
1287 struct packet_fanout *f, *match;
1288 u8 type = type_flags & 0xff;
1289 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1290 int err;
1291
1292 switch (type) {
1293 case PACKET_FANOUT_HASH:
1294 case PACKET_FANOUT_LB:
1295 case PACKET_FANOUT_CPU:
1296 break;
1297 default:
1298 return -EINVAL;
1299 }
1300
1301 if (!po->running)
1302 return -EINVAL;
1303
1304 if (po->fanout)
1305 return -EALREADY;
1306
1307 mutex_lock(&fanout_mutex);
1308 match = NULL;
1309 list_for_each_entry(f, &fanout_list, list) {
1310 if (f->id == id &&
1311 read_pnet(&f->net) == sock_net(sk)) {
1312 match = f;
1313 break;
1314 }
1315 }
1316 err = -EINVAL;
1317 if (match && match->defrag != defrag)
1318 goto out;
1319 if (!match) {
1320 err = -ENOMEM;
1321 match = kzalloc(sizeof(*match), GFP_KERNEL);
1322 if (!match)
1323 goto out;
1324 write_pnet(&match->net, sock_net(sk));
1325 match->id = id;
1326 match->type = type;
1327 match->defrag = defrag;
1328 atomic_set(&match->rr_cur, 0);
1329 INIT_LIST_HEAD(&match->list);
1330 spin_lock_init(&match->lock);
1331 atomic_set(&match->sk_ref, 0);
1332 match->prot_hook.type = po->prot_hook.type;
1333 match->prot_hook.dev = po->prot_hook.dev;
1334 match->prot_hook.func = packet_rcv_fanout;
1335 match->prot_hook.af_packet_priv = match;
1336 dev_add_pack(&match->prot_hook);
1337 list_add(&match->list, &fanout_list);
1338 }
1339 err = -EINVAL;
1340 if (match->type == type &&
1341 match->prot_hook.type == po->prot_hook.type &&
1342 match->prot_hook.dev == po->prot_hook.dev) {
1343 err = -ENOSPC;
1344 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1345 __dev_remove_pack(&po->prot_hook);
1346 po->fanout = match;
1347 atomic_inc(&match->sk_ref);
1348 __fanout_link(sk, po);
1349 err = 0;
1350 }
1351 }
1352 out:
1353 mutex_unlock(&fanout_mutex);
1354 return err;
1355 }
1356
1357 static void fanout_release(struct sock *sk)
1358 {
1359 struct packet_sock *po = pkt_sk(sk);
1360 struct packet_fanout *f;
1361
1362 f = po->fanout;
1363 if (!f)
1364 return;
1365
1366 po->fanout = NULL;
1367
1368 mutex_lock(&fanout_mutex);
1369 if (atomic_dec_and_test(&f->sk_ref)) {
1370 list_del(&f->list);
1371 dev_remove_pack(&f->prot_hook);
1372 kfree(f);
1373 }
1374 mutex_unlock(&fanout_mutex);
1375 }
1376
1377 static const struct proto_ops packet_ops;
1378
1379 static const struct proto_ops packet_ops_spkt;
1380
1381 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1382 struct packet_type *pt, struct net_device *orig_dev)
1383 {
1384 struct sock *sk;
1385 struct sockaddr_pkt *spkt;
1386
1387 /*
1388 * When we registered the protocol we saved the socket in the data
1389 * field for just this event.
1390 */
1391
1392 sk = pt->af_packet_priv;
1393
1394 /*
1395 * Yank back the headers [hope the device set this
1396 * right or kerboom...]
1397 *
1398 * Incoming packets have ll header pulled,
1399 * push it back.
1400 *
1401 * For outgoing ones skb->data == skb_mac_header(skb)
1402 * so that this procedure is noop.
1403 */
1404
1405 if (skb->pkt_type == PACKET_LOOPBACK)
1406 goto out;
1407
1408 if (!net_eq(dev_net(dev), sock_net(sk)))
1409 goto out;
1410
1411 skb = skb_share_check(skb, GFP_ATOMIC);
1412 if (skb == NULL)
1413 goto oom;
1414
1415 /* drop any routing info */
1416 skb_dst_drop(skb);
1417
1418 /* drop conntrack reference */
1419 nf_reset(skb);
1420
1421 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1422
1423 skb_push(skb, skb->data - skb_mac_header(skb));
1424
1425 /*
1426 * The SOCK_PACKET socket receives _all_ frames.
1427 */
1428
1429 spkt->spkt_family = dev->type;
1430 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1431 spkt->spkt_protocol = skb->protocol;
1432
1433 /*
1434 * Charge the memory to the socket. This is done specifically
1435 * to prevent sockets using all the memory up.
1436 */
1437
1438 if (sock_queue_rcv_skb(sk, skb) == 0)
1439 return 0;
1440
1441 out:
1442 kfree_skb(skb);
1443 oom:
1444 return 0;
1445 }
1446
1447
1448 /*
1449 * Output a raw packet to a device layer. This bypasses all the other
1450 * protocol layers and you must therefore supply it with a complete frame
1451 */
1452
1453 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1454 struct msghdr *msg, size_t len)
1455 {
1456 struct sock *sk = sock->sk;
1457 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1458 struct sk_buff *skb = NULL;
1459 struct net_device *dev;
1460 __be16 proto = 0;
1461 int err;
1462
1463 /*
1464 * Get and verify the address.
1465 */
1466
1467 if (saddr) {
1468 if (msg->msg_namelen < sizeof(struct sockaddr))
1469 return -EINVAL;
1470 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1471 proto = saddr->spkt_protocol;
1472 } else
1473 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1474
1475 /*
1476 * Find the device first to size check it
1477 */
1478
1479 saddr->spkt_device[13] = 0;
1480 retry:
1481 rcu_read_lock();
1482 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1483 err = -ENODEV;
1484 if (dev == NULL)
1485 goto out_unlock;
1486
1487 err = -ENETDOWN;
1488 if (!(dev->flags & IFF_UP))
1489 goto out_unlock;
1490
1491 /*
1492 * You may not queue a frame bigger than the mtu. This is the lowest level
1493 * raw protocol and you must do your own fragmentation at this level.
1494 */
1495
1496 err = -EMSGSIZE;
1497 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1498 goto out_unlock;
1499
1500 if (!skb) {
1501 size_t reserved = LL_RESERVED_SPACE(dev);
1502 int tlen = dev->needed_tailroom;
1503 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1504
1505 rcu_read_unlock();
1506 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1507 if (skb == NULL)
1508 return -ENOBUFS;
1509 /* FIXME: Save some space for broken drivers that write a hard
1510 * header at transmission time by themselves. PPP is the notable
1511 * one here. This should really be fixed at the driver level.
1512 */
1513 skb_reserve(skb, reserved);
1514 skb_reset_network_header(skb);
1515
1516 /* Try to align data part correctly */
1517 if (hhlen) {
1518 skb->data -= hhlen;
1519 skb->tail -= hhlen;
1520 if (len < hhlen)
1521 skb_reset_network_header(skb);
1522 }
1523 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1524 if (err)
1525 goto out_free;
1526 goto retry;
1527 }
1528
1529 if (len > (dev->mtu + dev->hard_header_len)) {
1530 /* Earlier code assumed this would be a VLAN pkt,
1531 * double-check this now that we have the actual
1532 * packet in hand.
1533 */
1534 struct ethhdr *ehdr;
1535 skb_reset_mac_header(skb);
1536 ehdr = eth_hdr(skb);
1537 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1538 err = -EMSGSIZE;
1539 goto out_unlock;
1540 }
1541 }
1542
1543 skb->protocol = proto;
1544 skb->dev = dev;
1545 skb->priority = sk->sk_priority;
1546 skb->mark = sk->sk_mark;
1547 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1548 if (err < 0)
1549 goto out_unlock;
1550
1551 dev_queue_xmit(skb);
1552 rcu_read_unlock();
1553 return len;
1554
1555 out_unlock:
1556 rcu_read_unlock();
1557 out_free:
1558 kfree_skb(skb);
1559 return err;
1560 }
1561
1562 static unsigned int run_filter(const struct sk_buff *skb,
1563 const struct sock *sk,
1564 unsigned int res)
1565 {
1566 struct sk_filter *filter;
1567
1568 rcu_read_lock();
1569 filter = rcu_dereference(sk->sk_filter);
1570 if (filter != NULL)
1571 res = SK_RUN_FILTER(filter, skb);
1572 rcu_read_unlock();
1573
1574 return res;
1575 }
1576
1577 /*
1578 * This function makes lazy skb cloning in hope that most of packets
1579 * are discarded by BPF.
1580 *
1581 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1582 * and skb->cb are mangled. It works because (and until) packets
1583 * falling here are owned by current CPU. Output packets are cloned
1584 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1585 * sequencially, so that if we return skb to original state on exit,
1586 * we will not harm anyone.
1587 */
1588
1589 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1590 struct packet_type *pt, struct net_device *orig_dev)
1591 {
1592 struct sock *sk;
1593 struct sockaddr_ll *sll;
1594 struct packet_sock *po;
1595 u8 *skb_head = skb->data;
1596 int skb_len = skb->len;
1597 unsigned int snaplen, res;
1598
1599 if (skb->pkt_type == PACKET_LOOPBACK)
1600 goto drop;
1601
1602 sk = pt->af_packet_priv;
1603 po = pkt_sk(sk);
1604
1605 if (!net_eq(dev_net(dev), sock_net(sk)))
1606 goto drop;
1607
1608 skb->dev = dev;
1609
1610 if (dev->header_ops) {
1611 /* The device has an explicit notion of ll header,
1612 * exported to higher levels.
1613 *
1614 * Otherwise, the device hides details of its frame
1615 * structure, so that corresponding packet head is
1616 * never delivered to user.
1617 */
1618 if (sk->sk_type != SOCK_DGRAM)
1619 skb_push(skb, skb->data - skb_mac_header(skb));
1620 else if (skb->pkt_type == PACKET_OUTGOING) {
1621 /* Special case: outgoing packets have ll header at head */
1622 skb_pull(skb, skb_network_offset(skb));
1623 }
1624 }
1625
1626 snaplen = skb->len;
1627
1628 res = run_filter(skb, sk, snaplen);
1629 if (!res)
1630 goto drop_n_restore;
1631 if (snaplen > res)
1632 snaplen = res;
1633
1634 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1635 goto drop_n_acct;
1636
1637 if (skb_shared(skb)) {
1638 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1639 if (nskb == NULL)
1640 goto drop_n_acct;
1641
1642 if (skb_head != skb->data) {
1643 skb->data = skb_head;
1644 skb->len = skb_len;
1645 }
1646 kfree_skb(skb);
1647 skb = nskb;
1648 }
1649
1650 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1651 sizeof(skb->cb));
1652
1653 sll = &PACKET_SKB_CB(skb)->sa.ll;
1654 sll->sll_family = AF_PACKET;
1655 sll->sll_hatype = dev->type;
1656 sll->sll_protocol = skb->protocol;
1657 sll->sll_pkttype = skb->pkt_type;
1658 if (unlikely(po->origdev))
1659 sll->sll_ifindex = orig_dev->ifindex;
1660 else
1661 sll->sll_ifindex = dev->ifindex;
1662
1663 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1664
1665 PACKET_SKB_CB(skb)->origlen = skb->len;
1666
1667 if (pskb_trim(skb, snaplen))
1668 goto drop_n_acct;
1669
1670 skb_set_owner_r(skb, sk);
1671 skb->dev = NULL;
1672 skb_dst_drop(skb);
1673
1674 /* drop conntrack reference */
1675 nf_reset(skb);
1676
1677 spin_lock(&sk->sk_receive_queue.lock);
1678 po->stats.tp_packets++;
1679 skb->dropcount = atomic_read(&sk->sk_drops);
1680 __skb_queue_tail(&sk->sk_receive_queue, skb);
1681 spin_unlock(&sk->sk_receive_queue.lock);
1682 sk->sk_data_ready(sk, skb->len);
1683 return 0;
1684
1685 drop_n_acct:
1686 spin_lock(&sk->sk_receive_queue.lock);
1687 po->stats.tp_drops++;
1688 atomic_inc(&sk->sk_drops);
1689 spin_unlock(&sk->sk_receive_queue.lock);
1690
1691 drop_n_restore:
1692 if (skb_head != skb->data && skb_shared(skb)) {
1693 skb->data = skb_head;
1694 skb->len = skb_len;
1695 }
1696 drop:
1697 consume_skb(skb);
1698 return 0;
1699 }
1700
1701 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1702 struct packet_type *pt, struct net_device *orig_dev)
1703 {
1704 struct sock *sk;
1705 struct packet_sock *po;
1706 struct sockaddr_ll *sll;
1707 union {
1708 struct tpacket_hdr *h1;
1709 struct tpacket2_hdr *h2;
1710 struct tpacket3_hdr *h3;
1711 void *raw;
1712 } h;
1713 u8 *skb_head = skb->data;
1714 int skb_len = skb->len;
1715 unsigned int snaplen, res;
1716 unsigned long status = TP_STATUS_USER;
1717 unsigned short macoff, netoff, hdrlen;
1718 struct sk_buff *copy_skb = NULL;
1719 struct timeval tv;
1720 struct timespec ts;
1721 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1722
1723 if (skb->pkt_type == PACKET_LOOPBACK)
1724 goto drop;
1725
1726 sk = pt->af_packet_priv;
1727 po = pkt_sk(sk);
1728
1729 if (!net_eq(dev_net(dev), sock_net(sk)))
1730 goto drop;
1731
1732 if (dev->header_ops) {
1733 if (sk->sk_type != SOCK_DGRAM)
1734 skb_push(skb, skb->data - skb_mac_header(skb));
1735 else if (skb->pkt_type == PACKET_OUTGOING) {
1736 /* Special case: outgoing packets have ll header at head */
1737 skb_pull(skb, skb_network_offset(skb));
1738 }
1739 }
1740
1741 if (skb->ip_summed == CHECKSUM_PARTIAL)
1742 status |= TP_STATUS_CSUMNOTREADY;
1743
1744 snaplen = skb->len;
1745
1746 res = run_filter(skb, sk, snaplen);
1747 if (!res)
1748 goto drop_n_restore;
1749 if (snaplen > res)
1750 snaplen = res;
1751
1752 if (sk->sk_type == SOCK_DGRAM) {
1753 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1754 po->tp_reserve;
1755 } else {
1756 unsigned maclen = skb_network_offset(skb);
1757 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1758 (maclen < 16 ? 16 : maclen)) +
1759 po->tp_reserve;
1760 macoff = netoff - maclen;
1761 }
1762 if (po->tp_version <= TPACKET_V2) {
1763 if (macoff + snaplen > po->rx_ring.frame_size) {
1764 if (po->copy_thresh &&
1765 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1766 if (skb_shared(skb)) {
1767 copy_skb = skb_clone(skb, GFP_ATOMIC);
1768 } else {
1769 copy_skb = skb_get(skb);
1770 skb_head = skb->data;
1771 }
1772 if (copy_skb)
1773 skb_set_owner_r(copy_skb, sk);
1774 }
1775 snaplen = po->rx_ring.frame_size - macoff;
1776 if ((int)snaplen < 0)
1777 snaplen = 0;
1778 }
1779 }
1780 spin_lock(&sk->sk_receive_queue.lock);
1781 h.raw = packet_current_rx_frame(po, skb,
1782 TP_STATUS_KERNEL, (macoff+snaplen));
1783 if (!h.raw)
1784 goto ring_is_full;
1785 if (po->tp_version <= TPACKET_V2) {
1786 packet_increment_rx_head(po, &po->rx_ring);
1787 /*
1788 * LOSING will be reported till you read the stats,
1789 * because it's COR - Clear On Read.
1790 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1791 * at packet level.
1792 */
1793 if (po->stats.tp_drops)
1794 status |= TP_STATUS_LOSING;
1795 }
1796 po->stats.tp_packets++;
1797 if (copy_skb) {
1798 status |= TP_STATUS_COPY;
1799 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1800 }
1801 spin_unlock(&sk->sk_receive_queue.lock);
1802
1803 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1804
1805 switch (po->tp_version) {
1806 case TPACKET_V1:
1807 h.h1->tp_len = skb->len;
1808 h.h1->tp_snaplen = snaplen;
1809 h.h1->tp_mac = macoff;
1810 h.h1->tp_net = netoff;
1811 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1812 && shhwtstamps->syststamp.tv64)
1813 tv = ktime_to_timeval(shhwtstamps->syststamp);
1814 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1815 && shhwtstamps->hwtstamp.tv64)
1816 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1817 else if (skb->tstamp.tv64)
1818 tv = ktime_to_timeval(skb->tstamp);
1819 else
1820 do_gettimeofday(&tv);
1821 h.h1->tp_sec = tv.tv_sec;
1822 h.h1->tp_usec = tv.tv_usec;
1823 hdrlen = sizeof(*h.h1);
1824 break;
1825 case TPACKET_V2:
1826 h.h2->tp_len = skb->len;
1827 h.h2->tp_snaplen = snaplen;
1828 h.h2->tp_mac = macoff;
1829 h.h2->tp_net = netoff;
1830 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1831 && shhwtstamps->syststamp.tv64)
1832 ts = ktime_to_timespec(shhwtstamps->syststamp);
1833 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1834 && shhwtstamps->hwtstamp.tv64)
1835 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1836 else if (skb->tstamp.tv64)
1837 ts = ktime_to_timespec(skb->tstamp);
1838 else
1839 getnstimeofday(&ts);
1840 h.h2->tp_sec = ts.tv_sec;
1841 h.h2->tp_nsec = ts.tv_nsec;
1842 if (vlan_tx_tag_present(skb)) {
1843 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1844 status |= TP_STATUS_VLAN_VALID;
1845 } else {
1846 h.h2->tp_vlan_tci = 0;
1847 }
1848 h.h2->tp_padding = 0;
1849 hdrlen = sizeof(*h.h2);
1850 break;
1851 case TPACKET_V3:
1852 /* tp_nxt_offset,vlan are already populated above.
1853 * So DONT clear those fields here
1854 */
1855 h.h3->tp_status |= status;
1856 h.h3->tp_len = skb->len;
1857 h.h3->tp_snaplen = snaplen;
1858 h.h3->tp_mac = macoff;
1859 h.h3->tp_net = netoff;
1860 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1861 && shhwtstamps->syststamp.tv64)
1862 ts = ktime_to_timespec(shhwtstamps->syststamp);
1863 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1864 && shhwtstamps->hwtstamp.tv64)
1865 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1866 else if (skb->tstamp.tv64)
1867 ts = ktime_to_timespec(skb->tstamp);
1868 else
1869 getnstimeofday(&ts);
1870 h.h3->tp_sec = ts.tv_sec;
1871 h.h3->tp_nsec = ts.tv_nsec;
1872 hdrlen = sizeof(*h.h3);
1873 break;
1874 default:
1875 BUG();
1876 }
1877
1878 sll = h.raw + TPACKET_ALIGN(hdrlen);
1879 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1880 sll->sll_family = AF_PACKET;
1881 sll->sll_hatype = dev->type;
1882 sll->sll_protocol = skb->protocol;
1883 sll->sll_pkttype = skb->pkt_type;
1884 if (unlikely(po->origdev))
1885 sll->sll_ifindex = orig_dev->ifindex;
1886 else
1887 sll->sll_ifindex = dev->ifindex;
1888
1889 smp_mb();
1890 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1891 {
1892 u8 *start, *end;
1893
1894 if (po->tp_version <= TPACKET_V2) {
1895 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1896 + macoff + snaplen);
1897 for (start = h.raw; start < end; start += PAGE_SIZE)
1898 flush_dcache_page(pgv_to_page(start));
1899 }
1900 smp_wmb();
1901 }
1902 #endif
1903 if (po->tp_version <= TPACKET_V2)
1904 __packet_set_status(po, h.raw, status);
1905 else
1906 prb_clear_blk_fill_status(&po->rx_ring);
1907
1908 sk->sk_data_ready(sk, 0);
1909
1910 drop_n_restore:
1911 if (skb_head != skb->data && skb_shared(skb)) {
1912 skb->data = skb_head;
1913 skb->len = skb_len;
1914 }
1915 drop:
1916 kfree_skb(skb);
1917 return 0;
1918
1919 ring_is_full:
1920 po->stats.tp_drops++;
1921 spin_unlock(&sk->sk_receive_queue.lock);
1922
1923 sk->sk_data_ready(sk, 0);
1924 kfree_skb(copy_skb);
1925 goto drop_n_restore;
1926 }
1927
1928 static void tpacket_destruct_skb(struct sk_buff *skb)
1929 {
1930 struct packet_sock *po = pkt_sk(skb->sk);
1931 void *ph;
1932
1933 if (likely(po->tx_ring.pg_vec)) {
1934 ph = skb_shinfo(skb)->destructor_arg;
1935 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1936 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1937 atomic_dec(&po->tx_ring.pending);
1938 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1939 }
1940
1941 sock_wfree(skb);
1942 }
1943
1944 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1945 void *frame, struct net_device *dev, int size_max,
1946 __be16 proto, unsigned char *addr, int hlen)
1947 {
1948 union {
1949 struct tpacket_hdr *h1;
1950 struct tpacket2_hdr *h2;
1951 void *raw;
1952 } ph;
1953 int to_write, offset, len, tp_len, nr_frags, len_max;
1954 struct socket *sock = po->sk.sk_socket;
1955 struct page *page;
1956 void *data;
1957 int err;
1958
1959 ph.raw = frame;
1960
1961 skb->protocol = proto;
1962 skb->dev = dev;
1963 skb->priority = po->sk.sk_priority;
1964 skb->mark = po->sk.sk_mark;
1965 skb_shinfo(skb)->destructor_arg = ph.raw;
1966
1967 switch (po->tp_version) {
1968 case TPACKET_V2:
1969 tp_len = ph.h2->tp_len;
1970 break;
1971 default:
1972 tp_len = ph.h1->tp_len;
1973 break;
1974 }
1975 if (unlikely(tp_len > size_max)) {
1976 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1977 return -EMSGSIZE;
1978 }
1979
1980 skb_reserve(skb, hlen);
1981 skb_reset_network_header(skb);
1982
1983 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1984 to_write = tp_len;
1985
1986 if (sock->type == SOCK_DGRAM) {
1987 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1988 NULL, tp_len);
1989 if (unlikely(err < 0))
1990 return -EINVAL;
1991 } else if (dev->hard_header_len) {
1992 /* net device doesn't like empty head */
1993 if (unlikely(tp_len <= dev->hard_header_len)) {
1994 pr_err("packet size is too short (%d < %d)\n",
1995 tp_len, dev->hard_header_len);
1996 return -EINVAL;
1997 }
1998
1999 skb_push(skb, dev->hard_header_len);
2000 err = skb_store_bits(skb, 0, data,
2001 dev->hard_header_len);
2002 if (unlikely(err))
2003 return err;
2004
2005 data += dev->hard_header_len;
2006 to_write -= dev->hard_header_len;
2007 }
2008
2009 err = -EFAULT;
2010 offset = offset_in_page(data);
2011 len_max = PAGE_SIZE - offset;
2012 len = ((to_write > len_max) ? len_max : to_write);
2013
2014 skb->data_len = to_write;
2015 skb->len += to_write;
2016 skb->truesize += to_write;
2017 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2018
2019 while (likely(to_write)) {
2020 nr_frags = skb_shinfo(skb)->nr_frags;
2021
2022 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2023 pr_err("Packet exceed the number of skb frags(%lu)\n",
2024 MAX_SKB_FRAGS);
2025 return -EFAULT;
2026 }
2027
2028 page = pgv_to_page(data);
2029 data += len;
2030 flush_dcache_page(page);
2031 get_page(page);
2032 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2033 to_write -= len;
2034 offset = 0;
2035 len_max = PAGE_SIZE;
2036 len = ((to_write > len_max) ? len_max : to_write);
2037 }
2038
2039 return tp_len;
2040 }
2041
2042 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2043 {
2044 struct sk_buff *skb;
2045 struct net_device *dev;
2046 __be16 proto;
2047 bool need_rls_dev = false;
2048 int err, reserve = 0;
2049 void *ph;
2050 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2051 int tp_len, size_max;
2052 unsigned char *addr;
2053 int len_sum = 0;
2054 int status = 0;
2055 int hlen, tlen;
2056
2057 mutex_lock(&po->pg_vec_lock);
2058
2059 err = -EBUSY;
2060 if (saddr == NULL) {
2061 dev = po->prot_hook.dev;
2062 proto = po->num;
2063 addr = NULL;
2064 } else {
2065 err = -EINVAL;
2066 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2067 goto out;
2068 if (msg->msg_namelen < (saddr->sll_halen
2069 + offsetof(struct sockaddr_ll,
2070 sll_addr)))
2071 goto out;
2072 proto = saddr->sll_protocol;
2073 addr = saddr->sll_addr;
2074 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2075 need_rls_dev = true;
2076 }
2077
2078 err = -ENXIO;
2079 if (unlikely(dev == NULL))
2080 goto out;
2081
2082 reserve = dev->hard_header_len;
2083
2084 err = -ENETDOWN;
2085 if (unlikely(!(dev->flags & IFF_UP)))
2086 goto out_put;
2087
2088 size_max = po->tx_ring.frame_size
2089 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2090
2091 if (size_max > dev->mtu + reserve)
2092 size_max = dev->mtu + reserve;
2093
2094 do {
2095 ph = packet_current_frame(po, &po->tx_ring,
2096 TP_STATUS_SEND_REQUEST);
2097
2098 if (unlikely(ph == NULL)) {
2099 schedule();
2100 continue;
2101 }
2102
2103 status = TP_STATUS_SEND_REQUEST;
2104 hlen = LL_RESERVED_SPACE(dev);
2105 tlen = dev->needed_tailroom;
2106 skb = sock_alloc_send_skb(&po->sk,
2107 hlen + tlen + sizeof(struct sockaddr_ll),
2108 0, &err);
2109
2110 if (unlikely(skb == NULL))
2111 goto out_status;
2112
2113 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2114 addr, hlen);
2115
2116 if (unlikely(tp_len < 0)) {
2117 if (po->tp_loss) {
2118 __packet_set_status(po, ph,
2119 TP_STATUS_AVAILABLE);
2120 packet_increment_head(&po->tx_ring);
2121 kfree_skb(skb);
2122 continue;
2123 } else {
2124 status = TP_STATUS_WRONG_FORMAT;
2125 err = tp_len;
2126 goto out_status;
2127 }
2128 }
2129
2130 skb->destructor = tpacket_destruct_skb;
2131 __packet_set_status(po, ph, TP_STATUS_SENDING);
2132 atomic_inc(&po->tx_ring.pending);
2133
2134 status = TP_STATUS_SEND_REQUEST;
2135 err = dev_queue_xmit(skb);
2136 if (unlikely(err > 0)) {
2137 err = net_xmit_errno(err);
2138 if (err && __packet_get_status(po, ph) ==
2139 TP_STATUS_AVAILABLE) {
2140 /* skb was destructed already */
2141 skb = NULL;
2142 goto out_status;
2143 }
2144 /*
2145 * skb was dropped but not destructed yet;
2146 * let's treat it like congestion or err < 0
2147 */
2148 err = 0;
2149 }
2150 packet_increment_head(&po->tx_ring);
2151 len_sum += tp_len;
2152 } while (likely((ph != NULL) ||
2153 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2154 (atomic_read(&po->tx_ring.pending))))
2155 );
2156
2157 err = len_sum;
2158 goto out_put;
2159
2160 out_status:
2161 __packet_set_status(po, ph, status);
2162 kfree_skb(skb);
2163 out_put:
2164 if (need_rls_dev)
2165 dev_put(dev);
2166 out:
2167 mutex_unlock(&po->pg_vec_lock);
2168 return err;
2169 }
2170
2171 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2172 size_t reserve, size_t len,
2173 size_t linear, int noblock,
2174 int *err)
2175 {
2176 struct sk_buff *skb;
2177
2178 /* Under a page? Don't bother with paged skb. */
2179 if (prepad + len < PAGE_SIZE || !linear)
2180 linear = len;
2181
2182 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2183 err);
2184 if (!skb)
2185 return NULL;
2186
2187 skb_reserve(skb, reserve);
2188 skb_put(skb, linear);
2189 skb->data_len = len - linear;
2190 skb->len += len - linear;
2191
2192 return skb;
2193 }
2194
2195 static int packet_snd(struct socket *sock,
2196 struct msghdr *msg, size_t len)
2197 {
2198 struct sock *sk = sock->sk;
2199 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2200 struct sk_buff *skb;
2201 struct net_device *dev;
2202 __be16 proto;
2203 bool need_rls_dev = false;
2204 unsigned char *addr;
2205 int err, reserve = 0;
2206 struct virtio_net_hdr vnet_hdr = { 0 };
2207 int offset = 0;
2208 int vnet_hdr_len;
2209 struct packet_sock *po = pkt_sk(sk);
2210 unsigned short gso_type = 0;
2211 int hlen, tlen;
2212
2213 /*
2214 * Get and verify the address.
2215 */
2216
2217 if (saddr == NULL) {
2218 dev = po->prot_hook.dev;
2219 proto = po->num;
2220 addr = NULL;
2221 } else {
2222 err = -EINVAL;
2223 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2224 goto out;
2225 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2226 goto out;
2227 proto = saddr->sll_protocol;
2228 addr = saddr->sll_addr;
2229 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2230 need_rls_dev = true;
2231 }
2232
2233 err = -ENXIO;
2234 if (dev == NULL)
2235 goto out_unlock;
2236 if (sock->type == SOCK_RAW)
2237 reserve = dev->hard_header_len;
2238
2239 err = -ENETDOWN;
2240 if (!(dev->flags & IFF_UP))
2241 goto out_unlock;
2242
2243 if (po->has_vnet_hdr) {
2244 vnet_hdr_len = sizeof(vnet_hdr);
2245
2246 err = -EINVAL;
2247 if (len < vnet_hdr_len)
2248 goto out_unlock;
2249
2250 len -= vnet_hdr_len;
2251
2252 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2253 vnet_hdr_len);
2254 if (err < 0)
2255 goto out_unlock;
2256
2257 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2258 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2259 vnet_hdr.hdr_len))
2260 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2261 vnet_hdr.csum_offset + 2;
2262
2263 err = -EINVAL;
2264 if (vnet_hdr.hdr_len > len)
2265 goto out_unlock;
2266
2267 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2268 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2269 case VIRTIO_NET_HDR_GSO_TCPV4:
2270 gso_type = SKB_GSO_TCPV4;
2271 break;
2272 case VIRTIO_NET_HDR_GSO_TCPV6:
2273 gso_type = SKB_GSO_TCPV6;
2274 break;
2275 case VIRTIO_NET_HDR_GSO_UDP:
2276 gso_type = SKB_GSO_UDP;
2277 break;
2278 default:
2279 goto out_unlock;
2280 }
2281
2282 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2283 gso_type |= SKB_GSO_TCP_ECN;
2284
2285 if (vnet_hdr.gso_size == 0)
2286 goto out_unlock;
2287
2288 }
2289 }
2290
2291 err = -EMSGSIZE;
2292 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2293 goto out_unlock;
2294
2295 err = -ENOBUFS;
2296 hlen = LL_RESERVED_SPACE(dev);
2297 tlen = dev->needed_tailroom;
2298 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2299 msg->msg_flags & MSG_DONTWAIT, &err);
2300 if (skb == NULL)
2301 goto out_unlock;
2302
2303 skb_set_network_header(skb, reserve);
2304
2305 err = -EINVAL;
2306 if (sock->type == SOCK_DGRAM &&
2307 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2308 goto out_free;
2309
2310 /* Returns -EFAULT on error */
2311 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2312 if (err)
2313 goto out_free;
2314 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2315 if (err < 0)
2316 goto out_free;
2317
2318 if (!gso_type && (len > dev->mtu + reserve)) {
2319 /* Earlier code assumed this would be a VLAN pkt,
2320 * double-check this now that we have the actual
2321 * packet in hand.
2322 */
2323 struct ethhdr *ehdr;
2324 skb_reset_mac_header(skb);
2325 ehdr = eth_hdr(skb);
2326 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2327 err = -EMSGSIZE;
2328 goto out_free;
2329 }
2330 }
2331
2332 skb->protocol = proto;
2333 skb->dev = dev;
2334 skb->priority = sk->sk_priority;
2335 skb->mark = sk->sk_mark;
2336
2337 if (po->has_vnet_hdr) {
2338 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2339 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2340 vnet_hdr.csum_offset)) {
2341 err = -EINVAL;
2342 goto out_free;
2343 }
2344 }
2345
2346 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2347 skb_shinfo(skb)->gso_type = gso_type;
2348
2349 /* Header must be checked, and gso_segs computed. */
2350 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2351 skb_shinfo(skb)->gso_segs = 0;
2352
2353 len += vnet_hdr_len;
2354 }
2355
2356 /*
2357 * Now send it
2358 */
2359
2360 err = dev_queue_xmit(skb);
2361 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2362 goto out_unlock;
2363
2364 if (need_rls_dev)
2365 dev_put(dev);
2366
2367 return len;
2368
2369 out_free:
2370 kfree_skb(skb);
2371 out_unlock:
2372 if (dev && need_rls_dev)
2373 dev_put(dev);
2374 out:
2375 return err;
2376 }
2377
2378 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2379 struct msghdr *msg, size_t len)
2380 {
2381 struct sock *sk = sock->sk;
2382 struct packet_sock *po = pkt_sk(sk);
2383 if (po->tx_ring.pg_vec)
2384 return tpacket_snd(po, msg);
2385 else
2386 return packet_snd(sock, msg, len);
2387 }
2388
2389 /*
2390 * Close a PACKET socket. This is fairly simple. We immediately go
2391 * to 'closed' state and remove our protocol entry in the device list.
2392 */
2393
2394 static int packet_release(struct socket *sock)
2395 {
2396 struct sock *sk = sock->sk;
2397 struct packet_sock *po;
2398 struct net *net;
2399 union tpacket_req_u req_u;
2400
2401 if (!sk)
2402 return 0;
2403
2404 net = sock_net(sk);
2405 po = pkt_sk(sk);
2406
2407 spin_lock_bh(&net->packet.sklist_lock);
2408 sk_del_node_init_rcu(sk);
2409 sock_prot_inuse_add(net, sk->sk_prot, -1);
2410 spin_unlock_bh(&net->packet.sklist_lock);
2411
2412 spin_lock(&po->bind_lock);
2413 unregister_prot_hook(sk, false);
2414 if (po->prot_hook.dev) {
2415 dev_put(po->prot_hook.dev);
2416 po->prot_hook.dev = NULL;
2417 }
2418 spin_unlock(&po->bind_lock);
2419
2420 packet_flush_mclist(sk);
2421
2422 memset(&req_u, 0, sizeof(req_u));
2423
2424 if (po->rx_ring.pg_vec)
2425 packet_set_ring(sk, &req_u, 1, 0);
2426
2427 if (po->tx_ring.pg_vec)
2428 packet_set_ring(sk, &req_u, 1, 1);
2429
2430 fanout_release(sk);
2431
2432 synchronize_net();
2433 /*
2434 * Now the socket is dead. No more input will appear.
2435 */
2436 sock_orphan(sk);
2437 sock->sk = NULL;
2438
2439 /* Purge queues */
2440
2441 skb_queue_purge(&sk->sk_receive_queue);
2442 sk_refcnt_debug_release(sk);
2443
2444 sock_put(sk);
2445 return 0;
2446 }
2447
2448 /*
2449 * Attach a packet hook.
2450 */
2451
2452 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2453 {
2454 struct packet_sock *po = pkt_sk(sk);
2455
2456 if (po->fanout) {
2457 if (dev)
2458 dev_put(dev);
2459
2460 return -EINVAL;
2461 }
2462
2463 lock_sock(sk);
2464
2465 spin_lock(&po->bind_lock);
2466 unregister_prot_hook(sk, true);
2467 po->num = protocol;
2468 po->prot_hook.type = protocol;
2469 if (po->prot_hook.dev)
2470 dev_put(po->prot_hook.dev);
2471 po->prot_hook.dev = dev;
2472
2473 po->ifindex = dev ? dev->ifindex : 0;
2474
2475 if (protocol == 0)
2476 goto out_unlock;
2477
2478 if (!dev || (dev->flags & IFF_UP)) {
2479 register_prot_hook(sk);
2480 } else {
2481 sk->sk_err = ENETDOWN;
2482 if (!sock_flag(sk, SOCK_DEAD))
2483 sk->sk_error_report(sk);
2484 }
2485
2486 out_unlock:
2487 spin_unlock(&po->bind_lock);
2488 release_sock(sk);
2489 return 0;
2490 }
2491
2492 /*
2493 * Bind a packet socket to a device
2494 */
2495
2496 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2497 int addr_len)
2498 {
2499 struct sock *sk = sock->sk;
2500 char name[15];
2501 struct net_device *dev;
2502 int err = -ENODEV;
2503
2504 /*
2505 * Check legality
2506 */
2507
2508 if (addr_len != sizeof(struct sockaddr))
2509 return -EINVAL;
2510 strlcpy(name, uaddr->sa_data, sizeof(name));
2511
2512 dev = dev_get_by_name(sock_net(sk), name);
2513 if (dev)
2514 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2515 return err;
2516 }
2517
2518 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2519 {
2520 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2521 struct sock *sk = sock->sk;
2522 struct net_device *dev = NULL;
2523 int err;
2524
2525
2526 /*
2527 * Check legality
2528 */
2529
2530 if (addr_len < sizeof(struct sockaddr_ll))
2531 return -EINVAL;
2532 if (sll->sll_family != AF_PACKET)
2533 return -EINVAL;
2534
2535 if (sll->sll_ifindex) {
2536 err = -ENODEV;
2537 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2538 if (dev == NULL)
2539 goto out;
2540 }
2541 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2542
2543 out:
2544 return err;
2545 }
2546
2547 static struct proto packet_proto = {
2548 .name = "PACKET",
2549 .owner = THIS_MODULE,
2550 .obj_size = sizeof(struct packet_sock),
2551 };
2552
2553 /*
2554 * Create a packet of type SOCK_PACKET.
2555 */
2556
2557 static int packet_create(struct net *net, struct socket *sock, int protocol,
2558 int kern)
2559 {
2560 struct sock *sk;
2561 struct packet_sock *po;
2562 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2563 int err;
2564
2565 if (!capable(CAP_NET_RAW))
2566 return -EPERM;
2567 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2568 sock->type != SOCK_PACKET)
2569 return -ESOCKTNOSUPPORT;
2570
2571 sock->state = SS_UNCONNECTED;
2572
2573 err = -ENOBUFS;
2574 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2575 if (sk == NULL)
2576 goto out;
2577
2578 sock->ops = &packet_ops;
2579 if (sock->type == SOCK_PACKET)
2580 sock->ops = &packet_ops_spkt;
2581
2582 sock_init_data(sock, sk);
2583
2584 po = pkt_sk(sk);
2585 sk->sk_family = PF_PACKET;
2586 po->num = proto;
2587
2588 sk->sk_destruct = packet_sock_destruct;
2589 sk_refcnt_debug_inc(sk);
2590
2591 /*
2592 * Attach a protocol block
2593 */
2594
2595 spin_lock_init(&po->bind_lock);
2596 mutex_init(&po->pg_vec_lock);
2597 po->prot_hook.func = packet_rcv;
2598
2599 if (sock->type == SOCK_PACKET)
2600 po->prot_hook.func = packet_rcv_spkt;
2601
2602 po->prot_hook.af_packet_priv = sk;
2603
2604 if (proto) {
2605 po->prot_hook.type = proto;
2606 register_prot_hook(sk);
2607 }
2608
2609 spin_lock_bh(&net->packet.sklist_lock);
2610 sk_add_node_rcu(sk, &net->packet.sklist);
2611 sock_prot_inuse_add(net, &packet_proto, 1);
2612 spin_unlock_bh(&net->packet.sklist_lock);
2613
2614 return 0;
2615 out:
2616 return err;
2617 }
2618
2619 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2620 {
2621 struct sock_exterr_skb *serr;
2622 struct sk_buff *skb, *skb2;
2623 int copied, err;
2624
2625 err = -EAGAIN;
2626 skb = skb_dequeue(&sk->sk_error_queue);
2627 if (skb == NULL)
2628 goto out;
2629
2630 copied = skb->len;
2631 if (copied > len) {
2632 msg->msg_flags |= MSG_TRUNC;
2633 copied = len;
2634 }
2635 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2636 if (err)
2637 goto out_free_skb;
2638
2639 sock_recv_timestamp(msg, sk, skb);
2640
2641 serr = SKB_EXT_ERR(skb);
2642 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2643 sizeof(serr->ee), &serr->ee);
2644
2645 msg->msg_flags |= MSG_ERRQUEUE;
2646 err = copied;
2647
2648 /* Reset and regenerate socket error */
2649 spin_lock_bh(&sk->sk_error_queue.lock);
2650 sk->sk_err = 0;
2651 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2652 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2653 spin_unlock_bh(&sk->sk_error_queue.lock);
2654 sk->sk_error_report(sk);
2655 } else
2656 spin_unlock_bh(&sk->sk_error_queue.lock);
2657
2658 out_free_skb:
2659 kfree_skb(skb);
2660 out:
2661 return err;
2662 }
2663
2664 /*
2665 * Pull a packet from our receive queue and hand it to the user.
2666 * If necessary we block.
2667 */
2668
2669 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2670 struct msghdr *msg, size_t len, int flags)
2671 {
2672 struct sock *sk = sock->sk;
2673 struct sk_buff *skb;
2674 int copied, err;
2675 struct sockaddr_ll *sll;
2676 int vnet_hdr_len = 0;
2677
2678 err = -EINVAL;
2679 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2680 goto out;
2681
2682 #if 0
2683 /* What error should we return now? EUNATTACH? */
2684 if (pkt_sk(sk)->ifindex < 0)
2685 return -ENODEV;
2686 #endif
2687
2688 if (flags & MSG_ERRQUEUE) {
2689 err = packet_recv_error(sk, msg, len);
2690 goto out;
2691 }
2692
2693 /*
2694 * Call the generic datagram receiver. This handles all sorts
2695 * of horrible races and re-entrancy so we can forget about it
2696 * in the protocol layers.
2697 *
2698 * Now it will return ENETDOWN, if device have just gone down,
2699 * but then it will block.
2700 */
2701
2702 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2703
2704 /*
2705 * An error occurred so return it. Because skb_recv_datagram()
2706 * handles the blocking we don't see and worry about blocking
2707 * retries.
2708 */
2709
2710 if (skb == NULL)
2711 goto out;
2712
2713 if (pkt_sk(sk)->has_vnet_hdr) {
2714 struct virtio_net_hdr vnet_hdr = { 0 };
2715
2716 err = -EINVAL;
2717 vnet_hdr_len = sizeof(vnet_hdr);
2718 if (len < vnet_hdr_len)
2719 goto out_free;
2720
2721 len -= vnet_hdr_len;
2722
2723 if (skb_is_gso(skb)) {
2724 struct skb_shared_info *sinfo = skb_shinfo(skb);
2725
2726 /* This is a hint as to how much should be linear. */
2727 vnet_hdr.hdr_len = skb_headlen(skb);
2728 vnet_hdr.gso_size = sinfo->gso_size;
2729 if (sinfo->gso_type & SKB_GSO_TCPV4)
2730 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2731 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2732 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2733 else if (sinfo->gso_type & SKB_GSO_UDP)
2734 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2735 else if (sinfo->gso_type & SKB_GSO_FCOE)
2736 goto out_free;
2737 else
2738 BUG();
2739 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2740 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2741 } else
2742 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2743
2744 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2745 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2746 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2747 vnet_hdr.csum_offset = skb->csum_offset;
2748 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2749 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2750 } /* else everything is zero */
2751
2752 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2753 vnet_hdr_len);
2754 if (err < 0)
2755 goto out_free;
2756 }
2757
2758 /*
2759 * If the address length field is there to be filled in, we fill
2760 * it in now.
2761 */
2762
2763 sll = &PACKET_SKB_CB(skb)->sa.ll;
2764 if (sock->type == SOCK_PACKET)
2765 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2766 else
2767 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2768
2769 /*
2770 * You lose any data beyond the buffer you gave. If it worries a
2771 * user program they can ask the device for its MTU anyway.
2772 */
2773
2774 copied = skb->len;
2775 if (copied > len) {
2776 copied = len;
2777 msg->msg_flags |= MSG_TRUNC;
2778 }
2779
2780 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2781 if (err)
2782 goto out_free;
2783
2784 sock_recv_ts_and_drops(msg, sk, skb);
2785
2786 if (msg->msg_name)
2787 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2788 msg->msg_namelen);
2789
2790 if (pkt_sk(sk)->auxdata) {
2791 struct tpacket_auxdata aux;
2792
2793 aux.tp_status = TP_STATUS_USER;
2794 if (skb->ip_summed == CHECKSUM_PARTIAL)
2795 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2796 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2797 aux.tp_snaplen = skb->len;
2798 aux.tp_mac = 0;
2799 aux.tp_net = skb_network_offset(skb);
2800 if (vlan_tx_tag_present(skb)) {
2801 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2802 aux.tp_status |= TP_STATUS_VLAN_VALID;
2803 } else {
2804 aux.tp_vlan_tci = 0;
2805 }
2806 aux.tp_padding = 0;
2807 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2808 }
2809
2810 /*
2811 * Free or return the buffer as appropriate. Again this
2812 * hides all the races and re-entrancy issues from us.
2813 */
2814 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2815
2816 out_free:
2817 skb_free_datagram(sk, skb);
2818 out:
2819 return err;
2820 }
2821
2822 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2823 int *uaddr_len, int peer)
2824 {
2825 struct net_device *dev;
2826 struct sock *sk = sock->sk;
2827
2828 if (peer)
2829 return -EOPNOTSUPP;
2830
2831 uaddr->sa_family = AF_PACKET;
2832 rcu_read_lock();
2833 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2834 if (dev)
2835 strncpy(uaddr->sa_data, dev->name, 14);
2836 else
2837 memset(uaddr->sa_data, 0, 14);
2838 rcu_read_unlock();
2839 *uaddr_len = sizeof(*uaddr);
2840
2841 return 0;
2842 }
2843
2844 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2845 int *uaddr_len, int peer)
2846 {
2847 struct net_device *dev;
2848 struct sock *sk = sock->sk;
2849 struct packet_sock *po = pkt_sk(sk);
2850 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2851
2852 if (peer)
2853 return -EOPNOTSUPP;
2854
2855 sll->sll_family = AF_PACKET;
2856 sll->sll_ifindex = po->ifindex;
2857 sll->sll_protocol = po->num;
2858 sll->sll_pkttype = 0;
2859 rcu_read_lock();
2860 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2861 if (dev) {
2862 sll->sll_hatype = dev->type;
2863 sll->sll_halen = dev->addr_len;
2864 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2865 } else {
2866 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2867 sll->sll_halen = 0;
2868 }
2869 rcu_read_unlock();
2870 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2871
2872 return 0;
2873 }
2874
2875 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2876 int what)
2877 {
2878 switch (i->type) {
2879 case PACKET_MR_MULTICAST:
2880 if (i->alen != dev->addr_len)
2881 return -EINVAL;
2882 if (what > 0)
2883 return dev_mc_add(dev, i->addr);
2884 else
2885 return dev_mc_del(dev, i->addr);
2886 break;
2887 case PACKET_MR_PROMISC:
2888 return dev_set_promiscuity(dev, what);
2889 break;
2890 case PACKET_MR_ALLMULTI:
2891 return dev_set_allmulti(dev, what);
2892 break;
2893 case PACKET_MR_UNICAST:
2894 if (i->alen != dev->addr_len)
2895 return -EINVAL;
2896 if (what > 0)
2897 return dev_uc_add(dev, i->addr);
2898 else
2899 return dev_uc_del(dev, i->addr);
2900 break;
2901 default:
2902 break;
2903 }
2904 return 0;
2905 }
2906
2907 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2908 {
2909 for ( ; i; i = i->next) {
2910 if (i->ifindex == dev->ifindex)
2911 packet_dev_mc(dev, i, what);
2912 }
2913 }
2914
2915 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2916 {
2917 struct packet_sock *po = pkt_sk(sk);
2918 struct packet_mclist *ml, *i;
2919 struct net_device *dev;
2920 int err;
2921
2922 rtnl_lock();
2923
2924 err = -ENODEV;
2925 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2926 if (!dev)
2927 goto done;
2928
2929 err = -EINVAL;
2930 if (mreq->mr_alen > dev->addr_len)
2931 goto done;
2932
2933 err = -ENOBUFS;
2934 i = kmalloc(sizeof(*i), GFP_KERNEL);
2935 if (i == NULL)
2936 goto done;
2937
2938 err = 0;
2939 for (ml = po->mclist; ml; ml = ml->next) {
2940 if (ml->ifindex == mreq->mr_ifindex &&
2941 ml->type == mreq->mr_type &&
2942 ml->alen == mreq->mr_alen &&
2943 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2944 ml->count++;
2945 /* Free the new element ... */
2946 kfree(i);
2947 goto done;
2948 }
2949 }
2950
2951 i->type = mreq->mr_type;
2952 i->ifindex = mreq->mr_ifindex;
2953 i->alen = mreq->mr_alen;
2954 memcpy(i->addr, mreq->mr_address, i->alen);
2955 i->count = 1;
2956 i->next = po->mclist;
2957 po->mclist = i;
2958 err = packet_dev_mc(dev, i, 1);
2959 if (err) {
2960 po->mclist = i->next;
2961 kfree(i);
2962 }
2963
2964 done:
2965 rtnl_unlock();
2966 return err;
2967 }
2968
2969 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2970 {
2971 struct packet_mclist *ml, **mlp;
2972
2973 rtnl_lock();
2974
2975 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2976 if (ml->ifindex == mreq->mr_ifindex &&
2977 ml->type == mreq->mr_type &&
2978 ml->alen == mreq->mr_alen &&
2979 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2980 if (--ml->count == 0) {
2981 struct net_device *dev;
2982 *mlp = ml->next;
2983 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2984 if (dev)
2985 packet_dev_mc(dev, ml, -1);
2986 kfree(ml);
2987 }
2988 rtnl_unlock();
2989 return 0;
2990 }
2991 }
2992 rtnl_unlock();
2993 return -EADDRNOTAVAIL;
2994 }
2995
2996 static void packet_flush_mclist(struct sock *sk)
2997 {
2998 struct packet_sock *po = pkt_sk(sk);
2999 struct packet_mclist *ml;
3000
3001 if (!po->mclist)
3002 return;
3003
3004 rtnl_lock();
3005 while ((ml = po->mclist) != NULL) {
3006 struct net_device *dev;
3007
3008 po->mclist = ml->next;
3009 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3010 if (dev != NULL)
3011 packet_dev_mc(dev, ml, -1);
3012 kfree(ml);
3013 }
3014 rtnl_unlock();
3015 }
3016
3017 static int
3018 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3019 {
3020 struct sock *sk = sock->sk;
3021 struct packet_sock *po = pkt_sk(sk);
3022 int ret;
3023
3024 if (level != SOL_PACKET)
3025 return -ENOPROTOOPT;
3026
3027 switch (optname) {
3028 case PACKET_ADD_MEMBERSHIP:
3029 case PACKET_DROP_MEMBERSHIP:
3030 {
3031 struct packet_mreq_max mreq;
3032 int len = optlen;
3033 memset(&mreq, 0, sizeof(mreq));
3034 if (len < sizeof(struct packet_mreq))
3035 return -EINVAL;
3036 if (len > sizeof(mreq))
3037 len = sizeof(mreq);
3038 if (copy_from_user(&mreq, optval, len))
3039 return -EFAULT;
3040 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3041 return -EINVAL;
3042 if (optname == PACKET_ADD_MEMBERSHIP)
3043 ret = packet_mc_add(sk, &mreq);
3044 else
3045 ret = packet_mc_drop(sk, &mreq);
3046 return ret;
3047 }
3048
3049 case PACKET_RX_RING:
3050 case PACKET_TX_RING:
3051 {
3052 union tpacket_req_u req_u;
3053 int len;
3054
3055 switch (po->tp_version) {
3056 case TPACKET_V1:
3057 case TPACKET_V2:
3058 len = sizeof(req_u.req);
3059 break;
3060 case TPACKET_V3:
3061 default:
3062 len = sizeof(req_u.req3);
3063 break;
3064 }
3065 if (optlen < len)
3066 return -EINVAL;
3067 if (pkt_sk(sk)->has_vnet_hdr)
3068 return -EINVAL;
3069 if (copy_from_user(&req_u.req, optval, len))
3070 return -EFAULT;
3071 return packet_set_ring(sk, &req_u, 0,
3072 optname == PACKET_TX_RING);
3073 }
3074 case PACKET_COPY_THRESH:
3075 {
3076 int val;
3077
3078 if (optlen != sizeof(val))
3079 return -EINVAL;
3080 if (copy_from_user(&val, optval, sizeof(val)))
3081 return -EFAULT;
3082
3083 pkt_sk(sk)->copy_thresh = val;
3084 return 0;
3085 }
3086 case PACKET_VERSION:
3087 {
3088 int val;
3089
3090 if (optlen != sizeof(val))
3091 return -EINVAL;
3092 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3093 return -EBUSY;
3094 if (copy_from_user(&val, optval, sizeof(val)))
3095 return -EFAULT;
3096 switch (val) {
3097 case TPACKET_V1:
3098 case TPACKET_V2:
3099 case TPACKET_V3:
3100 po->tp_version = val;
3101 return 0;
3102 default:
3103 return -EINVAL;
3104 }
3105 }
3106 case PACKET_RESERVE:
3107 {
3108 unsigned int val;
3109
3110 if (optlen != sizeof(val))
3111 return -EINVAL;
3112 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3113 return -EBUSY;
3114 if (copy_from_user(&val, optval, sizeof(val)))
3115 return -EFAULT;
3116 po->tp_reserve = val;
3117 return 0;
3118 }
3119 case PACKET_LOSS:
3120 {
3121 unsigned int val;
3122
3123 if (optlen != sizeof(val))
3124 return -EINVAL;
3125 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3126 return -EBUSY;
3127 if (copy_from_user(&val, optval, sizeof(val)))
3128 return -EFAULT;
3129 po->tp_loss = !!val;
3130 return 0;
3131 }
3132 case PACKET_AUXDATA:
3133 {
3134 int val;
3135
3136 if (optlen < sizeof(val))
3137 return -EINVAL;
3138 if (copy_from_user(&val, optval, sizeof(val)))
3139 return -EFAULT;
3140
3141 po->auxdata = !!val;
3142 return 0;
3143 }
3144 case PACKET_ORIGDEV:
3145 {
3146 int val;
3147
3148 if (optlen < sizeof(val))
3149 return -EINVAL;
3150 if (copy_from_user(&val, optval, sizeof(val)))
3151 return -EFAULT;
3152
3153 po->origdev = !!val;
3154 return 0;
3155 }
3156 case PACKET_VNET_HDR:
3157 {
3158 int val;
3159
3160 if (sock->type != SOCK_RAW)
3161 return -EINVAL;
3162 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3163 return -EBUSY;
3164 if (optlen < sizeof(val))
3165 return -EINVAL;
3166 if (copy_from_user(&val, optval, sizeof(val)))
3167 return -EFAULT;
3168
3169 po->has_vnet_hdr = !!val;
3170 return 0;
3171 }
3172 case PACKET_TIMESTAMP:
3173 {
3174 int val;
3175
3176 if (optlen != sizeof(val))
3177 return -EINVAL;
3178 if (copy_from_user(&val, optval, sizeof(val)))
3179 return -EFAULT;
3180
3181 po->tp_tstamp = val;
3182 return 0;
3183 }
3184 case PACKET_FANOUT:
3185 {
3186 int val;
3187
3188 if (optlen != sizeof(val))
3189 return -EINVAL;
3190 if (copy_from_user(&val, optval, sizeof(val)))
3191 return -EFAULT;
3192
3193 return fanout_add(sk, val & 0xffff, val >> 16);
3194 }
3195 default:
3196 return -ENOPROTOOPT;
3197 }
3198 }
3199
3200 static int packet_getsockopt(struct socket *sock, int level, int optname,
3201 char __user *optval, int __user *optlen)
3202 {
3203 int len;
3204 int val;
3205 struct sock *sk = sock->sk;
3206 struct packet_sock *po = pkt_sk(sk);
3207 void *data;
3208 struct tpacket_stats st;
3209 union tpacket_stats_u st_u;
3210
3211 if (level != SOL_PACKET)
3212 return -ENOPROTOOPT;
3213
3214 if (get_user(len, optlen))
3215 return -EFAULT;
3216
3217 if (len < 0)
3218 return -EINVAL;
3219
3220 switch (optname) {
3221 case PACKET_STATISTICS:
3222 if (po->tp_version == TPACKET_V3) {
3223 len = sizeof(struct tpacket_stats_v3);
3224 } else {
3225 if (len > sizeof(struct tpacket_stats))
3226 len = sizeof(struct tpacket_stats);
3227 }
3228 spin_lock_bh(&sk->sk_receive_queue.lock);
3229 if (po->tp_version == TPACKET_V3) {
3230 memcpy(&st_u.stats3, &po->stats,
3231 sizeof(struct tpacket_stats));
3232 st_u.stats3.tp_freeze_q_cnt =
3233 po->stats_u.stats3.tp_freeze_q_cnt;
3234 st_u.stats3.tp_packets += po->stats.tp_drops;
3235 data = &st_u.stats3;
3236 } else {
3237 st = po->stats;
3238 st.tp_packets += st.tp_drops;
3239 data = &st;
3240 }
3241 memset(&po->stats, 0, sizeof(st));
3242 spin_unlock_bh(&sk->sk_receive_queue.lock);
3243 break;
3244 case PACKET_AUXDATA:
3245 if (len > sizeof(int))
3246 len = sizeof(int);
3247 val = po->auxdata;
3248
3249 data = &val;
3250 break;
3251 case PACKET_ORIGDEV:
3252 if (len > sizeof(int))
3253 len = sizeof(int);
3254 val = po->origdev;
3255
3256 data = &val;
3257 break;
3258 case PACKET_VNET_HDR:
3259 if (len > sizeof(int))
3260 len = sizeof(int);
3261 val = po->has_vnet_hdr;
3262
3263 data = &val;
3264 break;
3265 case PACKET_VERSION:
3266 if (len > sizeof(int))
3267 len = sizeof(int);
3268 val = po->tp_version;
3269 data = &val;
3270 break;
3271 case PACKET_HDRLEN:
3272 if (len > sizeof(int))
3273 len = sizeof(int);
3274 if (copy_from_user(&val, optval, len))
3275 return -EFAULT;
3276 switch (val) {
3277 case TPACKET_V1:
3278 val = sizeof(struct tpacket_hdr);
3279 break;
3280 case TPACKET_V2:
3281 val = sizeof(struct tpacket2_hdr);
3282 break;
3283 case TPACKET_V3:
3284 val = sizeof(struct tpacket3_hdr);
3285 break;
3286 default:
3287 return -EINVAL;
3288 }
3289 data = &val;
3290 break;
3291 case PACKET_RESERVE:
3292 if (len > sizeof(unsigned int))
3293 len = sizeof(unsigned int);
3294 val = po->tp_reserve;
3295 data = &val;
3296 break;
3297 case PACKET_LOSS:
3298 if (len > sizeof(unsigned int))
3299 len = sizeof(unsigned int);
3300 val = po->tp_loss;
3301 data = &val;
3302 break;
3303 case PACKET_TIMESTAMP:
3304 if (len > sizeof(int))
3305 len = sizeof(int);
3306 val = po->tp_tstamp;
3307 data = &val;
3308 break;
3309 case PACKET_FANOUT:
3310 if (len > sizeof(int))
3311 len = sizeof(int);
3312 val = (po->fanout ?
3313 ((u32)po->fanout->id |
3314 ((u32)po->fanout->type << 16)) :
3315 0);
3316 data = &val;
3317 break;
3318 default:
3319 return -ENOPROTOOPT;
3320 }
3321
3322 if (put_user(len, optlen))
3323 return -EFAULT;
3324 if (copy_to_user(optval, data, len))
3325 return -EFAULT;
3326 return 0;
3327 }
3328
3329
3330 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3331 {
3332 struct sock *sk;
3333 struct hlist_node *node;
3334 struct net_device *dev = data;
3335 struct net *net = dev_net(dev);
3336
3337 rcu_read_lock();
3338 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3339 struct packet_sock *po = pkt_sk(sk);
3340
3341 switch (msg) {
3342 case NETDEV_UNREGISTER:
3343 if (po->mclist)
3344 packet_dev_mclist(dev, po->mclist, -1);
3345 /* fallthrough */
3346
3347 case NETDEV_DOWN:
3348 if (dev->ifindex == po->ifindex) {
3349 spin_lock(&po->bind_lock);
3350 if (po->running) {
3351 __unregister_prot_hook(sk, false);
3352 sk->sk_err = ENETDOWN;
3353 if (!sock_flag(sk, SOCK_DEAD))
3354 sk->sk_error_report(sk);
3355 }
3356 if (msg == NETDEV_UNREGISTER) {
3357 po->ifindex = -1;
3358 if (po->prot_hook.dev)
3359 dev_put(po->prot_hook.dev);
3360 po->prot_hook.dev = NULL;
3361 }
3362 spin_unlock(&po->bind_lock);
3363 }
3364 break;
3365 case NETDEV_UP:
3366 if (dev->ifindex == po->ifindex) {
3367 spin_lock(&po->bind_lock);
3368 if (po->num)
3369 register_prot_hook(sk);
3370 spin_unlock(&po->bind_lock);
3371 }
3372 break;
3373 }
3374 }
3375 rcu_read_unlock();
3376 return NOTIFY_DONE;
3377 }
3378
3379
3380 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3381 unsigned long arg)
3382 {
3383 struct sock *sk = sock->sk;
3384
3385 switch (cmd) {
3386 case SIOCOUTQ:
3387 {
3388 int amount = sk_wmem_alloc_get(sk);
3389
3390 return put_user(amount, (int __user *)arg);
3391 }
3392 case SIOCINQ:
3393 {
3394 struct sk_buff *skb;
3395 int amount = 0;
3396
3397 spin_lock_bh(&sk->sk_receive_queue.lock);
3398 skb = skb_peek(&sk->sk_receive_queue);
3399 if (skb)
3400 amount = skb->len;
3401 spin_unlock_bh(&sk->sk_receive_queue.lock);
3402 return put_user(amount, (int __user *)arg);
3403 }
3404 case SIOCGSTAMP:
3405 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3406 case SIOCGSTAMPNS:
3407 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3408
3409 #ifdef CONFIG_INET
3410 case SIOCADDRT:
3411 case SIOCDELRT:
3412 case SIOCDARP:
3413 case SIOCGARP:
3414 case SIOCSARP:
3415 case SIOCGIFADDR:
3416 case SIOCSIFADDR:
3417 case SIOCGIFBRDADDR:
3418 case SIOCSIFBRDADDR:
3419 case SIOCGIFNETMASK:
3420 case SIOCSIFNETMASK:
3421 case SIOCGIFDSTADDR:
3422 case SIOCSIFDSTADDR:
3423 case SIOCSIFFLAGS:
3424 return inet_dgram_ops.ioctl(sock, cmd, arg);
3425 #endif
3426
3427 default:
3428 return -ENOIOCTLCMD;
3429 }
3430 return 0;
3431 }
3432
3433 static unsigned int packet_poll(struct file *file, struct socket *sock,
3434 poll_table *wait)
3435 {
3436 struct sock *sk = sock->sk;
3437 struct packet_sock *po = pkt_sk(sk);
3438 unsigned int mask = datagram_poll(file, sock, wait);
3439
3440 spin_lock_bh(&sk->sk_receive_queue.lock);
3441 if (po->rx_ring.pg_vec) {
3442 if (!packet_previous_rx_frame(po, &po->rx_ring,
3443 TP_STATUS_KERNEL))
3444 mask |= POLLIN | POLLRDNORM;
3445 }
3446 spin_unlock_bh(&sk->sk_receive_queue.lock);
3447 spin_lock_bh(&sk->sk_write_queue.lock);
3448 if (po->tx_ring.pg_vec) {
3449 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3450 mask |= POLLOUT | POLLWRNORM;
3451 }
3452 spin_unlock_bh(&sk->sk_write_queue.lock);
3453 return mask;
3454 }
3455
3456
3457 /* Dirty? Well, I still did not learn better way to account
3458 * for user mmaps.
3459 */
3460
3461 static void packet_mm_open(struct vm_area_struct *vma)
3462 {
3463 struct file *file = vma->vm_file;
3464 struct socket *sock = file->private_data;
3465 struct sock *sk = sock->sk;
3466
3467 if (sk)
3468 atomic_inc(&pkt_sk(sk)->mapped);
3469 }
3470
3471 static void packet_mm_close(struct vm_area_struct *vma)
3472 {
3473 struct file *file = vma->vm_file;
3474 struct socket *sock = file->private_data;
3475 struct sock *sk = sock->sk;
3476
3477 if (sk)
3478 atomic_dec(&pkt_sk(sk)->mapped);
3479 }
3480
3481 static const struct vm_operations_struct packet_mmap_ops = {
3482 .open = packet_mm_open,
3483 .close = packet_mm_close,
3484 };
3485
3486 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3487 unsigned int len)
3488 {
3489 int i;
3490
3491 for (i = 0; i < len; i++) {
3492 if (likely(pg_vec[i].buffer)) {
3493 if (is_vmalloc_addr(pg_vec[i].buffer))
3494 vfree(pg_vec[i].buffer);
3495 else
3496 free_pages((unsigned long)pg_vec[i].buffer,
3497 order);
3498 pg_vec[i].buffer = NULL;
3499 }
3500 }
3501 kfree(pg_vec);
3502 }
3503
3504 static char *alloc_one_pg_vec_page(unsigned long order)
3505 {
3506 char *buffer = NULL;
3507 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3508 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3509
3510 buffer = (char *) __get_free_pages(gfp_flags, order);
3511
3512 if (buffer)
3513 return buffer;
3514
3515 /*
3516 * __get_free_pages failed, fall back to vmalloc
3517 */
3518 buffer = vzalloc((1 << order) * PAGE_SIZE);
3519
3520 if (buffer)
3521 return buffer;
3522
3523 /*
3524 * vmalloc failed, lets dig into swap here
3525 */
3526 gfp_flags &= ~__GFP_NORETRY;
3527 buffer = (char *)__get_free_pages(gfp_flags, order);
3528 if (buffer)
3529 return buffer;
3530
3531 /*
3532 * complete and utter failure
3533 */
3534 return NULL;
3535 }
3536
3537 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3538 {
3539 unsigned int block_nr = req->tp_block_nr;
3540 struct pgv *pg_vec;
3541 int i;
3542
3543 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3544 if (unlikely(!pg_vec))
3545 goto out;
3546
3547 for (i = 0; i < block_nr; i++) {
3548 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3549 if (unlikely(!pg_vec[i].buffer))
3550 goto out_free_pgvec;
3551 }
3552
3553 out:
3554 return pg_vec;
3555
3556 out_free_pgvec:
3557 free_pg_vec(pg_vec, order, block_nr);
3558 pg_vec = NULL;
3559 goto out;
3560 }
3561
3562 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3563 int closing, int tx_ring)
3564 {
3565 struct pgv *pg_vec = NULL;
3566 struct packet_sock *po = pkt_sk(sk);
3567 int was_running, order = 0;
3568 struct packet_ring_buffer *rb;
3569 struct sk_buff_head *rb_queue;
3570 __be16 num;
3571 int err = -EINVAL;
3572 /* Added to avoid minimal code churn */
3573 struct tpacket_req *req = &req_u->req;
3574
3575 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3576 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3577 WARN(1, "Tx-ring is not supported.\n");
3578 goto out;
3579 }
3580
3581 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3582 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3583
3584 err = -EBUSY;
3585 if (!closing) {
3586 if (atomic_read(&po->mapped))
3587 goto out;
3588 if (atomic_read(&rb->pending))
3589 goto out;
3590 }
3591
3592 if (req->tp_block_nr) {
3593 /* Sanity tests and some calculations */
3594 err = -EBUSY;
3595 if (unlikely(rb->pg_vec))
3596 goto out;
3597
3598 switch (po->tp_version) {
3599 case TPACKET_V1:
3600 po->tp_hdrlen = TPACKET_HDRLEN;
3601 break;
3602 case TPACKET_V2:
3603 po->tp_hdrlen = TPACKET2_HDRLEN;
3604 break;
3605 case TPACKET_V3:
3606 po->tp_hdrlen = TPACKET3_HDRLEN;
3607 break;
3608 }
3609
3610 err = -EINVAL;
3611 if (unlikely((int)req->tp_block_size <= 0))
3612 goto out;
3613 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3614 goto out;
3615 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3616 po->tp_reserve))
3617 goto out;
3618 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3619 goto out;
3620
3621 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3622 if (unlikely(rb->frames_per_block <= 0))
3623 goto out;
3624 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3625 req->tp_frame_nr))
3626 goto out;
3627
3628 err = -ENOMEM;
3629 order = get_order(req->tp_block_size);
3630 pg_vec = alloc_pg_vec(req, order);
3631 if (unlikely(!pg_vec))
3632 goto out;
3633 switch (po->tp_version) {
3634 case TPACKET_V3:
3635 /* Transmit path is not supported. We checked
3636 * it above but just being paranoid
3637 */
3638 if (!tx_ring)
3639 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3640 break;
3641 default:
3642 break;
3643 }
3644 }
3645 /* Done */
3646 else {
3647 err = -EINVAL;
3648 if (unlikely(req->tp_frame_nr))
3649 goto out;
3650 }
3651
3652 lock_sock(sk);
3653
3654 /* Detach socket from network */
3655 spin_lock(&po->bind_lock);
3656 was_running = po->running;
3657 num = po->num;
3658 if (was_running) {
3659 po->num = 0;
3660 __unregister_prot_hook(sk, false);
3661 }
3662 spin_unlock(&po->bind_lock);
3663
3664 synchronize_net();
3665
3666 err = -EBUSY;
3667 mutex_lock(&po->pg_vec_lock);
3668 if (closing || atomic_read(&po->mapped) == 0) {
3669 err = 0;
3670 spin_lock_bh(&rb_queue->lock);
3671 swap(rb->pg_vec, pg_vec);
3672 rb->frame_max = (req->tp_frame_nr - 1);
3673 rb->head = 0;
3674 rb->frame_size = req->tp_frame_size;
3675 spin_unlock_bh(&rb_queue->lock);
3676
3677 swap(rb->pg_vec_order, order);
3678 swap(rb->pg_vec_len, req->tp_block_nr);
3679
3680 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3681 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3682 tpacket_rcv : packet_rcv;
3683 skb_queue_purge(rb_queue);
3684 if (atomic_read(&po->mapped))
3685 pr_err("packet_mmap: vma is busy: %d\n",
3686 atomic_read(&po->mapped));
3687 }
3688 mutex_unlock(&po->pg_vec_lock);
3689
3690 spin_lock(&po->bind_lock);
3691 if (was_running) {
3692 po->num = num;
3693 register_prot_hook(sk);
3694 }
3695 spin_unlock(&po->bind_lock);
3696 if (closing && (po->tp_version > TPACKET_V2)) {
3697 /* Because we don't support block-based V3 on tx-ring */
3698 if (!tx_ring)
3699 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3700 }
3701 release_sock(sk);
3702
3703 if (pg_vec)
3704 free_pg_vec(pg_vec, order, req->tp_block_nr);
3705 out:
3706 return err;
3707 }
3708
3709 static int packet_mmap(struct file *file, struct socket *sock,
3710 struct vm_area_struct *vma)
3711 {
3712 struct sock *sk = sock->sk;
3713 struct packet_sock *po = pkt_sk(sk);
3714 unsigned long size, expected_size;
3715 struct packet_ring_buffer *rb;
3716 unsigned long start;
3717 int err = -EINVAL;
3718 int i;
3719
3720 if (vma->vm_pgoff)
3721 return -EINVAL;
3722
3723 mutex_lock(&po->pg_vec_lock);
3724
3725 expected_size = 0;
3726 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3727 if (rb->pg_vec) {
3728 expected_size += rb->pg_vec_len
3729 * rb->pg_vec_pages
3730 * PAGE_SIZE;
3731 }
3732 }
3733
3734 if (expected_size == 0)
3735 goto out;
3736
3737 size = vma->vm_end - vma->vm_start;
3738 if (size != expected_size)
3739 goto out;
3740
3741 start = vma->vm_start;
3742 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3743 if (rb->pg_vec == NULL)
3744 continue;
3745
3746 for (i = 0; i < rb->pg_vec_len; i++) {
3747 struct page *page;
3748 void *kaddr = rb->pg_vec[i].buffer;
3749 int pg_num;
3750
3751 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3752 page = pgv_to_page(kaddr);
3753 err = vm_insert_page(vma, start, page);
3754 if (unlikely(err))
3755 goto out;
3756 start += PAGE_SIZE;
3757 kaddr += PAGE_SIZE;
3758 }
3759 }
3760 }
3761
3762 atomic_inc(&po->mapped);
3763 vma->vm_ops = &packet_mmap_ops;
3764 err = 0;
3765
3766 out:
3767 mutex_unlock(&po->pg_vec_lock);
3768 return err;
3769 }
3770
3771 static const struct proto_ops packet_ops_spkt = {
3772 .family = PF_PACKET,
3773 .owner = THIS_MODULE,
3774 .release = packet_release,
3775 .bind = packet_bind_spkt,
3776 .connect = sock_no_connect,
3777 .socketpair = sock_no_socketpair,
3778 .accept = sock_no_accept,
3779 .getname = packet_getname_spkt,
3780 .poll = datagram_poll,
3781 .ioctl = packet_ioctl,
3782 .listen = sock_no_listen,
3783 .shutdown = sock_no_shutdown,
3784 .setsockopt = sock_no_setsockopt,
3785 .getsockopt = sock_no_getsockopt,
3786 .sendmsg = packet_sendmsg_spkt,
3787 .recvmsg = packet_recvmsg,
3788 .mmap = sock_no_mmap,
3789 .sendpage = sock_no_sendpage,
3790 };
3791
3792 static const struct proto_ops packet_ops = {
3793 .family = PF_PACKET,
3794 .owner = THIS_MODULE,
3795 .release = packet_release,
3796 .bind = packet_bind,
3797 .connect = sock_no_connect,
3798 .socketpair = sock_no_socketpair,
3799 .accept = sock_no_accept,
3800 .getname = packet_getname,
3801 .poll = packet_poll,
3802 .ioctl = packet_ioctl,
3803 .listen = sock_no_listen,
3804 .shutdown = sock_no_shutdown,
3805 .setsockopt = packet_setsockopt,
3806 .getsockopt = packet_getsockopt,
3807 .sendmsg = packet_sendmsg,
3808 .recvmsg = packet_recvmsg,
3809 .mmap = packet_mmap,
3810 .sendpage = sock_no_sendpage,
3811 };
3812
3813 static const struct net_proto_family packet_family_ops = {
3814 .family = PF_PACKET,
3815 .create = packet_create,
3816 .owner = THIS_MODULE,
3817 };
3818
3819 static struct notifier_block packet_netdev_notifier = {
3820 .notifier_call = packet_notifier,
3821 };
3822
3823 #ifdef CONFIG_PROC_FS
3824
3825 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3826 __acquires(RCU)
3827 {
3828 struct net *net = seq_file_net(seq);
3829
3830 rcu_read_lock();
3831 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3832 }
3833
3834 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3835 {
3836 struct net *net = seq_file_net(seq);
3837 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3838 }
3839
3840 static void packet_seq_stop(struct seq_file *seq, void *v)
3841 __releases(RCU)
3842 {
3843 rcu_read_unlock();
3844 }
3845
3846 static int packet_seq_show(struct seq_file *seq, void *v)
3847 {
3848 if (v == SEQ_START_TOKEN)
3849 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3850 else {
3851 struct sock *s = sk_entry(v);
3852 const struct packet_sock *po = pkt_sk(s);
3853
3854 seq_printf(seq,
3855 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3856 s,
3857 atomic_read(&s->sk_refcnt),
3858 s->sk_type,
3859 ntohs(po->num),
3860 po->ifindex,
3861 po->running,
3862 atomic_read(&s->sk_rmem_alloc),
3863 sock_i_uid(s),
3864 sock_i_ino(s));
3865 }
3866
3867 return 0;
3868 }
3869
3870 static const struct seq_operations packet_seq_ops = {
3871 .start = packet_seq_start,
3872 .next = packet_seq_next,
3873 .stop = packet_seq_stop,
3874 .show = packet_seq_show,
3875 };
3876
3877 static int packet_seq_open(struct inode *inode, struct file *file)
3878 {
3879 return seq_open_net(inode, file, &packet_seq_ops,
3880 sizeof(struct seq_net_private));
3881 }
3882
3883 static const struct file_operations packet_seq_fops = {
3884 .owner = THIS_MODULE,
3885 .open = packet_seq_open,
3886 .read = seq_read,
3887 .llseek = seq_lseek,
3888 .release = seq_release_net,
3889 };
3890
3891 #endif
3892
3893 static int __net_init packet_net_init(struct net *net)
3894 {
3895 spin_lock_init(&net->packet.sklist_lock);
3896 INIT_HLIST_HEAD(&net->packet.sklist);
3897
3898 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3899 return -ENOMEM;
3900
3901 return 0;
3902 }
3903
3904 static void __net_exit packet_net_exit(struct net *net)
3905 {
3906 proc_net_remove(net, "packet");
3907 }
3908
3909 static struct pernet_operations packet_net_ops = {
3910 .init = packet_net_init,
3911 .exit = packet_net_exit,
3912 };
3913
3914
3915 static void __exit packet_exit(void)
3916 {
3917 unregister_netdevice_notifier(&packet_netdev_notifier);
3918 unregister_pernet_subsys(&packet_net_ops);
3919 sock_unregister(PF_PACKET);
3920 proto_unregister(&packet_proto);
3921 }
3922
3923 static int __init packet_init(void)
3924 {
3925 int rc = proto_register(&packet_proto, 0);
3926
3927 if (rc != 0)
3928 goto out;
3929
3930 sock_register(&packet_family_ops);
3931 register_pernet_subsys(&packet_net_ops);
3932 register_netdevice_notifier(&packet_netdev_notifier);
3933 out:
3934 return rc;
3935 }
3936
3937 module_init(packet_init);
3938 module_exit(packet_exit);
3939 MODULE_LICENSE("GPL");
3940 MODULE_ALIAS_NETPROTO(PF_PACKET);
Linus' kernel tree