projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'v3.10.84' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / packet / af_packet.c
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/uaccess.h>
77 #include <asm/ioctls.h>
78 #include <asm/page.h>
79 #include <asm/cacheflush.h>
80 #include <asm/io.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91
92 #ifdef CONFIG_INET
93 #include <net/inet_common.h>
94 #endif
95
96 #include "internal.h"
97
98 /*
99 Assumptions:
100 - if device has no dev->hard_header routine, it adds and removes ll header
101 inside itself. In this case ll header is invisible outside of device,
102 but higher levels still should reserve dev->hard_header_len.
103 Some devices are enough clever to reallocate skb, when header
104 will not fit to reserved space (tunnel), another ones are silly
105 (PPP).
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
108
109 On receive:
110 -----------
111
112 Incoming, dev->hard_header!=NULL
113 mac_header -> ll header
114 data -> data
115
116 Outgoing, dev->hard_header!=NULL
117 mac_header -> ll header
118 data -> ll header
119
120 Incoming, dev->hard_header==NULL
121 mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 header. PPP makes it, that is wrong, because introduce
123 assymetry between rx and tx paths.
124 data -> data
125
126 Outgoing, dev->hard_header==NULL
127 mac_header -> data. ll header is still not built!
128 data -> data
129
130 Resume
131 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
132
133
134 On transmit:
135 ------------
136
137 dev->hard_header != NULL
138 mac_header -> ll header
139 data -> ll header
140
141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
142 mac_header -> data
143 data -> data
144
145 We should set nh.raw on output to correct posistion,
146 packet classifier depends on it.
147 */
148
149 /* Private packet socket structures. */
150
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
153 */
154 struct packet_mreq_max {
155 int mr_ifindex;
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
159 };
160
161 union tpacket_uhdr {
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
165 void *raw;
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 #define V3_ALIGNMENT (8)
172
173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
174
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
177
178 #define PGV_FROM_VMALLOC 1
179
180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
187
188 struct packet_sock;
189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 struct packet_type *pt, struct net_device *orig_dev);
192
193 static void *packet_previous_frame(struct packet_sock *po,
194 struct packet_ring_buffer *rb,
195 int status);
196 static void packet_increment_head(struct packet_ring_buffer *buff);
197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
198 struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(unsigned long);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_init_blk_timer(struct packet_sock *,
209 struct tpacket_kbdq_core *,
210 void (*func) (unsigned long));
211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void packet_flush_mclist(struct sock *sk);
217
218 struct packet_skb_cb {
219 unsigned int origlen;
220 union {
221 struct sockaddr_pkt pkt;
222 struct sockaddr_ll ll;
223 } sa;
224 };
225
226 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
227
228 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
229 #define GET_PBLOCK_DESC(x, bid) \
230 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
232 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
233 #define GET_NEXT_PRB_BLK_NUM(x) \
234 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
235 ((x)->kactive_blk_num+1) : 0)
236
237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
239
240 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
241 {
242 struct net_device *dev;
243
244 rcu_read_lock();
245 dev = rcu_dereference(po->cached_dev);
246 if (likely(dev))
247 dev_hold(dev);
248 rcu_read_unlock();
249
250 return dev;
251 }
252
253 static void packet_cached_dev_assign(struct packet_sock *po,
254 struct net_device *dev)
255 {
256 rcu_assign_pointer(po->cached_dev, dev);
257 }
258
259 static void packet_cached_dev_reset(struct packet_sock *po)
260 {
261 RCU_INIT_POINTER(po->cached_dev, NULL);
262 }
263
264 /* register_prot_hook must be invoked with the po->bind_lock held,
265 * or from a context in which asynchronous accesses to the packet
266 * socket is not possible (packet_create()).
267 */
268 static void register_prot_hook(struct sock *sk)
269 {
270 struct packet_sock *po = pkt_sk(sk);
271
272 if (!po->running) {
273 if (po->fanout)
274 __fanout_link(sk, po);
275 else
276 dev_add_pack(&po->prot_hook);
277
278 sock_hold(sk);
279 po->running = 1;
280 }
281 }
282
283 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
284 * held. If the sync parameter is true, we will temporarily drop
285 * the po->bind_lock and do a synchronize_net to make sure no
286 * asynchronous packet processing paths still refer to the elements
287 * of po->prot_hook. If the sync parameter is false, it is the
288 * callers responsibility to take care of this.
289 */
290 static void __unregister_prot_hook(struct sock *sk, bool sync)
291 {
292 struct packet_sock *po = pkt_sk(sk);
293
294 po->running = 0;
295
296 if (po->fanout)
297 __fanout_unlink(sk, po);
298 else
299 __dev_remove_pack(&po->prot_hook);
300
301 __sock_put(sk);
302
303 if (sync) {
304 spin_unlock(&po->bind_lock);
305 synchronize_net();
306 spin_lock(&po->bind_lock);
307 }
308 }
309
310 static void unregister_prot_hook(struct sock *sk, bool sync)
311 {
312 struct packet_sock *po = pkt_sk(sk);
313
314 if (po->running)
315 __unregister_prot_hook(sk, sync);
316 }
317
318 static inline __pure struct page *pgv_to_page(void *addr)
319 {
320 if (is_vmalloc_addr(addr))
321 return vmalloc_to_page(addr);
322 return virt_to_page(addr);
323 }
324
325 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
326 {
327 union tpacket_uhdr h;
328
329 h.raw = frame;
330 switch (po->tp_version) {
331 case TPACKET_V1:
332 h.h1->tp_status = status;
333 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
334 break;
335 case TPACKET_V2:
336 h.h2->tp_status = status;
337 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
338 break;
339 case TPACKET_V3:
340 default:
341 WARN(1, "TPACKET version not supported.\n");
342 BUG();
343 }
344
345 smp_wmb();
346 }
347
348 static int __packet_get_status(struct packet_sock *po, void *frame)
349 {
350 union tpacket_uhdr h;
351
352 smp_rmb();
353
354 h.raw = frame;
355 switch (po->tp_version) {
356 case TPACKET_V1:
357 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
358 return h.h1->tp_status;
359 case TPACKET_V2:
360 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
361 return h.h2->tp_status;
362 case TPACKET_V3:
363 default:
364 WARN(1, "TPACKET version not supported.\n");
365 BUG();
366 return 0;
367 }
368 }
369
370 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
371 unsigned int flags)
372 {
373 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
374
375 if (shhwtstamps) {
376 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
377 ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
378 return TP_STATUS_TS_SYS_HARDWARE;
379 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
380 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
381 return TP_STATUS_TS_RAW_HARDWARE;
382 }
383
384 if (ktime_to_timespec_cond(skb->tstamp, ts))
385 return TP_STATUS_TS_SOFTWARE;
386
387 return 0;
388 }
389
390 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
391 struct sk_buff *skb)
392 {
393 union tpacket_uhdr h;
394 struct timespec ts;
395 __u32 ts_status;
396
397 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
398 return 0;
399
400 h.raw = frame;
401 switch (po->tp_version) {
402 case TPACKET_V1:
403 h.h1->tp_sec = ts.tv_sec;
404 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
405 break;
406 case TPACKET_V2:
407 h.h2->tp_sec = ts.tv_sec;
408 h.h2->tp_nsec = ts.tv_nsec;
409 break;
410 case TPACKET_V3:
411 default:
412 WARN(1, "TPACKET version not supported.\n");
413 BUG();
414 }
415
416 /* one flush is safe, as both fields always lie on the same cacheline */
417 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
418 smp_wmb();
419
420 return ts_status;
421 }
422
423 static void *packet_lookup_frame(struct packet_sock *po,
424 struct packet_ring_buffer *rb,
425 unsigned int position,
426 int status)
427 {
428 unsigned int pg_vec_pos, frame_offset;
429 union tpacket_uhdr h;
430
431 pg_vec_pos = position / rb->frames_per_block;
432 frame_offset = position % rb->frames_per_block;
433
434 h.raw = rb->pg_vec[pg_vec_pos].buffer +
435 (frame_offset * rb->frame_size);
436
437 if (status != __packet_get_status(po, h.raw))
438 return NULL;
439
440 return h.raw;
441 }
442
443 static void *packet_current_frame(struct packet_sock *po,
444 struct packet_ring_buffer *rb,
445 int status)
446 {
447 return packet_lookup_frame(po, rb, rb->head, status);
448 }
449
450 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
451 {
452 del_timer_sync(&pkc->retire_blk_timer);
453 }
454
455 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
456 int tx_ring,
457 struct sk_buff_head *rb_queue)
458 {
459 struct tpacket_kbdq_core *pkc;
460
461 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
462
463 spin_lock_bh(&rb_queue->lock);
464 pkc->delete_blk_timer = 1;
465 spin_unlock_bh(&rb_queue->lock);
466
467 prb_del_retire_blk_timer(pkc);
468 }
469
470 static void prb_init_blk_timer(struct packet_sock *po,
471 struct tpacket_kbdq_core *pkc,
472 void (*func) (unsigned long))
473 {
474 init_timer(&pkc->retire_blk_timer);
475 pkc->retire_blk_timer.data = (long)po;
476 pkc->retire_blk_timer.function = func;
477 pkc->retire_blk_timer.expires = jiffies;
478 }
479
480 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
481 {
482 struct tpacket_kbdq_core *pkc;
483
484 if (tx_ring)
485 BUG();
486
487 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
488 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
489 }
490
491 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
492 int blk_size_in_bytes)
493 {
494 struct net_device *dev;
495 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
496 struct ethtool_cmd ecmd;
497 int err;
498 u32 speed;
499
500 rtnl_lock();
501 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
502 if (unlikely(!dev)) {
503 rtnl_unlock();
504 return DEFAULT_PRB_RETIRE_TOV;
505 }
506 err = __ethtool_get_settings(dev, &ecmd);
507 speed = ethtool_cmd_speed(&ecmd);
508 rtnl_unlock();
509 if (!err) {
510 /*
511 * If the link speed is so slow you don't really
512 * need to worry about perf anyways
513 */
514 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
515 return DEFAULT_PRB_RETIRE_TOV;
516 } else {
517 msec = 1;
518 div = speed / 1000;
519 }
520 }
521
522 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
523
524 if (div)
525 mbits /= div;
526
527 tmo = mbits * msec;
528
529 if (div)
530 return tmo+1;
531 return tmo;
532 }
533
534 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
535 union tpacket_req_u *req_u)
536 {
537 p1->feature_req_word = req_u->req3.tp_feature_req_word;
538 }
539
540 static void init_prb_bdqc(struct packet_sock *po,
541 struct packet_ring_buffer *rb,
542 struct pgv *pg_vec,
543 union tpacket_req_u *req_u, int tx_ring)
544 {
545 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
546 struct tpacket_block_desc *pbd;
547
548 memset(p1, 0x0, sizeof(*p1));
549
550 p1->knxt_seq_num = 1;
551 p1->pkbdq = pg_vec;
552 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
553 p1->pkblk_start = pg_vec[0].buffer;
554 p1->kblk_size = req_u->req3.tp_block_size;
555 p1->knum_blocks = req_u->req3.tp_block_nr;
556 p1->hdrlen = po->tp_hdrlen;
557 p1->version = po->tp_version;
558 p1->last_kactive_blk_num = 0;
559 po->stats.stats3.tp_freeze_q_cnt = 0;
560 if (req_u->req3.tp_retire_blk_tov)
561 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
562 else
563 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
564 req_u->req3.tp_block_size);
565 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
566 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
567
568 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
569 prb_init_ft_ops(p1, req_u);
570 prb_setup_retire_blk_timer(po, tx_ring);
571 prb_open_block(p1, pbd);
572 }
573
574 /* Do NOT update the last_blk_num first.
575 * Assumes sk_buff_head lock is held.
576 */
577 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
578 {
579 mod_timer(&pkc->retire_blk_timer,
580 jiffies + pkc->tov_in_jiffies);
581 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
582 }
583
584 /*
585 * Timer logic:
586 * 1) We refresh the timer only when we open a block.
587 * By doing this we don't waste cycles refreshing the timer
588 * on packet-by-packet basis.
589 *
590 * With a 1MB block-size, on a 1Gbps line, it will take
591 * i) ~8 ms to fill a block + ii) memcpy etc.
592 * In this cut we are not accounting for the memcpy time.
593 *
594 * So, if the user sets the 'tmo' to 10ms then the timer
595 * will never fire while the block is still getting filled
596 * (which is what we want). However, the user could choose
597 * to close a block early and that's fine.
598 *
599 * But when the timer does fire, we check whether or not to refresh it.
600 * Since the tmo granularity is in msecs, it is not too expensive
601 * to refresh the timer, lets say every '8' msecs.
602 * Either the user can set the 'tmo' or we can derive it based on
603 * a) line-speed and b) block-size.
604 * prb_calc_retire_blk_tmo() calculates the tmo.
605 *
606 */
607 static void prb_retire_rx_blk_timer_expired(unsigned long data)
608 {
609 struct packet_sock *po = (struct packet_sock *)data;
610 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
611 unsigned int frozen;
612 struct tpacket_block_desc *pbd;
613
614 spin_lock(&po->sk.sk_receive_queue.lock);
615
616 frozen = prb_queue_frozen(pkc);
617 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
618
619 if (unlikely(pkc->delete_blk_timer))
620 goto out;
621
622 /* We only need to plug the race when the block is partially filled.
623 * tpacket_rcv:
624 * lock(); increment BLOCK_NUM_PKTS; unlock()
625 * copy_bits() is in progress ...
626 * timer fires on other cpu:
627 * we can't retire the current block because copy_bits
628 * is in progress.
629 *
630 */
631 if (BLOCK_NUM_PKTS(pbd)) {
632 while (atomic_read(&pkc->blk_fill_in_prog)) {
633 /* Waiting for skb_copy_bits to finish... */
634 cpu_relax();
635 }
636 }
637
638 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
639 if (!frozen) {
640 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
641 if (!prb_dispatch_next_block(pkc, po))
642 goto refresh_timer;
643 else
644 goto out;
645 } else {
646 /* Case 1. Queue was frozen because user-space was
647 * lagging behind.
648 */
649 if (prb_curr_blk_in_use(pkc, pbd)) {
650 /*
651 * Ok, user-space is still behind.
652 * So just refresh the timer.
653 */
654 goto refresh_timer;
655 } else {
656 /* Case 2. queue was frozen,user-space caught up,
657 * now the link went idle && the timer fired.
658 * We don't have a block to close.So we open this
659 * block and restart the timer.
660 * opening a block thaws the queue,restarts timer
661 * Thawing/timer-refresh is a side effect.
662 */
663 prb_open_block(pkc, pbd);
664 goto out;
665 }
666 }
667 }
668
669 refresh_timer:
670 _prb_refresh_rx_retire_blk_timer(pkc);
671
672 out:
673 spin_unlock(&po->sk.sk_receive_queue.lock);
674 }
675
676 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
677 struct tpacket_block_desc *pbd1, __u32 status)
678 {
679 /* Flush everything minus the block header */
680
681 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
682 u8 *start, *end;
683
684 start = (u8 *)pbd1;
685
686 /* Skip the block header(we know header WILL fit in 4K) */
687 start += PAGE_SIZE;
688
689 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
690 for (; start < end; start += PAGE_SIZE)
691 flush_dcache_page(pgv_to_page(start));
692
693 smp_wmb();
694 #endif
695
696 /* Now update the block status. */
697
698 BLOCK_STATUS(pbd1) = status;
699
700 /* Flush the block header */
701
702 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
703 start = (u8 *)pbd1;
704 flush_dcache_page(pgv_to_page(start));
705
706 smp_wmb();
707 #endif
708 }
709
710 /*
711 * Side effect:
712 *
713 * 1) flush the block
714 * 2) Increment active_blk_num
715 *
716 * Note:We DONT refresh the timer on purpose.
717 * Because almost always the next block will be opened.
718 */
719 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
720 struct tpacket_block_desc *pbd1,
721 struct packet_sock *po, unsigned int stat)
722 {
723 __u32 status = TP_STATUS_USER | stat;
724
725 struct tpacket3_hdr *last_pkt;
726 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
727
728 if (po->stats.stats3.tp_drops)
729 status |= TP_STATUS_LOSING;
730
731 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
732 last_pkt->tp_next_offset = 0;
733
734 /* Get the ts of the last pkt */
735 if (BLOCK_NUM_PKTS(pbd1)) {
736 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
737 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
738 } else {
739 /* Ok, we tmo'd - so get the current time */
740 struct timespec ts;
741 getnstimeofday(&ts);
742 h1->ts_last_pkt.ts_sec = ts.tv_sec;
743 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
744 }
745
746 smp_wmb();
747
748 /* Flush the block */
749 prb_flush_block(pkc1, pbd1, status);
750
751 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
752 }
753
754 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
755 {
756 pkc->reset_pending_on_curr_blk = 0;
757 }
758
759 /*
760 * Side effect of opening a block:
761 *
762 * 1) prb_queue is thawed.
763 * 2) retire_blk_timer is refreshed.
764 *
765 */
766 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
767 struct tpacket_block_desc *pbd1)
768 {
769 struct timespec ts;
770 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
771
772 smp_rmb();
773
774 /* We could have just memset this but we will lose the
775 * flexibility of making the priv area sticky
776 */
777
778 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
779 BLOCK_NUM_PKTS(pbd1) = 0;
780 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
781
782 getnstimeofday(&ts);
783
784 h1->ts_first_pkt.ts_sec = ts.tv_sec;
785 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
786
787 pkc1->pkblk_start = (char *)pbd1;
788 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
789
790 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
791 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
792
793 pbd1->version = pkc1->version;
794 pkc1->prev = pkc1->nxt_offset;
795 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
796
797 prb_thaw_queue(pkc1);
798 _prb_refresh_rx_retire_blk_timer(pkc1);
799
800 smp_wmb();
801 }
802
803 /*
804 * Queue freeze logic:
805 * 1) Assume tp_block_nr = 8 blocks.
806 * 2) At time 't0', user opens Rx ring.
807 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
808 * 4) user-space is either sleeping or processing block '0'.
809 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
810 * it will close block-7,loop around and try to fill block '0'.
811 * call-flow:
812 * __packet_lookup_frame_in_block
813 * prb_retire_current_block()
814 * prb_dispatch_next_block()
815 * |->(BLOCK_STATUS == USER) evaluates to true
816 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
817 * 6) Now there are two cases:
818 * 6.1) Link goes idle right after the queue is frozen.
819 * But remember, the last open_block() refreshed the timer.
820 * When this timer expires,it will refresh itself so that we can
821 * re-open block-0 in near future.
822 * 6.2) Link is busy and keeps on receiving packets. This is a simple
823 * case and __packet_lookup_frame_in_block will check if block-0
824 * is free and can now be re-used.
825 */
826 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
827 struct packet_sock *po)
828 {
829 pkc->reset_pending_on_curr_blk = 1;
830 po->stats.stats3.tp_freeze_q_cnt++;
831 }
832
833 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
834
835 /*
836 * If the next block is free then we will dispatch it
837 * and return a good offset.
838 * Else, we will freeze the queue.
839 * So, caller must check the return value.
840 */
841 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
842 struct packet_sock *po)
843 {
844 struct tpacket_block_desc *pbd;
845
846 smp_rmb();
847
848 /* 1. Get current block num */
849 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
850
851 /* 2. If this block is currently in_use then freeze the queue */
852 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
853 prb_freeze_queue(pkc, po);
854 return NULL;
855 }
856
857 /*
858 * 3.
859 * open this block and return the offset where the first packet
860 * needs to get stored.
861 */
862 prb_open_block(pkc, pbd);
863 return (void *)pkc->nxt_offset;
864 }
865
866 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
867 struct packet_sock *po, unsigned int status)
868 {
869 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
870
871 /* retire/close the current block */
872 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
873 /*
874 * Plug the case where copy_bits() is in progress on
875 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
876 * have space to copy the pkt in the current block and
877 * called prb_retire_current_block()
878 *
879 * We don't need to worry about the TMO case because
880 * the timer-handler already handled this case.
881 */
882 if (!(status & TP_STATUS_BLK_TMO)) {
883 while (atomic_read(&pkc->blk_fill_in_prog)) {
884 /* Waiting for skb_copy_bits to finish... */
885 cpu_relax();
886 }
887 }
888 prb_close_block(pkc, pbd, po, status);
889 return;
890 }
891 }
892
893 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
894 struct tpacket_block_desc *pbd)
895 {
896 return TP_STATUS_USER & BLOCK_STATUS(pbd);
897 }
898
899 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
900 {
901 return pkc->reset_pending_on_curr_blk;
902 }
903
904 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
905 {
906 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
907 atomic_dec(&pkc->blk_fill_in_prog);
908 }
909
910 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
911 struct tpacket3_hdr *ppd)
912 {
913 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
914 }
915
916 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
917 struct tpacket3_hdr *ppd)
918 {
919 ppd->hv1.tp_rxhash = 0;
920 }
921
922 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
923 struct tpacket3_hdr *ppd)
924 {
925 if (vlan_tx_tag_present(pkc->skb)) {
926 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
927 ppd->tp_status = TP_STATUS_VLAN_VALID;
928 } else {
929 ppd->hv1.tp_vlan_tci = 0;
930 ppd->tp_status = TP_STATUS_AVAILABLE;
931 }
932 }
933
934 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
935 struct tpacket3_hdr *ppd)
936 {
937 prb_fill_vlan_info(pkc, ppd);
938
939 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
940 prb_fill_rxhash(pkc, ppd);
941 else
942 prb_clear_rxhash(pkc, ppd);
943 }
944
945 static void prb_fill_curr_block(char *curr,
946 struct tpacket_kbdq_core *pkc,
947 struct tpacket_block_desc *pbd,
948 unsigned int len)
949 {
950 struct tpacket3_hdr *ppd;
951
952 ppd = (struct tpacket3_hdr *)curr;
953 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
954 pkc->prev = curr;
955 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
956 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
957 BLOCK_NUM_PKTS(pbd) += 1;
958 atomic_inc(&pkc->blk_fill_in_prog);
959 prb_run_all_ft_ops(pkc, ppd);
960 }
961
962 /* Assumes caller has the sk->rx_queue.lock */
963 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
964 struct sk_buff *skb,
965 int status,
966 unsigned int len
967 )
968 {
969 struct tpacket_kbdq_core *pkc;
970 struct tpacket_block_desc *pbd;
971 char *curr, *end;
972
973 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
974 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
975
976 /* Queue is frozen when user space is lagging behind */
977 if (prb_queue_frozen(pkc)) {
978 /*
979 * Check if that last block which caused the queue to freeze,
980 * is still in_use by user-space.
981 */
982 if (prb_curr_blk_in_use(pkc, pbd)) {
983 /* Can't record this packet */
984 return NULL;
985 } else {
986 /*
987 * Ok, the block was released by user-space.
988 * Now let's open that block.
989 * opening a block also thaws the queue.
990 * Thawing is a side effect.
991 */
992 prb_open_block(pkc, pbd);
993 }
994 }
995
996 smp_mb();
997 curr = pkc->nxt_offset;
998 pkc->skb = skb;
999 end = (char *)pbd + pkc->kblk_size;
1000
1001 /* first try the current block */
1002 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1003 prb_fill_curr_block(curr, pkc, pbd, len);
1004 return (void *)curr;
1005 }
1006
1007 /* Ok, close the current block */
1008 prb_retire_current_block(pkc, po, 0);
1009
1010 /* Now, try to dispatch the next block */
1011 curr = (char *)prb_dispatch_next_block(pkc, po);
1012 if (curr) {
1013 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1014 prb_fill_curr_block(curr, pkc, pbd, len);
1015 return (void *)curr;
1016 }
1017
1018 /*
1019 * No free blocks are available.user_space hasn't caught up yet.
1020 * Queue was just frozen and now this packet will get dropped.
1021 */
1022 return NULL;
1023 }
1024
1025 static void *packet_current_rx_frame(struct packet_sock *po,
1026 struct sk_buff *skb,
1027 int status, unsigned int len)
1028 {
1029 char *curr = NULL;
1030 switch (po->tp_version) {
1031 case TPACKET_V1:
1032 case TPACKET_V2:
1033 curr = packet_lookup_frame(po, &po->rx_ring,
1034 po->rx_ring.head, status);
1035 return curr;
1036 case TPACKET_V3:
1037 return __packet_lookup_frame_in_block(po, skb, status, len);
1038 default:
1039 WARN(1, "TPACKET version not supported\n");
1040 BUG();
1041 return NULL;
1042 }
1043 }
1044
1045 static void *prb_lookup_block(struct packet_sock *po,
1046 struct packet_ring_buffer *rb,
1047 unsigned int idx,
1048 int status)
1049 {
1050 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1051 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1052
1053 if (status != BLOCK_STATUS(pbd))
1054 return NULL;
1055 return pbd;
1056 }
1057
1058 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1059 {
1060 unsigned int prev;
1061 if (rb->prb_bdqc.kactive_blk_num)
1062 prev = rb->prb_bdqc.kactive_blk_num-1;
1063 else
1064 prev = rb->prb_bdqc.knum_blocks-1;
1065 return prev;
1066 }
1067
1068 /* Assumes caller has held the rx_queue.lock */
1069 static void *__prb_previous_block(struct packet_sock *po,
1070 struct packet_ring_buffer *rb,
1071 int status)
1072 {
1073 unsigned int previous = prb_previous_blk_num(rb);
1074 return prb_lookup_block(po, rb, previous, status);
1075 }
1076
1077 static void *packet_previous_rx_frame(struct packet_sock *po,
1078 struct packet_ring_buffer *rb,
1079 int status)
1080 {
1081 if (po->tp_version <= TPACKET_V2)
1082 return packet_previous_frame(po, rb, status);
1083
1084 return __prb_previous_block(po, rb, status);
1085 }
1086
1087 static void packet_increment_rx_head(struct packet_sock *po,
1088 struct packet_ring_buffer *rb)
1089 {
1090 switch (po->tp_version) {
1091 case TPACKET_V1:
1092 case TPACKET_V2:
1093 return packet_increment_head(rb);
1094 case TPACKET_V3:
1095 default:
1096 WARN(1, "TPACKET version not supported.\n");
1097 BUG();
1098 return;
1099 }
1100 }
1101
1102 static void *packet_previous_frame(struct packet_sock *po,
1103 struct packet_ring_buffer *rb,
1104 int status)
1105 {
1106 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1107 return packet_lookup_frame(po, rb, previous, status);
1108 }
1109
1110 static void packet_increment_head(struct packet_ring_buffer *buff)
1111 {
1112 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1113 }
1114
1115 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1116 {
1117 struct sock *sk = &po->sk;
1118 bool has_room;
1119
1120 if (po->prot_hook.func != tpacket_rcv)
1121 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1122 <= sk->sk_rcvbuf;
1123
1124 spin_lock(&sk->sk_receive_queue.lock);
1125 if (po->tp_version == TPACKET_V3)
1126 has_room = prb_lookup_block(po, &po->rx_ring,
1127 po->rx_ring.prb_bdqc.kactive_blk_num,
1128 TP_STATUS_KERNEL);
1129 else
1130 has_room = packet_lookup_frame(po, &po->rx_ring,
1131 po->rx_ring.head,
1132 TP_STATUS_KERNEL);
1133 spin_unlock(&sk->sk_receive_queue.lock);
1134
1135 return has_room;
1136 }
1137
1138 static void packet_sock_destruct(struct sock *sk)
1139 {
1140 skb_queue_purge(&sk->sk_error_queue);
1141
1142 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1143 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1144
1145 if (!sock_flag(sk, SOCK_DEAD)) {
1146 pr_err("Attempt to release alive packet socket: %p\n", sk);
1147 return;
1148 }
1149
1150 sk_refcnt_debug_dec(sk);
1151 }
1152
1153 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1154 struct sk_buff *skb,
1155 unsigned int num)
1156 {
1157 return (((u64)skb->rxhash) * num) >> 32;
1158 }
1159
1160 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1161 struct sk_buff *skb,
1162 unsigned int num)
1163 {
1164 unsigned int val = atomic_inc_return(&f->rr_cur);
1165
1166 return val % num;
1167 }
1168
1169 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1170 struct sk_buff *skb,
1171 unsigned int num)
1172 {
1173 return smp_processor_id() % num;
1174 }
1175
1176 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1177 struct sk_buff *skb,
1178 unsigned int idx, unsigned int skip,
1179 unsigned int num)
1180 {
1181 unsigned int i, j;
1182
1183 i = j = min_t(int, f->next[idx], num - 1);
1184 do {
1185 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1186 if (i != j)
1187 f->next[idx] = i;
1188 return i;
1189 }
1190 if (++i == num)
1191 i = 0;
1192 } while (i != j);
1193
1194 return idx;
1195 }
1196
1197 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1198 {
1199 return f->flags & (flag >> 8);
1200 }
1201
1202 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1203 struct packet_type *pt, struct net_device *orig_dev)
1204 {
1205 struct packet_fanout *f = pt->af_packet_priv;
1206 unsigned int num = ACCESS_ONCE(f->num_members);
1207 struct packet_sock *po;
1208 unsigned int idx;
1209
1210 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1211 !num) {
1212 kfree_skb(skb);
1213 return 0;
1214 }
1215
1216 switch (f->type) {
1217 case PACKET_FANOUT_HASH:
1218 default:
1219 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1220 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1221 if (!skb)
1222 return 0;
1223 }
1224 skb_get_rxhash(skb);
1225 idx = fanout_demux_hash(f, skb, num);
1226 break;
1227 case PACKET_FANOUT_LB:
1228 idx = fanout_demux_lb(f, skb, num);
1229 break;
1230 case PACKET_FANOUT_CPU:
1231 idx = fanout_demux_cpu(f, skb, num);
1232 break;
1233 case PACKET_FANOUT_ROLLOVER:
1234 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1235 break;
1236 }
1237
1238 po = pkt_sk(f->arr[idx]);
1239 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1240 unlikely(!packet_rcv_has_room(po, skb))) {
1241 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1242 po = pkt_sk(f->arr[idx]);
1243 }
1244
1245 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1246 }
1247
1248 DEFINE_MUTEX(fanout_mutex);
1249 EXPORT_SYMBOL_GPL(fanout_mutex);
1250 static LIST_HEAD(fanout_list);
1251
1252 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1253 {
1254 struct packet_fanout *f = po->fanout;
1255
1256 spin_lock(&f->lock);
1257 f->arr[f->num_members] = sk;
1258 smp_wmb();
1259 f->num_members++;
1260 spin_unlock(&f->lock);
1261 }
1262
1263 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1264 {
1265 struct packet_fanout *f = po->fanout;
1266 int i;
1267
1268 spin_lock(&f->lock);
1269 for (i = 0; i < f->num_members; i++) {
1270 if (f->arr[i] == sk)
1271 break;
1272 }
1273 BUG_ON(i >= f->num_members);
1274 f->arr[i] = f->arr[f->num_members - 1];
1275 f->num_members--;
1276 spin_unlock(&f->lock);
1277 }
1278
1279 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1280 {
1281 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1282 return true;
1283
1284 return false;
1285 }
1286
1287 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1288 {
1289 struct packet_sock *po = pkt_sk(sk);
1290 struct packet_fanout *f, *match;
1291 u8 type = type_flags & 0xff;
1292 u8 flags = type_flags >> 8;
1293 int err;
1294
1295 switch (type) {
1296 case PACKET_FANOUT_ROLLOVER:
1297 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1298 return -EINVAL;
1299 case PACKET_FANOUT_HASH:
1300 case PACKET_FANOUT_LB:
1301 case PACKET_FANOUT_CPU:
1302 break;
1303 default:
1304 return -EINVAL;
1305 }
1306
1307 if (!po->running)
1308 return -EINVAL;
1309
1310 if (po->fanout)
1311 return -EALREADY;
1312
1313 mutex_lock(&fanout_mutex);
1314 match = NULL;
1315 list_for_each_entry(f, &fanout_list, list) {
1316 if (f->id == id &&
1317 read_pnet(&f->net) == sock_net(sk)) {
1318 match = f;
1319 break;
1320 }
1321 }
1322 err = -EINVAL;
1323 if (match && match->flags != flags)
1324 goto out;
1325 if (!match) {
1326 err = -ENOMEM;
1327 match = kzalloc(sizeof(*match), GFP_KERNEL);
1328 if (!match)
1329 goto out;
1330 write_pnet(&match->net, sock_net(sk));
1331 match->id = id;
1332 match->type = type;
1333 match->flags = flags;
1334 atomic_set(&match->rr_cur, 0);
1335 INIT_LIST_HEAD(&match->list);
1336 spin_lock_init(&match->lock);
1337 atomic_set(&match->sk_ref, 0);
1338 match->prot_hook.type = po->prot_hook.type;
1339 match->prot_hook.dev = po->prot_hook.dev;
1340 match->prot_hook.func = packet_rcv_fanout;
1341 match->prot_hook.af_packet_priv = match;
1342 match->prot_hook.id_match = match_fanout_group;
1343 dev_add_pack(&match->prot_hook);
1344 list_add(&match->list, &fanout_list);
1345 }
1346 err = -EINVAL;
1347 if (match->type == type &&
1348 match->prot_hook.type == po->prot_hook.type &&
1349 match->prot_hook.dev == po->prot_hook.dev) {
1350 err = -ENOSPC;
1351 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1352 __dev_remove_pack(&po->prot_hook);
1353 po->fanout = match;
1354 atomic_inc(&match->sk_ref);
1355 __fanout_link(sk, po);
1356 err = 0;
1357 }
1358 }
1359 out:
1360 mutex_unlock(&fanout_mutex);
1361 return err;
1362 }
1363
1364 static void fanout_release(struct sock *sk)
1365 {
1366 struct packet_sock *po = pkt_sk(sk);
1367 struct packet_fanout *f;
1368
1369 f = po->fanout;
1370 if (!f)
1371 return;
1372
1373 mutex_lock(&fanout_mutex);
1374 po->fanout = NULL;
1375
1376 if (atomic_dec_and_test(&f->sk_ref)) {
1377 list_del(&f->list);
1378 dev_remove_pack(&f->prot_hook);
1379 kfree(f);
1380 }
1381 mutex_unlock(&fanout_mutex);
1382 }
1383
1384 static const struct proto_ops packet_ops;
1385
1386 static const struct proto_ops packet_ops_spkt;
1387
1388 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1389 struct packet_type *pt, struct net_device *orig_dev)
1390 {
1391 struct sock *sk;
1392 struct sockaddr_pkt *spkt;
1393
1394 /*
1395 * When we registered the protocol we saved the socket in the data
1396 * field for just this event.
1397 */
1398
1399 sk = pt->af_packet_priv;
1400
1401 /*
1402 * Yank back the headers [hope the device set this
1403 * right or kerboom...]
1404 *
1405 * Incoming packets have ll header pulled,
1406 * push it back.
1407 *
1408 * For outgoing ones skb->data == skb_mac_header(skb)
1409 * so that this procedure is noop.
1410 */
1411
1412 if (skb->pkt_type == PACKET_LOOPBACK)
1413 goto out;
1414
1415 if (!net_eq(dev_net(dev), sock_net(sk)))
1416 goto out;
1417
1418 skb = skb_share_check(skb, GFP_ATOMIC);
1419 if (skb == NULL)
1420 goto oom;
1421
1422 /* drop any routing info */
1423 skb_dst_drop(skb);
1424
1425 /* drop conntrack reference */
1426 nf_reset(skb);
1427
1428 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1429
1430 skb_push(skb, skb->data - skb_mac_header(skb));
1431
1432 /*
1433 * The SOCK_PACKET socket receives _all_ frames.
1434 */
1435
1436 spkt->spkt_family = dev->type;
1437 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1438 spkt->spkt_protocol = skb->protocol;
1439
1440 /*
1441 * Charge the memory to the socket. This is done specifically
1442 * to prevent sockets using all the memory up.
1443 */
1444
1445 if (sock_queue_rcv_skb(sk, skb) == 0)
1446 return 0;
1447
1448 out:
1449 kfree_skb(skb);
1450 oom:
1451 return 0;
1452 }
1453
1454
1455 /*
1456 * Output a raw packet to a device layer. This bypasses all the other
1457 * protocol layers and you must therefore supply it with a complete frame
1458 */
1459
1460 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1461 struct msghdr *msg, size_t len)
1462 {
1463 struct sock *sk = sock->sk;
1464 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1465 struct sk_buff *skb = NULL;
1466 struct net_device *dev;
1467 __be16 proto = 0;
1468 int err;
1469 int extra_len = 0;
1470
1471 /*
1472 * Get and verify the address.
1473 */
1474
1475 if (saddr) {
1476 if (msg->msg_namelen < sizeof(struct sockaddr))
1477 return -EINVAL;
1478 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1479 proto = saddr->spkt_protocol;
1480 } else
1481 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1482
1483 /*
1484 * Find the device first to size check it
1485 */
1486
1487 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1488 retry:
1489 rcu_read_lock();
1490 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1491 err = -ENODEV;
1492 if (dev == NULL)
1493 goto out_unlock;
1494
1495 err = -ENETDOWN;
1496 if (!(dev->flags & IFF_UP))
1497 goto out_unlock;
1498
1499 /*
1500 * You may not queue a frame bigger than the mtu. This is the lowest level
1501 * raw protocol and you must do your own fragmentation at this level.
1502 */
1503
1504 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1505 if (!netif_supports_nofcs(dev)) {
1506 err = -EPROTONOSUPPORT;
1507 goto out_unlock;
1508 }
1509 extra_len = 4; /* We're doing our own CRC */
1510 }
1511
1512 err = -EMSGSIZE;
1513 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1514 goto out_unlock;
1515
1516 if (!skb) {
1517 size_t reserved = LL_RESERVED_SPACE(dev);
1518 int tlen = dev->needed_tailroom;
1519 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1520
1521 rcu_read_unlock();
1522 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1523 if (skb == NULL)
1524 return -ENOBUFS;
1525 /* FIXME: Save some space for broken drivers that write a hard
1526 * header at transmission time by themselves. PPP is the notable
1527 * one here. This should really be fixed at the driver level.
1528 */
1529 skb_reserve(skb, reserved);
1530 skb_reset_network_header(skb);
1531
1532 /* Try to align data part correctly */
1533 if (hhlen) {
1534 skb->data -= hhlen;
1535 skb->tail -= hhlen;
1536 if (len < hhlen)
1537 skb_reset_network_header(skb);
1538 }
1539 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1540 if (err)
1541 goto out_free;
1542 goto retry;
1543 }
1544
1545 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1546 /* Earlier code assumed this would be a VLAN pkt,
1547 * double-check this now that we have the actual
1548 * packet in hand.
1549 */
1550 struct ethhdr *ehdr;
1551 skb_reset_mac_header(skb);
1552 ehdr = eth_hdr(skb);
1553 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1554 err = -EMSGSIZE;
1555 goto out_unlock;
1556 }
1557 }
1558
1559 skb->protocol = proto;
1560 skb->dev = dev;
1561 skb->priority = sk->sk_priority;
1562 skb->mark = sk->sk_mark;
1563
1564 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1565
1566 if (unlikely(extra_len == 4))
1567 skb->no_fcs = 1;
1568
1569 skb_probe_transport_header(skb, 0);
1570
1571 dev_queue_xmit(skb);
1572 rcu_read_unlock();
1573 return len;
1574
1575 out_unlock:
1576 rcu_read_unlock();
1577 out_free:
1578 kfree_skb(skb);
1579 return err;
1580 }
1581
1582 static unsigned int run_filter(const struct sk_buff *skb,
1583 const struct sock *sk,
1584 unsigned int res)
1585 {
1586 struct sk_filter *filter;
1587
1588 rcu_read_lock();
1589 filter = rcu_dereference(sk->sk_filter);
1590 if (filter != NULL)
1591 res = SK_RUN_FILTER(filter, skb);
1592 rcu_read_unlock();
1593
1594 return res;
1595 }
1596
1597 /*
1598 * This function makes lazy skb cloning in hope that most of packets
1599 * are discarded by BPF.
1600 *
1601 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1602 * and skb->cb are mangled. It works because (and until) packets
1603 * falling here are owned by current CPU. Output packets are cloned
1604 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1605 * sequencially, so that if we return skb to original state on exit,
1606 * we will not harm anyone.
1607 */
1608
1609 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1610 struct packet_type *pt, struct net_device *orig_dev)
1611 {
1612 struct sock *sk;
1613 struct sockaddr_ll *sll;
1614 struct packet_sock *po;
1615 u8 *skb_head = skb->data;
1616 int skb_len = skb->len;
1617 unsigned int snaplen, res;
1618
1619 if (skb->pkt_type == PACKET_LOOPBACK)
1620 goto drop;
1621
1622 sk = pt->af_packet_priv;
1623 po = pkt_sk(sk);
1624
1625 if (!net_eq(dev_net(dev), sock_net(sk)))
1626 goto drop;
1627
1628 skb->dev = dev;
1629
1630 if (dev->header_ops) {
1631 /* The device has an explicit notion of ll header,
1632 * exported to higher levels.
1633 *
1634 * Otherwise, the device hides details of its frame
1635 * structure, so that corresponding packet head is
1636 * never delivered to user.
1637 */
1638 if (sk->sk_type != SOCK_DGRAM)
1639 skb_push(skb, skb->data - skb_mac_header(skb));
1640 else if (skb->pkt_type == PACKET_OUTGOING) {
1641 /* Special case: outgoing packets have ll header at head */
1642 skb_pull(skb, skb_network_offset(skb));
1643 }
1644 }
1645
1646 snaplen = skb->len;
1647
1648 res = run_filter(skb, sk, snaplen);
1649 if (!res)
1650 goto drop_n_restore;
1651 if (snaplen > res)
1652 snaplen = res;
1653
1654 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1655 goto drop_n_acct;
1656
1657 if (skb_shared(skb)) {
1658 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1659 if (nskb == NULL)
1660 goto drop_n_acct;
1661
1662 if (skb_head != skb->data) {
1663 skb->data = skb_head;
1664 skb->len = skb_len;
1665 }
1666 consume_skb(skb);
1667 skb = nskb;
1668 }
1669
1670 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1671 sizeof(skb->cb));
1672
1673 sll = &PACKET_SKB_CB(skb)->sa.ll;
1674 sll->sll_family = AF_PACKET;
1675 sll->sll_hatype = dev->type;
1676 sll->sll_protocol = skb->protocol;
1677 sll->sll_pkttype = skb->pkt_type;
1678 if (unlikely(po->origdev))
1679 sll->sll_ifindex = orig_dev->ifindex;
1680 else
1681 sll->sll_ifindex = dev->ifindex;
1682
1683 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1684
1685 PACKET_SKB_CB(skb)->origlen = skb->len;
1686
1687 if (pskb_trim(skb, snaplen))
1688 goto drop_n_acct;
1689
1690 skb_set_owner_r(skb, sk);
1691 skb->dev = NULL;
1692 skb_dst_drop(skb);
1693
1694 /* drop conntrack reference */
1695 nf_reset(skb);
1696
1697 spin_lock(&sk->sk_receive_queue.lock);
1698 po->stats.stats1.tp_packets++;
1699 skb->dropcount = atomic_read(&sk->sk_drops);
1700 __skb_queue_tail(&sk->sk_receive_queue, skb);
1701 spin_unlock(&sk->sk_receive_queue.lock);
1702 sk->sk_data_ready(sk, skb->len);
1703 return 0;
1704
1705 drop_n_acct:
1706 spin_lock(&sk->sk_receive_queue.lock);
1707 po->stats.stats1.tp_drops++;
1708 atomic_inc(&sk->sk_drops);
1709 spin_unlock(&sk->sk_receive_queue.lock);
1710
1711 drop_n_restore:
1712 if (skb_head != skb->data && skb_shared(skb)) {
1713 skb->data = skb_head;
1714 skb->len = skb_len;
1715 }
1716 drop:
1717 consume_skb(skb);
1718 return 0;
1719 }
1720
1721 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1722 struct packet_type *pt, struct net_device *orig_dev)
1723 {
1724 struct sock *sk;
1725 struct packet_sock *po;
1726 struct sockaddr_ll *sll;
1727 union tpacket_uhdr h;
1728 u8 *skb_head = skb->data;
1729 int skb_len = skb->len;
1730 unsigned int snaplen, res;
1731 unsigned long status = TP_STATUS_USER;
1732 unsigned short macoff, netoff, hdrlen;
1733 struct sk_buff *copy_skb = NULL;
1734 struct timespec ts;
1735 __u32 ts_status;
1736
1737 if (skb->pkt_type == PACKET_LOOPBACK)
1738 goto drop;
1739
1740 sk = pt->af_packet_priv;
1741 po = pkt_sk(sk);
1742
1743 if (!net_eq(dev_net(dev), sock_net(sk)))
1744 goto drop;
1745
1746 if (dev->header_ops) {
1747 if (sk->sk_type != SOCK_DGRAM)
1748 skb_push(skb, skb->data - skb_mac_header(skb));
1749 else if (skb->pkt_type == PACKET_OUTGOING) {
1750 /* Special case: outgoing packets have ll header at head */
1751 skb_pull(skb, skb_network_offset(skb));
1752 }
1753 }
1754
1755 if (skb->ip_summed == CHECKSUM_PARTIAL)
1756 status |= TP_STATUS_CSUMNOTREADY;
1757
1758 snaplen = skb->len;
1759
1760 res = run_filter(skb, sk, snaplen);
1761 if (!res)
1762 goto drop_n_restore;
1763 if (snaplen > res)
1764 snaplen = res;
1765
1766 if (sk->sk_type == SOCK_DGRAM) {
1767 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1768 po->tp_reserve;
1769 } else {
1770 unsigned int maclen = skb_network_offset(skb);
1771 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1772 (maclen < 16 ? 16 : maclen)) +
1773 po->tp_reserve;
1774 macoff = netoff - maclen;
1775 }
1776 if (po->tp_version <= TPACKET_V2) {
1777 if (macoff + snaplen > po->rx_ring.frame_size) {
1778 if (po->copy_thresh &&
1779 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1780 if (skb_shared(skb)) {
1781 copy_skb = skb_clone(skb, GFP_ATOMIC);
1782 } else {
1783 copy_skb = skb_get(skb);
1784 skb_head = skb->data;
1785 }
1786 if (copy_skb)
1787 skb_set_owner_r(copy_skb, sk);
1788 }
1789 snaplen = po->rx_ring.frame_size - macoff;
1790 if ((int)snaplen < 0)
1791 snaplen = 0;
1792 }
1793 } else if (unlikely(macoff + snaplen >
1794 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1795 u32 nval;
1796
1797 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1798 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1799 snaplen, nval, macoff);
1800 snaplen = nval;
1801 if (unlikely((int)snaplen < 0)) {
1802 snaplen = 0;
1803 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1804 }
1805 }
1806 spin_lock(&sk->sk_receive_queue.lock);
1807 h.raw = packet_current_rx_frame(po, skb,
1808 TP_STATUS_KERNEL, (macoff+snaplen));
1809 if (!h.raw)
1810 goto ring_is_full;
1811 if (po->tp_version <= TPACKET_V2) {
1812 packet_increment_rx_head(po, &po->rx_ring);
1813 /*
1814 * LOSING will be reported till you read the stats,
1815 * because it's COR - Clear On Read.
1816 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1817 * at packet level.
1818 */
1819 if (po->stats.stats1.tp_drops)
1820 status |= TP_STATUS_LOSING;
1821 }
1822 po->stats.stats1.tp_packets++;
1823 if (copy_skb) {
1824 status |= TP_STATUS_COPY;
1825 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1826 }
1827 spin_unlock(&sk->sk_receive_queue.lock);
1828
1829 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1830
1831 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1832 getnstimeofday(&ts);
1833
1834 status |= ts_status;
1835
1836 switch (po->tp_version) {
1837 case TPACKET_V1:
1838 h.h1->tp_len = skb->len;
1839 h.h1->tp_snaplen = snaplen;
1840 h.h1->tp_mac = macoff;
1841 h.h1->tp_net = netoff;
1842 h.h1->tp_sec = ts.tv_sec;
1843 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1844 hdrlen = sizeof(*h.h1);
1845 break;
1846 case TPACKET_V2:
1847 h.h2->tp_len = skb->len;
1848 h.h2->tp_snaplen = snaplen;
1849 h.h2->tp_mac = macoff;
1850 h.h2->tp_net = netoff;
1851 h.h2->tp_sec = ts.tv_sec;
1852 h.h2->tp_nsec = ts.tv_nsec;
1853 if (vlan_tx_tag_present(skb)) {
1854 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1855 status |= TP_STATUS_VLAN_VALID;
1856 } else {
1857 h.h2->tp_vlan_tci = 0;
1858 }
1859 h.h2->tp_padding = 0;
1860 hdrlen = sizeof(*h.h2);
1861 break;
1862 case TPACKET_V3:
1863 /* tp_nxt_offset,vlan are already populated above.
1864 * So DONT clear those fields here
1865 */
1866 h.h3->tp_status |= status;
1867 h.h3->tp_len = skb->len;
1868 h.h3->tp_snaplen = snaplen;
1869 h.h3->tp_mac = macoff;
1870 h.h3->tp_net = netoff;
1871 h.h3->tp_sec = ts.tv_sec;
1872 h.h3->tp_nsec = ts.tv_nsec;
1873 hdrlen = sizeof(*h.h3);
1874 break;
1875 default:
1876 BUG();
1877 }
1878
1879 sll = h.raw + TPACKET_ALIGN(hdrlen);
1880 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1881 sll->sll_family = AF_PACKET;
1882 sll->sll_hatype = dev->type;
1883 sll->sll_protocol = skb->protocol;
1884 sll->sll_pkttype = skb->pkt_type;
1885 if (unlikely(po->origdev))
1886 sll->sll_ifindex = orig_dev->ifindex;
1887 else
1888 sll->sll_ifindex = dev->ifindex;
1889
1890 smp_mb();
1891 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1892 {
1893 u8 *start, *end;
1894
1895 if (po->tp_version <= TPACKET_V2) {
1896 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1897 + macoff + snaplen);
1898 for (start = h.raw; start < end; start += PAGE_SIZE)
1899 flush_dcache_page(pgv_to_page(start));
1900 }
1901 smp_wmb();
1902 }
1903 #endif
1904 if (po->tp_version <= TPACKET_V2)
1905 __packet_set_status(po, h.raw, status);
1906 else
1907 prb_clear_blk_fill_status(&po->rx_ring);
1908
1909 sk->sk_data_ready(sk, 0);
1910
1911 drop_n_restore:
1912 if (skb_head != skb->data && skb_shared(skb)) {
1913 skb->data = skb_head;
1914 skb->len = skb_len;
1915 }
1916 drop:
1917 kfree_skb(skb);
1918 return 0;
1919
1920 ring_is_full:
1921 po->stats.stats1.tp_drops++;
1922 spin_unlock(&sk->sk_receive_queue.lock);
1923
1924 sk->sk_data_ready(sk, 0);
1925 kfree_skb(copy_skb);
1926 goto drop_n_restore;
1927 }
1928
1929 static void tpacket_destruct_skb(struct sk_buff *skb)
1930 {
1931 struct packet_sock *po = pkt_sk(skb->sk);
1932 void *ph;
1933
1934 if (likely(po->tx_ring.pg_vec)) {
1935 __u32 ts;
1936
1937 ph = skb_shinfo(skb)->destructor_arg;
1938 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1939 atomic_dec(&po->tx_ring.pending);
1940
1941 ts = __packet_set_timestamp(po, ph, skb);
1942 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1943 }
1944
1945 sock_wfree(skb);
1946 }
1947
1948 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1949 void *frame, struct net_device *dev, int size_max,
1950 __be16 proto, unsigned char *addr, int hlen)
1951 {
1952 union tpacket_uhdr 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 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1966 skb_shinfo(skb)->destructor_arg = ph.raw;
1967
1968 switch (po->tp_version) {
1969 case TPACKET_V2:
1970 tp_len = ph.h2->tp_len;
1971 break;
1972 default:
1973 tp_len = ph.h1->tp_len;
1974 break;
1975 }
1976 if (unlikely(tp_len > size_max)) {
1977 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1978 return -EMSGSIZE;
1979 }
1980
1981 skb_reserve(skb, hlen);
1982 skb_reset_network_header(skb);
1983 skb_probe_transport_header(skb, 0);
1984
1985 if (po->tp_tx_has_off) {
1986 int off_min, off_max, off;
1987 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1988 off_max = po->tx_ring.frame_size - tp_len;
1989 if (sock->type == SOCK_DGRAM) {
1990 switch (po->tp_version) {
1991 case TPACKET_V2:
1992 off = ph.h2->tp_net;
1993 break;
1994 default:
1995 off = ph.h1->tp_net;
1996 break;
1997 }
1998 } else {
1999 switch (po->tp_version) {
2000 case TPACKET_V2:
2001 off = ph.h2->tp_mac;
2002 break;
2003 default:
2004 off = ph.h1->tp_mac;
2005 break;
2006 }
2007 }
2008 if (unlikely((off < off_min) || (off_max < off)))
2009 return -EINVAL;
2010 data = ph.raw + off;
2011 } else {
2012 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2013 }
2014 to_write = tp_len;
2015
2016 if (sock->type == SOCK_DGRAM) {
2017 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2018 NULL, tp_len);
2019 if (unlikely(err < 0))
2020 return -EINVAL;
2021 } else if (dev->hard_header_len) {
2022 /* net device doesn't like empty head */
2023 if (unlikely(tp_len <= dev->hard_header_len)) {
2024 pr_err("packet size is too short (%d < %d)\n",
2025 tp_len, dev->hard_header_len);
2026 return -EINVAL;
2027 }
2028
2029 skb_push(skb, dev->hard_header_len);
2030 err = skb_store_bits(skb, 0, data,
2031 dev->hard_header_len);
2032 if (unlikely(err))
2033 return err;
2034
2035 data += dev->hard_header_len;
2036 to_write -= dev->hard_header_len;
2037 }
2038
2039 offset = offset_in_page(data);
2040 len_max = PAGE_SIZE - offset;
2041 len = ((to_write > len_max) ? len_max : to_write);
2042
2043 skb->data_len = to_write;
2044 skb->len += to_write;
2045 skb->truesize += to_write;
2046 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2047
2048 while (likely(to_write)) {
2049 nr_frags = skb_shinfo(skb)->nr_frags;
2050
2051 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2052 pr_err("Packet exceed the number of skb frags(%lu)\n",
2053 MAX_SKB_FRAGS);
2054 return -EFAULT;
2055 }
2056
2057 page = pgv_to_page(data);
2058 data += len;
2059 flush_dcache_page(page);
2060 get_page(page);
2061 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2062 to_write -= len;
2063 offset = 0;
2064 len_max = PAGE_SIZE;
2065 len = ((to_write > len_max) ? len_max : to_write);
2066 }
2067
2068 return tp_len;
2069 }
2070
2071 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2072 {
2073 struct sk_buff *skb;
2074 struct net_device *dev;
2075 __be16 proto;
2076 int err, reserve = 0;
2077 void *ph;
2078 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2079 int tp_len, size_max;
2080 unsigned char *addr;
2081 int len_sum = 0;
2082 int status = TP_STATUS_AVAILABLE;
2083 int hlen, tlen;
2084
2085 mutex_lock(&po->pg_vec_lock);
2086
2087 if (likely(saddr == NULL)) {
2088 dev = packet_cached_dev_get(po);
2089 proto = po->num;
2090 addr = NULL;
2091 } else {
2092 err = -EINVAL;
2093 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2094 goto out;
2095 if (msg->msg_namelen < (saddr->sll_halen
2096 + offsetof(struct sockaddr_ll,
2097 sll_addr)))
2098 goto out;
2099 proto = saddr->sll_protocol;
2100 addr = saddr->sll_addr;
2101 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2102 }
2103
2104 err = -ENXIO;
2105 if (unlikely(dev == NULL))
2106 goto out;
2107 err = -ENETDOWN;
2108 if (unlikely(!(dev->flags & IFF_UP)))
2109 goto out_put;
2110
2111 reserve = dev->hard_header_len;
2112
2113 size_max = po->tx_ring.frame_size
2114 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2115
2116 if (size_max > dev->mtu + reserve)
2117 size_max = dev->mtu + reserve;
2118
2119 do {
2120 ph = packet_current_frame(po, &po->tx_ring,
2121 TP_STATUS_SEND_REQUEST);
2122
2123 if (unlikely(ph == NULL)) {
2124 schedule();
2125 continue;
2126 }
2127
2128 status = TP_STATUS_SEND_REQUEST;
2129 hlen = LL_RESERVED_SPACE(dev);
2130 tlen = dev->needed_tailroom;
2131 skb = sock_alloc_send_skb(&po->sk,
2132 hlen + tlen + sizeof(struct sockaddr_ll),
2133 0, &err);
2134
2135 if (unlikely(skb == NULL))
2136 goto out_status;
2137
2138 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2139 addr, hlen);
2140
2141 if (unlikely(tp_len < 0)) {
2142 if (po->tp_loss) {
2143 __packet_set_status(po, ph,
2144 TP_STATUS_AVAILABLE);
2145 packet_increment_head(&po->tx_ring);
2146 kfree_skb(skb);
2147 continue;
2148 } else {
2149 status = TP_STATUS_WRONG_FORMAT;
2150 err = tp_len;
2151 goto out_status;
2152 }
2153 }
2154
2155 skb->destructor = tpacket_destruct_skb;
2156 __packet_set_status(po, ph, TP_STATUS_SENDING);
2157 atomic_inc(&po->tx_ring.pending);
2158
2159 status = TP_STATUS_SEND_REQUEST;
2160 err = dev_queue_xmit(skb);
2161 if (unlikely(err > 0)) {
2162 err = net_xmit_errno(err);
2163 if (err && __packet_get_status(po, ph) ==
2164 TP_STATUS_AVAILABLE) {
2165 /* skb was destructed already */
2166 skb = NULL;
2167 goto out_status;
2168 }
2169 /*
2170 * skb was dropped but not destructed yet;
2171 * let's treat it like congestion or err < 0
2172 */
2173 err = 0;
2174 }
2175 packet_increment_head(&po->tx_ring);
2176 len_sum += tp_len;
2177 } while (likely((ph != NULL) ||
2178 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2179 (atomic_read(&po->tx_ring.pending))))
2180 );
2181
2182 err = len_sum;
2183 goto out_put;
2184
2185 out_status:
2186 __packet_set_status(po, ph, status);
2187 kfree_skb(skb);
2188 out_put:
2189 dev_put(dev);
2190 out:
2191 mutex_unlock(&po->pg_vec_lock);
2192 return err;
2193 }
2194
2195 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2196 size_t reserve, size_t len,
2197 size_t linear, int noblock,
2198 int *err)
2199 {
2200 struct sk_buff *skb;
2201
2202 /* Under a page? Don't bother with paged skb. */
2203 if (prepad + len < PAGE_SIZE || !linear)
2204 linear = len;
2205
2206 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2207 err);
2208 if (!skb)
2209 return NULL;
2210
2211 skb_reserve(skb, reserve);
2212 skb_put(skb, linear);
2213 skb->data_len = len - linear;
2214 skb->len += len - linear;
2215
2216 return skb;
2217 }
2218
2219 static int packet_snd(struct socket *sock,
2220 struct msghdr *msg, size_t len)
2221 {
2222 struct sock *sk = sock->sk;
2223 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2224 struct sk_buff *skb;
2225 struct net_device *dev;
2226 __be16 proto;
2227 unsigned char *addr;
2228 int err, reserve = 0;
2229 struct virtio_net_hdr vnet_hdr = { 0 };
2230 int offset = 0;
2231 int vnet_hdr_len;
2232 struct packet_sock *po = pkt_sk(sk);
2233 unsigned short gso_type = 0;
2234 int hlen, tlen;
2235 int extra_len = 0;
2236
2237 /*
2238 * Get and verify the address.
2239 */
2240
2241 if (likely(saddr == NULL)) {
2242 dev = packet_cached_dev_get(po);
2243 proto = po->num;
2244 addr = NULL;
2245 } else {
2246 err = -EINVAL;
2247 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2248 goto out;
2249 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2250 goto out;
2251 proto = saddr->sll_protocol;
2252 addr = saddr->sll_addr;
2253 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2254 }
2255
2256 err = -ENXIO;
2257 if (unlikely(dev == NULL))
2258 goto out_unlock;
2259 err = -ENETDOWN;
2260 if (unlikely(!(dev->flags & IFF_UP)))
2261 goto out_unlock;
2262
2263 if (sock->type == SOCK_RAW)
2264 reserve = dev->hard_header_len;
2265 if (po->has_vnet_hdr) {
2266 vnet_hdr_len = sizeof(vnet_hdr);
2267
2268 err = -EINVAL;
2269 if (len < vnet_hdr_len)
2270 goto out_unlock;
2271
2272 len -= vnet_hdr_len;
2273
2274 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2275 vnet_hdr_len);
2276 if (err < 0)
2277 goto out_unlock;
2278
2279 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2280 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2281 vnet_hdr.hdr_len))
2282 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2283 vnet_hdr.csum_offset + 2;
2284
2285 err = -EINVAL;
2286 if (vnet_hdr.hdr_len > len)
2287 goto out_unlock;
2288
2289 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2290 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2291 case VIRTIO_NET_HDR_GSO_TCPV4:
2292 gso_type = SKB_GSO_TCPV4;
2293 break;
2294 case VIRTIO_NET_HDR_GSO_TCPV6:
2295 gso_type = SKB_GSO_TCPV6;
2296 break;
2297 case VIRTIO_NET_HDR_GSO_UDP:
2298 gso_type = SKB_GSO_UDP;
2299 break;
2300 default:
2301 goto out_unlock;
2302 }
2303
2304 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2305 gso_type |= SKB_GSO_TCP_ECN;
2306
2307 if (vnet_hdr.gso_size == 0)
2308 goto out_unlock;
2309
2310 }
2311 }
2312
2313 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2314 if (!netif_supports_nofcs(dev)) {
2315 err = -EPROTONOSUPPORT;
2316 goto out_unlock;
2317 }
2318 extra_len = 4; /* We're doing our own CRC */
2319 }
2320
2321 err = -EMSGSIZE;
2322 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2323 goto out_unlock;
2324
2325 err = -ENOBUFS;
2326 hlen = LL_RESERVED_SPACE(dev);
2327 tlen = dev->needed_tailroom;
2328 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2329 msg->msg_flags & MSG_DONTWAIT, &err);
2330 if (skb == NULL)
2331 goto out_unlock;
2332
2333 skb_set_network_header(skb, reserve);
2334
2335 err = -EINVAL;
2336 if (sock->type == SOCK_DGRAM &&
2337 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2338 goto out_free;
2339
2340 /* Returns -EFAULT on error */
2341 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2342 if (err)
2343 goto out_free;
2344
2345 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2346
2347 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2348 /* Earlier code assumed this would be a VLAN pkt,
2349 * double-check this now that we have the actual
2350 * packet in hand.
2351 */
2352 struct ethhdr *ehdr;
2353 skb_reset_mac_header(skb);
2354 ehdr = eth_hdr(skb);
2355 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2356 err = -EMSGSIZE;
2357 goto out_free;
2358 }
2359 }
2360
2361 skb->protocol = proto;
2362 skb->dev = dev;
2363 skb->priority = sk->sk_priority;
2364 skb->mark = sk->sk_mark;
2365
2366 if (po->has_vnet_hdr) {
2367 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2368 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2369 vnet_hdr.csum_offset)) {
2370 err = -EINVAL;
2371 goto out_free;
2372 }
2373 }
2374
2375 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2376 skb_shinfo(skb)->gso_type = gso_type;
2377
2378 /* Header must be checked, and gso_segs computed. */
2379 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2380 skb_shinfo(skb)->gso_segs = 0;
2381
2382 len += vnet_hdr_len;
2383 }
2384
2385 skb_probe_transport_header(skb, reserve);
2386
2387 if (unlikely(extra_len == 4))
2388 skb->no_fcs = 1;
2389
2390 /*
2391 * Now send it
2392 */
2393
2394 err = dev_queue_xmit(skb);
2395 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2396 goto out_unlock;
2397
2398 dev_put(dev);
2399
2400 return len;
2401
2402 out_free:
2403 kfree_skb(skb);
2404 out_unlock:
2405 if (dev)
2406 dev_put(dev);
2407 out:
2408 return err;
2409 }
2410
2411 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2412 struct msghdr *msg, size_t len)
2413 {
2414 struct sock *sk = sock->sk;
2415 struct packet_sock *po = pkt_sk(sk);
2416 if (po->tx_ring.pg_vec)
2417 return tpacket_snd(po, msg);
2418 else
2419 return packet_snd(sock, msg, len);
2420 }
2421
2422 /*
2423 * Close a PACKET socket. This is fairly simple. We immediately go
2424 * to 'closed' state and remove our protocol entry in the device list.
2425 */
2426
2427 static int packet_release(struct socket *sock)
2428 {
2429 struct sock *sk = sock->sk;
2430 struct packet_sock *po;
2431 struct net *net;
2432 union tpacket_req_u req_u;
2433
2434 if (!sk)
2435 return 0;
2436
2437 net = sock_net(sk);
2438 po = pkt_sk(sk);
2439
2440 mutex_lock(&net->packet.sklist_lock);
2441 sk_del_node_init_rcu(sk);
2442 mutex_unlock(&net->packet.sklist_lock);
2443
2444 preempt_disable();
2445 sock_prot_inuse_add(net, sk->sk_prot, -1);
2446 preempt_enable();
2447
2448 spin_lock(&po->bind_lock);
2449 unregister_prot_hook(sk, false);
2450 packet_cached_dev_reset(po);
2451
2452 if (po->prot_hook.dev) {
2453 dev_put(po->prot_hook.dev);
2454 po->prot_hook.dev = NULL;
2455 }
2456 spin_unlock(&po->bind_lock);
2457
2458 packet_flush_mclist(sk);
2459
2460 if (po->rx_ring.pg_vec) {
2461 memset(&req_u, 0, sizeof(req_u));
2462 packet_set_ring(sk, &req_u, 1, 0);
2463 }
2464
2465 if (po->tx_ring.pg_vec) {
2466 memset(&req_u, 0, sizeof(req_u));
2467 packet_set_ring(sk, &req_u, 1, 1);
2468 }
2469
2470 fanout_release(sk);
2471
2472 synchronize_net();
2473 /*
2474 * Now the socket is dead. No more input will appear.
2475 */
2476 sock_orphan(sk);
2477 sock->sk = NULL;
2478
2479 /* Purge queues */
2480
2481 skb_queue_purge(&sk->sk_receive_queue);
2482 sk_refcnt_debug_release(sk);
2483
2484 sock_put(sk);
2485 return 0;
2486 }
2487
2488 /*
2489 * Attach a packet hook.
2490 */
2491
2492 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2493 {
2494 struct packet_sock *po = pkt_sk(sk);
2495
2496 if (po->fanout) {
2497 if (dev)
2498 dev_put(dev);
2499
2500 return -EINVAL;
2501 }
2502
2503 lock_sock(sk);
2504
2505 spin_lock(&po->bind_lock);
2506 unregister_prot_hook(sk, true);
2507
2508 po->num = protocol;
2509 po->prot_hook.type = protocol;
2510 if (po->prot_hook.dev)
2511 dev_put(po->prot_hook.dev);
2512
2513 po->prot_hook.dev = dev;
2514 po->ifindex = dev ? dev->ifindex : 0;
2515
2516 packet_cached_dev_assign(po, dev);
2517
2518 if (protocol == 0)
2519 goto out_unlock;
2520
2521 if (!dev || (dev->flags & IFF_UP)) {
2522 register_prot_hook(sk);
2523 } else {
2524 sk->sk_err = ENETDOWN;
2525 if (!sock_flag(sk, SOCK_DEAD))
2526 sk->sk_error_report(sk);
2527 }
2528
2529 out_unlock:
2530 spin_unlock(&po->bind_lock);
2531 release_sock(sk);
2532 return 0;
2533 }
2534
2535 /*
2536 * Bind a packet socket to a device
2537 */
2538
2539 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2540 int addr_len)
2541 {
2542 struct sock *sk = sock->sk;
2543 char name[15];
2544 struct net_device *dev;
2545 int err = -ENODEV;
2546
2547 /*
2548 * Check legality
2549 */
2550
2551 if (addr_len != sizeof(struct sockaddr))
2552 return -EINVAL;
2553 strlcpy(name, uaddr->sa_data, sizeof(name));
2554
2555 dev = dev_get_by_name(sock_net(sk), name);
2556 if (dev)
2557 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2558 return err;
2559 }
2560
2561 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2562 {
2563 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2564 struct sock *sk = sock->sk;
2565 struct net_device *dev = NULL;
2566 int err;
2567
2568
2569 /*
2570 * Check legality
2571 */
2572
2573 if (addr_len < sizeof(struct sockaddr_ll))
2574 return -EINVAL;
2575 if (sll->sll_family != AF_PACKET)
2576 return -EINVAL;
2577
2578 if (sll->sll_ifindex) {
2579 err = -ENODEV;
2580 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2581 if (dev == NULL)
2582 goto out;
2583 }
2584 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2585
2586 out:
2587 return err;
2588 }
2589
2590 static struct proto packet_proto = {
2591 .name = "PACKET",
2592 .owner = THIS_MODULE,
2593 .obj_size = sizeof(struct packet_sock),
2594 };
2595
2596 /*
2597 * Create a packet of type SOCK_PACKET.
2598 */
2599
2600 static int packet_create(struct net *net, struct socket *sock, int protocol,
2601 int kern)
2602 {
2603 struct sock *sk;
2604 struct packet_sock *po;
2605 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2606 int err;
2607
2608 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2609 return -EPERM;
2610 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2611 sock->type != SOCK_PACKET)
2612 return -ESOCKTNOSUPPORT;
2613
2614 sock->state = SS_UNCONNECTED;
2615
2616 err = -ENOBUFS;
2617 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2618 if (sk == NULL)
2619 goto out;
2620
2621 sock->ops = &packet_ops;
2622 if (sock->type == SOCK_PACKET)
2623 sock->ops = &packet_ops_spkt;
2624
2625 sock_init_data(sock, sk);
2626
2627 po = pkt_sk(sk);
2628 sk->sk_family = PF_PACKET;
2629 po->num = proto;
2630
2631 packet_cached_dev_reset(po);
2632
2633 sk->sk_destruct = packet_sock_destruct;
2634 sk_refcnt_debug_inc(sk);
2635
2636 /*
2637 * Attach a protocol block
2638 */
2639
2640 spin_lock_init(&po->bind_lock);
2641 mutex_init(&po->pg_vec_lock);
2642 po->prot_hook.func = packet_rcv;
2643
2644 if (sock->type == SOCK_PACKET)
2645 po->prot_hook.func = packet_rcv_spkt;
2646
2647 po->prot_hook.af_packet_priv = sk;
2648
2649 if (proto) {
2650 po->prot_hook.type = proto;
2651 register_prot_hook(sk);
2652 }
2653
2654 mutex_lock(&net->packet.sklist_lock);
2655 sk_add_node_rcu(sk, &net->packet.sklist);
2656 mutex_unlock(&net->packet.sklist_lock);
2657
2658 preempt_disable();
2659 sock_prot_inuse_add(net, &packet_proto, 1);
2660 preempt_enable();
2661
2662 return 0;
2663 out:
2664 return err;
2665 }
2666
2667 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2668 {
2669 struct sock_exterr_skb *serr;
2670 struct sk_buff *skb, *skb2;
2671 int copied, err;
2672
2673 err = -EAGAIN;
2674 skb = skb_dequeue(&sk->sk_error_queue);
2675 if (skb == NULL)
2676 goto out;
2677
2678 copied = skb->len;
2679 if (copied > len) {
2680 msg->msg_flags |= MSG_TRUNC;
2681 copied = len;
2682 }
2683 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2684 if (err)
2685 goto out_free_skb;
2686
2687 sock_recv_timestamp(msg, sk, skb);
2688
2689 serr = SKB_EXT_ERR(skb);
2690 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2691 sizeof(serr->ee), &serr->ee);
2692
2693 msg->msg_flags |= MSG_ERRQUEUE;
2694 err = copied;
2695
2696 /* Reset and regenerate socket error */
2697 spin_lock_bh(&sk->sk_error_queue.lock);
2698 sk->sk_err = 0;
2699 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2700 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2701 spin_unlock_bh(&sk->sk_error_queue.lock);
2702 sk->sk_error_report(sk);
2703 } else
2704 spin_unlock_bh(&sk->sk_error_queue.lock);
2705
2706 out_free_skb:
2707 kfree_skb(skb);
2708 out:
2709 return err;
2710 }
2711
2712 /*
2713 * Pull a packet from our receive queue and hand it to the user.
2714 * If necessary we block.
2715 */
2716
2717 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2718 struct msghdr *msg, size_t len, int flags)
2719 {
2720 struct sock *sk = sock->sk;
2721 struct sk_buff *skb;
2722 int copied, err;
2723 int vnet_hdr_len = 0;
2724
2725 err = -EINVAL;
2726 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2727 goto out;
2728
2729 #if 0
2730 /* What error should we return now? EUNATTACH? */
2731 if (pkt_sk(sk)->ifindex < 0)
2732 return -ENODEV;
2733 #endif
2734
2735 if (flags & MSG_ERRQUEUE) {
2736 err = packet_recv_error(sk, msg, len);
2737 goto out;
2738 }
2739
2740 /*
2741 * Call the generic datagram receiver. This handles all sorts
2742 * of horrible races and re-entrancy so we can forget about it
2743 * in the protocol layers.
2744 *
2745 * Now it will return ENETDOWN, if device have just gone down,
2746 * but then it will block.
2747 */
2748
2749 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2750
2751 /*
2752 * An error occurred so return it. Because skb_recv_datagram()
2753 * handles the blocking we don't see and worry about blocking
2754 * retries.
2755 */
2756
2757 if (skb == NULL)
2758 goto out;
2759
2760 if (pkt_sk(sk)->has_vnet_hdr) {
2761 struct virtio_net_hdr vnet_hdr = { 0 };
2762
2763 err = -EINVAL;
2764 vnet_hdr_len = sizeof(vnet_hdr);
2765 if (len < vnet_hdr_len)
2766 goto out_free;
2767
2768 len -= vnet_hdr_len;
2769
2770 if (skb_is_gso(skb)) {
2771 struct skb_shared_info *sinfo = skb_shinfo(skb);
2772
2773 /* This is a hint as to how much should be linear. */
2774 vnet_hdr.hdr_len = skb_headlen(skb);
2775 vnet_hdr.gso_size = sinfo->gso_size;
2776 if (sinfo->gso_type & SKB_GSO_TCPV4)
2777 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2778 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2779 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2780 else if (sinfo->gso_type & SKB_GSO_UDP)
2781 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2782 else if (sinfo->gso_type & SKB_GSO_FCOE)
2783 goto out_free;
2784 else
2785 BUG();
2786 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2787 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2788 } else
2789 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2790
2791 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2792 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2793 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2794 vnet_hdr.csum_offset = skb->csum_offset;
2795 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2796 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2797 } /* else everything is zero */
2798
2799 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2800 vnet_hdr_len);
2801 if (err < 0)
2802 goto out_free;
2803 }
2804
2805 /* You lose any data beyond the buffer you gave. If it worries
2806 * a user program they can ask the device for its MTU
2807 * anyway.
2808 */
2809 copied = skb->len;
2810 if (copied > len) {
2811 copied = len;
2812 msg->msg_flags |= MSG_TRUNC;
2813 }
2814
2815 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2816 if (err)
2817 goto out_free;
2818
2819 sock_recv_ts_and_drops(msg, sk, skb);
2820
2821 if (msg->msg_name) {
2822 /* If the address length field is there to be filled
2823 * in, we fill it in now.
2824 */
2825 if (sock->type == SOCK_PACKET) {
2826 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2827 } else {
2828 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2829 msg->msg_namelen = sll->sll_halen +
2830 offsetof(struct sockaddr_ll, sll_addr);
2831 }
2832 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2833 msg->msg_namelen);
2834 }
2835
2836 if (pkt_sk(sk)->auxdata) {
2837 struct tpacket_auxdata aux;
2838
2839 aux.tp_status = TP_STATUS_USER;
2840 if (skb->ip_summed == CHECKSUM_PARTIAL)
2841 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2842 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2843 aux.tp_snaplen = skb->len;
2844 aux.tp_mac = 0;
2845 aux.tp_net = skb_network_offset(skb);
2846 if (vlan_tx_tag_present(skb)) {
2847 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2848 aux.tp_status |= TP_STATUS_VLAN_VALID;
2849 } else {
2850 aux.tp_vlan_tci = 0;
2851 }
2852 aux.tp_padding = 0;
2853 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2854 }
2855
2856 /*
2857 * Free or return the buffer as appropriate. Again this
2858 * hides all the races and re-entrancy issues from us.
2859 */
2860 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2861
2862 out_free:
2863 skb_free_datagram(sk, skb);
2864 out:
2865 return err;
2866 }
2867
2868 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2869 int *uaddr_len, int peer)
2870 {
2871 struct net_device *dev;
2872 struct sock *sk = sock->sk;
2873
2874 if (peer)
2875 return -EOPNOTSUPP;
2876
2877 uaddr->sa_family = AF_PACKET;
2878 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2879 rcu_read_lock();
2880 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2881 if (dev)
2882 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2883 rcu_read_unlock();
2884 *uaddr_len = sizeof(*uaddr);
2885
2886 return 0;
2887 }
2888
2889 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2890 int *uaddr_len, int peer)
2891 {
2892 struct net_device *dev;
2893 struct sock *sk = sock->sk;
2894 struct packet_sock *po = pkt_sk(sk);
2895 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2896
2897 if (peer)
2898 return -EOPNOTSUPP;
2899
2900 sll->sll_family = AF_PACKET;
2901 sll->sll_ifindex = po->ifindex;
2902 sll->sll_protocol = po->num;
2903 sll->sll_pkttype = 0;
2904 rcu_read_lock();
2905 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2906 if (dev) {
2907 sll->sll_hatype = dev->type;
2908 sll->sll_halen = dev->addr_len;
2909 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2910 } else {
2911 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2912 sll->sll_halen = 0;
2913 }
2914 rcu_read_unlock();
2915 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2916
2917 return 0;
2918 }
2919
2920 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2921 int what)
2922 {
2923 switch (i->type) {
2924 case PACKET_MR_MULTICAST:
2925 if (i->alen != dev->addr_len)
2926 return -EINVAL;
2927 if (what > 0)
2928 return dev_mc_add(dev, i->addr);
2929 else
2930 return dev_mc_del(dev, i->addr);
2931 break;
2932 case PACKET_MR_PROMISC:
2933 return dev_set_promiscuity(dev, what);
2934 break;
2935 case PACKET_MR_ALLMULTI:
2936 return dev_set_allmulti(dev, what);
2937 break;
2938 case PACKET_MR_UNICAST:
2939 if (i->alen != dev->addr_len)
2940 return -EINVAL;
2941 if (what > 0)
2942 return dev_uc_add(dev, i->addr);
2943 else
2944 return dev_uc_del(dev, i->addr);
2945 break;
2946 default:
2947 break;
2948 }
2949 return 0;
2950 }
2951
2952 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2953 {
2954 for ( ; i; i = i->next) {
2955 if (i->ifindex == dev->ifindex)
2956 packet_dev_mc(dev, i, what);
2957 }
2958 }
2959
2960 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2961 {
2962 struct packet_sock *po = pkt_sk(sk);
2963 struct packet_mclist *ml, *i;
2964 struct net_device *dev;
2965 int err;
2966
2967 rtnl_lock();
2968
2969 err = -ENODEV;
2970 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2971 if (!dev)
2972 goto done;
2973
2974 err = -EINVAL;
2975 if (mreq->mr_alen > dev->addr_len)
2976 goto done;
2977
2978 err = -ENOBUFS;
2979 i = kmalloc(sizeof(*i), GFP_KERNEL);
2980 if (i == NULL)
2981 goto done;
2982
2983 err = 0;
2984 for (ml = po->mclist; ml; ml = ml->next) {
2985 if (ml->ifindex == mreq->mr_ifindex &&
2986 ml->type == mreq->mr_type &&
2987 ml->alen == mreq->mr_alen &&
2988 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2989 ml->count++;
2990 /* Free the new element ... */
2991 kfree(i);
2992 goto done;
2993 }
2994 }
2995
2996 i->type = mreq->mr_type;
2997 i->ifindex = mreq->mr_ifindex;
2998 i->alen = mreq->mr_alen;
2999 memcpy(i->addr, mreq->mr_address, i->alen);
3000 i->count = 1;
3001 i->next = po->mclist;
3002 po->mclist = i;
3003 err = packet_dev_mc(dev, i, 1);
3004 if (err) {
3005 po->mclist = i->next;
3006 kfree(i);
3007 }
3008
3009 done:
3010 rtnl_unlock();
3011 return err;
3012 }
3013
3014 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3015 {
3016 struct packet_mclist *ml, **mlp;
3017
3018 rtnl_lock();
3019
3020 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3021 if (ml->ifindex == mreq->mr_ifindex &&
3022 ml->type == mreq->mr_type &&
3023 ml->alen == mreq->mr_alen &&
3024 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3025 if (--ml->count == 0) {
3026 struct net_device *dev;
3027 *mlp = ml->next;
3028 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3029 if (dev)
3030 packet_dev_mc(dev, ml, -1);
3031 kfree(ml);
3032 }
3033 rtnl_unlock();
3034 return 0;
3035 }
3036 }
3037 rtnl_unlock();
3038 return -EADDRNOTAVAIL;
3039 }
3040
3041 static void packet_flush_mclist(struct sock *sk)
3042 {
3043 struct packet_sock *po = pkt_sk(sk);
3044 struct packet_mclist *ml;
3045
3046 if (!po->mclist)
3047 return;
3048
3049 rtnl_lock();
3050 while ((ml = po->mclist) != NULL) {
3051 struct net_device *dev;
3052
3053 po->mclist = ml->next;
3054 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3055 if (dev != NULL)
3056 packet_dev_mc(dev, ml, -1);
3057 kfree(ml);
3058 }
3059 rtnl_unlock();
3060 }
3061
3062 static int
3063 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3064 {
3065 struct sock *sk = sock->sk;
3066 struct packet_sock *po = pkt_sk(sk);
3067 int ret;
3068
3069 if (level != SOL_PACKET)
3070 return -ENOPROTOOPT;
3071
3072 switch (optname) {
3073 case PACKET_ADD_MEMBERSHIP:
3074 case PACKET_DROP_MEMBERSHIP:
3075 {
3076 struct packet_mreq_max mreq;
3077 int len = optlen;
3078 memset(&mreq, 0, sizeof(mreq));
3079 if (len < sizeof(struct packet_mreq))
3080 return -EINVAL;
3081 if (len > sizeof(mreq))
3082 len = sizeof(mreq);
3083 if (copy_from_user(&mreq, optval, len))
3084 return -EFAULT;
3085 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3086 return -EINVAL;
3087 if (optname == PACKET_ADD_MEMBERSHIP)
3088 ret = packet_mc_add(sk, &mreq);
3089 else
3090 ret = packet_mc_drop(sk, &mreq);
3091 return ret;
3092 }
3093
3094 case PACKET_RX_RING:
3095 case PACKET_TX_RING:
3096 {
3097 union tpacket_req_u req_u;
3098 int len;
3099
3100 switch (po->tp_version) {
3101 case TPACKET_V1:
3102 case TPACKET_V2:
3103 len = sizeof(req_u.req);
3104 break;
3105 case TPACKET_V3:
3106 default:
3107 len = sizeof(req_u.req3);
3108 break;
3109 }
3110 if (optlen < len)
3111 return -EINVAL;
3112 if (pkt_sk(sk)->has_vnet_hdr)
3113 return -EINVAL;
3114 if (copy_from_user(&req_u.req, optval, len))
3115 return -EFAULT;
3116 return packet_set_ring(sk, &req_u, 0,
3117 optname == PACKET_TX_RING);
3118 }
3119 case PACKET_COPY_THRESH:
3120 {
3121 int val;
3122
3123 if (optlen != sizeof(val))
3124 return -EINVAL;
3125 if (copy_from_user(&val, optval, sizeof(val)))
3126 return -EFAULT;
3127
3128 pkt_sk(sk)->copy_thresh = val;
3129 return 0;
3130 }
3131 case PACKET_VERSION:
3132 {
3133 int val;
3134
3135 if (optlen != sizeof(val))
3136 return -EINVAL;
3137 if (copy_from_user(&val, optval, sizeof(val)))
3138 return -EFAULT;
3139 switch (val) {
3140 case TPACKET_V1:
3141 case TPACKET_V2:
3142 case TPACKET_V3:
3143 break;
3144 default:
3145 return -EINVAL;
3146 }
3147 lock_sock(sk);
3148 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3149 ret = -EBUSY;
3150 } else {
3151 po->tp_version = val;
3152 ret = 0;
3153 }
3154 release_sock(sk);
3155 return ret;
3156 }
3157 case PACKET_RESERVE:
3158 {
3159 unsigned int val;
3160
3161 if (optlen != sizeof(val))
3162 return -EINVAL;
3163 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3164 return -EBUSY;
3165 if (copy_from_user(&val, optval, sizeof(val)))
3166 return -EFAULT;
3167 if (val > INT_MAX)
3168 return -EINVAL;
3169 po->tp_reserve = val;
3170 return 0;
3171 }
3172 case PACKET_LOSS:
3173 {
3174 unsigned int val;
3175
3176 if (optlen != sizeof(val))
3177 return -EINVAL;
3178 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3179 return -EBUSY;
3180 if (copy_from_user(&val, optval, sizeof(val)))
3181 return -EFAULT;
3182 po->tp_loss = !!val;
3183 return 0;
3184 }
3185 case PACKET_AUXDATA:
3186 {
3187 int val;
3188
3189 if (optlen < sizeof(val))
3190 return -EINVAL;
3191 if (copy_from_user(&val, optval, sizeof(val)))
3192 return -EFAULT;
3193
3194 po->auxdata = !!val;
3195 return 0;
3196 }
3197 case PACKET_ORIGDEV:
3198 {
3199 int val;
3200
3201 if (optlen < sizeof(val))
3202 return -EINVAL;
3203 if (copy_from_user(&val, optval, sizeof(val)))
3204 return -EFAULT;
3205
3206 po->origdev = !!val;
3207 return 0;
3208 }
3209 case PACKET_VNET_HDR:
3210 {
3211 int val;
3212
3213 if (sock->type != SOCK_RAW)
3214 return -EINVAL;
3215 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3216 return -EBUSY;
3217 if (optlen < sizeof(val))
3218 return -EINVAL;
3219 if (copy_from_user(&val, optval, sizeof(val)))
3220 return -EFAULT;
3221
3222 po->has_vnet_hdr = !!val;
3223 return 0;
3224 }
3225 case PACKET_TIMESTAMP:
3226 {
3227 int val;
3228
3229 if (optlen != sizeof(val))
3230 return -EINVAL;
3231 if (copy_from_user(&val, optval, sizeof(val)))
3232 return -EFAULT;
3233
3234 po->tp_tstamp = val;
3235 return 0;
3236 }
3237 case PACKET_FANOUT:
3238 {
3239 int val;
3240
3241 if (optlen != sizeof(val))
3242 return -EINVAL;
3243 if (copy_from_user(&val, optval, sizeof(val)))
3244 return -EFAULT;
3245
3246 return fanout_add(sk, val & 0xffff, val >> 16);
3247 }
3248 case PACKET_TX_HAS_OFF:
3249 {
3250 unsigned int val;
3251
3252 if (optlen != sizeof(val))
3253 return -EINVAL;
3254 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3255 return -EBUSY;
3256 if (copy_from_user(&val, optval, sizeof(val)))
3257 return -EFAULT;
3258 po->tp_tx_has_off = !!val;
3259 return 0;
3260 }
3261 default:
3262 return -ENOPROTOOPT;
3263 }
3264 }
3265
3266 static int packet_getsockopt(struct socket *sock, int level, int optname,
3267 char __user *optval, int __user *optlen)
3268 {
3269 int len;
3270 int val, lv = sizeof(val);
3271 struct sock *sk = sock->sk;
3272 struct packet_sock *po = pkt_sk(sk);
3273 void *data = &val;
3274 union tpacket_stats_u st;
3275
3276 if (level != SOL_PACKET)
3277 return -ENOPROTOOPT;
3278
3279 if (get_user(len, optlen))
3280 return -EFAULT;
3281
3282 if (len < 0)
3283 return -EINVAL;
3284
3285 switch (optname) {
3286 case PACKET_STATISTICS:
3287 spin_lock_bh(&sk->sk_receive_queue.lock);
3288 memcpy(&st, &po->stats, sizeof(st));
3289 memset(&po->stats, 0, sizeof(po->stats));
3290 spin_unlock_bh(&sk->sk_receive_queue.lock);
3291
3292 if (po->tp_version == TPACKET_V3) {
3293 lv = sizeof(struct tpacket_stats_v3);
3294 st.stats3.tp_packets += st.stats3.tp_drops;
3295 data = &st.stats3;
3296 } else {
3297 lv = sizeof(struct tpacket_stats);
3298 st.stats1.tp_packets += st.stats1.tp_drops;
3299 data = &st.stats1;
3300 }
3301
3302 break;
3303 case PACKET_AUXDATA:
3304 val = po->auxdata;
3305 break;
3306 case PACKET_ORIGDEV:
3307 val = po->origdev;
3308 break;
3309 case PACKET_VNET_HDR:
3310 val = po->has_vnet_hdr;
3311 break;
3312 case PACKET_VERSION:
3313 val = po->tp_version;
3314 break;
3315 case PACKET_HDRLEN:
3316 if (len > sizeof(int))
3317 len = sizeof(int);
3318 if (copy_from_user(&val, optval, len))
3319 return -EFAULT;
3320 switch (val) {
3321 case TPACKET_V1:
3322 val = sizeof(struct tpacket_hdr);
3323 break;
3324 case TPACKET_V2:
3325 val = sizeof(struct tpacket2_hdr);
3326 break;
3327 case TPACKET_V3:
3328 val = sizeof(struct tpacket3_hdr);
3329 break;
3330 default:
3331 return -EINVAL;
3332 }
3333 break;
3334 case PACKET_RESERVE:
3335 val = po->tp_reserve;
3336 break;
3337 case PACKET_LOSS:
3338 val = po->tp_loss;
3339 break;
3340 case PACKET_TIMESTAMP:
3341 val = po->tp_tstamp;
3342 break;
3343 case PACKET_FANOUT:
3344 val = (po->fanout ?
3345 ((u32)po->fanout->id |
3346 ((u32)po->fanout->type << 16) |
3347 ((u32)po->fanout->flags << 24)) :
3348 0);
3349 break;
3350 case PACKET_TX_HAS_OFF:
3351 val = po->tp_tx_has_off;
3352 break;
3353 default:
3354 return -ENOPROTOOPT;
3355 }
3356
3357 if (len > lv)
3358 len = lv;
3359 if (put_user(len, optlen))
3360 return -EFAULT;
3361 if (copy_to_user(optval, data, len))
3362 return -EFAULT;
3363 return 0;
3364 }
3365
3366
3367 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3368 {
3369 struct sock *sk;
3370 struct net_device *dev = data;
3371 struct net *net = dev_net(dev);
3372
3373 rcu_read_lock();
3374 sk_for_each_rcu(sk, &net->packet.sklist) {
3375 struct packet_sock *po = pkt_sk(sk);
3376
3377 switch (msg) {
3378 case NETDEV_UNREGISTER:
3379 if (po->mclist)
3380 packet_dev_mclist(dev, po->mclist, -1);
3381 /* fallthrough */
3382
3383 case NETDEV_DOWN:
3384 if (dev->ifindex == po->ifindex) {
3385 spin_lock(&po->bind_lock);
3386 if (po->running) {
3387 __unregister_prot_hook(sk, false);
3388 sk->sk_err = ENETDOWN;
3389 if (!sock_flag(sk, SOCK_DEAD))
3390 sk->sk_error_report(sk);
3391 }
3392 if (msg == NETDEV_UNREGISTER) {
3393 packet_cached_dev_reset(po);
3394 po->ifindex = -1;
3395 if (po->prot_hook.dev)
3396 dev_put(po->prot_hook.dev);
3397 po->prot_hook.dev = NULL;
3398 }
3399 spin_unlock(&po->bind_lock);
3400 }
3401 break;
3402 case NETDEV_UP:
3403 if (dev->ifindex == po->ifindex) {
3404 spin_lock(&po->bind_lock);
3405 if (po->num)
3406 register_prot_hook(sk);
3407 spin_unlock(&po->bind_lock);
3408 }
3409 break;
3410 }
3411 }
3412 rcu_read_unlock();
3413 return NOTIFY_DONE;
3414 }
3415
3416
3417 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3418 unsigned long arg)
3419 {
3420 struct sock *sk = sock->sk;
3421
3422 switch (cmd) {
3423 case SIOCOUTQ:
3424 {
3425 int amount = sk_wmem_alloc_get(sk);
3426
3427 return put_user(amount, (int __user *)arg);
3428 }
3429 case SIOCINQ:
3430 {
3431 struct sk_buff *skb;
3432 int amount = 0;
3433
3434 spin_lock_bh(&sk->sk_receive_queue.lock);
3435 skb = skb_peek(&sk->sk_receive_queue);
3436 if (skb)
3437 amount = skb->len;
3438 spin_unlock_bh(&sk->sk_receive_queue.lock);
3439 return put_user(amount, (int __user *)arg);
3440 }
3441 case SIOCGSTAMP:
3442 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3443 case SIOCGSTAMPNS:
3444 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3445
3446 #ifdef CONFIG_INET
3447 case SIOCADDRT:
3448 case SIOCDELRT:
3449 case SIOCDARP:
3450 case SIOCGARP:
3451 case SIOCSARP:
3452 case SIOCGIFADDR:
3453 case SIOCSIFADDR:
3454 case SIOCGIFBRDADDR:
3455 case SIOCSIFBRDADDR:
3456 case SIOCGIFNETMASK:
3457 case SIOCSIFNETMASK:
3458 case SIOCGIFDSTADDR:
3459 case SIOCSIFDSTADDR:
3460 case SIOCSIFFLAGS:
3461 return inet_dgram_ops.ioctl(sock, cmd, arg);
3462 #endif
3463
3464 default:
3465 return -ENOIOCTLCMD;
3466 }
3467 return 0;
3468 }
3469
3470 static unsigned int packet_poll(struct file *file, struct socket *sock,
3471 poll_table *wait)
3472 {
3473 struct sock *sk = sock->sk;
3474 struct packet_sock *po = pkt_sk(sk);
3475 unsigned int mask = datagram_poll(file, sock, wait);
3476
3477 spin_lock_bh(&sk->sk_receive_queue.lock);
3478 if (po->rx_ring.pg_vec) {
3479 if (!packet_previous_rx_frame(po, &po->rx_ring,
3480 TP_STATUS_KERNEL))
3481 mask |= POLLIN | POLLRDNORM;
3482 }
3483 spin_unlock_bh(&sk->sk_receive_queue.lock);
3484 spin_lock_bh(&sk->sk_write_queue.lock);
3485 if (po->tx_ring.pg_vec) {
3486 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3487 mask |= POLLOUT | POLLWRNORM;
3488 }
3489 spin_unlock_bh(&sk->sk_write_queue.lock);
3490 return mask;
3491 }
3492
3493
3494 /* Dirty? Well, I still did not learn better way to account
3495 * for user mmaps.
3496 */
3497
3498 static void packet_mm_open(struct vm_area_struct *vma)
3499 {
3500 struct file *file = vma->vm_file;
3501 struct socket *sock = file->private_data;
3502 struct sock *sk = sock->sk;
3503
3504 if (sk)
3505 atomic_inc(&pkt_sk(sk)->mapped);
3506 }
3507
3508 static void packet_mm_close(struct vm_area_struct *vma)
3509 {
3510 struct file *file = vma->vm_file;
3511 struct socket *sock = file->private_data;
3512 struct sock *sk = sock->sk;
3513
3514 if (sk)
3515 atomic_dec(&pkt_sk(sk)->mapped);
3516 }
3517
3518 static const struct vm_operations_struct packet_mmap_ops = {
3519 .open = packet_mm_open,
3520 .close = packet_mm_close,
3521 };
3522
3523 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3524 unsigned int len)
3525 {
3526 int i;
3527
3528 for (i = 0; i < len; i++) {
3529 if (likely(pg_vec[i].buffer)) {
3530 if (is_vmalloc_addr(pg_vec[i].buffer))
3531 vfree(pg_vec[i].buffer);
3532 else
3533 free_pages((unsigned long)pg_vec[i].buffer,
3534 order);
3535 pg_vec[i].buffer = NULL;
3536 }
3537 }
3538 kfree(pg_vec);
3539 }
3540
3541 static char *alloc_one_pg_vec_page(unsigned long order)
3542 {
3543 char *buffer = NULL;
3544 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3545 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3546
3547 buffer = (char *) __get_free_pages(gfp_flags, order);
3548
3549 if (buffer)
3550 return buffer;
3551
3552 /*
3553 * __get_free_pages failed, fall back to vmalloc
3554 */
3555 buffer = vzalloc((1 << order) * PAGE_SIZE);
3556
3557 if (buffer)
3558 return buffer;
3559
3560 /*
3561 * vmalloc failed, lets dig into swap here
3562 */
3563 gfp_flags &= ~__GFP_NORETRY;
3564 buffer = (char *)__get_free_pages(gfp_flags, order);
3565 if (buffer)
3566 return buffer;
3567
3568 /*
3569 * complete and utter failure
3570 */
3571 return NULL;
3572 }
3573
3574 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3575 {
3576 unsigned int block_nr = req->tp_block_nr;
3577 struct pgv *pg_vec;
3578 int i;
3579
3580 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3581 if (unlikely(!pg_vec))
3582 goto out;
3583
3584 for (i = 0; i < block_nr; i++) {
3585 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3586 if (unlikely(!pg_vec[i].buffer))
3587 goto out_free_pgvec;
3588 }
3589
3590 out:
3591 return pg_vec;
3592
3593 out_free_pgvec:
3594 free_pg_vec(pg_vec, order, block_nr);
3595 pg_vec = NULL;
3596 goto out;
3597 }
3598
3599 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3600 int closing, int tx_ring)
3601 {
3602 struct pgv *pg_vec = NULL;
3603 struct packet_sock *po = pkt_sk(sk);
3604 int was_running, order = 0;
3605 struct packet_ring_buffer *rb;
3606 struct sk_buff_head *rb_queue;
3607 __be16 num;
3608 int err = -EINVAL;
3609 /* Added to avoid minimal code churn */
3610 struct tpacket_req *req = &req_u->req;
3611
3612 lock_sock(sk);
3613 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3614 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3615 WARN(1, "Tx-ring is not supported.\n");
3616 goto out;
3617 }
3618
3619 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3620 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3621
3622 err = -EBUSY;
3623 if (!closing) {
3624 if (atomic_read(&po->mapped))
3625 goto out;
3626 if (atomic_read(&rb->pending))
3627 goto out;
3628 }
3629
3630 if (req->tp_block_nr) {
3631 /* Sanity tests and some calculations */
3632 err = -EBUSY;
3633 if (unlikely(rb->pg_vec))
3634 goto out;
3635
3636 switch (po->tp_version) {
3637 case TPACKET_V1:
3638 po->tp_hdrlen = TPACKET_HDRLEN;
3639 break;
3640 case TPACKET_V2:
3641 po->tp_hdrlen = TPACKET2_HDRLEN;
3642 break;
3643 case TPACKET_V3:
3644 po->tp_hdrlen = TPACKET3_HDRLEN;
3645 break;
3646 }
3647
3648 err = -EINVAL;
3649 if (unlikely((int)req->tp_block_size <= 0))
3650 goto out;
3651 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3652 goto out;
3653 if (po->tp_version >= TPACKET_V3 &&
3654 (int)(req->tp_block_size -
3655 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3656 goto out;
3657 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3658 po->tp_reserve))
3659 goto out;
3660 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3661 goto out;
3662
3663 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3664 if (unlikely(rb->frames_per_block <= 0))
3665 goto out;
3666 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
3667 goto out;
3668 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3669 req->tp_frame_nr))
3670 goto out;
3671
3672 err = -ENOMEM;
3673 order = get_order(req->tp_block_size);
3674 pg_vec = alloc_pg_vec(req, order);
3675 if (unlikely(!pg_vec))
3676 goto out;
3677 switch (po->tp_version) {
3678 case TPACKET_V3:
3679 /* Transmit path is not supported. We checked
3680 * it above but just being paranoid
3681 */
3682 if (!tx_ring)
3683 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3684 break;
3685 default:
3686 break;
3687 }
3688 }
3689 /* Done */
3690 else {
3691 err = -EINVAL;
3692 if (unlikely(req->tp_frame_nr))
3693 goto out;
3694 }
3695
3696
3697 /* Detach socket from network */
3698 spin_lock(&po->bind_lock);
3699 was_running = po->running;
3700 num = po->num;
3701 if (was_running) {
3702 po->num = 0;
3703 __unregister_prot_hook(sk, false);
3704 }
3705 spin_unlock(&po->bind_lock);
3706
3707 synchronize_net();
3708
3709 err = -EBUSY;
3710 mutex_lock(&po->pg_vec_lock);
3711 if (closing || atomic_read(&po->mapped) == 0) {
3712 err = 0;
3713 spin_lock_bh(&rb_queue->lock);
3714 swap(rb->pg_vec, pg_vec);
3715 rb->frame_max = (req->tp_frame_nr - 1);
3716 rb->head = 0;
3717 rb->frame_size = req->tp_frame_size;
3718 spin_unlock_bh(&rb_queue->lock);
3719
3720 swap(rb->pg_vec_order, order);
3721 swap(rb->pg_vec_len, req->tp_block_nr);
3722
3723 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3724 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3725 tpacket_rcv : packet_rcv;
3726 skb_queue_purge(rb_queue);
3727 if (atomic_read(&po->mapped))
3728 pr_err("packet_mmap: vma is busy: %d\n",
3729 atomic_read(&po->mapped));
3730 }
3731 mutex_unlock(&po->pg_vec_lock);
3732
3733 spin_lock(&po->bind_lock);
3734 if (was_running) {
3735 po->num = num;
3736 register_prot_hook(sk);
3737 }
3738 spin_unlock(&po->bind_lock);
3739 if (closing && (po->tp_version > TPACKET_V2)) {
3740 /* Because we don't support block-based V3 on tx-ring */
3741 if (!tx_ring)
3742 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3743 }
3744
3745 if (pg_vec)
3746 free_pg_vec(pg_vec, order, req->tp_block_nr);
3747 out:
3748 release_sock(sk);
3749 return err;
3750 }
3751
3752 static int packet_mmap(struct file *file, struct socket *sock,
3753 struct vm_area_struct *vma)
3754 {
3755 struct sock *sk = sock->sk;
3756 struct packet_sock *po = pkt_sk(sk);
3757 unsigned long size, expected_size;
3758 struct packet_ring_buffer *rb;
3759 unsigned long start;
3760 int err = -EINVAL;
3761 int i;
3762
3763 if (vma->vm_pgoff)
3764 return -EINVAL;
3765
3766 mutex_lock(&po->pg_vec_lock);
3767
3768 expected_size = 0;
3769 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3770 if (rb->pg_vec) {
3771 expected_size += rb->pg_vec_len
3772 * rb->pg_vec_pages
3773 * PAGE_SIZE;
3774 }
3775 }
3776
3777 if (expected_size == 0)
3778 goto out;
3779
3780 size = vma->vm_end - vma->vm_start;
3781 if (size != expected_size)
3782 goto out;
3783
3784 start = vma->vm_start;
3785 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3786 if (rb->pg_vec == NULL)
3787 continue;
3788
3789 for (i = 0; i < rb->pg_vec_len; i++) {
3790 struct page *page;
3791 void *kaddr = rb->pg_vec[i].buffer;
3792 int pg_num;
3793
3794 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3795 page = pgv_to_page(kaddr);
3796 err = vm_insert_page(vma, start, page);
3797 if (unlikely(err))
3798 goto out;
3799 start += PAGE_SIZE;
3800 kaddr += PAGE_SIZE;
3801 }
3802 }
3803 }
3804
3805 atomic_inc(&po->mapped);
3806 vma->vm_ops = &packet_mmap_ops;
3807 err = 0;
3808
3809 out:
3810 mutex_unlock(&po->pg_vec_lock);
3811 return err;
3812 }
3813
3814 static const struct proto_ops packet_ops_spkt = {
3815 .family = PF_PACKET,
3816 .owner = THIS_MODULE,
3817 .release = packet_release,
3818 .bind = packet_bind_spkt,
3819 .connect = sock_no_connect,
3820 .socketpair = sock_no_socketpair,
3821 .accept = sock_no_accept,
3822 .getname = packet_getname_spkt,
3823 .poll = datagram_poll,
3824 .ioctl = packet_ioctl,
3825 .listen = sock_no_listen,
3826 .shutdown = sock_no_shutdown,
3827 .setsockopt = sock_no_setsockopt,
3828 .getsockopt = sock_no_getsockopt,
3829 .sendmsg = packet_sendmsg_spkt,
3830 .recvmsg = packet_recvmsg,
3831 .mmap = sock_no_mmap,
3832 .sendpage = sock_no_sendpage,
3833 };
3834
3835 static const struct proto_ops packet_ops = {
3836 .family = PF_PACKET,
3837 .owner = THIS_MODULE,
3838 .release = packet_release,
3839 .bind = packet_bind,
3840 .connect = sock_no_connect,
3841 .socketpair = sock_no_socketpair,
3842 .accept = sock_no_accept,
3843 .getname = packet_getname,
3844 .poll = packet_poll,
3845 .ioctl = packet_ioctl,
3846 .listen = sock_no_listen,
3847 .shutdown = sock_no_shutdown,
3848 .setsockopt = packet_setsockopt,
3849 .getsockopt = packet_getsockopt,
3850 .sendmsg = packet_sendmsg,
3851 .recvmsg = packet_recvmsg,
3852 .mmap = packet_mmap,
3853 .sendpage = sock_no_sendpage,
3854 };
3855
3856 static const struct net_proto_family packet_family_ops = {
3857 .family = PF_PACKET,
3858 .create = packet_create,
3859 .owner = THIS_MODULE,
3860 };
3861
3862 static struct notifier_block packet_netdev_notifier = {
3863 .notifier_call = packet_notifier,
3864 };
3865
3866 #ifdef CONFIG_PROC_FS
3867
3868 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3869 __acquires(RCU)
3870 {
3871 struct net *net = seq_file_net(seq);
3872
3873 rcu_read_lock();
3874 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3875 }
3876
3877 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3878 {
3879 struct net *net = seq_file_net(seq);
3880 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3881 }
3882
3883 static void packet_seq_stop(struct seq_file *seq, void *v)
3884 __releases(RCU)
3885 {
3886 rcu_read_unlock();
3887 }
3888
3889 static int packet_seq_show(struct seq_file *seq, void *v)
3890 {
3891 if (v == SEQ_START_TOKEN)
3892 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3893 else {
3894 struct sock *s = sk_entry(v);
3895 const struct packet_sock *po = pkt_sk(s);
3896
3897 seq_printf(seq,
3898 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3899 s,
3900 atomic_read(&s->sk_refcnt),
3901 s->sk_type,
3902 ntohs(po->num),
3903 po->ifindex,
3904 po->running,
3905 atomic_read(&s->sk_rmem_alloc),
3906 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3907 sock_i_ino(s));
3908 }
3909
3910 return 0;
3911 }
3912
3913 static const struct seq_operations packet_seq_ops = {
3914 .start = packet_seq_start,
3915 .next = packet_seq_next,
3916 .stop = packet_seq_stop,
3917 .show = packet_seq_show,
3918 };
3919
3920 static int packet_seq_open(struct inode *inode, struct file *file)
3921 {
3922 return seq_open_net(inode, file, &packet_seq_ops,
3923 sizeof(struct seq_net_private));
3924 }
3925
3926 static const struct file_operations packet_seq_fops = {
3927 .owner = THIS_MODULE,
3928 .open = packet_seq_open,
3929 .read = seq_read,
3930 .llseek = seq_lseek,
3931 .release = seq_release_net,
3932 };
3933
3934 #endif
3935
3936 static int __net_init packet_net_init(struct net *net)
3937 {
3938 mutex_init(&net->packet.sklist_lock);
3939 INIT_HLIST_HEAD(&net->packet.sklist);
3940
3941 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3942 return -ENOMEM;
3943
3944 return 0;
3945 }
3946
3947 static void __net_exit packet_net_exit(struct net *net)
3948 {
3949 remove_proc_entry("packet", net->proc_net);
3950 }
3951
3952 static struct pernet_operations packet_net_ops = {
3953 .init = packet_net_init,
3954 .exit = packet_net_exit,
3955 };
3956
3957
3958 static void __exit packet_exit(void)
3959 {
3960 unregister_netdevice_notifier(&packet_netdev_notifier);
3961 unregister_pernet_subsys(&packet_net_ops);
3962 sock_unregister(PF_PACKET);
3963 proto_unregister(&packet_proto);
3964 }
3965
3966 static int __init packet_init(void)
3967 {
3968 int rc = proto_register(&packet_proto, 0);
3969
3970 if (rc != 0)
3971 goto out;
3972
3973 sock_register(&packet_family_ops);
3974 register_pernet_subsys(&packet_net_ops);
3975 register_netdevice_notifier(&packet_netdev_notifier);
3976 out:
3977 return rc;
3978 }
3979
3980 module_init(packet_init);
3981 module_exit(packet_exit);
3982 MODULE_LICENSE("GPL");
3983 MODULE_ALIAS_NETPROTO(PF_PACKET);
kernel source for telekom puls (alcatel/mediatek)