projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:


0bbb3470fa7854aabd9c0b07499945127c019901
[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 if (f->num_members == 1)
1261 dev_add_pack(&f->prot_hook);
1262 spin_unlock(&f->lock);
1263 }
1264
1265 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1266 {
1267 struct packet_fanout *f = po->fanout;
1268 int i;
1269
1270 spin_lock(&f->lock);
1271 for (i = 0; i < f->num_members; i++) {
1272 if (f->arr[i] == sk)
1273 break;
1274 }
1275 BUG_ON(i >= f->num_members);
1276 f->arr[i] = f->arr[f->num_members - 1];
1277 f->num_members--;
1278 if (f->num_members == 0)
1279 __dev_remove_pack(&f->prot_hook);
1280 spin_unlock(&f->lock);
1281 }
1282
1283 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1284 {
1285 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1286 return true;
1287
1288 return false;
1289 }
1290
1291 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1292 {
1293 struct packet_sock *po = pkt_sk(sk);
1294 struct packet_fanout *f, *match;
1295 u8 type = type_flags & 0xff;
1296 u8 flags = type_flags >> 8;
1297 int err;
1298
1299 switch (type) {
1300 case PACKET_FANOUT_ROLLOVER:
1301 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1302 return -EINVAL;
1303 case PACKET_FANOUT_HASH:
1304 case PACKET_FANOUT_LB:
1305 case PACKET_FANOUT_CPU:
1306 break;
1307 default:
1308 return -EINVAL;
1309 }
1310
1311 mutex_lock(&fanout_mutex);
1312
1313 err = -EINVAL;
1314 if (!po->running)
1315 goto out;
1316
1317 err = -EALREADY;
1318 if (po->fanout)
1319 goto out;
1320
1321 match = NULL;
1322 list_for_each_entry(f, &fanout_list, list) {
1323 if (f->id == id &&
1324 read_pnet(&f->net) == sock_net(sk)) {
1325 match = f;
1326 break;
1327 }
1328 }
1329 err = -EINVAL;
1330 if (match && match->flags != flags)
1331 goto out;
1332 if (!match) {
1333 err = -ENOMEM;
1334 match = kzalloc(sizeof(*match), GFP_KERNEL);
1335 if (!match)
1336 goto out;
1337 write_pnet(&match->net, sock_net(sk));
1338 match->id = id;
1339 match->type = type;
1340 match->flags = flags;
1341 atomic_set(&match->rr_cur, 0);
1342 INIT_LIST_HEAD(&match->list);
1343 spin_lock_init(&match->lock);
1344 atomic_set(&match->sk_ref, 0);
1345 match->prot_hook.type = po->prot_hook.type;
1346 match->prot_hook.dev = po->prot_hook.dev;
1347 match->prot_hook.func = packet_rcv_fanout;
1348 match->prot_hook.af_packet_priv = match;
1349 match->prot_hook.id_match = match_fanout_group;
1350 list_add(&match->list, &fanout_list);
1351 }
1352 err = -EINVAL;
1353 if (match->type == type &&
1354 match->prot_hook.type == po->prot_hook.type &&
1355 match->prot_hook.dev == po->prot_hook.dev) {
1356 err = -ENOSPC;
1357 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1358 __dev_remove_pack(&po->prot_hook);
1359 po->fanout = match;
1360 atomic_inc(&match->sk_ref);
1361 __fanout_link(sk, po);
1362 err = 0;
1363 }
1364 }
1365 out:
1366 mutex_unlock(&fanout_mutex);
1367 return err;
1368 }
1369
1370 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1371 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1372 * It is the responsibility of the caller to call fanout_release_data() and
1373 * free the returned packet_fanout (after synchronize_net())
1374 */
1375 static struct packet_fanout *fanout_release(struct sock *sk)
1376 {
1377 struct packet_sock *po = pkt_sk(sk);
1378 struct packet_fanout *f;
1379
1380 mutex_lock(&fanout_mutex);
1381 f = po->fanout;
1382 if (f) {
1383 po->fanout = NULL;
1384
1385 if (atomic_dec_and_test(&f->sk_ref))
1386 list_del(&f->list);
1387 else
1388 f = NULL;
1389 }
1390 mutex_unlock(&fanout_mutex);
1391
1392 return f;
1393 }
1394
1395 static const struct proto_ops packet_ops;
1396
1397 static const struct proto_ops packet_ops_spkt;
1398
1399 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1400 struct packet_type *pt, struct net_device *orig_dev)
1401 {
1402 struct sock *sk;
1403 struct sockaddr_pkt *spkt;
1404
1405 /*
1406 * When we registered the protocol we saved the socket in the data
1407 * field for just this event.
1408 */
1409
1410 sk = pt->af_packet_priv;
1411
1412 /*
1413 * Yank back the headers [hope the device set this
1414 * right or kerboom...]
1415 *
1416 * Incoming packets have ll header pulled,
1417 * push it back.
1418 *
1419 * For outgoing ones skb->data == skb_mac_header(skb)
1420 * so that this procedure is noop.
1421 */
1422
1423 if (skb->pkt_type == PACKET_LOOPBACK)
1424 goto out;
1425
1426 if (!net_eq(dev_net(dev), sock_net(sk)))
1427 goto out;
1428
1429 skb = skb_share_check(skb, GFP_ATOMIC);
1430 if (skb == NULL)
1431 goto oom;
1432
1433 /* drop any routing info */
1434 skb_dst_drop(skb);
1435
1436 /* drop conntrack reference */
1437 nf_reset(skb);
1438
1439 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1440
1441 skb_push(skb, skb->data - skb_mac_header(skb));
1442
1443 /*
1444 * The SOCK_PACKET socket receives _all_ frames.
1445 */
1446
1447 spkt->spkt_family = dev->type;
1448 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1449 spkt->spkt_protocol = skb->protocol;
1450
1451 /*
1452 * Charge the memory to the socket. This is done specifically
1453 * to prevent sockets using all the memory up.
1454 */
1455
1456 if (sock_queue_rcv_skb(sk, skb) == 0)
1457 return 0;
1458
1459 out:
1460 kfree_skb(skb);
1461 oom:
1462 return 0;
1463 }
1464
1465
1466 /*
1467 * Output a raw packet to a device layer. This bypasses all the other
1468 * protocol layers and you must therefore supply it with a complete frame
1469 */
1470
1471 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1472 struct msghdr *msg, size_t len)
1473 {
1474 struct sock *sk = sock->sk;
1475 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1476 struct sk_buff *skb = NULL;
1477 struct net_device *dev;
1478 __be16 proto = 0;
1479 int err;
1480 int extra_len = 0;
1481
1482 /*
1483 * Get and verify the address.
1484 */
1485
1486 if (saddr) {
1487 if (msg->msg_namelen < sizeof(struct sockaddr))
1488 return -EINVAL;
1489 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1490 proto = saddr->spkt_protocol;
1491 } else
1492 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1493
1494 /*
1495 * Find the device first to size check it
1496 */
1497
1498 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1499 retry:
1500 rcu_read_lock();
1501 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1502 err = -ENODEV;
1503 if (dev == NULL)
1504 goto out_unlock;
1505
1506 err = -ENETDOWN;
1507 if (!(dev->flags & IFF_UP))
1508 goto out_unlock;
1509
1510 /*
1511 * You may not queue a frame bigger than the mtu. This is the lowest level
1512 * raw protocol and you must do your own fragmentation at this level.
1513 */
1514
1515 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1516 if (!netif_supports_nofcs(dev)) {
1517 err = -EPROTONOSUPPORT;
1518 goto out_unlock;
1519 }
1520 extra_len = 4; /* We're doing our own CRC */
1521 }
1522
1523 err = -EMSGSIZE;
1524 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1525 goto out_unlock;
1526
1527 if (!skb) {
1528 size_t reserved = LL_RESERVED_SPACE(dev);
1529 int tlen = dev->needed_tailroom;
1530 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1531
1532 rcu_read_unlock();
1533 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1534 if (skb == NULL)
1535 return -ENOBUFS;
1536 /* FIXME: Save some space for broken drivers that write a hard
1537 * header at transmission time by themselves. PPP is the notable
1538 * one here. This should really be fixed at the driver level.
1539 */
1540 skb_reserve(skb, reserved);
1541 skb_reset_network_header(skb);
1542
1543 /* Try to align data part correctly */
1544 if (hhlen) {
1545 skb->data -= hhlen;
1546 skb->tail -= hhlen;
1547 if (len < hhlen)
1548 skb_reset_network_header(skb);
1549 }
1550 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1551 if (err)
1552 goto out_free;
1553 goto retry;
1554 }
1555
1556 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1557 /* Earlier code assumed this would be a VLAN pkt,
1558 * double-check this now that we have the actual
1559 * packet in hand.
1560 */
1561 struct ethhdr *ehdr;
1562 skb_reset_mac_header(skb);
1563 ehdr = eth_hdr(skb);
1564 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1565 err = -EMSGSIZE;
1566 goto out_unlock;
1567 }
1568 }
1569
1570 skb->protocol = proto;
1571 skb->dev = dev;
1572 skb->priority = sk->sk_priority;
1573 skb->mark = sk->sk_mark;
1574
1575 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1576
1577 if (unlikely(extra_len == 4))
1578 skb->no_fcs = 1;
1579
1580 skb_probe_transport_header(skb, 0);
1581
1582 dev_queue_xmit(skb);
1583 rcu_read_unlock();
1584 return len;
1585
1586 out_unlock:
1587 rcu_read_unlock();
1588 out_free:
1589 kfree_skb(skb);
1590 return err;
1591 }
1592
1593 static unsigned int run_filter(const struct sk_buff *skb,
1594 const struct sock *sk,
1595 unsigned int res)
1596 {
1597 struct sk_filter *filter;
1598
1599 rcu_read_lock();
1600 filter = rcu_dereference(sk->sk_filter);
1601 if (filter != NULL)
1602 res = SK_RUN_FILTER(filter, skb);
1603 rcu_read_unlock();
1604
1605 return res;
1606 }
1607
1608 /*
1609 * This function makes lazy skb cloning in hope that most of packets
1610 * are discarded by BPF.
1611 *
1612 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1613 * and skb->cb are mangled. It works because (and until) packets
1614 * falling here are owned by current CPU. Output packets are cloned
1615 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1616 * sequencially, so that if we return skb to original state on exit,
1617 * we will not harm anyone.
1618 */
1619
1620 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1621 struct packet_type *pt, struct net_device *orig_dev)
1622 {
1623 struct sock *sk;
1624 struct sockaddr_ll *sll;
1625 struct packet_sock *po;
1626 u8 *skb_head = skb->data;
1627 int skb_len = skb->len;
1628 unsigned int snaplen, res;
1629
1630 if (skb->pkt_type == PACKET_LOOPBACK)
1631 goto drop;
1632
1633 sk = pt->af_packet_priv;
1634 po = pkt_sk(sk);
1635
1636 if (!net_eq(dev_net(dev), sock_net(sk)))
1637 goto drop;
1638
1639 skb->dev = dev;
1640
1641 if (dev->header_ops) {
1642 /* The device has an explicit notion of ll header,
1643 * exported to higher levels.
1644 *
1645 * Otherwise, the device hides details of its frame
1646 * structure, so that corresponding packet head is
1647 * never delivered to user.
1648 */
1649 if (sk->sk_type != SOCK_DGRAM)
1650 skb_push(skb, skb->data - skb_mac_header(skb));
1651 else if (skb->pkt_type == PACKET_OUTGOING) {
1652 /* Special case: outgoing packets have ll header at head */
1653 skb_pull(skb, skb_network_offset(skb));
1654 }
1655 }
1656
1657 snaplen = skb->len;
1658
1659 res = run_filter(skb, sk, snaplen);
1660 if (!res)
1661 goto drop_n_restore;
1662 if (snaplen > res)
1663 snaplen = res;
1664
1665 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1666 goto drop_n_acct;
1667
1668 if (skb_shared(skb)) {
1669 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1670 if (nskb == NULL)
1671 goto drop_n_acct;
1672
1673 if (skb_head != skb->data) {
1674 skb->data = skb_head;
1675 skb->len = skb_len;
1676 }
1677 consume_skb(skb);
1678 skb = nskb;
1679 }
1680
1681 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1682 sizeof(skb->cb));
1683
1684 sll = &PACKET_SKB_CB(skb)->sa.ll;
1685 sll->sll_family = AF_PACKET;
1686 sll->sll_hatype = dev->type;
1687 sll->sll_protocol = skb->protocol;
1688 sll->sll_pkttype = skb->pkt_type;
1689 if (unlikely(po->origdev))
1690 sll->sll_ifindex = orig_dev->ifindex;
1691 else
1692 sll->sll_ifindex = dev->ifindex;
1693
1694 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1695
1696 PACKET_SKB_CB(skb)->origlen = skb->len;
1697
1698 if (pskb_trim(skb, snaplen))
1699 goto drop_n_acct;
1700
1701 skb_set_owner_r(skb, sk);
1702 skb->dev = NULL;
1703 skb_dst_drop(skb);
1704
1705 /* drop conntrack reference */
1706 nf_reset(skb);
1707
1708 spin_lock(&sk->sk_receive_queue.lock);
1709 po->stats.stats1.tp_packets++;
1710 skb->dropcount = atomic_read(&sk->sk_drops);
1711 __skb_queue_tail(&sk->sk_receive_queue, skb);
1712 spin_unlock(&sk->sk_receive_queue.lock);
1713 sk->sk_data_ready(sk, skb->len);
1714 return 0;
1715
1716 drop_n_acct:
1717 spin_lock(&sk->sk_receive_queue.lock);
1718 po->stats.stats1.tp_drops++;
1719 atomic_inc(&sk->sk_drops);
1720 spin_unlock(&sk->sk_receive_queue.lock);
1721
1722 drop_n_restore:
1723 if (skb_head != skb->data && skb_shared(skb)) {
1724 skb->data = skb_head;
1725 skb->len = skb_len;
1726 }
1727 drop:
1728 consume_skb(skb);
1729 return 0;
1730 }
1731
1732 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1733 struct packet_type *pt, struct net_device *orig_dev)
1734 {
1735 struct sock *sk;
1736 struct packet_sock *po;
1737 struct sockaddr_ll *sll;
1738 union tpacket_uhdr h;
1739 u8 *skb_head = skb->data;
1740 int skb_len = skb->len;
1741 unsigned int snaplen, res;
1742 unsigned long status = TP_STATUS_USER;
1743 unsigned short macoff, netoff, hdrlen;
1744 struct sk_buff *copy_skb = NULL;
1745 struct timespec ts;
1746 __u32 ts_status;
1747
1748 if (skb->pkt_type == PACKET_LOOPBACK)
1749 goto drop;
1750
1751 sk = pt->af_packet_priv;
1752 po = pkt_sk(sk);
1753
1754 if (!net_eq(dev_net(dev), sock_net(sk)))
1755 goto drop;
1756
1757 if (dev->header_ops) {
1758 if (sk->sk_type != SOCK_DGRAM)
1759 skb_push(skb, skb->data - skb_mac_header(skb));
1760 else if (skb->pkt_type == PACKET_OUTGOING) {
1761 /* Special case: outgoing packets have ll header at head */
1762 skb_pull(skb, skb_network_offset(skb));
1763 }
1764 }
1765
1766 if (skb->ip_summed == CHECKSUM_PARTIAL)
1767 status |= TP_STATUS_CSUMNOTREADY;
1768
1769 snaplen = skb->len;
1770
1771 res = run_filter(skb, sk, snaplen);
1772 if (!res)
1773 goto drop_n_restore;
1774 if (snaplen > res)
1775 snaplen = res;
1776
1777 if (sk->sk_type == SOCK_DGRAM) {
1778 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1779 po->tp_reserve;
1780 } else {
1781 unsigned int maclen = skb_network_offset(skb);
1782 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1783 (maclen < 16 ? 16 : maclen)) +
1784 po->tp_reserve;
1785 macoff = netoff - maclen;
1786 }
1787 if (po->tp_version <= TPACKET_V2) {
1788 if (macoff + snaplen > po->rx_ring.frame_size) {
1789 if (po->copy_thresh &&
1790 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1791 if (skb_shared(skb)) {
1792 copy_skb = skb_clone(skb, GFP_ATOMIC);
1793 } else {
1794 copy_skb = skb_get(skb);
1795 skb_head = skb->data;
1796 }
1797 if (copy_skb)
1798 skb_set_owner_r(copy_skb, sk);
1799 }
1800 snaplen = po->rx_ring.frame_size - macoff;
1801 if ((int)snaplen < 0)
1802 snaplen = 0;
1803 }
1804 } else if (unlikely(macoff + snaplen >
1805 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1806 u32 nval;
1807
1808 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1809 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1810 snaplen, nval, macoff);
1811 snaplen = nval;
1812 if (unlikely((int)snaplen < 0)) {
1813 snaplen = 0;
1814 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1815 }
1816 }
1817 spin_lock(&sk->sk_receive_queue.lock);
1818 h.raw = packet_current_rx_frame(po, skb,
1819 TP_STATUS_KERNEL, (macoff+snaplen));
1820 if (!h.raw)
1821 goto ring_is_full;
1822 if (po->tp_version <= TPACKET_V2) {
1823 packet_increment_rx_head(po, &po->rx_ring);
1824 /*
1825 * LOSING will be reported till you read the stats,
1826 * because it's COR - Clear On Read.
1827 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1828 * at packet level.
1829 */
1830 if (po->stats.stats1.tp_drops)
1831 status |= TP_STATUS_LOSING;
1832 }
1833 po->stats.stats1.tp_packets++;
1834 if (copy_skb) {
1835 status |= TP_STATUS_COPY;
1836 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1837 }
1838 spin_unlock(&sk->sk_receive_queue.lock);
1839
1840 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1841
1842 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1843 getnstimeofday(&ts);
1844
1845 status |= ts_status;
1846
1847 switch (po->tp_version) {
1848 case TPACKET_V1:
1849 h.h1->tp_len = skb->len;
1850 h.h1->tp_snaplen = snaplen;
1851 h.h1->tp_mac = macoff;
1852 h.h1->tp_net = netoff;
1853 h.h1->tp_sec = ts.tv_sec;
1854 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1855 hdrlen = sizeof(*h.h1);
1856 break;
1857 case TPACKET_V2:
1858 h.h2->tp_len = skb->len;
1859 h.h2->tp_snaplen = snaplen;
1860 h.h2->tp_mac = macoff;
1861 h.h2->tp_net = netoff;
1862 h.h2->tp_sec = ts.tv_sec;
1863 h.h2->tp_nsec = ts.tv_nsec;
1864 if (vlan_tx_tag_present(skb)) {
1865 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1866 status |= TP_STATUS_VLAN_VALID;
1867 } else {
1868 h.h2->tp_vlan_tci = 0;
1869 }
1870 h.h2->tp_padding = 0;
1871 hdrlen = sizeof(*h.h2);
1872 break;
1873 case TPACKET_V3:
1874 /* tp_nxt_offset,vlan are already populated above.
1875 * So DONT clear those fields here
1876 */
1877 h.h3->tp_status |= status;
1878 h.h3->tp_len = skb->len;
1879 h.h3->tp_snaplen = snaplen;
1880 h.h3->tp_mac = macoff;
1881 h.h3->tp_net = netoff;
1882 h.h3->tp_sec = ts.tv_sec;
1883 h.h3->tp_nsec = ts.tv_nsec;
1884 hdrlen = sizeof(*h.h3);
1885 break;
1886 default:
1887 BUG();
1888 }
1889
1890 sll = h.raw + TPACKET_ALIGN(hdrlen);
1891 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1892 sll->sll_family = AF_PACKET;
1893 sll->sll_hatype = dev->type;
1894 sll->sll_protocol = skb->protocol;
1895 sll->sll_pkttype = skb->pkt_type;
1896 if (unlikely(po->origdev))
1897 sll->sll_ifindex = orig_dev->ifindex;
1898 else
1899 sll->sll_ifindex = dev->ifindex;
1900
1901 smp_mb();
1902 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1903 {
1904 u8 *start, *end;
1905
1906 if (po->tp_version <= TPACKET_V2) {
1907 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1908 + macoff + snaplen);
1909 for (start = h.raw; start < end; start += PAGE_SIZE)
1910 flush_dcache_page(pgv_to_page(start));
1911 }
1912 smp_wmb();
1913 }
1914 #endif
1915 if (po->tp_version <= TPACKET_V2)
1916 __packet_set_status(po, h.raw, status);
1917 else
1918 prb_clear_blk_fill_status(&po->rx_ring);
1919
1920 sk->sk_data_ready(sk, 0);
1921
1922 drop_n_restore:
1923 if (skb_head != skb->data && skb_shared(skb)) {
1924 skb->data = skb_head;
1925 skb->len = skb_len;
1926 }
1927 drop:
1928 kfree_skb(skb);
1929 return 0;
1930
1931 ring_is_full:
1932 po->stats.stats1.tp_drops++;
1933 spin_unlock(&sk->sk_receive_queue.lock);
1934
1935 sk->sk_data_ready(sk, 0);
1936 kfree_skb(copy_skb);
1937 goto drop_n_restore;
1938 }
1939
1940 static void tpacket_destruct_skb(struct sk_buff *skb)
1941 {
1942 struct packet_sock *po = pkt_sk(skb->sk);
1943 void *ph;
1944
1945 if (likely(po->tx_ring.pg_vec)) {
1946 __u32 ts;
1947
1948 ph = skb_shinfo(skb)->destructor_arg;
1949 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1950 atomic_dec(&po->tx_ring.pending);
1951
1952 ts = __packet_set_timestamp(po, ph, skb);
1953 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1954 }
1955
1956 sock_wfree(skb);
1957 }
1958
1959 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1960 void *frame, struct net_device *dev, int size_max,
1961 __be16 proto, unsigned char *addr, int hlen)
1962 {
1963 union tpacket_uhdr ph;
1964 int to_write, offset, len, tp_len, nr_frags, len_max;
1965 struct socket *sock = po->sk.sk_socket;
1966 struct page *page;
1967 void *data;
1968 int err;
1969
1970 ph.raw = frame;
1971
1972 skb->protocol = proto;
1973 skb->dev = dev;
1974 skb->priority = po->sk.sk_priority;
1975 skb->mark = po->sk.sk_mark;
1976 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1977 skb_shinfo(skb)->destructor_arg = ph.raw;
1978
1979 switch (po->tp_version) {
1980 case TPACKET_V2:
1981 tp_len = ph.h2->tp_len;
1982 break;
1983 default:
1984 tp_len = ph.h1->tp_len;
1985 break;
1986 }
1987 if (unlikely(tp_len > size_max)) {
1988 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1989 return -EMSGSIZE;
1990 }
1991
1992 skb_reserve(skb, hlen);
1993 skb_reset_network_header(skb);
1994 skb_probe_transport_header(skb, 0);
1995
1996 if (po->tp_tx_has_off) {
1997 int off_min, off_max, off;
1998 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1999 off_max = po->tx_ring.frame_size - tp_len;
2000 if (sock->type == SOCK_DGRAM) {
2001 switch (po->tp_version) {
2002 case TPACKET_V2:
2003 off = ph.h2->tp_net;
2004 break;
2005 default:
2006 off = ph.h1->tp_net;
2007 break;
2008 }
2009 } else {
2010 switch (po->tp_version) {
2011 case TPACKET_V2:
2012 off = ph.h2->tp_mac;
2013 break;
2014 default:
2015 off = ph.h1->tp_mac;
2016 break;
2017 }
2018 }
2019 if (unlikely((off < off_min) || (off_max < off)))
2020 return -EINVAL;
2021 data = ph.raw + off;
2022 } else {
2023 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2024 }
2025 to_write = tp_len;
2026
2027 if (sock->type == SOCK_DGRAM) {
2028 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2029 NULL, tp_len);
2030 if (unlikely(err < 0))
2031 return -EINVAL;
2032 } else if (dev->hard_header_len) {
2033 /* net device doesn't like empty head */
2034 if (unlikely(tp_len <= dev->hard_header_len)) {
2035 pr_err("packet size is too short (%d < %d)\n",
2036 tp_len, dev->hard_header_len);
2037 return -EINVAL;
2038 }
2039
2040 skb_push(skb, dev->hard_header_len);
2041 err = skb_store_bits(skb, 0, data,
2042 dev->hard_header_len);
2043 if (unlikely(err))
2044 return err;
2045
2046 data += dev->hard_header_len;
2047 to_write -= dev->hard_header_len;
2048 }
2049
2050 offset = offset_in_page(data);
2051 len_max = PAGE_SIZE - offset;
2052 len = ((to_write > len_max) ? len_max : to_write);
2053
2054 skb->data_len = to_write;
2055 skb->len += to_write;
2056 skb->truesize += to_write;
2057 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2058
2059 while (likely(to_write)) {
2060 nr_frags = skb_shinfo(skb)->nr_frags;
2061
2062 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2063 pr_err("Packet exceed the number of skb frags(%lu)\n",
2064 MAX_SKB_FRAGS);
2065 return -EFAULT;
2066 }
2067
2068 page = pgv_to_page(data);
2069 data += len;
2070 flush_dcache_page(page);
2071 get_page(page);
2072 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2073 to_write -= len;
2074 offset = 0;
2075 len_max = PAGE_SIZE;
2076 len = ((to_write > len_max) ? len_max : to_write);
2077 }
2078
2079 return tp_len;
2080 }
2081
2082 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2083 {
2084 struct sk_buff *skb;
2085 struct net_device *dev;
2086 __be16 proto;
2087 int err, reserve = 0;
2088 void *ph;
2089 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2090 int tp_len, size_max;
2091 unsigned char *addr;
2092 int len_sum = 0;
2093 int status = TP_STATUS_AVAILABLE;
2094 int hlen, tlen;
2095
2096 mutex_lock(&po->pg_vec_lock);
2097
2098 if (likely(saddr == NULL)) {
2099 dev = packet_cached_dev_get(po);
2100 proto = po->num;
2101 addr = NULL;
2102 } else {
2103 err = -EINVAL;
2104 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2105 goto out;
2106 if (msg->msg_namelen < (saddr->sll_halen
2107 + offsetof(struct sockaddr_ll,
2108 sll_addr)))
2109 goto out;
2110 proto = saddr->sll_protocol;
2111 addr = saddr->sll_addr;
2112 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2113 }
2114
2115 err = -ENXIO;
2116 if (unlikely(dev == NULL))
2117 goto out;
2118 err = -ENETDOWN;
2119 if (unlikely(!(dev->flags & IFF_UP)))
2120 goto out_put;
2121
2122 reserve = dev->hard_header_len;
2123
2124 size_max = po->tx_ring.frame_size
2125 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2126
2127 if (size_max > dev->mtu + reserve)
2128 size_max = dev->mtu + reserve;
2129
2130 do {
2131 ph = packet_current_frame(po, &po->tx_ring,
2132 TP_STATUS_SEND_REQUEST);
2133
2134 if (unlikely(ph == NULL)) {
2135 schedule();
2136 continue;
2137 }
2138
2139 status = TP_STATUS_SEND_REQUEST;
2140 hlen = LL_RESERVED_SPACE(dev);
2141 tlen = dev->needed_tailroom;
2142 skb = sock_alloc_send_skb(&po->sk,
2143 hlen + tlen + sizeof(struct sockaddr_ll),
2144 0, &err);
2145
2146 if (unlikely(skb == NULL))
2147 goto out_status;
2148
2149 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2150 addr, hlen);
2151
2152 if (unlikely(tp_len < 0)) {
2153 if (po->tp_loss) {
2154 __packet_set_status(po, ph,
2155 TP_STATUS_AVAILABLE);
2156 packet_increment_head(&po->tx_ring);
2157 kfree_skb(skb);
2158 continue;
2159 } else {
2160 status = TP_STATUS_WRONG_FORMAT;
2161 err = tp_len;
2162 goto out_status;
2163 }
2164 }
2165
2166 skb->destructor = tpacket_destruct_skb;
2167 __packet_set_status(po, ph, TP_STATUS_SENDING);
2168 atomic_inc(&po->tx_ring.pending);
2169
2170 status = TP_STATUS_SEND_REQUEST;
2171 err = dev_queue_xmit(skb);
2172 if (unlikely(err > 0)) {
2173 err = net_xmit_errno(err);
2174 if (err && __packet_get_status(po, ph) ==
2175 TP_STATUS_AVAILABLE) {
2176 /* skb was destructed already */
2177 skb = NULL;
2178 goto out_status;
2179 }
2180 /*
2181 * skb was dropped but not destructed yet;
2182 * let's treat it like congestion or err < 0
2183 */
2184 err = 0;
2185 }
2186 packet_increment_head(&po->tx_ring);
2187 len_sum += tp_len;
2188 } while (likely((ph != NULL) ||
2189 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2190 (atomic_read(&po->tx_ring.pending))))
2191 );
2192
2193 err = len_sum;
2194 goto out_put;
2195
2196 out_status:
2197 __packet_set_status(po, ph, status);
2198 kfree_skb(skb);
2199 out_put:
2200 dev_put(dev);
2201 out:
2202 mutex_unlock(&po->pg_vec_lock);
2203 return err;
2204 }
2205
2206 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2207 size_t reserve, size_t len,
2208 size_t linear, int noblock,
2209 int *err)
2210 {
2211 struct sk_buff *skb;
2212
2213 /* Under a page? Don't bother with paged skb. */
2214 if (prepad + len < PAGE_SIZE || !linear)
2215 linear = len;
2216
2217 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2218 err);
2219 if (!skb)
2220 return NULL;
2221
2222 skb_reserve(skb, reserve);
2223 skb_put(skb, linear);
2224 skb->data_len = len - linear;
2225 skb->len += len - linear;
2226
2227 return skb;
2228 }
2229
2230 static int packet_snd(struct socket *sock,
2231 struct msghdr *msg, size_t len)
2232 {
2233 struct sock *sk = sock->sk;
2234 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2235 struct sk_buff *skb;
2236 struct net_device *dev;
2237 __be16 proto;
2238 unsigned char *addr;
2239 int err, reserve = 0;
2240 struct virtio_net_hdr vnet_hdr = { 0 };
2241 int offset = 0;
2242 int vnet_hdr_len;
2243 struct packet_sock *po = pkt_sk(sk);
2244 unsigned short gso_type = 0;
2245 int hlen, tlen, linear;
2246 int extra_len = 0;
2247
2248 /*
2249 * Get and verify the address.
2250 */
2251
2252 if (likely(saddr == NULL)) {
2253 dev = packet_cached_dev_get(po);
2254 proto = po->num;
2255 addr = NULL;
2256 } else {
2257 err = -EINVAL;
2258 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2259 goto out;
2260 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2261 goto out;
2262 proto = saddr->sll_protocol;
2263 addr = saddr->sll_addr;
2264 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2265 }
2266
2267 err = -ENXIO;
2268 if (unlikely(dev == NULL))
2269 goto out_unlock;
2270 err = -ENETDOWN;
2271 if (unlikely(!(dev->flags & IFF_UP)))
2272 goto out_unlock;
2273
2274 if (sock->type == SOCK_RAW)
2275 reserve = dev->hard_header_len;
2276 if (po->has_vnet_hdr) {
2277 vnet_hdr_len = sizeof(vnet_hdr);
2278
2279 err = -EINVAL;
2280 if (len < vnet_hdr_len)
2281 goto out_unlock;
2282
2283 len -= vnet_hdr_len;
2284
2285 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2286 vnet_hdr_len);
2287 if (err < 0)
2288 goto out_unlock;
2289
2290 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2291 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2292 vnet_hdr.hdr_len))
2293 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2294 vnet_hdr.csum_offset + 2;
2295
2296 err = -EINVAL;
2297 if (vnet_hdr.hdr_len > len)
2298 goto out_unlock;
2299
2300 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2301 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2302 case VIRTIO_NET_HDR_GSO_TCPV4:
2303 gso_type = SKB_GSO_TCPV4;
2304 break;
2305 case VIRTIO_NET_HDR_GSO_TCPV6:
2306 gso_type = SKB_GSO_TCPV6;
2307 break;
2308 case VIRTIO_NET_HDR_GSO_UDP:
2309 gso_type = SKB_GSO_UDP;
2310 break;
2311 default:
2312 goto out_unlock;
2313 }
2314
2315 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2316 gso_type |= SKB_GSO_TCP_ECN;
2317
2318 if (vnet_hdr.gso_size == 0)
2319 goto out_unlock;
2320
2321 }
2322 }
2323
2324 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2325 if (!netif_supports_nofcs(dev)) {
2326 err = -EPROTONOSUPPORT;
2327 goto out_unlock;
2328 }
2329 extra_len = 4; /* We're doing our own CRC */
2330 }
2331
2332 err = -EMSGSIZE;
2333 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2334 goto out_unlock;
2335
2336 err = -ENOBUFS;
2337 hlen = LL_RESERVED_SPACE(dev);
2338 tlen = dev->needed_tailroom;
2339 linear = vnet_hdr.hdr_len;
2340 linear = max(linear, min_t(int, len, dev->hard_header_len));
2341 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2342 msg->msg_flags & MSG_DONTWAIT, &err);
2343 if (skb == NULL)
2344 goto out_unlock;
2345
2346 skb_set_network_header(skb, reserve);
2347
2348 err = -EINVAL;
2349 if (sock->type == SOCK_DGRAM &&
2350 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2351 goto out_free;
2352
2353 /* Returns -EFAULT on error */
2354 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2355 if (err)
2356 goto out_free;
2357
2358 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2359
2360 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2361 /* Earlier code assumed this would be a VLAN pkt,
2362 * double-check this now that we have the actual
2363 * packet in hand.
2364 */
2365 struct ethhdr *ehdr;
2366 skb_reset_mac_header(skb);
2367 ehdr = eth_hdr(skb);
2368 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2369 err = -EMSGSIZE;
2370 goto out_free;
2371 }
2372 }
2373
2374 skb->protocol = proto;
2375 skb->dev = dev;
2376 skb->priority = sk->sk_priority;
2377 skb->mark = sk->sk_mark;
2378
2379 if (po->has_vnet_hdr) {
2380 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2381 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2382 vnet_hdr.csum_offset)) {
2383 err = -EINVAL;
2384 goto out_free;
2385 }
2386 }
2387
2388 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2389 skb_shinfo(skb)->gso_type = gso_type;
2390
2391 /* Header must be checked, and gso_segs computed. */
2392 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2393 skb_shinfo(skb)->gso_segs = 0;
2394
2395 len += vnet_hdr_len;
2396 }
2397
2398 skb_probe_transport_header(skb, reserve);
2399
2400 if (unlikely(extra_len == 4))
2401 skb->no_fcs = 1;
2402
2403 /*
2404 * Now send it
2405 */
2406
2407 err = dev_queue_xmit(skb);
2408 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2409 goto out_unlock;
2410
2411 dev_put(dev);
2412
2413 return len;
2414
2415 out_free:
2416 kfree_skb(skb);
2417 out_unlock:
2418 if (dev)
2419 dev_put(dev);
2420 out:
2421 return err;
2422 }
2423
2424 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2425 struct msghdr *msg, size_t len)
2426 {
2427 struct sock *sk = sock->sk;
2428 struct packet_sock *po = pkt_sk(sk);
2429 if (po->tx_ring.pg_vec)
2430 return tpacket_snd(po, msg);
2431 else
2432 return packet_snd(sock, msg, len);
2433 }
2434
2435 /*
2436 * Close a PACKET socket. This is fairly simple. We immediately go
2437 * to 'closed' state and remove our protocol entry in the device list.
2438 */
2439
2440 static int packet_release(struct socket *sock)
2441 {
2442 struct sock *sk = sock->sk;
2443 struct packet_sock *po;
2444 struct packet_fanout *f;
2445 struct net *net;
2446 union tpacket_req_u req_u;
2447
2448 if (!sk)
2449 return 0;
2450
2451 net = sock_net(sk);
2452 po = pkt_sk(sk);
2453
2454 mutex_lock(&net->packet.sklist_lock);
2455 sk_del_node_init_rcu(sk);
2456 mutex_unlock(&net->packet.sklist_lock);
2457
2458 preempt_disable();
2459 sock_prot_inuse_add(net, sk->sk_prot, -1);
2460 preempt_enable();
2461
2462 spin_lock(&po->bind_lock);
2463 unregister_prot_hook(sk, false);
2464 packet_cached_dev_reset(po);
2465
2466 if (po->prot_hook.dev) {
2467 dev_put(po->prot_hook.dev);
2468 po->prot_hook.dev = NULL;
2469 }
2470 spin_unlock(&po->bind_lock);
2471
2472 packet_flush_mclist(sk);
2473
2474 if (po->rx_ring.pg_vec) {
2475 memset(&req_u, 0, sizeof(req_u));
2476 packet_set_ring(sk, &req_u, 1, 0);
2477 }
2478
2479 if (po->tx_ring.pg_vec) {
2480 memset(&req_u, 0, sizeof(req_u));
2481 packet_set_ring(sk, &req_u, 1, 1);
2482 }
2483
2484 f = fanout_release(sk);
2485
2486 synchronize_net();
2487
2488 if (f) {
2489 kfree(f);
2490 }
2491 /*
2492 * Now the socket is dead. No more input will appear.
2493 */
2494 sock_orphan(sk);
2495 sock->sk = NULL;
2496
2497 /* Purge queues */
2498
2499 skb_queue_purge(&sk->sk_receive_queue);
2500 sk_refcnt_debug_release(sk);
2501
2502 sock_put(sk);
2503 return 0;
2504 }
2505
2506 /*
2507 * Attach a packet hook.
2508 */
2509
2510 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2511 {
2512 struct packet_sock *po = pkt_sk(sk);
2513
2514 if (po->fanout) {
2515 if (dev)
2516 dev_put(dev);
2517
2518 return -EINVAL;
2519 }
2520
2521 lock_sock(sk);
2522
2523 spin_lock(&po->bind_lock);
2524 unregister_prot_hook(sk, true);
2525
2526 po->num = protocol;
2527 po->prot_hook.type = protocol;
2528 if (po->prot_hook.dev)
2529 dev_put(po->prot_hook.dev);
2530
2531 po->prot_hook.dev = dev;
2532 po->ifindex = dev ? dev->ifindex : 0;
2533
2534 packet_cached_dev_assign(po, dev);
2535
2536 if (protocol == 0)
2537 goto out_unlock;
2538
2539 if (!dev || (dev->flags & IFF_UP)) {
2540 register_prot_hook(sk);
2541 } else {
2542 sk->sk_err = ENETDOWN;
2543 if (!sock_flag(sk, SOCK_DEAD))
2544 sk->sk_error_report(sk);
2545 }
2546
2547 out_unlock:
2548 spin_unlock(&po->bind_lock);
2549 release_sock(sk);
2550 return 0;
2551 }
2552
2553 /*
2554 * Bind a packet socket to a device
2555 */
2556
2557 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2558 int addr_len)
2559 {
2560 struct sock *sk = sock->sk;
2561 char name[sizeof(uaddr->sa_data) + 1];
2562 struct net_device *dev;
2563 int err = -ENODEV;
2564
2565 /*
2566 * Check legality
2567 */
2568
2569 if (addr_len != sizeof(struct sockaddr))
2570 return -EINVAL;
2571 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
2572 * zero-terminated.
2573 */
2574 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
2575 name[sizeof(uaddr->sa_data)] = 0;
2576
2577 dev = dev_get_by_name(sock_net(sk), name);
2578 if (dev)
2579 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2580 return err;
2581 }
2582
2583 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2584 {
2585 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2586 struct sock *sk = sock->sk;
2587 struct net_device *dev = NULL;
2588 int err;
2589
2590
2591 /*
2592 * Check legality
2593 */
2594
2595 if (addr_len < sizeof(struct sockaddr_ll))
2596 return -EINVAL;
2597 if (sll->sll_family != AF_PACKET)
2598 return -EINVAL;
2599
2600 if (sll->sll_ifindex) {
2601 err = -ENODEV;
2602 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2603 if (dev == NULL)
2604 goto out;
2605 }
2606 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2607
2608 out:
2609 return err;
2610 }
2611
2612 static struct proto packet_proto = {
2613 .name = "PACKET",
2614 .owner = THIS_MODULE,
2615 .obj_size = sizeof(struct packet_sock),
2616 };
2617
2618 /*
2619 * Create a packet of type SOCK_PACKET.
2620 */
2621
2622 static int packet_create(struct net *net, struct socket *sock, int protocol,
2623 int kern)
2624 {
2625 struct sock *sk;
2626 struct packet_sock *po;
2627 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2628 int err;
2629
2630 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2631 return -EPERM;
2632 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2633 sock->type != SOCK_PACKET)
2634 return -ESOCKTNOSUPPORT;
2635
2636 sock->state = SS_UNCONNECTED;
2637
2638 err = -ENOBUFS;
2639 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2640 if (sk == NULL)
2641 goto out;
2642
2643 sock->ops = &packet_ops;
2644 if (sock->type == SOCK_PACKET)
2645 sock->ops = &packet_ops_spkt;
2646
2647 sock_init_data(sock, sk);
2648
2649 po = pkt_sk(sk);
2650 sk->sk_family = PF_PACKET;
2651 po->num = proto;
2652
2653 packet_cached_dev_reset(po);
2654
2655 sk->sk_destruct = packet_sock_destruct;
2656 sk_refcnt_debug_inc(sk);
2657
2658 /*
2659 * Attach a protocol block
2660 */
2661
2662 spin_lock_init(&po->bind_lock);
2663 mutex_init(&po->pg_vec_lock);
2664 po->prot_hook.func = packet_rcv;
2665
2666 if (sock->type == SOCK_PACKET)
2667 po->prot_hook.func = packet_rcv_spkt;
2668
2669 po->prot_hook.af_packet_priv = sk;
2670
2671 if (proto) {
2672 po->prot_hook.type = proto;
2673 register_prot_hook(sk);
2674 }
2675
2676 mutex_lock(&net->packet.sklist_lock);
2677 sk_add_node_rcu(sk, &net->packet.sklist);
2678 mutex_unlock(&net->packet.sklist_lock);
2679
2680 preempt_disable();
2681 sock_prot_inuse_add(net, &packet_proto, 1);
2682 preempt_enable();
2683
2684 return 0;
2685 out:
2686 return err;
2687 }
2688
2689 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2690 {
2691 struct sock_exterr_skb *serr;
2692 struct sk_buff *skb, *skb2;
2693 int copied, err;
2694
2695 err = -EAGAIN;
2696 skb = skb_dequeue(&sk->sk_error_queue);
2697 if (skb == NULL)
2698 goto out;
2699
2700 copied = skb->len;
2701 if (copied > len) {
2702 msg->msg_flags |= MSG_TRUNC;
2703 copied = len;
2704 }
2705 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2706 if (err)
2707 goto out_free_skb;
2708
2709 sock_recv_timestamp(msg, sk, skb);
2710
2711 serr = SKB_EXT_ERR(skb);
2712 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2713 sizeof(serr->ee), &serr->ee);
2714
2715 msg->msg_flags |= MSG_ERRQUEUE;
2716 err = copied;
2717
2718 /* Reset and regenerate socket error */
2719 spin_lock_bh(&sk->sk_error_queue.lock);
2720 sk->sk_err = 0;
2721 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2722 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2723 spin_unlock_bh(&sk->sk_error_queue.lock);
2724 sk->sk_error_report(sk);
2725 } else
2726 spin_unlock_bh(&sk->sk_error_queue.lock);
2727
2728 out_free_skb:
2729 kfree_skb(skb);
2730 out:
2731 return err;
2732 }
2733
2734 /*
2735 * Pull a packet from our receive queue and hand it to the user.
2736 * If necessary we block.
2737 */
2738
2739 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2740 struct msghdr *msg, size_t len, int flags)
2741 {
2742 struct sock *sk = sock->sk;
2743 struct sk_buff *skb;
2744 int copied, err;
2745 int vnet_hdr_len = 0;
2746
2747 err = -EINVAL;
2748 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2749 goto out;
2750
2751 #if 0
2752 /* What error should we return now? EUNATTACH? */
2753 if (pkt_sk(sk)->ifindex < 0)
2754 return -ENODEV;
2755 #endif
2756
2757 if (flags & MSG_ERRQUEUE) {
2758 err = packet_recv_error(sk, msg, len);
2759 goto out;
2760 }
2761
2762 /*
2763 * Call the generic datagram receiver. This handles all sorts
2764 * of horrible races and re-entrancy so we can forget about it
2765 * in the protocol layers.
2766 *
2767 * Now it will return ENETDOWN, if device have just gone down,
2768 * but then it will block.
2769 */
2770
2771 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2772
2773 /*
2774 * An error occurred so return it. Because skb_recv_datagram()
2775 * handles the blocking we don't see and worry about blocking
2776 * retries.
2777 */
2778
2779 if (skb == NULL)
2780 goto out;
2781
2782 if (pkt_sk(sk)->has_vnet_hdr) {
2783 struct virtio_net_hdr vnet_hdr = { 0 };
2784
2785 err = -EINVAL;
2786 vnet_hdr_len = sizeof(vnet_hdr);
2787 if (len < vnet_hdr_len)
2788 goto out_free;
2789
2790 len -= vnet_hdr_len;
2791
2792 if (skb_is_gso(skb)) {
2793 struct skb_shared_info *sinfo = skb_shinfo(skb);
2794
2795 /* This is a hint as to how much should be linear. */
2796 vnet_hdr.hdr_len = skb_headlen(skb);
2797 vnet_hdr.gso_size = sinfo->gso_size;
2798 if (sinfo->gso_type & SKB_GSO_TCPV4)
2799 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2800 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2801 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2802 else if (sinfo->gso_type & SKB_GSO_UDP)
2803 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2804 else if (sinfo->gso_type & SKB_GSO_FCOE)
2805 goto out_free;
2806 else
2807 BUG();
2808 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2809 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2810 } else
2811 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2812
2813 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2814 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2815 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2816 vnet_hdr.csum_offset = skb->csum_offset;
2817 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2818 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2819 } /* else everything is zero */
2820
2821 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2822 vnet_hdr_len);
2823 if (err < 0)
2824 goto out_free;
2825 }
2826
2827 /* You lose any data beyond the buffer you gave. If it worries
2828 * a user program they can ask the device for its MTU
2829 * anyway.
2830 */
2831 copied = skb->len;
2832 if (copied > len) {
2833 copied = len;
2834 msg->msg_flags |= MSG_TRUNC;
2835 }
2836
2837 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2838 if (err)
2839 goto out_free;
2840
2841 sock_recv_ts_and_drops(msg, sk, skb);
2842
2843 if (msg->msg_name) {
2844 /* If the address length field is there to be filled
2845 * in, we fill it in now.
2846 */
2847 if (sock->type == SOCK_PACKET) {
2848 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2849 } else {
2850 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2851 msg->msg_namelen = sll->sll_halen +
2852 offsetof(struct sockaddr_ll, sll_addr);
2853 }
2854 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2855 msg->msg_namelen);
2856 }
2857
2858 if (pkt_sk(sk)->auxdata) {
2859 struct tpacket_auxdata aux;
2860
2861 aux.tp_status = TP_STATUS_USER;
2862 if (skb->ip_summed == CHECKSUM_PARTIAL)
2863 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2864 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2865 aux.tp_snaplen = skb->len;
2866 aux.tp_mac = 0;
2867 aux.tp_net = skb_network_offset(skb);
2868 if (vlan_tx_tag_present(skb)) {
2869 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2870 aux.tp_status |= TP_STATUS_VLAN_VALID;
2871 } else {
2872 aux.tp_vlan_tci = 0;
2873 }
2874 aux.tp_padding = 0;
2875 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2876 }
2877
2878 /*
2879 * Free or return the buffer as appropriate. Again this
2880 * hides all the races and re-entrancy issues from us.
2881 */
2882 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2883
2884 out_free:
2885 skb_free_datagram(sk, skb);
2886 out:
2887 return err;
2888 }
2889
2890 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2891 int *uaddr_len, int peer)
2892 {
2893 struct net_device *dev;
2894 struct sock *sk = sock->sk;
2895
2896 if (peer)
2897 return -EOPNOTSUPP;
2898
2899 uaddr->sa_family = AF_PACKET;
2900 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2901 rcu_read_lock();
2902 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2903 if (dev)
2904 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2905 rcu_read_unlock();
2906 *uaddr_len = sizeof(*uaddr);
2907
2908 return 0;
2909 }
2910
2911 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2912 int *uaddr_len, int peer)
2913 {
2914 struct net_device *dev;
2915 struct sock *sk = sock->sk;
2916 struct packet_sock *po = pkt_sk(sk);
2917 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2918
2919 if (peer)
2920 return -EOPNOTSUPP;
2921
2922 sll->sll_family = AF_PACKET;
2923 sll->sll_ifindex = po->ifindex;
2924 sll->sll_protocol = po->num;
2925 sll->sll_pkttype = 0;
2926 rcu_read_lock();
2927 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2928 if (dev) {
2929 sll->sll_hatype = dev->type;
2930 sll->sll_halen = dev->addr_len;
2931 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2932 } else {
2933 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2934 sll->sll_halen = 0;
2935 }
2936 rcu_read_unlock();
2937 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2938
2939 return 0;
2940 }
2941
2942 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2943 int what)
2944 {
2945 switch (i->type) {
2946 case PACKET_MR_MULTICAST:
2947 if (i->alen != dev->addr_len)
2948 return -EINVAL;
2949 if (what > 0)
2950 return dev_mc_add(dev, i->addr);
2951 else
2952 return dev_mc_del(dev, i->addr);
2953 break;
2954 case PACKET_MR_PROMISC:
2955 return dev_set_promiscuity(dev, what);
2956 break;
2957 case PACKET_MR_ALLMULTI:
2958 return dev_set_allmulti(dev, what);
2959 break;
2960 case PACKET_MR_UNICAST:
2961 if (i->alen != dev->addr_len)
2962 return -EINVAL;
2963 if (what > 0)
2964 return dev_uc_add(dev, i->addr);
2965 else
2966 return dev_uc_del(dev, i->addr);
2967 break;
2968 default:
2969 break;
2970 }
2971 return 0;
2972 }
2973
2974 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2975 {
2976 for ( ; i; i = i->next) {
2977 if (i->ifindex == dev->ifindex)
2978 packet_dev_mc(dev, i, what);
2979 }
2980 }
2981
2982 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2983 {
2984 struct packet_sock *po = pkt_sk(sk);
2985 struct packet_mclist *ml, *i;
2986 struct net_device *dev;
2987 int err;
2988
2989 rtnl_lock();
2990
2991 err = -ENODEV;
2992 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2993 if (!dev)
2994 goto done;
2995
2996 err = -EINVAL;
2997 if (mreq->mr_alen > dev->addr_len)
2998 goto done;
2999
3000 err = -ENOBUFS;
3001 i = kmalloc(sizeof(*i), GFP_KERNEL);
3002 if (i == NULL)
3003 goto done;
3004
3005 err = 0;
3006 for (ml = po->mclist; ml; ml = ml->next) {
3007 if (ml->ifindex == mreq->mr_ifindex &&
3008 ml->type == mreq->mr_type &&
3009 ml->alen == mreq->mr_alen &&
3010 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3011 ml->count++;
3012 /* Free the new element ... */
3013 kfree(i);
3014 goto done;
3015 }
3016 }
3017
3018 i->type = mreq->mr_type;
3019 i->ifindex = mreq->mr_ifindex;
3020 i->alen = mreq->mr_alen;
3021 memcpy(i->addr, mreq->mr_address, i->alen);
3022 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3023 i->count = 1;
3024 i->next = po->mclist;
3025 po->mclist = i;
3026 err = packet_dev_mc(dev, i, 1);
3027 if (err) {
3028 po->mclist = i->next;
3029 kfree(i);
3030 }
3031
3032 done:
3033 rtnl_unlock();
3034 return err;
3035 }
3036
3037 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3038 {
3039 struct packet_mclist *ml, **mlp;
3040
3041 rtnl_lock();
3042
3043 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3044 if (ml->ifindex == mreq->mr_ifindex &&
3045 ml->type == mreq->mr_type &&
3046 ml->alen == mreq->mr_alen &&
3047 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3048 if (--ml->count == 0) {
3049 struct net_device *dev;
3050 *mlp = ml->next;
3051 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3052 if (dev)
3053 packet_dev_mc(dev, ml, -1);
3054 kfree(ml);
3055 }
3056 rtnl_unlock();
3057 return 0;
3058 }
3059 }
3060 rtnl_unlock();
3061 return -EADDRNOTAVAIL;
3062 }
3063
3064 static void packet_flush_mclist(struct sock *sk)
3065 {
3066 struct packet_sock *po = pkt_sk(sk);
3067 struct packet_mclist *ml;
3068
3069 if (!po->mclist)
3070 return;
3071
3072 rtnl_lock();
3073 while ((ml = po->mclist) != NULL) {
3074 struct net_device *dev;
3075
3076 po->mclist = ml->next;
3077 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3078 if (dev != NULL)
3079 packet_dev_mc(dev, ml, -1);
3080 kfree(ml);
3081 }
3082 rtnl_unlock();
3083 }
3084
3085 static int
3086 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3087 {
3088 struct sock *sk = sock->sk;
3089 struct packet_sock *po = pkt_sk(sk);
3090 int ret;
3091
3092 if (level != SOL_PACKET)
3093 return -ENOPROTOOPT;
3094
3095 switch (optname) {
3096 case PACKET_ADD_MEMBERSHIP:
3097 case PACKET_DROP_MEMBERSHIP:
3098 {
3099 struct packet_mreq_max mreq;
3100 int len = optlen;
3101 memset(&mreq, 0, sizeof(mreq));
3102 if (len < sizeof(struct packet_mreq))
3103 return -EINVAL;
3104 if (len > sizeof(mreq))
3105 len = sizeof(mreq);
3106 if (copy_from_user(&mreq, optval, len))
3107 return -EFAULT;
3108 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3109 return -EINVAL;
3110 if (optname == PACKET_ADD_MEMBERSHIP)
3111 ret = packet_mc_add(sk, &mreq);
3112 else
3113 ret = packet_mc_drop(sk, &mreq);
3114 return ret;
3115 }
3116
3117 case PACKET_RX_RING:
3118 case PACKET_TX_RING:
3119 {
3120 union tpacket_req_u req_u;
3121 int len;
3122
3123 switch (po->tp_version) {
3124 case TPACKET_V1:
3125 case TPACKET_V2:
3126 len = sizeof(req_u.req);
3127 break;
3128 case TPACKET_V3:
3129 default:
3130 len = sizeof(req_u.req3);
3131 break;
3132 }
3133 if (optlen < len)
3134 return -EINVAL;
3135 if (pkt_sk(sk)->has_vnet_hdr)
3136 return -EINVAL;
3137 if (copy_from_user(&req_u.req, optval, len))
3138 return -EFAULT;
3139 return packet_set_ring(sk, &req_u, 0,
3140 optname == PACKET_TX_RING);
3141 }
3142 case PACKET_COPY_THRESH:
3143 {
3144 int val;
3145
3146 if (optlen != sizeof(val))
3147 return -EINVAL;
3148 if (copy_from_user(&val, optval, sizeof(val)))
3149 return -EFAULT;
3150
3151 pkt_sk(sk)->copy_thresh = val;
3152 return 0;
3153 }
3154 case PACKET_VERSION:
3155 {
3156 int val;
3157
3158 if (optlen != sizeof(val))
3159 return -EINVAL;
3160 if (copy_from_user(&val, optval, sizeof(val)))
3161 return -EFAULT;
3162 switch (val) {
3163 case TPACKET_V1:
3164 case TPACKET_V2:
3165 case TPACKET_V3:
3166 break;
3167 default:
3168 return -EINVAL;
3169 }
3170 lock_sock(sk);
3171 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3172 ret = -EBUSY;
3173 } else {
3174 po->tp_version = val;
3175 ret = 0;
3176 }
3177 release_sock(sk);
3178 return ret;
3179 }
3180 case PACKET_RESERVE:
3181 {
3182 unsigned int val;
3183
3184 if (optlen != sizeof(val))
3185 return -EINVAL;
3186 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3187 return -EBUSY;
3188 if (copy_from_user(&val, optval, sizeof(val)))
3189 return -EFAULT;
3190 if (val > INT_MAX)
3191 return -EINVAL;
3192 po->tp_reserve = val;
3193 return 0;
3194 }
3195 case PACKET_LOSS:
3196 {
3197 unsigned int val;
3198
3199 if (optlen != sizeof(val))
3200 return -EINVAL;
3201 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3202 return -EBUSY;
3203 if (copy_from_user(&val, optval, sizeof(val)))
3204 return -EFAULT;
3205 po->tp_loss = !!val;
3206 return 0;
3207 }
3208 case PACKET_AUXDATA:
3209 {
3210 int val;
3211
3212 if (optlen < sizeof(val))
3213 return -EINVAL;
3214 if (copy_from_user(&val, optval, sizeof(val)))
3215 return -EFAULT;
3216
3217 po->auxdata = !!val;
3218 return 0;
3219 }
3220 case PACKET_ORIGDEV:
3221 {
3222 int val;
3223
3224 if (optlen < sizeof(val))
3225 return -EINVAL;
3226 if (copy_from_user(&val, optval, sizeof(val)))
3227 return -EFAULT;
3228
3229 po->origdev = !!val;
3230 return 0;
3231 }
3232 case PACKET_VNET_HDR:
3233 {
3234 int val;
3235
3236 if (sock->type != SOCK_RAW)
3237 return -EINVAL;
3238 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3239 return -EBUSY;
3240 if (optlen < sizeof(val))
3241 return -EINVAL;
3242 if (copy_from_user(&val, optval, sizeof(val)))
3243 return -EFAULT;
3244
3245 po->has_vnet_hdr = !!val;
3246 return 0;
3247 }
3248 case PACKET_TIMESTAMP:
3249 {
3250 int val;
3251
3252 if (optlen != sizeof(val))
3253 return -EINVAL;
3254 if (copy_from_user(&val, optval, sizeof(val)))
3255 return -EFAULT;
3256
3257 po->tp_tstamp = val;
3258 return 0;
3259 }
3260 case PACKET_FANOUT:
3261 {
3262 int val;
3263
3264 if (optlen != sizeof(val))
3265 return -EINVAL;
3266 if (copy_from_user(&val, optval, sizeof(val)))
3267 return -EFAULT;
3268
3269 return fanout_add(sk, val & 0xffff, val >> 16);
3270 }
3271 case PACKET_TX_HAS_OFF:
3272 {
3273 unsigned int val;
3274
3275 if (optlen != sizeof(val))
3276 return -EINVAL;
3277 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3278 return -EBUSY;
3279 if (copy_from_user(&val, optval, sizeof(val)))
3280 return -EFAULT;
3281 po->tp_tx_has_off = !!val;
3282 return 0;
3283 }
3284 default:
3285 return -ENOPROTOOPT;
3286 }
3287 }
3288
3289 static int packet_getsockopt(struct socket *sock, int level, int optname,
3290 char __user *optval, int __user *optlen)
3291 {
3292 int len;
3293 int val, lv = sizeof(val);
3294 struct sock *sk = sock->sk;
3295 struct packet_sock *po = pkt_sk(sk);
3296 void *data = &val;
3297 union tpacket_stats_u st;
3298
3299 if (level != SOL_PACKET)
3300 return -ENOPROTOOPT;
3301
3302 if (get_user(len, optlen))
3303 return -EFAULT;
3304
3305 if (len < 0)
3306 return -EINVAL;
3307
3308 switch (optname) {
3309 case PACKET_STATISTICS:
3310 spin_lock_bh(&sk->sk_receive_queue.lock);
3311 memcpy(&st, &po->stats, sizeof(st));
3312 memset(&po->stats, 0, sizeof(po->stats));
3313 spin_unlock_bh(&sk->sk_receive_queue.lock);
3314
3315 if (po->tp_version == TPACKET_V3) {
3316 lv = sizeof(struct tpacket_stats_v3);
3317 st.stats3.tp_packets += st.stats3.tp_drops;
3318 data = &st.stats3;
3319 } else {
3320 lv = sizeof(struct tpacket_stats);
3321 st.stats1.tp_packets += st.stats1.tp_drops;
3322 data = &st.stats1;
3323 }
3324
3325 break;
3326 case PACKET_AUXDATA:
3327 val = po->auxdata;
3328 break;
3329 case PACKET_ORIGDEV:
3330 val = po->origdev;
3331 break;
3332 case PACKET_VNET_HDR:
3333 val = po->has_vnet_hdr;
3334 break;
3335 case PACKET_VERSION:
3336 val = po->tp_version;
3337 break;
3338 case PACKET_HDRLEN:
3339 if (len > sizeof(int))
3340 len = sizeof(int);
3341 if (copy_from_user(&val, optval, len))
3342 return -EFAULT;
3343 switch (val) {
3344 case TPACKET_V1:
3345 val = sizeof(struct tpacket_hdr);
3346 break;
3347 case TPACKET_V2:
3348 val = sizeof(struct tpacket2_hdr);
3349 break;
3350 case TPACKET_V3:
3351 val = sizeof(struct tpacket3_hdr);
3352 break;
3353 default:
3354 return -EINVAL;
3355 }
3356 break;
3357 case PACKET_RESERVE:
3358 val = po->tp_reserve;
3359 break;
3360 case PACKET_LOSS:
3361 val = po->tp_loss;
3362 break;
3363 case PACKET_TIMESTAMP:
3364 val = po->tp_tstamp;
3365 break;
3366 case PACKET_FANOUT:
3367 val = (po->fanout ?
3368 ((u32)po->fanout->id |
3369 ((u32)po->fanout->type << 16) |
3370 ((u32)po->fanout->flags << 24)) :
3371 0);
3372 break;
3373 case PACKET_TX_HAS_OFF:
3374 val = po->tp_tx_has_off;
3375 break;
3376 default:
3377 return -ENOPROTOOPT;
3378 }
3379
3380 if (len > lv)
3381 len = lv;
3382 if (put_user(len, optlen))
3383 return -EFAULT;
3384 if (copy_to_user(optval, data, len))
3385 return -EFAULT;
3386 return 0;
3387 }
3388
3389
3390 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3391 {
3392 struct sock *sk;
3393 struct net_device *dev = data;
3394 struct net *net = dev_net(dev);
3395
3396 rcu_read_lock();
3397 sk_for_each_rcu(sk, &net->packet.sklist) {
3398 struct packet_sock *po = pkt_sk(sk);
3399
3400 switch (msg) {
3401 case NETDEV_UNREGISTER:
3402 if (po->mclist)
3403 packet_dev_mclist(dev, po->mclist, -1);
3404 /* fallthrough */
3405
3406 case NETDEV_DOWN:
3407 if (dev->ifindex == po->ifindex) {
3408 spin_lock(&po->bind_lock);
3409 if (po->running) {
3410 __unregister_prot_hook(sk, false);
3411 sk->sk_err = ENETDOWN;
3412 if (!sock_flag(sk, SOCK_DEAD))
3413 sk->sk_error_report(sk);
3414 }
3415 if (msg == NETDEV_UNREGISTER) {
3416 packet_cached_dev_reset(po);
3417 po->ifindex = -1;
3418 if (po->prot_hook.dev)
3419 dev_put(po->prot_hook.dev);
3420 po->prot_hook.dev = NULL;
3421 }
3422 spin_unlock(&po->bind_lock);
3423 }
3424 break;
3425 case NETDEV_UP:
3426 if (dev->ifindex == po->ifindex) {
3427 spin_lock(&po->bind_lock);
3428 if (po->num)
3429 register_prot_hook(sk);
3430 spin_unlock(&po->bind_lock);
3431 }
3432 break;
3433 }
3434 }
3435 rcu_read_unlock();
3436 return NOTIFY_DONE;
3437 }
3438
3439
3440 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3441 unsigned long arg)
3442 {
3443 struct sock *sk = sock->sk;
3444
3445 switch (cmd) {
3446 case SIOCOUTQ:
3447 {
3448 int amount = sk_wmem_alloc_get(sk);
3449
3450 return put_user(amount, (int __user *)arg);
3451 }
3452 case SIOCINQ:
3453 {
3454 struct sk_buff *skb;
3455 int amount = 0;
3456
3457 spin_lock_bh(&sk->sk_receive_queue.lock);
3458 skb = skb_peek(&sk->sk_receive_queue);
3459 if (skb)
3460 amount = skb->len;
3461 spin_unlock_bh(&sk->sk_receive_queue.lock);
3462 return put_user(amount, (int __user *)arg);
3463 }
3464 case SIOCGSTAMP:
3465 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3466 case SIOCGSTAMPNS:
3467 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3468
3469 #ifdef CONFIG_INET
3470 case SIOCADDRT:
3471 case SIOCDELRT:
3472 case SIOCDARP:
3473 case SIOCGARP:
3474 case SIOCSARP:
3475 case SIOCGIFADDR:
3476 case SIOCSIFADDR:
3477 case SIOCGIFBRDADDR:
3478 case SIOCSIFBRDADDR:
3479 case SIOCGIFNETMASK:
3480 case SIOCSIFNETMASK:
3481 case SIOCGIFDSTADDR:
3482 case SIOCSIFDSTADDR:
3483 case SIOCSIFFLAGS:
3484 return inet_dgram_ops.ioctl(sock, cmd, arg);
3485 #endif
3486
3487 default:
3488 return -ENOIOCTLCMD;
3489 }
3490 return 0;
3491 }
3492
3493 static unsigned int packet_poll(struct file *file, struct socket *sock,
3494 poll_table *wait)
3495 {
3496 struct sock *sk = sock->sk;
3497 struct packet_sock *po = pkt_sk(sk);
3498 unsigned int mask = datagram_poll(file, sock, wait);
3499
3500 spin_lock_bh(&sk->sk_receive_queue.lock);
3501 if (po->rx_ring.pg_vec) {
3502 if (!packet_previous_rx_frame(po, &po->rx_ring,
3503 TP_STATUS_KERNEL))
3504 mask |= POLLIN | POLLRDNORM;
3505 }
3506 spin_unlock_bh(&sk->sk_receive_queue.lock);
3507 spin_lock_bh(&sk->sk_write_queue.lock);
3508 if (po->tx_ring.pg_vec) {
3509 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3510 mask |= POLLOUT | POLLWRNORM;
3511 }
3512 spin_unlock_bh(&sk->sk_write_queue.lock);
3513 return mask;
3514 }
3515
3516
3517 /* Dirty? Well, I still did not learn better way to account
3518 * for user mmaps.
3519 */
3520
3521 static void packet_mm_open(struct vm_area_struct *vma)
3522 {
3523 struct file *file = vma->vm_file;
3524 struct socket *sock = file->private_data;
3525 struct sock *sk = sock->sk;
3526
3527 if (sk)
3528 atomic_inc(&pkt_sk(sk)->mapped);
3529 }
3530
3531 static void packet_mm_close(struct vm_area_struct *vma)
3532 {
3533 struct file *file = vma->vm_file;
3534 struct socket *sock = file->private_data;
3535 struct sock *sk = sock->sk;
3536
3537 if (sk)
3538 atomic_dec(&pkt_sk(sk)->mapped);
3539 }
3540
3541 static const struct vm_operations_struct packet_mmap_ops = {
3542 .open = packet_mm_open,
3543 .close = packet_mm_close,
3544 };
3545
3546 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3547 unsigned int len)
3548 {
3549 int i;
3550
3551 for (i = 0; i < len; i++) {
3552 if (likely(pg_vec[i].buffer)) {
3553 if (is_vmalloc_addr(pg_vec[i].buffer))
3554 vfree(pg_vec[i].buffer);
3555 else
3556 free_pages((unsigned long)pg_vec[i].buffer,
3557 order);
3558 pg_vec[i].buffer = NULL;
3559 }
3560 }
3561 kfree(pg_vec);
3562 }
3563
3564 static char *alloc_one_pg_vec_page(unsigned long order)
3565 {
3566 char *buffer = NULL;
3567 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3568 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3569
3570 buffer = (char *) __get_free_pages(gfp_flags, order);
3571
3572 if (buffer)
3573 return buffer;
3574
3575 /*
3576 * __get_free_pages failed, fall back to vmalloc
3577 */
3578 buffer = vzalloc((1 << order) * PAGE_SIZE);
3579
3580 if (buffer)
3581 return buffer;
3582
3583 /*
3584 * vmalloc failed, lets dig into swap here
3585 */
3586 gfp_flags &= ~__GFP_NORETRY;
3587 buffer = (char *)__get_free_pages(gfp_flags, order);
3588 if (buffer)
3589 return buffer;
3590
3591 /*
3592 * complete and utter failure
3593 */
3594 return NULL;
3595 }
3596
3597 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3598 {
3599 unsigned int block_nr = req->tp_block_nr;
3600 struct pgv *pg_vec;
3601 int i;
3602
3603 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3604 if (unlikely(!pg_vec))
3605 goto out;
3606
3607 for (i = 0; i < block_nr; i++) {
3608 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3609 if (unlikely(!pg_vec[i].buffer))
3610 goto out_free_pgvec;
3611 }
3612
3613 out:
3614 return pg_vec;
3615
3616 out_free_pgvec:
3617 free_pg_vec(pg_vec, order, block_nr);
3618 pg_vec = NULL;
3619 goto out;
3620 }
3621
3622 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3623 int closing, int tx_ring)
3624 {
3625 struct pgv *pg_vec = NULL;
3626 struct packet_sock *po = pkt_sk(sk);
3627 int was_running, order = 0;
3628 struct packet_ring_buffer *rb;
3629 struct sk_buff_head *rb_queue;
3630 __be16 num;
3631 int err = -EINVAL;
3632 /* Added to avoid minimal code churn */
3633 struct tpacket_req *req = &req_u->req;
3634
3635 lock_sock(sk);
3636 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3637 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3638 WARN(1, "Tx-ring is not supported.\n");
3639 goto out;
3640 }
3641
3642 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3643 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3644
3645 err = -EBUSY;
3646 if (!closing) {
3647 if (atomic_read(&po->mapped))
3648 goto out;
3649 if (atomic_read(&rb->pending))
3650 goto out;
3651 }
3652
3653 if (req->tp_block_nr) {
3654 /* Sanity tests and some calculations */
3655 err = -EBUSY;
3656 if (unlikely(rb->pg_vec))
3657 goto out;
3658
3659 switch (po->tp_version) {
3660 case TPACKET_V1:
3661 po->tp_hdrlen = TPACKET_HDRLEN;
3662 break;
3663 case TPACKET_V2:
3664 po->tp_hdrlen = TPACKET2_HDRLEN;
3665 break;
3666 case TPACKET_V3:
3667 po->tp_hdrlen = TPACKET3_HDRLEN;
3668 break;
3669 }
3670
3671 err = -EINVAL;
3672 if (unlikely((int)req->tp_block_size <= 0))
3673 goto out;
3674 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3675 goto out;
3676 if (po->tp_version >= TPACKET_V3 &&
3677 req->tp_block_size <=
3678 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
3679 goto out;
3680 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3681 po->tp_reserve))
3682 goto out;
3683 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3684 goto out;
3685
3686 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3687 if (unlikely(rb->frames_per_block <= 0))
3688 goto out;
3689 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
3690 goto out;
3691 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3692 req->tp_frame_nr))
3693 goto out;
3694
3695 err = -ENOMEM;
3696 order = get_order(req->tp_block_size);
3697 pg_vec = alloc_pg_vec(req, order);
3698 if (unlikely(!pg_vec))
3699 goto out;
3700 switch (po->tp_version) {
3701 case TPACKET_V3:
3702 /* Transmit path is not supported. We checked
3703 * it above but just being paranoid
3704 */
3705 if (!tx_ring)
3706 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3707 break;
3708 default:
3709 break;
3710 }
3711 }
3712 /* Done */
3713 else {
3714 err = -EINVAL;
3715 if (unlikely(req->tp_frame_nr))
3716 goto out;
3717 }
3718
3719
3720 /* Detach socket from network */
3721 spin_lock(&po->bind_lock);
3722 was_running = po->running;
3723 num = po->num;
3724 if (was_running) {
3725 po->num = 0;
3726 __unregister_prot_hook(sk, false);
3727 }
3728 spin_unlock(&po->bind_lock);
3729
3730 synchronize_net();
3731
3732 err = -EBUSY;
3733 mutex_lock(&po->pg_vec_lock);
3734 if (closing || atomic_read(&po->mapped) == 0) {
3735 err = 0;
3736 spin_lock_bh(&rb_queue->lock);
3737 swap(rb->pg_vec, pg_vec);
3738 rb->frame_max = (req->tp_frame_nr - 1);
3739 rb->head = 0;
3740 rb->frame_size = req->tp_frame_size;
3741 spin_unlock_bh(&rb_queue->lock);
3742
3743 swap(rb->pg_vec_order, order);
3744 swap(rb->pg_vec_len, req->tp_block_nr);
3745
3746 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3747 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3748 tpacket_rcv : packet_rcv;
3749 skb_queue_purge(rb_queue);
3750 if (atomic_read(&po->mapped))
3751 pr_err("packet_mmap: vma is busy: %d\n",
3752 atomic_read(&po->mapped));
3753 }
3754 mutex_unlock(&po->pg_vec_lock);
3755
3756 spin_lock(&po->bind_lock);
3757 if (was_running) {
3758 po->num = num;
3759 register_prot_hook(sk);
3760 }
3761 spin_unlock(&po->bind_lock);
3762 if (closing && (po->tp_version > TPACKET_V2)) {
3763 /* Because we don't support block-based V3 on tx-ring */
3764 if (!tx_ring)
3765 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3766 }
3767
3768 if (pg_vec)
3769 free_pg_vec(pg_vec, order, req->tp_block_nr);
3770 out:
3771 release_sock(sk);
3772 return err;
3773 }
3774
3775 static int packet_mmap(struct file *file, struct socket *sock,
3776 struct vm_area_struct *vma)
3777 {
3778 struct sock *sk = sock->sk;
3779 struct packet_sock *po = pkt_sk(sk);
3780 unsigned long size, expected_size;
3781 struct packet_ring_buffer *rb;
3782 unsigned long start;
3783 int err = -EINVAL;
3784 int i;
3785
3786 if (vma->vm_pgoff)
3787 return -EINVAL;
3788
3789 mutex_lock(&po->pg_vec_lock);
3790
3791 expected_size = 0;
3792 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3793 if (rb->pg_vec) {
3794 expected_size += rb->pg_vec_len
3795 * rb->pg_vec_pages
3796 * PAGE_SIZE;
3797 }
3798 }
3799
3800 if (expected_size == 0)
3801 goto out;
3802
3803 size = vma->vm_end - vma->vm_start;
3804 if (size != expected_size)
3805 goto out;
3806
3807 start = vma->vm_start;
3808 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3809 if (rb->pg_vec == NULL)
3810 continue;
3811
3812 for (i = 0; i < rb->pg_vec_len; i++) {
3813 struct page *page;
3814 void *kaddr = rb->pg_vec[i].buffer;
3815 int pg_num;
3816
3817 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3818 page = pgv_to_page(kaddr);
3819 err = vm_insert_page(vma, start, page);
3820 if (unlikely(err))
3821 goto out;
3822 start += PAGE_SIZE;
3823 kaddr += PAGE_SIZE;
3824 }
3825 }
3826 }
3827
3828 atomic_inc(&po->mapped);
3829 vma->vm_ops = &packet_mmap_ops;
3830 err = 0;
3831
3832 out:
3833 mutex_unlock(&po->pg_vec_lock);
3834 return err;
3835 }
3836
3837 static const struct proto_ops packet_ops_spkt = {
3838 .family = PF_PACKET,
3839 .owner = THIS_MODULE,
3840 .release = packet_release,
3841 .bind = packet_bind_spkt,
3842 .connect = sock_no_connect,
3843 .socketpair = sock_no_socketpair,
3844 .accept = sock_no_accept,
3845 .getname = packet_getname_spkt,
3846 .poll = datagram_poll,
3847 .ioctl = packet_ioctl,
3848 .listen = sock_no_listen,
3849 .shutdown = sock_no_shutdown,
3850 .setsockopt = sock_no_setsockopt,
3851 .getsockopt = sock_no_getsockopt,
3852 .sendmsg = packet_sendmsg_spkt,
3853 .recvmsg = packet_recvmsg,
3854 .mmap = sock_no_mmap,
3855 .sendpage = sock_no_sendpage,
3856 };
3857
3858 static const struct proto_ops packet_ops = {
3859 .family = PF_PACKET,
3860 .owner = THIS_MODULE,
3861 .release = packet_release,
3862 .bind = packet_bind,
3863 .connect = sock_no_connect,
3864 .socketpair = sock_no_socketpair,
3865 .accept = sock_no_accept,
3866 .getname = packet_getname,
3867 .poll = packet_poll,
3868 .ioctl = packet_ioctl,
3869 .listen = sock_no_listen,
3870 .shutdown = sock_no_shutdown,
3871 .setsockopt = packet_setsockopt,
3872 .getsockopt = packet_getsockopt,
3873 .sendmsg = packet_sendmsg,
3874 .recvmsg = packet_recvmsg,
3875 .mmap = packet_mmap,
3876 .sendpage = sock_no_sendpage,
3877 };
3878
3879 static const struct net_proto_family packet_family_ops = {
3880 .family = PF_PACKET,
3881 .create = packet_create,
3882 .owner = THIS_MODULE,
3883 };
3884
3885 static struct notifier_block packet_netdev_notifier = {
3886 .notifier_call = packet_notifier,
3887 };
3888
3889 #ifdef CONFIG_PROC_FS
3890
3891 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3892 __acquires(RCU)
3893 {
3894 struct net *net = seq_file_net(seq);
3895
3896 rcu_read_lock();
3897 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3898 }
3899
3900 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3901 {
3902 struct net *net = seq_file_net(seq);
3903 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3904 }
3905
3906 static void packet_seq_stop(struct seq_file *seq, void *v)
3907 __releases(RCU)
3908 {
3909 rcu_read_unlock();
3910 }
3911
3912 static int packet_seq_show(struct seq_file *seq, void *v)
3913 {
3914 if (v == SEQ_START_TOKEN)
3915 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3916 else {
3917 struct sock *s = sk_entry(v);
3918 const struct packet_sock *po = pkt_sk(s);
3919
3920 seq_printf(seq,
3921 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3922 s,
3923 atomic_read(&s->sk_refcnt),
3924 s->sk_type,
3925 ntohs(po->num),
3926 po->ifindex,
3927 po->running,
3928 atomic_read(&s->sk_rmem_alloc),
3929 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3930 sock_i_ino(s));
3931 }
3932
3933 return 0;
3934 }
3935
3936 static const struct seq_operations packet_seq_ops = {
3937 .start = packet_seq_start,
3938 .next = packet_seq_next,
3939 .stop = packet_seq_stop,
3940 .show = packet_seq_show,
3941 };
3942
3943 static int packet_seq_open(struct inode *inode, struct file *file)
3944 {
3945 return seq_open_net(inode, file, &packet_seq_ops,
3946 sizeof(struct seq_net_private));
3947 }
3948
3949 static const struct file_operations packet_seq_fops = {
3950 .owner = THIS_MODULE,
3951 .open = packet_seq_open,
3952 .read = seq_read,
3953 .llseek = seq_lseek,
3954 .release = seq_release_net,
3955 };
3956
3957 #endif
3958
3959 static int __net_init packet_net_init(struct net *net)
3960 {
3961 mutex_init(&net->packet.sklist_lock);
3962 INIT_HLIST_HEAD(&net->packet.sklist);
3963
3964 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3965 return -ENOMEM;
3966
3967 return 0;
3968 }
3969
3970 static void __net_exit packet_net_exit(struct net *net)
3971 {
3972 remove_proc_entry("packet", net->proc_net);
3973 }
3974
3975 static struct pernet_operations packet_net_ops = {
3976 .init = packet_net_init,
3977 .exit = packet_net_exit,
3978 };
3979
3980
3981 static void __exit packet_exit(void)
3982 {
3983 unregister_netdevice_notifier(&packet_netdev_notifier);
3984 unregister_pernet_subsys(&packet_net_ops);
3985 sock_unregister(PF_PACKET);
3986 proto_unregister(&packet_proto);
3987 }
3988
3989 static int __init packet_init(void)
3990 {
3991 int rc = proto_register(&packet_proto, 0);
3992
3993 if (rc != 0)
3994 goto out;
3995
3996 sock_register(&packet_family_ops);
3997 register_pernet_subsys(&packet_net_ops);
3998 register_netdevice_notifier(&packet_netdev_notifier);
3999 out:
4000 return rc;
4001 }
4002
4003 module_init(packet_init);
4004 module_exit(packet_exit);
4005 MODULE_LICENSE("GPL");
4006 MODULE_ALIAS_NETPROTO(PF_PACKET);
kernel source for telekom puls (alcatel/mediatek)