projects / deliverable / linux.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
net: socket: move ktime2ts to ktime header api
[deliverable/linux.git] / net / packet / af_packet.c
... / ...
CommitLineData
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
109On receive:
110-----------
111
112Incoming, dev->hard_header!=NULL
113 mac_header -> ll header
114 data -> data
115
116Outgoing, dev->hard_header!=NULL
117 mac_header -> ll header
118 data -> ll header
119
120Incoming, 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
126Outgoing, dev->hard_header==NULL
127 mac_header -> data. ll header is still not built!
128 data -> data
129
130Resume
131 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
132
133
134On transmit:
135------------
136
137dev->hard_header != NULL
138 mac_header -> ll header
139 data -> ll header
140
141dev->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 */
154struct 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
161union tpacket_uhdr {
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
165 void *raw;
166};
167
168static 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
188struct packet_sock;
189static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
190static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 struct packet_type *pt, struct net_device *orig_dev);
192
193static void *packet_previous_frame(struct packet_sock *po,
194 struct packet_ring_buffer *rb,
195 int status);
196static void packet_increment_head(struct packet_ring_buffer *buff);
197static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
198 struct tpacket_block_desc *);
199static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203static int prb_queue_frozen(struct tpacket_kbdq_core *);
204static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206static void prb_retire_rx_blk_timer_expired(unsigned long);
207static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208static void prb_init_blk_timer(struct packet_sock *,
209 struct tpacket_kbdq_core *,
210 void (*func) (unsigned long));
211static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216static void packet_flush_mclist(struct sock *sk);
217
218struct 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
237static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
238static void __fanout_link(struct sock *sk, struct packet_sock *po);
239
240/* register_prot_hook must be invoked with the po->bind_lock held,
241 * or from a context in which asynchronous accesses to the packet
242 * socket is not possible (packet_create()).
243 */
244static void register_prot_hook(struct sock *sk)
245{
246 struct packet_sock *po = pkt_sk(sk);
247 if (!po->running) {
248 if (po->fanout)
249 __fanout_link(sk, po);
250 else
251 dev_add_pack(&po->prot_hook);
252 sock_hold(sk);
253 po->running = 1;
254 }
255}
256
257/* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
258 * held. If the sync parameter is true, we will temporarily drop
259 * the po->bind_lock and do a synchronize_net to make sure no
260 * asynchronous packet processing paths still refer to the elements
261 * of po->prot_hook. If the sync parameter is false, it is the
262 * callers responsibility to take care of this.
263 */
264static void __unregister_prot_hook(struct sock *sk, bool sync)
265{
266 struct packet_sock *po = pkt_sk(sk);
267
268 po->running = 0;
269 if (po->fanout)
270 __fanout_unlink(sk, po);
271 else
272 __dev_remove_pack(&po->prot_hook);
273 __sock_put(sk);
274
275 if (sync) {
276 spin_unlock(&po->bind_lock);
277 synchronize_net();
278 spin_lock(&po->bind_lock);
279 }
280}
281
282static void unregister_prot_hook(struct sock *sk, bool sync)
283{
284 struct packet_sock *po = pkt_sk(sk);
285
286 if (po->running)
287 __unregister_prot_hook(sk, sync);
288}
289
290static inline __pure struct page *pgv_to_page(void *addr)
291{
292 if (is_vmalloc_addr(addr))
293 return vmalloc_to_page(addr);
294 return virt_to_page(addr);
295}
296
297static void __packet_set_status(struct packet_sock *po, void *frame, int status)
298{
299 union tpacket_uhdr h;
300
301 h.raw = frame;
302 switch (po->tp_version) {
303 case TPACKET_V1:
304 h.h1->tp_status = status;
305 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
306 break;
307 case TPACKET_V2:
308 h.h2->tp_status = status;
309 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
310 break;
311 case TPACKET_V3:
312 default:
313 WARN(1, "TPACKET version not supported.\n");
314 BUG();
315 }
316
317 smp_wmb();
318}
319
320static int __packet_get_status(struct packet_sock *po, void *frame)
321{
322 union tpacket_uhdr h;
323
324 smp_rmb();
325
326 h.raw = frame;
327 switch (po->tp_version) {
328 case TPACKET_V1:
329 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
330 return h.h1->tp_status;
331 case TPACKET_V2:
332 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
333 return h.h2->tp_status;
334 case TPACKET_V3:
335 default:
336 WARN(1, "TPACKET version not supported.\n");
337 BUG();
338 return 0;
339 }
340}
341
342static void *packet_lookup_frame(struct packet_sock *po,
343 struct packet_ring_buffer *rb,
344 unsigned int position,
345 int status)
346{
347 unsigned int pg_vec_pos, frame_offset;
348 union tpacket_uhdr h;
349
350 pg_vec_pos = position / rb->frames_per_block;
351 frame_offset = position % rb->frames_per_block;
352
353 h.raw = rb->pg_vec[pg_vec_pos].buffer +
354 (frame_offset * rb->frame_size);
355
356 if (status != __packet_get_status(po, h.raw))
357 return NULL;
358
359 return h.raw;
360}
361
362static void *packet_current_frame(struct packet_sock *po,
363 struct packet_ring_buffer *rb,
364 int status)
365{
366 return packet_lookup_frame(po, rb, rb->head, status);
367}
368
369static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
370{
371 del_timer_sync(&pkc->retire_blk_timer);
372}
373
374static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
375 int tx_ring,
376 struct sk_buff_head *rb_queue)
377{
378 struct tpacket_kbdq_core *pkc;
379
380 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
381
382 spin_lock(&rb_queue->lock);
383 pkc->delete_blk_timer = 1;
384 spin_unlock(&rb_queue->lock);
385
386 prb_del_retire_blk_timer(pkc);
387}
388
389static void prb_init_blk_timer(struct packet_sock *po,
390 struct tpacket_kbdq_core *pkc,
391 void (*func) (unsigned long))
392{
393 init_timer(&pkc->retire_blk_timer);
394 pkc->retire_blk_timer.data = (long)po;
395 pkc->retire_blk_timer.function = func;
396 pkc->retire_blk_timer.expires = jiffies;
397}
398
399static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
400{
401 struct tpacket_kbdq_core *pkc;
402
403 if (tx_ring)
404 BUG();
405
406 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
407 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
408}
409
410static int prb_calc_retire_blk_tmo(struct packet_sock *po,
411 int blk_size_in_bytes)
412{
413 struct net_device *dev;
414 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
415 struct ethtool_cmd ecmd;
416 int err;
417 u32 speed;
418
419 rtnl_lock();
420 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
421 if (unlikely(!dev)) {
422 rtnl_unlock();
423 return DEFAULT_PRB_RETIRE_TOV;
424 }
425 err = __ethtool_get_settings(dev, &ecmd);
426 speed = ethtool_cmd_speed(&ecmd);
427 rtnl_unlock();
428 if (!err) {
429 /*
430 * If the link speed is so slow you don't really
431 * need to worry about perf anyways
432 */
433 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
434 return DEFAULT_PRB_RETIRE_TOV;
435 } else {
436 msec = 1;
437 div = speed / 1000;
438 }
439 }
440
441 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
442
443 if (div)
444 mbits /= div;
445
446 tmo = mbits * msec;
447
448 if (div)
449 return tmo+1;
450 return tmo;
451}
452
453static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
454 union tpacket_req_u *req_u)
455{
456 p1->feature_req_word = req_u->req3.tp_feature_req_word;
457}
458
459static void init_prb_bdqc(struct packet_sock *po,
460 struct packet_ring_buffer *rb,
461 struct pgv *pg_vec,
462 union tpacket_req_u *req_u, int tx_ring)
463{
464 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
465 struct tpacket_block_desc *pbd;
466
467 memset(p1, 0x0, sizeof(*p1));
468
469 p1->knxt_seq_num = 1;
470 p1->pkbdq = pg_vec;
471 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
472 p1->pkblk_start = pg_vec[0].buffer;
473 p1->kblk_size = req_u->req3.tp_block_size;
474 p1->knum_blocks = req_u->req3.tp_block_nr;
475 p1->hdrlen = po->tp_hdrlen;
476 p1->version = po->tp_version;
477 p1->last_kactive_blk_num = 0;
478 po->stats_u.stats3.tp_freeze_q_cnt = 0;
479 if (req_u->req3.tp_retire_blk_tov)
480 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
481 else
482 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
483 req_u->req3.tp_block_size);
484 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
485 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
486
487 prb_init_ft_ops(p1, req_u);
488 prb_setup_retire_blk_timer(po, tx_ring);
489 prb_open_block(p1, pbd);
490}
491
492/* Do NOT update the last_blk_num first.
493 * Assumes sk_buff_head lock is held.
494 */
495static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
496{
497 mod_timer(&pkc->retire_blk_timer,
498 jiffies + pkc->tov_in_jiffies);
499 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
500}
501
502/*
503 * Timer logic:
504 * 1) We refresh the timer only when we open a block.
505 * By doing this we don't waste cycles refreshing the timer
506 * on packet-by-packet basis.
507 *
508 * With a 1MB block-size, on a 1Gbps line, it will take
509 * i) ~8 ms to fill a block + ii) memcpy etc.
510 * In this cut we are not accounting for the memcpy time.
511 *
512 * So, if the user sets the 'tmo' to 10ms then the timer
513 * will never fire while the block is still getting filled
514 * (which is what we want). However, the user could choose
515 * to close a block early and that's fine.
516 *
517 * But when the timer does fire, we check whether or not to refresh it.
518 * Since the tmo granularity is in msecs, it is not too expensive
519 * to refresh the timer, lets say every '8' msecs.
520 * Either the user can set the 'tmo' or we can derive it based on
521 * a) line-speed and b) block-size.
522 * prb_calc_retire_blk_tmo() calculates the tmo.
523 *
524 */
525static void prb_retire_rx_blk_timer_expired(unsigned long data)
526{
527 struct packet_sock *po = (struct packet_sock *)data;
528 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
529 unsigned int frozen;
530 struct tpacket_block_desc *pbd;
531
532 spin_lock(&po->sk.sk_receive_queue.lock);
533
534 frozen = prb_queue_frozen(pkc);
535 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
536
537 if (unlikely(pkc->delete_blk_timer))
538 goto out;
539
540 /* We only need to plug the race when the block is partially filled.
541 * tpacket_rcv:
542 * lock(); increment BLOCK_NUM_PKTS; unlock()
543 * copy_bits() is in progress ...
544 * timer fires on other cpu:
545 * we can't retire the current block because copy_bits
546 * is in progress.
547 *
548 */
549 if (BLOCK_NUM_PKTS(pbd)) {
550 while (atomic_read(&pkc->blk_fill_in_prog)) {
551 /* Waiting for skb_copy_bits to finish... */
552 cpu_relax();
553 }
554 }
555
556 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
557 if (!frozen) {
558 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
559 if (!prb_dispatch_next_block(pkc, po))
560 goto refresh_timer;
561 else
562 goto out;
563 } else {
564 /* Case 1. Queue was frozen because user-space was
565 * lagging behind.
566 */
567 if (prb_curr_blk_in_use(pkc, pbd)) {
568 /*
569 * Ok, user-space is still behind.
570 * So just refresh the timer.
571 */
572 goto refresh_timer;
573 } else {
574 /* Case 2. queue was frozen,user-space caught up,
575 * now the link went idle && the timer fired.
576 * We don't have a block to close.So we open this
577 * block and restart the timer.
578 * opening a block thaws the queue,restarts timer
579 * Thawing/timer-refresh is a side effect.
580 */
581 prb_open_block(pkc, pbd);
582 goto out;
583 }
584 }
585 }
586
587refresh_timer:
588 _prb_refresh_rx_retire_blk_timer(pkc);
589
590out:
591 spin_unlock(&po->sk.sk_receive_queue.lock);
592}
593
594static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
595 struct tpacket_block_desc *pbd1, __u32 status)
596{
597 /* Flush everything minus the block header */
598
599#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
600 u8 *start, *end;
601
602 start = (u8 *)pbd1;
603
604 /* Skip the block header(we know header WILL fit in 4K) */
605 start += PAGE_SIZE;
606
607 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
608 for (; start < end; start += PAGE_SIZE)
609 flush_dcache_page(pgv_to_page(start));
610
611 smp_wmb();
612#endif
613
614 /* Now update the block status. */
615
616 BLOCK_STATUS(pbd1) = status;
617
618 /* Flush the block header */
619
620#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
621 start = (u8 *)pbd1;
622 flush_dcache_page(pgv_to_page(start));
623
624 smp_wmb();
625#endif
626}
627
628/*
629 * Side effect:
630 *
631 * 1) flush the block
632 * 2) Increment active_blk_num
633 *
634 * Note:We DONT refresh the timer on purpose.
635 * Because almost always the next block will be opened.
636 */
637static void prb_close_block(struct tpacket_kbdq_core *pkc1,
638 struct tpacket_block_desc *pbd1,
639 struct packet_sock *po, unsigned int stat)
640{
641 __u32 status = TP_STATUS_USER | stat;
642
643 struct tpacket3_hdr *last_pkt;
644 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
645
646 if (po->stats.tp_drops)
647 status |= TP_STATUS_LOSING;
648
649 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
650 last_pkt->tp_next_offset = 0;
651
652 /* Get the ts of the last pkt */
653 if (BLOCK_NUM_PKTS(pbd1)) {
654 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
655 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
656 } else {
657 /* Ok, we tmo'd - so get the current time */
658 struct timespec ts;
659 getnstimeofday(&ts);
660 h1->ts_last_pkt.ts_sec = ts.tv_sec;
661 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
662 }
663
664 smp_wmb();
665
666 /* Flush the block */
667 prb_flush_block(pkc1, pbd1, status);
668
669 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
670}
671
672static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
673{
674 pkc->reset_pending_on_curr_blk = 0;
675}
676
677/*
678 * Side effect of opening a block:
679 *
680 * 1) prb_queue is thawed.
681 * 2) retire_blk_timer is refreshed.
682 *
683 */
684static void prb_open_block(struct tpacket_kbdq_core *pkc1,
685 struct tpacket_block_desc *pbd1)
686{
687 struct timespec ts;
688 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
689
690 smp_rmb();
691
692 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
693
694 /* We could have just memset this but we will lose the
695 * flexibility of making the priv area sticky
696 */
697 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
698 BLOCK_NUM_PKTS(pbd1) = 0;
699 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
700 getnstimeofday(&ts);
701 h1->ts_first_pkt.ts_sec = ts.tv_sec;
702 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
703 pkc1->pkblk_start = (char *)pbd1;
704 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
705 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
706 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
707 pbd1->version = pkc1->version;
708 pkc1->prev = pkc1->nxt_offset;
709 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
710 prb_thaw_queue(pkc1);
711 _prb_refresh_rx_retire_blk_timer(pkc1);
712
713 smp_wmb();
714
715 return;
716 }
717
718 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
719 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
720 dump_stack();
721 BUG();
722}
723
724/*
725 * Queue freeze logic:
726 * 1) Assume tp_block_nr = 8 blocks.
727 * 2) At time 't0', user opens Rx ring.
728 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
729 * 4) user-space is either sleeping or processing block '0'.
730 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
731 * it will close block-7,loop around and try to fill block '0'.
732 * call-flow:
733 * __packet_lookup_frame_in_block
734 * prb_retire_current_block()
735 * prb_dispatch_next_block()
736 * |->(BLOCK_STATUS == USER) evaluates to true
737 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
738 * 6) Now there are two cases:
739 * 6.1) Link goes idle right after the queue is frozen.
740 * But remember, the last open_block() refreshed the timer.
741 * When this timer expires,it will refresh itself so that we can
742 * re-open block-0 in near future.
743 * 6.2) Link is busy and keeps on receiving packets. This is a simple
744 * case and __packet_lookup_frame_in_block will check if block-0
745 * is free and can now be re-used.
746 */
747static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
748 struct packet_sock *po)
749{
750 pkc->reset_pending_on_curr_blk = 1;
751 po->stats_u.stats3.tp_freeze_q_cnt++;
752}
753
754#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
755
756/*
757 * If the next block is free then we will dispatch it
758 * and return a good offset.
759 * Else, we will freeze the queue.
760 * So, caller must check the return value.
761 */
762static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
763 struct packet_sock *po)
764{
765 struct tpacket_block_desc *pbd;
766
767 smp_rmb();
768
769 /* 1. Get current block num */
770 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
771
772 /* 2. If this block is currently in_use then freeze the queue */
773 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
774 prb_freeze_queue(pkc, po);
775 return NULL;
776 }
777
778 /*
779 * 3.
780 * open this block and return the offset where the first packet
781 * needs to get stored.
782 */
783 prb_open_block(pkc, pbd);
784 return (void *)pkc->nxt_offset;
785}
786
787static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
788 struct packet_sock *po, unsigned int status)
789{
790 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
791
792 /* retire/close the current block */
793 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
794 /*
795 * Plug the case where copy_bits() is in progress on
796 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
797 * have space to copy the pkt in the current block and
798 * called prb_retire_current_block()
799 *
800 * We don't need to worry about the TMO case because
801 * the timer-handler already handled this case.
802 */
803 if (!(status & TP_STATUS_BLK_TMO)) {
804 while (atomic_read(&pkc->blk_fill_in_prog)) {
805 /* Waiting for skb_copy_bits to finish... */
806 cpu_relax();
807 }
808 }
809 prb_close_block(pkc, pbd, po, status);
810 return;
811 }
812
813 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
814 dump_stack();
815 BUG();
816}
817
818static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
819 struct tpacket_block_desc *pbd)
820{
821 return TP_STATUS_USER & BLOCK_STATUS(pbd);
822}
823
824static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
825{
826 return pkc->reset_pending_on_curr_blk;
827}
828
829static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
830{
831 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
832 atomic_dec(&pkc->blk_fill_in_prog);
833}
834
835static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
836 struct tpacket3_hdr *ppd)
837{
838 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
839}
840
841static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
842 struct tpacket3_hdr *ppd)
843{
844 ppd->hv1.tp_rxhash = 0;
845}
846
847static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
848 struct tpacket3_hdr *ppd)
849{
850 if (vlan_tx_tag_present(pkc->skb)) {
851 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
852 ppd->tp_status = TP_STATUS_VLAN_VALID;
853 } else {
854 ppd->hv1.tp_vlan_tci = 0;
855 ppd->tp_status = TP_STATUS_AVAILABLE;
856 }
857}
858
859static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
860 struct tpacket3_hdr *ppd)
861{
862 prb_fill_vlan_info(pkc, ppd);
863
864 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
865 prb_fill_rxhash(pkc, ppd);
866 else
867 prb_clear_rxhash(pkc, ppd);
868}
869
870static void prb_fill_curr_block(char *curr,
871 struct tpacket_kbdq_core *pkc,
872 struct tpacket_block_desc *pbd,
873 unsigned int len)
874{
875 struct tpacket3_hdr *ppd;
876
877 ppd = (struct tpacket3_hdr *)curr;
878 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
879 pkc->prev = curr;
880 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
881 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
882 BLOCK_NUM_PKTS(pbd) += 1;
883 atomic_inc(&pkc->blk_fill_in_prog);
884 prb_run_all_ft_ops(pkc, ppd);
885}
886
887/* Assumes caller has the sk->rx_queue.lock */
888static void *__packet_lookup_frame_in_block(struct packet_sock *po,
889 struct sk_buff *skb,
890 int status,
891 unsigned int len
892 )
893{
894 struct tpacket_kbdq_core *pkc;
895 struct tpacket_block_desc *pbd;
896 char *curr, *end;
897
898 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
899 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
900
901 /* Queue is frozen when user space is lagging behind */
902 if (prb_queue_frozen(pkc)) {
903 /*
904 * Check if that last block which caused the queue to freeze,
905 * is still in_use by user-space.
906 */
907 if (prb_curr_blk_in_use(pkc, pbd)) {
908 /* Can't record this packet */
909 return NULL;
910 } else {
911 /*
912 * Ok, the block was released by user-space.
913 * Now let's open that block.
914 * opening a block also thaws the queue.
915 * Thawing is a side effect.
916 */
917 prb_open_block(pkc, pbd);
918 }
919 }
920
921 smp_mb();
922 curr = pkc->nxt_offset;
923 pkc->skb = skb;
924 end = (char *)pbd + pkc->kblk_size;
925
926 /* first try the current block */
927 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
928 prb_fill_curr_block(curr, pkc, pbd, len);
929 return (void *)curr;
930 }
931
932 /* Ok, close the current block */
933 prb_retire_current_block(pkc, po, 0);
934
935 /* Now, try to dispatch the next block */
936 curr = (char *)prb_dispatch_next_block(pkc, po);
937 if (curr) {
938 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
939 prb_fill_curr_block(curr, pkc, pbd, len);
940 return (void *)curr;
941 }
942
943 /*
944 * No free blocks are available.user_space hasn't caught up yet.
945 * Queue was just frozen and now this packet will get dropped.
946 */
947 return NULL;
948}
949
950static void *packet_current_rx_frame(struct packet_sock *po,
951 struct sk_buff *skb,
952 int status, unsigned int len)
953{
954 char *curr = NULL;
955 switch (po->tp_version) {
956 case TPACKET_V1:
957 case TPACKET_V2:
958 curr = packet_lookup_frame(po, &po->rx_ring,
959 po->rx_ring.head, status);
960 return curr;
961 case TPACKET_V3:
962 return __packet_lookup_frame_in_block(po, skb, status, len);
963 default:
964 WARN(1, "TPACKET version not supported\n");
965 BUG();
966 return NULL;
967 }
968}
969
970static void *prb_lookup_block(struct packet_sock *po,
971 struct packet_ring_buffer *rb,
972 unsigned int idx,
973 int status)
974{
975 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
976 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
977
978 if (status != BLOCK_STATUS(pbd))
979 return NULL;
980 return pbd;
981}
982
983static int prb_previous_blk_num(struct packet_ring_buffer *rb)
984{
985 unsigned int prev;
986 if (rb->prb_bdqc.kactive_blk_num)
987 prev = rb->prb_bdqc.kactive_blk_num-1;
988 else
989 prev = rb->prb_bdqc.knum_blocks-1;
990 return prev;
991}
992
993/* Assumes caller has held the rx_queue.lock */
994static void *__prb_previous_block(struct packet_sock *po,
995 struct packet_ring_buffer *rb,
996 int status)
997{
998 unsigned int previous = prb_previous_blk_num(rb);
999 return prb_lookup_block(po, rb, previous, status);
1000}
1001
1002static void *packet_previous_rx_frame(struct packet_sock *po,
1003 struct packet_ring_buffer *rb,
1004 int status)
1005{
1006 if (po->tp_version <= TPACKET_V2)
1007 return packet_previous_frame(po, rb, status);
1008
1009 return __prb_previous_block(po, rb, status);
1010}
1011
1012static void packet_increment_rx_head(struct packet_sock *po,
1013 struct packet_ring_buffer *rb)
1014{
1015 switch (po->tp_version) {
1016 case TPACKET_V1:
1017 case TPACKET_V2:
1018 return packet_increment_head(rb);
1019 case TPACKET_V3:
1020 default:
1021 WARN(1, "TPACKET version not supported.\n");
1022 BUG();
1023 return;
1024 }
1025}
1026
1027static void *packet_previous_frame(struct packet_sock *po,
1028 struct packet_ring_buffer *rb,
1029 int status)
1030{
1031 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1032 return packet_lookup_frame(po, rb, previous, status);
1033}
1034
1035static void packet_increment_head(struct packet_ring_buffer *buff)
1036{
1037 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1038}
1039
1040static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1041{
1042 struct sock *sk = &po->sk;
1043 bool has_room;
1044
1045 if (po->prot_hook.func != tpacket_rcv)
1046 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1047 <= sk->sk_rcvbuf;
1048
1049 spin_lock(&sk->sk_receive_queue.lock);
1050 if (po->tp_version == TPACKET_V3)
1051 has_room = prb_lookup_block(po, &po->rx_ring,
1052 po->rx_ring.prb_bdqc.kactive_blk_num,
1053 TP_STATUS_KERNEL);
1054 else
1055 has_room = packet_lookup_frame(po, &po->rx_ring,
1056 po->rx_ring.head,
1057 TP_STATUS_KERNEL);
1058 spin_unlock(&sk->sk_receive_queue.lock);
1059
1060 return has_room;
1061}
1062
1063static void packet_sock_destruct(struct sock *sk)
1064{
1065 skb_queue_purge(&sk->sk_error_queue);
1066
1067 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1068 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1069
1070 if (!sock_flag(sk, SOCK_DEAD)) {
1071 pr_err("Attempt to release alive packet socket: %p\n", sk);
1072 return;
1073 }
1074
1075 sk_refcnt_debug_dec(sk);
1076}
1077
1078static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1079{
1080 int x = atomic_read(&f->rr_cur) + 1;
1081
1082 if (x >= num)
1083 x = 0;
1084
1085 return x;
1086}
1087
1088static unsigned int fanout_demux_hash(struct packet_fanout *f,
1089 struct sk_buff *skb,
1090 unsigned int num)
1091{
1092 return (((u64)skb->rxhash) * num) >> 32;
1093}
1094
1095static unsigned int fanout_demux_lb(struct packet_fanout *f,
1096 struct sk_buff *skb,
1097 unsigned int num)
1098{
1099 int cur, old;
1100
1101 cur = atomic_read(&f->rr_cur);
1102 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1103 fanout_rr_next(f, num))) != cur)
1104 cur = old;
1105 return cur;
1106}
1107
1108static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1109 struct sk_buff *skb,
1110 unsigned int num)
1111{
1112 return smp_processor_id() % num;
1113}
1114
1115static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1116 struct sk_buff *skb,
1117 unsigned int idx, unsigned int skip,
1118 unsigned int num)
1119{
1120 unsigned int i, j;
1121
1122 i = j = min_t(int, f->next[idx], num - 1);
1123 do {
1124 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1125 if (i != j)
1126 f->next[idx] = i;
1127 return i;
1128 }
1129 if (++i == num)
1130 i = 0;
1131 } while (i != j);
1132
1133 return idx;
1134}
1135
1136static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1137{
1138 return f->flags & (flag >> 8);
1139}
1140
1141static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1142 struct packet_type *pt, struct net_device *orig_dev)
1143{
1144 struct packet_fanout *f = pt->af_packet_priv;
1145 unsigned int num = f->num_members;
1146 struct packet_sock *po;
1147 unsigned int idx;
1148
1149 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1150 !num) {
1151 kfree_skb(skb);
1152 return 0;
1153 }
1154
1155 switch (f->type) {
1156 case PACKET_FANOUT_HASH:
1157 default:
1158 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1159 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1160 if (!skb)
1161 return 0;
1162 }
1163 skb_get_rxhash(skb);
1164 idx = fanout_demux_hash(f, skb, num);
1165 break;
1166 case PACKET_FANOUT_LB:
1167 idx = fanout_demux_lb(f, skb, num);
1168 break;
1169 case PACKET_FANOUT_CPU:
1170 idx = fanout_demux_cpu(f, skb, num);
1171 break;
1172 case PACKET_FANOUT_ROLLOVER:
1173 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1174 break;
1175 }
1176
1177 po = pkt_sk(f->arr[idx]);
1178 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1179 unlikely(!packet_rcv_has_room(po, skb))) {
1180 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1181 po = pkt_sk(f->arr[idx]);
1182 }
1183
1184 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1185}
1186
1187DEFINE_MUTEX(fanout_mutex);
1188EXPORT_SYMBOL_GPL(fanout_mutex);
1189static LIST_HEAD(fanout_list);
1190
1191static void __fanout_link(struct sock *sk, struct packet_sock *po)
1192{
1193 struct packet_fanout *f = po->fanout;
1194
1195 spin_lock(&f->lock);
1196 f->arr[f->num_members] = sk;
1197 smp_wmb();
1198 f->num_members++;
1199 spin_unlock(&f->lock);
1200}
1201
1202static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1203{
1204 struct packet_fanout *f = po->fanout;
1205 int i;
1206
1207 spin_lock(&f->lock);
1208 for (i = 0; i < f->num_members; i++) {
1209 if (f->arr[i] == sk)
1210 break;
1211 }
1212 BUG_ON(i >= f->num_members);
1213 f->arr[i] = f->arr[f->num_members - 1];
1214 f->num_members--;
1215 spin_unlock(&f->lock);
1216}
1217
1218static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1219{
1220 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1221 return true;
1222
1223 return false;
1224}
1225
1226static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1227{
1228 struct packet_sock *po = pkt_sk(sk);
1229 struct packet_fanout *f, *match;
1230 u8 type = type_flags & 0xff;
1231 u8 flags = type_flags >> 8;
1232 int err;
1233
1234 switch (type) {
1235 case PACKET_FANOUT_ROLLOVER:
1236 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1237 return -EINVAL;
1238 case PACKET_FANOUT_HASH:
1239 case PACKET_FANOUT_LB:
1240 case PACKET_FANOUT_CPU:
1241 break;
1242 default:
1243 return -EINVAL;
1244 }
1245
1246 if (!po->running)
1247 return -EINVAL;
1248
1249 if (po->fanout)
1250 return -EALREADY;
1251
1252 mutex_lock(&fanout_mutex);
1253 match = NULL;
1254 list_for_each_entry(f, &fanout_list, list) {
1255 if (f->id == id &&
1256 read_pnet(&f->net) == sock_net(sk)) {
1257 match = f;
1258 break;
1259 }
1260 }
1261 err = -EINVAL;
1262 if (match && match->flags != flags)
1263 goto out;
1264 if (!match) {
1265 err = -ENOMEM;
1266 match = kzalloc(sizeof(*match), GFP_KERNEL);
1267 if (!match)
1268 goto out;
1269 write_pnet(&match->net, sock_net(sk));
1270 match->id = id;
1271 match->type = type;
1272 match->flags = flags;
1273 atomic_set(&match->rr_cur, 0);
1274 INIT_LIST_HEAD(&match->list);
1275 spin_lock_init(&match->lock);
1276 atomic_set(&match->sk_ref, 0);
1277 match->prot_hook.type = po->prot_hook.type;
1278 match->prot_hook.dev = po->prot_hook.dev;
1279 match->prot_hook.func = packet_rcv_fanout;
1280 match->prot_hook.af_packet_priv = match;
1281 match->prot_hook.id_match = match_fanout_group;
1282 dev_add_pack(&match->prot_hook);
1283 list_add(&match->list, &fanout_list);
1284 }
1285 err = -EINVAL;
1286 if (match->type == type &&
1287 match->prot_hook.type == po->prot_hook.type &&
1288 match->prot_hook.dev == po->prot_hook.dev) {
1289 err = -ENOSPC;
1290 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1291 __dev_remove_pack(&po->prot_hook);
1292 po->fanout = match;
1293 atomic_inc(&match->sk_ref);
1294 __fanout_link(sk, po);
1295 err = 0;
1296 }
1297 }
1298out:
1299 mutex_unlock(&fanout_mutex);
1300 return err;
1301}
1302
1303static void fanout_release(struct sock *sk)
1304{
1305 struct packet_sock *po = pkt_sk(sk);
1306 struct packet_fanout *f;
1307
1308 f = po->fanout;
1309 if (!f)
1310 return;
1311
1312 mutex_lock(&fanout_mutex);
1313 po->fanout = NULL;
1314
1315 if (atomic_dec_and_test(&f->sk_ref)) {
1316 list_del(&f->list);
1317 dev_remove_pack(&f->prot_hook);
1318 kfree(f);
1319 }
1320 mutex_unlock(&fanout_mutex);
1321}
1322
1323static const struct proto_ops packet_ops;
1324
1325static const struct proto_ops packet_ops_spkt;
1326
1327static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1328 struct packet_type *pt, struct net_device *orig_dev)
1329{
1330 struct sock *sk;
1331 struct sockaddr_pkt *spkt;
1332
1333 /*
1334 * When we registered the protocol we saved the socket in the data
1335 * field for just this event.
1336 */
1337
1338 sk = pt->af_packet_priv;
1339
1340 /*
1341 * Yank back the headers [hope the device set this
1342 * right or kerboom...]
1343 *
1344 * Incoming packets have ll header pulled,
1345 * push it back.
1346 *
1347 * For outgoing ones skb->data == skb_mac_header(skb)
1348 * so that this procedure is noop.
1349 */
1350
1351 if (skb->pkt_type == PACKET_LOOPBACK)
1352 goto out;
1353
1354 if (!net_eq(dev_net(dev), sock_net(sk)))
1355 goto out;
1356
1357 skb = skb_share_check(skb, GFP_ATOMIC);
1358 if (skb == NULL)
1359 goto oom;
1360
1361 /* drop any routing info */
1362 skb_dst_drop(skb);
1363
1364 /* drop conntrack reference */
1365 nf_reset(skb);
1366
1367 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1368
1369 skb_push(skb, skb->data - skb_mac_header(skb));
1370
1371 /*
1372 * The SOCK_PACKET socket receives _all_ frames.
1373 */
1374
1375 spkt->spkt_family = dev->type;
1376 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1377 spkt->spkt_protocol = skb->protocol;
1378
1379 /*
1380 * Charge the memory to the socket. This is done specifically
1381 * to prevent sockets using all the memory up.
1382 */
1383
1384 if (sock_queue_rcv_skb(sk, skb) == 0)
1385 return 0;
1386
1387out:
1388 kfree_skb(skb);
1389oom:
1390 return 0;
1391}
1392
1393
1394/*
1395 * Output a raw packet to a device layer. This bypasses all the other
1396 * protocol layers and you must therefore supply it with a complete frame
1397 */
1398
1399static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1400 struct msghdr *msg, size_t len)
1401{
1402 struct sock *sk = sock->sk;
1403 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1404 struct sk_buff *skb = NULL;
1405 struct net_device *dev;
1406 __be16 proto = 0;
1407 int err;
1408 int extra_len = 0;
1409
1410 /*
1411 * Get and verify the address.
1412 */
1413
1414 if (saddr) {
1415 if (msg->msg_namelen < sizeof(struct sockaddr))
1416 return -EINVAL;
1417 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1418 proto = saddr->spkt_protocol;
1419 } else
1420 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1421
1422 /*
1423 * Find the device first to size check it
1424 */
1425
1426 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1427retry:
1428 rcu_read_lock();
1429 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1430 err = -ENODEV;
1431 if (dev == NULL)
1432 goto out_unlock;
1433
1434 err = -ENETDOWN;
1435 if (!(dev->flags & IFF_UP))
1436 goto out_unlock;
1437
1438 /*
1439 * You may not queue a frame bigger than the mtu. This is the lowest level
1440 * raw protocol and you must do your own fragmentation at this level.
1441 */
1442
1443 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1444 if (!netif_supports_nofcs(dev)) {
1445 err = -EPROTONOSUPPORT;
1446 goto out_unlock;
1447 }
1448 extra_len = 4; /* We're doing our own CRC */
1449 }
1450
1451 err = -EMSGSIZE;
1452 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1453 goto out_unlock;
1454
1455 if (!skb) {
1456 size_t reserved = LL_RESERVED_SPACE(dev);
1457 int tlen = dev->needed_tailroom;
1458 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1459
1460 rcu_read_unlock();
1461 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1462 if (skb == NULL)
1463 return -ENOBUFS;
1464 /* FIXME: Save some space for broken drivers that write a hard
1465 * header at transmission time by themselves. PPP is the notable
1466 * one here. This should really be fixed at the driver level.
1467 */
1468 skb_reserve(skb, reserved);
1469 skb_reset_network_header(skb);
1470
1471 /* Try to align data part correctly */
1472 if (hhlen) {
1473 skb->data -= hhlen;
1474 skb->tail -= hhlen;
1475 if (len < hhlen)
1476 skb_reset_network_header(skb);
1477 }
1478 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1479 if (err)
1480 goto out_free;
1481 goto retry;
1482 }
1483
1484 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1485 /* Earlier code assumed this would be a VLAN pkt,
1486 * double-check this now that we have the actual
1487 * packet in hand.
1488 */
1489 struct ethhdr *ehdr;
1490 skb_reset_mac_header(skb);
1491 ehdr = eth_hdr(skb);
1492 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1493 err = -EMSGSIZE;
1494 goto out_unlock;
1495 }
1496 }
1497
1498 skb->protocol = proto;
1499 skb->dev = dev;
1500 skb->priority = sk->sk_priority;
1501 skb->mark = sk->sk_mark;
1502
1503 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1504
1505 if (unlikely(extra_len == 4))
1506 skb->no_fcs = 1;
1507
1508 skb_probe_transport_header(skb, 0);
1509
1510 dev_queue_xmit(skb);
1511 rcu_read_unlock();
1512 return len;
1513
1514out_unlock:
1515 rcu_read_unlock();
1516out_free:
1517 kfree_skb(skb);
1518 return err;
1519}
1520
1521static unsigned int run_filter(const struct sk_buff *skb,
1522 const struct sock *sk,
1523 unsigned int res)
1524{
1525 struct sk_filter *filter;
1526
1527 rcu_read_lock();
1528 filter = rcu_dereference(sk->sk_filter);
1529 if (filter != NULL)
1530 res = SK_RUN_FILTER(filter, skb);
1531 rcu_read_unlock();
1532
1533 return res;
1534}
1535
1536/*
1537 * This function makes lazy skb cloning in hope that most of packets
1538 * are discarded by BPF.
1539 *
1540 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1541 * and skb->cb are mangled. It works because (and until) packets
1542 * falling here are owned by current CPU. Output packets are cloned
1543 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1544 * sequencially, so that if we return skb to original state on exit,
1545 * we will not harm anyone.
1546 */
1547
1548static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1549 struct packet_type *pt, struct net_device *orig_dev)
1550{
1551 struct sock *sk;
1552 struct sockaddr_ll *sll;
1553 struct packet_sock *po;
1554 u8 *skb_head = skb->data;
1555 int skb_len = skb->len;
1556 unsigned int snaplen, res;
1557
1558 if (skb->pkt_type == PACKET_LOOPBACK)
1559 goto drop;
1560
1561 sk = pt->af_packet_priv;
1562 po = pkt_sk(sk);
1563
1564 if (!net_eq(dev_net(dev), sock_net(sk)))
1565 goto drop;
1566
1567 skb->dev = dev;
1568
1569 if (dev->header_ops) {
1570 /* The device has an explicit notion of ll header,
1571 * exported to higher levels.
1572 *
1573 * Otherwise, the device hides details of its frame
1574 * structure, so that corresponding packet head is
1575 * never delivered to user.
1576 */
1577 if (sk->sk_type != SOCK_DGRAM)
1578 skb_push(skb, skb->data - skb_mac_header(skb));
1579 else if (skb->pkt_type == PACKET_OUTGOING) {
1580 /* Special case: outgoing packets have ll header at head */
1581 skb_pull(skb, skb_network_offset(skb));
1582 }
1583 }
1584
1585 snaplen = skb->len;
1586
1587 res = run_filter(skb, sk, snaplen);
1588 if (!res)
1589 goto drop_n_restore;
1590 if (snaplen > res)
1591 snaplen = res;
1592
1593 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1594 goto drop_n_acct;
1595
1596 if (skb_shared(skb)) {
1597 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1598 if (nskb == NULL)
1599 goto drop_n_acct;
1600
1601 if (skb_head != skb->data) {
1602 skb->data = skb_head;
1603 skb->len = skb_len;
1604 }
1605 consume_skb(skb);
1606 skb = nskb;
1607 }
1608
1609 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1610 sizeof(skb->cb));
1611
1612 sll = &PACKET_SKB_CB(skb)->sa.ll;
1613 sll->sll_family = AF_PACKET;
1614 sll->sll_hatype = dev->type;
1615 sll->sll_protocol = skb->protocol;
1616 sll->sll_pkttype = skb->pkt_type;
1617 if (unlikely(po->origdev))
1618 sll->sll_ifindex = orig_dev->ifindex;
1619 else
1620 sll->sll_ifindex = dev->ifindex;
1621
1622 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1623
1624 PACKET_SKB_CB(skb)->origlen = skb->len;
1625
1626 if (pskb_trim(skb, snaplen))
1627 goto drop_n_acct;
1628
1629 skb_set_owner_r(skb, sk);
1630 skb->dev = NULL;
1631 skb_dst_drop(skb);
1632
1633 /* drop conntrack reference */
1634 nf_reset(skb);
1635
1636 spin_lock(&sk->sk_receive_queue.lock);
1637 po->stats.tp_packets++;
1638 skb->dropcount = atomic_read(&sk->sk_drops);
1639 __skb_queue_tail(&sk->sk_receive_queue, skb);
1640 spin_unlock(&sk->sk_receive_queue.lock);
1641 sk->sk_data_ready(sk, skb->len);
1642 return 0;
1643
1644drop_n_acct:
1645 spin_lock(&sk->sk_receive_queue.lock);
1646 po->stats.tp_drops++;
1647 atomic_inc(&sk->sk_drops);
1648 spin_unlock(&sk->sk_receive_queue.lock);
1649
1650drop_n_restore:
1651 if (skb_head != skb->data && skb_shared(skb)) {
1652 skb->data = skb_head;
1653 skb->len = skb_len;
1654 }
1655drop:
1656 consume_skb(skb);
1657 return 0;
1658}
1659
1660static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1661 struct packet_type *pt, struct net_device *orig_dev)
1662{
1663 struct sock *sk;
1664 struct packet_sock *po;
1665 struct sockaddr_ll *sll;
1666 union tpacket_uhdr h;
1667 u8 *skb_head = skb->data;
1668 int skb_len = skb->len;
1669 unsigned int snaplen, res;
1670 unsigned long status = TP_STATUS_USER;
1671 unsigned short macoff, netoff, hdrlen;
1672 struct sk_buff *copy_skb = NULL;
1673 struct timeval tv;
1674 struct timespec ts;
1675 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1676
1677 if (skb->pkt_type == PACKET_LOOPBACK)
1678 goto drop;
1679
1680 sk = pt->af_packet_priv;
1681 po = pkt_sk(sk);
1682
1683 if (!net_eq(dev_net(dev), sock_net(sk)))
1684 goto drop;
1685
1686 if (dev->header_ops) {
1687 if (sk->sk_type != SOCK_DGRAM)
1688 skb_push(skb, skb->data - skb_mac_header(skb));
1689 else if (skb->pkt_type == PACKET_OUTGOING) {
1690 /* Special case: outgoing packets have ll header at head */
1691 skb_pull(skb, skb_network_offset(skb));
1692 }
1693 }
1694
1695 if (skb->ip_summed == CHECKSUM_PARTIAL)
1696 status |= TP_STATUS_CSUMNOTREADY;
1697
1698 snaplen = skb->len;
1699
1700 res = run_filter(skb, sk, snaplen);
1701 if (!res)
1702 goto drop_n_restore;
1703 if (snaplen > res)
1704 snaplen = res;
1705
1706 if (sk->sk_type == SOCK_DGRAM) {
1707 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1708 po->tp_reserve;
1709 } else {
1710 unsigned int maclen = skb_network_offset(skb);
1711 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1712 (maclen < 16 ? 16 : maclen)) +
1713 po->tp_reserve;
1714 macoff = netoff - maclen;
1715 }
1716 if (po->tp_version <= TPACKET_V2) {
1717 if (macoff + snaplen > po->rx_ring.frame_size) {
1718 if (po->copy_thresh &&
1719 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1720 if (skb_shared(skb)) {
1721 copy_skb = skb_clone(skb, GFP_ATOMIC);
1722 } else {
1723 copy_skb = skb_get(skb);
1724 skb_head = skb->data;
1725 }
1726 if (copy_skb)
1727 skb_set_owner_r(copy_skb, sk);
1728 }
1729 snaplen = po->rx_ring.frame_size - macoff;
1730 if ((int)snaplen < 0)
1731 snaplen = 0;
1732 }
1733 }
1734 spin_lock(&sk->sk_receive_queue.lock);
1735 h.raw = packet_current_rx_frame(po, skb,
1736 TP_STATUS_KERNEL, (macoff+snaplen));
1737 if (!h.raw)
1738 goto ring_is_full;
1739 if (po->tp_version <= TPACKET_V2) {
1740 packet_increment_rx_head(po, &po->rx_ring);
1741 /*
1742 * LOSING will be reported till you read the stats,
1743 * because it's COR - Clear On Read.
1744 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1745 * at packet level.
1746 */
1747 if (po->stats.tp_drops)
1748 status |= TP_STATUS_LOSING;
1749 }
1750 po->stats.tp_packets++;
1751 if (copy_skb) {
1752 status |= TP_STATUS_COPY;
1753 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1754 }
1755 spin_unlock(&sk->sk_receive_queue.lock);
1756
1757 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1758
1759 switch (po->tp_version) {
1760 case TPACKET_V1:
1761 h.h1->tp_len = skb->len;
1762 h.h1->tp_snaplen = snaplen;
1763 h.h1->tp_mac = macoff;
1764 h.h1->tp_net = netoff;
1765 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1766 && shhwtstamps->syststamp.tv64)
1767 tv = ktime_to_timeval(shhwtstamps->syststamp);
1768 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1769 && shhwtstamps->hwtstamp.tv64)
1770 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1771 else if (skb->tstamp.tv64)
1772 tv = ktime_to_timeval(skb->tstamp);
1773 else
1774 do_gettimeofday(&tv);
1775 h.h1->tp_sec = tv.tv_sec;
1776 h.h1->tp_usec = tv.tv_usec;
1777 hdrlen = sizeof(*h.h1);
1778 break;
1779 case TPACKET_V2:
1780 h.h2->tp_len = skb->len;
1781 h.h2->tp_snaplen = snaplen;
1782 h.h2->tp_mac = macoff;
1783 h.h2->tp_net = netoff;
1784 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1785 && shhwtstamps->syststamp.tv64)
1786 ts = ktime_to_timespec(shhwtstamps->syststamp);
1787 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1788 && shhwtstamps->hwtstamp.tv64)
1789 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1790 else if (skb->tstamp.tv64)
1791 ts = ktime_to_timespec(skb->tstamp);
1792 else
1793 getnstimeofday(&ts);
1794 h.h2->tp_sec = ts.tv_sec;
1795 h.h2->tp_nsec = ts.tv_nsec;
1796 if (vlan_tx_tag_present(skb)) {
1797 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1798 status |= TP_STATUS_VLAN_VALID;
1799 } else {
1800 h.h2->tp_vlan_tci = 0;
1801 }
1802 h.h2->tp_padding = 0;
1803 hdrlen = sizeof(*h.h2);
1804 break;
1805 case TPACKET_V3:
1806 /* tp_nxt_offset,vlan are already populated above.
1807 * So DONT clear those fields here
1808 */
1809 h.h3->tp_status |= status;
1810 h.h3->tp_len = skb->len;
1811 h.h3->tp_snaplen = snaplen;
1812 h.h3->tp_mac = macoff;
1813 h.h3->tp_net = netoff;
1814 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1815 && shhwtstamps->syststamp.tv64)
1816 ts = ktime_to_timespec(shhwtstamps->syststamp);
1817 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1818 && shhwtstamps->hwtstamp.tv64)
1819 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1820 else if (skb->tstamp.tv64)
1821 ts = ktime_to_timespec(skb->tstamp);
1822 else
1823 getnstimeofday(&ts);
1824 h.h3->tp_sec = ts.tv_sec;
1825 h.h3->tp_nsec = ts.tv_nsec;
1826 hdrlen = sizeof(*h.h3);
1827 break;
1828 default:
1829 BUG();
1830 }
1831
1832 sll = h.raw + TPACKET_ALIGN(hdrlen);
1833 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1834 sll->sll_family = AF_PACKET;
1835 sll->sll_hatype = dev->type;
1836 sll->sll_protocol = skb->protocol;
1837 sll->sll_pkttype = skb->pkt_type;
1838 if (unlikely(po->origdev))
1839 sll->sll_ifindex = orig_dev->ifindex;
1840 else
1841 sll->sll_ifindex = dev->ifindex;
1842
1843 smp_mb();
1844#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1845 {
1846 u8 *start, *end;
1847
1848 if (po->tp_version <= TPACKET_V2) {
1849 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1850 + macoff + snaplen);
1851 for (start = h.raw; start < end; start += PAGE_SIZE)
1852 flush_dcache_page(pgv_to_page(start));
1853 }
1854 smp_wmb();
1855 }
1856#endif
1857 if (po->tp_version <= TPACKET_V2)
1858 __packet_set_status(po, h.raw, status);
1859 else
1860 prb_clear_blk_fill_status(&po->rx_ring);
1861
1862 sk->sk_data_ready(sk, 0);
1863
1864drop_n_restore:
1865 if (skb_head != skb->data && skb_shared(skb)) {
1866 skb->data = skb_head;
1867 skb->len = skb_len;
1868 }
1869drop:
1870 kfree_skb(skb);
1871 return 0;
1872
1873ring_is_full:
1874 po->stats.tp_drops++;
1875 spin_unlock(&sk->sk_receive_queue.lock);
1876
1877 sk->sk_data_ready(sk, 0);
1878 kfree_skb(copy_skb);
1879 goto drop_n_restore;
1880}
1881
1882static void tpacket_destruct_skb(struct sk_buff *skb)
1883{
1884 struct packet_sock *po = pkt_sk(skb->sk);
1885 void *ph;
1886
1887 if (likely(po->tx_ring.pg_vec)) {
1888 ph = skb_shinfo(skb)->destructor_arg;
1889 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1890 atomic_dec(&po->tx_ring.pending);
1891 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1892 }
1893
1894 sock_wfree(skb);
1895}
1896
1897static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1898 void *frame, struct net_device *dev, int size_max,
1899 __be16 proto, unsigned char *addr, int hlen)
1900{
1901 union tpacket_uhdr ph;
1902 int to_write, offset, len, tp_len, nr_frags, len_max;
1903 struct socket *sock = po->sk.sk_socket;
1904 struct page *page;
1905 void *data;
1906 int err;
1907
1908 ph.raw = frame;
1909
1910 skb->protocol = proto;
1911 skb->dev = dev;
1912 skb->priority = po->sk.sk_priority;
1913 skb->mark = po->sk.sk_mark;
1914 skb_shinfo(skb)->destructor_arg = ph.raw;
1915
1916 switch (po->tp_version) {
1917 case TPACKET_V2:
1918 tp_len = ph.h2->tp_len;
1919 break;
1920 default:
1921 tp_len = ph.h1->tp_len;
1922 break;
1923 }
1924 if (unlikely(tp_len > size_max)) {
1925 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1926 return -EMSGSIZE;
1927 }
1928
1929 skb_reserve(skb, hlen);
1930 skb_reset_network_header(skb);
1931 skb_probe_transport_header(skb, 0);
1932
1933 if (po->tp_tx_has_off) {
1934 int off_min, off_max, off;
1935 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1936 off_max = po->tx_ring.frame_size - tp_len;
1937 if (sock->type == SOCK_DGRAM) {
1938 switch (po->tp_version) {
1939 case TPACKET_V2:
1940 off = ph.h2->tp_net;
1941 break;
1942 default:
1943 off = ph.h1->tp_net;
1944 break;
1945 }
1946 } else {
1947 switch (po->tp_version) {
1948 case TPACKET_V2:
1949 off = ph.h2->tp_mac;
1950 break;
1951 default:
1952 off = ph.h1->tp_mac;
1953 break;
1954 }
1955 }
1956 if (unlikely((off < off_min) || (off_max < off)))
1957 return -EINVAL;
1958 data = ph.raw + off;
1959 } else {
1960 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1961 }
1962 to_write = tp_len;
1963
1964 if (sock->type == SOCK_DGRAM) {
1965 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1966 NULL, tp_len);
1967 if (unlikely(err < 0))
1968 return -EINVAL;
1969 } else if (dev->hard_header_len) {
1970 /* net device doesn't like empty head */
1971 if (unlikely(tp_len <= dev->hard_header_len)) {
1972 pr_err("packet size is too short (%d < %d)\n",
1973 tp_len, dev->hard_header_len);
1974 return -EINVAL;
1975 }
1976
1977 skb_push(skb, dev->hard_header_len);
1978 err = skb_store_bits(skb, 0, data,
1979 dev->hard_header_len);
1980 if (unlikely(err))
1981 return err;
1982
1983 data += dev->hard_header_len;
1984 to_write -= dev->hard_header_len;
1985 }
1986
1987 offset = offset_in_page(data);
1988 len_max = PAGE_SIZE - offset;
1989 len = ((to_write > len_max) ? len_max : to_write);
1990
1991 skb->data_len = to_write;
1992 skb->len += to_write;
1993 skb->truesize += to_write;
1994 atomic_add(to_write, &po->sk.sk_wmem_alloc);
1995
1996 while (likely(to_write)) {
1997 nr_frags = skb_shinfo(skb)->nr_frags;
1998
1999 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2000 pr_err("Packet exceed the number of skb frags(%lu)\n",
2001 MAX_SKB_FRAGS);
2002 return -EFAULT;
2003 }
2004
2005 page = pgv_to_page(data);
2006 data += len;
2007 flush_dcache_page(page);
2008 get_page(page);
2009 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2010 to_write -= len;
2011 offset = 0;
2012 len_max = PAGE_SIZE;
2013 len = ((to_write > len_max) ? len_max : to_write);
2014 }
2015
2016 return tp_len;
2017}
2018
2019static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2020{
2021 struct sk_buff *skb;
2022 struct net_device *dev;
2023 __be16 proto;
2024 bool need_rls_dev = false;
2025 int err, reserve = 0;
2026 void *ph;
2027 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2028 int tp_len, size_max;
2029 unsigned char *addr;
2030 int len_sum = 0;
2031 int status = TP_STATUS_AVAILABLE;
2032 int hlen, tlen;
2033
2034 mutex_lock(&po->pg_vec_lock);
2035
2036 if (saddr == NULL) {
2037 dev = po->prot_hook.dev;
2038 proto = po->num;
2039 addr = NULL;
2040 } else {
2041 err = -EINVAL;
2042 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2043 goto out;
2044 if (msg->msg_namelen < (saddr->sll_halen
2045 + offsetof(struct sockaddr_ll,
2046 sll_addr)))
2047 goto out;
2048 proto = saddr->sll_protocol;
2049 addr = saddr->sll_addr;
2050 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2051 need_rls_dev = true;
2052 }
2053
2054 err = -ENXIO;
2055 if (unlikely(dev == NULL))
2056 goto out;
2057
2058 reserve = dev->hard_header_len;
2059
2060 err = -ENETDOWN;
2061 if (unlikely(!(dev->flags & IFF_UP)))
2062 goto out_put;
2063
2064 size_max = po->tx_ring.frame_size
2065 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2066
2067 if (size_max > dev->mtu + reserve)
2068 size_max = dev->mtu + reserve;
2069
2070 do {
2071 ph = packet_current_frame(po, &po->tx_ring,
2072 TP_STATUS_SEND_REQUEST);
2073
2074 if (unlikely(ph == NULL)) {
2075 schedule();
2076 continue;
2077 }
2078
2079 status = TP_STATUS_SEND_REQUEST;
2080 hlen = LL_RESERVED_SPACE(dev);
2081 tlen = dev->needed_tailroom;
2082 skb = sock_alloc_send_skb(&po->sk,
2083 hlen + tlen + sizeof(struct sockaddr_ll),
2084 0, &err);
2085
2086 if (unlikely(skb == NULL))
2087 goto out_status;
2088
2089 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2090 addr, hlen);
2091
2092 if (unlikely(tp_len < 0)) {
2093 if (po->tp_loss) {
2094 __packet_set_status(po, ph,
2095 TP_STATUS_AVAILABLE);
2096 packet_increment_head(&po->tx_ring);
2097 kfree_skb(skb);
2098 continue;
2099 } else {
2100 status = TP_STATUS_WRONG_FORMAT;
2101 err = tp_len;
2102 goto out_status;
2103 }
2104 }
2105
2106 skb->destructor = tpacket_destruct_skb;
2107 __packet_set_status(po, ph, TP_STATUS_SENDING);
2108 atomic_inc(&po->tx_ring.pending);
2109
2110 status = TP_STATUS_SEND_REQUEST;
2111 err = dev_queue_xmit(skb);
2112 if (unlikely(err > 0)) {
2113 err = net_xmit_errno(err);
2114 if (err && __packet_get_status(po, ph) ==
2115 TP_STATUS_AVAILABLE) {
2116 /* skb was destructed already */
2117 skb = NULL;
2118 goto out_status;
2119 }
2120 /*
2121 * skb was dropped but not destructed yet;
2122 * let's treat it like congestion or err < 0
2123 */
2124 err = 0;
2125 }
2126 packet_increment_head(&po->tx_ring);
2127 len_sum += tp_len;
2128 } while (likely((ph != NULL) ||
2129 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2130 (atomic_read(&po->tx_ring.pending))))
2131 );
2132
2133 err = len_sum;
2134 goto out_put;
2135
2136out_status:
2137 __packet_set_status(po, ph, status);
2138 kfree_skb(skb);
2139out_put:
2140 if (need_rls_dev)
2141 dev_put(dev);
2142out:
2143 mutex_unlock(&po->pg_vec_lock);
2144 return err;
2145}
2146
2147static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2148 size_t reserve, size_t len,
2149 size_t linear, int noblock,
2150 int *err)
2151{
2152 struct sk_buff *skb;
2153
2154 /* Under a page? Don't bother with paged skb. */
2155 if (prepad + len < PAGE_SIZE || !linear)
2156 linear = len;
2157
2158 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2159 err);
2160 if (!skb)
2161 return NULL;
2162
2163 skb_reserve(skb, reserve);
2164 skb_put(skb, linear);
2165 skb->data_len = len - linear;
2166 skb->len += len - linear;
2167
2168 return skb;
2169}
2170
2171static int packet_snd(struct socket *sock,
2172 struct msghdr *msg, size_t len)
2173{
2174 struct sock *sk = sock->sk;
2175 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2176 struct sk_buff *skb;
2177 struct net_device *dev;
2178 __be16 proto;
2179 bool need_rls_dev = false;
2180 unsigned char *addr;
2181 int err, reserve = 0;
2182 struct virtio_net_hdr vnet_hdr = { 0 };
2183 int offset = 0;
2184 int vnet_hdr_len;
2185 struct packet_sock *po = pkt_sk(sk);
2186 unsigned short gso_type = 0;
2187 int hlen, tlen;
2188 int extra_len = 0;
2189
2190 /*
2191 * Get and verify the address.
2192 */
2193
2194 if (saddr == NULL) {
2195 dev = po->prot_hook.dev;
2196 proto = po->num;
2197 addr = NULL;
2198 } else {
2199 err = -EINVAL;
2200 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2201 goto out;
2202 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2203 goto out;
2204 proto = saddr->sll_protocol;
2205 addr = saddr->sll_addr;
2206 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2207 need_rls_dev = true;
2208 }
2209
2210 err = -ENXIO;
2211 if (dev == NULL)
2212 goto out_unlock;
2213 if (sock->type == SOCK_RAW)
2214 reserve = dev->hard_header_len;
2215
2216 err = -ENETDOWN;
2217 if (!(dev->flags & IFF_UP))
2218 goto out_unlock;
2219
2220 if (po->has_vnet_hdr) {
2221 vnet_hdr_len = sizeof(vnet_hdr);
2222
2223 err = -EINVAL;
2224 if (len < vnet_hdr_len)
2225 goto out_unlock;
2226
2227 len -= vnet_hdr_len;
2228
2229 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2230 vnet_hdr_len);
2231 if (err < 0)
2232 goto out_unlock;
2233
2234 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2235 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2236 vnet_hdr.hdr_len))
2237 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2238 vnet_hdr.csum_offset + 2;
2239
2240 err = -EINVAL;
2241 if (vnet_hdr.hdr_len > len)
2242 goto out_unlock;
2243
2244 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2245 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2246 case VIRTIO_NET_HDR_GSO_TCPV4:
2247 gso_type = SKB_GSO_TCPV4;
2248 break;
2249 case VIRTIO_NET_HDR_GSO_TCPV6:
2250 gso_type = SKB_GSO_TCPV6;
2251 break;
2252 case VIRTIO_NET_HDR_GSO_UDP:
2253 gso_type = SKB_GSO_UDP;
2254 break;
2255 default:
2256 goto out_unlock;
2257 }
2258
2259 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2260 gso_type |= SKB_GSO_TCP_ECN;
2261
2262 if (vnet_hdr.gso_size == 0)
2263 goto out_unlock;
2264
2265 }
2266 }
2267
2268 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2269 if (!netif_supports_nofcs(dev)) {
2270 err = -EPROTONOSUPPORT;
2271 goto out_unlock;
2272 }
2273 extra_len = 4; /* We're doing our own CRC */
2274 }
2275
2276 err = -EMSGSIZE;
2277 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2278 goto out_unlock;
2279
2280 err = -ENOBUFS;
2281 hlen = LL_RESERVED_SPACE(dev);
2282 tlen = dev->needed_tailroom;
2283 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2284 msg->msg_flags & MSG_DONTWAIT, &err);
2285 if (skb == NULL)
2286 goto out_unlock;
2287
2288 skb_set_network_header(skb, reserve);
2289
2290 err = -EINVAL;
2291 if (sock->type == SOCK_DGRAM &&
2292 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2293 goto out_free;
2294
2295 /* Returns -EFAULT on error */
2296 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2297 if (err)
2298 goto out_free;
2299
2300 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2301
2302 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2303 /* Earlier code assumed this would be a VLAN pkt,
2304 * double-check this now that we have the actual
2305 * packet in hand.
2306 */
2307 struct ethhdr *ehdr;
2308 skb_reset_mac_header(skb);
2309 ehdr = eth_hdr(skb);
2310 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2311 err = -EMSGSIZE;
2312 goto out_free;
2313 }
2314 }
2315
2316 skb->protocol = proto;
2317 skb->dev = dev;
2318 skb->priority = sk->sk_priority;
2319 skb->mark = sk->sk_mark;
2320
2321 if (po->has_vnet_hdr) {
2322 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2323 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2324 vnet_hdr.csum_offset)) {
2325 err = -EINVAL;
2326 goto out_free;
2327 }
2328 }
2329
2330 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2331 skb_shinfo(skb)->gso_type = gso_type;
2332
2333 /* Header must be checked, and gso_segs computed. */
2334 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2335 skb_shinfo(skb)->gso_segs = 0;
2336
2337 len += vnet_hdr_len;
2338 }
2339
2340 skb_probe_transport_header(skb, reserve);
2341
2342 if (unlikely(extra_len == 4))
2343 skb->no_fcs = 1;
2344
2345 /*
2346 * Now send it
2347 */
2348
2349 err = dev_queue_xmit(skb);
2350 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2351 goto out_unlock;
2352
2353 if (need_rls_dev)
2354 dev_put(dev);
2355
2356 return len;
2357
2358out_free:
2359 kfree_skb(skb);
2360out_unlock:
2361 if (dev && need_rls_dev)
2362 dev_put(dev);
2363out:
2364 return err;
2365}
2366
2367static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2368 struct msghdr *msg, size_t len)
2369{
2370 struct sock *sk = sock->sk;
2371 struct packet_sock *po = pkt_sk(sk);
2372 if (po->tx_ring.pg_vec)
2373 return tpacket_snd(po, msg);
2374 else
2375 return packet_snd(sock, msg, len);
2376}
2377
2378/*
2379 * Close a PACKET socket. This is fairly simple. We immediately go
2380 * to 'closed' state and remove our protocol entry in the device list.
2381 */
2382
2383static int packet_release(struct socket *sock)
2384{
2385 struct sock *sk = sock->sk;
2386 struct packet_sock *po;
2387 struct net *net;
2388 union tpacket_req_u req_u;
2389
2390 if (!sk)
2391 return 0;
2392
2393 net = sock_net(sk);
2394 po = pkt_sk(sk);
2395
2396 mutex_lock(&net->packet.sklist_lock);
2397 sk_del_node_init_rcu(sk);
2398 mutex_unlock(&net->packet.sklist_lock);
2399
2400 preempt_disable();
2401 sock_prot_inuse_add(net, sk->sk_prot, -1);
2402 preempt_enable();
2403
2404 spin_lock(&po->bind_lock);
2405 unregister_prot_hook(sk, false);
2406 if (po->prot_hook.dev) {
2407 dev_put(po->prot_hook.dev);
2408 po->prot_hook.dev = NULL;
2409 }
2410 spin_unlock(&po->bind_lock);
2411
2412 packet_flush_mclist(sk);
2413
2414 if (po->rx_ring.pg_vec) {
2415 memset(&req_u, 0, sizeof(req_u));
2416 packet_set_ring(sk, &req_u, 1, 0);
2417 }
2418
2419 if (po->tx_ring.pg_vec) {
2420 memset(&req_u, 0, sizeof(req_u));
2421 packet_set_ring(sk, &req_u, 1, 1);
2422 }
2423
2424 fanout_release(sk);
2425
2426 synchronize_net();
2427 /*
2428 * Now the socket is dead. No more input will appear.
2429 */
2430 sock_orphan(sk);
2431 sock->sk = NULL;
2432
2433 /* Purge queues */
2434
2435 skb_queue_purge(&sk->sk_receive_queue);
2436 sk_refcnt_debug_release(sk);
2437
2438 sock_put(sk);
2439 return 0;
2440}
2441
2442/*
2443 * Attach a packet hook.
2444 */
2445
2446static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2447{
2448 struct packet_sock *po = pkt_sk(sk);
2449
2450 if (po->fanout) {
2451 if (dev)
2452 dev_put(dev);
2453
2454 return -EINVAL;
2455 }
2456
2457 lock_sock(sk);
2458
2459 spin_lock(&po->bind_lock);
2460 unregister_prot_hook(sk, true);
2461 po->num = protocol;
2462 po->prot_hook.type = protocol;
2463 if (po->prot_hook.dev)
2464 dev_put(po->prot_hook.dev);
2465 po->prot_hook.dev = dev;
2466
2467 po->ifindex = dev ? dev->ifindex : 0;
2468
2469 if (protocol == 0)
2470 goto out_unlock;
2471
2472 if (!dev || (dev->flags & IFF_UP)) {
2473 register_prot_hook(sk);
2474 } else {
2475 sk->sk_err = ENETDOWN;
2476 if (!sock_flag(sk, SOCK_DEAD))
2477 sk->sk_error_report(sk);
2478 }
2479
2480out_unlock:
2481 spin_unlock(&po->bind_lock);
2482 release_sock(sk);
2483 return 0;
2484}
2485
2486/*
2487 * Bind a packet socket to a device
2488 */
2489
2490static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2491 int addr_len)
2492{
2493 struct sock *sk = sock->sk;
2494 char name[15];
2495 struct net_device *dev;
2496 int err = -ENODEV;
2497
2498 /*
2499 * Check legality
2500 */
2501
2502 if (addr_len != sizeof(struct sockaddr))
2503 return -EINVAL;
2504 strlcpy(name, uaddr->sa_data, sizeof(name));
2505
2506 dev = dev_get_by_name(sock_net(sk), name);
2507 if (dev)
2508 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2509 return err;
2510}
2511
2512static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2513{
2514 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2515 struct sock *sk = sock->sk;
2516 struct net_device *dev = NULL;
2517 int err;
2518
2519
2520 /*
2521 * Check legality
2522 */
2523
2524 if (addr_len < sizeof(struct sockaddr_ll))
2525 return -EINVAL;
2526 if (sll->sll_family != AF_PACKET)
2527 return -EINVAL;
2528
2529 if (sll->sll_ifindex) {
2530 err = -ENODEV;
2531 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2532 if (dev == NULL)
2533 goto out;
2534 }
2535 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2536
2537out:
2538 return err;
2539}
2540
2541static struct proto packet_proto = {
2542 .name = "PACKET",
2543 .owner = THIS_MODULE,
2544 .obj_size = sizeof(struct packet_sock),
2545};
2546
2547/*
2548 * Create a packet of type SOCK_PACKET.
2549 */
2550
2551static int packet_create(struct net *net, struct socket *sock, int protocol,
2552 int kern)
2553{
2554 struct sock *sk;
2555 struct packet_sock *po;
2556 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2557 int err;
2558
2559 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2560 return -EPERM;
2561 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2562 sock->type != SOCK_PACKET)
2563 return -ESOCKTNOSUPPORT;
2564
2565 sock->state = SS_UNCONNECTED;
2566
2567 err = -ENOBUFS;
2568 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2569 if (sk == NULL)
2570 goto out;
2571
2572 sock->ops = &packet_ops;
2573 if (sock->type == SOCK_PACKET)
2574 sock->ops = &packet_ops_spkt;
2575
2576 sock_init_data(sock, sk);
2577
2578 po = pkt_sk(sk);
2579 sk->sk_family = PF_PACKET;
2580 po->num = proto;
2581
2582 sk->sk_destruct = packet_sock_destruct;
2583 sk_refcnt_debug_inc(sk);
2584
2585 /*
2586 * Attach a protocol block
2587 */
2588
2589 spin_lock_init(&po->bind_lock);
2590 mutex_init(&po->pg_vec_lock);
2591 po->prot_hook.func = packet_rcv;
2592
2593 if (sock->type == SOCK_PACKET)
2594 po->prot_hook.func = packet_rcv_spkt;
2595
2596 po->prot_hook.af_packet_priv = sk;
2597
2598 if (proto) {
2599 po->prot_hook.type = proto;
2600 register_prot_hook(sk);
2601 }
2602
2603 mutex_lock(&net->packet.sklist_lock);
2604 sk_add_node_rcu(sk, &net->packet.sklist);
2605 mutex_unlock(&net->packet.sklist_lock);
2606
2607 preempt_disable();
2608 sock_prot_inuse_add(net, &packet_proto, 1);
2609 preempt_enable();
2610
2611 return 0;
2612out:
2613 return err;
2614}
2615
2616static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2617{
2618 struct sock_exterr_skb *serr;
2619 struct sk_buff *skb, *skb2;
2620 int copied, err;
2621
2622 err = -EAGAIN;
2623 skb = skb_dequeue(&sk->sk_error_queue);
2624 if (skb == NULL)
2625 goto out;
2626
2627 copied = skb->len;
2628 if (copied > len) {
2629 msg->msg_flags |= MSG_TRUNC;
2630 copied = len;
2631 }
2632 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2633 if (err)
2634 goto out_free_skb;
2635
2636 sock_recv_timestamp(msg, sk, skb);
2637
2638 serr = SKB_EXT_ERR(skb);
2639 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2640 sizeof(serr->ee), &serr->ee);
2641
2642 msg->msg_flags |= MSG_ERRQUEUE;
2643 err = copied;
2644
2645 /* Reset and regenerate socket error */
2646 spin_lock_bh(&sk->sk_error_queue.lock);
2647 sk->sk_err = 0;
2648 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2649 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2650 spin_unlock_bh(&sk->sk_error_queue.lock);
2651 sk->sk_error_report(sk);
2652 } else
2653 spin_unlock_bh(&sk->sk_error_queue.lock);
2654
2655out_free_skb:
2656 kfree_skb(skb);
2657out:
2658 return err;
2659}
2660
2661/*
2662 * Pull a packet from our receive queue and hand it to the user.
2663 * If necessary we block.
2664 */
2665
2666static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2667 struct msghdr *msg, size_t len, int flags)
2668{
2669 struct sock *sk = sock->sk;
2670 struct sk_buff *skb;
2671 int copied, err;
2672 struct sockaddr_ll *sll;
2673 int vnet_hdr_len = 0;
2674
2675 err = -EINVAL;
2676 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2677 goto out;
2678
2679#if 0
2680 /* What error should we return now? EUNATTACH? */
2681 if (pkt_sk(sk)->ifindex < 0)
2682 return -ENODEV;
2683#endif
2684
2685 if (flags & MSG_ERRQUEUE) {
2686 err = packet_recv_error(sk, msg, len);
2687 goto out;
2688 }
2689
2690 /*
2691 * Call the generic datagram receiver. This handles all sorts
2692 * of horrible races and re-entrancy so we can forget about it
2693 * in the protocol layers.
2694 *
2695 * Now it will return ENETDOWN, if device have just gone down,
2696 * but then it will block.
2697 */
2698
2699 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2700
2701 /*
2702 * An error occurred so return it. Because skb_recv_datagram()
2703 * handles the blocking we don't see and worry about blocking
2704 * retries.
2705 */
2706
2707 if (skb == NULL)
2708 goto out;
2709
2710 if (pkt_sk(sk)->has_vnet_hdr) {
2711 struct virtio_net_hdr vnet_hdr = { 0 };
2712
2713 err = -EINVAL;
2714 vnet_hdr_len = sizeof(vnet_hdr);
2715 if (len < vnet_hdr_len)
2716 goto out_free;
2717
2718 len -= vnet_hdr_len;
2719
2720 if (skb_is_gso(skb)) {
2721 struct skb_shared_info *sinfo = skb_shinfo(skb);
2722
2723 /* This is a hint as to how much should be linear. */
2724 vnet_hdr.hdr_len = skb_headlen(skb);
2725 vnet_hdr.gso_size = sinfo->gso_size;
2726 if (sinfo->gso_type & SKB_GSO_TCPV4)
2727 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2728 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2729 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2730 else if (sinfo->gso_type & SKB_GSO_UDP)
2731 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2732 else if (sinfo->gso_type & SKB_GSO_FCOE)
2733 goto out_free;
2734 else
2735 BUG();
2736 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2737 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2738 } else
2739 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2740
2741 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2742 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2743 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2744 vnet_hdr.csum_offset = skb->csum_offset;
2745 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2746 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2747 } /* else everything is zero */
2748
2749 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2750 vnet_hdr_len);
2751 if (err < 0)
2752 goto out_free;
2753 }
2754
2755 /*
2756 * If the address length field is there to be filled in, we fill
2757 * it in now.
2758 */
2759
2760 sll = &PACKET_SKB_CB(skb)->sa.ll;
2761 if (sock->type == SOCK_PACKET)
2762 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2763 else
2764 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2765
2766 /*
2767 * You lose any data beyond the buffer you gave. If it worries a
2768 * user program they can ask the device for its MTU anyway.
2769 */
2770
2771 copied = skb->len;
2772 if (copied > len) {
2773 copied = len;
2774 msg->msg_flags |= MSG_TRUNC;
2775 }
2776
2777 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2778 if (err)
2779 goto out_free;
2780
2781 sock_recv_ts_and_drops(msg, sk, skb);
2782
2783 if (msg->msg_name)
2784 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2785 msg->msg_namelen);
2786
2787 if (pkt_sk(sk)->auxdata) {
2788 struct tpacket_auxdata aux;
2789
2790 aux.tp_status = TP_STATUS_USER;
2791 if (skb->ip_summed == CHECKSUM_PARTIAL)
2792 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2793 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2794 aux.tp_snaplen = skb->len;
2795 aux.tp_mac = 0;
2796 aux.tp_net = skb_network_offset(skb);
2797 if (vlan_tx_tag_present(skb)) {
2798 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2799 aux.tp_status |= TP_STATUS_VLAN_VALID;
2800 } else {
2801 aux.tp_vlan_tci = 0;
2802 }
2803 aux.tp_padding = 0;
2804 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2805 }
2806
2807 /*
2808 * Free or return the buffer as appropriate. Again this
2809 * hides all the races and re-entrancy issues from us.
2810 */
2811 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2812
2813out_free:
2814 skb_free_datagram(sk, skb);
2815out:
2816 return err;
2817}
2818
2819static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2820 int *uaddr_len, int peer)
2821{
2822 struct net_device *dev;
2823 struct sock *sk = sock->sk;
2824
2825 if (peer)
2826 return -EOPNOTSUPP;
2827
2828 uaddr->sa_family = AF_PACKET;
2829 rcu_read_lock();
2830 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2831 if (dev)
2832 strncpy(uaddr->sa_data, dev->name, 14);
2833 else
2834 memset(uaddr->sa_data, 0, 14);
2835 rcu_read_unlock();
2836 *uaddr_len = sizeof(*uaddr);
2837
2838 return 0;
2839}
2840
2841static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2842 int *uaddr_len, int peer)
2843{
2844 struct net_device *dev;
2845 struct sock *sk = sock->sk;
2846 struct packet_sock *po = pkt_sk(sk);
2847 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2848
2849 if (peer)
2850 return -EOPNOTSUPP;
2851
2852 sll->sll_family = AF_PACKET;
2853 sll->sll_ifindex = po->ifindex;
2854 sll->sll_protocol = po->num;
2855 sll->sll_pkttype = 0;
2856 rcu_read_lock();
2857 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2858 if (dev) {
2859 sll->sll_hatype = dev->type;
2860 sll->sll_halen = dev->addr_len;
2861 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2862 } else {
2863 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2864 sll->sll_halen = 0;
2865 }
2866 rcu_read_unlock();
2867 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2868
2869 return 0;
2870}
2871
2872static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2873 int what)
2874{
2875 switch (i->type) {
2876 case PACKET_MR_MULTICAST:
2877 if (i->alen != dev->addr_len)
2878 return -EINVAL;
2879 if (what > 0)
2880 return dev_mc_add(dev, i->addr);
2881 else
2882 return dev_mc_del(dev, i->addr);
2883 break;
2884 case PACKET_MR_PROMISC:
2885 return dev_set_promiscuity(dev, what);
2886 break;
2887 case PACKET_MR_ALLMULTI:
2888 return dev_set_allmulti(dev, what);
2889 break;
2890 case PACKET_MR_UNICAST:
2891 if (i->alen != dev->addr_len)
2892 return -EINVAL;
2893 if (what > 0)
2894 return dev_uc_add(dev, i->addr);
2895 else
2896 return dev_uc_del(dev, i->addr);
2897 break;
2898 default:
2899 break;
2900 }
2901 return 0;
2902}
2903
2904static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2905{
2906 for ( ; i; i = i->next) {
2907 if (i->ifindex == dev->ifindex)
2908 packet_dev_mc(dev, i, what);
2909 }
2910}
2911
2912static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2913{
2914 struct packet_sock *po = pkt_sk(sk);
2915 struct packet_mclist *ml, *i;
2916 struct net_device *dev;
2917 int err;
2918
2919 rtnl_lock();
2920
2921 err = -ENODEV;
2922 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2923 if (!dev)
2924 goto done;
2925
2926 err = -EINVAL;
2927 if (mreq->mr_alen > dev->addr_len)
2928 goto done;
2929
2930 err = -ENOBUFS;
2931 i = kmalloc(sizeof(*i), GFP_KERNEL);
2932 if (i == NULL)
2933 goto done;
2934
2935 err = 0;
2936 for (ml = po->mclist; ml; ml = ml->next) {
2937 if (ml->ifindex == mreq->mr_ifindex &&
2938 ml->type == mreq->mr_type &&
2939 ml->alen == mreq->mr_alen &&
2940 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2941 ml->count++;
2942 /* Free the new element ... */
2943 kfree(i);
2944 goto done;
2945 }
2946 }
2947
2948 i->type = mreq->mr_type;
2949 i->ifindex = mreq->mr_ifindex;
2950 i->alen = mreq->mr_alen;
2951 memcpy(i->addr, mreq->mr_address, i->alen);
2952 i->count = 1;
2953 i->next = po->mclist;
2954 po->mclist = i;
2955 err = packet_dev_mc(dev, i, 1);
2956 if (err) {
2957 po->mclist = i->next;
2958 kfree(i);
2959 }
2960
2961done:
2962 rtnl_unlock();
2963 return err;
2964}
2965
2966static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2967{
2968 struct packet_mclist *ml, **mlp;
2969
2970 rtnl_lock();
2971
2972 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2973 if (ml->ifindex == mreq->mr_ifindex &&
2974 ml->type == mreq->mr_type &&
2975 ml->alen == mreq->mr_alen &&
2976 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2977 if (--ml->count == 0) {
2978 struct net_device *dev;
2979 *mlp = ml->next;
2980 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2981 if (dev)
2982 packet_dev_mc(dev, ml, -1);
2983 kfree(ml);
2984 }
2985 rtnl_unlock();
2986 return 0;
2987 }
2988 }
2989 rtnl_unlock();
2990 return -EADDRNOTAVAIL;
2991}
2992
2993static void packet_flush_mclist(struct sock *sk)
2994{
2995 struct packet_sock *po = pkt_sk(sk);
2996 struct packet_mclist *ml;
2997
2998 if (!po->mclist)
2999 return;
3000
3001 rtnl_lock();
3002 while ((ml = po->mclist) != NULL) {
3003 struct net_device *dev;
3004
3005 po->mclist = ml->next;
3006 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3007 if (dev != NULL)
3008 packet_dev_mc(dev, ml, -1);
3009 kfree(ml);
3010 }
3011 rtnl_unlock();
3012}
3013
3014static int
3015packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3016{
3017 struct sock *sk = sock->sk;
3018 struct packet_sock *po = pkt_sk(sk);
3019 int ret;
3020
3021 if (level != SOL_PACKET)
3022 return -ENOPROTOOPT;
3023
3024 switch (optname) {
3025 case PACKET_ADD_MEMBERSHIP:
3026 case PACKET_DROP_MEMBERSHIP:
3027 {
3028 struct packet_mreq_max mreq;
3029 int len = optlen;
3030 memset(&mreq, 0, sizeof(mreq));
3031 if (len < sizeof(struct packet_mreq))
3032 return -EINVAL;
3033 if (len > sizeof(mreq))
3034 len = sizeof(mreq);
3035 if (copy_from_user(&mreq, optval, len))
3036 return -EFAULT;
3037 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3038 return -EINVAL;
3039 if (optname == PACKET_ADD_MEMBERSHIP)
3040 ret = packet_mc_add(sk, &mreq);
3041 else
3042 ret = packet_mc_drop(sk, &mreq);
3043 return ret;
3044 }
3045
3046 case PACKET_RX_RING:
3047 case PACKET_TX_RING:
3048 {
3049 union tpacket_req_u req_u;
3050 int len;
3051
3052 switch (po->tp_version) {
3053 case TPACKET_V1:
3054 case TPACKET_V2:
3055 len = sizeof(req_u.req);
3056 break;
3057 case TPACKET_V3:
3058 default:
3059 len = sizeof(req_u.req3);
3060 break;
3061 }
3062 if (optlen < len)
3063 return -EINVAL;
3064 if (pkt_sk(sk)->has_vnet_hdr)
3065 return -EINVAL;
3066 if (copy_from_user(&req_u.req, optval, len))
3067 return -EFAULT;
3068 return packet_set_ring(sk, &req_u, 0,
3069 optname == PACKET_TX_RING);
3070 }
3071 case PACKET_COPY_THRESH:
3072 {
3073 int val;
3074
3075 if (optlen != sizeof(val))
3076 return -EINVAL;
3077 if (copy_from_user(&val, optval, sizeof(val)))
3078 return -EFAULT;
3079
3080 pkt_sk(sk)->copy_thresh = val;
3081 return 0;
3082 }
3083 case PACKET_VERSION:
3084 {
3085 int val;
3086
3087 if (optlen != sizeof(val))
3088 return -EINVAL;
3089 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3090 return -EBUSY;
3091 if (copy_from_user(&val, optval, sizeof(val)))
3092 return -EFAULT;
3093 switch (val) {
3094 case TPACKET_V1:
3095 case TPACKET_V2:
3096 case TPACKET_V3:
3097 po->tp_version = val;
3098 return 0;
3099 default:
3100 return -EINVAL;
3101 }
3102 }
3103 case PACKET_RESERVE:
3104 {
3105 unsigned int val;
3106
3107 if (optlen != sizeof(val))
3108 return -EINVAL;
3109 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3110 return -EBUSY;
3111 if (copy_from_user(&val, optval, sizeof(val)))
3112 return -EFAULT;
3113 po->tp_reserve = val;
3114 return 0;
3115 }
3116 case PACKET_LOSS:
3117 {
3118 unsigned int val;
3119
3120 if (optlen != sizeof(val))
3121 return -EINVAL;
3122 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3123 return -EBUSY;
3124 if (copy_from_user(&val, optval, sizeof(val)))
3125 return -EFAULT;
3126 po->tp_loss = !!val;
3127 return 0;
3128 }
3129 case PACKET_AUXDATA:
3130 {
3131 int val;
3132
3133 if (optlen < sizeof(val))
3134 return -EINVAL;
3135 if (copy_from_user(&val, optval, sizeof(val)))
3136 return -EFAULT;
3137
3138 po->auxdata = !!val;
3139 return 0;
3140 }
3141 case PACKET_ORIGDEV:
3142 {
3143 int val;
3144
3145 if (optlen < sizeof(val))
3146 return -EINVAL;
3147 if (copy_from_user(&val, optval, sizeof(val)))
3148 return -EFAULT;
3149
3150 po->origdev = !!val;
3151 return 0;
3152 }
3153 case PACKET_VNET_HDR:
3154 {
3155 int val;
3156
3157 if (sock->type != SOCK_RAW)
3158 return -EINVAL;
3159 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3160 return -EBUSY;
3161 if (optlen < sizeof(val))
3162 return -EINVAL;
3163 if (copy_from_user(&val, optval, sizeof(val)))
3164 return -EFAULT;
3165
3166 po->has_vnet_hdr = !!val;
3167 return 0;
3168 }
3169 case PACKET_TIMESTAMP:
3170 {
3171 int val;
3172
3173 if (optlen != sizeof(val))
3174 return -EINVAL;
3175 if (copy_from_user(&val, optval, sizeof(val)))
3176 return -EFAULT;
3177
3178 po->tp_tstamp = val;
3179 return 0;
3180 }
3181 case PACKET_FANOUT:
3182 {
3183 int val;
3184
3185 if (optlen != sizeof(val))
3186 return -EINVAL;
3187 if (copy_from_user(&val, optval, sizeof(val)))
3188 return -EFAULT;
3189
3190 return fanout_add(sk, val & 0xffff, val >> 16);
3191 }
3192 case PACKET_TX_HAS_OFF:
3193 {
3194 unsigned int val;
3195
3196 if (optlen != sizeof(val))
3197 return -EINVAL;
3198 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3199 return -EBUSY;
3200 if (copy_from_user(&val, optval, sizeof(val)))
3201 return -EFAULT;
3202 po->tp_tx_has_off = !!val;
3203 return 0;
3204 }
3205 default:
3206 return -ENOPROTOOPT;
3207 }
3208}
3209
3210static int packet_getsockopt(struct socket *sock, int level, int optname,
3211 char __user *optval, int __user *optlen)
3212{
3213 int len;
3214 int val, lv = sizeof(val);
3215 struct sock *sk = sock->sk;
3216 struct packet_sock *po = pkt_sk(sk);
3217 void *data = &val;
3218 struct tpacket_stats st;
3219 union tpacket_stats_u st_u;
3220
3221 if (level != SOL_PACKET)
3222 return -ENOPROTOOPT;
3223
3224 if (get_user(len, optlen))
3225 return -EFAULT;
3226
3227 if (len < 0)
3228 return -EINVAL;
3229
3230 switch (optname) {
3231 case PACKET_STATISTICS:
3232 spin_lock_bh(&sk->sk_receive_queue.lock);
3233 if (po->tp_version == TPACKET_V3) {
3234 lv = sizeof(struct tpacket_stats_v3);
3235 memcpy(&st_u.stats3, &po->stats,
3236 sizeof(struct tpacket_stats));
3237 st_u.stats3.tp_freeze_q_cnt =
3238 po->stats_u.stats3.tp_freeze_q_cnt;
3239 st_u.stats3.tp_packets += po->stats.tp_drops;
3240 data = &st_u.stats3;
3241 } else {
3242 lv = sizeof(struct tpacket_stats);
3243 st = po->stats;
3244 st.tp_packets += st.tp_drops;
3245 data = &st;
3246 }
3247 memset(&po->stats, 0, sizeof(st));
3248 spin_unlock_bh(&sk->sk_receive_queue.lock);
3249 break;
3250 case PACKET_AUXDATA:
3251 val = po->auxdata;
3252 break;
3253 case PACKET_ORIGDEV:
3254 val = po->origdev;
3255 break;
3256 case PACKET_VNET_HDR:
3257 val = po->has_vnet_hdr;
3258 break;
3259 case PACKET_VERSION:
3260 val = po->tp_version;
3261 break;
3262 case PACKET_HDRLEN:
3263 if (len > sizeof(int))
3264 len = sizeof(int);
3265 if (copy_from_user(&val, optval, len))
3266 return -EFAULT;
3267 switch (val) {
3268 case TPACKET_V1:
3269 val = sizeof(struct tpacket_hdr);
3270 break;
3271 case TPACKET_V2:
3272 val = sizeof(struct tpacket2_hdr);
3273 break;
3274 case TPACKET_V3:
3275 val = sizeof(struct tpacket3_hdr);
3276 break;
3277 default:
3278 return -EINVAL;
3279 }
3280 break;
3281 case PACKET_RESERVE:
3282 val = po->tp_reserve;
3283 break;
3284 case PACKET_LOSS:
3285 val = po->tp_loss;
3286 break;
3287 case PACKET_TIMESTAMP:
3288 val = po->tp_tstamp;
3289 break;
3290 case PACKET_FANOUT:
3291 val = (po->fanout ?
3292 ((u32)po->fanout->id |
3293 ((u32)po->fanout->type << 16) |
3294 ((u32)po->fanout->flags << 24)) :
3295 0);
3296 break;
3297 case PACKET_TX_HAS_OFF:
3298 val = po->tp_tx_has_off;
3299 break;
3300 default:
3301 return -ENOPROTOOPT;
3302 }
3303
3304 if (len > lv)
3305 len = lv;
3306 if (put_user(len, optlen))
3307 return -EFAULT;
3308 if (copy_to_user(optval, data, len))
3309 return -EFAULT;
3310 return 0;
3311}
3312
3313
3314static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3315{
3316 struct sock *sk;
3317 struct net_device *dev = data;
3318 struct net *net = dev_net(dev);
3319
3320 rcu_read_lock();
3321 sk_for_each_rcu(sk, &net->packet.sklist) {
3322 struct packet_sock *po = pkt_sk(sk);
3323
3324 switch (msg) {
3325 case NETDEV_UNREGISTER:
3326 if (po->mclist)
3327 packet_dev_mclist(dev, po->mclist, -1);
3328 /* fallthrough */
3329
3330 case NETDEV_DOWN:
3331 if (dev->ifindex == po->ifindex) {
3332 spin_lock(&po->bind_lock);
3333 if (po->running) {
3334 __unregister_prot_hook(sk, false);
3335 sk->sk_err = ENETDOWN;
3336 if (!sock_flag(sk, SOCK_DEAD))
3337 sk->sk_error_report(sk);
3338 }
3339 if (msg == NETDEV_UNREGISTER) {
3340 po->ifindex = -1;
3341 if (po->prot_hook.dev)
3342 dev_put(po->prot_hook.dev);
3343 po->prot_hook.dev = NULL;
3344 }
3345 spin_unlock(&po->bind_lock);
3346 }
3347 break;
3348 case NETDEV_UP:
3349 if (dev->ifindex == po->ifindex) {
3350 spin_lock(&po->bind_lock);
3351 if (po->num)
3352 register_prot_hook(sk);
3353 spin_unlock(&po->bind_lock);
3354 }
3355 break;
3356 }
3357 }
3358 rcu_read_unlock();
3359 return NOTIFY_DONE;
3360}
3361
3362
3363static int packet_ioctl(struct socket *sock, unsigned int cmd,
3364 unsigned long arg)
3365{
3366 struct sock *sk = sock->sk;
3367
3368 switch (cmd) {
3369 case SIOCOUTQ:
3370 {
3371 int amount = sk_wmem_alloc_get(sk);
3372
3373 return put_user(amount, (int __user *)arg);
3374 }
3375 case SIOCINQ:
3376 {
3377 struct sk_buff *skb;
3378 int amount = 0;
3379
3380 spin_lock_bh(&sk->sk_receive_queue.lock);
3381 skb = skb_peek(&sk->sk_receive_queue);
3382 if (skb)
3383 amount = skb->len;
3384 spin_unlock_bh(&sk->sk_receive_queue.lock);
3385 return put_user(amount, (int __user *)arg);
3386 }
3387 case SIOCGSTAMP:
3388 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3389 case SIOCGSTAMPNS:
3390 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3391
3392#ifdef CONFIG_INET
3393 case SIOCADDRT:
3394 case SIOCDELRT:
3395 case SIOCDARP:
3396 case SIOCGARP:
3397 case SIOCSARP:
3398 case SIOCGIFADDR:
3399 case SIOCSIFADDR:
3400 case SIOCGIFBRDADDR:
3401 case SIOCSIFBRDADDR:
3402 case SIOCGIFNETMASK:
3403 case SIOCSIFNETMASK:
3404 case SIOCGIFDSTADDR:
3405 case SIOCSIFDSTADDR:
3406 case SIOCSIFFLAGS:
3407 return inet_dgram_ops.ioctl(sock, cmd, arg);
3408#endif
3409
3410 default:
3411 return -ENOIOCTLCMD;
3412 }
3413 return 0;
3414}
3415
3416static unsigned int packet_poll(struct file *file, struct socket *sock,
3417 poll_table *wait)
3418{
3419 struct sock *sk = sock->sk;
3420 struct packet_sock *po = pkt_sk(sk);
3421 unsigned int mask = datagram_poll(file, sock, wait);
3422
3423 spin_lock_bh(&sk->sk_receive_queue.lock);
3424 if (po->rx_ring.pg_vec) {
3425 if (!packet_previous_rx_frame(po, &po->rx_ring,
3426 TP_STATUS_KERNEL))
3427 mask |= POLLIN | POLLRDNORM;
3428 }
3429 spin_unlock_bh(&sk->sk_receive_queue.lock);
3430 spin_lock_bh(&sk->sk_write_queue.lock);
3431 if (po->tx_ring.pg_vec) {
3432 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3433 mask |= POLLOUT | POLLWRNORM;
3434 }
3435 spin_unlock_bh(&sk->sk_write_queue.lock);
3436 return mask;
3437}
3438
3439
3440/* Dirty? Well, I still did not learn better way to account
3441 * for user mmaps.
3442 */
3443
3444static void packet_mm_open(struct vm_area_struct *vma)
3445{
3446 struct file *file = vma->vm_file;
3447 struct socket *sock = file->private_data;
3448 struct sock *sk = sock->sk;
3449
3450 if (sk)
3451 atomic_inc(&pkt_sk(sk)->mapped);
3452}
3453
3454static void packet_mm_close(struct vm_area_struct *vma)
3455{
3456 struct file *file = vma->vm_file;
3457 struct socket *sock = file->private_data;
3458 struct sock *sk = sock->sk;
3459
3460 if (sk)
3461 atomic_dec(&pkt_sk(sk)->mapped);
3462}
3463
3464static const struct vm_operations_struct packet_mmap_ops = {
3465 .open = packet_mm_open,
3466 .close = packet_mm_close,
3467};
3468
3469static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3470 unsigned int len)
3471{
3472 int i;
3473
3474 for (i = 0; i < len; i++) {
3475 if (likely(pg_vec[i].buffer)) {
3476 if (is_vmalloc_addr(pg_vec[i].buffer))
3477 vfree(pg_vec[i].buffer);
3478 else
3479 free_pages((unsigned long)pg_vec[i].buffer,
3480 order);
3481 pg_vec[i].buffer = NULL;
3482 }
3483 }
3484 kfree(pg_vec);
3485}
3486
3487static char *alloc_one_pg_vec_page(unsigned long order)
3488{
3489 char *buffer = NULL;
3490 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3491 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3492
3493 buffer = (char *) __get_free_pages(gfp_flags, order);
3494
3495 if (buffer)
3496 return buffer;
3497
3498 /*
3499 * __get_free_pages failed, fall back to vmalloc
3500 */
3501 buffer = vzalloc((1 << order) * PAGE_SIZE);
3502
3503 if (buffer)
3504 return buffer;
3505
3506 /*
3507 * vmalloc failed, lets dig into swap here
3508 */
3509 gfp_flags &= ~__GFP_NORETRY;
3510 buffer = (char *)__get_free_pages(gfp_flags, order);
3511 if (buffer)
3512 return buffer;
3513
3514 /*
3515 * complete and utter failure
3516 */
3517 return NULL;
3518}
3519
3520static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3521{
3522 unsigned int block_nr = req->tp_block_nr;
3523 struct pgv *pg_vec;
3524 int i;
3525
3526 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3527 if (unlikely(!pg_vec))
3528 goto out;
3529
3530 for (i = 0; i < block_nr; i++) {
3531 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3532 if (unlikely(!pg_vec[i].buffer))
3533 goto out_free_pgvec;
3534 }
3535
3536out:
3537 return pg_vec;
3538
3539out_free_pgvec:
3540 free_pg_vec(pg_vec, order, block_nr);
3541 pg_vec = NULL;
3542 goto out;
3543}
3544
3545static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3546 int closing, int tx_ring)
3547{
3548 struct pgv *pg_vec = NULL;
3549 struct packet_sock *po = pkt_sk(sk);
3550 int was_running, order = 0;
3551 struct packet_ring_buffer *rb;
3552 struct sk_buff_head *rb_queue;
3553 __be16 num;
3554 int err = -EINVAL;
3555 /* Added to avoid minimal code churn */
3556 struct tpacket_req *req = &req_u->req;
3557
3558 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3559 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3560 WARN(1, "Tx-ring is not supported.\n");
3561 goto out;
3562 }
3563
3564 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3565 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3566
3567 err = -EBUSY;
3568 if (!closing) {
3569 if (atomic_read(&po->mapped))
3570 goto out;
3571 if (atomic_read(&rb->pending))
3572 goto out;
3573 }
3574
3575 if (req->tp_block_nr) {
3576 /* Sanity tests and some calculations */
3577 err = -EBUSY;
3578 if (unlikely(rb->pg_vec))
3579 goto out;
3580
3581 switch (po->tp_version) {
3582 case TPACKET_V1:
3583 po->tp_hdrlen = TPACKET_HDRLEN;
3584 break;
3585 case TPACKET_V2:
3586 po->tp_hdrlen = TPACKET2_HDRLEN;
3587 break;
3588 case TPACKET_V3:
3589 po->tp_hdrlen = TPACKET3_HDRLEN;
3590 break;
3591 }
3592
3593 err = -EINVAL;
3594 if (unlikely((int)req->tp_block_size <= 0))
3595 goto out;
3596 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3597 goto out;
3598 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3599 po->tp_reserve))
3600 goto out;
3601 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3602 goto out;
3603
3604 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3605 if (unlikely(rb->frames_per_block <= 0))
3606 goto out;
3607 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3608 req->tp_frame_nr))
3609 goto out;
3610
3611 err = -ENOMEM;
3612 order = get_order(req->tp_block_size);
3613 pg_vec = alloc_pg_vec(req, order);
3614 if (unlikely(!pg_vec))
3615 goto out;
3616 switch (po->tp_version) {
3617 case TPACKET_V3:
3618 /* Transmit path is not supported. We checked
3619 * it above but just being paranoid
3620 */
3621 if (!tx_ring)
3622 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3623 break;
3624 default:
3625 break;
3626 }
3627 }
3628 /* Done */
3629 else {
3630 err = -EINVAL;
3631 if (unlikely(req->tp_frame_nr))
3632 goto out;
3633 }
3634
3635 lock_sock(sk);
3636
3637 /* Detach socket from network */
3638 spin_lock(&po->bind_lock);
3639 was_running = po->running;
3640 num = po->num;
3641 if (was_running) {
3642 po->num = 0;
3643 __unregister_prot_hook(sk, false);
3644 }
3645 spin_unlock(&po->bind_lock);
3646
3647 synchronize_net();
3648
3649 err = -EBUSY;
3650 mutex_lock(&po->pg_vec_lock);
3651 if (closing || atomic_read(&po->mapped) == 0) {
3652 err = 0;
3653 spin_lock_bh(&rb_queue->lock);
3654 swap(rb->pg_vec, pg_vec);
3655 rb->frame_max = (req->tp_frame_nr - 1);
3656 rb->head = 0;
3657 rb->frame_size = req->tp_frame_size;
3658 spin_unlock_bh(&rb_queue->lock);
3659
3660 swap(rb->pg_vec_order, order);
3661 swap(rb->pg_vec_len, req->tp_block_nr);
3662
3663 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3664 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3665 tpacket_rcv : packet_rcv;
3666 skb_queue_purge(rb_queue);
3667 if (atomic_read(&po->mapped))
3668 pr_err("packet_mmap: vma is busy: %d\n",
3669 atomic_read(&po->mapped));
3670 }
3671 mutex_unlock(&po->pg_vec_lock);
3672
3673 spin_lock(&po->bind_lock);
3674 if (was_running) {
3675 po->num = num;
3676 register_prot_hook(sk);
3677 }
3678 spin_unlock(&po->bind_lock);
3679 if (closing && (po->tp_version > TPACKET_V2)) {
3680 /* Because we don't support block-based V3 on tx-ring */
3681 if (!tx_ring)
3682 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3683 }
3684 release_sock(sk);
3685
3686 if (pg_vec)
3687 free_pg_vec(pg_vec, order, req->tp_block_nr);
3688out:
3689 return err;
3690}
3691
3692static int packet_mmap(struct file *file, struct socket *sock,
3693 struct vm_area_struct *vma)
3694{
3695 struct sock *sk = sock->sk;
3696 struct packet_sock *po = pkt_sk(sk);
3697 unsigned long size, expected_size;
3698 struct packet_ring_buffer *rb;
3699 unsigned long start;
3700 int err = -EINVAL;
3701 int i;
3702
3703 if (vma->vm_pgoff)
3704 return -EINVAL;
3705
3706 mutex_lock(&po->pg_vec_lock);
3707
3708 expected_size = 0;
3709 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3710 if (rb->pg_vec) {
3711 expected_size += rb->pg_vec_len
3712 * rb->pg_vec_pages
3713 * PAGE_SIZE;
3714 }
3715 }
3716
3717 if (expected_size == 0)
3718 goto out;
3719
3720 size = vma->vm_end - vma->vm_start;
3721 if (size != expected_size)
3722 goto out;
3723
3724 start = vma->vm_start;
3725 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3726 if (rb->pg_vec == NULL)
3727 continue;
3728
3729 for (i = 0; i < rb->pg_vec_len; i++) {
3730 struct page *page;
3731 void *kaddr = rb->pg_vec[i].buffer;
3732 int pg_num;
3733
3734 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3735 page = pgv_to_page(kaddr);
3736 err = vm_insert_page(vma, start, page);
3737 if (unlikely(err))
3738 goto out;
3739 start += PAGE_SIZE;
3740 kaddr += PAGE_SIZE;
3741 }
3742 }
3743 }
3744
3745 atomic_inc(&po->mapped);
3746 vma->vm_ops = &packet_mmap_ops;
3747 err = 0;
3748
3749out:
3750 mutex_unlock(&po->pg_vec_lock);
3751 return err;
3752}
3753
3754static const struct proto_ops packet_ops_spkt = {
3755 .family = PF_PACKET,
3756 .owner = THIS_MODULE,
3757 .release = packet_release,
3758 .bind = packet_bind_spkt,
3759 .connect = sock_no_connect,
3760 .socketpair = sock_no_socketpair,
3761 .accept = sock_no_accept,
3762 .getname = packet_getname_spkt,
3763 .poll = datagram_poll,
3764 .ioctl = packet_ioctl,
3765 .listen = sock_no_listen,
3766 .shutdown = sock_no_shutdown,
3767 .setsockopt = sock_no_setsockopt,
3768 .getsockopt = sock_no_getsockopt,
3769 .sendmsg = packet_sendmsg_spkt,
3770 .recvmsg = packet_recvmsg,
3771 .mmap = sock_no_mmap,
3772 .sendpage = sock_no_sendpage,
3773};
3774
3775static const struct proto_ops packet_ops = {
3776 .family = PF_PACKET,
3777 .owner = THIS_MODULE,
3778 .release = packet_release,
3779 .bind = packet_bind,
3780 .connect = sock_no_connect,
3781 .socketpair = sock_no_socketpair,
3782 .accept = sock_no_accept,
3783 .getname = packet_getname,
3784 .poll = packet_poll,
3785 .ioctl = packet_ioctl,
3786 .listen = sock_no_listen,
3787 .shutdown = sock_no_shutdown,
3788 .setsockopt = packet_setsockopt,
3789 .getsockopt = packet_getsockopt,
3790 .sendmsg = packet_sendmsg,
3791 .recvmsg = packet_recvmsg,
3792 .mmap = packet_mmap,
3793 .sendpage = sock_no_sendpage,
3794};
3795
3796static const struct net_proto_family packet_family_ops = {
3797 .family = PF_PACKET,
3798 .create = packet_create,
3799 .owner = THIS_MODULE,
3800};
3801
3802static struct notifier_block packet_netdev_notifier = {
3803 .notifier_call = packet_notifier,
3804};
3805
3806#ifdef CONFIG_PROC_FS
3807
3808static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3809 __acquires(RCU)
3810{
3811 struct net *net = seq_file_net(seq);
3812
3813 rcu_read_lock();
3814 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3815}
3816
3817static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3818{
3819 struct net *net = seq_file_net(seq);
3820 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3821}
3822
3823static void packet_seq_stop(struct seq_file *seq, void *v)
3824 __releases(RCU)
3825{
3826 rcu_read_unlock();
3827}
3828
3829static int packet_seq_show(struct seq_file *seq, void *v)
3830{
3831 if (v == SEQ_START_TOKEN)
3832 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3833 else {
3834 struct sock *s = sk_entry(v);
3835 const struct packet_sock *po = pkt_sk(s);
3836
3837 seq_printf(seq,
3838 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3839 s,
3840 atomic_read(&s->sk_refcnt),
3841 s->sk_type,
3842 ntohs(po->num),
3843 po->ifindex,
3844 po->running,
3845 atomic_read(&s->sk_rmem_alloc),
3846 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3847 sock_i_ino(s));
3848 }
3849
3850 return 0;
3851}
3852
3853static const struct seq_operations packet_seq_ops = {
3854 .start = packet_seq_start,
3855 .next = packet_seq_next,
3856 .stop = packet_seq_stop,
3857 .show = packet_seq_show,
3858};
3859
3860static int packet_seq_open(struct inode *inode, struct file *file)
3861{
3862 return seq_open_net(inode, file, &packet_seq_ops,
3863 sizeof(struct seq_net_private));
3864}
3865
3866static const struct file_operations packet_seq_fops = {
3867 .owner = THIS_MODULE,
3868 .open = packet_seq_open,
3869 .read = seq_read,
3870 .llseek = seq_lseek,
3871 .release = seq_release_net,
3872};
3873
3874#endif
3875
3876static int __net_init packet_net_init(struct net *net)
3877{
3878 mutex_init(&net->packet.sklist_lock);
3879 INIT_HLIST_HEAD(&net->packet.sklist);
3880
3881 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3882 return -ENOMEM;
3883
3884 return 0;
3885}
3886
3887static void __net_exit packet_net_exit(struct net *net)
3888{
3889 remove_proc_entry("packet", net->proc_net);
3890}
3891
3892static struct pernet_operations packet_net_ops = {
3893 .init = packet_net_init,
3894 .exit = packet_net_exit,
3895};
3896
3897
3898static void __exit packet_exit(void)
3899{
3900 unregister_netdevice_notifier(&packet_netdev_notifier);
3901 unregister_pernet_subsys(&packet_net_ops);
3902 sock_unregister(PF_PACKET);
3903 proto_unregister(&packet_proto);
3904}
3905
3906static int __init packet_init(void)
3907{
3908 int rc = proto_register(&packet_proto, 0);
3909
3910 if (rc != 0)
3911 goto out;
3912
3913 sock_register(&packet_family_ops);
3914 register_pernet_subsys(&packet_net_ops);
3915 register_netdevice_notifier(&packet_netdev_notifier);
3916out:
3917 return rc;
3918}
3919
3920module_init(packet_init);
3921module_exit(packet_exit);
3922MODULE_LICENSE("GPL");
3923MODULE_ALIAS_NETPROTO(PF_PACKET);
Linux kernel - Deliverables
This page took 0.065533 seconds and 5 git commands to generate.