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.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
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
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
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.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.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>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.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 #include <linux/reciprocal_div.h>
93 #include <net/inet_common.h>
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
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
112 Incoming, dev->hard_header!=NULL
113 mac_header -> ll header
116 Outgoing, dev->hard_header!=NULL
117 mac_header -> ll header
120 Incoming, dev->hard_header==NULL
121 mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 header. PPP makes it, that is wrong, because introduce
123 assymetry between rx and tx paths.
126 Outgoing, dev->hard_header==NULL
127 mac_header -> data. ll header is still not built!
131 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
137 dev->hard_header != NULL
138 mac_header -> ll header
141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
145 We should set nh.raw on output to correct posistion,
146 packet classifier depends on it.
149 /* Private packet socket structures. */
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
154 struct packet_mreq_max
{
156 unsigned short mr_type
;
157 unsigned short mr_alen
;
158 unsigned char mr_address
[MAX_ADDR_LEN
];
162 struct tpacket_hdr
*h1
;
163 struct tpacket2_hdr
*h2
;
164 struct tpacket3_hdr
*h3
;
168 static int packet_set_ring(struct sock
*sk
, union tpacket_req_u
*req_u
,
169 int closing
, int tx_ring
);
171 #define V3_ALIGNMENT (8)
173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 #define PGV_FROM_VMALLOC 1
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)))
189 static int tpacket_snd(struct packet_sock
*po
, struct msghdr
*msg
);
190 static int tpacket_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
191 struct packet_type
*pt
, struct net_device
*orig_dev
);
193 static void *packet_previous_frame(struct packet_sock
*po
,
194 struct packet_ring_buffer
*rb
,
196 static void packet_increment_head(struct packet_ring_buffer
*buff
);
197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core
*,
198 struct tpacket_block_desc
*);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core
*,
200 struct packet_sock
*);
201 static void prb_retire_current_block(struct tpacket_kbdq_core
*,
202 struct packet_sock
*, unsigned int status
);
203 static int prb_queue_frozen(struct tpacket_kbdq_core
*);
204 static void prb_open_block(struct tpacket_kbdq_core
*,
205 struct tpacket_block_desc
*);
206 static void prb_retire_rx_blk_timer_expired(unsigned long);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core
*);
208 static void prb_init_blk_timer(struct packet_sock
*,
209 struct tpacket_kbdq_core
*,
210 void (*func
) (unsigned long));
211 static void prb_fill_rxhash(struct tpacket_kbdq_core
*, struct tpacket3_hdr
*);
212 static void prb_clear_rxhash(struct tpacket_kbdq_core
*,
213 struct tpacket3_hdr
*);
214 static void prb_fill_vlan_info(struct tpacket_kbdq_core
*,
215 struct tpacket3_hdr
*);
216 static void packet_flush_mclist(struct sock
*sk
);
218 struct packet_skb_cb
{
219 unsigned int origlen
;
221 struct sockaddr_pkt pkt
;
222 struct sockaddr_ll ll
;
226 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
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)
237 static void __fanout_unlink(struct sock
*sk
, struct packet_sock
*po
);
238 static void __fanout_link(struct sock
*sk
, struct packet_sock
*po
);
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()).
244 static void register_prot_hook(struct sock
*sk
)
246 struct packet_sock
*po
= pkt_sk(sk
);
249 __fanout_link(sk
, po
);
251 dev_add_pack(&po
->prot_hook
);
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.
264 static void __unregister_prot_hook(struct sock
*sk
, bool sync
)
266 struct packet_sock
*po
= pkt_sk(sk
);
270 __fanout_unlink(sk
, po
);
272 __dev_remove_pack(&po
->prot_hook
);
276 spin_unlock(&po
->bind_lock
);
278 spin_lock(&po
->bind_lock
);
282 static void unregister_prot_hook(struct sock
*sk
, bool sync
)
284 struct packet_sock
*po
= pkt_sk(sk
);
287 __unregister_prot_hook(sk
, sync
);
290 static inline __pure
struct page
*pgv_to_page(void *addr
)
292 if (is_vmalloc_addr(addr
))
293 return vmalloc_to_page(addr
);
294 return virt_to_page(addr
);
297 static void __packet_set_status(struct packet_sock
*po
, void *frame
, int status
)
299 union tpacket_uhdr h
;
302 switch (po
->tp_version
) {
304 h
.h1
->tp_status
= status
;
305 flush_dcache_page(pgv_to_page(&h
.h1
->tp_status
));
308 h
.h2
->tp_status
= status
;
309 flush_dcache_page(pgv_to_page(&h
.h2
->tp_status
));
313 WARN(1, "TPACKET version not supported.\n");
320 static int __packet_get_status(struct packet_sock
*po
, void *frame
)
322 union tpacket_uhdr h
;
327 switch (po
->tp_version
) {
329 flush_dcache_page(pgv_to_page(&h
.h1
->tp_status
));
330 return h
.h1
->tp_status
;
332 flush_dcache_page(pgv_to_page(&h
.h2
->tp_status
));
333 return h
.h2
->tp_status
;
336 WARN(1, "TPACKET version not supported.\n");
342 static __u32
tpacket_get_timestamp(struct sk_buff
*skb
, struct timespec
*ts
,
345 struct skb_shared_hwtstamps
*shhwtstamps
= skb_hwtstamps(skb
);
348 if ((flags
& SOF_TIMESTAMPING_SYS_HARDWARE
) &&
349 ktime_to_timespec_cond(shhwtstamps
->syststamp
, ts
))
350 return TP_STATUS_TS_SYS_HARDWARE
;
351 if ((flags
& SOF_TIMESTAMPING_RAW_HARDWARE
) &&
352 ktime_to_timespec_cond(shhwtstamps
->hwtstamp
, ts
))
353 return TP_STATUS_TS_RAW_HARDWARE
;
356 if (ktime_to_timespec_cond(skb
->tstamp
, ts
))
357 return TP_STATUS_TS_SOFTWARE
;
362 static __u32
__packet_set_timestamp(struct packet_sock
*po
, void *frame
,
365 union tpacket_uhdr h
;
369 if (!(ts_status
= tpacket_get_timestamp(skb
, &ts
, po
->tp_tstamp
)))
373 switch (po
->tp_version
) {
375 h
.h1
->tp_sec
= ts
.tv_sec
;
376 h
.h1
->tp_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
379 h
.h2
->tp_sec
= ts
.tv_sec
;
380 h
.h2
->tp_nsec
= ts
.tv_nsec
;
384 WARN(1, "TPACKET version not supported.\n");
388 /* one flush is safe, as both fields always lie on the same cacheline */
389 flush_dcache_page(pgv_to_page(&h
.h1
->tp_sec
));
395 static void *packet_lookup_frame(struct packet_sock
*po
,
396 struct packet_ring_buffer
*rb
,
397 unsigned int position
,
400 unsigned int pg_vec_pos
, frame_offset
;
401 union tpacket_uhdr h
;
403 pg_vec_pos
= position
/ rb
->frames_per_block
;
404 frame_offset
= position
% rb
->frames_per_block
;
406 h
.raw
= rb
->pg_vec
[pg_vec_pos
].buffer
+
407 (frame_offset
* rb
->frame_size
);
409 if (status
!= __packet_get_status(po
, h
.raw
))
415 static void *packet_current_frame(struct packet_sock
*po
,
416 struct packet_ring_buffer
*rb
,
419 return packet_lookup_frame(po
, rb
, rb
->head
, status
);
422 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core
*pkc
)
424 del_timer_sync(&pkc
->retire_blk_timer
);
427 static void prb_shutdown_retire_blk_timer(struct packet_sock
*po
,
429 struct sk_buff_head
*rb_queue
)
431 struct tpacket_kbdq_core
*pkc
;
433 pkc
= tx_ring
? &po
->tx_ring
.prb_bdqc
: &po
->rx_ring
.prb_bdqc
;
435 spin_lock(&rb_queue
->lock
);
436 pkc
->delete_blk_timer
= 1;
437 spin_unlock(&rb_queue
->lock
);
439 prb_del_retire_blk_timer(pkc
);
442 static void prb_init_blk_timer(struct packet_sock
*po
,
443 struct tpacket_kbdq_core
*pkc
,
444 void (*func
) (unsigned long))
446 init_timer(&pkc
->retire_blk_timer
);
447 pkc
->retire_blk_timer
.data
= (long)po
;
448 pkc
->retire_blk_timer
.function
= func
;
449 pkc
->retire_blk_timer
.expires
= jiffies
;
452 static void prb_setup_retire_blk_timer(struct packet_sock
*po
, int tx_ring
)
454 struct tpacket_kbdq_core
*pkc
;
459 pkc
= tx_ring
? &po
->tx_ring
.prb_bdqc
: &po
->rx_ring
.prb_bdqc
;
460 prb_init_blk_timer(po
, pkc
, prb_retire_rx_blk_timer_expired
);
463 static int prb_calc_retire_blk_tmo(struct packet_sock
*po
,
464 int blk_size_in_bytes
)
466 struct net_device
*dev
;
467 unsigned int mbits
= 0, msec
= 0, div
= 0, tmo
= 0;
468 struct ethtool_cmd ecmd
;
473 dev
= __dev_get_by_index(sock_net(&po
->sk
), po
->ifindex
);
474 if (unlikely(!dev
)) {
476 return DEFAULT_PRB_RETIRE_TOV
;
478 err
= __ethtool_get_settings(dev
, &ecmd
);
479 speed
= ethtool_cmd_speed(&ecmd
);
483 * If the link speed is so slow you don't really
484 * need to worry about perf anyways
486 if (speed
< SPEED_1000
|| speed
== SPEED_UNKNOWN
) {
487 return DEFAULT_PRB_RETIRE_TOV
;
494 mbits
= (blk_size_in_bytes
* 8) / (1024 * 1024);
506 static void prb_init_ft_ops(struct tpacket_kbdq_core
*p1
,
507 union tpacket_req_u
*req_u
)
509 p1
->feature_req_word
= req_u
->req3
.tp_feature_req_word
;
512 static void init_prb_bdqc(struct packet_sock
*po
,
513 struct packet_ring_buffer
*rb
,
515 union tpacket_req_u
*req_u
, int tx_ring
)
517 struct tpacket_kbdq_core
*p1
= &rb
->prb_bdqc
;
518 struct tpacket_block_desc
*pbd
;
520 memset(p1
, 0x0, sizeof(*p1
));
522 p1
->knxt_seq_num
= 1;
524 pbd
= (struct tpacket_block_desc
*)pg_vec
[0].buffer
;
525 p1
->pkblk_start
= pg_vec
[0].buffer
;
526 p1
->kblk_size
= req_u
->req3
.tp_block_size
;
527 p1
->knum_blocks
= req_u
->req3
.tp_block_nr
;
528 p1
->hdrlen
= po
->tp_hdrlen
;
529 p1
->version
= po
->tp_version
;
530 p1
->last_kactive_blk_num
= 0;
531 po
->stats
.stats3
.tp_freeze_q_cnt
= 0;
532 if (req_u
->req3
.tp_retire_blk_tov
)
533 p1
->retire_blk_tov
= req_u
->req3
.tp_retire_blk_tov
;
535 p1
->retire_blk_tov
= prb_calc_retire_blk_tmo(po
,
536 req_u
->req3
.tp_block_size
);
537 p1
->tov_in_jiffies
= msecs_to_jiffies(p1
->retire_blk_tov
);
538 p1
->blk_sizeof_priv
= req_u
->req3
.tp_sizeof_priv
;
540 prb_init_ft_ops(p1
, req_u
);
541 prb_setup_retire_blk_timer(po
, tx_ring
);
542 prb_open_block(p1
, pbd
);
545 /* Do NOT update the last_blk_num first.
546 * Assumes sk_buff_head lock is held.
548 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core
*pkc
)
550 mod_timer(&pkc
->retire_blk_timer
,
551 jiffies
+ pkc
->tov_in_jiffies
);
552 pkc
->last_kactive_blk_num
= pkc
->kactive_blk_num
;
557 * 1) We refresh the timer only when we open a block.
558 * By doing this we don't waste cycles refreshing the timer
559 * on packet-by-packet basis.
561 * With a 1MB block-size, on a 1Gbps line, it will take
562 * i) ~8 ms to fill a block + ii) memcpy etc.
563 * In this cut we are not accounting for the memcpy time.
565 * So, if the user sets the 'tmo' to 10ms then the timer
566 * will never fire while the block is still getting filled
567 * (which is what we want). However, the user could choose
568 * to close a block early and that's fine.
570 * But when the timer does fire, we check whether or not to refresh it.
571 * Since the tmo granularity is in msecs, it is not too expensive
572 * to refresh the timer, lets say every '8' msecs.
573 * Either the user can set the 'tmo' or we can derive it based on
574 * a) line-speed and b) block-size.
575 * prb_calc_retire_blk_tmo() calculates the tmo.
578 static void prb_retire_rx_blk_timer_expired(unsigned long data
)
580 struct packet_sock
*po
= (struct packet_sock
*)data
;
581 struct tpacket_kbdq_core
*pkc
= &po
->rx_ring
.prb_bdqc
;
583 struct tpacket_block_desc
*pbd
;
585 spin_lock(&po
->sk
.sk_receive_queue
.lock
);
587 frozen
= prb_queue_frozen(pkc
);
588 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
590 if (unlikely(pkc
->delete_blk_timer
))
593 /* We only need to plug the race when the block is partially filled.
595 * lock(); increment BLOCK_NUM_PKTS; unlock()
596 * copy_bits() is in progress ...
597 * timer fires on other cpu:
598 * we can't retire the current block because copy_bits
602 if (BLOCK_NUM_PKTS(pbd
)) {
603 while (atomic_read(&pkc
->blk_fill_in_prog
)) {
604 /* Waiting for skb_copy_bits to finish... */
609 if (pkc
->last_kactive_blk_num
== pkc
->kactive_blk_num
) {
611 prb_retire_current_block(pkc
, po
, TP_STATUS_BLK_TMO
);
612 if (!prb_dispatch_next_block(pkc
, po
))
617 /* Case 1. Queue was frozen because user-space was
620 if (prb_curr_blk_in_use(pkc
, pbd
)) {
622 * Ok, user-space is still behind.
623 * So just refresh the timer.
627 /* Case 2. queue was frozen,user-space caught up,
628 * now the link went idle && the timer fired.
629 * We don't have a block to close.So we open this
630 * block and restart the timer.
631 * opening a block thaws the queue,restarts timer
632 * Thawing/timer-refresh is a side effect.
634 prb_open_block(pkc
, pbd
);
641 _prb_refresh_rx_retire_blk_timer(pkc
);
644 spin_unlock(&po
->sk
.sk_receive_queue
.lock
);
647 static void prb_flush_block(struct tpacket_kbdq_core
*pkc1
,
648 struct tpacket_block_desc
*pbd1
, __u32 status
)
650 /* Flush everything minus the block header */
652 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
657 /* Skip the block header(we know header WILL fit in 4K) */
660 end
= (u8
*)PAGE_ALIGN((unsigned long)pkc1
->pkblk_end
);
661 for (; start
< end
; start
+= PAGE_SIZE
)
662 flush_dcache_page(pgv_to_page(start
));
667 /* Now update the block status. */
669 BLOCK_STATUS(pbd1
) = status
;
671 /* Flush the block header */
673 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
675 flush_dcache_page(pgv_to_page(start
));
685 * 2) Increment active_blk_num
687 * Note:We DONT refresh the timer on purpose.
688 * Because almost always the next block will be opened.
690 static void prb_close_block(struct tpacket_kbdq_core
*pkc1
,
691 struct tpacket_block_desc
*pbd1
,
692 struct packet_sock
*po
, unsigned int stat
)
694 __u32 status
= TP_STATUS_USER
| stat
;
696 struct tpacket3_hdr
*last_pkt
;
697 struct tpacket_hdr_v1
*h1
= &pbd1
->hdr
.bh1
;
699 if (po
->stats
.stats3
.tp_drops
)
700 status
|= TP_STATUS_LOSING
;
702 last_pkt
= (struct tpacket3_hdr
*)pkc1
->prev
;
703 last_pkt
->tp_next_offset
= 0;
705 /* Get the ts of the last pkt */
706 if (BLOCK_NUM_PKTS(pbd1
)) {
707 h1
->ts_last_pkt
.ts_sec
= last_pkt
->tp_sec
;
708 h1
->ts_last_pkt
.ts_nsec
= last_pkt
->tp_nsec
;
710 /* Ok, we tmo'd - so get the current time */
713 h1
->ts_last_pkt
.ts_sec
= ts
.tv_sec
;
714 h1
->ts_last_pkt
.ts_nsec
= ts
.tv_nsec
;
719 /* Flush the block */
720 prb_flush_block(pkc1
, pbd1
, status
);
722 pkc1
->kactive_blk_num
= GET_NEXT_PRB_BLK_NUM(pkc1
);
725 static void prb_thaw_queue(struct tpacket_kbdq_core
*pkc
)
727 pkc
->reset_pending_on_curr_blk
= 0;
731 * Side effect of opening a block:
733 * 1) prb_queue is thawed.
734 * 2) retire_blk_timer is refreshed.
737 static void prb_open_block(struct tpacket_kbdq_core
*pkc1
,
738 struct tpacket_block_desc
*pbd1
)
741 struct tpacket_hdr_v1
*h1
= &pbd1
->hdr
.bh1
;
745 /* We could have just memset this but we will lose the
746 * flexibility of making the priv area sticky
749 BLOCK_SNUM(pbd1
) = pkc1
->knxt_seq_num
++;
750 BLOCK_NUM_PKTS(pbd1
) = 0;
751 BLOCK_LEN(pbd1
) = BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
755 h1
->ts_first_pkt
.ts_sec
= ts
.tv_sec
;
756 h1
->ts_first_pkt
.ts_nsec
= ts
.tv_nsec
;
758 pkc1
->pkblk_start
= (char *)pbd1
;
759 pkc1
->nxt_offset
= pkc1
->pkblk_start
+ BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
761 BLOCK_O2FP(pbd1
) = (__u32
)BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
762 BLOCK_O2PRIV(pbd1
) = BLK_HDR_LEN
;
764 pbd1
->version
= pkc1
->version
;
765 pkc1
->prev
= pkc1
->nxt_offset
;
766 pkc1
->pkblk_end
= pkc1
->pkblk_start
+ pkc1
->kblk_size
;
768 prb_thaw_queue(pkc1
);
769 _prb_refresh_rx_retire_blk_timer(pkc1
);
775 * Queue freeze logic:
776 * 1) Assume tp_block_nr = 8 blocks.
777 * 2) At time 't0', user opens Rx ring.
778 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
779 * 4) user-space is either sleeping or processing block '0'.
780 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
781 * it will close block-7,loop around and try to fill block '0'.
783 * __packet_lookup_frame_in_block
784 * prb_retire_current_block()
785 * prb_dispatch_next_block()
786 * |->(BLOCK_STATUS == USER) evaluates to true
787 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
788 * 6) Now there are two cases:
789 * 6.1) Link goes idle right after the queue is frozen.
790 * But remember, the last open_block() refreshed the timer.
791 * When this timer expires,it will refresh itself so that we can
792 * re-open block-0 in near future.
793 * 6.2) Link is busy and keeps on receiving packets. This is a simple
794 * case and __packet_lookup_frame_in_block will check if block-0
795 * is free and can now be re-used.
797 static void prb_freeze_queue(struct tpacket_kbdq_core
*pkc
,
798 struct packet_sock
*po
)
800 pkc
->reset_pending_on_curr_blk
= 1;
801 po
->stats
.stats3
.tp_freeze_q_cnt
++;
804 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
807 * If the next block is free then we will dispatch it
808 * and return a good offset.
809 * Else, we will freeze the queue.
810 * So, caller must check the return value.
812 static void *prb_dispatch_next_block(struct tpacket_kbdq_core
*pkc
,
813 struct packet_sock
*po
)
815 struct tpacket_block_desc
*pbd
;
819 /* 1. Get current block num */
820 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
822 /* 2. If this block is currently in_use then freeze the queue */
823 if (TP_STATUS_USER
& BLOCK_STATUS(pbd
)) {
824 prb_freeze_queue(pkc
, po
);
830 * open this block and return the offset where the first packet
831 * needs to get stored.
833 prb_open_block(pkc
, pbd
);
834 return (void *)pkc
->nxt_offset
;
837 static void prb_retire_current_block(struct tpacket_kbdq_core
*pkc
,
838 struct packet_sock
*po
, unsigned int status
)
840 struct tpacket_block_desc
*pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
842 /* retire/close the current block */
843 if (likely(TP_STATUS_KERNEL
== BLOCK_STATUS(pbd
))) {
845 * Plug the case where copy_bits() is in progress on
846 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
847 * have space to copy the pkt in the current block and
848 * called prb_retire_current_block()
850 * We don't need to worry about the TMO case because
851 * the timer-handler already handled this case.
853 if (!(status
& TP_STATUS_BLK_TMO
)) {
854 while (atomic_read(&pkc
->blk_fill_in_prog
)) {
855 /* Waiting for skb_copy_bits to finish... */
859 prb_close_block(pkc
, pbd
, po
, status
);
864 static int prb_curr_blk_in_use(struct tpacket_kbdq_core
*pkc
,
865 struct tpacket_block_desc
*pbd
)
867 return TP_STATUS_USER
& BLOCK_STATUS(pbd
);
870 static int prb_queue_frozen(struct tpacket_kbdq_core
*pkc
)
872 return pkc
->reset_pending_on_curr_blk
;
875 static void prb_clear_blk_fill_status(struct packet_ring_buffer
*rb
)
877 struct tpacket_kbdq_core
*pkc
= GET_PBDQC_FROM_RB(rb
);
878 atomic_dec(&pkc
->blk_fill_in_prog
);
881 static void prb_fill_rxhash(struct tpacket_kbdq_core
*pkc
,
882 struct tpacket3_hdr
*ppd
)
884 ppd
->hv1
.tp_rxhash
= skb_get_rxhash(pkc
->skb
);
887 static void prb_clear_rxhash(struct tpacket_kbdq_core
*pkc
,
888 struct tpacket3_hdr
*ppd
)
890 ppd
->hv1
.tp_rxhash
= 0;
893 static void prb_fill_vlan_info(struct tpacket_kbdq_core
*pkc
,
894 struct tpacket3_hdr
*ppd
)
896 if (vlan_tx_tag_present(pkc
->skb
)) {
897 ppd
->hv1
.tp_vlan_tci
= vlan_tx_tag_get(pkc
->skb
);
898 ppd
->tp_status
= TP_STATUS_VLAN_VALID
;
900 ppd
->hv1
.tp_vlan_tci
= 0;
901 ppd
->tp_status
= TP_STATUS_AVAILABLE
;
905 static void prb_run_all_ft_ops(struct tpacket_kbdq_core
*pkc
,
906 struct tpacket3_hdr
*ppd
)
908 prb_fill_vlan_info(pkc
, ppd
);
910 if (pkc
->feature_req_word
& TP_FT_REQ_FILL_RXHASH
)
911 prb_fill_rxhash(pkc
, ppd
);
913 prb_clear_rxhash(pkc
, ppd
);
916 static void prb_fill_curr_block(char *curr
,
917 struct tpacket_kbdq_core
*pkc
,
918 struct tpacket_block_desc
*pbd
,
921 struct tpacket3_hdr
*ppd
;
923 ppd
= (struct tpacket3_hdr
*)curr
;
924 ppd
->tp_next_offset
= TOTAL_PKT_LEN_INCL_ALIGN(len
);
926 pkc
->nxt_offset
+= TOTAL_PKT_LEN_INCL_ALIGN(len
);
927 BLOCK_LEN(pbd
) += TOTAL_PKT_LEN_INCL_ALIGN(len
);
928 BLOCK_NUM_PKTS(pbd
) += 1;
929 atomic_inc(&pkc
->blk_fill_in_prog
);
930 prb_run_all_ft_ops(pkc
, ppd
);
933 /* Assumes caller has the sk->rx_queue.lock */
934 static void *__packet_lookup_frame_in_block(struct packet_sock
*po
,
940 struct tpacket_kbdq_core
*pkc
;
941 struct tpacket_block_desc
*pbd
;
944 pkc
= GET_PBDQC_FROM_RB(&po
->rx_ring
);
945 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
947 /* Queue is frozen when user space is lagging behind */
948 if (prb_queue_frozen(pkc
)) {
950 * Check if that last block which caused the queue to freeze,
951 * is still in_use by user-space.
953 if (prb_curr_blk_in_use(pkc
, pbd
)) {
954 /* Can't record this packet */
958 * Ok, the block was released by user-space.
959 * Now let's open that block.
960 * opening a block also thaws the queue.
961 * Thawing is a side effect.
963 prb_open_block(pkc
, pbd
);
968 curr
= pkc
->nxt_offset
;
970 end
= (char *)pbd
+ pkc
->kblk_size
;
972 /* first try the current block */
973 if (curr
+TOTAL_PKT_LEN_INCL_ALIGN(len
) < end
) {
974 prb_fill_curr_block(curr
, pkc
, pbd
, len
);
978 /* Ok, close the current block */
979 prb_retire_current_block(pkc
, po
, 0);
981 /* Now, try to dispatch the next block */
982 curr
= (char *)prb_dispatch_next_block(pkc
, po
);
984 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
985 prb_fill_curr_block(curr
, pkc
, pbd
, len
);
990 * No free blocks are available.user_space hasn't caught up yet.
991 * Queue was just frozen and now this packet will get dropped.
996 static void *packet_current_rx_frame(struct packet_sock
*po
,
998 int status
, unsigned int len
)
1001 switch (po
->tp_version
) {
1004 curr
= packet_lookup_frame(po
, &po
->rx_ring
,
1005 po
->rx_ring
.head
, status
);
1008 return __packet_lookup_frame_in_block(po
, skb
, status
, len
);
1010 WARN(1, "TPACKET version not supported\n");
1016 static void *prb_lookup_block(struct packet_sock
*po
,
1017 struct packet_ring_buffer
*rb
,
1021 struct tpacket_kbdq_core
*pkc
= GET_PBDQC_FROM_RB(rb
);
1022 struct tpacket_block_desc
*pbd
= GET_PBLOCK_DESC(pkc
, idx
);
1024 if (status
!= BLOCK_STATUS(pbd
))
1029 static int prb_previous_blk_num(struct packet_ring_buffer
*rb
)
1032 if (rb
->prb_bdqc
.kactive_blk_num
)
1033 prev
= rb
->prb_bdqc
.kactive_blk_num
-1;
1035 prev
= rb
->prb_bdqc
.knum_blocks
-1;
1039 /* Assumes caller has held the rx_queue.lock */
1040 static void *__prb_previous_block(struct packet_sock
*po
,
1041 struct packet_ring_buffer
*rb
,
1044 unsigned int previous
= prb_previous_blk_num(rb
);
1045 return prb_lookup_block(po
, rb
, previous
, status
);
1048 static void *packet_previous_rx_frame(struct packet_sock
*po
,
1049 struct packet_ring_buffer
*rb
,
1052 if (po
->tp_version
<= TPACKET_V2
)
1053 return packet_previous_frame(po
, rb
, status
);
1055 return __prb_previous_block(po
, rb
, status
);
1058 static void packet_increment_rx_head(struct packet_sock
*po
,
1059 struct packet_ring_buffer
*rb
)
1061 switch (po
->tp_version
) {
1064 return packet_increment_head(rb
);
1067 WARN(1, "TPACKET version not supported.\n");
1073 static void *packet_previous_frame(struct packet_sock
*po
,
1074 struct packet_ring_buffer
*rb
,
1077 unsigned int previous
= rb
->head
? rb
->head
- 1 : rb
->frame_max
;
1078 return packet_lookup_frame(po
, rb
, previous
, status
);
1081 static void packet_increment_head(struct packet_ring_buffer
*buff
)
1083 buff
->head
= buff
->head
!= buff
->frame_max
? buff
->head
+1 : 0;
1086 static bool packet_rcv_has_room(struct packet_sock
*po
, struct sk_buff
*skb
)
1088 struct sock
*sk
= &po
->sk
;
1091 if (po
->prot_hook
.func
!= tpacket_rcv
)
1092 return (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
)
1095 spin_lock(&sk
->sk_receive_queue
.lock
);
1096 if (po
->tp_version
== TPACKET_V3
)
1097 has_room
= prb_lookup_block(po
, &po
->rx_ring
,
1098 po
->rx_ring
.prb_bdqc
.kactive_blk_num
,
1101 has_room
= packet_lookup_frame(po
, &po
->rx_ring
,
1104 spin_unlock(&sk
->sk_receive_queue
.lock
);
1109 static void packet_sock_destruct(struct sock
*sk
)
1111 skb_queue_purge(&sk
->sk_error_queue
);
1113 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
1114 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
1116 if (!sock_flag(sk
, SOCK_DEAD
)) {
1117 pr_err("Attempt to release alive packet socket: %p\n", sk
);
1121 sk_refcnt_debug_dec(sk
);
1124 static int fanout_rr_next(struct packet_fanout
*f
, unsigned int num
)
1126 int x
= atomic_read(&f
->rr_cur
) + 1;
1134 static unsigned int fanout_demux_hash(struct packet_fanout
*f
,
1135 struct sk_buff
*skb
,
1138 return reciprocal_divide(skb
->rxhash
, num
);
1141 static unsigned int fanout_demux_lb(struct packet_fanout
*f
,
1142 struct sk_buff
*skb
,
1147 cur
= atomic_read(&f
->rr_cur
);
1148 while ((old
= atomic_cmpxchg(&f
->rr_cur
, cur
,
1149 fanout_rr_next(f
, num
))) != cur
)
1154 static unsigned int fanout_demux_cpu(struct packet_fanout
*f
,
1155 struct sk_buff
*skb
,
1158 return smp_processor_id() % num
;
1161 static unsigned int fanout_demux_rnd(struct packet_fanout
*f
,
1162 struct sk_buff
*skb
,
1165 return reciprocal_divide(prandom_u32(), num
);
1168 static unsigned int fanout_demux_rollover(struct packet_fanout
*f
,
1169 struct sk_buff
*skb
,
1170 unsigned int idx
, unsigned int skip
,
1175 i
= j
= min_t(int, f
->next
[idx
], num
- 1);
1177 if (i
!= skip
&& packet_rcv_has_room(pkt_sk(f
->arr
[i
]), skb
)) {
1189 static bool fanout_has_flag(struct packet_fanout
*f
, u16 flag
)
1191 return f
->flags
& (flag
>> 8);
1194 static int packet_rcv_fanout(struct sk_buff
*skb
, struct net_device
*dev
,
1195 struct packet_type
*pt
, struct net_device
*orig_dev
)
1197 struct packet_fanout
*f
= pt
->af_packet_priv
;
1198 unsigned int num
= f
->num_members
;
1199 struct packet_sock
*po
;
1202 if (!net_eq(dev_net(dev
), read_pnet(&f
->net
)) ||
1209 case PACKET_FANOUT_HASH
:
1211 if (fanout_has_flag(f
, PACKET_FANOUT_FLAG_DEFRAG
)) {
1212 skb
= ip_check_defrag(skb
, IP_DEFRAG_AF_PACKET
);
1216 skb_get_rxhash(skb
);
1217 idx
= fanout_demux_hash(f
, skb
, num
);
1219 case PACKET_FANOUT_LB
:
1220 idx
= fanout_demux_lb(f
, skb
, num
);
1222 case PACKET_FANOUT_CPU
:
1223 idx
= fanout_demux_cpu(f
, skb
, num
);
1225 case PACKET_FANOUT_RND
:
1226 idx
= fanout_demux_rnd(f
, skb
, num
);
1228 case PACKET_FANOUT_ROLLOVER
:
1229 idx
= fanout_demux_rollover(f
, skb
, 0, (unsigned int) -1, num
);
1233 po
= pkt_sk(f
->arr
[idx
]);
1234 if (fanout_has_flag(f
, PACKET_FANOUT_FLAG_ROLLOVER
) &&
1235 unlikely(!packet_rcv_has_room(po
, skb
))) {
1236 idx
= fanout_demux_rollover(f
, skb
, idx
, idx
, num
);
1237 po
= pkt_sk(f
->arr
[idx
]);
1240 return po
->prot_hook
.func(skb
, dev
, &po
->prot_hook
, orig_dev
);
1243 DEFINE_MUTEX(fanout_mutex
);
1244 EXPORT_SYMBOL_GPL(fanout_mutex
);
1245 static LIST_HEAD(fanout_list
);
1247 static void __fanout_link(struct sock
*sk
, struct packet_sock
*po
)
1249 struct packet_fanout
*f
= po
->fanout
;
1251 spin_lock(&f
->lock
);
1252 f
->arr
[f
->num_members
] = sk
;
1255 spin_unlock(&f
->lock
);
1258 static void __fanout_unlink(struct sock
*sk
, struct packet_sock
*po
)
1260 struct packet_fanout
*f
= po
->fanout
;
1263 spin_lock(&f
->lock
);
1264 for (i
= 0; i
< f
->num_members
; i
++) {
1265 if (f
->arr
[i
] == sk
)
1268 BUG_ON(i
>= f
->num_members
);
1269 f
->arr
[i
] = f
->arr
[f
->num_members
- 1];
1271 spin_unlock(&f
->lock
);
1274 static bool match_fanout_group(struct packet_type
*ptype
, struct sock
* sk
)
1276 if (ptype
->af_packet_priv
== (void*)((struct packet_sock
*)sk
)->fanout
)
1282 static int fanout_add(struct sock
*sk
, u16 id
, u16 type_flags
)
1284 struct packet_sock
*po
= pkt_sk(sk
);
1285 struct packet_fanout
*f
, *match
;
1286 u8 type
= type_flags
& 0xff;
1287 u8 flags
= type_flags
>> 8;
1291 case PACKET_FANOUT_ROLLOVER
:
1292 if (type_flags
& PACKET_FANOUT_FLAG_ROLLOVER
)
1294 case PACKET_FANOUT_HASH
:
1295 case PACKET_FANOUT_LB
:
1296 case PACKET_FANOUT_CPU
:
1297 case PACKET_FANOUT_RND
:
1309 mutex_lock(&fanout_mutex
);
1311 list_for_each_entry(f
, &fanout_list
, list
) {
1313 read_pnet(&f
->net
) == sock_net(sk
)) {
1319 if (match
&& match
->flags
!= flags
)
1323 match
= kzalloc(sizeof(*match
), GFP_KERNEL
);
1326 write_pnet(&match
->net
, sock_net(sk
));
1329 match
->flags
= flags
;
1330 atomic_set(&match
->rr_cur
, 0);
1331 INIT_LIST_HEAD(&match
->list
);
1332 spin_lock_init(&match
->lock
);
1333 atomic_set(&match
->sk_ref
, 0);
1334 match
->prot_hook
.type
= po
->prot_hook
.type
;
1335 match
->prot_hook
.dev
= po
->prot_hook
.dev
;
1336 match
->prot_hook
.func
= packet_rcv_fanout
;
1337 match
->prot_hook
.af_packet_priv
= match
;
1338 match
->prot_hook
.id_match
= match_fanout_group
;
1339 dev_add_pack(&match
->prot_hook
);
1340 list_add(&match
->list
, &fanout_list
);
1343 if (match
->type
== type
&&
1344 match
->prot_hook
.type
== po
->prot_hook
.type
&&
1345 match
->prot_hook
.dev
== po
->prot_hook
.dev
) {
1347 if (atomic_read(&match
->sk_ref
) < PACKET_FANOUT_MAX
) {
1348 __dev_remove_pack(&po
->prot_hook
);
1350 atomic_inc(&match
->sk_ref
);
1351 __fanout_link(sk
, po
);
1356 mutex_unlock(&fanout_mutex
);
1360 static void fanout_release(struct sock
*sk
)
1362 struct packet_sock
*po
= pkt_sk(sk
);
1363 struct packet_fanout
*f
;
1369 mutex_lock(&fanout_mutex
);
1372 if (atomic_dec_and_test(&f
->sk_ref
)) {
1374 dev_remove_pack(&f
->prot_hook
);
1377 mutex_unlock(&fanout_mutex
);
1380 static const struct proto_ops packet_ops
;
1382 static const struct proto_ops packet_ops_spkt
;
1384 static int packet_rcv_spkt(struct sk_buff
*skb
, struct net_device
*dev
,
1385 struct packet_type
*pt
, struct net_device
*orig_dev
)
1388 struct sockaddr_pkt
*spkt
;
1391 * When we registered the protocol we saved the socket in the data
1392 * field for just this event.
1395 sk
= pt
->af_packet_priv
;
1398 * Yank back the headers [hope the device set this
1399 * right or kerboom...]
1401 * Incoming packets have ll header pulled,
1404 * For outgoing ones skb->data == skb_mac_header(skb)
1405 * so that this procedure is noop.
1408 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1411 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1414 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1418 /* drop any routing info */
1421 /* drop conntrack reference */
1424 spkt
= &PACKET_SKB_CB(skb
)->sa
.pkt
;
1426 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1429 * The SOCK_PACKET socket receives _all_ frames.
1432 spkt
->spkt_family
= dev
->type
;
1433 strlcpy(spkt
->spkt_device
, dev
->name
, sizeof(spkt
->spkt_device
));
1434 spkt
->spkt_protocol
= skb
->protocol
;
1437 * Charge the memory to the socket. This is done specifically
1438 * to prevent sockets using all the memory up.
1441 if (sock_queue_rcv_skb(sk
, skb
) == 0)
1452 * Output a raw packet to a device layer. This bypasses all the other
1453 * protocol layers and you must therefore supply it with a complete frame
1456 static int packet_sendmsg_spkt(struct kiocb
*iocb
, struct socket
*sock
,
1457 struct msghdr
*msg
, size_t len
)
1459 struct sock
*sk
= sock
->sk
;
1460 struct sockaddr_pkt
*saddr
= (struct sockaddr_pkt
*)msg
->msg_name
;
1461 struct sk_buff
*skb
= NULL
;
1462 struct net_device
*dev
;
1468 * Get and verify the address.
1472 if (msg
->msg_namelen
< sizeof(struct sockaddr
))
1474 if (msg
->msg_namelen
== sizeof(struct sockaddr_pkt
))
1475 proto
= saddr
->spkt_protocol
;
1477 return -ENOTCONN
; /* SOCK_PACKET must be sent giving an address */
1480 * Find the device first to size check it
1483 saddr
->spkt_device
[sizeof(saddr
->spkt_device
) - 1] = 0;
1486 dev
= dev_get_by_name_rcu(sock_net(sk
), saddr
->spkt_device
);
1492 if (!(dev
->flags
& IFF_UP
))
1496 * You may not queue a frame bigger than the mtu. This is the lowest level
1497 * raw protocol and you must do your own fragmentation at this level.
1500 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
1501 if (!netif_supports_nofcs(dev
)) {
1502 err
= -EPROTONOSUPPORT
;
1505 extra_len
= 4; /* We're doing our own CRC */
1509 if (len
> dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
+ extra_len
)
1513 size_t reserved
= LL_RESERVED_SPACE(dev
);
1514 int tlen
= dev
->needed_tailroom
;
1515 unsigned int hhlen
= dev
->header_ops
? dev
->hard_header_len
: 0;
1518 skb
= sock_wmalloc(sk
, len
+ reserved
+ tlen
, 0, GFP_KERNEL
);
1521 /* FIXME: Save some space for broken drivers that write a hard
1522 * header at transmission time by themselves. PPP is the notable
1523 * one here. This should really be fixed at the driver level.
1525 skb_reserve(skb
, reserved
);
1526 skb_reset_network_header(skb
);
1528 /* Try to align data part correctly */
1533 skb_reset_network_header(skb
);
1535 err
= memcpy_fromiovec(skb_put(skb
, len
), msg
->msg_iov
, len
);
1541 if (len
> (dev
->mtu
+ dev
->hard_header_len
+ extra_len
)) {
1542 /* Earlier code assumed this would be a VLAN pkt,
1543 * double-check this now that we have the actual
1546 struct ethhdr
*ehdr
;
1547 skb_reset_mac_header(skb
);
1548 ehdr
= eth_hdr(skb
);
1549 if (ehdr
->h_proto
!= htons(ETH_P_8021Q
)) {
1555 skb
->protocol
= proto
;
1557 skb
->priority
= sk
->sk_priority
;
1558 skb
->mark
= sk
->sk_mark
;
1560 sock_tx_timestamp(sk
, &skb_shinfo(skb
)->tx_flags
);
1562 if (unlikely(extra_len
== 4))
1565 skb_probe_transport_header(skb
, 0);
1567 dev_queue_xmit(skb
);
1578 static unsigned int run_filter(const struct sk_buff
*skb
,
1579 const struct sock
*sk
,
1582 struct sk_filter
*filter
;
1585 filter
= rcu_dereference(sk
->sk_filter
);
1587 res
= SK_RUN_FILTER(filter
, skb
);
1594 * This function makes lazy skb cloning in hope that most of packets
1595 * are discarded by BPF.
1597 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1598 * and skb->cb are mangled. It works because (and until) packets
1599 * falling here are owned by current CPU. Output packets are cloned
1600 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1601 * sequencially, so that if we return skb to original state on exit,
1602 * we will not harm anyone.
1605 static int packet_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
1606 struct packet_type
*pt
, struct net_device
*orig_dev
)
1609 struct sockaddr_ll
*sll
;
1610 struct packet_sock
*po
;
1611 u8
*skb_head
= skb
->data
;
1612 int skb_len
= skb
->len
;
1613 unsigned int snaplen
, res
;
1615 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1618 sk
= pt
->af_packet_priv
;
1621 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1626 if (dev
->header_ops
) {
1627 /* The device has an explicit notion of ll header,
1628 * exported to higher levels.
1630 * Otherwise, the device hides details of its frame
1631 * structure, so that corresponding packet head is
1632 * never delivered to user.
1634 if (sk
->sk_type
!= SOCK_DGRAM
)
1635 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1636 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
1637 /* Special case: outgoing packets have ll header at head */
1638 skb_pull(skb
, skb_network_offset(skb
));
1644 res
= run_filter(skb
, sk
, snaplen
);
1646 goto drop_n_restore
;
1650 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
1653 if (skb_shared(skb
)) {
1654 struct sk_buff
*nskb
= skb_clone(skb
, GFP_ATOMIC
);
1658 if (skb_head
!= skb
->data
) {
1659 skb
->data
= skb_head
;
1666 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb
)) + MAX_ADDR_LEN
- 8 >
1669 sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
1670 sll
->sll_family
= AF_PACKET
;
1671 sll
->sll_hatype
= dev
->type
;
1672 sll
->sll_protocol
= skb
->protocol
;
1673 sll
->sll_pkttype
= skb
->pkt_type
;
1674 if (unlikely(po
->origdev
))
1675 sll
->sll_ifindex
= orig_dev
->ifindex
;
1677 sll
->sll_ifindex
= dev
->ifindex
;
1679 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
1681 PACKET_SKB_CB(skb
)->origlen
= skb
->len
;
1683 if (pskb_trim(skb
, snaplen
))
1686 skb_set_owner_r(skb
, sk
);
1690 /* drop conntrack reference */
1693 spin_lock(&sk
->sk_receive_queue
.lock
);
1694 po
->stats
.stats1
.tp_packets
++;
1695 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
1696 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1697 spin_unlock(&sk
->sk_receive_queue
.lock
);
1698 sk
->sk_data_ready(sk
, skb
->len
);
1702 spin_lock(&sk
->sk_receive_queue
.lock
);
1703 po
->stats
.stats1
.tp_drops
++;
1704 atomic_inc(&sk
->sk_drops
);
1705 spin_unlock(&sk
->sk_receive_queue
.lock
);
1708 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
1709 skb
->data
= skb_head
;
1717 static int tpacket_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
1718 struct packet_type
*pt
, struct net_device
*orig_dev
)
1721 struct packet_sock
*po
;
1722 struct sockaddr_ll
*sll
;
1723 union tpacket_uhdr h
;
1724 u8
*skb_head
= skb
->data
;
1725 int skb_len
= skb
->len
;
1726 unsigned int snaplen
, res
;
1727 unsigned long status
= TP_STATUS_USER
;
1728 unsigned short macoff
, netoff
, hdrlen
;
1729 struct sk_buff
*copy_skb
= NULL
;
1733 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1736 sk
= pt
->af_packet_priv
;
1739 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1742 if (dev
->header_ops
) {
1743 if (sk
->sk_type
!= SOCK_DGRAM
)
1744 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1745 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
1746 /* Special case: outgoing packets have ll header at head */
1747 skb_pull(skb
, skb_network_offset(skb
));
1751 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1752 status
|= TP_STATUS_CSUMNOTREADY
;
1756 res
= run_filter(skb
, sk
, snaplen
);
1758 goto drop_n_restore
;
1762 if (sk
->sk_type
== SOCK_DGRAM
) {
1763 macoff
= netoff
= TPACKET_ALIGN(po
->tp_hdrlen
) + 16 +
1766 unsigned int maclen
= skb_network_offset(skb
);
1767 netoff
= TPACKET_ALIGN(po
->tp_hdrlen
+
1768 (maclen
< 16 ? 16 : maclen
)) +
1770 macoff
= netoff
- maclen
;
1772 if (po
->tp_version
<= TPACKET_V2
) {
1773 if (macoff
+ snaplen
> po
->rx_ring
.frame_size
) {
1774 if (po
->copy_thresh
&&
1775 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1776 if (skb_shared(skb
)) {
1777 copy_skb
= skb_clone(skb
, GFP_ATOMIC
);
1779 copy_skb
= skb_get(skb
);
1780 skb_head
= skb
->data
;
1783 skb_set_owner_r(copy_skb
, sk
);
1785 snaplen
= po
->rx_ring
.frame_size
- macoff
;
1786 if ((int)snaplen
< 0)
1790 spin_lock(&sk
->sk_receive_queue
.lock
);
1791 h
.raw
= packet_current_rx_frame(po
, skb
,
1792 TP_STATUS_KERNEL
, (macoff
+snaplen
));
1795 if (po
->tp_version
<= TPACKET_V2
) {
1796 packet_increment_rx_head(po
, &po
->rx_ring
);
1798 * LOSING will be reported till you read the stats,
1799 * because it's COR - Clear On Read.
1800 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1803 if (po
->stats
.stats1
.tp_drops
)
1804 status
|= TP_STATUS_LOSING
;
1806 po
->stats
.stats1
.tp_packets
++;
1808 status
|= TP_STATUS_COPY
;
1809 __skb_queue_tail(&sk
->sk_receive_queue
, copy_skb
);
1811 spin_unlock(&sk
->sk_receive_queue
.lock
);
1813 skb_copy_bits(skb
, 0, h
.raw
+ macoff
, snaplen
);
1815 if (!(ts_status
= tpacket_get_timestamp(skb
, &ts
, po
->tp_tstamp
)))
1816 getnstimeofday(&ts
);
1818 status
|= ts_status
;
1820 switch (po
->tp_version
) {
1822 h
.h1
->tp_len
= skb
->len
;
1823 h
.h1
->tp_snaplen
= snaplen
;
1824 h
.h1
->tp_mac
= macoff
;
1825 h
.h1
->tp_net
= netoff
;
1826 h
.h1
->tp_sec
= ts
.tv_sec
;
1827 h
.h1
->tp_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
1828 hdrlen
= sizeof(*h
.h1
);
1831 h
.h2
->tp_len
= skb
->len
;
1832 h
.h2
->tp_snaplen
= snaplen
;
1833 h
.h2
->tp_mac
= macoff
;
1834 h
.h2
->tp_net
= netoff
;
1835 h
.h2
->tp_sec
= ts
.tv_sec
;
1836 h
.h2
->tp_nsec
= ts
.tv_nsec
;
1837 if (vlan_tx_tag_present(skb
)) {
1838 h
.h2
->tp_vlan_tci
= vlan_tx_tag_get(skb
);
1839 status
|= TP_STATUS_VLAN_VALID
;
1841 h
.h2
->tp_vlan_tci
= 0;
1843 h
.h2
->tp_padding
= 0;
1844 hdrlen
= sizeof(*h
.h2
);
1847 /* tp_nxt_offset,vlan are already populated above.
1848 * So DONT clear those fields here
1850 h
.h3
->tp_status
|= status
;
1851 h
.h3
->tp_len
= skb
->len
;
1852 h
.h3
->tp_snaplen
= snaplen
;
1853 h
.h3
->tp_mac
= macoff
;
1854 h
.h3
->tp_net
= netoff
;
1855 h
.h3
->tp_sec
= ts
.tv_sec
;
1856 h
.h3
->tp_nsec
= ts
.tv_nsec
;
1857 hdrlen
= sizeof(*h
.h3
);
1863 sll
= h
.raw
+ TPACKET_ALIGN(hdrlen
);
1864 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
1865 sll
->sll_family
= AF_PACKET
;
1866 sll
->sll_hatype
= dev
->type
;
1867 sll
->sll_protocol
= skb
->protocol
;
1868 sll
->sll_pkttype
= skb
->pkt_type
;
1869 if (unlikely(po
->origdev
))
1870 sll
->sll_ifindex
= orig_dev
->ifindex
;
1872 sll
->sll_ifindex
= dev
->ifindex
;
1875 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1879 if (po
->tp_version
<= TPACKET_V2
) {
1880 end
= (u8
*)PAGE_ALIGN((unsigned long)h
.raw
1881 + macoff
+ snaplen
);
1882 for (start
= h
.raw
; start
< end
; start
+= PAGE_SIZE
)
1883 flush_dcache_page(pgv_to_page(start
));
1888 if (po
->tp_version
<= TPACKET_V2
)
1889 __packet_set_status(po
, h
.raw
, status
);
1891 prb_clear_blk_fill_status(&po
->rx_ring
);
1893 sk
->sk_data_ready(sk
, 0);
1896 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
1897 skb
->data
= skb_head
;
1905 po
->stats
.stats1
.tp_drops
++;
1906 spin_unlock(&sk
->sk_receive_queue
.lock
);
1908 sk
->sk_data_ready(sk
, 0);
1909 kfree_skb(copy_skb
);
1910 goto drop_n_restore
;
1913 static void tpacket_destruct_skb(struct sk_buff
*skb
)
1915 struct packet_sock
*po
= pkt_sk(skb
->sk
);
1918 if (likely(po
->tx_ring
.pg_vec
)) {
1921 ph
= skb_shinfo(skb
)->destructor_arg
;
1922 BUG_ON(atomic_read(&po
->tx_ring
.pending
) == 0);
1923 atomic_dec(&po
->tx_ring
.pending
);
1925 ts
= __packet_set_timestamp(po
, ph
, skb
);
1926 __packet_set_status(po
, ph
, TP_STATUS_AVAILABLE
| ts
);
1932 static int tpacket_fill_skb(struct packet_sock
*po
, struct sk_buff
*skb
,
1933 void *frame
, struct net_device
*dev
, int size_max
,
1934 __be16 proto
, unsigned char *addr
, int hlen
)
1936 union tpacket_uhdr ph
;
1937 int to_write
, offset
, len
, tp_len
, nr_frags
, len_max
;
1938 struct socket
*sock
= po
->sk
.sk_socket
;
1945 skb
->protocol
= proto
;
1947 skb
->priority
= po
->sk
.sk_priority
;
1948 skb
->mark
= po
->sk
.sk_mark
;
1949 sock_tx_timestamp(&po
->sk
, &skb_shinfo(skb
)->tx_flags
);
1950 skb_shinfo(skb
)->destructor_arg
= ph
.raw
;
1952 switch (po
->tp_version
) {
1954 tp_len
= ph
.h2
->tp_len
;
1957 tp_len
= ph
.h1
->tp_len
;
1960 if (unlikely(tp_len
> size_max
)) {
1961 pr_err("packet size is too long (%d > %d)\n", tp_len
, size_max
);
1965 skb_reserve(skb
, hlen
);
1966 skb_reset_network_header(skb
);
1967 skb_probe_transport_header(skb
, 0);
1969 if (po
->tp_tx_has_off
) {
1970 int off_min
, off_max
, off
;
1971 off_min
= po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
1972 off_max
= po
->tx_ring
.frame_size
- tp_len
;
1973 if (sock
->type
== SOCK_DGRAM
) {
1974 switch (po
->tp_version
) {
1976 off
= ph
.h2
->tp_net
;
1979 off
= ph
.h1
->tp_net
;
1983 switch (po
->tp_version
) {
1985 off
= ph
.h2
->tp_mac
;
1988 off
= ph
.h1
->tp_mac
;
1992 if (unlikely((off
< off_min
) || (off_max
< off
)))
1994 data
= ph
.raw
+ off
;
1996 data
= ph
.raw
+ po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
2000 if (sock
->type
== SOCK_DGRAM
) {
2001 err
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
,
2003 if (unlikely(err
< 0))
2005 } else if (dev
->hard_header_len
) {
2006 /* net device doesn't like empty head */
2007 if (unlikely(tp_len
<= dev
->hard_header_len
)) {
2008 pr_err("packet size is too short (%d < %d)\n",
2009 tp_len
, dev
->hard_header_len
);
2013 skb_push(skb
, dev
->hard_header_len
);
2014 err
= skb_store_bits(skb
, 0, data
,
2015 dev
->hard_header_len
);
2019 data
+= dev
->hard_header_len
;
2020 to_write
-= dev
->hard_header_len
;
2023 offset
= offset_in_page(data
);
2024 len_max
= PAGE_SIZE
- offset
;
2025 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2027 skb
->data_len
= to_write
;
2028 skb
->len
+= to_write
;
2029 skb
->truesize
+= to_write
;
2030 atomic_add(to_write
, &po
->sk
.sk_wmem_alloc
);
2032 while (likely(to_write
)) {
2033 nr_frags
= skb_shinfo(skb
)->nr_frags
;
2035 if (unlikely(nr_frags
>= MAX_SKB_FRAGS
)) {
2036 pr_err("Packet exceed the number of skb frags(%lu)\n",
2041 page
= pgv_to_page(data
);
2043 flush_dcache_page(page
);
2045 skb_fill_page_desc(skb
, nr_frags
, page
, offset
, len
);
2048 len_max
= PAGE_SIZE
;
2049 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2055 static int tpacket_snd(struct packet_sock
*po
, struct msghdr
*msg
)
2057 struct sk_buff
*skb
;
2058 struct net_device
*dev
;
2060 bool need_rls_dev
= false;
2061 int err
, reserve
= 0;
2063 struct sockaddr_ll
*saddr
= (struct sockaddr_ll
*)msg
->msg_name
;
2064 int tp_len
, size_max
;
2065 unsigned char *addr
;
2067 int status
= TP_STATUS_AVAILABLE
;
2070 mutex_lock(&po
->pg_vec_lock
);
2072 if (saddr
== NULL
) {
2073 dev
= po
->prot_hook
.dev
;
2078 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2080 if (msg
->msg_namelen
< (saddr
->sll_halen
2081 + offsetof(struct sockaddr_ll
,
2084 proto
= saddr
->sll_protocol
;
2085 addr
= saddr
->sll_addr
;
2086 dev
= dev_get_by_index(sock_net(&po
->sk
), saddr
->sll_ifindex
);
2087 need_rls_dev
= true;
2091 if (unlikely(dev
== NULL
))
2094 reserve
= dev
->hard_header_len
;
2097 if (unlikely(!(dev
->flags
& IFF_UP
)))
2100 size_max
= po
->tx_ring
.frame_size
2101 - (po
->tp_hdrlen
- sizeof(struct sockaddr_ll
));
2103 if (size_max
> dev
->mtu
+ reserve
)
2104 size_max
= dev
->mtu
+ reserve
;
2107 ph
= packet_current_frame(po
, &po
->tx_ring
,
2108 TP_STATUS_SEND_REQUEST
);
2110 if (unlikely(ph
== NULL
)) {
2115 status
= TP_STATUS_SEND_REQUEST
;
2116 hlen
= LL_RESERVED_SPACE(dev
);
2117 tlen
= dev
->needed_tailroom
;
2118 skb
= sock_alloc_send_skb(&po
->sk
,
2119 hlen
+ tlen
+ sizeof(struct sockaddr_ll
),
2122 if (unlikely(skb
== NULL
))
2125 tp_len
= tpacket_fill_skb(po
, skb
, ph
, dev
, size_max
, proto
,
2128 if (unlikely(tp_len
< 0)) {
2130 __packet_set_status(po
, ph
,
2131 TP_STATUS_AVAILABLE
);
2132 packet_increment_head(&po
->tx_ring
);
2136 status
= TP_STATUS_WRONG_FORMAT
;
2142 skb
->destructor
= tpacket_destruct_skb
;
2143 __packet_set_status(po
, ph
, TP_STATUS_SENDING
);
2144 atomic_inc(&po
->tx_ring
.pending
);
2146 status
= TP_STATUS_SEND_REQUEST
;
2147 err
= dev_queue_xmit(skb
);
2148 if (unlikely(err
> 0)) {
2149 err
= net_xmit_errno(err
);
2150 if (err
&& __packet_get_status(po
, ph
) ==
2151 TP_STATUS_AVAILABLE
) {
2152 /* skb was destructed already */
2157 * skb was dropped but not destructed yet;
2158 * let's treat it like congestion or err < 0
2162 packet_increment_head(&po
->tx_ring
);
2164 } while (likely((ph
!= NULL
) ||
2165 ((!(msg
->msg_flags
& MSG_DONTWAIT
)) &&
2166 (atomic_read(&po
->tx_ring
.pending
))))
2173 __packet_set_status(po
, ph
, status
);
2179 mutex_unlock(&po
->pg_vec_lock
);
2183 static struct sk_buff
*packet_alloc_skb(struct sock
*sk
, size_t prepad
,
2184 size_t reserve
, size_t len
,
2185 size_t linear
, int noblock
,
2188 struct sk_buff
*skb
;
2190 /* Under a page? Don't bother with paged skb. */
2191 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
2194 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
2199 skb_reserve(skb
, reserve
);
2200 skb_put(skb
, linear
);
2201 skb
->data_len
= len
- linear
;
2202 skb
->len
+= len
- linear
;
2207 static int packet_snd(struct socket
*sock
,
2208 struct msghdr
*msg
, size_t len
)
2210 struct sock
*sk
= sock
->sk
;
2211 struct sockaddr_ll
*saddr
= (struct sockaddr_ll
*)msg
->msg_name
;
2212 struct sk_buff
*skb
;
2213 struct net_device
*dev
;
2215 bool need_rls_dev
= false;
2216 unsigned char *addr
;
2217 int err
, reserve
= 0;
2218 struct virtio_net_hdr vnet_hdr
= { 0 };
2221 struct packet_sock
*po
= pkt_sk(sk
);
2222 unsigned short gso_type
= 0;
2227 * Get and verify the address.
2230 if (saddr
== NULL
) {
2231 dev
= po
->prot_hook
.dev
;
2236 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2238 if (msg
->msg_namelen
< (saddr
->sll_halen
+ offsetof(struct sockaddr_ll
, sll_addr
)))
2240 proto
= saddr
->sll_protocol
;
2241 addr
= saddr
->sll_addr
;
2242 dev
= dev_get_by_index(sock_net(sk
), saddr
->sll_ifindex
);
2243 need_rls_dev
= true;
2249 if (sock
->type
== SOCK_RAW
)
2250 reserve
= dev
->hard_header_len
;
2253 if (!(dev
->flags
& IFF_UP
))
2256 if (po
->has_vnet_hdr
) {
2257 vnet_hdr_len
= sizeof(vnet_hdr
);
2260 if (len
< vnet_hdr_len
)
2263 len
-= vnet_hdr_len
;
2265 err
= memcpy_fromiovec((void *)&vnet_hdr
, msg
->msg_iov
,
2270 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
2271 (vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
2273 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
2274 vnet_hdr
.csum_offset
+ 2;
2277 if (vnet_hdr
.hdr_len
> len
)
2280 if (vnet_hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
2281 switch (vnet_hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
2282 case VIRTIO_NET_HDR_GSO_TCPV4
:
2283 gso_type
= SKB_GSO_TCPV4
;
2285 case VIRTIO_NET_HDR_GSO_TCPV6
:
2286 gso_type
= SKB_GSO_TCPV6
;
2288 case VIRTIO_NET_HDR_GSO_UDP
:
2289 gso_type
= SKB_GSO_UDP
;
2295 if (vnet_hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
2296 gso_type
|= SKB_GSO_TCP_ECN
;
2298 if (vnet_hdr
.gso_size
== 0)
2304 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
2305 if (!netif_supports_nofcs(dev
)) {
2306 err
= -EPROTONOSUPPORT
;
2309 extra_len
= 4; /* We're doing our own CRC */
2313 if (!gso_type
&& (len
> dev
->mtu
+ reserve
+ VLAN_HLEN
+ extra_len
))
2317 hlen
= LL_RESERVED_SPACE(dev
);
2318 tlen
= dev
->needed_tailroom
;
2319 skb
= packet_alloc_skb(sk
, hlen
+ tlen
, hlen
, len
, vnet_hdr
.hdr_len
,
2320 msg
->msg_flags
& MSG_DONTWAIT
, &err
);
2324 skb_set_network_header(skb
, reserve
);
2327 if (sock
->type
== SOCK_DGRAM
&&
2328 (offset
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
, NULL
, len
)) < 0)
2331 /* Returns -EFAULT on error */
2332 err
= skb_copy_datagram_from_iovec(skb
, offset
, msg
->msg_iov
, 0, len
);
2336 sock_tx_timestamp(sk
, &skb_shinfo(skb
)->tx_flags
);
2338 if (!gso_type
&& (len
> dev
->mtu
+ reserve
+ extra_len
)) {
2339 /* Earlier code assumed this would be a VLAN pkt,
2340 * double-check this now that we have the actual
2343 struct ethhdr
*ehdr
;
2344 skb_reset_mac_header(skb
);
2345 ehdr
= eth_hdr(skb
);
2346 if (ehdr
->h_proto
!= htons(ETH_P_8021Q
)) {
2352 skb
->protocol
= proto
;
2354 skb
->priority
= sk
->sk_priority
;
2355 skb
->mark
= sk
->sk_mark
;
2357 if (po
->has_vnet_hdr
) {
2358 if (vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
2359 if (!skb_partial_csum_set(skb
, vnet_hdr
.csum_start
,
2360 vnet_hdr
.csum_offset
)) {
2366 skb_shinfo(skb
)->gso_size
= vnet_hdr
.gso_size
;
2367 skb_shinfo(skb
)->gso_type
= gso_type
;
2369 /* Header must be checked, and gso_segs computed. */
2370 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
2371 skb_shinfo(skb
)->gso_segs
= 0;
2373 len
+= vnet_hdr_len
;
2376 skb_probe_transport_header(skb
, reserve
);
2378 if (unlikely(extra_len
== 4))
2385 err
= dev_queue_xmit(skb
);
2386 if (err
> 0 && (err
= net_xmit_errno(err
)) != 0)
2397 if (dev
&& need_rls_dev
)
2403 static int packet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
2404 struct msghdr
*msg
, size_t len
)
2406 struct sock
*sk
= sock
->sk
;
2407 struct packet_sock
*po
= pkt_sk(sk
);
2408 if (po
->tx_ring
.pg_vec
)
2409 return tpacket_snd(po
, msg
);
2411 return packet_snd(sock
, msg
, len
);
2415 * Close a PACKET socket. This is fairly simple. We immediately go
2416 * to 'closed' state and remove our protocol entry in the device list.
2419 static int packet_release(struct socket
*sock
)
2421 struct sock
*sk
= sock
->sk
;
2422 struct packet_sock
*po
;
2424 union tpacket_req_u req_u
;
2432 mutex_lock(&net
->packet
.sklist_lock
);
2433 sk_del_node_init_rcu(sk
);
2434 mutex_unlock(&net
->packet
.sklist_lock
);
2437 sock_prot_inuse_add(net
, sk
->sk_prot
, -1);
2440 spin_lock(&po
->bind_lock
);
2441 unregister_prot_hook(sk
, false);
2442 if (po
->prot_hook
.dev
) {
2443 dev_put(po
->prot_hook
.dev
);
2444 po
->prot_hook
.dev
= NULL
;
2446 spin_unlock(&po
->bind_lock
);
2448 packet_flush_mclist(sk
);
2450 if (po
->rx_ring
.pg_vec
) {
2451 memset(&req_u
, 0, sizeof(req_u
));
2452 packet_set_ring(sk
, &req_u
, 1, 0);
2455 if (po
->tx_ring
.pg_vec
) {
2456 memset(&req_u
, 0, sizeof(req_u
));
2457 packet_set_ring(sk
, &req_u
, 1, 1);
2464 * Now the socket is dead. No more input will appear.
2471 skb_queue_purge(&sk
->sk_receive_queue
);
2472 sk_refcnt_debug_release(sk
);
2479 * Attach a packet hook.
2482 static int packet_do_bind(struct sock
*sk
, struct net_device
*dev
, __be16 protocol
)
2484 struct packet_sock
*po
= pkt_sk(sk
);
2495 spin_lock(&po
->bind_lock
);
2496 unregister_prot_hook(sk
, true);
2498 po
->prot_hook
.type
= protocol
;
2499 if (po
->prot_hook
.dev
)
2500 dev_put(po
->prot_hook
.dev
);
2501 po
->prot_hook
.dev
= dev
;
2503 po
->ifindex
= dev
? dev
->ifindex
: 0;
2508 if (!dev
|| (dev
->flags
& IFF_UP
)) {
2509 register_prot_hook(sk
);
2511 sk
->sk_err
= ENETDOWN
;
2512 if (!sock_flag(sk
, SOCK_DEAD
))
2513 sk
->sk_error_report(sk
);
2517 spin_unlock(&po
->bind_lock
);
2523 * Bind a packet socket to a device
2526 static int packet_bind_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
2529 struct sock
*sk
= sock
->sk
;
2531 struct net_device
*dev
;
2538 if (addr_len
!= sizeof(struct sockaddr
))
2540 strlcpy(name
, uaddr
->sa_data
, sizeof(name
));
2542 dev
= dev_get_by_name(sock_net(sk
), name
);
2544 err
= packet_do_bind(sk
, dev
, pkt_sk(sk
)->num
);
2548 static int packet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
2550 struct sockaddr_ll
*sll
= (struct sockaddr_ll
*)uaddr
;
2551 struct sock
*sk
= sock
->sk
;
2552 struct net_device
*dev
= NULL
;
2560 if (addr_len
< sizeof(struct sockaddr_ll
))
2562 if (sll
->sll_family
!= AF_PACKET
)
2565 if (sll
->sll_ifindex
) {
2567 dev
= dev_get_by_index(sock_net(sk
), sll
->sll_ifindex
);
2571 err
= packet_do_bind(sk
, dev
, sll
->sll_protocol
? : pkt_sk(sk
)->num
);
2577 static struct proto packet_proto
= {
2579 .owner
= THIS_MODULE
,
2580 .obj_size
= sizeof(struct packet_sock
),
2584 * Create a packet of type SOCK_PACKET.
2587 static int packet_create(struct net
*net
, struct socket
*sock
, int protocol
,
2591 struct packet_sock
*po
;
2592 __be16 proto
= (__force __be16
)protocol
; /* weird, but documented */
2595 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
2597 if (sock
->type
!= SOCK_DGRAM
&& sock
->type
!= SOCK_RAW
&&
2598 sock
->type
!= SOCK_PACKET
)
2599 return -ESOCKTNOSUPPORT
;
2601 sock
->state
= SS_UNCONNECTED
;
2604 sk
= sk_alloc(net
, PF_PACKET
, GFP_KERNEL
, &packet_proto
);
2608 sock
->ops
= &packet_ops
;
2609 if (sock
->type
== SOCK_PACKET
)
2610 sock
->ops
= &packet_ops_spkt
;
2612 sock_init_data(sock
, sk
);
2615 sk
->sk_family
= PF_PACKET
;
2618 sk
->sk_destruct
= packet_sock_destruct
;
2619 sk_refcnt_debug_inc(sk
);
2622 * Attach a protocol block
2625 spin_lock_init(&po
->bind_lock
);
2626 mutex_init(&po
->pg_vec_lock
);
2627 po
->prot_hook
.func
= packet_rcv
;
2629 if (sock
->type
== SOCK_PACKET
)
2630 po
->prot_hook
.func
= packet_rcv_spkt
;
2632 po
->prot_hook
.af_packet_priv
= sk
;
2635 po
->prot_hook
.type
= proto
;
2636 register_prot_hook(sk
);
2639 mutex_lock(&net
->packet
.sklist_lock
);
2640 sk_add_node_rcu(sk
, &net
->packet
.sklist
);
2641 mutex_unlock(&net
->packet
.sklist_lock
);
2644 sock_prot_inuse_add(net
, &packet_proto
, 1);
2653 * Pull a packet from our receive queue and hand it to the user.
2654 * If necessary we block.
2657 static int packet_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2658 struct msghdr
*msg
, size_t len
, int flags
)
2660 struct sock
*sk
= sock
->sk
;
2661 struct sk_buff
*skb
;
2663 int vnet_hdr_len
= 0;
2666 if (flags
& ~(MSG_PEEK
|MSG_DONTWAIT
|MSG_TRUNC
|MSG_CMSG_COMPAT
|MSG_ERRQUEUE
))
2670 /* What error should we return now? EUNATTACH? */
2671 if (pkt_sk(sk
)->ifindex
< 0)
2675 if (flags
& MSG_ERRQUEUE
) {
2676 err
= sock_recv_errqueue(sk
, msg
, len
,
2677 SOL_PACKET
, PACKET_TX_TIMESTAMP
);
2682 * Call the generic datagram receiver. This handles all sorts
2683 * of horrible races and re-entrancy so we can forget about it
2684 * in the protocol layers.
2686 * Now it will return ENETDOWN, if device have just gone down,
2687 * but then it will block.
2690 skb
= skb_recv_datagram(sk
, flags
, flags
& MSG_DONTWAIT
, &err
);
2693 * An error occurred so return it. Because skb_recv_datagram()
2694 * handles the blocking we don't see and worry about blocking
2701 if (pkt_sk(sk
)->has_vnet_hdr
) {
2702 struct virtio_net_hdr vnet_hdr
= { 0 };
2705 vnet_hdr_len
= sizeof(vnet_hdr
);
2706 if (len
< vnet_hdr_len
)
2709 len
-= vnet_hdr_len
;
2711 if (skb_is_gso(skb
)) {
2712 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
2714 /* This is a hint as to how much should be linear. */
2715 vnet_hdr
.hdr_len
= skb_headlen(skb
);
2716 vnet_hdr
.gso_size
= sinfo
->gso_size
;
2717 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
2718 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
2719 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
2720 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
2721 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
2722 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
2723 else if (sinfo
->gso_type
& SKB_GSO_FCOE
)
2727 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
2728 vnet_hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
2730 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
2732 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2733 vnet_hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
2734 vnet_hdr
.csum_start
= skb_checksum_start_offset(skb
);
2735 vnet_hdr
.csum_offset
= skb
->csum_offset
;
2736 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
2737 vnet_hdr
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
2738 } /* else everything is zero */
2740 err
= memcpy_toiovec(msg
->msg_iov
, (void *)&vnet_hdr
,
2746 /* You lose any data beyond the buffer you gave. If it worries
2747 * a user program they can ask the device for its MTU
2753 msg
->msg_flags
|= MSG_TRUNC
;
2756 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
2760 sock_recv_ts_and_drops(msg
, sk
, skb
);
2762 if (msg
->msg_name
) {
2763 /* If the address length field is there to be filled
2764 * in, we fill it in now.
2766 if (sock
->type
== SOCK_PACKET
) {
2767 msg
->msg_namelen
= sizeof(struct sockaddr_pkt
);
2769 struct sockaddr_ll
*sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
2770 msg
->msg_namelen
= sll
->sll_halen
+
2771 offsetof(struct sockaddr_ll
, sll_addr
);
2773 memcpy(msg
->msg_name
, &PACKET_SKB_CB(skb
)->sa
,
2777 if (pkt_sk(sk
)->auxdata
) {
2778 struct tpacket_auxdata aux
;
2780 aux
.tp_status
= TP_STATUS_USER
;
2781 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2782 aux
.tp_status
|= TP_STATUS_CSUMNOTREADY
;
2783 aux
.tp_len
= PACKET_SKB_CB(skb
)->origlen
;
2784 aux
.tp_snaplen
= skb
->len
;
2786 aux
.tp_net
= skb_network_offset(skb
);
2787 if (vlan_tx_tag_present(skb
)) {
2788 aux
.tp_vlan_tci
= vlan_tx_tag_get(skb
);
2789 aux
.tp_status
|= TP_STATUS_VLAN_VALID
;
2791 aux
.tp_vlan_tci
= 0;
2794 put_cmsg(msg
, SOL_PACKET
, PACKET_AUXDATA
, sizeof(aux
), &aux
);
2798 * Free or return the buffer as appropriate. Again this
2799 * hides all the races and re-entrancy issues from us.
2801 err
= vnet_hdr_len
+ ((flags
&MSG_TRUNC
) ? skb
->len
: copied
);
2804 skb_free_datagram(sk
, skb
);
2809 static int packet_getname_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
2810 int *uaddr_len
, int peer
)
2812 struct net_device
*dev
;
2813 struct sock
*sk
= sock
->sk
;
2818 uaddr
->sa_family
= AF_PACKET
;
2819 memset(uaddr
->sa_data
, 0, sizeof(uaddr
->sa_data
));
2821 dev
= dev_get_by_index_rcu(sock_net(sk
), pkt_sk(sk
)->ifindex
);
2823 strlcpy(uaddr
->sa_data
, dev
->name
, sizeof(uaddr
->sa_data
));
2825 *uaddr_len
= sizeof(*uaddr
);
2830 static int packet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
2831 int *uaddr_len
, int peer
)
2833 struct net_device
*dev
;
2834 struct sock
*sk
= sock
->sk
;
2835 struct packet_sock
*po
= pkt_sk(sk
);
2836 DECLARE_SOCKADDR(struct sockaddr_ll
*, sll
, uaddr
);
2841 sll
->sll_family
= AF_PACKET
;
2842 sll
->sll_ifindex
= po
->ifindex
;
2843 sll
->sll_protocol
= po
->num
;
2844 sll
->sll_pkttype
= 0;
2846 dev
= dev_get_by_index_rcu(sock_net(sk
), po
->ifindex
);
2848 sll
->sll_hatype
= dev
->type
;
2849 sll
->sll_halen
= dev
->addr_len
;
2850 memcpy(sll
->sll_addr
, dev
->dev_addr
, dev
->addr_len
);
2852 sll
->sll_hatype
= 0; /* Bad: we have no ARPHRD_UNSPEC */
2856 *uaddr_len
= offsetof(struct sockaddr_ll
, sll_addr
) + sll
->sll_halen
;
2861 static int packet_dev_mc(struct net_device
*dev
, struct packet_mclist
*i
,
2865 case PACKET_MR_MULTICAST
:
2866 if (i
->alen
!= dev
->addr_len
)
2869 return dev_mc_add(dev
, i
->addr
);
2871 return dev_mc_del(dev
, i
->addr
);
2873 case PACKET_MR_PROMISC
:
2874 return dev_set_promiscuity(dev
, what
);
2876 case PACKET_MR_ALLMULTI
:
2877 return dev_set_allmulti(dev
, what
);
2879 case PACKET_MR_UNICAST
:
2880 if (i
->alen
!= dev
->addr_len
)
2883 return dev_uc_add(dev
, i
->addr
);
2885 return dev_uc_del(dev
, i
->addr
);
2893 static void packet_dev_mclist(struct net_device
*dev
, struct packet_mclist
*i
, int what
)
2895 for ( ; i
; i
= i
->next
) {
2896 if (i
->ifindex
== dev
->ifindex
)
2897 packet_dev_mc(dev
, i
, what
);
2901 static int packet_mc_add(struct sock
*sk
, struct packet_mreq_max
*mreq
)
2903 struct packet_sock
*po
= pkt_sk(sk
);
2904 struct packet_mclist
*ml
, *i
;
2905 struct net_device
*dev
;
2911 dev
= __dev_get_by_index(sock_net(sk
), mreq
->mr_ifindex
);
2916 if (mreq
->mr_alen
> dev
->addr_len
)
2920 i
= kmalloc(sizeof(*i
), GFP_KERNEL
);
2925 for (ml
= po
->mclist
; ml
; ml
= ml
->next
) {
2926 if (ml
->ifindex
== mreq
->mr_ifindex
&&
2927 ml
->type
== mreq
->mr_type
&&
2928 ml
->alen
== mreq
->mr_alen
&&
2929 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
2931 /* Free the new element ... */
2937 i
->type
= mreq
->mr_type
;
2938 i
->ifindex
= mreq
->mr_ifindex
;
2939 i
->alen
= mreq
->mr_alen
;
2940 memcpy(i
->addr
, mreq
->mr_address
, i
->alen
);
2942 i
->next
= po
->mclist
;
2944 err
= packet_dev_mc(dev
, i
, 1);
2946 po
->mclist
= i
->next
;
2955 static int packet_mc_drop(struct sock
*sk
, struct packet_mreq_max
*mreq
)
2957 struct packet_mclist
*ml
, **mlp
;
2961 for (mlp
= &pkt_sk(sk
)->mclist
; (ml
= *mlp
) != NULL
; mlp
= &ml
->next
) {
2962 if (ml
->ifindex
== mreq
->mr_ifindex
&&
2963 ml
->type
== mreq
->mr_type
&&
2964 ml
->alen
== mreq
->mr_alen
&&
2965 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
2966 if (--ml
->count
== 0) {
2967 struct net_device
*dev
;
2969 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
2971 packet_dev_mc(dev
, ml
, -1);
2979 return -EADDRNOTAVAIL
;
2982 static void packet_flush_mclist(struct sock
*sk
)
2984 struct packet_sock
*po
= pkt_sk(sk
);
2985 struct packet_mclist
*ml
;
2991 while ((ml
= po
->mclist
) != NULL
) {
2992 struct net_device
*dev
;
2994 po
->mclist
= ml
->next
;
2995 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
2997 packet_dev_mc(dev
, ml
, -1);
3004 packet_setsockopt(struct socket
*sock
, int level
, int optname
, char __user
*optval
, unsigned int optlen
)
3006 struct sock
*sk
= sock
->sk
;
3007 struct packet_sock
*po
= pkt_sk(sk
);
3010 if (level
!= SOL_PACKET
)
3011 return -ENOPROTOOPT
;
3014 case PACKET_ADD_MEMBERSHIP
:
3015 case PACKET_DROP_MEMBERSHIP
:
3017 struct packet_mreq_max mreq
;
3019 memset(&mreq
, 0, sizeof(mreq
));
3020 if (len
< sizeof(struct packet_mreq
))
3022 if (len
> sizeof(mreq
))
3024 if (copy_from_user(&mreq
, optval
, len
))
3026 if (len
< (mreq
.mr_alen
+ offsetof(struct packet_mreq
, mr_address
)))
3028 if (optname
== PACKET_ADD_MEMBERSHIP
)
3029 ret
= packet_mc_add(sk
, &mreq
);
3031 ret
= packet_mc_drop(sk
, &mreq
);
3035 case PACKET_RX_RING
:
3036 case PACKET_TX_RING
:
3038 union tpacket_req_u req_u
;
3041 switch (po
->tp_version
) {
3044 len
= sizeof(req_u
.req
);
3048 len
= sizeof(req_u
.req3
);
3053 if (pkt_sk(sk
)->has_vnet_hdr
)
3055 if (copy_from_user(&req_u
.req
, optval
, len
))
3057 return packet_set_ring(sk
, &req_u
, 0,
3058 optname
== PACKET_TX_RING
);
3060 case PACKET_COPY_THRESH
:
3064 if (optlen
!= sizeof(val
))
3066 if (copy_from_user(&val
, optval
, sizeof(val
)))
3069 pkt_sk(sk
)->copy_thresh
= val
;
3072 case PACKET_VERSION
:
3076 if (optlen
!= sizeof(val
))
3078 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3080 if (copy_from_user(&val
, optval
, sizeof(val
)))
3086 po
->tp_version
= val
;
3092 case PACKET_RESERVE
:
3096 if (optlen
!= sizeof(val
))
3098 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3100 if (copy_from_user(&val
, optval
, sizeof(val
)))
3102 po
->tp_reserve
= val
;
3109 if (optlen
!= sizeof(val
))
3111 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3113 if (copy_from_user(&val
, optval
, sizeof(val
)))
3115 po
->tp_loss
= !!val
;
3118 case PACKET_AUXDATA
:
3122 if (optlen
< sizeof(val
))
3124 if (copy_from_user(&val
, optval
, sizeof(val
)))
3127 po
->auxdata
= !!val
;
3130 case PACKET_ORIGDEV
:
3134 if (optlen
< sizeof(val
))
3136 if (copy_from_user(&val
, optval
, sizeof(val
)))
3139 po
->origdev
= !!val
;
3142 case PACKET_VNET_HDR
:
3146 if (sock
->type
!= SOCK_RAW
)
3148 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3150 if (optlen
< sizeof(val
))
3152 if (copy_from_user(&val
, optval
, sizeof(val
)))
3155 po
->has_vnet_hdr
= !!val
;
3158 case PACKET_TIMESTAMP
:
3162 if (optlen
!= sizeof(val
))
3164 if (copy_from_user(&val
, optval
, sizeof(val
)))
3167 po
->tp_tstamp
= val
;
3174 if (optlen
!= sizeof(val
))
3176 if (copy_from_user(&val
, optval
, sizeof(val
)))
3179 return fanout_add(sk
, val
& 0xffff, val
>> 16);
3181 case PACKET_TX_HAS_OFF
:
3185 if (optlen
!= sizeof(val
))
3187 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3189 if (copy_from_user(&val
, optval
, sizeof(val
)))
3191 po
->tp_tx_has_off
= !!val
;
3195 return -ENOPROTOOPT
;
3199 static int packet_getsockopt(struct socket
*sock
, int level
, int optname
,
3200 char __user
*optval
, int __user
*optlen
)
3203 int val
, lv
= sizeof(val
);
3204 struct sock
*sk
= sock
->sk
;
3205 struct packet_sock
*po
= pkt_sk(sk
);
3207 union tpacket_stats_u st
;
3209 if (level
!= SOL_PACKET
)
3210 return -ENOPROTOOPT
;
3212 if (get_user(len
, optlen
))
3219 case PACKET_STATISTICS
:
3220 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3221 memcpy(&st
, &po
->stats
, sizeof(st
));
3222 memset(&po
->stats
, 0, sizeof(po
->stats
));
3223 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3225 if (po
->tp_version
== TPACKET_V3
) {
3226 lv
= sizeof(struct tpacket_stats_v3
);
3227 st
.stats3
.tp_packets
+= st
.stats3
.tp_drops
;
3230 lv
= sizeof(struct tpacket_stats
);
3231 st
.stats1
.tp_packets
+= st
.stats1
.tp_drops
;
3236 case PACKET_AUXDATA
:
3239 case PACKET_ORIGDEV
:
3242 case PACKET_VNET_HDR
:
3243 val
= po
->has_vnet_hdr
;
3245 case PACKET_VERSION
:
3246 val
= po
->tp_version
;
3249 if (len
> sizeof(int))
3251 if (copy_from_user(&val
, optval
, len
))
3255 val
= sizeof(struct tpacket_hdr
);
3258 val
= sizeof(struct tpacket2_hdr
);
3261 val
= sizeof(struct tpacket3_hdr
);
3267 case PACKET_RESERVE
:
3268 val
= po
->tp_reserve
;
3273 case PACKET_TIMESTAMP
:
3274 val
= po
->tp_tstamp
;
3278 ((u32
)po
->fanout
->id
|
3279 ((u32
)po
->fanout
->type
<< 16) |
3280 ((u32
)po
->fanout
->flags
<< 24)) :
3283 case PACKET_TX_HAS_OFF
:
3284 val
= po
->tp_tx_has_off
;
3287 return -ENOPROTOOPT
;
3292 if (put_user(len
, optlen
))
3294 if (copy_to_user(optval
, data
, len
))
3300 static int packet_notifier(struct notifier_block
*this,
3301 unsigned long msg
, void *ptr
)
3304 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
3305 struct net
*net
= dev_net(dev
);
3308 sk_for_each_rcu(sk
, &net
->packet
.sklist
) {
3309 struct packet_sock
*po
= pkt_sk(sk
);
3312 case NETDEV_UNREGISTER
:
3314 packet_dev_mclist(dev
, po
->mclist
, -1);
3318 if (dev
->ifindex
== po
->ifindex
) {
3319 spin_lock(&po
->bind_lock
);
3321 __unregister_prot_hook(sk
, false);
3322 sk
->sk_err
= ENETDOWN
;
3323 if (!sock_flag(sk
, SOCK_DEAD
))
3324 sk
->sk_error_report(sk
);
3326 if (msg
== NETDEV_UNREGISTER
) {
3328 if (po
->prot_hook
.dev
)
3329 dev_put(po
->prot_hook
.dev
);
3330 po
->prot_hook
.dev
= NULL
;
3332 spin_unlock(&po
->bind_lock
);
3336 if (dev
->ifindex
== po
->ifindex
) {
3337 spin_lock(&po
->bind_lock
);
3339 register_prot_hook(sk
);
3340 spin_unlock(&po
->bind_lock
);
3350 static int packet_ioctl(struct socket
*sock
, unsigned int cmd
,
3353 struct sock
*sk
= sock
->sk
;
3358 int amount
= sk_wmem_alloc_get(sk
);
3360 return put_user(amount
, (int __user
*)arg
);
3364 struct sk_buff
*skb
;
3367 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3368 skb
= skb_peek(&sk
->sk_receive_queue
);
3371 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3372 return put_user(amount
, (int __user
*)arg
);
3375 return sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
3377 return sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
3387 case SIOCGIFBRDADDR
:
3388 case SIOCSIFBRDADDR
:
3389 case SIOCGIFNETMASK
:
3390 case SIOCSIFNETMASK
:
3391 case SIOCGIFDSTADDR
:
3392 case SIOCSIFDSTADDR
:
3394 return inet_dgram_ops
.ioctl(sock
, cmd
, arg
);
3398 return -ENOIOCTLCMD
;
3403 static unsigned int packet_poll(struct file
*file
, struct socket
*sock
,
3406 struct sock
*sk
= sock
->sk
;
3407 struct packet_sock
*po
= pkt_sk(sk
);
3408 unsigned int mask
= datagram_poll(file
, sock
, wait
);
3410 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3411 if (po
->rx_ring
.pg_vec
) {
3412 if (!packet_previous_rx_frame(po
, &po
->rx_ring
,
3414 mask
|= POLLIN
| POLLRDNORM
;
3416 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3417 spin_lock_bh(&sk
->sk_write_queue
.lock
);
3418 if (po
->tx_ring
.pg_vec
) {
3419 if (packet_current_frame(po
, &po
->tx_ring
, TP_STATUS_AVAILABLE
))
3420 mask
|= POLLOUT
| POLLWRNORM
;
3422 spin_unlock_bh(&sk
->sk_write_queue
.lock
);
3427 /* Dirty? Well, I still did not learn better way to account
3431 static void packet_mm_open(struct vm_area_struct
*vma
)
3433 struct file
*file
= vma
->vm_file
;
3434 struct socket
*sock
= file
->private_data
;
3435 struct sock
*sk
= sock
->sk
;
3438 atomic_inc(&pkt_sk(sk
)->mapped
);
3441 static void packet_mm_close(struct vm_area_struct
*vma
)
3443 struct file
*file
= vma
->vm_file
;
3444 struct socket
*sock
= file
->private_data
;
3445 struct sock
*sk
= sock
->sk
;
3448 atomic_dec(&pkt_sk(sk
)->mapped
);
3451 static const struct vm_operations_struct packet_mmap_ops
= {
3452 .open
= packet_mm_open
,
3453 .close
= packet_mm_close
,
3456 static void free_pg_vec(struct pgv
*pg_vec
, unsigned int order
,
3461 for (i
= 0; i
< len
; i
++) {
3462 if (likely(pg_vec
[i
].buffer
)) {
3463 if (is_vmalloc_addr(pg_vec
[i
].buffer
))
3464 vfree(pg_vec
[i
].buffer
);
3466 free_pages((unsigned long)pg_vec
[i
].buffer
,
3468 pg_vec
[i
].buffer
= NULL
;
3474 static char *alloc_one_pg_vec_page(unsigned long order
)
3476 char *buffer
= NULL
;
3477 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_COMP
|
3478 __GFP_ZERO
| __GFP_NOWARN
| __GFP_NORETRY
;
3480 buffer
= (char *) __get_free_pages(gfp_flags
, order
);
3486 * __get_free_pages failed, fall back to vmalloc
3488 buffer
= vzalloc((1 << order
) * PAGE_SIZE
);
3494 * vmalloc failed, lets dig into swap here
3496 gfp_flags
&= ~__GFP_NORETRY
;
3497 buffer
= (char *)__get_free_pages(gfp_flags
, order
);
3502 * complete and utter failure
3507 static struct pgv
*alloc_pg_vec(struct tpacket_req
*req
, int order
)
3509 unsigned int block_nr
= req
->tp_block_nr
;
3513 pg_vec
= kcalloc(block_nr
, sizeof(struct pgv
), GFP_KERNEL
);
3514 if (unlikely(!pg_vec
))
3517 for (i
= 0; i
< block_nr
; i
++) {
3518 pg_vec
[i
].buffer
= alloc_one_pg_vec_page(order
);
3519 if (unlikely(!pg_vec
[i
].buffer
))
3520 goto out_free_pgvec
;
3527 free_pg_vec(pg_vec
, order
, block_nr
);
3532 static int packet_set_ring(struct sock
*sk
, union tpacket_req_u
*req_u
,
3533 int closing
, int tx_ring
)
3535 struct pgv
*pg_vec
= NULL
;
3536 struct packet_sock
*po
= pkt_sk(sk
);
3537 int was_running
, order
= 0;
3538 struct packet_ring_buffer
*rb
;
3539 struct sk_buff_head
*rb_queue
;
3542 /* Added to avoid minimal code churn */
3543 struct tpacket_req
*req
= &req_u
->req
;
3545 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3546 if (!closing
&& tx_ring
&& (po
->tp_version
> TPACKET_V2
)) {
3547 WARN(1, "Tx-ring is not supported.\n");
3551 rb
= tx_ring
? &po
->tx_ring
: &po
->rx_ring
;
3552 rb_queue
= tx_ring
? &sk
->sk_write_queue
: &sk
->sk_receive_queue
;
3556 if (atomic_read(&po
->mapped
))
3558 if (atomic_read(&rb
->pending
))
3562 if (req
->tp_block_nr
) {
3563 /* Sanity tests and some calculations */
3565 if (unlikely(rb
->pg_vec
))
3568 switch (po
->tp_version
) {
3570 po
->tp_hdrlen
= TPACKET_HDRLEN
;
3573 po
->tp_hdrlen
= TPACKET2_HDRLEN
;
3576 po
->tp_hdrlen
= TPACKET3_HDRLEN
;
3581 if (unlikely((int)req
->tp_block_size
<= 0))
3583 if (unlikely(req
->tp_block_size
& (PAGE_SIZE
- 1)))
3585 if (unlikely(req
->tp_frame_size
< po
->tp_hdrlen
+
3588 if (unlikely(req
->tp_frame_size
& (TPACKET_ALIGNMENT
- 1)))
3591 rb
->frames_per_block
= req
->tp_block_size
/req
->tp_frame_size
;
3592 if (unlikely(rb
->frames_per_block
<= 0))
3594 if (unlikely((rb
->frames_per_block
* req
->tp_block_nr
) !=
3599 order
= get_order(req
->tp_block_size
);
3600 pg_vec
= alloc_pg_vec(req
, order
);
3601 if (unlikely(!pg_vec
))
3603 switch (po
->tp_version
) {
3605 /* Transmit path is not supported. We checked
3606 * it above but just being paranoid
3609 init_prb_bdqc(po
, rb
, pg_vec
, req_u
, tx_ring
);
3618 if (unlikely(req
->tp_frame_nr
))
3624 /* Detach socket from network */
3625 spin_lock(&po
->bind_lock
);
3626 was_running
= po
->running
;
3630 __unregister_prot_hook(sk
, false);
3632 spin_unlock(&po
->bind_lock
);
3637 mutex_lock(&po
->pg_vec_lock
);
3638 if (closing
|| atomic_read(&po
->mapped
) == 0) {
3640 spin_lock_bh(&rb_queue
->lock
);
3641 swap(rb
->pg_vec
, pg_vec
);
3642 rb
->frame_max
= (req
->tp_frame_nr
- 1);
3644 rb
->frame_size
= req
->tp_frame_size
;
3645 spin_unlock_bh(&rb_queue
->lock
);
3647 swap(rb
->pg_vec_order
, order
);
3648 swap(rb
->pg_vec_len
, req
->tp_block_nr
);
3650 rb
->pg_vec_pages
= req
->tp_block_size
/PAGE_SIZE
;
3651 po
->prot_hook
.func
= (po
->rx_ring
.pg_vec
) ?
3652 tpacket_rcv
: packet_rcv
;
3653 skb_queue_purge(rb_queue
);
3654 if (atomic_read(&po
->mapped
))
3655 pr_err("packet_mmap: vma is busy: %d\n",
3656 atomic_read(&po
->mapped
));
3658 mutex_unlock(&po
->pg_vec_lock
);
3660 spin_lock(&po
->bind_lock
);
3663 register_prot_hook(sk
);
3665 spin_unlock(&po
->bind_lock
);
3666 if (closing
&& (po
->tp_version
> TPACKET_V2
)) {
3667 /* Because we don't support block-based V3 on tx-ring */
3669 prb_shutdown_retire_blk_timer(po
, tx_ring
, rb_queue
);
3674 free_pg_vec(pg_vec
, order
, req
->tp_block_nr
);
3679 static int packet_mmap(struct file
*file
, struct socket
*sock
,
3680 struct vm_area_struct
*vma
)
3682 struct sock
*sk
= sock
->sk
;
3683 struct packet_sock
*po
= pkt_sk(sk
);
3684 unsigned long size
, expected_size
;
3685 struct packet_ring_buffer
*rb
;
3686 unsigned long start
;
3693 mutex_lock(&po
->pg_vec_lock
);
3696 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
3698 expected_size
+= rb
->pg_vec_len
3704 if (expected_size
== 0)
3707 size
= vma
->vm_end
- vma
->vm_start
;
3708 if (size
!= expected_size
)
3711 start
= vma
->vm_start
;
3712 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
3713 if (rb
->pg_vec
== NULL
)
3716 for (i
= 0; i
< rb
->pg_vec_len
; i
++) {
3718 void *kaddr
= rb
->pg_vec
[i
].buffer
;
3721 for (pg_num
= 0; pg_num
< rb
->pg_vec_pages
; pg_num
++) {
3722 page
= pgv_to_page(kaddr
);
3723 err
= vm_insert_page(vma
, start
, page
);
3732 atomic_inc(&po
->mapped
);
3733 vma
->vm_ops
= &packet_mmap_ops
;
3737 mutex_unlock(&po
->pg_vec_lock
);
3741 static const struct proto_ops packet_ops_spkt
= {
3742 .family
= PF_PACKET
,
3743 .owner
= THIS_MODULE
,
3744 .release
= packet_release
,
3745 .bind
= packet_bind_spkt
,
3746 .connect
= sock_no_connect
,
3747 .socketpair
= sock_no_socketpair
,
3748 .accept
= sock_no_accept
,
3749 .getname
= packet_getname_spkt
,
3750 .poll
= datagram_poll
,
3751 .ioctl
= packet_ioctl
,
3752 .listen
= sock_no_listen
,
3753 .shutdown
= sock_no_shutdown
,
3754 .setsockopt
= sock_no_setsockopt
,
3755 .getsockopt
= sock_no_getsockopt
,
3756 .sendmsg
= packet_sendmsg_spkt
,
3757 .recvmsg
= packet_recvmsg
,
3758 .mmap
= sock_no_mmap
,
3759 .sendpage
= sock_no_sendpage
,
3762 static const struct proto_ops packet_ops
= {
3763 .family
= PF_PACKET
,
3764 .owner
= THIS_MODULE
,
3765 .release
= packet_release
,
3766 .bind
= packet_bind
,
3767 .connect
= sock_no_connect
,
3768 .socketpair
= sock_no_socketpair
,
3769 .accept
= sock_no_accept
,
3770 .getname
= packet_getname
,
3771 .poll
= packet_poll
,
3772 .ioctl
= packet_ioctl
,
3773 .listen
= sock_no_listen
,
3774 .shutdown
= sock_no_shutdown
,
3775 .setsockopt
= packet_setsockopt
,
3776 .getsockopt
= packet_getsockopt
,
3777 .sendmsg
= packet_sendmsg
,
3778 .recvmsg
= packet_recvmsg
,
3779 .mmap
= packet_mmap
,
3780 .sendpage
= sock_no_sendpage
,
3783 static const struct net_proto_family packet_family_ops
= {
3784 .family
= PF_PACKET
,
3785 .create
= packet_create
,
3786 .owner
= THIS_MODULE
,
3789 static struct notifier_block packet_netdev_notifier
= {
3790 .notifier_call
= packet_notifier
,
3793 #ifdef CONFIG_PROC_FS
3795 static void *packet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3798 struct net
*net
= seq_file_net(seq
);
3801 return seq_hlist_start_head_rcu(&net
->packet
.sklist
, *pos
);
3804 static void *packet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3806 struct net
*net
= seq_file_net(seq
);
3807 return seq_hlist_next_rcu(v
, &net
->packet
.sklist
, pos
);
3810 static void packet_seq_stop(struct seq_file
*seq
, void *v
)
3816 static int packet_seq_show(struct seq_file
*seq
, void *v
)
3818 if (v
== SEQ_START_TOKEN
)
3819 seq_puts(seq
, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3821 struct sock
*s
= sk_entry(v
);
3822 const struct packet_sock
*po
= pkt_sk(s
);
3825 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3827 atomic_read(&s
->sk_refcnt
),
3832 atomic_read(&s
->sk_rmem_alloc
),
3833 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(s
)),
3840 static const struct seq_operations packet_seq_ops
= {
3841 .start
= packet_seq_start
,
3842 .next
= packet_seq_next
,
3843 .stop
= packet_seq_stop
,
3844 .show
= packet_seq_show
,
3847 static int packet_seq_open(struct inode
*inode
, struct file
*file
)
3849 return seq_open_net(inode
, file
, &packet_seq_ops
,
3850 sizeof(struct seq_net_private
));
3853 static const struct file_operations packet_seq_fops
= {
3854 .owner
= THIS_MODULE
,
3855 .open
= packet_seq_open
,
3857 .llseek
= seq_lseek
,
3858 .release
= seq_release_net
,
3863 static int __net_init
packet_net_init(struct net
*net
)
3865 mutex_init(&net
->packet
.sklist_lock
);
3866 INIT_HLIST_HEAD(&net
->packet
.sklist
);
3868 if (!proc_create("packet", 0, net
->proc_net
, &packet_seq_fops
))
3874 static void __net_exit
packet_net_exit(struct net
*net
)
3876 remove_proc_entry("packet", net
->proc_net
);
3879 static struct pernet_operations packet_net_ops
= {
3880 .init
= packet_net_init
,
3881 .exit
= packet_net_exit
,
3885 static void __exit
packet_exit(void)
3887 unregister_netdevice_notifier(&packet_netdev_notifier
);
3888 unregister_pernet_subsys(&packet_net_ops
);
3889 sock_unregister(PF_PACKET
);
3890 proto_unregister(&packet_proto
);
3893 static int __init
packet_init(void)
3895 int rc
= proto_register(&packet_proto
, 0);
3900 sock_register(&packet_family_ops
);
3901 register_pernet_subsys(&packet_net_ops
);
3902 register_netdevice_notifier(&packet_netdev_notifier
);
3907 module_init(packet_init
);
3908 module_exit(packet_exit
);
3909 MODULE_LICENSE("GPL");
3910 MODULE_ALIAS_NETPROTO(PF_PACKET
);