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>
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 static struct net_device
*packet_cached_dev_get(struct packet_sock
*po
)
242 struct net_device
*dev
;
245 dev
= rcu_dereference(po
->cached_dev
);
253 static void packet_cached_dev_assign(struct packet_sock
*po
,
254 struct net_device
*dev
)
256 rcu_assign_pointer(po
->cached_dev
, dev
);
259 static void packet_cached_dev_reset(struct packet_sock
*po
)
261 RCU_INIT_POINTER(po
->cached_dev
, NULL
);
264 /* register_prot_hook must be invoked with the po->bind_lock held,
265 * or from a context in which asynchronous accesses to the packet
266 * socket is not possible (packet_create()).
268 static void register_prot_hook(struct sock
*sk
)
270 struct packet_sock
*po
= pkt_sk(sk
);
274 __fanout_link(sk
, po
);
276 dev_add_pack(&po
->prot_hook
);
283 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
284 * held. If the sync parameter is true, we will temporarily drop
285 * the po->bind_lock and do a synchronize_net to make sure no
286 * asynchronous packet processing paths still refer to the elements
287 * of po->prot_hook. If the sync parameter is false, it is the
288 * callers responsibility to take care of this.
290 static void __unregister_prot_hook(struct sock
*sk
, bool sync
)
292 struct packet_sock
*po
= pkt_sk(sk
);
297 __fanout_unlink(sk
, po
);
299 __dev_remove_pack(&po
->prot_hook
);
304 spin_unlock(&po
->bind_lock
);
306 spin_lock(&po
->bind_lock
);
310 static void unregister_prot_hook(struct sock
*sk
, bool sync
)
312 struct packet_sock
*po
= pkt_sk(sk
);
315 __unregister_prot_hook(sk
, sync
);
318 static inline __pure
struct page
*pgv_to_page(void *addr
)
320 if (is_vmalloc_addr(addr
))
321 return vmalloc_to_page(addr
);
322 return virt_to_page(addr
);
325 static void __packet_set_status(struct packet_sock
*po
, void *frame
, int status
)
327 union tpacket_uhdr h
;
330 switch (po
->tp_version
) {
332 h
.h1
->tp_status
= status
;
333 flush_dcache_page(pgv_to_page(&h
.h1
->tp_status
));
336 h
.h2
->tp_status
= status
;
337 flush_dcache_page(pgv_to_page(&h
.h2
->tp_status
));
341 WARN(1, "TPACKET version not supported.\n");
348 static int __packet_get_status(struct packet_sock
*po
, void *frame
)
350 union tpacket_uhdr h
;
355 switch (po
->tp_version
) {
357 flush_dcache_page(pgv_to_page(&h
.h1
->tp_status
));
358 return h
.h1
->tp_status
;
360 flush_dcache_page(pgv_to_page(&h
.h2
->tp_status
));
361 return h
.h2
->tp_status
;
364 WARN(1, "TPACKET version not supported.\n");
370 static __u32
tpacket_get_timestamp(struct sk_buff
*skb
, struct timespec
*ts
,
373 struct skb_shared_hwtstamps
*shhwtstamps
= skb_hwtstamps(skb
);
376 if ((flags
& SOF_TIMESTAMPING_SYS_HARDWARE
) &&
377 ktime_to_timespec_cond(shhwtstamps
->syststamp
, ts
))
378 return TP_STATUS_TS_SYS_HARDWARE
;
379 if ((flags
& SOF_TIMESTAMPING_RAW_HARDWARE
) &&
380 ktime_to_timespec_cond(shhwtstamps
->hwtstamp
, ts
))
381 return TP_STATUS_TS_RAW_HARDWARE
;
384 if (ktime_to_timespec_cond(skb
->tstamp
, ts
))
385 return TP_STATUS_TS_SOFTWARE
;
390 static __u32
__packet_set_timestamp(struct packet_sock
*po
, void *frame
,
393 union tpacket_uhdr h
;
397 if (!(ts_status
= tpacket_get_timestamp(skb
, &ts
, po
->tp_tstamp
)))
401 switch (po
->tp_version
) {
403 h
.h1
->tp_sec
= ts
.tv_sec
;
404 h
.h1
->tp_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
407 h
.h2
->tp_sec
= ts
.tv_sec
;
408 h
.h2
->tp_nsec
= ts
.tv_nsec
;
412 WARN(1, "TPACKET version not supported.\n");
416 /* one flush is safe, as both fields always lie on the same cacheline */
417 flush_dcache_page(pgv_to_page(&h
.h1
->tp_sec
));
423 static void *packet_lookup_frame(struct packet_sock
*po
,
424 struct packet_ring_buffer
*rb
,
425 unsigned int position
,
428 unsigned int pg_vec_pos
, frame_offset
;
429 union tpacket_uhdr h
;
431 pg_vec_pos
= position
/ rb
->frames_per_block
;
432 frame_offset
= position
% rb
->frames_per_block
;
434 h
.raw
= rb
->pg_vec
[pg_vec_pos
].buffer
+
435 (frame_offset
* rb
->frame_size
);
437 if (status
!= __packet_get_status(po
, h
.raw
))
443 static void *packet_current_frame(struct packet_sock
*po
,
444 struct packet_ring_buffer
*rb
,
447 return packet_lookup_frame(po
, rb
, rb
->head
, status
);
450 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core
*pkc
)
452 del_timer_sync(&pkc
->retire_blk_timer
);
455 static void prb_shutdown_retire_blk_timer(struct packet_sock
*po
,
457 struct sk_buff_head
*rb_queue
)
459 struct tpacket_kbdq_core
*pkc
;
461 pkc
= tx_ring
? &po
->tx_ring
.prb_bdqc
: &po
->rx_ring
.prb_bdqc
;
463 spin_lock_bh(&rb_queue
->lock
);
464 pkc
->delete_blk_timer
= 1;
465 spin_unlock_bh(&rb_queue
->lock
);
467 prb_del_retire_blk_timer(pkc
);
470 static void prb_init_blk_timer(struct packet_sock
*po
,
471 struct tpacket_kbdq_core
*pkc
,
472 void (*func
) (unsigned long))
474 init_timer(&pkc
->retire_blk_timer
);
475 pkc
->retire_blk_timer
.data
= (long)po
;
476 pkc
->retire_blk_timer
.function
= func
;
477 pkc
->retire_blk_timer
.expires
= jiffies
;
480 static void prb_setup_retire_blk_timer(struct packet_sock
*po
, int tx_ring
)
482 struct tpacket_kbdq_core
*pkc
;
487 pkc
= tx_ring
? &po
->tx_ring
.prb_bdqc
: &po
->rx_ring
.prb_bdqc
;
488 prb_init_blk_timer(po
, pkc
, prb_retire_rx_blk_timer_expired
);
491 static int prb_calc_retire_blk_tmo(struct packet_sock
*po
,
492 int blk_size_in_bytes
)
494 struct net_device
*dev
;
495 unsigned int mbits
= 0, msec
= 0, div
= 0, tmo
= 0;
496 struct ethtool_cmd ecmd
;
501 dev
= __dev_get_by_index(sock_net(&po
->sk
), po
->ifindex
);
502 if (unlikely(!dev
)) {
504 return DEFAULT_PRB_RETIRE_TOV
;
506 err
= __ethtool_get_settings(dev
, &ecmd
);
507 speed
= ethtool_cmd_speed(&ecmd
);
511 * If the link speed is so slow you don't really
512 * need to worry about perf anyways
514 if (speed
< SPEED_1000
|| speed
== SPEED_UNKNOWN
) {
515 return DEFAULT_PRB_RETIRE_TOV
;
522 mbits
= (blk_size_in_bytes
* 8) / (1024 * 1024);
534 static void prb_init_ft_ops(struct tpacket_kbdq_core
*p1
,
535 union tpacket_req_u
*req_u
)
537 p1
->feature_req_word
= req_u
->req3
.tp_feature_req_word
;
540 static void init_prb_bdqc(struct packet_sock
*po
,
541 struct packet_ring_buffer
*rb
,
543 union tpacket_req_u
*req_u
, int tx_ring
)
545 struct tpacket_kbdq_core
*p1
= &rb
->prb_bdqc
;
546 struct tpacket_block_desc
*pbd
;
548 memset(p1
, 0x0, sizeof(*p1
));
550 p1
->knxt_seq_num
= 1;
552 pbd
= (struct tpacket_block_desc
*)pg_vec
[0].buffer
;
553 p1
->pkblk_start
= pg_vec
[0].buffer
;
554 p1
->kblk_size
= req_u
->req3
.tp_block_size
;
555 p1
->knum_blocks
= req_u
->req3
.tp_block_nr
;
556 p1
->hdrlen
= po
->tp_hdrlen
;
557 p1
->version
= po
->tp_version
;
558 p1
->last_kactive_blk_num
= 0;
559 po
->stats
.stats3
.tp_freeze_q_cnt
= 0;
560 if (req_u
->req3
.tp_retire_blk_tov
)
561 p1
->retire_blk_tov
= req_u
->req3
.tp_retire_blk_tov
;
563 p1
->retire_blk_tov
= prb_calc_retire_blk_tmo(po
,
564 req_u
->req3
.tp_block_size
);
565 p1
->tov_in_jiffies
= msecs_to_jiffies(p1
->retire_blk_tov
);
566 p1
->blk_sizeof_priv
= req_u
->req3
.tp_sizeof_priv
;
568 p1
->max_frame_len
= p1
->kblk_size
- BLK_PLUS_PRIV(p1
->blk_sizeof_priv
);
569 prb_init_ft_ops(p1
, req_u
);
570 prb_setup_retire_blk_timer(po
, tx_ring
);
571 prb_open_block(p1
, pbd
);
574 /* Do NOT update the last_blk_num first.
575 * Assumes sk_buff_head lock is held.
577 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core
*pkc
)
579 mod_timer(&pkc
->retire_blk_timer
,
580 jiffies
+ pkc
->tov_in_jiffies
);
581 pkc
->last_kactive_blk_num
= pkc
->kactive_blk_num
;
586 * 1) We refresh the timer only when we open a block.
587 * By doing this we don't waste cycles refreshing the timer
588 * on packet-by-packet basis.
590 * With a 1MB block-size, on a 1Gbps line, it will take
591 * i) ~8 ms to fill a block + ii) memcpy etc.
592 * In this cut we are not accounting for the memcpy time.
594 * So, if the user sets the 'tmo' to 10ms then the timer
595 * will never fire while the block is still getting filled
596 * (which is what we want). However, the user could choose
597 * to close a block early and that's fine.
599 * But when the timer does fire, we check whether or not to refresh it.
600 * Since the tmo granularity is in msecs, it is not too expensive
601 * to refresh the timer, lets say every '8' msecs.
602 * Either the user can set the 'tmo' or we can derive it based on
603 * a) line-speed and b) block-size.
604 * prb_calc_retire_blk_tmo() calculates the tmo.
607 static void prb_retire_rx_blk_timer_expired(unsigned long data
)
609 struct packet_sock
*po
= (struct packet_sock
*)data
;
610 struct tpacket_kbdq_core
*pkc
= &po
->rx_ring
.prb_bdqc
;
612 struct tpacket_block_desc
*pbd
;
614 spin_lock(&po
->sk
.sk_receive_queue
.lock
);
616 frozen
= prb_queue_frozen(pkc
);
617 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
619 if (unlikely(pkc
->delete_blk_timer
))
622 /* We only need to plug the race when the block is partially filled.
624 * lock(); increment BLOCK_NUM_PKTS; unlock()
625 * copy_bits() is in progress ...
626 * timer fires on other cpu:
627 * we can't retire the current block because copy_bits
631 if (BLOCK_NUM_PKTS(pbd
)) {
632 while (atomic_read(&pkc
->blk_fill_in_prog
)) {
633 /* Waiting for skb_copy_bits to finish... */
638 if (pkc
->last_kactive_blk_num
== pkc
->kactive_blk_num
) {
640 prb_retire_current_block(pkc
, po
, TP_STATUS_BLK_TMO
);
641 if (!prb_dispatch_next_block(pkc
, po
))
646 /* Case 1. Queue was frozen because user-space was
649 if (prb_curr_blk_in_use(pkc
, pbd
)) {
651 * Ok, user-space is still behind.
652 * So just refresh the timer.
656 /* Case 2. queue was frozen,user-space caught up,
657 * now the link went idle && the timer fired.
658 * We don't have a block to close.So we open this
659 * block and restart the timer.
660 * opening a block thaws the queue,restarts timer
661 * Thawing/timer-refresh is a side effect.
663 prb_open_block(pkc
, pbd
);
670 _prb_refresh_rx_retire_blk_timer(pkc
);
673 spin_unlock(&po
->sk
.sk_receive_queue
.lock
);
676 static void prb_flush_block(struct tpacket_kbdq_core
*pkc1
,
677 struct tpacket_block_desc
*pbd1
, __u32 status
)
679 /* Flush everything minus the block header */
681 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
686 /* Skip the block header(we know header WILL fit in 4K) */
689 end
= (u8
*)PAGE_ALIGN((unsigned long)pkc1
->pkblk_end
);
690 for (; start
< end
; start
+= PAGE_SIZE
)
691 flush_dcache_page(pgv_to_page(start
));
696 /* Now update the block status. */
698 BLOCK_STATUS(pbd1
) = status
;
700 /* Flush the block header */
702 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
704 flush_dcache_page(pgv_to_page(start
));
714 * 2) Increment active_blk_num
716 * Note:We DONT refresh the timer on purpose.
717 * Because almost always the next block will be opened.
719 static void prb_close_block(struct tpacket_kbdq_core
*pkc1
,
720 struct tpacket_block_desc
*pbd1
,
721 struct packet_sock
*po
, unsigned int stat
)
723 __u32 status
= TP_STATUS_USER
| stat
;
725 struct tpacket3_hdr
*last_pkt
;
726 struct tpacket_hdr_v1
*h1
= &pbd1
->hdr
.bh1
;
728 if (po
->stats
.stats3
.tp_drops
)
729 status
|= TP_STATUS_LOSING
;
731 last_pkt
= (struct tpacket3_hdr
*)pkc1
->prev
;
732 last_pkt
->tp_next_offset
= 0;
734 /* Get the ts of the last pkt */
735 if (BLOCK_NUM_PKTS(pbd1
)) {
736 h1
->ts_last_pkt
.ts_sec
= last_pkt
->tp_sec
;
737 h1
->ts_last_pkt
.ts_nsec
= last_pkt
->tp_nsec
;
739 /* Ok, we tmo'd - so get the current time */
742 h1
->ts_last_pkt
.ts_sec
= ts
.tv_sec
;
743 h1
->ts_last_pkt
.ts_nsec
= ts
.tv_nsec
;
748 /* Flush the block */
749 prb_flush_block(pkc1
, pbd1
, status
);
751 pkc1
->kactive_blk_num
= GET_NEXT_PRB_BLK_NUM(pkc1
);
754 static void prb_thaw_queue(struct tpacket_kbdq_core
*pkc
)
756 pkc
->reset_pending_on_curr_blk
= 0;
760 * Side effect of opening a block:
762 * 1) prb_queue is thawed.
763 * 2) retire_blk_timer is refreshed.
766 static void prb_open_block(struct tpacket_kbdq_core
*pkc1
,
767 struct tpacket_block_desc
*pbd1
)
770 struct tpacket_hdr_v1
*h1
= &pbd1
->hdr
.bh1
;
774 /* We could have just memset this but we will lose the
775 * flexibility of making the priv area sticky
778 BLOCK_SNUM(pbd1
) = pkc1
->knxt_seq_num
++;
779 BLOCK_NUM_PKTS(pbd1
) = 0;
780 BLOCK_LEN(pbd1
) = BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
784 h1
->ts_first_pkt
.ts_sec
= ts
.tv_sec
;
785 h1
->ts_first_pkt
.ts_nsec
= ts
.tv_nsec
;
787 pkc1
->pkblk_start
= (char *)pbd1
;
788 pkc1
->nxt_offset
= pkc1
->pkblk_start
+ BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
790 BLOCK_O2FP(pbd1
) = (__u32
)BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
791 BLOCK_O2PRIV(pbd1
) = BLK_HDR_LEN
;
793 pbd1
->version
= pkc1
->version
;
794 pkc1
->prev
= pkc1
->nxt_offset
;
795 pkc1
->pkblk_end
= pkc1
->pkblk_start
+ pkc1
->kblk_size
;
797 prb_thaw_queue(pkc1
);
798 _prb_refresh_rx_retire_blk_timer(pkc1
);
804 * Queue freeze logic:
805 * 1) Assume tp_block_nr = 8 blocks.
806 * 2) At time 't0', user opens Rx ring.
807 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
808 * 4) user-space is either sleeping or processing block '0'.
809 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
810 * it will close block-7,loop around and try to fill block '0'.
812 * __packet_lookup_frame_in_block
813 * prb_retire_current_block()
814 * prb_dispatch_next_block()
815 * |->(BLOCK_STATUS == USER) evaluates to true
816 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
817 * 6) Now there are two cases:
818 * 6.1) Link goes idle right after the queue is frozen.
819 * But remember, the last open_block() refreshed the timer.
820 * When this timer expires,it will refresh itself so that we can
821 * re-open block-0 in near future.
822 * 6.2) Link is busy and keeps on receiving packets. This is a simple
823 * case and __packet_lookup_frame_in_block will check if block-0
824 * is free and can now be re-used.
826 static void prb_freeze_queue(struct tpacket_kbdq_core
*pkc
,
827 struct packet_sock
*po
)
829 pkc
->reset_pending_on_curr_blk
= 1;
830 po
->stats
.stats3
.tp_freeze_q_cnt
++;
833 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
836 * If the next block is free then we will dispatch it
837 * and return a good offset.
838 * Else, we will freeze the queue.
839 * So, caller must check the return value.
841 static void *prb_dispatch_next_block(struct tpacket_kbdq_core
*pkc
,
842 struct packet_sock
*po
)
844 struct tpacket_block_desc
*pbd
;
848 /* 1. Get current block num */
849 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
851 /* 2. If this block is currently in_use then freeze the queue */
852 if (TP_STATUS_USER
& BLOCK_STATUS(pbd
)) {
853 prb_freeze_queue(pkc
, po
);
859 * open this block and return the offset where the first packet
860 * needs to get stored.
862 prb_open_block(pkc
, pbd
);
863 return (void *)pkc
->nxt_offset
;
866 static void prb_retire_current_block(struct tpacket_kbdq_core
*pkc
,
867 struct packet_sock
*po
, unsigned int status
)
869 struct tpacket_block_desc
*pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
871 /* retire/close the current block */
872 if (likely(TP_STATUS_KERNEL
== BLOCK_STATUS(pbd
))) {
874 * Plug the case where copy_bits() is in progress on
875 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
876 * have space to copy the pkt in the current block and
877 * called prb_retire_current_block()
879 * We don't need to worry about the TMO case because
880 * the timer-handler already handled this case.
882 if (!(status
& TP_STATUS_BLK_TMO
)) {
883 while (atomic_read(&pkc
->blk_fill_in_prog
)) {
884 /* Waiting for skb_copy_bits to finish... */
888 prb_close_block(pkc
, pbd
, po
, status
);
893 static int prb_curr_blk_in_use(struct tpacket_kbdq_core
*pkc
,
894 struct tpacket_block_desc
*pbd
)
896 return TP_STATUS_USER
& BLOCK_STATUS(pbd
);
899 static int prb_queue_frozen(struct tpacket_kbdq_core
*pkc
)
901 return pkc
->reset_pending_on_curr_blk
;
904 static void prb_clear_blk_fill_status(struct packet_ring_buffer
*rb
)
906 struct tpacket_kbdq_core
*pkc
= GET_PBDQC_FROM_RB(rb
);
907 atomic_dec(&pkc
->blk_fill_in_prog
);
910 static void prb_fill_rxhash(struct tpacket_kbdq_core
*pkc
,
911 struct tpacket3_hdr
*ppd
)
913 ppd
->hv1
.tp_rxhash
= skb_get_rxhash(pkc
->skb
);
916 static void prb_clear_rxhash(struct tpacket_kbdq_core
*pkc
,
917 struct tpacket3_hdr
*ppd
)
919 ppd
->hv1
.tp_rxhash
= 0;
922 static void prb_fill_vlan_info(struct tpacket_kbdq_core
*pkc
,
923 struct tpacket3_hdr
*ppd
)
925 if (vlan_tx_tag_present(pkc
->skb
)) {
926 ppd
->hv1
.tp_vlan_tci
= vlan_tx_tag_get(pkc
->skb
);
927 ppd
->tp_status
= TP_STATUS_VLAN_VALID
;
929 ppd
->hv1
.tp_vlan_tci
= 0;
930 ppd
->tp_status
= TP_STATUS_AVAILABLE
;
934 static void prb_run_all_ft_ops(struct tpacket_kbdq_core
*pkc
,
935 struct tpacket3_hdr
*ppd
)
937 prb_fill_vlan_info(pkc
, ppd
);
939 if (pkc
->feature_req_word
& TP_FT_REQ_FILL_RXHASH
)
940 prb_fill_rxhash(pkc
, ppd
);
942 prb_clear_rxhash(pkc
, ppd
);
945 static void prb_fill_curr_block(char *curr
,
946 struct tpacket_kbdq_core
*pkc
,
947 struct tpacket_block_desc
*pbd
,
950 struct tpacket3_hdr
*ppd
;
952 ppd
= (struct tpacket3_hdr
*)curr
;
953 ppd
->tp_next_offset
= TOTAL_PKT_LEN_INCL_ALIGN(len
);
955 pkc
->nxt_offset
+= TOTAL_PKT_LEN_INCL_ALIGN(len
);
956 BLOCK_LEN(pbd
) += TOTAL_PKT_LEN_INCL_ALIGN(len
);
957 BLOCK_NUM_PKTS(pbd
) += 1;
958 atomic_inc(&pkc
->blk_fill_in_prog
);
959 prb_run_all_ft_ops(pkc
, ppd
);
962 /* Assumes caller has the sk->rx_queue.lock */
963 static void *__packet_lookup_frame_in_block(struct packet_sock
*po
,
969 struct tpacket_kbdq_core
*pkc
;
970 struct tpacket_block_desc
*pbd
;
973 pkc
= GET_PBDQC_FROM_RB(&po
->rx_ring
);
974 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
976 /* Queue is frozen when user space is lagging behind */
977 if (prb_queue_frozen(pkc
)) {
979 * Check if that last block which caused the queue to freeze,
980 * is still in_use by user-space.
982 if (prb_curr_blk_in_use(pkc
, pbd
)) {
983 /* Can't record this packet */
987 * Ok, the block was released by user-space.
988 * Now let's open that block.
989 * opening a block also thaws the queue.
990 * Thawing is a side effect.
992 prb_open_block(pkc
, pbd
);
997 curr
= pkc
->nxt_offset
;
999 end
= (char *)pbd
+ pkc
->kblk_size
;
1001 /* first try the current block */
1002 if (curr
+TOTAL_PKT_LEN_INCL_ALIGN(len
) < end
) {
1003 prb_fill_curr_block(curr
, pkc
, pbd
, len
);
1004 return (void *)curr
;
1007 /* Ok, close the current block */
1008 prb_retire_current_block(pkc
, po
, 0);
1010 /* Now, try to dispatch the next block */
1011 curr
= (char *)prb_dispatch_next_block(pkc
, po
);
1013 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
1014 prb_fill_curr_block(curr
, pkc
, pbd
, len
);
1015 return (void *)curr
;
1019 * No free blocks are available.user_space hasn't caught up yet.
1020 * Queue was just frozen and now this packet will get dropped.
1025 static void *packet_current_rx_frame(struct packet_sock
*po
,
1026 struct sk_buff
*skb
,
1027 int status
, unsigned int len
)
1030 switch (po
->tp_version
) {
1033 curr
= packet_lookup_frame(po
, &po
->rx_ring
,
1034 po
->rx_ring
.head
, status
);
1037 return __packet_lookup_frame_in_block(po
, skb
, status
, len
);
1039 WARN(1, "TPACKET version not supported\n");
1045 static void *prb_lookup_block(struct packet_sock
*po
,
1046 struct packet_ring_buffer
*rb
,
1050 struct tpacket_kbdq_core
*pkc
= GET_PBDQC_FROM_RB(rb
);
1051 struct tpacket_block_desc
*pbd
= GET_PBLOCK_DESC(pkc
, idx
);
1053 if (status
!= BLOCK_STATUS(pbd
))
1058 static int prb_previous_blk_num(struct packet_ring_buffer
*rb
)
1061 if (rb
->prb_bdqc
.kactive_blk_num
)
1062 prev
= rb
->prb_bdqc
.kactive_blk_num
-1;
1064 prev
= rb
->prb_bdqc
.knum_blocks
-1;
1068 /* Assumes caller has held the rx_queue.lock */
1069 static void *__prb_previous_block(struct packet_sock
*po
,
1070 struct packet_ring_buffer
*rb
,
1073 unsigned int previous
= prb_previous_blk_num(rb
);
1074 return prb_lookup_block(po
, rb
, previous
, status
);
1077 static void *packet_previous_rx_frame(struct packet_sock
*po
,
1078 struct packet_ring_buffer
*rb
,
1081 if (po
->tp_version
<= TPACKET_V2
)
1082 return packet_previous_frame(po
, rb
, status
);
1084 return __prb_previous_block(po
, rb
, status
);
1087 static void packet_increment_rx_head(struct packet_sock
*po
,
1088 struct packet_ring_buffer
*rb
)
1090 switch (po
->tp_version
) {
1093 return packet_increment_head(rb
);
1096 WARN(1, "TPACKET version not supported.\n");
1102 static void *packet_previous_frame(struct packet_sock
*po
,
1103 struct packet_ring_buffer
*rb
,
1106 unsigned int previous
= rb
->head
? rb
->head
- 1 : rb
->frame_max
;
1107 return packet_lookup_frame(po
, rb
, previous
, status
);
1110 static void packet_increment_head(struct packet_ring_buffer
*buff
)
1112 buff
->head
= buff
->head
!= buff
->frame_max
? buff
->head
+1 : 0;
1115 static bool packet_rcv_has_room(struct packet_sock
*po
, struct sk_buff
*skb
)
1117 struct sock
*sk
= &po
->sk
;
1120 if (po
->prot_hook
.func
!= tpacket_rcv
)
1121 return (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
)
1124 spin_lock(&sk
->sk_receive_queue
.lock
);
1125 if (po
->tp_version
== TPACKET_V3
)
1126 has_room
= prb_lookup_block(po
, &po
->rx_ring
,
1127 po
->rx_ring
.prb_bdqc
.kactive_blk_num
,
1130 has_room
= packet_lookup_frame(po
, &po
->rx_ring
,
1133 spin_unlock(&sk
->sk_receive_queue
.lock
);
1138 static void packet_sock_destruct(struct sock
*sk
)
1140 skb_queue_purge(&sk
->sk_error_queue
);
1142 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
1143 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
1145 if (!sock_flag(sk
, SOCK_DEAD
)) {
1146 pr_err("Attempt to release alive packet socket: %p\n", sk
);
1150 sk_refcnt_debug_dec(sk
);
1153 static unsigned int fanout_demux_hash(struct packet_fanout
*f
,
1154 struct sk_buff
*skb
,
1157 return (((u64
)skb
->rxhash
) * num
) >> 32;
1160 static unsigned int fanout_demux_lb(struct packet_fanout
*f
,
1161 struct sk_buff
*skb
,
1164 unsigned int val
= atomic_inc_return(&f
->rr_cur
);
1169 static unsigned int fanout_demux_cpu(struct packet_fanout
*f
,
1170 struct sk_buff
*skb
,
1173 return smp_processor_id() % num
;
1176 static unsigned int fanout_demux_rollover(struct packet_fanout
*f
,
1177 struct sk_buff
*skb
,
1178 unsigned int idx
, unsigned int skip
,
1183 i
= j
= min_t(int, f
->next
[idx
], num
- 1);
1185 if (i
!= skip
&& packet_rcv_has_room(pkt_sk(f
->arr
[i
]), skb
)) {
1197 static bool fanout_has_flag(struct packet_fanout
*f
, u16 flag
)
1199 return f
->flags
& (flag
>> 8);
1202 static int packet_rcv_fanout(struct sk_buff
*skb
, struct net_device
*dev
,
1203 struct packet_type
*pt
, struct net_device
*orig_dev
)
1205 struct packet_fanout
*f
= pt
->af_packet_priv
;
1206 unsigned int num
= ACCESS_ONCE(f
->num_members
);
1207 struct packet_sock
*po
;
1210 if (!net_eq(dev_net(dev
), read_pnet(&f
->net
)) ||
1217 case PACKET_FANOUT_HASH
:
1219 if (fanout_has_flag(f
, PACKET_FANOUT_FLAG_DEFRAG
)) {
1220 skb
= ip_check_defrag(skb
, IP_DEFRAG_AF_PACKET
);
1224 skb_get_rxhash(skb
);
1225 idx
= fanout_demux_hash(f
, skb
, num
);
1227 case PACKET_FANOUT_LB
:
1228 idx
= fanout_demux_lb(f
, skb
, num
);
1230 case PACKET_FANOUT_CPU
:
1231 idx
= fanout_demux_cpu(f
, skb
, num
);
1233 case PACKET_FANOUT_ROLLOVER
:
1234 idx
= fanout_demux_rollover(f
, skb
, 0, (unsigned int) -1, num
);
1238 po
= pkt_sk(f
->arr
[idx
]);
1239 if (fanout_has_flag(f
, PACKET_FANOUT_FLAG_ROLLOVER
) &&
1240 unlikely(!packet_rcv_has_room(po
, skb
))) {
1241 idx
= fanout_demux_rollover(f
, skb
, idx
, idx
, num
);
1242 po
= pkt_sk(f
->arr
[idx
]);
1245 return po
->prot_hook
.func(skb
, dev
, &po
->prot_hook
, orig_dev
);
1248 DEFINE_MUTEX(fanout_mutex
);
1249 EXPORT_SYMBOL_GPL(fanout_mutex
);
1250 static LIST_HEAD(fanout_list
);
1252 static void __fanout_link(struct sock
*sk
, struct packet_sock
*po
)
1254 struct packet_fanout
*f
= po
->fanout
;
1256 spin_lock(&f
->lock
);
1257 f
->arr
[f
->num_members
] = sk
;
1260 spin_unlock(&f
->lock
);
1263 static void __fanout_unlink(struct sock
*sk
, struct packet_sock
*po
)
1265 struct packet_fanout
*f
= po
->fanout
;
1268 spin_lock(&f
->lock
);
1269 for (i
= 0; i
< f
->num_members
; i
++) {
1270 if (f
->arr
[i
] == sk
)
1273 BUG_ON(i
>= f
->num_members
);
1274 f
->arr
[i
] = f
->arr
[f
->num_members
- 1];
1276 spin_unlock(&f
->lock
);
1279 static bool match_fanout_group(struct packet_type
*ptype
, struct sock
* sk
)
1281 if (ptype
->af_packet_priv
== (void*)((struct packet_sock
*)sk
)->fanout
)
1287 static int fanout_add(struct sock
*sk
, u16 id
, u16 type_flags
)
1289 struct packet_sock
*po
= pkt_sk(sk
);
1290 struct packet_fanout
*f
, *match
;
1291 u8 type
= type_flags
& 0xff;
1292 u8 flags
= type_flags
>> 8;
1296 case PACKET_FANOUT_ROLLOVER
:
1297 if (type_flags
& PACKET_FANOUT_FLAG_ROLLOVER
)
1299 case PACKET_FANOUT_HASH
:
1300 case PACKET_FANOUT_LB
:
1301 case PACKET_FANOUT_CPU
:
1313 mutex_lock(&fanout_mutex
);
1315 list_for_each_entry(f
, &fanout_list
, list
) {
1317 read_pnet(&f
->net
) == sock_net(sk
)) {
1323 if (match
&& match
->flags
!= flags
)
1327 match
= kzalloc(sizeof(*match
), GFP_KERNEL
);
1330 write_pnet(&match
->net
, sock_net(sk
));
1333 match
->flags
= flags
;
1334 atomic_set(&match
->rr_cur
, 0);
1335 INIT_LIST_HEAD(&match
->list
);
1336 spin_lock_init(&match
->lock
);
1337 atomic_set(&match
->sk_ref
, 0);
1338 match
->prot_hook
.type
= po
->prot_hook
.type
;
1339 match
->prot_hook
.dev
= po
->prot_hook
.dev
;
1340 match
->prot_hook
.func
= packet_rcv_fanout
;
1341 match
->prot_hook
.af_packet_priv
= match
;
1342 match
->prot_hook
.id_match
= match_fanout_group
;
1343 dev_add_pack(&match
->prot_hook
);
1344 list_add(&match
->list
, &fanout_list
);
1347 if (match
->type
== type
&&
1348 match
->prot_hook
.type
== po
->prot_hook
.type
&&
1349 match
->prot_hook
.dev
== po
->prot_hook
.dev
) {
1351 if (atomic_read(&match
->sk_ref
) < PACKET_FANOUT_MAX
) {
1352 __dev_remove_pack(&po
->prot_hook
);
1354 atomic_inc(&match
->sk_ref
);
1355 __fanout_link(sk
, po
);
1360 mutex_unlock(&fanout_mutex
);
1364 static void fanout_release(struct sock
*sk
)
1366 struct packet_sock
*po
= pkt_sk(sk
);
1367 struct packet_fanout
*f
;
1373 mutex_lock(&fanout_mutex
);
1376 if (atomic_dec_and_test(&f
->sk_ref
)) {
1378 dev_remove_pack(&f
->prot_hook
);
1381 mutex_unlock(&fanout_mutex
);
1384 static const struct proto_ops packet_ops
;
1386 static const struct proto_ops packet_ops_spkt
;
1388 static int packet_rcv_spkt(struct sk_buff
*skb
, struct net_device
*dev
,
1389 struct packet_type
*pt
, struct net_device
*orig_dev
)
1392 struct sockaddr_pkt
*spkt
;
1395 * When we registered the protocol we saved the socket in the data
1396 * field for just this event.
1399 sk
= pt
->af_packet_priv
;
1402 * Yank back the headers [hope the device set this
1403 * right or kerboom...]
1405 * Incoming packets have ll header pulled,
1408 * For outgoing ones skb->data == skb_mac_header(skb)
1409 * so that this procedure is noop.
1412 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1415 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1418 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1422 /* drop any routing info */
1425 /* drop conntrack reference */
1428 spkt
= &PACKET_SKB_CB(skb
)->sa
.pkt
;
1430 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1433 * The SOCK_PACKET socket receives _all_ frames.
1436 spkt
->spkt_family
= dev
->type
;
1437 strlcpy(spkt
->spkt_device
, dev
->name
, sizeof(spkt
->spkt_device
));
1438 spkt
->spkt_protocol
= skb
->protocol
;
1441 * Charge the memory to the socket. This is done specifically
1442 * to prevent sockets using all the memory up.
1445 if (sock_queue_rcv_skb(sk
, skb
) == 0)
1456 * Output a raw packet to a device layer. This bypasses all the other
1457 * protocol layers and you must therefore supply it with a complete frame
1460 static int packet_sendmsg_spkt(struct kiocb
*iocb
, struct socket
*sock
,
1461 struct msghdr
*msg
, size_t len
)
1463 struct sock
*sk
= sock
->sk
;
1464 struct sockaddr_pkt
*saddr
= (struct sockaddr_pkt
*)msg
->msg_name
;
1465 struct sk_buff
*skb
= NULL
;
1466 struct net_device
*dev
;
1472 * Get and verify the address.
1476 if (msg
->msg_namelen
< sizeof(struct sockaddr
))
1478 if (msg
->msg_namelen
== sizeof(struct sockaddr_pkt
))
1479 proto
= saddr
->spkt_protocol
;
1481 return -ENOTCONN
; /* SOCK_PACKET must be sent giving an address */
1484 * Find the device first to size check it
1487 saddr
->spkt_device
[sizeof(saddr
->spkt_device
) - 1] = 0;
1490 dev
= dev_get_by_name_rcu(sock_net(sk
), saddr
->spkt_device
);
1496 if (!(dev
->flags
& IFF_UP
))
1500 * You may not queue a frame bigger than the mtu. This is the lowest level
1501 * raw protocol and you must do your own fragmentation at this level.
1504 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
1505 if (!netif_supports_nofcs(dev
)) {
1506 err
= -EPROTONOSUPPORT
;
1509 extra_len
= 4; /* We're doing our own CRC */
1513 if (len
> dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
+ extra_len
)
1517 size_t reserved
= LL_RESERVED_SPACE(dev
);
1518 int tlen
= dev
->needed_tailroom
;
1519 unsigned int hhlen
= dev
->header_ops
? dev
->hard_header_len
: 0;
1522 skb
= sock_wmalloc(sk
, len
+ reserved
+ tlen
, 0, GFP_KERNEL
);
1525 /* FIXME: Save some space for broken drivers that write a hard
1526 * header at transmission time by themselves. PPP is the notable
1527 * one here. This should really be fixed at the driver level.
1529 skb_reserve(skb
, reserved
);
1530 skb_reset_network_header(skb
);
1532 /* Try to align data part correctly */
1537 skb_reset_network_header(skb
);
1539 err
= memcpy_fromiovec(skb_put(skb
, len
), msg
->msg_iov
, len
);
1545 if (len
> (dev
->mtu
+ dev
->hard_header_len
+ extra_len
)) {
1546 /* Earlier code assumed this would be a VLAN pkt,
1547 * double-check this now that we have the actual
1550 struct ethhdr
*ehdr
;
1551 skb_reset_mac_header(skb
);
1552 ehdr
= eth_hdr(skb
);
1553 if (ehdr
->h_proto
!= htons(ETH_P_8021Q
)) {
1559 skb
->protocol
= proto
;
1561 skb
->priority
= sk
->sk_priority
;
1562 skb
->mark
= sk
->sk_mark
;
1564 sock_tx_timestamp(sk
, &skb_shinfo(skb
)->tx_flags
);
1566 if (unlikely(extra_len
== 4))
1569 skb_probe_transport_header(skb
, 0);
1571 dev_queue_xmit(skb
);
1582 static unsigned int run_filter(const struct sk_buff
*skb
,
1583 const struct sock
*sk
,
1586 struct sk_filter
*filter
;
1589 filter
= rcu_dereference(sk
->sk_filter
);
1591 res
= SK_RUN_FILTER(filter
, skb
);
1598 * This function makes lazy skb cloning in hope that most of packets
1599 * are discarded by BPF.
1601 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1602 * and skb->cb are mangled. It works because (and until) packets
1603 * falling here are owned by current CPU. Output packets are cloned
1604 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1605 * sequencially, so that if we return skb to original state on exit,
1606 * we will not harm anyone.
1609 static int packet_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
1610 struct packet_type
*pt
, struct net_device
*orig_dev
)
1613 struct sockaddr_ll
*sll
;
1614 struct packet_sock
*po
;
1615 u8
*skb_head
= skb
->data
;
1616 int skb_len
= skb
->len
;
1617 unsigned int snaplen
, res
;
1619 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1622 sk
= pt
->af_packet_priv
;
1625 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1630 if (dev
->header_ops
) {
1631 /* The device has an explicit notion of ll header,
1632 * exported to higher levels.
1634 * Otherwise, the device hides details of its frame
1635 * structure, so that corresponding packet head is
1636 * never delivered to user.
1638 if (sk
->sk_type
!= SOCK_DGRAM
)
1639 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1640 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
1641 /* Special case: outgoing packets have ll header at head */
1642 skb_pull(skb
, skb_network_offset(skb
));
1648 res
= run_filter(skb
, sk
, snaplen
);
1650 goto drop_n_restore
;
1654 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
1657 if (skb_shared(skb
)) {
1658 struct sk_buff
*nskb
= skb_clone(skb
, GFP_ATOMIC
);
1662 if (skb_head
!= skb
->data
) {
1663 skb
->data
= skb_head
;
1670 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb
)) + MAX_ADDR_LEN
- 8 >
1673 sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
1674 sll
->sll_family
= AF_PACKET
;
1675 sll
->sll_hatype
= dev
->type
;
1676 sll
->sll_protocol
= skb
->protocol
;
1677 sll
->sll_pkttype
= skb
->pkt_type
;
1678 if (unlikely(po
->origdev
))
1679 sll
->sll_ifindex
= orig_dev
->ifindex
;
1681 sll
->sll_ifindex
= dev
->ifindex
;
1683 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
1685 PACKET_SKB_CB(skb
)->origlen
= skb
->len
;
1687 if (pskb_trim(skb
, snaplen
))
1690 skb_set_owner_r(skb
, sk
);
1694 /* drop conntrack reference */
1697 spin_lock(&sk
->sk_receive_queue
.lock
);
1698 po
->stats
.stats1
.tp_packets
++;
1699 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
1700 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1701 spin_unlock(&sk
->sk_receive_queue
.lock
);
1702 sk
->sk_data_ready(sk
, skb
->len
);
1706 spin_lock(&sk
->sk_receive_queue
.lock
);
1707 po
->stats
.stats1
.tp_drops
++;
1708 atomic_inc(&sk
->sk_drops
);
1709 spin_unlock(&sk
->sk_receive_queue
.lock
);
1712 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
1713 skb
->data
= skb_head
;
1721 static int tpacket_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
1722 struct packet_type
*pt
, struct net_device
*orig_dev
)
1725 struct packet_sock
*po
;
1726 struct sockaddr_ll
*sll
;
1727 union tpacket_uhdr h
;
1728 u8
*skb_head
= skb
->data
;
1729 int skb_len
= skb
->len
;
1730 unsigned int snaplen
, res
;
1731 unsigned long status
= TP_STATUS_USER
;
1732 unsigned short macoff
, netoff
, hdrlen
;
1733 struct sk_buff
*copy_skb
= NULL
;
1737 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1740 sk
= pt
->af_packet_priv
;
1743 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1746 if (dev
->header_ops
) {
1747 if (sk
->sk_type
!= SOCK_DGRAM
)
1748 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1749 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
1750 /* Special case: outgoing packets have ll header at head */
1751 skb_pull(skb
, skb_network_offset(skb
));
1755 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1756 status
|= TP_STATUS_CSUMNOTREADY
;
1760 res
= run_filter(skb
, sk
, snaplen
);
1762 goto drop_n_restore
;
1766 if (sk
->sk_type
== SOCK_DGRAM
) {
1767 macoff
= netoff
= TPACKET_ALIGN(po
->tp_hdrlen
) + 16 +
1770 unsigned int maclen
= skb_network_offset(skb
);
1771 netoff
= TPACKET_ALIGN(po
->tp_hdrlen
+
1772 (maclen
< 16 ? 16 : maclen
)) +
1774 macoff
= netoff
- maclen
;
1776 if (po
->tp_version
<= TPACKET_V2
) {
1777 if (macoff
+ snaplen
> po
->rx_ring
.frame_size
) {
1778 if (po
->copy_thresh
&&
1779 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1780 if (skb_shared(skb
)) {
1781 copy_skb
= skb_clone(skb
, GFP_ATOMIC
);
1783 copy_skb
= skb_get(skb
);
1784 skb_head
= skb
->data
;
1787 skb_set_owner_r(copy_skb
, sk
);
1789 snaplen
= po
->rx_ring
.frame_size
- macoff
;
1790 if ((int)snaplen
< 0)
1793 } else if (unlikely(macoff
+ snaplen
>
1794 GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
)) {
1797 nval
= GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
- macoff
;
1798 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1799 snaplen
, nval
, macoff
);
1801 if (unlikely((int)snaplen
< 0)) {
1803 macoff
= GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
;
1806 spin_lock(&sk
->sk_receive_queue
.lock
);
1807 h
.raw
= packet_current_rx_frame(po
, skb
,
1808 TP_STATUS_KERNEL
, (macoff
+snaplen
));
1811 if (po
->tp_version
<= TPACKET_V2
) {
1812 packet_increment_rx_head(po
, &po
->rx_ring
);
1814 * LOSING will be reported till you read the stats,
1815 * because it's COR - Clear On Read.
1816 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1819 if (po
->stats
.stats1
.tp_drops
)
1820 status
|= TP_STATUS_LOSING
;
1822 po
->stats
.stats1
.tp_packets
++;
1824 status
|= TP_STATUS_COPY
;
1825 __skb_queue_tail(&sk
->sk_receive_queue
, copy_skb
);
1827 spin_unlock(&sk
->sk_receive_queue
.lock
);
1829 skb_copy_bits(skb
, 0, h
.raw
+ macoff
, snaplen
);
1831 if (!(ts_status
= tpacket_get_timestamp(skb
, &ts
, po
->tp_tstamp
)))
1832 getnstimeofday(&ts
);
1834 status
|= ts_status
;
1836 switch (po
->tp_version
) {
1838 h
.h1
->tp_len
= skb
->len
;
1839 h
.h1
->tp_snaplen
= snaplen
;
1840 h
.h1
->tp_mac
= macoff
;
1841 h
.h1
->tp_net
= netoff
;
1842 h
.h1
->tp_sec
= ts
.tv_sec
;
1843 h
.h1
->tp_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
1844 hdrlen
= sizeof(*h
.h1
);
1847 h
.h2
->tp_len
= skb
->len
;
1848 h
.h2
->tp_snaplen
= snaplen
;
1849 h
.h2
->tp_mac
= macoff
;
1850 h
.h2
->tp_net
= netoff
;
1851 h
.h2
->tp_sec
= ts
.tv_sec
;
1852 h
.h2
->tp_nsec
= ts
.tv_nsec
;
1853 if (vlan_tx_tag_present(skb
)) {
1854 h
.h2
->tp_vlan_tci
= vlan_tx_tag_get(skb
);
1855 status
|= TP_STATUS_VLAN_VALID
;
1857 h
.h2
->tp_vlan_tci
= 0;
1859 h
.h2
->tp_padding
= 0;
1860 hdrlen
= sizeof(*h
.h2
);
1863 /* tp_nxt_offset,vlan are already populated above.
1864 * So DONT clear those fields here
1866 h
.h3
->tp_status
|= status
;
1867 h
.h3
->tp_len
= skb
->len
;
1868 h
.h3
->tp_snaplen
= snaplen
;
1869 h
.h3
->tp_mac
= macoff
;
1870 h
.h3
->tp_net
= netoff
;
1871 h
.h3
->tp_sec
= ts
.tv_sec
;
1872 h
.h3
->tp_nsec
= ts
.tv_nsec
;
1873 hdrlen
= sizeof(*h
.h3
);
1879 sll
= h
.raw
+ TPACKET_ALIGN(hdrlen
);
1880 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
1881 sll
->sll_family
= AF_PACKET
;
1882 sll
->sll_hatype
= dev
->type
;
1883 sll
->sll_protocol
= skb
->protocol
;
1884 sll
->sll_pkttype
= skb
->pkt_type
;
1885 if (unlikely(po
->origdev
))
1886 sll
->sll_ifindex
= orig_dev
->ifindex
;
1888 sll
->sll_ifindex
= dev
->ifindex
;
1891 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1895 if (po
->tp_version
<= TPACKET_V2
) {
1896 end
= (u8
*)PAGE_ALIGN((unsigned long)h
.raw
1897 + macoff
+ snaplen
);
1898 for (start
= h
.raw
; start
< end
; start
+= PAGE_SIZE
)
1899 flush_dcache_page(pgv_to_page(start
));
1904 if (po
->tp_version
<= TPACKET_V2
)
1905 __packet_set_status(po
, h
.raw
, status
);
1907 prb_clear_blk_fill_status(&po
->rx_ring
);
1909 sk
->sk_data_ready(sk
, 0);
1912 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
1913 skb
->data
= skb_head
;
1921 po
->stats
.stats1
.tp_drops
++;
1922 spin_unlock(&sk
->sk_receive_queue
.lock
);
1924 sk
->sk_data_ready(sk
, 0);
1925 kfree_skb(copy_skb
);
1926 goto drop_n_restore
;
1929 static void tpacket_destruct_skb(struct sk_buff
*skb
)
1931 struct packet_sock
*po
= pkt_sk(skb
->sk
);
1934 if (likely(po
->tx_ring
.pg_vec
)) {
1937 ph
= skb_shinfo(skb
)->destructor_arg
;
1938 BUG_ON(atomic_read(&po
->tx_ring
.pending
) == 0);
1939 atomic_dec(&po
->tx_ring
.pending
);
1941 ts
= __packet_set_timestamp(po
, ph
, skb
);
1942 __packet_set_status(po
, ph
, TP_STATUS_AVAILABLE
| ts
);
1948 static int tpacket_fill_skb(struct packet_sock
*po
, struct sk_buff
*skb
,
1949 void *frame
, struct net_device
*dev
, int size_max
,
1950 __be16 proto
, unsigned char *addr
, int hlen
)
1952 union tpacket_uhdr ph
;
1953 int to_write
, offset
, len
, tp_len
, nr_frags
, len_max
;
1954 struct socket
*sock
= po
->sk
.sk_socket
;
1961 skb
->protocol
= proto
;
1963 skb
->priority
= po
->sk
.sk_priority
;
1964 skb
->mark
= po
->sk
.sk_mark
;
1965 sock_tx_timestamp(&po
->sk
, &skb_shinfo(skb
)->tx_flags
);
1966 skb_shinfo(skb
)->destructor_arg
= ph
.raw
;
1968 switch (po
->tp_version
) {
1970 tp_len
= ph
.h2
->tp_len
;
1973 tp_len
= ph
.h1
->tp_len
;
1976 if (unlikely(tp_len
> size_max
)) {
1977 pr_err("packet size is too long (%d > %d)\n", tp_len
, size_max
);
1981 skb_reserve(skb
, hlen
);
1982 skb_reset_network_header(skb
);
1983 skb_probe_transport_header(skb
, 0);
1985 if (po
->tp_tx_has_off
) {
1986 int off_min
, off_max
, off
;
1987 off_min
= po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
1988 off_max
= po
->tx_ring
.frame_size
- tp_len
;
1989 if (sock
->type
== SOCK_DGRAM
) {
1990 switch (po
->tp_version
) {
1992 off
= ph
.h2
->tp_net
;
1995 off
= ph
.h1
->tp_net
;
1999 switch (po
->tp_version
) {
2001 off
= ph
.h2
->tp_mac
;
2004 off
= ph
.h1
->tp_mac
;
2008 if (unlikely((off
< off_min
) || (off_max
< off
)))
2010 data
= ph
.raw
+ off
;
2012 data
= ph
.raw
+ po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
2016 if (sock
->type
== SOCK_DGRAM
) {
2017 err
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
,
2019 if (unlikely(err
< 0))
2021 } else if (dev
->hard_header_len
) {
2022 /* net device doesn't like empty head */
2023 if (unlikely(tp_len
<= dev
->hard_header_len
)) {
2024 pr_err("packet size is too short (%d < %d)\n",
2025 tp_len
, dev
->hard_header_len
);
2029 skb_push(skb
, dev
->hard_header_len
);
2030 err
= skb_store_bits(skb
, 0, data
,
2031 dev
->hard_header_len
);
2035 data
+= dev
->hard_header_len
;
2036 to_write
-= dev
->hard_header_len
;
2039 offset
= offset_in_page(data
);
2040 len_max
= PAGE_SIZE
- offset
;
2041 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2043 skb
->data_len
= to_write
;
2044 skb
->len
+= to_write
;
2045 skb
->truesize
+= to_write
;
2046 atomic_add(to_write
, &po
->sk
.sk_wmem_alloc
);
2048 while (likely(to_write
)) {
2049 nr_frags
= skb_shinfo(skb
)->nr_frags
;
2051 if (unlikely(nr_frags
>= MAX_SKB_FRAGS
)) {
2052 pr_err("Packet exceed the number of skb frags(%lu)\n",
2057 page
= pgv_to_page(data
);
2059 flush_dcache_page(page
);
2061 skb_fill_page_desc(skb
, nr_frags
, page
, offset
, len
);
2064 len_max
= PAGE_SIZE
;
2065 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2071 static int tpacket_snd(struct packet_sock
*po
, struct msghdr
*msg
)
2073 struct sk_buff
*skb
;
2074 struct net_device
*dev
;
2076 int err
, reserve
= 0;
2078 struct sockaddr_ll
*saddr
= (struct sockaddr_ll
*)msg
->msg_name
;
2079 int tp_len
, size_max
;
2080 unsigned char *addr
;
2082 int status
= TP_STATUS_AVAILABLE
;
2085 mutex_lock(&po
->pg_vec_lock
);
2087 if (likely(saddr
== NULL
)) {
2088 dev
= packet_cached_dev_get(po
);
2093 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2095 if (msg
->msg_namelen
< (saddr
->sll_halen
2096 + offsetof(struct sockaddr_ll
,
2099 proto
= saddr
->sll_protocol
;
2100 addr
= saddr
->sll_addr
;
2101 dev
= dev_get_by_index(sock_net(&po
->sk
), saddr
->sll_ifindex
);
2105 if (unlikely(dev
== NULL
))
2108 if (unlikely(!(dev
->flags
& IFF_UP
)))
2111 reserve
= dev
->hard_header_len
;
2113 size_max
= po
->tx_ring
.frame_size
2114 - (po
->tp_hdrlen
- sizeof(struct sockaddr_ll
));
2116 if (size_max
> dev
->mtu
+ reserve
)
2117 size_max
= dev
->mtu
+ reserve
;
2120 ph
= packet_current_frame(po
, &po
->tx_ring
,
2121 TP_STATUS_SEND_REQUEST
);
2123 if (unlikely(ph
== NULL
)) {
2128 status
= TP_STATUS_SEND_REQUEST
;
2129 hlen
= LL_RESERVED_SPACE(dev
);
2130 tlen
= dev
->needed_tailroom
;
2131 skb
= sock_alloc_send_skb(&po
->sk
,
2132 hlen
+ tlen
+ sizeof(struct sockaddr_ll
),
2135 if (unlikely(skb
== NULL
))
2138 tp_len
= tpacket_fill_skb(po
, skb
, ph
, dev
, size_max
, proto
,
2141 if (unlikely(tp_len
< 0)) {
2143 __packet_set_status(po
, ph
,
2144 TP_STATUS_AVAILABLE
);
2145 packet_increment_head(&po
->tx_ring
);
2149 status
= TP_STATUS_WRONG_FORMAT
;
2155 skb
->destructor
= tpacket_destruct_skb
;
2156 __packet_set_status(po
, ph
, TP_STATUS_SENDING
);
2157 atomic_inc(&po
->tx_ring
.pending
);
2159 status
= TP_STATUS_SEND_REQUEST
;
2160 err
= dev_queue_xmit(skb
);
2161 if (unlikely(err
> 0)) {
2162 err
= net_xmit_errno(err
);
2163 if (err
&& __packet_get_status(po
, ph
) ==
2164 TP_STATUS_AVAILABLE
) {
2165 /* skb was destructed already */
2170 * skb was dropped but not destructed yet;
2171 * let's treat it like congestion or err < 0
2175 packet_increment_head(&po
->tx_ring
);
2177 } while (likely((ph
!= NULL
) ||
2178 ((!(msg
->msg_flags
& MSG_DONTWAIT
)) &&
2179 (atomic_read(&po
->tx_ring
.pending
))))
2186 __packet_set_status(po
, ph
, status
);
2191 mutex_unlock(&po
->pg_vec_lock
);
2195 static struct sk_buff
*packet_alloc_skb(struct sock
*sk
, size_t prepad
,
2196 size_t reserve
, size_t len
,
2197 size_t linear
, int noblock
,
2200 struct sk_buff
*skb
;
2202 /* Under a page? Don't bother with paged skb. */
2203 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
2206 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
2211 skb_reserve(skb
, reserve
);
2212 skb_put(skb
, linear
);
2213 skb
->data_len
= len
- linear
;
2214 skb
->len
+= len
- linear
;
2219 static int packet_snd(struct socket
*sock
,
2220 struct msghdr
*msg
, size_t len
)
2222 struct sock
*sk
= sock
->sk
;
2223 struct sockaddr_ll
*saddr
= (struct sockaddr_ll
*)msg
->msg_name
;
2224 struct sk_buff
*skb
;
2225 struct net_device
*dev
;
2227 unsigned char *addr
;
2228 int err
, reserve
= 0;
2229 struct virtio_net_hdr vnet_hdr
= { 0 };
2232 struct packet_sock
*po
= pkt_sk(sk
);
2233 unsigned short gso_type
= 0;
2238 * Get and verify the address.
2241 if (likely(saddr
== NULL
)) {
2242 dev
= packet_cached_dev_get(po
);
2247 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2249 if (msg
->msg_namelen
< (saddr
->sll_halen
+ offsetof(struct sockaddr_ll
, sll_addr
)))
2251 proto
= saddr
->sll_protocol
;
2252 addr
= saddr
->sll_addr
;
2253 dev
= dev_get_by_index(sock_net(sk
), saddr
->sll_ifindex
);
2257 if (unlikely(dev
== NULL
))
2260 if (unlikely(!(dev
->flags
& IFF_UP
)))
2263 if (sock
->type
== SOCK_RAW
)
2264 reserve
= dev
->hard_header_len
;
2265 if (po
->has_vnet_hdr
) {
2266 vnet_hdr_len
= sizeof(vnet_hdr
);
2269 if (len
< vnet_hdr_len
)
2272 len
-= vnet_hdr_len
;
2274 err
= memcpy_fromiovec((void *)&vnet_hdr
, msg
->msg_iov
,
2279 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
2280 (vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
2282 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
2283 vnet_hdr
.csum_offset
+ 2;
2286 if (vnet_hdr
.hdr_len
> len
)
2289 if (vnet_hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
2290 switch (vnet_hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
2291 case VIRTIO_NET_HDR_GSO_TCPV4
:
2292 gso_type
= SKB_GSO_TCPV4
;
2294 case VIRTIO_NET_HDR_GSO_TCPV6
:
2295 gso_type
= SKB_GSO_TCPV6
;
2297 case VIRTIO_NET_HDR_GSO_UDP
:
2298 gso_type
= SKB_GSO_UDP
;
2304 if (vnet_hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
2305 gso_type
|= SKB_GSO_TCP_ECN
;
2307 if (vnet_hdr
.gso_size
== 0)
2313 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
2314 if (!netif_supports_nofcs(dev
)) {
2315 err
= -EPROTONOSUPPORT
;
2318 extra_len
= 4; /* We're doing our own CRC */
2322 if (!gso_type
&& (len
> dev
->mtu
+ reserve
+ VLAN_HLEN
+ extra_len
))
2326 hlen
= LL_RESERVED_SPACE(dev
);
2327 tlen
= dev
->needed_tailroom
;
2328 skb
= packet_alloc_skb(sk
, hlen
+ tlen
, hlen
, len
, vnet_hdr
.hdr_len
,
2329 msg
->msg_flags
& MSG_DONTWAIT
, &err
);
2333 skb_set_network_header(skb
, reserve
);
2336 if (sock
->type
== SOCK_DGRAM
&&
2337 (offset
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
, NULL
, len
)) < 0)
2340 /* Returns -EFAULT on error */
2341 err
= skb_copy_datagram_from_iovec(skb
, offset
, msg
->msg_iov
, 0, len
);
2345 sock_tx_timestamp(sk
, &skb_shinfo(skb
)->tx_flags
);
2347 if (!gso_type
&& (len
> dev
->mtu
+ reserve
+ extra_len
)) {
2348 /* Earlier code assumed this would be a VLAN pkt,
2349 * double-check this now that we have the actual
2352 struct ethhdr
*ehdr
;
2353 skb_reset_mac_header(skb
);
2354 ehdr
= eth_hdr(skb
);
2355 if (ehdr
->h_proto
!= htons(ETH_P_8021Q
)) {
2361 skb
->protocol
= proto
;
2363 skb
->priority
= sk
->sk_priority
;
2364 skb
->mark
= sk
->sk_mark
;
2366 if (po
->has_vnet_hdr
) {
2367 if (vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
2368 if (!skb_partial_csum_set(skb
, vnet_hdr
.csum_start
,
2369 vnet_hdr
.csum_offset
)) {
2375 skb_shinfo(skb
)->gso_size
= vnet_hdr
.gso_size
;
2376 skb_shinfo(skb
)->gso_type
= gso_type
;
2378 /* Header must be checked, and gso_segs computed. */
2379 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
2380 skb_shinfo(skb
)->gso_segs
= 0;
2382 len
+= vnet_hdr_len
;
2385 skb_probe_transport_header(skb
, reserve
);
2387 if (unlikely(extra_len
== 4))
2394 err
= dev_queue_xmit(skb
);
2395 if (err
> 0 && (err
= net_xmit_errno(err
)) != 0)
2411 static int packet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
2412 struct msghdr
*msg
, size_t len
)
2414 struct sock
*sk
= sock
->sk
;
2415 struct packet_sock
*po
= pkt_sk(sk
);
2416 if (po
->tx_ring
.pg_vec
)
2417 return tpacket_snd(po
, msg
);
2419 return packet_snd(sock
, msg
, len
);
2423 * Close a PACKET socket. This is fairly simple. We immediately go
2424 * to 'closed' state and remove our protocol entry in the device list.
2427 static int packet_release(struct socket
*sock
)
2429 struct sock
*sk
= sock
->sk
;
2430 struct packet_sock
*po
;
2432 union tpacket_req_u req_u
;
2440 mutex_lock(&net
->packet
.sklist_lock
);
2441 sk_del_node_init_rcu(sk
);
2442 mutex_unlock(&net
->packet
.sklist_lock
);
2445 sock_prot_inuse_add(net
, sk
->sk_prot
, -1);
2448 spin_lock(&po
->bind_lock
);
2449 unregister_prot_hook(sk
, false);
2450 packet_cached_dev_reset(po
);
2452 if (po
->prot_hook
.dev
) {
2453 dev_put(po
->prot_hook
.dev
);
2454 po
->prot_hook
.dev
= NULL
;
2456 spin_unlock(&po
->bind_lock
);
2458 packet_flush_mclist(sk
);
2460 if (po
->rx_ring
.pg_vec
) {
2461 memset(&req_u
, 0, sizeof(req_u
));
2462 packet_set_ring(sk
, &req_u
, 1, 0);
2465 if (po
->tx_ring
.pg_vec
) {
2466 memset(&req_u
, 0, sizeof(req_u
));
2467 packet_set_ring(sk
, &req_u
, 1, 1);
2474 * Now the socket is dead. No more input will appear.
2481 skb_queue_purge(&sk
->sk_receive_queue
);
2482 sk_refcnt_debug_release(sk
);
2489 * Attach a packet hook.
2492 static int packet_do_bind(struct sock
*sk
, struct net_device
*dev
, __be16 protocol
)
2494 struct packet_sock
*po
= pkt_sk(sk
);
2505 spin_lock(&po
->bind_lock
);
2506 unregister_prot_hook(sk
, true);
2509 po
->prot_hook
.type
= protocol
;
2510 if (po
->prot_hook
.dev
)
2511 dev_put(po
->prot_hook
.dev
);
2513 po
->prot_hook
.dev
= dev
;
2514 po
->ifindex
= dev
? dev
->ifindex
: 0;
2516 packet_cached_dev_assign(po
, dev
);
2521 if (!dev
|| (dev
->flags
& IFF_UP
)) {
2522 register_prot_hook(sk
);
2524 sk
->sk_err
= ENETDOWN
;
2525 if (!sock_flag(sk
, SOCK_DEAD
))
2526 sk
->sk_error_report(sk
);
2530 spin_unlock(&po
->bind_lock
);
2536 * Bind a packet socket to a device
2539 static int packet_bind_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
2542 struct sock
*sk
= sock
->sk
;
2544 struct net_device
*dev
;
2551 if (addr_len
!= sizeof(struct sockaddr
))
2553 strlcpy(name
, uaddr
->sa_data
, sizeof(name
));
2555 dev
= dev_get_by_name(sock_net(sk
), name
);
2557 err
= packet_do_bind(sk
, dev
, pkt_sk(sk
)->num
);
2561 static int packet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
2563 struct sockaddr_ll
*sll
= (struct sockaddr_ll
*)uaddr
;
2564 struct sock
*sk
= sock
->sk
;
2565 struct net_device
*dev
= NULL
;
2573 if (addr_len
< sizeof(struct sockaddr_ll
))
2575 if (sll
->sll_family
!= AF_PACKET
)
2578 if (sll
->sll_ifindex
) {
2580 dev
= dev_get_by_index(sock_net(sk
), sll
->sll_ifindex
);
2584 err
= packet_do_bind(sk
, dev
, sll
->sll_protocol
? : pkt_sk(sk
)->num
);
2590 static struct proto packet_proto
= {
2592 .owner
= THIS_MODULE
,
2593 .obj_size
= sizeof(struct packet_sock
),
2597 * Create a packet of type SOCK_PACKET.
2600 static int packet_create(struct net
*net
, struct socket
*sock
, int protocol
,
2604 struct packet_sock
*po
;
2605 __be16 proto
= (__force __be16
)protocol
; /* weird, but documented */
2608 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
2610 if (sock
->type
!= SOCK_DGRAM
&& sock
->type
!= SOCK_RAW
&&
2611 sock
->type
!= SOCK_PACKET
)
2612 return -ESOCKTNOSUPPORT
;
2614 sock
->state
= SS_UNCONNECTED
;
2617 sk
= sk_alloc(net
, PF_PACKET
, GFP_KERNEL
, &packet_proto
);
2621 sock
->ops
= &packet_ops
;
2622 if (sock
->type
== SOCK_PACKET
)
2623 sock
->ops
= &packet_ops_spkt
;
2625 sock_init_data(sock
, sk
);
2628 sk
->sk_family
= PF_PACKET
;
2631 packet_cached_dev_reset(po
);
2633 sk
->sk_destruct
= packet_sock_destruct
;
2634 sk_refcnt_debug_inc(sk
);
2637 * Attach a protocol block
2640 spin_lock_init(&po
->bind_lock
);
2641 mutex_init(&po
->pg_vec_lock
);
2642 po
->prot_hook
.func
= packet_rcv
;
2644 if (sock
->type
== SOCK_PACKET
)
2645 po
->prot_hook
.func
= packet_rcv_spkt
;
2647 po
->prot_hook
.af_packet_priv
= sk
;
2650 po
->prot_hook
.type
= proto
;
2651 register_prot_hook(sk
);
2654 mutex_lock(&net
->packet
.sklist_lock
);
2655 sk_add_node_rcu(sk
, &net
->packet
.sklist
);
2656 mutex_unlock(&net
->packet
.sklist_lock
);
2659 sock_prot_inuse_add(net
, &packet_proto
, 1);
2667 static int packet_recv_error(struct sock
*sk
, struct msghdr
*msg
, int len
)
2669 struct sock_exterr_skb
*serr
;
2670 struct sk_buff
*skb
, *skb2
;
2674 skb
= skb_dequeue(&sk
->sk_error_queue
);
2680 msg
->msg_flags
|= MSG_TRUNC
;
2683 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
2687 sock_recv_timestamp(msg
, sk
, skb
);
2689 serr
= SKB_EXT_ERR(skb
);
2690 put_cmsg(msg
, SOL_PACKET
, PACKET_TX_TIMESTAMP
,
2691 sizeof(serr
->ee
), &serr
->ee
);
2693 msg
->msg_flags
|= MSG_ERRQUEUE
;
2696 /* Reset and regenerate socket error */
2697 spin_lock_bh(&sk
->sk_error_queue
.lock
);
2699 if ((skb2
= skb_peek(&sk
->sk_error_queue
)) != NULL
) {
2700 sk
->sk_err
= SKB_EXT_ERR(skb2
)->ee
.ee_errno
;
2701 spin_unlock_bh(&sk
->sk_error_queue
.lock
);
2702 sk
->sk_error_report(sk
);
2704 spin_unlock_bh(&sk
->sk_error_queue
.lock
);
2713 * Pull a packet from our receive queue and hand it to the user.
2714 * If necessary we block.
2717 static int packet_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2718 struct msghdr
*msg
, size_t len
, int flags
)
2720 struct sock
*sk
= sock
->sk
;
2721 struct sk_buff
*skb
;
2723 int vnet_hdr_len
= 0;
2726 if (flags
& ~(MSG_PEEK
|MSG_DONTWAIT
|MSG_TRUNC
|MSG_CMSG_COMPAT
|MSG_ERRQUEUE
))
2730 /* What error should we return now? EUNATTACH? */
2731 if (pkt_sk(sk
)->ifindex
< 0)
2735 if (flags
& MSG_ERRQUEUE
) {
2736 err
= packet_recv_error(sk
, msg
, len
);
2741 * Call the generic datagram receiver. This handles all sorts
2742 * of horrible races and re-entrancy so we can forget about it
2743 * in the protocol layers.
2745 * Now it will return ENETDOWN, if device have just gone down,
2746 * but then it will block.
2749 skb
= skb_recv_datagram(sk
, flags
, flags
& MSG_DONTWAIT
, &err
);
2752 * An error occurred so return it. Because skb_recv_datagram()
2753 * handles the blocking we don't see and worry about blocking
2760 if (pkt_sk(sk
)->has_vnet_hdr
) {
2761 struct virtio_net_hdr vnet_hdr
= { 0 };
2764 vnet_hdr_len
= sizeof(vnet_hdr
);
2765 if (len
< vnet_hdr_len
)
2768 len
-= vnet_hdr_len
;
2770 if (skb_is_gso(skb
)) {
2771 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
2773 /* This is a hint as to how much should be linear. */
2774 vnet_hdr
.hdr_len
= skb_headlen(skb
);
2775 vnet_hdr
.gso_size
= sinfo
->gso_size
;
2776 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
2777 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
2778 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
2779 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
2780 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
2781 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
2782 else if (sinfo
->gso_type
& SKB_GSO_FCOE
)
2786 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
2787 vnet_hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
2789 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
2791 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2792 vnet_hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
2793 vnet_hdr
.csum_start
= skb_checksum_start_offset(skb
);
2794 vnet_hdr
.csum_offset
= skb
->csum_offset
;
2795 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
2796 vnet_hdr
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
2797 } /* else everything is zero */
2799 err
= memcpy_toiovec(msg
->msg_iov
, (void *)&vnet_hdr
,
2805 /* You lose any data beyond the buffer you gave. If it worries
2806 * a user program they can ask the device for its MTU
2812 msg
->msg_flags
|= MSG_TRUNC
;
2815 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
2819 sock_recv_ts_and_drops(msg
, sk
, skb
);
2821 if (msg
->msg_name
) {
2822 /* If the address length field is there to be filled
2823 * in, we fill it in now.
2825 if (sock
->type
== SOCK_PACKET
) {
2826 msg
->msg_namelen
= sizeof(struct sockaddr_pkt
);
2828 struct sockaddr_ll
*sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
2829 msg
->msg_namelen
= sll
->sll_halen
+
2830 offsetof(struct sockaddr_ll
, sll_addr
);
2832 memcpy(msg
->msg_name
, &PACKET_SKB_CB(skb
)->sa
,
2836 if (pkt_sk(sk
)->auxdata
) {
2837 struct tpacket_auxdata aux
;
2839 aux
.tp_status
= TP_STATUS_USER
;
2840 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2841 aux
.tp_status
|= TP_STATUS_CSUMNOTREADY
;
2842 aux
.tp_len
= PACKET_SKB_CB(skb
)->origlen
;
2843 aux
.tp_snaplen
= skb
->len
;
2845 aux
.tp_net
= skb_network_offset(skb
);
2846 if (vlan_tx_tag_present(skb
)) {
2847 aux
.tp_vlan_tci
= vlan_tx_tag_get(skb
);
2848 aux
.tp_status
|= TP_STATUS_VLAN_VALID
;
2850 aux
.tp_vlan_tci
= 0;
2853 put_cmsg(msg
, SOL_PACKET
, PACKET_AUXDATA
, sizeof(aux
), &aux
);
2857 * Free or return the buffer as appropriate. Again this
2858 * hides all the races and re-entrancy issues from us.
2860 err
= vnet_hdr_len
+ ((flags
&MSG_TRUNC
) ? skb
->len
: copied
);
2863 skb_free_datagram(sk
, skb
);
2868 static int packet_getname_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
2869 int *uaddr_len
, int peer
)
2871 struct net_device
*dev
;
2872 struct sock
*sk
= sock
->sk
;
2877 uaddr
->sa_family
= AF_PACKET
;
2878 memset(uaddr
->sa_data
, 0, sizeof(uaddr
->sa_data
));
2880 dev
= dev_get_by_index_rcu(sock_net(sk
), pkt_sk(sk
)->ifindex
);
2882 strlcpy(uaddr
->sa_data
, dev
->name
, sizeof(uaddr
->sa_data
));
2884 *uaddr_len
= sizeof(*uaddr
);
2889 static int packet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
2890 int *uaddr_len
, int peer
)
2892 struct net_device
*dev
;
2893 struct sock
*sk
= sock
->sk
;
2894 struct packet_sock
*po
= pkt_sk(sk
);
2895 DECLARE_SOCKADDR(struct sockaddr_ll
*, sll
, uaddr
);
2900 sll
->sll_family
= AF_PACKET
;
2901 sll
->sll_ifindex
= po
->ifindex
;
2902 sll
->sll_protocol
= po
->num
;
2903 sll
->sll_pkttype
= 0;
2905 dev
= dev_get_by_index_rcu(sock_net(sk
), po
->ifindex
);
2907 sll
->sll_hatype
= dev
->type
;
2908 sll
->sll_halen
= dev
->addr_len
;
2909 memcpy(sll
->sll_addr
, dev
->dev_addr
, dev
->addr_len
);
2911 sll
->sll_hatype
= 0; /* Bad: we have no ARPHRD_UNSPEC */
2915 *uaddr_len
= offsetof(struct sockaddr_ll
, sll_addr
) + sll
->sll_halen
;
2920 static int packet_dev_mc(struct net_device
*dev
, struct packet_mclist
*i
,
2924 case PACKET_MR_MULTICAST
:
2925 if (i
->alen
!= dev
->addr_len
)
2928 return dev_mc_add(dev
, i
->addr
);
2930 return dev_mc_del(dev
, i
->addr
);
2932 case PACKET_MR_PROMISC
:
2933 return dev_set_promiscuity(dev
, what
);
2935 case PACKET_MR_ALLMULTI
:
2936 return dev_set_allmulti(dev
, what
);
2938 case PACKET_MR_UNICAST
:
2939 if (i
->alen
!= dev
->addr_len
)
2942 return dev_uc_add(dev
, i
->addr
);
2944 return dev_uc_del(dev
, i
->addr
);
2952 static void packet_dev_mclist(struct net_device
*dev
, struct packet_mclist
*i
, int what
)
2954 for ( ; i
; i
= i
->next
) {
2955 if (i
->ifindex
== dev
->ifindex
)
2956 packet_dev_mc(dev
, i
, what
);
2960 static int packet_mc_add(struct sock
*sk
, struct packet_mreq_max
*mreq
)
2962 struct packet_sock
*po
= pkt_sk(sk
);
2963 struct packet_mclist
*ml
, *i
;
2964 struct net_device
*dev
;
2970 dev
= __dev_get_by_index(sock_net(sk
), mreq
->mr_ifindex
);
2975 if (mreq
->mr_alen
> dev
->addr_len
)
2979 i
= kmalloc(sizeof(*i
), GFP_KERNEL
);
2984 for (ml
= po
->mclist
; ml
; ml
= ml
->next
) {
2985 if (ml
->ifindex
== mreq
->mr_ifindex
&&
2986 ml
->type
== mreq
->mr_type
&&
2987 ml
->alen
== mreq
->mr_alen
&&
2988 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
2990 /* Free the new element ... */
2996 i
->type
= mreq
->mr_type
;
2997 i
->ifindex
= mreq
->mr_ifindex
;
2998 i
->alen
= mreq
->mr_alen
;
2999 memcpy(i
->addr
, mreq
->mr_address
, i
->alen
);
3001 i
->next
= po
->mclist
;
3003 err
= packet_dev_mc(dev
, i
, 1);
3005 po
->mclist
= i
->next
;
3014 static int packet_mc_drop(struct sock
*sk
, struct packet_mreq_max
*mreq
)
3016 struct packet_mclist
*ml
, **mlp
;
3020 for (mlp
= &pkt_sk(sk
)->mclist
; (ml
= *mlp
) != NULL
; mlp
= &ml
->next
) {
3021 if (ml
->ifindex
== mreq
->mr_ifindex
&&
3022 ml
->type
== mreq
->mr_type
&&
3023 ml
->alen
== mreq
->mr_alen
&&
3024 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
3025 if (--ml
->count
== 0) {
3026 struct net_device
*dev
;
3028 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
3030 packet_dev_mc(dev
, ml
, -1);
3038 return -EADDRNOTAVAIL
;
3041 static void packet_flush_mclist(struct sock
*sk
)
3043 struct packet_sock
*po
= pkt_sk(sk
);
3044 struct packet_mclist
*ml
;
3050 while ((ml
= po
->mclist
) != NULL
) {
3051 struct net_device
*dev
;
3053 po
->mclist
= ml
->next
;
3054 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
3056 packet_dev_mc(dev
, ml
, -1);
3063 packet_setsockopt(struct socket
*sock
, int level
, int optname
, char __user
*optval
, unsigned int optlen
)
3065 struct sock
*sk
= sock
->sk
;
3066 struct packet_sock
*po
= pkt_sk(sk
);
3069 if (level
!= SOL_PACKET
)
3070 return -ENOPROTOOPT
;
3073 case PACKET_ADD_MEMBERSHIP
:
3074 case PACKET_DROP_MEMBERSHIP
:
3076 struct packet_mreq_max mreq
;
3078 memset(&mreq
, 0, sizeof(mreq
));
3079 if (len
< sizeof(struct packet_mreq
))
3081 if (len
> sizeof(mreq
))
3083 if (copy_from_user(&mreq
, optval
, len
))
3085 if (len
< (mreq
.mr_alen
+ offsetof(struct packet_mreq
, mr_address
)))
3087 if (optname
== PACKET_ADD_MEMBERSHIP
)
3088 ret
= packet_mc_add(sk
, &mreq
);
3090 ret
= packet_mc_drop(sk
, &mreq
);
3094 case PACKET_RX_RING
:
3095 case PACKET_TX_RING
:
3097 union tpacket_req_u req_u
;
3100 switch (po
->tp_version
) {
3103 len
= sizeof(req_u
.req
);
3107 len
= sizeof(req_u
.req3
);
3112 if (pkt_sk(sk
)->has_vnet_hdr
)
3114 if (copy_from_user(&req_u
.req
, optval
, len
))
3116 return packet_set_ring(sk
, &req_u
, 0,
3117 optname
== PACKET_TX_RING
);
3119 case PACKET_COPY_THRESH
:
3123 if (optlen
!= sizeof(val
))
3125 if (copy_from_user(&val
, optval
, sizeof(val
)))
3128 pkt_sk(sk
)->copy_thresh
= val
;
3131 case PACKET_VERSION
:
3135 if (optlen
!= sizeof(val
))
3137 if (copy_from_user(&val
, optval
, sizeof(val
)))
3148 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
) {
3151 po
->tp_version
= val
;
3157 case PACKET_RESERVE
:
3161 if (optlen
!= sizeof(val
))
3163 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3165 if (copy_from_user(&val
, optval
, sizeof(val
)))
3169 po
->tp_reserve
= val
;
3176 if (optlen
!= sizeof(val
))
3178 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3180 if (copy_from_user(&val
, optval
, sizeof(val
)))
3182 po
->tp_loss
= !!val
;
3185 case PACKET_AUXDATA
:
3189 if (optlen
< sizeof(val
))
3191 if (copy_from_user(&val
, optval
, sizeof(val
)))
3194 po
->auxdata
= !!val
;
3197 case PACKET_ORIGDEV
:
3201 if (optlen
< sizeof(val
))
3203 if (copy_from_user(&val
, optval
, sizeof(val
)))
3206 po
->origdev
= !!val
;
3209 case PACKET_VNET_HDR
:
3213 if (sock
->type
!= SOCK_RAW
)
3215 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3217 if (optlen
< sizeof(val
))
3219 if (copy_from_user(&val
, optval
, sizeof(val
)))
3222 po
->has_vnet_hdr
= !!val
;
3225 case PACKET_TIMESTAMP
:
3229 if (optlen
!= sizeof(val
))
3231 if (copy_from_user(&val
, optval
, sizeof(val
)))
3234 po
->tp_tstamp
= val
;
3241 if (optlen
!= sizeof(val
))
3243 if (copy_from_user(&val
, optval
, sizeof(val
)))
3246 return fanout_add(sk
, val
& 0xffff, val
>> 16);
3248 case PACKET_TX_HAS_OFF
:
3252 if (optlen
!= sizeof(val
))
3254 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3256 if (copy_from_user(&val
, optval
, sizeof(val
)))
3258 po
->tp_tx_has_off
= !!val
;
3262 return -ENOPROTOOPT
;
3266 static int packet_getsockopt(struct socket
*sock
, int level
, int optname
,
3267 char __user
*optval
, int __user
*optlen
)
3270 int val
, lv
= sizeof(val
);
3271 struct sock
*sk
= sock
->sk
;
3272 struct packet_sock
*po
= pkt_sk(sk
);
3274 union tpacket_stats_u st
;
3276 if (level
!= SOL_PACKET
)
3277 return -ENOPROTOOPT
;
3279 if (get_user(len
, optlen
))
3286 case PACKET_STATISTICS
:
3287 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3288 memcpy(&st
, &po
->stats
, sizeof(st
));
3289 memset(&po
->stats
, 0, sizeof(po
->stats
));
3290 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3292 if (po
->tp_version
== TPACKET_V3
) {
3293 lv
= sizeof(struct tpacket_stats_v3
);
3294 st
.stats3
.tp_packets
+= st
.stats3
.tp_drops
;
3297 lv
= sizeof(struct tpacket_stats
);
3298 st
.stats1
.tp_packets
+= st
.stats1
.tp_drops
;
3303 case PACKET_AUXDATA
:
3306 case PACKET_ORIGDEV
:
3309 case PACKET_VNET_HDR
:
3310 val
= po
->has_vnet_hdr
;
3312 case PACKET_VERSION
:
3313 val
= po
->tp_version
;
3316 if (len
> sizeof(int))
3318 if (copy_from_user(&val
, optval
, len
))
3322 val
= sizeof(struct tpacket_hdr
);
3325 val
= sizeof(struct tpacket2_hdr
);
3328 val
= sizeof(struct tpacket3_hdr
);
3334 case PACKET_RESERVE
:
3335 val
= po
->tp_reserve
;
3340 case PACKET_TIMESTAMP
:
3341 val
= po
->tp_tstamp
;
3345 ((u32
)po
->fanout
->id
|
3346 ((u32
)po
->fanout
->type
<< 16) |
3347 ((u32
)po
->fanout
->flags
<< 24)) :
3350 case PACKET_TX_HAS_OFF
:
3351 val
= po
->tp_tx_has_off
;
3354 return -ENOPROTOOPT
;
3359 if (put_user(len
, optlen
))
3361 if (copy_to_user(optval
, data
, len
))
3367 static int packet_notifier(struct notifier_block
*this, unsigned long msg
, void *data
)
3370 struct net_device
*dev
= data
;
3371 struct net
*net
= dev_net(dev
);
3374 sk_for_each_rcu(sk
, &net
->packet
.sklist
) {
3375 struct packet_sock
*po
= pkt_sk(sk
);
3378 case NETDEV_UNREGISTER
:
3380 packet_dev_mclist(dev
, po
->mclist
, -1);
3384 if (dev
->ifindex
== po
->ifindex
) {
3385 spin_lock(&po
->bind_lock
);
3387 __unregister_prot_hook(sk
, false);
3388 sk
->sk_err
= ENETDOWN
;
3389 if (!sock_flag(sk
, SOCK_DEAD
))
3390 sk
->sk_error_report(sk
);
3392 if (msg
== NETDEV_UNREGISTER
) {
3393 packet_cached_dev_reset(po
);
3395 if (po
->prot_hook
.dev
)
3396 dev_put(po
->prot_hook
.dev
);
3397 po
->prot_hook
.dev
= NULL
;
3399 spin_unlock(&po
->bind_lock
);
3403 if (dev
->ifindex
== po
->ifindex
) {
3404 spin_lock(&po
->bind_lock
);
3406 register_prot_hook(sk
);
3407 spin_unlock(&po
->bind_lock
);
3417 static int packet_ioctl(struct socket
*sock
, unsigned int cmd
,
3420 struct sock
*sk
= sock
->sk
;
3425 int amount
= sk_wmem_alloc_get(sk
);
3427 return put_user(amount
, (int __user
*)arg
);
3431 struct sk_buff
*skb
;
3434 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3435 skb
= skb_peek(&sk
->sk_receive_queue
);
3438 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3439 return put_user(amount
, (int __user
*)arg
);
3442 return sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
3444 return sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
3454 case SIOCGIFBRDADDR
:
3455 case SIOCSIFBRDADDR
:
3456 case SIOCGIFNETMASK
:
3457 case SIOCSIFNETMASK
:
3458 case SIOCGIFDSTADDR
:
3459 case SIOCSIFDSTADDR
:
3461 return inet_dgram_ops
.ioctl(sock
, cmd
, arg
);
3465 return -ENOIOCTLCMD
;
3470 static unsigned int packet_poll(struct file
*file
, struct socket
*sock
,
3473 struct sock
*sk
= sock
->sk
;
3474 struct packet_sock
*po
= pkt_sk(sk
);
3475 unsigned int mask
= datagram_poll(file
, sock
, wait
);
3477 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3478 if (po
->rx_ring
.pg_vec
) {
3479 if (!packet_previous_rx_frame(po
, &po
->rx_ring
,
3481 mask
|= POLLIN
| POLLRDNORM
;
3483 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3484 spin_lock_bh(&sk
->sk_write_queue
.lock
);
3485 if (po
->tx_ring
.pg_vec
) {
3486 if (packet_current_frame(po
, &po
->tx_ring
, TP_STATUS_AVAILABLE
))
3487 mask
|= POLLOUT
| POLLWRNORM
;
3489 spin_unlock_bh(&sk
->sk_write_queue
.lock
);
3494 /* Dirty? Well, I still did not learn better way to account
3498 static void packet_mm_open(struct vm_area_struct
*vma
)
3500 struct file
*file
= vma
->vm_file
;
3501 struct socket
*sock
= file
->private_data
;
3502 struct sock
*sk
= sock
->sk
;
3505 atomic_inc(&pkt_sk(sk
)->mapped
);
3508 static void packet_mm_close(struct vm_area_struct
*vma
)
3510 struct file
*file
= vma
->vm_file
;
3511 struct socket
*sock
= file
->private_data
;
3512 struct sock
*sk
= sock
->sk
;
3515 atomic_dec(&pkt_sk(sk
)->mapped
);
3518 static const struct vm_operations_struct packet_mmap_ops
= {
3519 .open
= packet_mm_open
,
3520 .close
= packet_mm_close
,
3523 static void free_pg_vec(struct pgv
*pg_vec
, unsigned int order
,
3528 for (i
= 0; i
< len
; i
++) {
3529 if (likely(pg_vec
[i
].buffer
)) {
3530 if (is_vmalloc_addr(pg_vec
[i
].buffer
))
3531 vfree(pg_vec
[i
].buffer
);
3533 free_pages((unsigned long)pg_vec
[i
].buffer
,
3535 pg_vec
[i
].buffer
= NULL
;
3541 static char *alloc_one_pg_vec_page(unsigned long order
)
3543 char *buffer
= NULL
;
3544 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_COMP
|
3545 __GFP_ZERO
| __GFP_NOWARN
| __GFP_NORETRY
;
3547 buffer
= (char *) __get_free_pages(gfp_flags
, order
);
3553 * __get_free_pages failed, fall back to vmalloc
3555 buffer
= vzalloc((1 << order
) * PAGE_SIZE
);
3561 * vmalloc failed, lets dig into swap here
3563 gfp_flags
&= ~__GFP_NORETRY
;
3564 buffer
= (char *)__get_free_pages(gfp_flags
, order
);
3569 * complete and utter failure
3574 static struct pgv
*alloc_pg_vec(struct tpacket_req
*req
, int order
)
3576 unsigned int block_nr
= req
->tp_block_nr
;
3580 pg_vec
= kcalloc(block_nr
, sizeof(struct pgv
), GFP_KERNEL
);
3581 if (unlikely(!pg_vec
))
3584 for (i
= 0; i
< block_nr
; i
++) {
3585 pg_vec
[i
].buffer
= alloc_one_pg_vec_page(order
);
3586 if (unlikely(!pg_vec
[i
].buffer
))
3587 goto out_free_pgvec
;
3594 free_pg_vec(pg_vec
, order
, block_nr
);
3599 static int packet_set_ring(struct sock
*sk
, union tpacket_req_u
*req_u
,
3600 int closing
, int tx_ring
)
3602 struct pgv
*pg_vec
= NULL
;
3603 struct packet_sock
*po
= pkt_sk(sk
);
3604 int was_running
, order
= 0;
3605 struct packet_ring_buffer
*rb
;
3606 struct sk_buff_head
*rb_queue
;
3609 /* Added to avoid minimal code churn */
3610 struct tpacket_req
*req
= &req_u
->req
;
3613 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3614 if (!closing
&& tx_ring
&& (po
->tp_version
> TPACKET_V2
)) {
3615 WARN(1, "Tx-ring is not supported.\n");
3619 rb
= tx_ring
? &po
->tx_ring
: &po
->rx_ring
;
3620 rb_queue
= tx_ring
? &sk
->sk_write_queue
: &sk
->sk_receive_queue
;
3624 if (atomic_read(&po
->mapped
))
3626 if (atomic_read(&rb
->pending
))
3630 if (req
->tp_block_nr
) {
3631 /* Sanity tests and some calculations */
3633 if (unlikely(rb
->pg_vec
))
3636 switch (po
->tp_version
) {
3638 po
->tp_hdrlen
= TPACKET_HDRLEN
;
3641 po
->tp_hdrlen
= TPACKET2_HDRLEN
;
3644 po
->tp_hdrlen
= TPACKET3_HDRLEN
;
3649 if (unlikely((int)req
->tp_block_size
<= 0))
3651 if (unlikely(req
->tp_block_size
& (PAGE_SIZE
- 1)))
3653 if (po
->tp_version
>= TPACKET_V3
&&
3654 (int)(req
->tp_block_size
-
3655 BLK_PLUS_PRIV(req_u
->req3
.tp_sizeof_priv
)) <= 0)
3657 if (unlikely(req
->tp_frame_size
< po
->tp_hdrlen
+
3660 if (unlikely(req
->tp_frame_size
& (TPACKET_ALIGNMENT
- 1)))
3663 rb
->frames_per_block
= req
->tp_block_size
/req
->tp_frame_size
;
3664 if (unlikely(rb
->frames_per_block
<= 0))
3666 if (unlikely(req
->tp_block_size
> UINT_MAX
/ req
->tp_block_nr
))
3668 if (unlikely((rb
->frames_per_block
* req
->tp_block_nr
) !=
3673 order
= get_order(req
->tp_block_size
);
3674 pg_vec
= alloc_pg_vec(req
, order
);
3675 if (unlikely(!pg_vec
))
3677 switch (po
->tp_version
) {
3679 /* Transmit path is not supported. We checked
3680 * it above but just being paranoid
3683 init_prb_bdqc(po
, rb
, pg_vec
, req_u
, tx_ring
);
3692 if (unlikely(req
->tp_frame_nr
))
3697 /* Detach socket from network */
3698 spin_lock(&po
->bind_lock
);
3699 was_running
= po
->running
;
3703 __unregister_prot_hook(sk
, false);
3705 spin_unlock(&po
->bind_lock
);
3710 mutex_lock(&po
->pg_vec_lock
);
3711 if (closing
|| atomic_read(&po
->mapped
) == 0) {
3713 spin_lock_bh(&rb_queue
->lock
);
3714 swap(rb
->pg_vec
, pg_vec
);
3715 rb
->frame_max
= (req
->tp_frame_nr
- 1);
3717 rb
->frame_size
= req
->tp_frame_size
;
3718 spin_unlock_bh(&rb_queue
->lock
);
3720 swap(rb
->pg_vec_order
, order
);
3721 swap(rb
->pg_vec_len
, req
->tp_block_nr
);
3723 rb
->pg_vec_pages
= req
->tp_block_size
/PAGE_SIZE
;
3724 po
->prot_hook
.func
= (po
->rx_ring
.pg_vec
) ?
3725 tpacket_rcv
: packet_rcv
;
3726 skb_queue_purge(rb_queue
);
3727 if (atomic_read(&po
->mapped
))
3728 pr_err("packet_mmap: vma is busy: %d\n",
3729 atomic_read(&po
->mapped
));
3731 mutex_unlock(&po
->pg_vec_lock
);
3733 spin_lock(&po
->bind_lock
);
3736 register_prot_hook(sk
);
3738 spin_unlock(&po
->bind_lock
);
3739 if (closing
&& (po
->tp_version
> TPACKET_V2
)) {
3740 /* Because we don't support block-based V3 on tx-ring */
3742 prb_shutdown_retire_blk_timer(po
, tx_ring
, rb_queue
);
3746 free_pg_vec(pg_vec
, order
, req
->tp_block_nr
);
3752 static int packet_mmap(struct file
*file
, struct socket
*sock
,
3753 struct vm_area_struct
*vma
)
3755 struct sock
*sk
= sock
->sk
;
3756 struct packet_sock
*po
= pkt_sk(sk
);
3757 unsigned long size
, expected_size
;
3758 struct packet_ring_buffer
*rb
;
3759 unsigned long start
;
3766 mutex_lock(&po
->pg_vec_lock
);
3769 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
3771 expected_size
+= rb
->pg_vec_len
3777 if (expected_size
== 0)
3780 size
= vma
->vm_end
- vma
->vm_start
;
3781 if (size
!= expected_size
)
3784 start
= vma
->vm_start
;
3785 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
3786 if (rb
->pg_vec
== NULL
)
3789 for (i
= 0; i
< rb
->pg_vec_len
; i
++) {
3791 void *kaddr
= rb
->pg_vec
[i
].buffer
;
3794 for (pg_num
= 0; pg_num
< rb
->pg_vec_pages
; pg_num
++) {
3795 page
= pgv_to_page(kaddr
);
3796 err
= vm_insert_page(vma
, start
, page
);
3805 atomic_inc(&po
->mapped
);
3806 vma
->vm_ops
= &packet_mmap_ops
;
3810 mutex_unlock(&po
->pg_vec_lock
);
3814 static const struct proto_ops packet_ops_spkt
= {
3815 .family
= PF_PACKET
,
3816 .owner
= THIS_MODULE
,
3817 .release
= packet_release
,
3818 .bind
= packet_bind_spkt
,
3819 .connect
= sock_no_connect
,
3820 .socketpair
= sock_no_socketpair
,
3821 .accept
= sock_no_accept
,
3822 .getname
= packet_getname_spkt
,
3823 .poll
= datagram_poll
,
3824 .ioctl
= packet_ioctl
,
3825 .listen
= sock_no_listen
,
3826 .shutdown
= sock_no_shutdown
,
3827 .setsockopt
= sock_no_setsockopt
,
3828 .getsockopt
= sock_no_getsockopt
,
3829 .sendmsg
= packet_sendmsg_spkt
,
3830 .recvmsg
= packet_recvmsg
,
3831 .mmap
= sock_no_mmap
,
3832 .sendpage
= sock_no_sendpage
,
3835 static const struct proto_ops packet_ops
= {
3836 .family
= PF_PACKET
,
3837 .owner
= THIS_MODULE
,
3838 .release
= packet_release
,
3839 .bind
= packet_bind
,
3840 .connect
= sock_no_connect
,
3841 .socketpair
= sock_no_socketpair
,
3842 .accept
= sock_no_accept
,
3843 .getname
= packet_getname
,
3844 .poll
= packet_poll
,
3845 .ioctl
= packet_ioctl
,
3846 .listen
= sock_no_listen
,
3847 .shutdown
= sock_no_shutdown
,
3848 .setsockopt
= packet_setsockopt
,
3849 .getsockopt
= packet_getsockopt
,
3850 .sendmsg
= packet_sendmsg
,
3851 .recvmsg
= packet_recvmsg
,
3852 .mmap
= packet_mmap
,
3853 .sendpage
= sock_no_sendpage
,
3856 static const struct net_proto_family packet_family_ops
= {
3857 .family
= PF_PACKET
,
3858 .create
= packet_create
,
3859 .owner
= THIS_MODULE
,
3862 static struct notifier_block packet_netdev_notifier
= {
3863 .notifier_call
= packet_notifier
,
3866 #ifdef CONFIG_PROC_FS
3868 static void *packet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3871 struct net
*net
= seq_file_net(seq
);
3874 return seq_hlist_start_head_rcu(&net
->packet
.sklist
, *pos
);
3877 static void *packet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3879 struct net
*net
= seq_file_net(seq
);
3880 return seq_hlist_next_rcu(v
, &net
->packet
.sklist
, pos
);
3883 static void packet_seq_stop(struct seq_file
*seq
, void *v
)
3889 static int packet_seq_show(struct seq_file
*seq
, void *v
)
3891 if (v
== SEQ_START_TOKEN
)
3892 seq_puts(seq
, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3894 struct sock
*s
= sk_entry(v
);
3895 const struct packet_sock
*po
= pkt_sk(s
);
3898 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3900 atomic_read(&s
->sk_refcnt
),
3905 atomic_read(&s
->sk_rmem_alloc
),
3906 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(s
)),
3913 static const struct seq_operations packet_seq_ops
= {
3914 .start
= packet_seq_start
,
3915 .next
= packet_seq_next
,
3916 .stop
= packet_seq_stop
,
3917 .show
= packet_seq_show
,
3920 static int packet_seq_open(struct inode
*inode
, struct file
*file
)
3922 return seq_open_net(inode
, file
, &packet_seq_ops
,
3923 sizeof(struct seq_net_private
));
3926 static const struct file_operations packet_seq_fops
= {
3927 .owner
= THIS_MODULE
,
3928 .open
= packet_seq_open
,
3930 .llseek
= seq_lseek
,
3931 .release
= seq_release_net
,
3936 static int __net_init
packet_net_init(struct net
*net
)
3938 mutex_init(&net
->packet
.sklist_lock
);
3939 INIT_HLIST_HEAD(&net
->packet
.sklist
);
3941 if (!proc_create("packet", 0, net
->proc_net
, &packet_seq_fops
))
3947 static void __net_exit
packet_net_exit(struct net
*net
)
3949 remove_proc_entry("packet", net
->proc_net
);
3952 static struct pernet_operations packet_net_ops
= {
3953 .init
= packet_net_init
,
3954 .exit
= packet_net_exit
,
3958 static void __exit
packet_exit(void)
3960 unregister_netdevice_notifier(&packet_netdev_notifier
);
3961 unregister_pernet_subsys(&packet_net_ops
);
3962 sock_unregister(PF_PACKET
);
3963 proto_unregister(&packet_proto
);
3966 static int __init
packet_init(void)
3968 int rc
= proto_register(&packet_proto
, 0);
3973 sock_register(&packet_family_ops
);
3974 register_pernet_subsys(&packet_net_ops
);
3975 register_netdevice_notifier(&packet_netdev_notifier
);
3980 module_init(packet_init
);
3981 module_exit(packet_exit
);
3982 MODULE_LICENSE("GPL");
3983 MODULE_ALIAS_NETPROTO(PF_PACKET
);