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 if (f
->num_members
== 1)
1261 dev_add_pack(&f
->prot_hook
);
1262 spin_unlock(&f
->lock
);
1265 static void __fanout_unlink(struct sock
*sk
, struct packet_sock
*po
)
1267 struct packet_fanout
*f
= po
->fanout
;
1270 spin_lock(&f
->lock
);
1271 for (i
= 0; i
< f
->num_members
; i
++) {
1272 if (f
->arr
[i
] == sk
)
1275 BUG_ON(i
>= f
->num_members
);
1276 f
->arr
[i
] = f
->arr
[f
->num_members
- 1];
1278 if (f
->num_members
== 0)
1279 __dev_remove_pack(&f
->prot_hook
);
1280 spin_unlock(&f
->lock
);
1283 static bool match_fanout_group(struct packet_type
*ptype
, struct sock
* sk
)
1285 if (ptype
->af_packet_priv
== (void*)((struct packet_sock
*)sk
)->fanout
)
1291 static int fanout_add(struct sock
*sk
, u16 id
, u16 type_flags
)
1293 struct packet_sock
*po
= pkt_sk(sk
);
1294 struct packet_fanout
*f
, *match
;
1295 u8 type
= type_flags
& 0xff;
1296 u8 flags
= type_flags
>> 8;
1300 case PACKET_FANOUT_ROLLOVER
:
1301 if (type_flags
& PACKET_FANOUT_FLAG_ROLLOVER
)
1303 case PACKET_FANOUT_HASH
:
1304 case PACKET_FANOUT_LB
:
1305 case PACKET_FANOUT_CPU
:
1311 mutex_lock(&fanout_mutex
);
1322 list_for_each_entry(f
, &fanout_list
, list
) {
1324 read_pnet(&f
->net
) == sock_net(sk
)) {
1330 if (match
&& match
->flags
!= flags
)
1334 match
= kzalloc(sizeof(*match
), GFP_KERNEL
);
1337 write_pnet(&match
->net
, sock_net(sk
));
1340 match
->flags
= flags
;
1341 atomic_set(&match
->rr_cur
, 0);
1342 INIT_LIST_HEAD(&match
->list
);
1343 spin_lock_init(&match
->lock
);
1344 atomic_set(&match
->sk_ref
, 0);
1345 match
->prot_hook
.type
= po
->prot_hook
.type
;
1346 match
->prot_hook
.dev
= po
->prot_hook
.dev
;
1347 match
->prot_hook
.func
= packet_rcv_fanout
;
1348 match
->prot_hook
.af_packet_priv
= match
;
1349 match
->prot_hook
.id_match
= match_fanout_group
;
1350 list_add(&match
->list
, &fanout_list
);
1354 spin_lock(&po
->bind_lock
);
1356 match
->type
== type
&&
1357 match
->prot_hook
.type
== po
->prot_hook
.type
&&
1358 match
->prot_hook
.dev
== po
->prot_hook
.dev
) {
1360 if (atomic_read(&match
->sk_ref
) < PACKET_FANOUT_MAX
) {
1361 __dev_remove_pack(&po
->prot_hook
);
1363 atomic_inc(&match
->sk_ref
);
1364 __fanout_link(sk
, po
);
1368 spin_unlock(&po
->bind_lock
);
1370 if (err
&& !atomic_read(&match
->sk_ref
)) {
1371 list_del(&match
->list
);
1376 mutex_unlock(&fanout_mutex
);
1380 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1381 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1382 * It is the responsibility of the caller to call fanout_release_data() and
1383 * free the returned packet_fanout (after synchronize_net())
1385 static struct packet_fanout
*fanout_release(struct sock
*sk
)
1387 struct packet_sock
*po
= pkt_sk(sk
);
1388 struct packet_fanout
*f
;
1390 mutex_lock(&fanout_mutex
);
1395 if (atomic_dec_and_test(&f
->sk_ref
))
1400 mutex_unlock(&fanout_mutex
);
1405 static const struct proto_ops packet_ops
;
1407 static const struct proto_ops packet_ops_spkt
;
1409 static int packet_rcv_spkt(struct sk_buff
*skb
, struct net_device
*dev
,
1410 struct packet_type
*pt
, struct net_device
*orig_dev
)
1413 struct sockaddr_pkt
*spkt
;
1416 * When we registered the protocol we saved the socket in the data
1417 * field for just this event.
1420 sk
= pt
->af_packet_priv
;
1423 * Yank back the headers [hope the device set this
1424 * right or kerboom...]
1426 * Incoming packets have ll header pulled,
1429 * For outgoing ones skb->data == skb_mac_header(skb)
1430 * so that this procedure is noop.
1433 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1436 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1439 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1443 /* drop any routing info */
1446 /* drop conntrack reference */
1449 spkt
= &PACKET_SKB_CB(skb
)->sa
.pkt
;
1451 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1454 * The SOCK_PACKET socket receives _all_ frames.
1457 spkt
->spkt_family
= dev
->type
;
1458 strlcpy(spkt
->spkt_device
, dev
->name
, sizeof(spkt
->spkt_device
));
1459 spkt
->spkt_protocol
= skb
->protocol
;
1462 * Charge the memory to the socket. This is done specifically
1463 * to prevent sockets using all the memory up.
1466 if (sock_queue_rcv_skb(sk
, skb
) == 0)
1477 * Output a raw packet to a device layer. This bypasses all the other
1478 * protocol layers and you must therefore supply it with a complete frame
1481 static int packet_sendmsg_spkt(struct kiocb
*iocb
, struct socket
*sock
,
1482 struct msghdr
*msg
, size_t len
)
1484 struct sock
*sk
= sock
->sk
;
1485 struct sockaddr_pkt
*saddr
= (struct sockaddr_pkt
*)msg
->msg_name
;
1486 struct sk_buff
*skb
= NULL
;
1487 struct net_device
*dev
;
1493 * Get and verify the address.
1497 if (msg
->msg_namelen
< sizeof(struct sockaddr
))
1499 if (msg
->msg_namelen
== sizeof(struct sockaddr_pkt
))
1500 proto
= saddr
->spkt_protocol
;
1502 return -ENOTCONN
; /* SOCK_PACKET must be sent giving an address */
1505 * Find the device first to size check it
1508 saddr
->spkt_device
[sizeof(saddr
->spkt_device
) - 1] = 0;
1511 dev
= dev_get_by_name_rcu(sock_net(sk
), saddr
->spkt_device
);
1517 if (!(dev
->flags
& IFF_UP
))
1521 * You may not queue a frame bigger than the mtu. This is the lowest level
1522 * raw protocol and you must do your own fragmentation at this level.
1525 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
1526 if (!netif_supports_nofcs(dev
)) {
1527 err
= -EPROTONOSUPPORT
;
1530 extra_len
= 4; /* We're doing our own CRC */
1534 if (len
> dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
+ extra_len
)
1538 size_t reserved
= LL_RESERVED_SPACE(dev
);
1539 int tlen
= dev
->needed_tailroom
;
1540 unsigned int hhlen
= dev
->header_ops
? dev
->hard_header_len
: 0;
1543 skb
= sock_wmalloc(sk
, len
+ reserved
+ tlen
, 0, GFP_KERNEL
);
1546 /* FIXME: Save some space for broken drivers that write a hard
1547 * header at transmission time by themselves. PPP is the notable
1548 * one here. This should really be fixed at the driver level.
1550 skb_reserve(skb
, reserved
);
1551 skb_reset_network_header(skb
);
1553 /* Try to align data part correctly */
1558 skb_reset_network_header(skb
);
1560 err
= memcpy_fromiovec(skb_put(skb
, len
), msg
->msg_iov
, len
);
1566 if (len
> (dev
->mtu
+ dev
->hard_header_len
+ extra_len
)) {
1567 /* Earlier code assumed this would be a VLAN pkt,
1568 * double-check this now that we have the actual
1571 struct ethhdr
*ehdr
;
1572 skb_reset_mac_header(skb
);
1573 ehdr
= eth_hdr(skb
);
1574 if (ehdr
->h_proto
!= htons(ETH_P_8021Q
)) {
1580 skb
->protocol
= proto
;
1582 skb
->priority
= sk
->sk_priority
;
1583 skb
->mark
= sk
->sk_mark
;
1585 sock_tx_timestamp(sk
, &skb_shinfo(skb
)->tx_flags
);
1587 if (unlikely(extra_len
== 4))
1590 skb_probe_transport_header(skb
, 0);
1592 dev_queue_xmit(skb
);
1603 static unsigned int run_filter(const struct sk_buff
*skb
,
1604 const struct sock
*sk
,
1607 struct sk_filter
*filter
;
1610 filter
= rcu_dereference(sk
->sk_filter
);
1612 res
= SK_RUN_FILTER(filter
, skb
);
1619 * This function makes lazy skb cloning in hope that most of packets
1620 * are discarded by BPF.
1622 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1623 * and skb->cb are mangled. It works because (and until) packets
1624 * falling here are owned by current CPU. Output packets are cloned
1625 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1626 * sequencially, so that if we return skb to original state on exit,
1627 * we will not harm anyone.
1630 static int packet_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
1631 struct packet_type
*pt
, struct net_device
*orig_dev
)
1634 struct sockaddr_ll
*sll
;
1635 struct packet_sock
*po
;
1636 u8
*skb_head
= skb
->data
;
1637 int skb_len
= skb
->len
;
1638 unsigned int snaplen
, res
;
1640 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1643 sk
= pt
->af_packet_priv
;
1646 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1651 if (dev
->header_ops
) {
1652 /* The device has an explicit notion of ll header,
1653 * exported to higher levels.
1655 * Otherwise, the device hides details of its frame
1656 * structure, so that corresponding packet head is
1657 * never delivered to user.
1659 if (sk
->sk_type
!= SOCK_DGRAM
)
1660 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1661 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
1662 /* Special case: outgoing packets have ll header at head */
1663 skb_pull(skb
, skb_network_offset(skb
));
1669 res
= run_filter(skb
, sk
, snaplen
);
1671 goto drop_n_restore
;
1675 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
1678 if (skb_shared(skb
)) {
1679 struct sk_buff
*nskb
= skb_clone(skb
, GFP_ATOMIC
);
1683 if (skb_head
!= skb
->data
) {
1684 skb
->data
= skb_head
;
1691 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb
)) + MAX_ADDR_LEN
- 8 >
1694 sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
1695 sll
->sll_family
= AF_PACKET
;
1696 sll
->sll_hatype
= dev
->type
;
1697 sll
->sll_protocol
= skb
->protocol
;
1698 sll
->sll_pkttype
= skb
->pkt_type
;
1699 if (unlikely(po
->origdev
))
1700 sll
->sll_ifindex
= orig_dev
->ifindex
;
1702 sll
->sll_ifindex
= dev
->ifindex
;
1704 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
1706 PACKET_SKB_CB(skb
)->origlen
= skb
->len
;
1708 if (pskb_trim(skb
, snaplen
))
1711 skb_set_owner_r(skb
, sk
);
1715 /* drop conntrack reference */
1718 spin_lock(&sk
->sk_receive_queue
.lock
);
1719 po
->stats
.stats1
.tp_packets
++;
1720 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
1721 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1722 spin_unlock(&sk
->sk_receive_queue
.lock
);
1723 sk
->sk_data_ready(sk
, skb
->len
);
1727 spin_lock(&sk
->sk_receive_queue
.lock
);
1728 po
->stats
.stats1
.tp_drops
++;
1729 atomic_inc(&sk
->sk_drops
);
1730 spin_unlock(&sk
->sk_receive_queue
.lock
);
1733 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
1734 skb
->data
= skb_head
;
1742 static int tpacket_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
1743 struct packet_type
*pt
, struct net_device
*orig_dev
)
1746 struct packet_sock
*po
;
1747 struct sockaddr_ll
*sll
;
1748 union tpacket_uhdr h
;
1749 u8
*skb_head
= skb
->data
;
1750 int skb_len
= skb
->len
;
1751 unsigned int snaplen
, res
;
1752 unsigned long status
= TP_STATUS_USER
;
1753 unsigned short macoff
, netoff
, hdrlen
;
1754 struct sk_buff
*copy_skb
= NULL
;
1758 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1761 sk
= pt
->af_packet_priv
;
1764 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1767 if (dev
->header_ops
) {
1768 if (sk
->sk_type
!= SOCK_DGRAM
)
1769 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1770 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
1771 /* Special case: outgoing packets have ll header at head */
1772 skb_pull(skb
, skb_network_offset(skb
));
1776 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1777 status
|= TP_STATUS_CSUMNOTREADY
;
1781 res
= run_filter(skb
, sk
, snaplen
);
1783 goto drop_n_restore
;
1787 if (sk
->sk_type
== SOCK_DGRAM
) {
1788 macoff
= netoff
= TPACKET_ALIGN(po
->tp_hdrlen
) + 16 +
1791 unsigned int maclen
= skb_network_offset(skb
);
1792 netoff
= TPACKET_ALIGN(po
->tp_hdrlen
+
1793 (maclen
< 16 ? 16 : maclen
)) +
1795 macoff
= netoff
- maclen
;
1797 if (po
->tp_version
<= TPACKET_V2
) {
1798 if (macoff
+ snaplen
> po
->rx_ring
.frame_size
) {
1799 if (po
->copy_thresh
&&
1800 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1801 if (skb_shared(skb
)) {
1802 copy_skb
= skb_clone(skb
, GFP_ATOMIC
);
1804 copy_skb
= skb_get(skb
);
1805 skb_head
= skb
->data
;
1808 skb_set_owner_r(copy_skb
, sk
);
1810 snaplen
= po
->rx_ring
.frame_size
- macoff
;
1811 if ((int)snaplen
< 0)
1814 } else if (unlikely(macoff
+ snaplen
>
1815 GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
)) {
1818 nval
= GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
- macoff
;
1819 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1820 snaplen
, nval
, macoff
);
1822 if (unlikely((int)snaplen
< 0)) {
1824 macoff
= GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
;
1827 spin_lock(&sk
->sk_receive_queue
.lock
);
1828 h
.raw
= packet_current_rx_frame(po
, skb
,
1829 TP_STATUS_KERNEL
, (macoff
+snaplen
));
1832 if (po
->tp_version
<= TPACKET_V2
) {
1833 packet_increment_rx_head(po
, &po
->rx_ring
);
1835 * LOSING will be reported till you read the stats,
1836 * because it's COR - Clear On Read.
1837 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1840 if (po
->stats
.stats1
.tp_drops
)
1841 status
|= TP_STATUS_LOSING
;
1843 po
->stats
.stats1
.tp_packets
++;
1845 status
|= TP_STATUS_COPY
;
1846 __skb_queue_tail(&sk
->sk_receive_queue
, copy_skb
);
1848 spin_unlock(&sk
->sk_receive_queue
.lock
);
1850 skb_copy_bits(skb
, 0, h
.raw
+ macoff
, snaplen
);
1852 if (!(ts_status
= tpacket_get_timestamp(skb
, &ts
, po
->tp_tstamp
)))
1853 getnstimeofday(&ts
);
1855 status
|= ts_status
;
1857 switch (po
->tp_version
) {
1859 h
.h1
->tp_len
= skb
->len
;
1860 h
.h1
->tp_snaplen
= snaplen
;
1861 h
.h1
->tp_mac
= macoff
;
1862 h
.h1
->tp_net
= netoff
;
1863 h
.h1
->tp_sec
= ts
.tv_sec
;
1864 h
.h1
->tp_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
1865 hdrlen
= sizeof(*h
.h1
);
1868 h
.h2
->tp_len
= skb
->len
;
1869 h
.h2
->tp_snaplen
= snaplen
;
1870 h
.h2
->tp_mac
= macoff
;
1871 h
.h2
->tp_net
= netoff
;
1872 h
.h2
->tp_sec
= ts
.tv_sec
;
1873 h
.h2
->tp_nsec
= ts
.tv_nsec
;
1874 if (vlan_tx_tag_present(skb
)) {
1875 h
.h2
->tp_vlan_tci
= vlan_tx_tag_get(skb
);
1876 status
|= TP_STATUS_VLAN_VALID
;
1878 h
.h2
->tp_vlan_tci
= 0;
1880 h
.h2
->tp_padding
= 0;
1881 hdrlen
= sizeof(*h
.h2
);
1884 /* tp_nxt_offset,vlan are already populated above.
1885 * So DONT clear those fields here
1887 h
.h3
->tp_status
|= status
;
1888 h
.h3
->tp_len
= skb
->len
;
1889 h
.h3
->tp_snaplen
= snaplen
;
1890 h
.h3
->tp_mac
= macoff
;
1891 h
.h3
->tp_net
= netoff
;
1892 h
.h3
->tp_sec
= ts
.tv_sec
;
1893 h
.h3
->tp_nsec
= ts
.tv_nsec
;
1894 hdrlen
= sizeof(*h
.h3
);
1900 sll
= h
.raw
+ TPACKET_ALIGN(hdrlen
);
1901 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
1902 sll
->sll_family
= AF_PACKET
;
1903 sll
->sll_hatype
= dev
->type
;
1904 sll
->sll_protocol
= skb
->protocol
;
1905 sll
->sll_pkttype
= skb
->pkt_type
;
1906 if (unlikely(po
->origdev
))
1907 sll
->sll_ifindex
= orig_dev
->ifindex
;
1909 sll
->sll_ifindex
= dev
->ifindex
;
1912 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1916 if (po
->tp_version
<= TPACKET_V2
) {
1917 end
= (u8
*)PAGE_ALIGN((unsigned long)h
.raw
1918 + macoff
+ snaplen
);
1919 for (start
= h
.raw
; start
< end
; start
+= PAGE_SIZE
)
1920 flush_dcache_page(pgv_to_page(start
));
1925 if (po
->tp_version
<= TPACKET_V2
)
1926 __packet_set_status(po
, h
.raw
, status
);
1928 prb_clear_blk_fill_status(&po
->rx_ring
);
1930 sk
->sk_data_ready(sk
, 0);
1933 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
1934 skb
->data
= skb_head
;
1942 po
->stats
.stats1
.tp_drops
++;
1943 spin_unlock(&sk
->sk_receive_queue
.lock
);
1945 sk
->sk_data_ready(sk
, 0);
1946 kfree_skb(copy_skb
);
1947 goto drop_n_restore
;
1950 static void tpacket_destruct_skb(struct sk_buff
*skb
)
1952 struct packet_sock
*po
= pkt_sk(skb
->sk
);
1955 if (likely(po
->tx_ring
.pg_vec
)) {
1958 ph
= skb_shinfo(skb
)->destructor_arg
;
1959 BUG_ON(atomic_read(&po
->tx_ring
.pending
) == 0);
1960 atomic_dec(&po
->tx_ring
.pending
);
1962 ts
= __packet_set_timestamp(po
, ph
, skb
);
1963 __packet_set_status(po
, ph
, TP_STATUS_AVAILABLE
| ts
);
1969 static int tpacket_fill_skb(struct packet_sock
*po
, struct sk_buff
*skb
,
1970 void *frame
, struct net_device
*dev
, int size_max
,
1971 __be16 proto
, unsigned char *addr
, int hlen
)
1973 union tpacket_uhdr ph
;
1974 int to_write
, offset
, len
, tp_len
, nr_frags
, len_max
;
1975 struct socket
*sock
= po
->sk
.sk_socket
;
1982 skb
->protocol
= proto
;
1984 skb
->priority
= po
->sk
.sk_priority
;
1985 skb
->mark
= po
->sk
.sk_mark
;
1986 sock_tx_timestamp(&po
->sk
, &skb_shinfo(skb
)->tx_flags
);
1987 skb_shinfo(skb
)->destructor_arg
= ph
.raw
;
1989 switch (po
->tp_version
) {
1991 tp_len
= ph
.h2
->tp_len
;
1994 tp_len
= ph
.h1
->tp_len
;
1997 if (unlikely(tp_len
> size_max
)) {
1998 pr_err("packet size is too long (%d > %d)\n", tp_len
, size_max
);
2002 skb_reserve(skb
, hlen
);
2003 skb_reset_network_header(skb
);
2004 skb_probe_transport_header(skb
, 0);
2006 if (po
->tp_tx_has_off
) {
2007 int off_min
, off_max
, off
;
2008 off_min
= po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
2009 off_max
= po
->tx_ring
.frame_size
- tp_len
;
2010 if (sock
->type
== SOCK_DGRAM
) {
2011 switch (po
->tp_version
) {
2013 off
= ph
.h2
->tp_net
;
2016 off
= ph
.h1
->tp_net
;
2020 switch (po
->tp_version
) {
2022 off
= ph
.h2
->tp_mac
;
2025 off
= ph
.h1
->tp_mac
;
2029 if (unlikely((off
< off_min
) || (off_max
< off
)))
2031 data
= ph
.raw
+ off
;
2033 data
= ph
.raw
+ po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
2037 if (sock
->type
== SOCK_DGRAM
) {
2038 err
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
,
2040 if (unlikely(err
< 0))
2042 } else if (dev
->hard_header_len
) {
2043 /* net device doesn't like empty head */
2044 if (unlikely(tp_len
<= dev
->hard_header_len
)) {
2045 pr_err("packet size is too short (%d < %d)\n",
2046 tp_len
, dev
->hard_header_len
);
2050 skb_push(skb
, dev
->hard_header_len
);
2051 err
= skb_store_bits(skb
, 0, data
,
2052 dev
->hard_header_len
);
2056 data
+= dev
->hard_header_len
;
2057 to_write
-= dev
->hard_header_len
;
2060 offset
= offset_in_page(data
);
2061 len_max
= PAGE_SIZE
- offset
;
2062 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2064 skb
->data_len
= to_write
;
2065 skb
->len
+= to_write
;
2066 skb
->truesize
+= to_write
;
2067 atomic_add(to_write
, &po
->sk
.sk_wmem_alloc
);
2069 while (likely(to_write
)) {
2070 nr_frags
= skb_shinfo(skb
)->nr_frags
;
2072 if (unlikely(nr_frags
>= MAX_SKB_FRAGS
)) {
2073 pr_err("Packet exceed the number of skb frags(%lu)\n",
2078 page
= pgv_to_page(data
);
2080 flush_dcache_page(page
);
2082 skb_fill_page_desc(skb
, nr_frags
, page
, offset
, len
);
2085 len_max
= PAGE_SIZE
;
2086 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2092 static int tpacket_snd(struct packet_sock
*po
, struct msghdr
*msg
)
2094 struct sk_buff
*skb
;
2095 struct net_device
*dev
;
2097 int err
, reserve
= 0;
2099 struct sockaddr_ll
*saddr
= (struct sockaddr_ll
*)msg
->msg_name
;
2100 int tp_len
, size_max
;
2101 unsigned char *addr
;
2103 int status
= TP_STATUS_AVAILABLE
;
2106 mutex_lock(&po
->pg_vec_lock
);
2108 if (likely(saddr
== NULL
)) {
2109 dev
= packet_cached_dev_get(po
);
2114 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2116 if (msg
->msg_namelen
< (saddr
->sll_halen
2117 + offsetof(struct sockaddr_ll
,
2120 proto
= saddr
->sll_protocol
;
2121 addr
= saddr
->sll_addr
;
2122 dev
= dev_get_by_index(sock_net(&po
->sk
), saddr
->sll_ifindex
);
2126 if (unlikely(dev
== NULL
))
2129 if (unlikely(!(dev
->flags
& IFF_UP
)))
2132 reserve
= dev
->hard_header_len
;
2134 size_max
= po
->tx_ring
.frame_size
2135 - (po
->tp_hdrlen
- sizeof(struct sockaddr_ll
));
2137 if (size_max
> dev
->mtu
+ reserve
)
2138 size_max
= dev
->mtu
+ reserve
;
2141 ph
= packet_current_frame(po
, &po
->tx_ring
,
2142 TP_STATUS_SEND_REQUEST
);
2144 if (unlikely(ph
== NULL
)) {
2149 status
= TP_STATUS_SEND_REQUEST
;
2150 hlen
= LL_RESERVED_SPACE(dev
);
2151 tlen
= dev
->needed_tailroom
;
2152 skb
= sock_alloc_send_skb(&po
->sk
,
2153 hlen
+ tlen
+ sizeof(struct sockaddr_ll
),
2156 if (unlikely(skb
== NULL
))
2159 tp_len
= tpacket_fill_skb(po
, skb
, ph
, dev
, size_max
, proto
,
2162 if (unlikely(tp_len
< 0)) {
2164 __packet_set_status(po
, ph
,
2165 TP_STATUS_AVAILABLE
);
2166 packet_increment_head(&po
->tx_ring
);
2170 status
= TP_STATUS_WRONG_FORMAT
;
2176 skb
->destructor
= tpacket_destruct_skb
;
2177 __packet_set_status(po
, ph
, TP_STATUS_SENDING
);
2178 atomic_inc(&po
->tx_ring
.pending
);
2180 status
= TP_STATUS_SEND_REQUEST
;
2181 err
= dev_queue_xmit(skb
);
2182 if (unlikely(err
> 0)) {
2183 err
= net_xmit_errno(err
);
2184 if (err
&& __packet_get_status(po
, ph
) ==
2185 TP_STATUS_AVAILABLE
) {
2186 /* skb was destructed already */
2191 * skb was dropped but not destructed yet;
2192 * let's treat it like congestion or err < 0
2196 packet_increment_head(&po
->tx_ring
);
2198 } while (likely((ph
!= NULL
) ||
2199 ((!(msg
->msg_flags
& MSG_DONTWAIT
)) &&
2200 (atomic_read(&po
->tx_ring
.pending
))))
2207 __packet_set_status(po
, ph
, status
);
2212 mutex_unlock(&po
->pg_vec_lock
);
2216 static struct sk_buff
*packet_alloc_skb(struct sock
*sk
, size_t prepad
,
2217 size_t reserve
, size_t len
,
2218 size_t linear
, int noblock
,
2221 struct sk_buff
*skb
;
2223 /* Under a page? Don't bother with paged skb. */
2224 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
2227 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
2232 skb_reserve(skb
, reserve
);
2233 skb_put(skb
, linear
);
2234 skb
->data_len
= len
- linear
;
2235 skb
->len
+= len
- linear
;
2240 static int packet_snd(struct socket
*sock
,
2241 struct msghdr
*msg
, size_t len
)
2243 struct sock
*sk
= sock
->sk
;
2244 struct sockaddr_ll
*saddr
= (struct sockaddr_ll
*)msg
->msg_name
;
2245 struct sk_buff
*skb
;
2246 struct net_device
*dev
;
2248 unsigned char *addr
;
2249 int err
, reserve
= 0;
2250 struct virtio_net_hdr vnet_hdr
= { 0 };
2253 struct packet_sock
*po
= pkt_sk(sk
);
2254 unsigned short gso_type
= 0;
2255 int hlen
, tlen
, linear
;
2259 * Get and verify the address.
2262 if (likely(saddr
== NULL
)) {
2263 dev
= packet_cached_dev_get(po
);
2268 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2270 if (msg
->msg_namelen
< (saddr
->sll_halen
+ offsetof(struct sockaddr_ll
, sll_addr
)))
2272 proto
= saddr
->sll_protocol
;
2273 addr
= saddr
->sll_addr
;
2274 dev
= dev_get_by_index(sock_net(sk
), saddr
->sll_ifindex
);
2278 if (unlikely(dev
== NULL
))
2281 if (unlikely(!(dev
->flags
& IFF_UP
)))
2284 if (sock
->type
== SOCK_RAW
)
2285 reserve
= dev
->hard_header_len
;
2286 if (po
->has_vnet_hdr
) {
2287 vnet_hdr_len
= sizeof(vnet_hdr
);
2290 if (len
< vnet_hdr_len
)
2293 len
-= vnet_hdr_len
;
2295 err
= memcpy_fromiovec((void *)&vnet_hdr
, msg
->msg_iov
,
2300 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
2301 (vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
2303 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
2304 vnet_hdr
.csum_offset
+ 2;
2307 if (vnet_hdr
.hdr_len
> len
)
2310 if (vnet_hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
2311 switch (vnet_hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
2312 case VIRTIO_NET_HDR_GSO_TCPV4
:
2313 gso_type
= SKB_GSO_TCPV4
;
2315 case VIRTIO_NET_HDR_GSO_TCPV6
:
2316 gso_type
= SKB_GSO_TCPV6
;
2318 case VIRTIO_NET_HDR_GSO_UDP
:
2319 gso_type
= SKB_GSO_UDP
;
2325 if (vnet_hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
2326 gso_type
|= SKB_GSO_TCP_ECN
;
2328 if (vnet_hdr
.gso_size
== 0)
2334 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
2335 if (!netif_supports_nofcs(dev
)) {
2336 err
= -EPROTONOSUPPORT
;
2339 extra_len
= 4; /* We're doing our own CRC */
2343 if (!gso_type
&& (len
> dev
->mtu
+ reserve
+ VLAN_HLEN
+ extra_len
))
2347 hlen
= LL_RESERVED_SPACE(dev
);
2348 tlen
= dev
->needed_tailroom
;
2349 linear
= vnet_hdr
.hdr_len
;
2350 linear
= max(linear
, min_t(int, len
, dev
->hard_header_len
));
2351 skb
= packet_alloc_skb(sk
, hlen
+ tlen
, hlen
, len
, linear
,
2352 msg
->msg_flags
& MSG_DONTWAIT
, &err
);
2356 skb_set_network_header(skb
, reserve
);
2359 if (sock
->type
== SOCK_DGRAM
&&
2360 (offset
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
, NULL
, len
)) < 0)
2363 /* Returns -EFAULT on error */
2364 err
= skb_copy_datagram_from_iovec(skb
, offset
, msg
->msg_iov
, 0, len
);
2368 sock_tx_timestamp(sk
, &skb_shinfo(skb
)->tx_flags
);
2370 if (!gso_type
&& (len
> dev
->mtu
+ reserve
+ extra_len
)) {
2371 /* Earlier code assumed this would be a VLAN pkt,
2372 * double-check this now that we have the actual
2375 struct ethhdr
*ehdr
;
2376 skb_reset_mac_header(skb
);
2377 ehdr
= eth_hdr(skb
);
2378 if (ehdr
->h_proto
!= htons(ETH_P_8021Q
)) {
2384 skb
->protocol
= proto
;
2386 skb
->priority
= sk
->sk_priority
;
2387 skb
->mark
= sk
->sk_mark
;
2389 if (po
->has_vnet_hdr
) {
2390 if (vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
2391 if (!skb_partial_csum_set(skb
, vnet_hdr
.csum_start
,
2392 vnet_hdr
.csum_offset
)) {
2398 skb_shinfo(skb
)->gso_size
= vnet_hdr
.gso_size
;
2399 skb_shinfo(skb
)->gso_type
= gso_type
;
2401 /* Header must be checked, and gso_segs computed. */
2402 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
2403 skb_shinfo(skb
)->gso_segs
= 0;
2405 len
+= vnet_hdr_len
;
2408 skb_probe_transport_header(skb
, reserve
);
2410 if (unlikely(extra_len
== 4))
2417 err
= dev_queue_xmit(skb
);
2418 if (err
> 0 && (err
= net_xmit_errno(err
)) != 0)
2434 static int packet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
2435 struct msghdr
*msg
, size_t len
)
2437 struct sock
*sk
= sock
->sk
;
2438 struct packet_sock
*po
= pkt_sk(sk
);
2439 if (po
->tx_ring
.pg_vec
)
2440 return tpacket_snd(po
, msg
);
2442 return packet_snd(sock
, msg
, len
);
2446 * Close a PACKET socket. This is fairly simple. We immediately go
2447 * to 'closed' state and remove our protocol entry in the device list.
2450 static int packet_release(struct socket
*sock
)
2452 struct sock
*sk
= sock
->sk
;
2453 struct packet_sock
*po
;
2454 struct packet_fanout
*f
;
2456 union tpacket_req_u req_u
;
2464 mutex_lock(&net
->packet
.sklist_lock
);
2465 sk_del_node_init_rcu(sk
);
2466 mutex_unlock(&net
->packet
.sklist_lock
);
2469 sock_prot_inuse_add(net
, sk
->sk_prot
, -1);
2472 spin_lock(&po
->bind_lock
);
2473 unregister_prot_hook(sk
, false);
2474 packet_cached_dev_reset(po
);
2476 if (po
->prot_hook
.dev
) {
2477 dev_put(po
->prot_hook
.dev
);
2478 po
->prot_hook
.dev
= NULL
;
2480 spin_unlock(&po
->bind_lock
);
2482 packet_flush_mclist(sk
);
2484 if (po
->rx_ring
.pg_vec
) {
2485 memset(&req_u
, 0, sizeof(req_u
));
2486 packet_set_ring(sk
, &req_u
, 1, 0);
2489 if (po
->tx_ring
.pg_vec
) {
2490 memset(&req_u
, 0, sizeof(req_u
));
2491 packet_set_ring(sk
, &req_u
, 1, 1);
2494 f
= fanout_release(sk
);
2502 * Now the socket is dead. No more input will appear.
2509 skb_queue_purge(&sk
->sk_receive_queue
);
2510 sk_refcnt_debug_release(sk
);
2517 * Attach a packet hook.
2520 static int packet_do_bind(struct sock
*sk
, struct net_device
*dev
, __be16 protocol
)
2522 struct packet_sock
*po
= pkt_sk(sk
);
2527 spin_lock(&po
->bind_lock
);
2537 unregister_prot_hook(sk
, true);
2540 po
->prot_hook
.type
= protocol
;
2541 if (po
->prot_hook
.dev
)
2542 dev_put(po
->prot_hook
.dev
);
2544 po
->prot_hook
.dev
= dev
;
2545 po
->ifindex
= dev
? dev
->ifindex
: 0;
2547 packet_cached_dev_assign(po
, dev
);
2552 if (!dev
|| (dev
->flags
& IFF_UP
)) {
2553 register_prot_hook(sk
);
2555 sk
->sk_err
= ENETDOWN
;
2556 if (!sock_flag(sk
, SOCK_DEAD
))
2557 sk
->sk_error_report(sk
);
2561 spin_unlock(&po
->bind_lock
);
2567 * Bind a packet socket to a device
2570 static int packet_bind_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
2573 struct sock
*sk
= sock
->sk
;
2574 char name
[sizeof(uaddr
->sa_data
) + 1];
2575 struct net_device
*dev
;
2582 if (addr_len
!= sizeof(struct sockaddr
))
2584 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
2587 memcpy(name
, uaddr
->sa_data
, sizeof(uaddr
->sa_data
));
2588 name
[sizeof(uaddr
->sa_data
)] = 0;
2590 dev
= dev_get_by_name(sock_net(sk
), name
);
2592 err
= packet_do_bind(sk
, dev
, pkt_sk(sk
)->num
);
2596 static int packet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
2598 struct sockaddr_ll
*sll
= (struct sockaddr_ll
*)uaddr
;
2599 struct sock
*sk
= sock
->sk
;
2600 struct net_device
*dev
= NULL
;
2608 if (addr_len
< sizeof(struct sockaddr_ll
))
2610 if (sll
->sll_family
!= AF_PACKET
)
2613 if (sll
->sll_ifindex
) {
2615 dev
= dev_get_by_index(sock_net(sk
), sll
->sll_ifindex
);
2619 err
= packet_do_bind(sk
, dev
, sll
->sll_protocol
? : pkt_sk(sk
)->num
);
2625 static struct proto packet_proto
= {
2627 .owner
= THIS_MODULE
,
2628 .obj_size
= sizeof(struct packet_sock
),
2632 * Create a packet of type SOCK_PACKET.
2635 static int packet_create(struct net
*net
, struct socket
*sock
, int protocol
,
2639 struct packet_sock
*po
;
2640 __be16 proto
= (__force __be16
)protocol
; /* weird, but documented */
2643 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
2645 if (sock
->type
!= SOCK_DGRAM
&& sock
->type
!= SOCK_RAW
&&
2646 sock
->type
!= SOCK_PACKET
)
2647 return -ESOCKTNOSUPPORT
;
2649 sock
->state
= SS_UNCONNECTED
;
2652 sk
= sk_alloc(net
, PF_PACKET
, GFP_KERNEL
, &packet_proto
);
2656 sock
->ops
= &packet_ops
;
2657 if (sock
->type
== SOCK_PACKET
)
2658 sock
->ops
= &packet_ops_spkt
;
2660 sock_init_data(sock
, sk
);
2663 sk
->sk_family
= PF_PACKET
;
2666 packet_cached_dev_reset(po
);
2668 sk
->sk_destruct
= packet_sock_destruct
;
2669 sk_refcnt_debug_inc(sk
);
2672 * Attach a protocol block
2675 spin_lock_init(&po
->bind_lock
);
2676 mutex_init(&po
->pg_vec_lock
);
2677 po
->prot_hook
.func
= packet_rcv
;
2679 if (sock
->type
== SOCK_PACKET
)
2680 po
->prot_hook
.func
= packet_rcv_spkt
;
2682 po
->prot_hook
.af_packet_priv
= sk
;
2685 po
->prot_hook
.type
= proto
;
2686 register_prot_hook(sk
);
2689 mutex_lock(&net
->packet
.sklist_lock
);
2690 sk_add_node_rcu(sk
, &net
->packet
.sklist
);
2691 mutex_unlock(&net
->packet
.sklist_lock
);
2694 sock_prot_inuse_add(net
, &packet_proto
, 1);
2702 static int packet_recv_error(struct sock
*sk
, struct msghdr
*msg
, int len
)
2704 struct sock_exterr_skb
*serr
;
2705 struct sk_buff
*skb
, *skb2
;
2709 skb
= skb_dequeue(&sk
->sk_error_queue
);
2715 msg
->msg_flags
|= MSG_TRUNC
;
2718 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
2722 sock_recv_timestamp(msg
, sk
, skb
);
2724 serr
= SKB_EXT_ERR(skb
);
2725 put_cmsg(msg
, SOL_PACKET
, PACKET_TX_TIMESTAMP
,
2726 sizeof(serr
->ee
), &serr
->ee
);
2728 msg
->msg_flags
|= MSG_ERRQUEUE
;
2731 /* Reset and regenerate socket error */
2732 spin_lock_bh(&sk
->sk_error_queue
.lock
);
2734 if ((skb2
= skb_peek(&sk
->sk_error_queue
)) != NULL
) {
2735 sk
->sk_err
= SKB_EXT_ERR(skb2
)->ee
.ee_errno
;
2736 spin_unlock_bh(&sk
->sk_error_queue
.lock
);
2737 sk
->sk_error_report(sk
);
2739 spin_unlock_bh(&sk
->sk_error_queue
.lock
);
2748 * Pull a packet from our receive queue and hand it to the user.
2749 * If necessary we block.
2752 static int packet_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2753 struct msghdr
*msg
, size_t len
, int flags
)
2755 struct sock
*sk
= sock
->sk
;
2756 struct sk_buff
*skb
;
2758 int vnet_hdr_len
= 0;
2761 if (flags
& ~(MSG_PEEK
|MSG_DONTWAIT
|MSG_TRUNC
|MSG_CMSG_COMPAT
|MSG_ERRQUEUE
))
2765 /* What error should we return now? EUNATTACH? */
2766 if (pkt_sk(sk
)->ifindex
< 0)
2770 if (flags
& MSG_ERRQUEUE
) {
2771 err
= packet_recv_error(sk
, msg
, len
);
2776 * Call the generic datagram receiver. This handles all sorts
2777 * of horrible races and re-entrancy so we can forget about it
2778 * in the protocol layers.
2780 * Now it will return ENETDOWN, if device have just gone down,
2781 * but then it will block.
2784 skb
= skb_recv_datagram(sk
, flags
, flags
& MSG_DONTWAIT
, &err
);
2787 * An error occurred so return it. Because skb_recv_datagram()
2788 * handles the blocking we don't see and worry about blocking
2795 if (pkt_sk(sk
)->has_vnet_hdr
) {
2796 struct virtio_net_hdr vnet_hdr
= { 0 };
2799 vnet_hdr_len
= sizeof(vnet_hdr
);
2800 if (len
< vnet_hdr_len
)
2803 len
-= vnet_hdr_len
;
2805 if (skb_is_gso(skb
)) {
2806 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
2808 /* This is a hint as to how much should be linear. */
2809 vnet_hdr
.hdr_len
= skb_headlen(skb
);
2810 vnet_hdr
.gso_size
= sinfo
->gso_size
;
2811 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
2812 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
2813 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
2814 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
2815 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
2816 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
2817 else if (sinfo
->gso_type
& SKB_GSO_FCOE
)
2821 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
2822 vnet_hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
2824 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
2826 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2827 vnet_hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
2828 vnet_hdr
.csum_start
= skb_checksum_start_offset(skb
);
2829 vnet_hdr
.csum_offset
= skb
->csum_offset
;
2830 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
2831 vnet_hdr
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
2832 } /* else everything is zero */
2834 err
= memcpy_toiovec(msg
->msg_iov
, (void *)&vnet_hdr
,
2840 /* You lose any data beyond the buffer you gave. If it worries
2841 * a user program they can ask the device for its MTU
2847 msg
->msg_flags
|= MSG_TRUNC
;
2850 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
2854 sock_recv_ts_and_drops(msg
, sk
, skb
);
2856 if (msg
->msg_name
) {
2857 /* If the address length field is there to be filled
2858 * in, we fill it in now.
2860 if (sock
->type
== SOCK_PACKET
) {
2861 msg
->msg_namelen
= sizeof(struct sockaddr_pkt
);
2863 struct sockaddr_ll
*sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
2864 msg
->msg_namelen
= sll
->sll_halen
+
2865 offsetof(struct sockaddr_ll
, sll_addr
);
2867 memcpy(msg
->msg_name
, &PACKET_SKB_CB(skb
)->sa
,
2871 if (pkt_sk(sk
)->auxdata
) {
2872 struct tpacket_auxdata aux
;
2874 aux
.tp_status
= TP_STATUS_USER
;
2875 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2876 aux
.tp_status
|= TP_STATUS_CSUMNOTREADY
;
2877 aux
.tp_len
= PACKET_SKB_CB(skb
)->origlen
;
2878 aux
.tp_snaplen
= skb
->len
;
2880 aux
.tp_net
= skb_network_offset(skb
);
2881 if (vlan_tx_tag_present(skb
)) {
2882 aux
.tp_vlan_tci
= vlan_tx_tag_get(skb
);
2883 aux
.tp_status
|= TP_STATUS_VLAN_VALID
;
2885 aux
.tp_vlan_tci
= 0;
2888 put_cmsg(msg
, SOL_PACKET
, PACKET_AUXDATA
, sizeof(aux
), &aux
);
2892 * Free or return the buffer as appropriate. Again this
2893 * hides all the races and re-entrancy issues from us.
2895 err
= vnet_hdr_len
+ ((flags
&MSG_TRUNC
) ? skb
->len
: copied
);
2898 skb_free_datagram(sk
, skb
);
2903 static int packet_getname_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
2904 int *uaddr_len
, int peer
)
2906 struct net_device
*dev
;
2907 struct sock
*sk
= sock
->sk
;
2912 uaddr
->sa_family
= AF_PACKET
;
2913 memset(uaddr
->sa_data
, 0, sizeof(uaddr
->sa_data
));
2915 dev
= dev_get_by_index_rcu(sock_net(sk
), pkt_sk(sk
)->ifindex
);
2917 strlcpy(uaddr
->sa_data
, dev
->name
, sizeof(uaddr
->sa_data
));
2919 *uaddr_len
= sizeof(*uaddr
);
2924 static int packet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
2925 int *uaddr_len
, int peer
)
2927 struct net_device
*dev
;
2928 struct sock
*sk
= sock
->sk
;
2929 struct packet_sock
*po
= pkt_sk(sk
);
2930 DECLARE_SOCKADDR(struct sockaddr_ll
*, sll
, uaddr
);
2935 sll
->sll_family
= AF_PACKET
;
2936 sll
->sll_ifindex
= po
->ifindex
;
2937 sll
->sll_protocol
= po
->num
;
2938 sll
->sll_pkttype
= 0;
2940 dev
= dev_get_by_index_rcu(sock_net(sk
), po
->ifindex
);
2942 sll
->sll_hatype
= dev
->type
;
2943 sll
->sll_halen
= dev
->addr_len
;
2944 memcpy(sll
->sll_addr
, dev
->dev_addr
, dev
->addr_len
);
2946 sll
->sll_hatype
= 0; /* Bad: we have no ARPHRD_UNSPEC */
2950 *uaddr_len
= offsetof(struct sockaddr_ll
, sll_addr
) + sll
->sll_halen
;
2955 static int packet_dev_mc(struct net_device
*dev
, struct packet_mclist
*i
,
2959 case PACKET_MR_MULTICAST
:
2960 if (i
->alen
!= dev
->addr_len
)
2963 return dev_mc_add(dev
, i
->addr
);
2965 return dev_mc_del(dev
, i
->addr
);
2967 case PACKET_MR_PROMISC
:
2968 return dev_set_promiscuity(dev
, what
);
2970 case PACKET_MR_ALLMULTI
:
2971 return dev_set_allmulti(dev
, what
);
2973 case PACKET_MR_UNICAST
:
2974 if (i
->alen
!= dev
->addr_len
)
2977 return dev_uc_add(dev
, i
->addr
);
2979 return dev_uc_del(dev
, i
->addr
);
2987 static void packet_dev_mclist(struct net_device
*dev
, struct packet_mclist
*i
, int what
)
2989 for ( ; i
; i
= i
->next
) {
2990 if (i
->ifindex
== dev
->ifindex
)
2991 packet_dev_mc(dev
, i
, what
);
2995 static int packet_mc_add(struct sock
*sk
, struct packet_mreq_max
*mreq
)
2997 struct packet_sock
*po
= pkt_sk(sk
);
2998 struct packet_mclist
*ml
, *i
;
2999 struct net_device
*dev
;
3005 dev
= __dev_get_by_index(sock_net(sk
), mreq
->mr_ifindex
);
3010 if (mreq
->mr_alen
> dev
->addr_len
)
3014 i
= kmalloc(sizeof(*i
), GFP_KERNEL
);
3019 for (ml
= po
->mclist
; ml
; ml
= ml
->next
) {
3020 if (ml
->ifindex
== mreq
->mr_ifindex
&&
3021 ml
->type
== mreq
->mr_type
&&
3022 ml
->alen
== mreq
->mr_alen
&&
3023 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
3025 /* Free the new element ... */
3031 i
->type
= mreq
->mr_type
;
3032 i
->ifindex
= mreq
->mr_ifindex
;
3033 i
->alen
= mreq
->mr_alen
;
3034 memcpy(i
->addr
, mreq
->mr_address
, i
->alen
);
3035 memset(i
->addr
+ i
->alen
, 0, sizeof(i
->addr
) - i
->alen
);
3037 i
->next
= po
->mclist
;
3039 err
= packet_dev_mc(dev
, i
, 1);
3041 po
->mclist
= i
->next
;
3050 static int packet_mc_drop(struct sock
*sk
, struct packet_mreq_max
*mreq
)
3052 struct packet_mclist
*ml
, **mlp
;
3056 for (mlp
= &pkt_sk(sk
)->mclist
; (ml
= *mlp
) != NULL
; mlp
= &ml
->next
) {
3057 if (ml
->ifindex
== mreq
->mr_ifindex
&&
3058 ml
->type
== mreq
->mr_type
&&
3059 ml
->alen
== mreq
->mr_alen
&&
3060 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
3061 if (--ml
->count
== 0) {
3062 struct net_device
*dev
;
3064 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
3066 packet_dev_mc(dev
, ml
, -1);
3074 return -EADDRNOTAVAIL
;
3077 static void packet_flush_mclist(struct sock
*sk
)
3079 struct packet_sock
*po
= pkt_sk(sk
);
3080 struct packet_mclist
*ml
;
3086 while ((ml
= po
->mclist
) != NULL
) {
3087 struct net_device
*dev
;
3089 po
->mclist
= ml
->next
;
3090 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
3092 packet_dev_mc(dev
, ml
, -1);
3099 packet_setsockopt(struct socket
*sock
, int level
, int optname
, char __user
*optval
, unsigned int optlen
)
3101 struct sock
*sk
= sock
->sk
;
3102 struct packet_sock
*po
= pkt_sk(sk
);
3105 if (level
!= SOL_PACKET
)
3106 return -ENOPROTOOPT
;
3109 case PACKET_ADD_MEMBERSHIP
:
3110 case PACKET_DROP_MEMBERSHIP
:
3112 struct packet_mreq_max mreq
;
3114 memset(&mreq
, 0, sizeof(mreq
));
3115 if (len
< sizeof(struct packet_mreq
))
3117 if (len
> sizeof(mreq
))
3119 if (copy_from_user(&mreq
, optval
, len
))
3121 if (len
< (mreq
.mr_alen
+ offsetof(struct packet_mreq
, mr_address
)))
3123 if (optname
== PACKET_ADD_MEMBERSHIP
)
3124 ret
= packet_mc_add(sk
, &mreq
);
3126 ret
= packet_mc_drop(sk
, &mreq
);
3130 case PACKET_RX_RING
:
3131 case PACKET_TX_RING
:
3133 union tpacket_req_u req_u
;
3136 switch (po
->tp_version
) {
3139 len
= sizeof(req_u
.req
);
3143 len
= sizeof(req_u
.req3
);
3148 if (pkt_sk(sk
)->has_vnet_hdr
)
3150 if (copy_from_user(&req_u
.req
, optval
, len
))
3152 return packet_set_ring(sk
, &req_u
, 0,
3153 optname
== PACKET_TX_RING
);
3155 case PACKET_COPY_THRESH
:
3159 if (optlen
!= sizeof(val
))
3161 if (copy_from_user(&val
, optval
, sizeof(val
)))
3164 pkt_sk(sk
)->copy_thresh
= val
;
3167 case PACKET_VERSION
:
3171 if (optlen
!= sizeof(val
))
3173 if (copy_from_user(&val
, optval
, sizeof(val
)))
3184 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
) {
3187 po
->tp_version
= val
;
3193 case PACKET_RESERVE
:
3197 if (optlen
!= sizeof(val
))
3199 if (copy_from_user(&val
, optval
, sizeof(val
)))
3204 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
) {
3207 po
->tp_reserve
= val
;
3217 if (optlen
!= sizeof(val
))
3219 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3221 if (copy_from_user(&val
, optval
, sizeof(val
)))
3223 po
->tp_loss
= !!val
;
3226 case PACKET_AUXDATA
:
3230 if (optlen
< sizeof(val
))
3232 if (copy_from_user(&val
, optval
, sizeof(val
)))
3235 po
->auxdata
= !!val
;
3238 case PACKET_ORIGDEV
:
3242 if (optlen
< sizeof(val
))
3244 if (copy_from_user(&val
, optval
, sizeof(val
)))
3247 po
->origdev
= !!val
;
3250 case PACKET_VNET_HDR
:
3254 if (sock
->type
!= SOCK_RAW
)
3256 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3258 if (optlen
< sizeof(val
))
3260 if (copy_from_user(&val
, optval
, sizeof(val
)))
3263 po
->has_vnet_hdr
= !!val
;
3266 case PACKET_TIMESTAMP
:
3270 if (optlen
!= sizeof(val
))
3272 if (copy_from_user(&val
, optval
, sizeof(val
)))
3275 po
->tp_tstamp
= val
;
3282 if (optlen
!= sizeof(val
))
3284 if (copy_from_user(&val
, optval
, sizeof(val
)))
3287 return fanout_add(sk
, val
& 0xffff, val
>> 16);
3289 case PACKET_TX_HAS_OFF
:
3293 if (optlen
!= sizeof(val
))
3295 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3297 if (copy_from_user(&val
, optval
, sizeof(val
)))
3299 po
->tp_tx_has_off
= !!val
;
3303 return -ENOPROTOOPT
;
3307 static int packet_getsockopt(struct socket
*sock
, int level
, int optname
,
3308 char __user
*optval
, int __user
*optlen
)
3311 int val
, lv
= sizeof(val
);
3312 struct sock
*sk
= sock
->sk
;
3313 struct packet_sock
*po
= pkt_sk(sk
);
3315 union tpacket_stats_u st
;
3317 if (level
!= SOL_PACKET
)
3318 return -ENOPROTOOPT
;
3320 if (get_user(len
, optlen
))
3327 case PACKET_STATISTICS
:
3328 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3329 memcpy(&st
, &po
->stats
, sizeof(st
));
3330 memset(&po
->stats
, 0, sizeof(po
->stats
));
3331 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3333 if (po
->tp_version
== TPACKET_V3
) {
3334 lv
= sizeof(struct tpacket_stats_v3
);
3335 st
.stats3
.tp_packets
+= st
.stats3
.tp_drops
;
3338 lv
= sizeof(struct tpacket_stats
);
3339 st
.stats1
.tp_packets
+= st
.stats1
.tp_drops
;
3344 case PACKET_AUXDATA
:
3347 case PACKET_ORIGDEV
:
3350 case PACKET_VNET_HDR
:
3351 val
= po
->has_vnet_hdr
;
3353 case PACKET_VERSION
:
3354 val
= po
->tp_version
;
3357 if (len
> sizeof(int))
3359 if (len
< sizeof(int))
3361 if (copy_from_user(&val
, optval
, len
))
3365 val
= sizeof(struct tpacket_hdr
);
3368 val
= sizeof(struct tpacket2_hdr
);
3371 val
= sizeof(struct tpacket3_hdr
);
3377 case PACKET_RESERVE
:
3378 val
= po
->tp_reserve
;
3383 case PACKET_TIMESTAMP
:
3384 val
= po
->tp_tstamp
;
3388 ((u32
)po
->fanout
->id
|
3389 ((u32
)po
->fanout
->type
<< 16) |
3390 ((u32
)po
->fanout
->flags
<< 24)) :
3393 case PACKET_TX_HAS_OFF
:
3394 val
= po
->tp_tx_has_off
;
3397 return -ENOPROTOOPT
;
3402 if (put_user(len
, optlen
))
3404 if (copy_to_user(optval
, data
, len
))
3410 static int packet_notifier(struct notifier_block
*this, unsigned long msg
, void *data
)
3413 struct net_device
*dev
= data
;
3414 struct net
*net
= dev_net(dev
);
3417 sk_for_each_rcu(sk
, &net
->packet
.sklist
) {
3418 struct packet_sock
*po
= pkt_sk(sk
);
3421 case NETDEV_UNREGISTER
:
3423 packet_dev_mclist(dev
, po
->mclist
, -1);
3427 if (dev
->ifindex
== po
->ifindex
) {
3428 spin_lock(&po
->bind_lock
);
3430 __unregister_prot_hook(sk
, false);
3431 sk
->sk_err
= ENETDOWN
;
3432 if (!sock_flag(sk
, SOCK_DEAD
))
3433 sk
->sk_error_report(sk
);
3435 if (msg
== NETDEV_UNREGISTER
) {
3436 packet_cached_dev_reset(po
);
3438 if (po
->prot_hook
.dev
)
3439 dev_put(po
->prot_hook
.dev
);
3440 po
->prot_hook
.dev
= NULL
;
3442 spin_unlock(&po
->bind_lock
);
3446 if (dev
->ifindex
== po
->ifindex
) {
3447 spin_lock(&po
->bind_lock
);
3449 register_prot_hook(sk
);
3450 spin_unlock(&po
->bind_lock
);
3460 static int packet_ioctl(struct socket
*sock
, unsigned int cmd
,
3463 struct sock
*sk
= sock
->sk
;
3468 int amount
= sk_wmem_alloc_get(sk
);
3470 return put_user(amount
, (int __user
*)arg
);
3474 struct sk_buff
*skb
;
3477 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3478 skb
= skb_peek(&sk
->sk_receive_queue
);
3481 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3482 return put_user(amount
, (int __user
*)arg
);
3485 return sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
3487 return sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
3497 case SIOCGIFBRDADDR
:
3498 case SIOCSIFBRDADDR
:
3499 case SIOCGIFNETMASK
:
3500 case SIOCSIFNETMASK
:
3501 case SIOCGIFDSTADDR
:
3502 case SIOCSIFDSTADDR
:
3504 return inet_dgram_ops
.ioctl(sock
, cmd
, arg
);
3508 return -ENOIOCTLCMD
;
3513 static unsigned int packet_poll(struct file
*file
, struct socket
*sock
,
3516 struct sock
*sk
= sock
->sk
;
3517 struct packet_sock
*po
= pkt_sk(sk
);
3518 unsigned int mask
= datagram_poll(file
, sock
, wait
);
3520 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3521 if (po
->rx_ring
.pg_vec
) {
3522 if (!packet_previous_rx_frame(po
, &po
->rx_ring
,
3524 mask
|= POLLIN
| POLLRDNORM
;
3526 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3527 spin_lock_bh(&sk
->sk_write_queue
.lock
);
3528 if (po
->tx_ring
.pg_vec
) {
3529 if (packet_current_frame(po
, &po
->tx_ring
, TP_STATUS_AVAILABLE
))
3530 mask
|= POLLOUT
| POLLWRNORM
;
3532 spin_unlock_bh(&sk
->sk_write_queue
.lock
);
3537 /* Dirty? Well, I still did not learn better way to account
3541 static void packet_mm_open(struct vm_area_struct
*vma
)
3543 struct file
*file
= vma
->vm_file
;
3544 struct socket
*sock
= file
->private_data
;
3545 struct sock
*sk
= sock
->sk
;
3548 atomic_inc(&pkt_sk(sk
)->mapped
);
3551 static void packet_mm_close(struct vm_area_struct
*vma
)
3553 struct file
*file
= vma
->vm_file
;
3554 struct socket
*sock
= file
->private_data
;
3555 struct sock
*sk
= sock
->sk
;
3558 atomic_dec(&pkt_sk(sk
)->mapped
);
3561 static const struct vm_operations_struct packet_mmap_ops
= {
3562 .open
= packet_mm_open
,
3563 .close
= packet_mm_close
,
3566 static void free_pg_vec(struct pgv
*pg_vec
, unsigned int order
,
3571 for (i
= 0; i
< len
; i
++) {
3572 if (likely(pg_vec
[i
].buffer
)) {
3573 if (is_vmalloc_addr(pg_vec
[i
].buffer
))
3574 vfree(pg_vec
[i
].buffer
);
3576 free_pages((unsigned long)pg_vec
[i
].buffer
,
3578 pg_vec
[i
].buffer
= NULL
;
3584 static char *alloc_one_pg_vec_page(unsigned long order
)
3586 char *buffer
= NULL
;
3587 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_COMP
|
3588 __GFP_ZERO
| __GFP_NOWARN
| __GFP_NORETRY
;
3590 buffer
= (char *) __get_free_pages(gfp_flags
, order
);
3596 * __get_free_pages failed, fall back to vmalloc
3598 buffer
= vzalloc((1 << order
) * PAGE_SIZE
);
3604 * vmalloc failed, lets dig into swap here
3606 gfp_flags
&= ~__GFP_NORETRY
;
3607 buffer
= (char *)__get_free_pages(gfp_flags
, order
);
3612 * complete and utter failure
3617 static struct pgv
*alloc_pg_vec(struct tpacket_req
*req
, int order
)
3619 unsigned int block_nr
= req
->tp_block_nr
;
3623 pg_vec
= kcalloc(block_nr
, sizeof(struct pgv
), GFP_KERNEL
);
3624 if (unlikely(!pg_vec
))
3627 for (i
= 0; i
< block_nr
; i
++) {
3628 pg_vec
[i
].buffer
= alloc_one_pg_vec_page(order
);
3629 if (unlikely(!pg_vec
[i
].buffer
))
3630 goto out_free_pgvec
;
3637 free_pg_vec(pg_vec
, order
, block_nr
);
3642 static int packet_set_ring(struct sock
*sk
, union tpacket_req_u
*req_u
,
3643 int closing
, int tx_ring
)
3645 struct pgv
*pg_vec
= NULL
;
3646 struct packet_sock
*po
= pkt_sk(sk
);
3647 int was_running
, order
= 0;
3648 struct packet_ring_buffer
*rb
;
3649 struct sk_buff_head
*rb_queue
;
3652 /* Added to avoid minimal code churn */
3653 struct tpacket_req
*req
= &req_u
->req
;
3656 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3657 if (!closing
&& tx_ring
&& (po
->tp_version
> TPACKET_V2
)) {
3658 WARN(1, "Tx-ring is not supported.\n");
3662 rb
= tx_ring
? &po
->tx_ring
: &po
->rx_ring
;
3663 rb_queue
= tx_ring
? &sk
->sk_write_queue
: &sk
->sk_receive_queue
;
3667 if (atomic_read(&po
->mapped
))
3669 if (atomic_read(&rb
->pending
))
3673 if (req
->tp_block_nr
) {
3674 /* Sanity tests and some calculations */
3676 if (unlikely(rb
->pg_vec
))
3679 switch (po
->tp_version
) {
3681 po
->tp_hdrlen
= TPACKET_HDRLEN
;
3684 po
->tp_hdrlen
= TPACKET2_HDRLEN
;
3687 po
->tp_hdrlen
= TPACKET3_HDRLEN
;
3692 if (unlikely((int)req
->tp_block_size
<= 0))
3694 if (unlikely(req
->tp_block_size
& (PAGE_SIZE
- 1)))
3696 if (po
->tp_version
>= TPACKET_V3
&&
3697 req
->tp_block_size
<=
3698 BLK_PLUS_PRIV((u64
)req_u
->req3
.tp_sizeof_priv
))
3700 if (unlikely(req
->tp_frame_size
< po
->tp_hdrlen
+
3703 if (unlikely(req
->tp_frame_size
& (TPACKET_ALIGNMENT
- 1)))
3706 rb
->frames_per_block
= req
->tp_block_size
/req
->tp_frame_size
;
3707 if (unlikely(rb
->frames_per_block
<= 0))
3709 if (unlikely(req
->tp_block_size
> UINT_MAX
/ req
->tp_block_nr
))
3711 if (unlikely((rb
->frames_per_block
* req
->tp_block_nr
) !=
3716 order
= get_order(req
->tp_block_size
);
3717 pg_vec
= alloc_pg_vec(req
, order
);
3718 if (unlikely(!pg_vec
))
3720 switch (po
->tp_version
) {
3722 /* Transmit path is not supported. We checked
3723 * it above but just being paranoid
3726 init_prb_bdqc(po
, rb
, pg_vec
, req_u
, tx_ring
);
3735 if (unlikely(req
->tp_frame_nr
))
3740 /* Detach socket from network */
3741 spin_lock(&po
->bind_lock
);
3742 was_running
= po
->running
;
3746 __unregister_prot_hook(sk
, false);
3748 spin_unlock(&po
->bind_lock
);
3753 mutex_lock(&po
->pg_vec_lock
);
3754 if (closing
|| atomic_read(&po
->mapped
) == 0) {
3756 spin_lock_bh(&rb_queue
->lock
);
3757 swap(rb
->pg_vec
, pg_vec
);
3758 rb
->frame_max
= (req
->tp_frame_nr
- 1);
3760 rb
->frame_size
= req
->tp_frame_size
;
3761 spin_unlock_bh(&rb_queue
->lock
);
3763 swap(rb
->pg_vec_order
, order
);
3764 swap(rb
->pg_vec_len
, req
->tp_block_nr
);
3766 rb
->pg_vec_pages
= req
->tp_block_size
/PAGE_SIZE
;
3767 po
->prot_hook
.func
= (po
->rx_ring
.pg_vec
) ?
3768 tpacket_rcv
: packet_rcv
;
3769 skb_queue_purge(rb_queue
);
3770 if (atomic_read(&po
->mapped
))
3771 pr_err("packet_mmap: vma is busy: %d\n",
3772 atomic_read(&po
->mapped
));
3774 mutex_unlock(&po
->pg_vec_lock
);
3776 spin_lock(&po
->bind_lock
);
3779 register_prot_hook(sk
);
3781 spin_unlock(&po
->bind_lock
);
3782 if (closing
&& (po
->tp_version
> TPACKET_V2
)) {
3783 /* Because we don't support block-based V3 on tx-ring */
3785 prb_shutdown_retire_blk_timer(po
, tx_ring
, rb_queue
);
3789 free_pg_vec(pg_vec
, order
, req
->tp_block_nr
);
3795 static int packet_mmap(struct file
*file
, struct socket
*sock
,
3796 struct vm_area_struct
*vma
)
3798 struct sock
*sk
= sock
->sk
;
3799 struct packet_sock
*po
= pkt_sk(sk
);
3800 unsigned long size
, expected_size
;
3801 struct packet_ring_buffer
*rb
;
3802 unsigned long start
;
3809 mutex_lock(&po
->pg_vec_lock
);
3812 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
3814 expected_size
+= rb
->pg_vec_len
3820 if (expected_size
== 0)
3823 size
= vma
->vm_end
- vma
->vm_start
;
3824 if (size
!= expected_size
)
3827 start
= vma
->vm_start
;
3828 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
3829 if (rb
->pg_vec
== NULL
)
3832 for (i
= 0; i
< rb
->pg_vec_len
; i
++) {
3834 void *kaddr
= rb
->pg_vec
[i
].buffer
;
3837 for (pg_num
= 0; pg_num
< rb
->pg_vec_pages
; pg_num
++) {
3838 page
= pgv_to_page(kaddr
);
3839 err
= vm_insert_page(vma
, start
, page
);
3848 atomic_inc(&po
->mapped
);
3849 vma
->vm_ops
= &packet_mmap_ops
;
3853 mutex_unlock(&po
->pg_vec_lock
);
3857 static const struct proto_ops packet_ops_spkt
= {
3858 .family
= PF_PACKET
,
3859 .owner
= THIS_MODULE
,
3860 .release
= packet_release
,
3861 .bind
= packet_bind_spkt
,
3862 .connect
= sock_no_connect
,
3863 .socketpair
= sock_no_socketpair
,
3864 .accept
= sock_no_accept
,
3865 .getname
= packet_getname_spkt
,
3866 .poll
= datagram_poll
,
3867 .ioctl
= packet_ioctl
,
3868 .listen
= sock_no_listen
,
3869 .shutdown
= sock_no_shutdown
,
3870 .setsockopt
= sock_no_setsockopt
,
3871 .getsockopt
= sock_no_getsockopt
,
3872 .sendmsg
= packet_sendmsg_spkt
,
3873 .recvmsg
= packet_recvmsg
,
3874 .mmap
= sock_no_mmap
,
3875 .sendpage
= sock_no_sendpage
,
3878 static const struct proto_ops packet_ops
= {
3879 .family
= PF_PACKET
,
3880 .owner
= THIS_MODULE
,
3881 .release
= packet_release
,
3882 .bind
= packet_bind
,
3883 .connect
= sock_no_connect
,
3884 .socketpair
= sock_no_socketpair
,
3885 .accept
= sock_no_accept
,
3886 .getname
= packet_getname
,
3887 .poll
= packet_poll
,
3888 .ioctl
= packet_ioctl
,
3889 .listen
= sock_no_listen
,
3890 .shutdown
= sock_no_shutdown
,
3891 .setsockopt
= packet_setsockopt
,
3892 .getsockopt
= packet_getsockopt
,
3893 .sendmsg
= packet_sendmsg
,
3894 .recvmsg
= packet_recvmsg
,
3895 .mmap
= packet_mmap
,
3896 .sendpage
= sock_no_sendpage
,
3899 static const struct net_proto_family packet_family_ops
= {
3900 .family
= PF_PACKET
,
3901 .create
= packet_create
,
3902 .owner
= THIS_MODULE
,
3905 static struct notifier_block packet_netdev_notifier
= {
3906 .notifier_call
= packet_notifier
,
3909 #ifdef CONFIG_PROC_FS
3911 static void *packet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3914 struct net
*net
= seq_file_net(seq
);
3917 return seq_hlist_start_head_rcu(&net
->packet
.sklist
, *pos
);
3920 static void *packet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3922 struct net
*net
= seq_file_net(seq
);
3923 return seq_hlist_next_rcu(v
, &net
->packet
.sklist
, pos
);
3926 static void packet_seq_stop(struct seq_file
*seq
, void *v
)
3932 static int packet_seq_show(struct seq_file
*seq
, void *v
)
3934 if (v
== SEQ_START_TOKEN
)
3935 seq_puts(seq
, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3937 struct sock
*s
= sk_entry(v
);
3938 const struct packet_sock
*po
= pkt_sk(s
);
3941 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3943 atomic_read(&s
->sk_refcnt
),
3948 atomic_read(&s
->sk_rmem_alloc
),
3949 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(s
)),
3956 static const struct seq_operations packet_seq_ops
= {
3957 .start
= packet_seq_start
,
3958 .next
= packet_seq_next
,
3959 .stop
= packet_seq_stop
,
3960 .show
= packet_seq_show
,
3963 static int packet_seq_open(struct inode
*inode
, struct file
*file
)
3965 return seq_open_net(inode
, file
, &packet_seq_ops
,
3966 sizeof(struct seq_net_private
));
3969 static const struct file_operations packet_seq_fops
= {
3970 .owner
= THIS_MODULE
,
3971 .open
= packet_seq_open
,
3973 .llseek
= seq_lseek
,
3974 .release
= seq_release_net
,
3979 static int __net_init
packet_net_init(struct net
*net
)
3981 mutex_init(&net
->packet
.sklist_lock
);
3982 INIT_HLIST_HEAD(&net
->packet
.sklist
);
3984 if (!proc_create("packet", 0, net
->proc_net
, &packet_seq_fops
))
3990 static void __net_exit
packet_net_exit(struct net
*net
)
3992 remove_proc_entry("packet", net
->proc_net
);
3995 static struct pernet_operations packet_net_ops
= {
3996 .init
= packet_net_init
,
3997 .exit
= packet_net_exit
,
4001 static void __exit
packet_exit(void)
4003 unregister_netdevice_notifier(&packet_netdev_notifier
);
4004 unregister_pernet_subsys(&packet_net_ops
);
4005 sock_unregister(PF_PACKET
);
4006 proto_unregister(&packet_proto
);
4009 static int __init
packet_init(void)
4011 int rc
= proto_register(&packet_proto
, 0);
4016 sock_register(&packet_family_ops
);
4017 register_pernet_subsys(&packet_net_ops
);
4018 register_netdevice_notifier(&packet_netdev_notifier
);
4023 module_init(packet_init
);
4024 module_exit(packet_exit
);
4025 MODULE_LICENSE("GPL");
4026 MODULE_ALIAS_NETPROTO(PF_PACKET
);