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
);
1353 if (match
->type
== type
&&
1354 match
->prot_hook
.type
== po
->prot_hook
.type
&&
1355 match
->prot_hook
.dev
== po
->prot_hook
.dev
) {
1357 if (atomic_read(&match
->sk_ref
) < PACKET_FANOUT_MAX
) {
1358 __dev_remove_pack(&po
->prot_hook
);
1360 atomic_inc(&match
->sk_ref
);
1361 __fanout_link(sk
, po
);
1366 mutex_unlock(&fanout_mutex
);
1370 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1371 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1372 * It is the responsibility of the caller to call fanout_release_data() and
1373 * free the returned packet_fanout (after synchronize_net())
1375 static struct packet_fanout
*fanout_release(struct sock
*sk
)
1377 struct packet_sock
*po
= pkt_sk(sk
);
1378 struct packet_fanout
*f
;
1380 mutex_lock(&fanout_mutex
);
1385 if (atomic_dec_and_test(&f
->sk_ref
))
1390 mutex_unlock(&fanout_mutex
);
1395 static const struct proto_ops packet_ops
;
1397 static const struct proto_ops packet_ops_spkt
;
1399 static int packet_rcv_spkt(struct sk_buff
*skb
, struct net_device
*dev
,
1400 struct packet_type
*pt
, struct net_device
*orig_dev
)
1403 struct sockaddr_pkt
*spkt
;
1406 * When we registered the protocol we saved the socket in the data
1407 * field for just this event.
1410 sk
= pt
->af_packet_priv
;
1413 * Yank back the headers [hope the device set this
1414 * right or kerboom...]
1416 * Incoming packets have ll header pulled,
1419 * For outgoing ones skb->data == skb_mac_header(skb)
1420 * so that this procedure is noop.
1423 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1426 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1429 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1433 /* drop any routing info */
1436 /* drop conntrack reference */
1439 spkt
= &PACKET_SKB_CB(skb
)->sa
.pkt
;
1441 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1444 * The SOCK_PACKET socket receives _all_ frames.
1447 spkt
->spkt_family
= dev
->type
;
1448 strlcpy(spkt
->spkt_device
, dev
->name
, sizeof(spkt
->spkt_device
));
1449 spkt
->spkt_protocol
= skb
->protocol
;
1452 * Charge the memory to the socket. This is done specifically
1453 * to prevent sockets using all the memory up.
1456 if (sock_queue_rcv_skb(sk
, skb
) == 0)
1467 * Output a raw packet to a device layer. This bypasses all the other
1468 * protocol layers and you must therefore supply it with a complete frame
1471 static int packet_sendmsg_spkt(struct kiocb
*iocb
, struct socket
*sock
,
1472 struct msghdr
*msg
, size_t len
)
1474 struct sock
*sk
= sock
->sk
;
1475 struct sockaddr_pkt
*saddr
= (struct sockaddr_pkt
*)msg
->msg_name
;
1476 struct sk_buff
*skb
= NULL
;
1477 struct net_device
*dev
;
1483 * Get and verify the address.
1487 if (msg
->msg_namelen
< sizeof(struct sockaddr
))
1489 if (msg
->msg_namelen
== sizeof(struct sockaddr_pkt
))
1490 proto
= saddr
->spkt_protocol
;
1492 return -ENOTCONN
; /* SOCK_PACKET must be sent giving an address */
1495 * Find the device first to size check it
1498 saddr
->spkt_device
[sizeof(saddr
->spkt_device
) - 1] = 0;
1501 dev
= dev_get_by_name_rcu(sock_net(sk
), saddr
->spkt_device
);
1507 if (!(dev
->flags
& IFF_UP
))
1511 * You may not queue a frame bigger than the mtu. This is the lowest level
1512 * raw protocol and you must do your own fragmentation at this level.
1515 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
1516 if (!netif_supports_nofcs(dev
)) {
1517 err
= -EPROTONOSUPPORT
;
1520 extra_len
= 4; /* We're doing our own CRC */
1524 if (len
> dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
+ extra_len
)
1528 size_t reserved
= LL_RESERVED_SPACE(dev
);
1529 int tlen
= dev
->needed_tailroom
;
1530 unsigned int hhlen
= dev
->header_ops
? dev
->hard_header_len
: 0;
1533 skb
= sock_wmalloc(sk
, len
+ reserved
+ tlen
, 0, GFP_KERNEL
);
1536 /* FIXME: Save some space for broken drivers that write a hard
1537 * header at transmission time by themselves. PPP is the notable
1538 * one here. This should really be fixed at the driver level.
1540 skb_reserve(skb
, reserved
);
1541 skb_reset_network_header(skb
);
1543 /* Try to align data part correctly */
1548 skb_reset_network_header(skb
);
1550 err
= memcpy_fromiovec(skb_put(skb
, len
), msg
->msg_iov
, len
);
1556 if (len
> (dev
->mtu
+ dev
->hard_header_len
+ extra_len
)) {
1557 /* Earlier code assumed this would be a VLAN pkt,
1558 * double-check this now that we have the actual
1561 struct ethhdr
*ehdr
;
1562 skb_reset_mac_header(skb
);
1563 ehdr
= eth_hdr(skb
);
1564 if (ehdr
->h_proto
!= htons(ETH_P_8021Q
)) {
1570 skb
->protocol
= proto
;
1572 skb
->priority
= sk
->sk_priority
;
1573 skb
->mark
= sk
->sk_mark
;
1575 sock_tx_timestamp(sk
, &skb_shinfo(skb
)->tx_flags
);
1577 if (unlikely(extra_len
== 4))
1580 skb_probe_transport_header(skb
, 0);
1582 dev_queue_xmit(skb
);
1593 static unsigned int run_filter(const struct sk_buff
*skb
,
1594 const struct sock
*sk
,
1597 struct sk_filter
*filter
;
1600 filter
= rcu_dereference(sk
->sk_filter
);
1602 res
= SK_RUN_FILTER(filter
, skb
);
1609 * This function makes lazy skb cloning in hope that most of packets
1610 * are discarded by BPF.
1612 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1613 * and skb->cb are mangled. It works because (and until) packets
1614 * falling here are owned by current CPU. Output packets are cloned
1615 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1616 * sequencially, so that if we return skb to original state on exit,
1617 * we will not harm anyone.
1620 static int packet_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
1621 struct packet_type
*pt
, struct net_device
*orig_dev
)
1624 struct sockaddr_ll
*sll
;
1625 struct packet_sock
*po
;
1626 u8
*skb_head
= skb
->data
;
1627 int skb_len
= skb
->len
;
1628 unsigned int snaplen
, res
;
1630 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1633 sk
= pt
->af_packet_priv
;
1636 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1641 if (dev
->header_ops
) {
1642 /* The device has an explicit notion of ll header,
1643 * exported to higher levels.
1645 * Otherwise, the device hides details of its frame
1646 * structure, so that corresponding packet head is
1647 * never delivered to user.
1649 if (sk
->sk_type
!= SOCK_DGRAM
)
1650 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1651 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
1652 /* Special case: outgoing packets have ll header at head */
1653 skb_pull(skb
, skb_network_offset(skb
));
1659 res
= run_filter(skb
, sk
, snaplen
);
1661 goto drop_n_restore
;
1665 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
1668 if (skb_shared(skb
)) {
1669 struct sk_buff
*nskb
= skb_clone(skb
, GFP_ATOMIC
);
1673 if (skb_head
!= skb
->data
) {
1674 skb
->data
= skb_head
;
1681 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb
)) + MAX_ADDR_LEN
- 8 >
1684 sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
1685 sll
->sll_family
= AF_PACKET
;
1686 sll
->sll_hatype
= dev
->type
;
1687 sll
->sll_protocol
= skb
->protocol
;
1688 sll
->sll_pkttype
= skb
->pkt_type
;
1689 if (unlikely(po
->origdev
))
1690 sll
->sll_ifindex
= orig_dev
->ifindex
;
1692 sll
->sll_ifindex
= dev
->ifindex
;
1694 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
1696 PACKET_SKB_CB(skb
)->origlen
= skb
->len
;
1698 if (pskb_trim(skb
, snaplen
))
1701 skb_set_owner_r(skb
, sk
);
1705 /* drop conntrack reference */
1708 spin_lock(&sk
->sk_receive_queue
.lock
);
1709 po
->stats
.stats1
.tp_packets
++;
1710 skb
->dropcount
= atomic_read(&sk
->sk_drops
);
1711 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1712 spin_unlock(&sk
->sk_receive_queue
.lock
);
1713 sk
->sk_data_ready(sk
, skb
->len
);
1717 spin_lock(&sk
->sk_receive_queue
.lock
);
1718 po
->stats
.stats1
.tp_drops
++;
1719 atomic_inc(&sk
->sk_drops
);
1720 spin_unlock(&sk
->sk_receive_queue
.lock
);
1723 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
1724 skb
->data
= skb_head
;
1732 static int tpacket_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
1733 struct packet_type
*pt
, struct net_device
*orig_dev
)
1736 struct packet_sock
*po
;
1737 struct sockaddr_ll
*sll
;
1738 union tpacket_uhdr h
;
1739 u8
*skb_head
= skb
->data
;
1740 int skb_len
= skb
->len
;
1741 unsigned int snaplen
, res
;
1742 unsigned long status
= TP_STATUS_USER
;
1743 unsigned short macoff
, netoff
, hdrlen
;
1744 struct sk_buff
*copy_skb
= NULL
;
1748 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1751 sk
= pt
->af_packet_priv
;
1754 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1757 if (dev
->header_ops
) {
1758 if (sk
->sk_type
!= SOCK_DGRAM
)
1759 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1760 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
1761 /* Special case: outgoing packets have ll header at head */
1762 skb_pull(skb
, skb_network_offset(skb
));
1766 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
1767 status
|= TP_STATUS_CSUMNOTREADY
;
1771 res
= run_filter(skb
, sk
, snaplen
);
1773 goto drop_n_restore
;
1777 if (sk
->sk_type
== SOCK_DGRAM
) {
1778 macoff
= netoff
= TPACKET_ALIGN(po
->tp_hdrlen
) + 16 +
1781 unsigned int maclen
= skb_network_offset(skb
);
1782 netoff
= TPACKET_ALIGN(po
->tp_hdrlen
+
1783 (maclen
< 16 ? 16 : maclen
)) +
1785 macoff
= netoff
- maclen
;
1787 if (po
->tp_version
<= TPACKET_V2
) {
1788 if (macoff
+ snaplen
> po
->rx_ring
.frame_size
) {
1789 if (po
->copy_thresh
&&
1790 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
1791 if (skb_shared(skb
)) {
1792 copy_skb
= skb_clone(skb
, GFP_ATOMIC
);
1794 copy_skb
= skb_get(skb
);
1795 skb_head
= skb
->data
;
1798 skb_set_owner_r(copy_skb
, sk
);
1800 snaplen
= po
->rx_ring
.frame_size
- macoff
;
1801 if ((int)snaplen
< 0)
1804 } else if (unlikely(macoff
+ snaplen
>
1805 GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
)) {
1808 nval
= GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
- macoff
;
1809 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1810 snaplen
, nval
, macoff
);
1812 if (unlikely((int)snaplen
< 0)) {
1814 macoff
= GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
;
1817 spin_lock(&sk
->sk_receive_queue
.lock
);
1818 h
.raw
= packet_current_rx_frame(po
, skb
,
1819 TP_STATUS_KERNEL
, (macoff
+snaplen
));
1822 if (po
->tp_version
<= TPACKET_V2
) {
1823 packet_increment_rx_head(po
, &po
->rx_ring
);
1825 * LOSING will be reported till you read the stats,
1826 * because it's COR - Clear On Read.
1827 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1830 if (po
->stats
.stats1
.tp_drops
)
1831 status
|= TP_STATUS_LOSING
;
1833 po
->stats
.stats1
.tp_packets
++;
1835 status
|= TP_STATUS_COPY
;
1836 __skb_queue_tail(&sk
->sk_receive_queue
, copy_skb
);
1838 spin_unlock(&sk
->sk_receive_queue
.lock
);
1840 skb_copy_bits(skb
, 0, h
.raw
+ macoff
, snaplen
);
1842 if (!(ts_status
= tpacket_get_timestamp(skb
, &ts
, po
->tp_tstamp
)))
1843 getnstimeofday(&ts
);
1845 status
|= ts_status
;
1847 switch (po
->tp_version
) {
1849 h
.h1
->tp_len
= skb
->len
;
1850 h
.h1
->tp_snaplen
= snaplen
;
1851 h
.h1
->tp_mac
= macoff
;
1852 h
.h1
->tp_net
= netoff
;
1853 h
.h1
->tp_sec
= ts
.tv_sec
;
1854 h
.h1
->tp_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
1855 hdrlen
= sizeof(*h
.h1
);
1858 h
.h2
->tp_len
= skb
->len
;
1859 h
.h2
->tp_snaplen
= snaplen
;
1860 h
.h2
->tp_mac
= macoff
;
1861 h
.h2
->tp_net
= netoff
;
1862 h
.h2
->tp_sec
= ts
.tv_sec
;
1863 h
.h2
->tp_nsec
= ts
.tv_nsec
;
1864 if (vlan_tx_tag_present(skb
)) {
1865 h
.h2
->tp_vlan_tci
= vlan_tx_tag_get(skb
);
1866 status
|= TP_STATUS_VLAN_VALID
;
1868 h
.h2
->tp_vlan_tci
= 0;
1870 h
.h2
->tp_padding
= 0;
1871 hdrlen
= sizeof(*h
.h2
);
1874 /* tp_nxt_offset,vlan are already populated above.
1875 * So DONT clear those fields here
1877 h
.h3
->tp_status
|= status
;
1878 h
.h3
->tp_len
= skb
->len
;
1879 h
.h3
->tp_snaplen
= snaplen
;
1880 h
.h3
->tp_mac
= macoff
;
1881 h
.h3
->tp_net
= netoff
;
1882 h
.h3
->tp_sec
= ts
.tv_sec
;
1883 h
.h3
->tp_nsec
= ts
.tv_nsec
;
1884 hdrlen
= sizeof(*h
.h3
);
1890 sll
= h
.raw
+ TPACKET_ALIGN(hdrlen
);
1891 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
1892 sll
->sll_family
= AF_PACKET
;
1893 sll
->sll_hatype
= dev
->type
;
1894 sll
->sll_protocol
= skb
->protocol
;
1895 sll
->sll_pkttype
= skb
->pkt_type
;
1896 if (unlikely(po
->origdev
))
1897 sll
->sll_ifindex
= orig_dev
->ifindex
;
1899 sll
->sll_ifindex
= dev
->ifindex
;
1902 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1906 if (po
->tp_version
<= TPACKET_V2
) {
1907 end
= (u8
*)PAGE_ALIGN((unsigned long)h
.raw
1908 + macoff
+ snaplen
);
1909 for (start
= h
.raw
; start
< end
; start
+= PAGE_SIZE
)
1910 flush_dcache_page(pgv_to_page(start
));
1915 if (po
->tp_version
<= TPACKET_V2
)
1916 __packet_set_status(po
, h
.raw
, status
);
1918 prb_clear_blk_fill_status(&po
->rx_ring
);
1920 sk
->sk_data_ready(sk
, 0);
1923 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
1924 skb
->data
= skb_head
;
1932 po
->stats
.stats1
.tp_drops
++;
1933 spin_unlock(&sk
->sk_receive_queue
.lock
);
1935 sk
->sk_data_ready(sk
, 0);
1936 kfree_skb(copy_skb
);
1937 goto drop_n_restore
;
1940 static void tpacket_destruct_skb(struct sk_buff
*skb
)
1942 struct packet_sock
*po
= pkt_sk(skb
->sk
);
1945 if (likely(po
->tx_ring
.pg_vec
)) {
1948 ph
= skb_shinfo(skb
)->destructor_arg
;
1949 BUG_ON(atomic_read(&po
->tx_ring
.pending
) == 0);
1950 atomic_dec(&po
->tx_ring
.pending
);
1952 ts
= __packet_set_timestamp(po
, ph
, skb
);
1953 __packet_set_status(po
, ph
, TP_STATUS_AVAILABLE
| ts
);
1959 static int tpacket_fill_skb(struct packet_sock
*po
, struct sk_buff
*skb
,
1960 void *frame
, struct net_device
*dev
, int size_max
,
1961 __be16 proto
, unsigned char *addr
, int hlen
)
1963 union tpacket_uhdr ph
;
1964 int to_write
, offset
, len
, tp_len
, nr_frags
, len_max
;
1965 struct socket
*sock
= po
->sk
.sk_socket
;
1972 skb
->protocol
= proto
;
1974 skb
->priority
= po
->sk
.sk_priority
;
1975 skb
->mark
= po
->sk
.sk_mark
;
1976 sock_tx_timestamp(&po
->sk
, &skb_shinfo(skb
)->tx_flags
);
1977 skb_shinfo(skb
)->destructor_arg
= ph
.raw
;
1979 switch (po
->tp_version
) {
1981 tp_len
= ph
.h2
->tp_len
;
1984 tp_len
= ph
.h1
->tp_len
;
1987 if (unlikely(tp_len
> size_max
)) {
1988 pr_err("packet size is too long (%d > %d)\n", tp_len
, size_max
);
1992 skb_reserve(skb
, hlen
);
1993 skb_reset_network_header(skb
);
1994 skb_probe_transport_header(skb
, 0);
1996 if (po
->tp_tx_has_off
) {
1997 int off_min
, off_max
, off
;
1998 off_min
= po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
1999 off_max
= po
->tx_ring
.frame_size
- tp_len
;
2000 if (sock
->type
== SOCK_DGRAM
) {
2001 switch (po
->tp_version
) {
2003 off
= ph
.h2
->tp_net
;
2006 off
= ph
.h1
->tp_net
;
2010 switch (po
->tp_version
) {
2012 off
= ph
.h2
->tp_mac
;
2015 off
= ph
.h1
->tp_mac
;
2019 if (unlikely((off
< off_min
) || (off_max
< off
)))
2021 data
= ph
.raw
+ off
;
2023 data
= ph
.raw
+ po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
2027 if (sock
->type
== SOCK_DGRAM
) {
2028 err
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
,
2030 if (unlikely(err
< 0))
2032 } else if (dev
->hard_header_len
) {
2033 /* net device doesn't like empty head */
2034 if (unlikely(tp_len
<= dev
->hard_header_len
)) {
2035 pr_err("packet size is too short (%d < %d)\n",
2036 tp_len
, dev
->hard_header_len
);
2040 skb_push(skb
, dev
->hard_header_len
);
2041 err
= skb_store_bits(skb
, 0, data
,
2042 dev
->hard_header_len
);
2046 data
+= dev
->hard_header_len
;
2047 to_write
-= dev
->hard_header_len
;
2050 offset
= offset_in_page(data
);
2051 len_max
= PAGE_SIZE
- offset
;
2052 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2054 skb
->data_len
= to_write
;
2055 skb
->len
+= to_write
;
2056 skb
->truesize
+= to_write
;
2057 atomic_add(to_write
, &po
->sk
.sk_wmem_alloc
);
2059 while (likely(to_write
)) {
2060 nr_frags
= skb_shinfo(skb
)->nr_frags
;
2062 if (unlikely(nr_frags
>= MAX_SKB_FRAGS
)) {
2063 pr_err("Packet exceed the number of skb frags(%lu)\n",
2068 page
= pgv_to_page(data
);
2070 flush_dcache_page(page
);
2072 skb_fill_page_desc(skb
, nr_frags
, page
, offset
, len
);
2075 len_max
= PAGE_SIZE
;
2076 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2082 static int tpacket_snd(struct packet_sock
*po
, struct msghdr
*msg
)
2084 struct sk_buff
*skb
;
2085 struct net_device
*dev
;
2087 int err
, reserve
= 0;
2089 struct sockaddr_ll
*saddr
= (struct sockaddr_ll
*)msg
->msg_name
;
2090 int tp_len
, size_max
;
2091 unsigned char *addr
;
2093 int status
= TP_STATUS_AVAILABLE
;
2096 mutex_lock(&po
->pg_vec_lock
);
2098 if (likely(saddr
== NULL
)) {
2099 dev
= packet_cached_dev_get(po
);
2104 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2106 if (msg
->msg_namelen
< (saddr
->sll_halen
2107 + offsetof(struct sockaddr_ll
,
2110 proto
= saddr
->sll_protocol
;
2111 addr
= saddr
->sll_addr
;
2112 dev
= dev_get_by_index(sock_net(&po
->sk
), saddr
->sll_ifindex
);
2116 if (unlikely(dev
== NULL
))
2119 if (unlikely(!(dev
->flags
& IFF_UP
)))
2122 reserve
= dev
->hard_header_len
;
2124 size_max
= po
->tx_ring
.frame_size
2125 - (po
->tp_hdrlen
- sizeof(struct sockaddr_ll
));
2127 if (size_max
> dev
->mtu
+ reserve
)
2128 size_max
= dev
->mtu
+ reserve
;
2131 ph
= packet_current_frame(po
, &po
->tx_ring
,
2132 TP_STATUS_SEND_REQUEST
);
2134 if (unlikely(ph
== NULL
)) {
2139 status
= TP_STATUS_SEND_REQUEST
;
2140 hlen
= LL_RESERVED_SPACE(dev
);
2141 tlen
= dev
->needed_tailroom
;
2142 skb
= sock_alloc_send_skb(&po
->sk
,
2143 hlen
+ tlen
+ sizeof(struct sockaddr_ll
),
2146 if (unlikely(skb
== NULL
))
2149 tp_len
= tpacket_fill_skb(po
, skb
, ph
, dev
, size_max
, proto
,
2152 if (unlikely(tp_len
< 0)) {
2154 __packet_set_status(po
, ph
,
2155 TP_STATUS_AVAILABLE
);
2156 packet_increment_head(&po
->tx_ring
);
2160 status
= TP_STATUS_WRONG_FORMAT
;
2166 skb
->destructor
= tpacket_destruct_skb
;
2167 __packet_set_status(po
, ph
, TP_STATUS_SENDING
);
2168 atomic_inc(&po
->tx_ring
.pending
);
2170 status
= TP_STATUS_SEND_REQUEST
;
2171 err
= dev_queue_xmit(skb
);
2172 if (unlikely(err
> 0)) {
2173 err
= net_xmit_errno(err
);
2174 if (err
&& __packet_get_status(po
, ph
) ==
2175 TP_STATUS_AVAILABLE
) {
2176 /* skb was destructed already */
2181 * skb was dropped but not destructed yet;
2182 * let's treat it like congestion or err < 0
2186 packet_increment_head(&po
->tx_ring
);
2188 } while (likely((ph
!= NULL
) ||
2189 ((!(msg
->msg_flags
& MSG_DONTWAIT
)) &&
2190 (atomic_read(&po
->tx_ring
.pending
))))
2197 __packet_set_status(po
, ph
, status
);
2202 mutex_unlock(&po
->pg_vec_lock
);
2206 static struct sk_buff
*packet_alloc_skb(struct sock
*sk
, size_t prepad
,
2207 size_t reserve
, size_t len
,
2208 size_t linear
, int noblock
,
2211 struct sk_buff
*skb
;
2213 /* Under a page? Don't bother with paged skb. */
2214 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
2217 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
2222 skb_reserve(skb
, reserve
);
2223 skb_put(skb
, linear
);
2224 skb
->data_len
= len
- linear
;
2225 skb
->len
+= len
- linear
;
2230 static int packet_snd(struct socket
*sock
,
2231 struct msghdr
*msg
, size_t len
)
2233 struct sock
*sk
= sock
->sk
;
2234 struct sockaddr_ll
*saddr
= (struct sockaddr_ll
*)msg
->msg_name
;
2235 struct sk_buff
*skb
;
2236 struct net_device
*dev
;
2238 unsigned char *addr
;
2239 int err
, reserve
= 0;
2240 struct virtio_net_hdr vnet_hdr
= { 0 };
2243 struct packet_sock
*po
= pkt_sk(sk
);
2244 unsigned short gso_type
= 0;
2245 int hlen
, tlen
, linear
;
2249 * Get and verify the address.
2252 if (likely(saddr
== NULL
)) {
2253 dev
= packet_cached_dev_get(po
);
2258 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2260 if (msg
->msg_namelen
< (saddr
->sll_halen
+ offsetof(struct sockaddr_ll
, sll_addr
)))
2262 proto
= saddr
->sll_protocol
;
2263 addr
= saddr
->sll_addr
;
2264 dev
= dev_get_by_index(sock_net(sk
), saddr
->sll_ifindex
);
2268 if (unlikely(dev
== NULL
))
2271 if (unlikely(!(dev
->flags
& IFF_UP
)))
2274 if (sock
->type
== SOCK_RAW
)
2275 reserve
= dev
->hard_header_len
;
2276 if (po
->has_vnet_hdr
) {
2277 vnet_hdr_len
= sizeof(vnet_hdr
);
2280 if (len
< vnet_hdr_len
)
2283 len
-= vnet_hdr_len
;
2285 err
= memcpy_fromiovec((void *)&vnet_hdr
, msg
->msg_iov
,
2290 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
2291 (vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
2293 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
2294 vnet_hdr
.csum_offset
+ 2;
2297 if (vnet_hdr
.hdr_len
> len
)
2300 if (vnet_hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
2301 switch (vnet_hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
2302 case VIRTIO_NET_HDR_GSO_TCPV4
:
2303 gso_type
= SKB_GSO_TCPV4
;
2305 case VIRTIO_NET_HDR_GSO_TCPV6
:
2306 gso_type
= SKB_GSO_TCPV6
;
2308 case VIRTIO_NET_HDR_GSO_UDP
:
2309 gso_type
= SKB_GSO_UDP
;
2315 if (vnet_hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
2316 gso_type
|= SKB_GSO_TCP_ECN
;
2318 if (vnet_hdr
.gso_size
== 0)
2324 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
2325 if (!netif_supports_nofcs(dev
)) {
2326 err
= -EPROTONOSUPPORT
;
2329 extra_len
= 4; /* We're doing our own CRC */
2333 if (!gso_type
&& (len
> dev
->mtu
+ reserve
+ VLAN_HLEN
+ extra_len
))
2337 hlen
= LL_RESERVED_SPACE(dev
);
2338 tlen
= dev
->needed_tailroom
;
2339 linear
= vnet_hdr
.hdr_len
;
2340 linear
= max(linear
, min_t(int, len
, dev
->hard_header_len
));
2341 skb
= packet_alloc_skb(sk
, hlen
+ tlen
, hlen
, len
, linear
,
2342 msg
->msg_flags
& MSG_DONTWAIT
, &err
);
2346 skb_set_network_header(skb
, reserve
);
2349 if (sock
->type
== SOCK_DGRAM
&&
2350 (offset
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
, NULL
, len
)) < 0)
2353 /* Returns -EFAULT on error */
2354 err
= skb_copy_datagram_from_iovec(skb
, offset
, msg
->msg_iov
, 0, len
);
2358 sock_tx_timestamp(sk
, &skb_shinfo(skb
)->tx_flags
);
2360 if (!gso_type
&& (len
> dev
->mtu
+ reserve
+ extra_len
)) {
2361 /* Earlier code assumed this would be a VLAN pkt,
2362 * double-check this now that we have the actual
2365 struct ethhdr
*ehdr
;
2366 skb_reset_mac_header(skb
);
2367 ehdr
= eth_hdr(skb
);
2368 if (ehdr
->h_proto
!= htons(ETH_P_8021Q
)) {
2374 skb
->protocol
= proto
;
2376 skb
->priority
= sk
->sk_priority
;
2377 skb
->mark
= sk
->sk_mark
;
2379 if (po
->has_vnet_hdr
) {
2380 if (vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
2381 if (!skb_partial_csum_set(skb
, vnet_hdr
.csum_start
,
2382 vnet_hdr
.csum_offset
)) {
2388 skb_shinfo(skb
)->gso_size
= vnet_hdr
.gso_size
;
2389 skb_shinfo(skb
)->gso_type
= gso_type
;
2391 /* Header must be checked, and gso_segs computed. */
2392 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
2393 skb_shinfo(skb
)->gso_segs
= 0;
2395 len
+= vnet_hdr_len
;
2398 skb_probe_transport_header(skb
, reserve
);
2400 if (unlikely(extra_len
== 4))
2407 err
= dev_queue_xmit(skb
);
2408 if (err
> 0 && (err
= net_xmit_errno(err
)) != 0)
2424 static int packet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
2425 struct msghdr
*msg
, size_t len
)
2427 struct sock
*sk
= sock
->sk
;
2428 struct packet_sock
*po
= pkt_sk(sk
);
2429 if (po
->tx_ring
.pg_vec
)
2430 return tpacket_snd(po
, msg
);
2432 return packet_snd(sock
, msg
, len
);
2436 * Close a PACKET socket. This is fairly simple. We immediately go
2437 * to 'closed' state and remove our protocol entry in the device list.
2440 static int packet_release(struct socket
*sock
)
2442 struct sock
*sk
= sock
->sk
;
2443 struct packet_sock
*po
;
2444 struct packet_fanout
*f
;
2446 union tpacket_req_u req_u
;
2454 mutex_lock(&net
->packet
.sklist_lock
);
2455 sk_del_node_init_rcu(sk
);
2456 mutex_unlock(&net
->packet
.sklist_lock
);
2459 sock_prot_inuse_add(net
, sk
->sk_prot
, -1);
2462 spin_lock(&po
->bind_lock
);
2463 unregister_prot_hook(sk
, false);
2464 packet_cached_dev_reset(po
);
2466 if (po
->prot_hook
.dev
) {
2467 dev_put(po
->prot_hook
.dev
);
2468 po
->prot_hook
.dev
= NULL
;
2470 spin_unlock(&po
->bind_lock
);
2472 packet_flush_mclist(sk
);
2474 if (po
->rx_ring
.pg_vec
) {
2475 memset(&req_u
, 0, sizeof(req_u
));
2476 packet_set_ring(sk
, &req_u
, 1, 0);
2479 if (po
->tx_ring
.pg_vec
) {
2480 memset(&req_u
, 0, sizeof(req_u
));
2481 packet_set_ring(sk
, &req_u
, 1, 1);
2484 f
= fanout_release(sk
);
2492 * Now the socket is dead. No more input will appear.
2499 skb_queue_purge(&sk
->sk_receive_queue
);
2500 sk_refcnt_debug_release(sk
);
2507 * Attach a packet hook.
2510 static int packet_do_bind(struct sock
*sk
, struct net_device
*dev
, __be16 protocol
)
2512 struct packet_sock
*po
= pkt_sk(sk
);
2523 spin_lock(&po
->bind_lock
);
2524 unregister_prot_hook(sk
, true);
2527 po
->prot_hook
.type
= protocol
;
2528 if (po
->prot_hook
.dev
)
2529 dev_put(po
->prot_hook
.dev
);
2531 po
->prot_hook
.dev
= dev
;
2532 po
->ifindex
= dev
? dev
->ifindex
: 0;
2534 packet_cached_dev_assign(po
, dev
);
2539 if (!dev
|| (dev
->flags
& IFF_UP
)) {
2540 register_prot_hook(sk
);
2542 sk
->sk_err
= ENETDOWN
;
2543 if (!sock_flag(sk
, SOCK_DEAD
))
2544 sk
->sk_error_report(sk
);
2548 spin_unlock(&po
->bind_lock
);
2554 * Bind a packet socket to a device
2557 static int packet_bind_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
2560 struct sock
*sk
= sock
->sk
;
2561 char name
[sizeof(uaddr
->sa_data
) + 1];
2562 struct net_device
*dev
;
2569 if (addr_len
!= sizeof(struct sockaddr
))
2571 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
2574 memcpy(name
, uaddr
->sa_data
, sizeof(uaddr
->sa_data
));
2575 name
[sizeof(uaddr
->sa_data
)] = 0;
2577 dev
= dev_get_by_name(sock_net(sk
), name
);
2579 err
= packet_do_bind(sk
, dev
, pkt_sk(sk
)->num
);
2583 static int packet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
2585 struct sockaddr_ll
*sll
= (struct sockaddr_ll
*)uaddr
;
2586 struct sock
*sk
= sock
->sk
;
2587 struct net_device
*dev
= NULL
;
2595 if (addr_len
< sizeof(struct sockaddr_ll
))
2597 if (sll
->sll_family
!= AF_PACKET
)
2600 if (sll
->sll_ifindex
) {
2602 dev
= dev_get_by_index(sock_net(sk
), sll
->sll_ifindex
);
2606 err
= packet_do_bind(sk
, dev
, sll
->sll_protocol
? : pkt_sk(sk
)->num
);
2612 static struct proto packet_proto
= {
2614 .owner
= THIS_MODULE
,
2615 .obj_size
= sizeof(struct packet_sock
),
2619 * Create a packet of type SOCK_PACKET.
2622 static int packet_create(struct net
*net
, struct socket
*sock
, int protocol
,
2626 struct packet_sock
*po
;
2627 __be16 proto
= (__force __be16
)protocol
; /* weird, but documented */
2630 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
2632 if (sock
->type
!= SOCK_DGRAM
&& sock
->type
!= SOCK_RAW
&&
2633 sock
->type
!= SOCK_PACKET
)
2634 return -ESOCKTNOSUPPORT
;
2636 sock
->state
= SS_UNCONNECTED
;
2639 sk
= sk_alloc(net
, PF_PACKET
, GFP_KERNEL
, &packet_proto
);
2643 sock
->ops
= &packet_ops
;
2644 if (sock
->type
== SOCK_PACKET
)
2645 sock
->ops
= &packet_ops_spkt
;
2647 sock_init_data(sock
, sk
);
2650 sk
->sk_family
= PF_PACKET
;
2653 packet_cached_dev_reset(po
);
2655 sk
->sk_destruct
= packet_sock_destruct
;
2656 sk_refcnt_debug_inc(sk
);
2659 * Attach a protocol block
2662 spin_lock_init(&po
->bind_lock
);
2663 mutex_init(&po
->pg_vec_lock
);
2664 po
->prot_hook
.func
= packet_rcv
;
2666 if (sock
->type
== SOCK_PACKET
)
2667 po
->prot_hook
.func
= packet_rcv_spkt
;
2669 po
->prot_hook
.af_packet_priv
= sk
;
2672 po
->prot_hook
.type
= proto
;
2673 register_prot_hook(sk
);
2676 mutex_lock(&net
->packet
.sklist_lock
);
2677 sk_add_node_rcu(sk
, &net
->packet
.sklist
);
2678 mutex_unlock(&net
->packet
.sklist_lock
);
2681 sock_prot_inuse_add(net
, &packet_proto
, 1);
2689 static int packet_recv_error(struct sock
*sk
, struct msghdr
*msg
, int len
)
2691 struct sock_exterr_skb
*serr
;
2692 struct sk_buff
*skb
, *skb2
;
2696 skb
= skb_dequeue(&sk
->sk_error_queue
);
2702 msg
->msg_flags
|= MSG_TRUNC
;
2705 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
2709 sock_recv_timestamp(msg
, sk
, skb
);
2711 serr
= SKB_EXT_ERR(skb
);
2712 put_cmsg(msg
, SOL_PACKET
, PACKET_TX_TIMESTAMP
,
2713 sizeof(serr
->ee
), &serr
->ee
);
2715 msg
->msg_flags
|= MSG_ERRQUEUE
;
2718 /* Reset and regenerate socket error */
2719 spin_lock_bh(&sk
->sk_error_queue
.lock
);
2721 if ((skb2
= skb_peek(&sk
->sk_error_queue
)) != NULL
) {
2722 sk
->sk_err
= SKB_EXT_ERR(skb2
)->ee
.ee_errno
;
2723 spin_unlock_bh(&sk
->sk_error_queue
.lock
);
2724 sk
->sk_error_report(sk
);
2726 spin_unlock_bh(&sk
->sk_error_queue
.lock
);
2735 * Pull a packet from our receive queue and hand it to the user.
2736 * If necessary we block.
2739 static int packet_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
2740 struct msghdr
*msg
, size_t len
, int flags
)
2742 struct sock
*sk
= sock
->sk
;
2743 struct sk_buff
*skb
;
2745 int vnet_hdr_len
= 0;
2748 if (flags
& ~(MSG_PEEK
|MSG_DONTWAIT
|MSG_TRUNC
|MSG_CMSG_COMPAT
|MSG_ERRQUEUE
))
2752 /* What error should we return now? EUNATTACH? */
2753 if (pkt_sk(sk
)->ifindex
< 0)
2757 if (flags
& MSG_ERRQUEUE
) {
2758 err
= packet_recv_error(sk
, msg
, len
);
2763 * Call the generic datagram receiver. This handles all sorts
2764 * of horrible races and re-entrancy so we can forget about it
2765 * in the protocol layers.
2767 * Now it will return ENETDOWN, if device have just gone down,
2768 * but then it will block.
2771 skb
= skb_recv_datagram(sk
, flags
, flags
& MSG_DONTWAIT
, &err
);
2774 * An error occurred so return it. Because skb_recv_datagram()
2775 * handles the blocking we don't see and worry about blocking
2782 if (pkt_sk(sk
)->has_vnet_hdr
) {
2783 struct virtio_net_hdr vnet_hdr
= { 0 };
2786 vnet_hdr_len
= sizeof(vnet_hdr
);
2787 if (len
< vnet_hdr_len
)
2790 len
-= vnet_hdr_len
;
2792 if (skb_is_gso(skb
)) {
2793 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
2795 /* This is a hint as to how much should be linear. */
2796 vnet_hdr
.hdr_len
= skb_headlen(skb
);
2797 vnet_hdr
.gso_size
= sinfo
->gso_size
;
2798 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
2799 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
2800 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
2801 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
2802 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
2803 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
2804 else if (sinfo
->gso_type
& SKB_GSO_FCOE
)
2808 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
2809 vnet_hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
2811 vnet_hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
2813 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2814 vnet_hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
2815 vnet_hdr
.csum_start
= skb_checksum_start_offset(skb
);
2816 vnet_hdr
.csum_offset
= skb
->csum_offset
;
2817 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
2818 vnet_hdr
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
2819 } /* else everything is zero */
2821 err
= memcpy_toiovec(msg
->msg_iov
, (void *)&vnet_hdr
,
2827 /* You lose any data beyond the buffer you gave. If it worries
2828 * a user program they can ask the device for its MTU
2834 msg
->msg_flags
|= MSG_TRUNC
;
2837 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
2841 sock_recv_ts_and_drops(msg
, sk
, skb
);
2843 if (msg
->msg_name
) {
2844 /* If the address length field is there to be filled
2845 * in, we fill it in now.
2847 if (sock
->type
== SOCK_PACKET
) {
2848 msg
->msg_namelen
= sizeof(struct sockaddr_pkt
);
2850 struct sockaddr_ll
*sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
2851 msg
->msg_namelen
= sll
->sll_halen
+
2852 offsetof(struct sockaddr_ll
, sll_addr
);
2854 memcpy(msg
->msg_name
, &PACKET_SKB_CB(skb
)->sa
,
2858 if (pkt_sk(sk
)->auxdata
) {
2859 struct tpacket_auxdata aux
;
2861 aux
.tp_status
= TP_STATUS_USER
;
2862 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2863 aux
.tp_status
|= TP_STATUS_CSUMNOTREADY
;
2864 aux
.tp_len
= PACKET_SKB_CB(skb
)->origlen
;
2865 aux
.tp_snaplen
= skb
->len
;
2867 aux
.tp_net
= skb_network_offset(skb
);
2868 if (vlan_tx_tag_present(skb
)) {
2869 aux
.tp_vlan_tci
= vlan_tx_tag_get(skb
);
2870 aux
.tp_status
|= TP_STATUS_VLAN_VALID
;
2872 aux
.tp_vlan_tci
= 0;
2875 put_cmsg(msg
, SOL_PACKET
, PACKET_AUXDATA
, sizeof(aux
), &aux
);
2879 * Free or return the buffer as appropriate. Again this
2880 * hides all the races and re-entrancy issues from us.
2882 err
= vnet_hdr_len
+ ((flags
&MSG_TRUNC
) ? skb
->len
: copied
);
2885 skb_free_datagram(sk
, skb
);
2890 static int packet_getname_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
2891 int *uaddr_len
, int peer
)
2893 struct net_device
*dev
;
2894 struct sock
*sk
= sock
->sk
;
2899 uaddr
->sa_family
= AF_PACKET
;
2900 memset(uaddr
->sa_data
, 0, sizeof(uaddr
->sa_data
));
2902 dev
= dev_get_by_index_rcu(sock_net(sk
), pkt_sk(sk
)->ifindex
);
2904 strlcpy(uaddr
->sa_data
, dev
->name
, sizeof(uaddr
->sa_data
));
2906 *uaddr_len
= sizeof(*uaddr
);
2911 static int packet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
2912 int *uaddr_len
, int peer
)
2914 struct net_device
*dev
;
2915 struct sock
*sk
= sock
->sk
;
2916 struct packet_sock
*po
= pkt_sk(sk
);
2917 DECLARE_SOCKADDR(struct sockaddr_ll
*, sll
, uaddr
);
2922 sll
->sll_family
= AF_PACKET
;
2923 sll
->sll_ifindex
= po
->ifindex
;
2924 sll
->sll_protocol
= po
->num
;
2925 sll
->sll_pkttype
= 0;
2927 dev
= dev_get_by_index_rcu(sock_net(sk
), po
->ifindex
);
2929 sll
->sll_hatype
= dev
->type
;
2930 sll
->sll_halen
= dev
->addr_len
;
2931 memcpy(sll
->sll_addr
, dev
->dev_addr
, dev
->addr_len
);
2933 sll
->sll_hatype
= 0; /* Bad: we have no ARPHRD_UNSPEC */
2937 *uaddr_len
= offsetof(struct sockaddr_ll
, sll_addr
) + sll
->sll_halen
;
2942 static int packet_dev_mc(struct net_device
*dev
, struct packet_mclist
*i
,
2946 case PACKET_MR_MULTICAST
:
2947 if (i
->alen
!= dev
->addr_len
)
2950 return dev_mc_add(dev
, i
->addr
);
2952 return dev_mc_del(dev
, i
->addr
);
2954 case PACKET_MR_PROMISC
:
2955 return dev_set_promiscuity(dev
, what
);
2957 case PACKET_MR_ALLMULTI
:
2958 return dev_set_allmulti(dev
, what
);
2960 case PACKET_MR_UNICAST
:
2961 if (i
->alen
!= dev
->addr_len
)
2964 return dev_uc_add(dev
, i
->addr
);
2966 return dev_uc_del(dev
, i
->addr
);
2974 static void packet_dev_mclist(struct net_device
*dev
, struct packet_mclist
*i
, int what
)
2976 for ( ; i
; i
= i
->next
) {
2977 if (i
->ifindex
== dev
->ifindex
)
2978 packet_dev_mc(dev
, i
, what
);
2982 static int packet_mc_add(struct sock
*sk
, struct packet_mreq_max
*mreq
)
2984 struct packet_sock
*po
= pkt_sk(sk
);
2985 struct packet_mclist
*ml
, *i
;
2986 struct net_device
*dev
;
2992 dev
= __dev_get_by_index(sock_net(sk
), mreq
->mr_ifindex
);
2997 if (mreq
->mr_alen
> dev
->addr_len
)
3001 i
= kmalloc(sizeof(*i
), GFP_KERNEL
);
3006 for (ml
= po
->mclist
; ml
; ml
= ml
->next
) {
3007 if (ml
->ifindex
== mreq
->mr_ifindex
&&
3008 ml
->type
== mreq
->mr_type
&&
3009 ml
->alen
== mreq
->mr_alen
&&
3010 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
3012 /* Free the new element ... */
3018 i
->type
= mreq
->mr_type
;
3019 i
->ifindex
= mreq
->mr_ifindex
;
3020 i
->alen
= mreq
->mr_alen
;
3021 memcpy(i
->addr
, mreq
->mr_address
, i
->alen
);
3022 memset(i
->addr
+ i
->alen
, 0, sizeof(i
->addr
) - i
->alen
);
3024 i
->next
= po
->mclist
;
3026 err
= packet_dev_mc(dev
, i
, 1);
3028 po
->mclist
= i
->next
;
3037 static int packet_mc_drop(struct sock
*sk
, struct packet_mreq_max
*mreq
)
3039 struct packet_mclist
*ml
, **mlp
;
3043 for (mlp
= &pkt_sk(sk
)->mclist
; (ml
= *mlp
) != NULL
; mlp
= &ml
->next
) {
3044 if (ml
->ifindex
== mreq
->mr_ifindex
&&
3045 ml
->type
== mreq
->mr_type
&&
3046 ml
->alen
== mreq
->mr_alen
&&
3047 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
3048 if (--ml
->count
== 0) {
3049 struct net_device
*dev
;
3051 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
3053 packet_dev_mc(dev
, ml
, -1);
3061 return -EADDRNOTAVAIL
;
3064 static void packet_flush_mclist(struct sock
*sk
)
3066 struct packet_sock
*po
= pkt_sk(sk
);
3067 struct packet_mclist
*ml
;
3073 while ((ml
= po
->mclist
) != NULL
) {
3074 struct net_device
*dev
;
3076 po
->mclist
= ml
->next
;
3077 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
3079 packet_dev_mc(dev
, ml
, -1);
3086 packet_setsockopt(struct socket
*sock
, int level
, int optname
, char __user
*optval
, unsigned int optlen
)
3088 struct sock
*sk
= sock
->sk
;
3089 struct packet_sock
*po
= pkt_sk(sk
);
3092 if (level
!= SOL_PACKET
)
3093 return -ENOPROTOOPT
;
3096 case PACKET_ADD_MEMBERSHIP
:
3097 case PACKET_DROP_MEMBERSHIP
:
3099 struct packet_mreq_max mreq
;
3101 memset(&mreq
, 0, sizeof(mreq
));
3102 if (len
< sizeof(struct packet_mreq
))
3104 if (len
> sizeof(mreq
))
3106 if (copy_from_user(&mreq
, optval
, len
))
3108 if (len
< (mreq
.mr_alen
+ offsetof(struct packet_mreq
, mr_address
)))
3110 if (optname
== PACKET_ADD_MEMBERSHIP
)
3111 ret
= packet_mc_add(sk
, &mreq
);
3113 ret
= packet_mc_drop(sk
, &mreq
);
3117 case PACKET_RX_RING
:
3118 case PACKET_TX_RING
:
3120 union tpacket_req_u req_u
;
3123 switch (po
->tp_version
) {
3126 len
= sizeof(req_u
.req
);
3130 len
= sizeof(req_u
.req3
);
3135 if (pkt_sk(sk
)->has_vnet_hdr
)
3137 if (copy_from_user(&req_u
.req
, optval
, len
))
3139 return packet_set_ring(sk
, &req_u
, 0,
3140 optname
== PACKET_TX_RING
);
3142 case PACKET_COPY_THRESH
:
3146 if (optlen
!= sizeof(val
))
3148 if (copy_from_user(&val
, optval
, sizeof(val
)))
3151 pkt_sk(sk
)->copy_thresh
= val
;
3154 case PACKET_VERSION
:
3158 if (optlen
!= sizeof(val
))
3160 if (copy_from_user(&val
, optval
, sizeof(val
)))
3171 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
) {
3174 po
->tp_version
= val
;
3180 case PACKET_RESERVE
:
3184 if (optlen
!= sizeof(val
))
3186 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3188 if (copy_from_user(&val
, optval
, sizeof(val
)))
3192 po
->tp_reserve
= val
;
3199 if (optlen
!= sizeof(val
))
3201 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3203 if (copy_from_user(&val
, optval
, sizeof(val
)))
3205 po
->tp_loss
= !!val
;
3208 case PACKET_AUXDATA
:
3212 if (optlen
< sizeof(val
))
3214 if (copy_from_user(&val
, optval
, sizeof(val
)))
3217 po
->auxdata
= !!val
;
3220 case PACKET_ORIGDEV
:
3224 if (optlen
< sizeof(val
))
3226 if (copy_from_user(&val
, optval
, sizeof(val
)))
3229 po
->origdev
= !!val
;
3232 case PACKET_VNET_HDR
:
3236 if (sock
->type
!= SOCK_RAW
)
3238 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3240 if (optlen
< sizeof(val
))
3242 if (copy_from_user(&val
, optval
, sizeof(val
)))
3245 po
->has_vnet_hdr
= !!val
;
3248 case PACKET_TIMESTAMP
:
3252 if (optlen
!= sizeof(val
))
3254 if (copy_from_user(&val
, optval
, sizeof(val
)))
3257 po
->tp_tstamp
= val
;
3264 if (optlen
!= sizeof(val
))
3266 if (copy_from_user(&val
, optval
, sizeof(val
)))
3269 return fanout_add(sk
, val
& 0xffff, val
>> 16);
3271 case PACKET_TX_HAS_OFF
:
3275 if (optlen
!= sizeof(val
))
3277 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3279 if (copy_from_user(&val
, optval
, sizeof(val
)))
3281 po
->tp_tx_has_off
= !!val
;
3285 return -ENOPROTOOPT
;
3289 static int packet_getsockopt(struct socket
*sock
, int level
, int optname
,
3290 char __user
*optval
, int __user
*optlen
)
3293 int val
, lv
= sizeof(val
);
3294 struct sock
*sk
= sock
->sk
;
3295 struct packet_sock
*po
= pkt_sk(sk
);
3297 union tpacket_stats_u st
;
3299 if (level
!= SOL_PACKET
)
3300 return -ENOPROTOOPT
;
3302 if (get_user(len
, optlen
))
3309 case PACKET_STATISTICS
:
3310 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3311 memcpy(&st
, &po
->stats
, sizeof(st
));
3312 memset(&po
->stats
, 0, sizeof(po
->stats
));
3313 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3315 if (po
->tp_version
== TPACKET_V3
) {
3316 lv
= sizeof(struct tpacket_stats_v3
);
3317 st
.stats3
.tp_packets
+= st
.stats3
.tp_drops
;
3320 lv
= sizeof(struct tpacket_stats
);
3321 st
.stats1
.tp_packets
+= st
.stats1
.tp_drops
;
3326 case PACKET_AUXDATA
:
3329 case PACKET_ORIGDEV
:
3332 case PACKET_VNET_HDR
:
3333 val
= po
->has_vnet_hdr
;
3335 case PACKET_VERSION
:
3336 val
= po
->tp_version
;
3339 if (len
> sizeof(int))
3341 if (copy_from_user(&val
, optval
, len
))
3345 val
= sizeof(struct tpacket_hdr
);
3348 val
= sizeof(struct tpacket2_hdr
);
3351 val
= sizeof(struct tpacket3_hdr
);
3357 case PACKET_RESERVE
:
3358 val
= po
->tp_reserve
;
3363 case PACKET_TIMESTAMP
:
3364 val
= po
->tp_tstamp
;
3368 ((u32
)po
->fanout
->id
|
3369 ((u32
)po
->fanout
->type
<< 16) |
3370 ((u32
)po
->fanout
->flags
<< 24)) :
3373 case PACKET_TX_HAS_OFF
:
3374 val
= po
->tp_tx_has_off
;
3377 return -ENOPROTOOPT
;
3382 if (put_user(len
, optlen
))
3384 if (copy_to_user(optval
, data
, len
))
3390 static int packet_notifier(struct notifier_block
*this, unsigned long msg
, void *data
)
3393 struct net_device
*dev
= data
;
3394 struct net
*net
= dev_net(dev
);
3397 sk_for_each_rcu(sk
, &net
->packet
.sklist
) {
3398 struct packet_sock
*po
= pkt_sk(sk
);
3401 case NETDEV_UNREGISTER
:
3403 packet_dev_mclist(dev
, po
->mclist
, -1);
3407 if (dev
->ifindex
== po
->ifindex
) {
3408 spin_lock(&po
->bind_lock
);
3410 __unregister_prot_hook(sk
, false);
3411 sk
->sk_err
= ENETDOWN
;
3412 if (!sock_flag(sk
, SOCK_DEAD
))
3413 sk
->sk_error_report(sk
);
3415 if (msg
== NETDEV_UNREGISTER
) {
3416 packet_cached_dev_reset(po
);
3418 if (po
->prot_hook
.dev
)
3419 dev_put(po
->prot_hook
.dev
);
3420 po
->prot_hook
.dev
= NULL
;
3422 spin_unlock(&po
->bind_lock
);
3426 if (dev
->ifindex
== po
->ifindex
) {
3427 spin_lock(&po
->bind_lock
);
3429 register_prot_hook(sk
);
3430 spin_unlock(&po
->bind_lock
);
3440 static int packet_ioctl(struct socket
*sock
, unsigned int cmd
,
3443 struct sock
*sk
= sock
->sk
;
3448 int amount
= sk_wmem_alloc_get(sk
);
3450 return put_user(amount
, (int __user
*)arg
);
3454 struct sk_buff
*skb
;
3457 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3458 skb
= skb_peek(&sk
->sk_receive_queue
);
3461 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3462 return put_user(amount
, (int __user
*)arg
);
3465 return sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
3467 return sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
3477 case SIOCGIFBRDADDR
:
3478 case SIOCSIFBRDADDR
:
3479 case SIOCGIFNETMASK
:
3480 case SIOCSIFNETMASK
:
3481 case SIOCGIFDSTADDR
:
3482 case SIOCSIFDSTADDR
:
3484 return inet_dgram_ops
.ioctl(sock
, cmd
, arg
);
3488 return -ENOIOCTLCMD
;
3493 static unsigned int packet_poll(struct file
*file
, struct socket
*sock
,
3496 struct sock
*sk
= sock
->sk
;
3497 struct packet_sock
*po
= pkt_sk(sk
);
3498 unsigned int mask
= datagram_poll(file
, sock
, wait
);
3500 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3501 if (po
->rx_ring
.pg_vec
) {
3502 if (!packet_previous_rx_frame(po
, &po
->rx_ring
,
3504 mask
|= POLLIN
| POLLRDNORM
;
3506 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3507 spin_lock_bh(&sk
->sk_write_queue
.lock
);
3508 if (po
->tx_ring
.pg_vec
) {
3509 if (packet_current_frame(po
, &po
->tx_ring
, TP_STATUS_AVAILABLE
))
3510 mask
|= POLLOUT
| POLLWRNORM
;
3512 spin_unlock_bh(&sk
->sk_write_queue
.lock
);
3517 /* Dirty? Well, I still did not learn better way to account
3521 static void packet_mm_open(struct vm_area_struct
*vma
)
3523 struct file
*file
= vma
->vm_file
;
3524 struct socket
*sock
= file
->private_data
;
3525 struct sock
*sk
= sock
->sk
;
3528 atomic_inc(&pkt_sk(sk
)->mapped
);
3531 static void packet_mm_close(struct vm_area_struct
*vma
)
3533 struct file
*file
= vma
->vm_file
;
3534 struct socket
*sock
= file
->private_data
;
3535 struct sock
*sk
= sock
->sk
;
3538 atomic_dec(&pkt_sk(sk
)->mapped
);
3541 static const struct vm_operations_struct packet_mmap_ops
= {
3542 .open
= packet_mm_open
,
3543 .close
= packet_mm_close
,
3546 static void free_pg_vec(struct pgv
*pg_vec
, unsigned int order
,
3551 for (i
= 0; i
< len
; i
++) {
3552 if (likely(pg_vec
[i
].buffer
)) {
3553 if (is_vmalloc_addr(pg_vec
[i
].buffer
))
3554 vfree(pg_vec
[i
].buffer
);
3556 free_pages((unsigned long)pg_vec
[i
].buffer
,
3558 pg_vec
[i
].buffer
= NULL
;
3564 static char *alloc_one_pg_vec_page(unsigned long order
)
3566 char *buffer
= NULL
;
3567 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_COMP
|
3568 __GFP_ZERO
| __GFP_NOWARN
| __GFP_NORETRY
;
3570 buffer
= (char *) __get_free_pages(gfp_flags
, order
);
3576 * __get_free_pages failed, fall back to vmalloc
3578 buffer
= vzalloc((1 << order
) * PAGE_SIZE
);
3584 * vmalloc failed, lets dig into swap here
3586 gfp_flags
&= ~__GFP_NORETRY
;
3587 buffer
= (char *)__get_free_pages(gfp_flags
, order
);
3592 * complete and utter failure
3597 static struct pgv
*alloc_pg_vec(struct tpacket_req
*req
, int order
)
3599 unsigned int block_nr
= req
->tp_block_nr
;
3603 pg_vec
= kcalloc(block_nr
, sizeof(struct pgv
), GFP_KERNEL
);
3604 if (unlikely(!pg_vec
))
3607 for (i
= 0; i
< block_nr
; i
++) {
3608 pg_vec
[i
].buffer
= alloc_one_pg_vec_page(order
);
3609 if (unlikely(!pg_vec
[i
].buffer
))
3610 goto out_free_pgvec
;
3617 free_pg_vec(pg_vec
, order
, block_nr
);
3622 static int packet_set_ring(struct sock
*sk
, union tpacket_req_u
*req_u
,
3623 int closing
, int tx_ring
)
3625 struct pgv
*pg_vec
= NULL
;
3626 struct packet_sock
*po
= pkt_sk(sk
);
3627 int was_running
, order
= 0;
3628 struct packet_ring_buffer
*rb
;
3629 struct sk_buff_head
*rb_queue
;
3632 /* Added to avoid minimal code churn */
3633 struct tpacket_req
*req
= &req_u
->req
;
3636 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3637 if (!closing
&& tx_ring
&& (po
->tp_version
> TPACKET_V2
)) {
3638 WARN(1, "Tx-ring is not supported.\n");
3642 rb
= tx_ring
? &po
->tx_ring
: &po
->rx_ring
;
3643 rb_queue
= tx_ring
? &sk
->sk_write_queue
: &sk
->sk_receive_queue
;
3647 if (atomic_read(&po
->mapped
))
3649 if (atomic_read(&rb
->pending
))
3653 if (req
->tp_block_nr
) {
3654 /* Sanity tests and some calculations */
3656 if (unlikely(rb
->pg_vec
))
3659 switch (po
->tp_version
) {
3661 po
->tp_hdrlen
= TPACKET_HDRLEN
;
3664 po
->tp_hdrlen
= TPACKET2_HDRLEN
;
3667 po
->tp_hdrlen
= TPACKET3_HDRLEN
;
3672 if (unlikely((int)req
->tp_block_size
<= 0))
3674 if (unlikely(req
->tp_block_size
& (PAGE_SIZE
- 1)))
3676 if (po
->tp_version
>= TPACKET_V3
&&
3677 req
->tp_block_size
<=
3678 BLK_PLUS_PRIV((u64
)req_u
->req3
.tp_sizeof_priv
))
3680 if (unlikely(req
->tp_frame_size
< po
->tp_hdrlen
+
3683 if (unlikely(req
->tp_frame_size
& (TPACKET_ALIGNMENT
- 1)))
3686 rb
->frames_per_block
= req
->tp_block_size
/req
->tp_frame_size
;
3687 if (unlikely(rb
->frames_per_block
<= 0))
3689 if (unlikely(req
->tp_block_size
> UINT_MAX
/ req
->tp_block_nr
))
3691 if (unlikely((rb
->frames_per_block
* req
->tp_block_nr
) !=
3696 order
= get_order(req
->tp_block_size
);
3697 pg_vec
= alloc_pg_vec(req
, order
);
3698 if (unlikely(!pg_vec
))
3700 switch (po
->tp_version
) {
3702 /* Transmit path is not supported. We checked
3703 * it above but just being paranoid
3706 init_prb_bdqc(po
, rb
, pg_vec
, req_u
, tx_ring
);
3715 if (unlikely(req
->tp_frame_nr
))
3720 /* Detach socket from network */
3721 spin_lock(&po
->bind_lock
);
3722 was_running
= po
->running
;
3726 __unregister_prot_hook(sk
, false);
3728 spin_unlock(&po
->bind_lock
);
3733 mutex_lock(&po
->pg_vec_lock
);
3734 if (closing
|| atomic_read(&po
->mapped
) == 0) {
3736 spin_lock_bh(&rb_queue
->lock
);
3737 swap(rb
->pg_vec
, pg_vec
);
3738 rb
->frame_max
= (req
->tp_frame_nr
- 1);
3740 rb
->frame_size
= req
->tp_frame_size
;
3741 spin_unlock_bh(&rb_queue
->lock
);
3743 swap(rb
->pg_vec_order
, order
);
3744 swap(rb
->pg_vec_len
, req
->tp_block_nr
);
3746 rb
->pg_vec_pages
= req
->tp_block_size
/PAGE_SIZE
;
3747 po
->prot_hook
.func
= (po
->rx_ring
.pg_vec
) ?
3748 tpacket_rcv
: packet_rcv
;
3749 skb_queue_purge(rb_queue
);
3750 if (atomic_read(&po
->mapped
))
3751 pr_err("packet_mmap: vma is busy: %d\n",
3752 atomic_read(&po
->mapped
));
3754 mutex_unlock(&po
->pg_vec_lock
);
3756 spin_lock(&po
->bind_lock
);
3759 register_prot_hook(sk
);
3761 spin_unlock(&po
->bind_lock
);
3762 if (closing
&& (po
->tp_version
> TPACKET_V2
)) {
3763 /* Because we don't support block-based V3 on tx-ring */
3765 prb_shutdown_retire_blk_timer(po
, tx_ring
, rb_queue
);
3769 free_pg_vec(pg_vec
, order
, req
->tp_block_nr
);
3775 static int packet_mmap(struct file
*file
, struct socket
*sock
,
3776 struct vm_area_struct
*vma
)
3778 struct sock
*sk
= sock
->sk
;
3779 struct packet_sock
*po
= pkt_sk(sk
);
3780 unsigned long size
, expected_size
;
3781 struct packet_ring_buffer
*rb
;
3782 unsigned long start
;
3789 mutex_lock(&po
->pg_vec_lock
);
3792 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
3794 expected_size
+= rb
->pg_vec_len
3800 if (expected_size
== 0)
3803 size
= vma
->vm_end
- vma
->vm_start
;
3804 if (size
!= expected_size
)
3807 start
= vma
->vm_start
;
3808 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
3809 if (rb
->pg_vec
== NULL
)
3812 for (i
= 0; i
< rb
->pg_vec_len
; i
++) {
3814 void *kaddr
= rb
->pg_vec
[i
].buffer
;
3817 for (pg_num
= 0; pg_num
< rb
->pg_vec_pages
; pg_num
++) {
3818 page
= pgv_to_page(kaddr
);
3819 err
= vm_insert_page(vma
, start
, page
);
3828 atomic_inc(&po
->mapped
);
3829 vma
->vm_ops
= &packet_mmap_ops
;
3833 mutex_unlock(&po
->pg_vec_lock
);
3837 static const struct proto_ops packet_ops_spkt
= {
3838 .family
= PF_PACKET
,
3839 .owner
= THIS_MODULE
,
3840 .release
= packet_release
,
3841 .bind
= packet_bind_spkt
,
3842 .connect
= sock_no_connect
,
3843 .socketpair
= sock_no_socketpair
,
3844 .accept
= sock_no_accept
,
3845 .getname
= packet_getname_spkt
,
3846 .poll
= datagram_poll
,
3847 .ioctl
= packet_ioctl
,
3848 .listen
= sock_no_listen
,
3849 .shutdown
= sock_no_shutdown
,
3850 .setsockopt
= sock_no_setsockopt
,
3851 .getsockopt
= sock_no_getsockopt
,
3852 .sendmsg
= packet_sendmsg_spkt
,
3853 .recvmsg
= packet_recvmsg
,
3854 .mmap
= sock_no_mmap
,
3855 .sendpage
= sock_no_sendpage
,
3858 static const struct proto_ops packet_ops
= {
3859 .family
= PF_PACKET
,
3860 .owner
= THIS_MODULE
,
3861 .release
= packet_release
,
3862 .bind
= packet_bind
,
3863 .connect
= sock_no_connect
,
3864 .socketpair
= sock_no_socketpair
,
3865 .accept
= sock_no_accept
,
3866 .getname
= packet_getname
,
3867 .poll
= packet_poll
,
3868 .ioctl
= packet_ioctl
,
3869 .listen
= sock_no_listen
,
3870 .shutdown
= sock_no_shutdown
,
3871 .setsockopt
= packet_setsockopt
,
3872 .getsockopt
= packet_getsockopt
,
3873 .sendmsg
= packet_sendmsg
,
3874 .recvmsg
= packet_recvmsg
,
3875 .mmap
= packet_mmap
,
3876 .sendpage
= sock_no_sendpage
,
3879 static const struct net_proto_family packet_family_ops
= {
3880 .family
= PF_PACKET
,
3881 .create
= packet_create
,
3882 .owner
= THIS_MODULE
,
3885 static struct notifier_block packet_netdev_notifier
= {
3886 .notifier_call
= packet_notifier
,
3889 #ifdef CONFIG_PROC_FS
3891 static void *packet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3894 struct net
*net
= seq_file_net(seq
);
3897 return seq_hlist_start_head_rcu(&net
->packet
.sklist
, *pos
);
3900 static void *packet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3902 struct net
*net
= seq_file_net(seq
);
3903 return seq_hlist_next_rcu(v
, &net
->packet
.sklist
, pos
);
3906 static void packet_seq_stop(struct seq_file
*seq
, void *v
)
3912 static int packet_seq_show(struct seq_file
*seq
, void *v
)
3914 if (v
== SEQ_START_TOKEN
)
3915 seq_puts(seq
, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3917 struct sock
*s
= sk_entry(v
);
3918 const struct packet_sock
*po
= pkt_sk(s
);
3921 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3923 atomic_read(&s
->sk_refcnt
),
3928 atomic_read(&s
->sk_rmem_alloc
),
3929 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(s
)),
3936 static const struct seq_operations packet_seq_ops
= {
3937 .start
= packet_seq_start
,
3938 .next
= packet_seq_next
,
3939 .stop
= packet_seq_stop
,
3940 .show
= packet_seq_show
,
3943 static int packet_seq_open(struct inode
*inode
, struct file
*file
)
3945 return seq_open_net(inode
, file
, &packet_seq_ops
,
3946 sizeof(struct seq_net_private
));
3949 static const struct file_operations packet_seq_fops
= {
3950 .owner
= THIS_MODULE
,
3951 .open
= packet_seq_open
,
3953 .llseek
= seq_lseek
,
3954 .release
= seq_release_net
,
3959 static int __net_init
packet_net_init(struct net
*net
)
3961 mutex_init(&net
->packet
.sklist_lock
);
3962 INIT_HLIST_HEAD(&net
->packet
.sklist
);
3964 if (!proc_create("packet", 0, net
->proc_net
, &packet_seq_fops
))
3970 static void __net_exit
packet_net_exit(struct net
*net
)
3972 remove_proc_entry("packet", net
->proc_net
);
3975 static struct pernet_operations packet_net_ops
= {
3976 .init
= packet_net_init
,
3977 .exit
= packet_net_exit
,
3981 static void __exit
packet_exit(void)
3983 unregister_netdevice_notifier(&packet_netdev_notifier
);
3984 unregister_pernet_subsys(&packet_net_ops
);
3985 sock_unregister(PF_PACKET
);
3986 proto_unregister(&packet_proto
);
3989 static int __init
packet_init(void)
3991 int rc
= proto_register(&packet_proto
, 0);
3996 sock_register(&packet_family_ops
);
3997 register_pernet_subsys(&packet_net_ops
);
3998 register_netdevice_notifier(&packet_netdev_notifier
);
4003 module_init(packet_init
);
4004 module_exit(packet_exit
);
4005 MODULE_LICENSE("GPL");
4006 MODULE_ALIAS_NETPROTO(PF_PACKET
);