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 <linux/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
95 #include <linux/bpf.h>
96 #include <net/compat.h>
102 - if device has no dev->hard_header routine, it adds and removes ll header
103 inside itself. In this case ll header is invisible outside of device,
104 but higher levels still should reserve dev->hard_header_len.
105 Some devices are enough clever to reallocate skb, when header
106 will not fit to reserved space (tunnel), another ones are silly
108 - packet socket receives packets with pulled ll header,
109 so that SOCK_RAW should push it back.
114 Incoming, dev->hard_header!=NULL
115 mac_header -> ll header
118 Outgoing, dev->hard_header!=NULL
119 mac_header -> ll header
122 Incoming, dev->hard_header==NULL
123 mac_header -> UNKNOWN position. It is very likely, that it points to ll
124 header. PPP makes it, that is wrong, because introduce
125 assymetry between rx and tx paths.
128 Outgoing, dev->hard_header==NULL
129 mac_header -> data. ll header is still not built!
133 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
139 dev->hard_header != NULL
140 mac_header -> ll header
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
147 We should set nh.raw on output to correct posistion,
148 packet classifier depends on it.
151 /* Private packet socket structures. */
153 /* identical to struct packet_mreq except it has
154 * a longer address field.
156 struct packet_mreq_max
{
158 unsigned short mr_type
;
159 unsigned short mr_alen
;
160 unsigned char mr_address
[MAX_ADDR_LEN
];
164 struct tpacket_hdr
*h1
;
165 struct tpacket2_hdr
*h2
;
166 struct tpacket3_hdr
*h3
;
170 static int packet_set_ring(struct sock
*sk
, union tpacket_req_u
*req_u
,
171 int closing
, int tx_ring
);
173 #define V3_ALIGNMENT (8)
175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
180 #define PGV_FROM_VMALLOC 1
182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
183 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
184 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
185 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
186 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
188 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
191 static int tpacket_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
192 struct packet_type
*pt
, struct net_device
*orig_dev
);
194 static void *packet_previous_frame(struct packet_sock
*po
,
195 struct packet_ring_buffer
*rb
,
197 static void packet_increment_head(struct packet_ring_buffer
*buff
);
198 static int prb_curr_blk_in_use(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
);
217 static void packet_pick_tx_queue(struct net_device
*dev
, struct sk_buff
*skb
);
219 struct packet_skb_cb
{
221 struct sockaddr_pkt pkt
;
223 /* Trick: alias skb original length with
224 * ll.sll_family and ll.protocol in order
227 unsigned int origlen
;
228 struct sockaddr_ll ll
;
233 #define vio_le() virtio_legacy_is_little_endian()
235 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
237 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
238 #define GET_PBLOCK_DESC(x, bid) \
239 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
241 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
242 #define GET_NEXT_PRB_BLK_NUM(x) \
243 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
244 ((x)->kactive_blk_num+1) : 0)
246 static void __fanout_unlink(struct sock
*sk
, struct packet_sock
*po
);
247 static void __fanout_link(struct sock
*sk
, struct packet_sock
*po
);
249 static int packet_direct_xmit(struct sk_buff
*skb
)
251 struct net_device
*dev
= skb
->dev
;
252 struct sk_buff
*orig_skb
= skb
;
253 struct netdev_queue
*txq
;
254 int ret
= NETDEV_TX_BUSY
;
256 if (unlikely(!netif_running(dev
) ||
257 !netif_carrier_ok(dev
)))
260 skb
= validate_xmit_skb_list(skb
, dev
);
264 packet_pick_tx_queue(dev
, skb
);
265 txq
= skb_get_tx_queue(dev
, skb
);
269 HARD_TX_LOCK(dev
, txq
, smp_processor_id());
270 if (!netif_xmit_frozen_or_drv_stopped(txq
))
271 ret
= netdev_start_xmit(skb
, dev
, txq
, false);
272 HARD_TX_UNLOCK(dev
, txq
);
276 if (!dev_xmit_complete(ret
))
281 atomic_long_inc(&dev
->tx_dropped
);
283 return NET_XMIT_DROP
;
286 static struct net_device
*packet_cached_dev_get(struct packet_sock
*po
)
288 struct net_device
*dev
;
291 dev
= rcu_dereference(po
->cached_dev
);
299 static void packet_cached_dev_assign(struct packet_sock
*po
,
300 struct net_device
*dev
)
302 rcu_assign_pointer(po
->cached_dev
, dev
);
305 static void packet_cached_dev_reset(struct packet_sock
*po
)
307 RCU_INIT_POINTER(po
->cached_dev
, NULL
);
310 static bool packet_use_direct_xmit(const struct packet_sock
*po
)
312 return po
->xmit
== packet_direct_xmit
;
315 static u16
__packet_pick_tx_queue(struct net_device
*dev
, struct sk_buff
*skb
)
317 return (u16
) raw_smp_processor_id() % dev
->real_num_tx_queues
;
320 static void packet_pick_tx_queue(struct net_device
*dev
, struct sk_buff
*skb
)
322 const struct net_device_ops
*ops
= dev
->netdev_ops
;
325 if (ops
->ndo_select_queue
) {
326 queue_index
= ops
->ndo_select_queue(dev
, skb
, NULL
,
327 __packet_pick_tx_queue
);
328 queue_index
= netdev_cap_txqueue(dev
, queue_index
);
330 queue_index
= __packet_pick_tx_queue(dev
, skb
);
333 skb_set_queue_mapping(skb
, queue_index
);
336 /* register_prot_hook must be invoked with the po->bind_lock held,
337 * or from a context in which asynchronous accesses to the packet
338 * socket is not possible (packet_create()).
340 static void register_prot_hook(struct sock
*sk
)
342 struct packet_sock
*po
= pkt_sk(sk
);
346 __fanout_link(sk
, po
);
348 dev_add_pack(&po
->prot_hook
);
355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356 * held. If the sync parameter is true, we will temporarily drop
357 * the po->bind_lock and do a synchronize_net to make sure no
358 * asynchronous packet processing paths still refer to the elements
359 * of po->prot_hook. If the sync parameter is false, it is the
360 * callers responsibility to take care of this.
362 static void __unregister_prot_hook(struct sock
*sk
, bool sync
)
364 struct packet_sock
*po
= pkt_sk(sk
);
369 __fanout_unlink(sk
, po
);
371 __dev_remove_pack(&po
->prot_hook
);
376 spin_unlock(&po
->bind_lock
);
378 spin_lock(&po
->bind_lock
);
382 static void unregister_prot_hook(struct sock
*sk
, bool sync
)
384 struct packet_sock
*po
= pkt_sk(sk
);
387 __unregister_prot_hook(sk
, sync
);
390 static inline struct page
* __pure
pgv_to_page(void *addr
)
392 if (is_vmalloc_addr(addr
))
393 return vmalloc_to_page(addr
);
394 return virt_to_page(addr
);
397 static void __packet_set_status(struct packet_sock
*po
, void *frame
, int status
)
399 union tpacket_uhdr h
;
402 switch (po
->tp_version
) {
404 h
.h1
->tp_status
= status
;
405 flush_dcache_page(pgv_to_page(&h
.h1
->tp_status
));
408 h
.h2
->tp_status
= status
;
409 flush_dcache_page(pgv_to_page(&h
.h2
->tp_status
));
412 h
.h3
->tp_status
= status
;
413 flush_dcache_page(pgv_to_page(&h
.h3
->tp_status
));
416 WARN(1, "TPACKET version not supported.\n");
423 static int __packet_get_status(struct packet_sock
*po
, void *frame
)
425 union tpacket_uhdr h
;
430 switch (po
->tp_version
) {
432 flush_dcache_page(pgv_to_page(&h
.h1
->tp_status
));
433 return h
.h1
->tp_status
;
435 flush_dcache_page(pgv_to_page(&h
.h2
->tp_status
));
436 return h
.h2
->tp_status
;
438 flush_dcache_page(pgv_to_page(&h
.h3
->tp_status
));
439 return h
.h3
->tp_status
;
441 WARN(1, "TPACKET version not supported.\n");
447 static __u32
tpacket_get_timestamp(struct sk_buff
*skb
, struct timespec
*ts
,
450 struct skb_shared_hwtstamps
*shhwtstamps
= skb_hwtstamps(skb
);
453 (flags
& SOF_TIMESTAMPING_RAW_HARDWARE
) &&
454 ktime_to_timespec_cond(shhwtstamps
->hwtstamp
, ts
))
455 return TP_STATUS_TS_RAW_HARDWARE
;
457 if (ktime_to_timespec_cond(skb
->tstamp
, ts
))
458 return TP_STATUS_TS_SOFTWARE
;
463 static __u32
__packet_set_timestamp(struct packet_sock
*po
, void *frame
,
466 union tpacket_uhdr h
;
470 if (!(ts_status
= tpacket_get_timestamp(skb
, &ts
, po
->tp_tstamp
)))
474 switch (po
->tp_version
) {
476 h
.h1
->tp_sec
= ts
.tv_sec
;
477 h
.h1
->tp_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
480 h
.h2
->tp_sec
= ts
.tv_sec
;
481 h
.h2
->tp_nsec
= ts
.tv_nsec
;
484 h
.h3
->tp_sec
= ts
.tv_sec
;
485 h
.h3
->tp_nsec
= ts
.tv_nsec
;
488 WARN(1, "TPACKET version not supported.\n");
492 /* one flush is safe, as both fields always lie on the same cacheline */
493 flush_dcache_page(pgv_to_page(&h
.h1
->tp_sec
));
499 static void *packet_lookup_frame(struct packet_sock
*po
,
500 struct packet_ring_buffer
*rb
,
501 unsigned int position
,
504 unsigned int pg_vec_pos
, frame_offset
;
505 union tpacket_uhdr h
;
507 pg_vec_pos
= position
/ rb
->frames_per_block
;
508 frame_offset
= position
% rb
->frames_per_block
;
510 h
.raw
= rb
->pg_vec
[pg_vec_pos
].buffer
+
511 (frame_offset
* rb
->frame_size
);
513 if (status
!= __packet_get_status(po
, h
.raw
))
519 static void *packet_current_frame(struct packet_sock
*po
,
520 struct packet_ring_buffer
*rb
,
523 return packet_lookup_frame(po
, rb
, rb
->head
, status
);
526 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core
*pkc
)
528 del_timer_sync(&pkc
->retire_blk_timer
);
531 static void prb_shutdown_retire_blk_timer(struct packet_sock
*po
,
532 struct sk_buff_head
*rb_queue
)
534 struct tpacket_kbdq_core
*pkc
;
536 pkc
= GET_PBDQC_FROM_RB(&po
->rx_ring
);
538 spin_lock_bh(&rb_queue
->lock
);
539 pkc
->delete_blk_timer
= 1;
540 spin_unlock_bh(&rb_queue
->lock
);
542 prb_del_retire_blk_timer(pkc
);
545 static void prb_init_blk_timer(struct packet_sock
*po
,
546 struct tpacket_kbdq_core
*pkc
,
547 void (*func
) (unsigned long))
549 init_timer(&pkc
->retire_blk_timer
);
550 pkc
->retire_blk_timer
.data
= (long)po
;
551 pkc
->retire_blk_timer
.function
= func
;
552 pkc
->retire_blk_timer
.expires
= jiffies
;
555 static void prb_setup_retire_blk_timer(struct packet_sock
*po
)
557 struct tpacket_kbdq_core
*pkc
;
559 pkc
= GET_PBDQC_FROM_RB(&po
->rx_ring
);
560 prb_init_blk_timer(po
, pkc
, prb_retire_rx_blk_timer_expired
);
563 static int prb_calc_retire_blk_tmo(struct packet_sock
*po
,
564 int blk_size_in_bytes
)
566 struct net_device
*dev
;
567 unsigned int mbits
= 0, msec
= 0, div
= 0, tmo
= 0;
568 struct ethtool_link_ksettings ecmd
;
572 dev
= __dev_get_by_index(sock_net(&po
->sk
), po
->ifindex
);
573 if (unlikely(!dev
)) {
575 return DEFAULT_PRB_RETIRE_TOV
;
577 err
= __ethtool_get_link_ksettings(dev
, &ecmd
);
581 * If the link speed is so slow you don't really
582 * need to worry about perf anyways
584 if (ecmd
.base
.speed
< SPEED_1000
||
585 ecmd
.base
.speed
== SPEED_UNKNOWN
) {
586 return DEFAULT_PRB_RETIRE_TOV
;
589 div
= ecmd
.base
.speed
/ 1000;
593 mbits
= (blk_size_in_bytes
* 8) / (1024 * 1024);
605 static void prb_init_ft_ops(struct tpacket_kbdq_core
*p1
,
606 union tpacket_req_u
*req_u
)
608 p1
->feature_req_word
= req_u
->req3
.tp_feature_req_word
;
611 static void init_prb_bdqc(struct packet_sock
*po
,
612 struct packet_ring_buffer
*rb
,
614 union tpacket_req_u
*req_u
)
616 struct tpacket_kbdq_core
*p1
= GET_PBDQC_FROM_RB(rb
);
617 struct tpacket_block_desc
*pbd
;
619 memset(p1
, 0x0, sizeof(*p1
));
621 p1
->knxt_seq_num
= 1;
623 pbd
= (struct tpacket_block_desc
*)pg_vec
[0].buffer
;
624 p1
->pkblk_start
= pg_vec
[0].buffer
;
625 p1
->kblk_size
= req_u
->req3
.tp_block_size
;
626 p1
->knum_blocks
= req_u
->req3
.tp_block_nr
;
627 p1
->hdrlen
= po
->tp_hdrlen
;
628 p1
->version
= po
->tp_version
;
629 p1
->last_kactive_blk_num
= 0;
630 po
->stats
.stats3
.tp_freeze_q_cnt
= 0;
631 if (req_u
->req3
.tp_retire_blk_tov
)
632 p1
->retire_blk_tov
= req_u
->req3
.tp_retire_blk_tov
;
634 p1
->retire_blk_tov
= prb_calc_retire_blk_tmo(po
,
635 req_u
->req3
.tp_block_size
);
636 p1
->tov_in_jiffies
= msecs_to_jiffies(p1
->retire_blk_tov
);
637 p1
->blk_sizeof_priv
= req_u
->req3
.tp_sizeof_priv
;
639 p1
->max_frame_len
= p1
->kblk_size
- BLK_PLUS_PRIV(p1
->blk_sizeof_priv
);
640 prb_init_ft_ops(p1
, req_u
);
641 prb_setup_retire_blk_timer(po
);
642 prb_open_block(p1
, pbd
);
645 /* Do NOT update the last_blk_num first.
646 * Assumes sk_buff_head lock is held.
648 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core
*pkc
)
650 mod_timer(&pkc
->retire_blk_timer
,
651 jiffies
+ pkc
->tov_in_jiffies
);
652 pkc
->last_kactive_blk_num
= pkc
->kactive_blk_num
;
657 * 1) We refresh the timer only when we open a block.
658 * By doing this we don't waste cycles refreshing the timer
659 * on packet-by-packet basis.
661 * With a 1MB block-size, on a 1Gbps line, it will take
662 * i) ~8 ms to fill a block + ii) memcpy etc.
663 * In this cut we are not accounting for the memcpy time.
665 * So, if the user sets the 'tmo' to 10ms then the timer
666 * will never fire while the block is still getting filled
667 * (which is what we want). However, the user could choose
668 * to close a block early and that's fine.
670 * But when the timer does fire, we check whether or not to refresh it.
671 * Since the tmo granularity is in msecs, it is not too expensive
672 * to refresh the timer, lets say every '8' msecs.
673 * Either the user can set the 'tmo' or we can derive it based on
674 * a) line-speed and b) block-size.
675 * prb_calc_retire_blk_tmo() calculates the tmo.
678 static void prb_retire_rx_blk_timer_expired(unsigned long data
)
680 struct packet_sock
*po
= (struct packet_sock
*)data
;
681 struct tpacket_kbdq_core
*pkc
= GET_PBDQC_FROM_RB(&po
->rx_ring
);
683 struct tpacket_block_desc
*pbd
;
685 spin_lock(&po
->sk
.sk_receive_queue
.lock
);
687 frozen
= prb_queue_frozen(pkc
);
688 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
690 if (unlikely(pkc
->delete_blk_timer
))
693 /* We only need to plug the race when the block is partially filled.
695 * lock(); increment BLOCK_NUM_PKTS; unlock()
696 * copy_bits() is in progress ...
697 * timer fires on other cpu:
698 * we can't retire the current block because copy_bits
702 if (BLOCK_NUM_PKTS(pbd
)) {
703 while (atomic_read(&pkc
->blk_fill_in_prog
)) {
704 /* Waiting for skb_copy_bits to finish... */
709 if (pkc
->last_kactive_blk_num
== pkc
->kactive_blk_num
) {
711 if (!BLOCK_NUM_PKTS(pbd
)) {
712 /* An empty block. Just refresh the timer. */
715 prb_retire_current_block(pkc
, po
, TP_STATUS_BLK_TMO
);
716 if (!prb_dispatch_next_block(pkc
, po
))
721 /* Case 1. Queue was frozen because user-space was
724 if (prb_curr_blk_in_use(pbd
)) {
726 * Ok, user-space is still behind.
727 * So just refresh the timer.
731 /* Case 2. queue was frozen,user-space caught up,
732 * now the link went idle && the timer fired.
733 * We don't have a block to close.So we open this
734 * block and restart the timer.
735 * opening a block thaws the queue,restarts timer
736 * Thawing/timer-refresh is a side effect.
738 prb_open_block(pkc
, pbd
);
745 _prb_refresh_rx_retire_blk_timer(pkc
);
748 spin_unlock(&po
->sk
.sk_receive_queue
.lock
);
751 static void prb_flush_block(struct tpacket_kbdq_core
*pkc1
,
752 struct tpacket_block_desc
*pbd1
, __u32 status
)
754 /* Flush everything minus the block header */
756 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
761 /* Skip the block header(we know header WILL fit in 4K) */
764 end
= (u8
*)PAGE_ALIGN((unsigned long)pkc1
->pkblk_end
);
765 for (; start
< end
; start
+= PAGE_SIZE
)
766 flush_dcache_page(pgv_to_page(start
));
771 /* Now update the block status. */
773 BLOCK_STATUS(pbd1
) = status
;
775 /* Flush the block header */
777 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
779 flush_dcache_page(pgv_to_page(start
));
789 * 2) Increment active_blk_num
791 * Note:We DONT refresh the timer on purpose.
792 * Because almost always the next block will be opened.
794 static void prb_close_block(struct tpacket_kbdq_core
*pkc1
,
795 struct tpacket_block_desc
*pbd1
,
796 struct packet_sock
*po
, unsigned int stat
)
798 __u32 status
= TP_STATUS_USER
| stat
;
800 struct tpacket3_hdr
*last_pkt
;
801 struct tpacket_hdr_v1
*h1
= &pbd1
->hdr
.bh1
;
802 struct sock
*sk
= &po
->sk
;
804 if (po
->stats
.stats3
.tp_drops
)
805 status
|= TP_STATUS_LOSING
;
807 last_pkt
= (struct tpacket3_hdr
*)pkc1
->prev
;
808 last_pkt
->tp_next_offset
= 0;
810 /* Get the ts of the last pkt */
811 if (BLOCK_NUM_PKTS(pbd1
)) {
812 h1
->ts_last_pkt
.ts_sec
= last_pkt
->tp_sec
;
813 h1
->ts_last_pkt
.ts_nsec
= last_pkt
->tp_nsec
;
815 /* Ok, we tmo'd - so get the current time.
817 * It shouldn't really happen as we don't close empty
818 * blocks. See prb_retire_rx_blk_timer_expired().
822 h1
->ts_last_pkt
.ts_sec
= ts
.tv_sec
;
823 h1
->ts_last_pkt
.ts_nsec
= ts
.tv_nsec
;
828 /* Flush the block */
829 prb_flush_block(pkc1
, pbd1
, status
);
831 sk
->sk_data_ready(sk
);
833 pkc1
->kactive_blk_num
= GET_NEXT_PRB_BLK_NUM(pkc1
);
836 static void prb_thaw_queue(struct tpacket_kbdq_core
*pkc
)
838 pkc
->reset_pending_on_curr_blk
= 0;
842 * Side effect of opening a block:
844 * 1) prb_queue is thawed.
845 * 2) retire_blk_timer is refreshed.
848 static void prb_open_block(struct tpacket_kbdq_core
*pkc1
,
849 struct tpacket_block_desc
*pbd1
)
852 struct tpacket_hdr_v1
*h1
= &pbd1
->hdr
.bh1
;
856 /* We could have just memset this but we will lose the
857 * flexibility of making the priv area sticky
860 BLOCK_SNUM(pbd1
) = pkc1
->knxt_seq_num
++;
861 BLOCK_NUM_PKTS(pbd1
) = 0;
862 BLOCK_LEN(pbd1
) = BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
866 h1
->ts_first_pkt
.ts_sec
= ts
.tv_sec
;
867 h1
->ts_first_pkt
.ts_nsec
= ts
.tv_nsec
;
869 pkc1
->pkblk_start
= (char *)pbd1
;
870 pkc1
->nxt_offset
= pkc1
->pkblk_start
+ BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
872 BLOCK_O2FP(pbd1
) = (__u32
)BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
873 BLOCK_O2PRIV(pbd1
) = BLK_HDR_LEN
;
875 pbd1
->version
= pkc1
->version
;
876 pkc1
->prev
= pkc1
->nxt_offset
;
877 pkc1
->pkblk_end
= pkc1
->pkblk_start
+ pkc1
->kblk_size
;
879 prb_thaw_queue(pkc1
);
880 _prb_refresh_rx_retire_blk_timer(pkc1
);
886 * Queue freeze logic:
887 * 1) Assume tp_block_nr = 8 blocks.
888 * 2) At time 't0', user opens Rx ring.
889 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
890 * 4) user-space is either sleeping or processing block '0'.
891 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
892 * it will close block-7,loop around and try to fill block '0'.
894 * __packet_lookup_frame_in_block
895 * prb_retire_current_block()
896 * prb_dispatch_next_block()
897 * |->(BLOCK_STATUS == USER) evaluates to true
898 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
899 * 6) Now there are two cases:
900 * 6.1) Link goes idle right after the queue is frozen.
901 * But remember, the last open_block() refreshed the timer.
902 * When this timer expires,it will refresh itself so that we can
903 * re-open block-0 in near future.
904 * 6.2) Link is busy and keeps on receiving packets. This is a simple
905 * case and __packet_lookup_frame_in_block will check if block-0
906 * is free and can now be re-used.
908 static void prb_freeze_queue(struct tpacket_kbdq_core
*pkc
,
909 struct packet_sock
*po
)
911 pkc
->reset_pending_on_curr_blk
= 1;
912 po
->stats
.stats3
.tp_freeze_q_cnt
++;
915 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
918 * If the next block is free then we will dispatch it
919 * and return a good offset.
920 * Else, we will freeze the queue.
921 * So, caller must check the return value.
923 static void *prb_dispatch_next_block(struct tpacket_kbdq_core
*pkc
,
924 struct packet_sock
*po
)
926 struct tpacket_block_desc
*pbd
;
930 /* 1. Get current block num */
931 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
933 /* 2. If this block is currently in_use then freeze the queue */
934 if (TP_STATUS_USER
& BLOCK_STATUS(pbd
)) {
935 prb_freeze_queue(pkc
, po
);
941 * open this block and return the offset where the first packet
942 * needs to get stored.
944 prb_open_block(pkc
, pbd
);
945 return (void *)pkc
->nxt_offset
;
948 static void prb_retire_current_block(struct tpacket_kbdq_core
*pkc
,
949 struct packet_sock
*po
, unsigned int status
)
951 struct tpacket_block_desc
*pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
953 /* retire/close the current block */
954 if (likely(TP_STATUS_KERNEL
== BLOCK_STATUS(pbd
))) {
956 * Plug the case where copy_bits() is in progress on
957 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
958 * have space to copy the pkt in the current block and
959 * called prb_retire_current_block()
961 * We don't need to worry about the TMO case because
962 * the timer-handler already handled this case.
964 if (!(status
& TP_STATUS_BLK_TMO
)) {
965 while (atomic_read(&pkc
->blk_fill_in_prog
)) {
966 /* Waiting for skb_copy_bits to finish... */
970 prb_close_block(pkc
, pbd
, po
, status
);
975 static int prb_curr_blk_in_use(struct tpacket_block_desc
*pbd
)
977 return TP_STATUS_USER
& BLOCK_STATUS(pbd
);
980 static int prb_queue_frozen(struct tpacket_kbdq_core
*pkc
)
982 return pkc
->reset_pending_on_curr_blk
;
985 static void prb_clear_blk_fill_status(struct packet_ring_buffer
*rb
)
987 struct tpacket_kbdq_core
*pkc
= GET_PBDQC_FROM_RB(rb
);
988 atomic_dec(&pkc
->blk_fill_in_prog
);
991 static void prb_fill_rxhash(struct tpacket_kbdq_core
*pkc
,
992 struct tpacket3_hdr
*ppd
)
994 ppd
->hv1
.tp_rxhash
= skb_get_hash(pkc
->skb
);
997 static void prb_clear_rxhash(struct tpacket_kbdq_core
*pkc
,
998 struct tpacket3_hdr
*ppd
)
1000 ppd
->hv1
.tp_rxhash
= 0;
1003 static void prb_fill_vlan_info(struct tpacket_kbdq_core
*pkc
,
1004 struct tpacket3_hdr
*ppd
)
1006 if (skb_vlan_tag_present(pkc
->skb
)) {
1007 ppd
->hv1
.tp_vlan_tci
= skb_vlan_tag_get(pkc
->skb
);
1008 ppd
->hv1
.tp_vlan_tpid
= ntohs(pkc
->skb
->vlan_proto
);
1009 ppd
->tp_status
= TP_STATUS_VLAN_VALID
| TP_STATUS_VLAN_TPID_VALID
;
1011 ppd
->hv1
.tp_vlan_tci
= 0;
1012 ppd
->hv1
.tp_vlan_tpid
= 0;
1013 ppd
->tp_status
= TP_STATUS_AVAILABLE
;
1017 static void prb_run_all_ft_ops(struct tpacket_kbdq_core
*pkc
,
1018 struct tpacket3_hdr
*ppd
)
1020 ppd
->hv1
.tp_padding
= 0;
1021 prb_fill_vlan_info(pkc
, ppd
);
1023 if (pkc
->feature_req_word
& TP_FT_REQ_FILL_RXHASH
)
1024 prb_fill_rxhash(pkc
, ppd
);
1026 prb_clear_rxhash(pkc
, ppd
);
1029 static void prb_fill_curr_block(char *curr
,
1030 struct tpacket_kbdq_core
*pkc
,
1031 struct tpacket_block_desc
*pbd
,
1034 struct tpacket3_hdr
*ppd
;
1036 ppd
= (struct tpacket3_hdr
*)curr
;
1037 ppd
->tp_next_offset
= TOTAL_PKT_LEN_INCL_ALIGN(len
);
1039 pkc
->nxt_offset
+= TOTAL_PKT_LEN_INCL_ALIGN(len
);
1040 BLOCK_LEN(pbd
) += TOTAL_PKT_LEN_INCL_ALIGN(len
);
1041 BLOCK_NUM_PKTS(pbd
) += 1;
1042 atomic_inc(&pkc
->blk_fill_in_prog
);
1043 prb_run_all_ft_ops(pkc
, ppd
);
1046 /* Assumes caller has the sk->rx_queue.lock */
1047 static void *__packet_lookup_frame_in_block(struct packet_sock
*po
,
1048 struct sk_buff
*skb
,
1053 struct tpacket_kbdq_core
*pkc
;
1054 struct tpacket_block_desc
*pbd
;
1057 pkc
= GET_PBDQC_FROM_RB(&po
->rx_ring
);
1058 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
1060 /* Queue is frozen when user space is lagging behind */
1061 if (prb_queue_frozen(pkc
)) {
1063 * Check if that last block which caused the queue to freeze,
1064 * is still in_use by user-space.
1066 if (prb_curr_blk_in_use(pbd
)) {
1067 /* Can't record this packet */
1071 * Ok, the block was released by user-space.
1072 * Now let's open that block.
1073 * opening a block also thaws the queue.
1074 * Thawing is a side effect.
1076 prb_open_block(pkc
, pbd
);
1081 curr
= pkc
->nxt_offset
;
1083 end
= (char *)pbd
+ pkc
->kblk_size
;
1085 /* first try the current block */
1086 if (curr
+TOTAL_PKT_LEN_INCL_ALIGN(len
) < end
) {
1087 prb_fill_curr_block(curr
, pkc
, pbd
, len
);
1088 return (void *)curr
;
1091 /* Ok, close the current block */
1092 prb_retire_current_block(pkc
, po
, 0);
1094 /* Now, try to dispatch the next block */
1095 curr
= (char *)prb_dispatch_next_block(pkc
, po
);
1097 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
1098 prb_fill_curr_block(curr
, pkc
, pbd
, len
);
1099 return (void *)curr
;
1103 * No free blocks are available.user_space hasn't caught up yet.
1104 * Queue was just frozen and now this packet will get dropped.
1109 static void *packet_current_rx_frame(struct packet_sock
*po
,
1110 struct sk_buff
*skb
,
1111 int status
, unsigned int len
)
1114 switch (po
->tp_version
) {
1117 curr
= packet_lookup_frame(po
, &po
->rx_ring
,
1118 po
->rx_ring
.head
, status
);
1121 return __packet_lookup_frame_in_block(po
, skb
, status
, len
);
1123 WARN(1, "TPACKET version not supported\n");
1129 static void *prb_lookup_block(struct packet_sock
*po
,
1130 struct packet_ring_buffer
*rb
,
1134 struct tpacket_kbdq_core
*pkc
= GET_PBDQC_FROM_RB(rb
);
1135 struct tpacket_block_desc
*pbd
= GET_PBLOCK_DESC(pkc
, idx
);
1137 if (status
!= BLOCK_STATUS(pbd
))
1142 static int prb_previous_blk_num(struct packet_ring_buffer
*rb
)
1145 if (rb
->prb_bdqc
.kactive_blk_num
)
1146 prev
= rb
->prb_bdqc
.kactive_blk_num
-1;
1148 prev
= rb
->prb_bdqc
.knum_blocks
-1;
1152 /* Assumes caller has held the rx_queue.lock */
1153 static void *__prb_previous_block(struct packet_sock
*po
,
1154 struct packet_ring_buffer
*rb
,
1157 unsigned int previous
= prb_previous_blk_num(rb
);
1158 return prb_lookup_block(po
, rb
, previous
, status
);
1161 static void *packet_previous_rx_frame(struct packet_sock
*po
,
1162 struct packet_ring_buffer
*rb
,
1165 if (po
->tp_version
<= TPACKET_V2
)
1166 return packet_previous_frame(po
, rb
, status
);
1168 return __prb_previous_block(po
, rb
, status
);
1171 static void packet_increment_rx_head(struct packet_sock
*po
,
1172 struct packet_ring_buffer
*rb
)
1174 switch (po
->tp_version
) {
1177 return packet_increment_head(rb
);
1180 WARN(1, "TPACKET version not supported.\n");
1186 static void *packet_previous_frame(struct packet_sock
*po
,
1187 struct packet_ring_buffer
*rb
,
1190 unsigned int previous
= rb
->head
? rb
->head
- 1 : rb
->frame_max
;
1191 return packet_lookup_frame(po
, rb
, previous
, status
);
1194 static void packet_increment_head(struct packet_ring_buffer
*buff
)
1196 buff
->head
= buff
->head
!= buff
->frame_max
? buff
->head
+1 : 0;
1199 static void packet_inc_pending(struct packet_ring_buffer
*rb
)
1201 this_cpu_inc(*rb
->pending_refcnt
);
1204 static void packet_dec_pending(struct packet_ring_buffer
*rb
)
1206 this_cpu_dec(*rb
->pending_refcnt
);
1209 static unsigned int packet_read_pending(const struct packet_ring_buffer
*rb
)
1211 unsigned int refcnt
= 0;
1214 /* We don't use pending refcount in rx_ring. */
1215 if (rb
->pending_refcnt
== NULL
)
1218 for_each_possible_cpu(cpu
)
1219 refcnt
+= *per_cpu_ptr(rb
->pending_refcnt
, cpu
);
1224 static int packet_alloc_pending(struct packet_sock
*po
)
1226 po
->rx_ring
.pending_refcnt
= NULL
;
1228 po
->tx_ring
.pending_refcnt
= alloc_percpu(unsigned int);
1229 if (unlikely(po
->tx_ring
.pending_refcnt
== NULL
))
1235 static void packet_free_pending(struct packet_sock
*po
)
1237 free_percpu(po
->tx_ring
.pending_refcnt
);
1240 #define ROOM_POW_OFF 2
1241 #define ROOM_NONE 0x0
1242 #define ROOM_LOW 0x1
1243 #define ROOM_NORMAL 0x2
1245 static bool __tpacket_has_room(struct packet_sock
*po
, int pow_off
)
1249 len
= po
->rx_ring
.frame_max
+ 1;
1250 idx
= po
->rx_ring
.head
;
1252 idx
+= len
>> pow_off
;
1255 return packet_lookup_frame(po
, &po
->rx_ring
, idx
, TP_STATUS_KERNEL
);
1258 static bool __tpacket_v3_has_room(struct packet_sock
*po
, int pow_off
)
1262 len
= po
->rx_ring
.prb_bdqc
.knum_blocks
;
1263 idx
= po
->rx_ring
.prb_bdqc
.kactive_blk_num
;
1265 idx
+= len
>> pow_off
;
1268 return prb_lookup_block(po
, &po
->rx_ring
, idx
, TP_STATUS_KERNEL
);
1271 static int __packet_rcv_has_room(struct packet_sock
*po
, struct sk_buff
*skb
)
1273 struct sock
*sk
= &po
->sk
;
1274 int ret
= ROOM_NONE
;
1276 if (po
->prot_hook
.func
!= tpacket_rcv
) {
1277 int avail
= sk
->sk_rcvbuf
- atomic_read(&sk
->sk_rmem_alloc
)
1278 - (skb
? skb
->truesize
: 0);
1279 if (avail
> (sk
->sk_rcvbuf
>> ROOM_POW_OFF
))
1287 if (po
->tp_version
== TPACKET_V3
) {
1288 if (__tpacket_v3_has_room(po
, ROOM_POW_OFF
))
1290 else if (__tpacket_v3_has_room(po
, 0))
1293 if (__tpacket_has_room(po
, ROOM_POW_OFF
))
1295 else if (__tpacket_has_room(po
, 0))
1302 static int packet_rcv_has_room(struct packet_sock
*po
, struct sk_buff
*skb
)
1307 spin_lock_bh(&po
->sk
.sk_receive_queue
.lock
);
1308 ret
= __packet_rcv_has_room(po
, skb
);
1309 has_room
= ret
== ROOM_NORMAL
;
1310 if (po
->pressure
== has_room
)
1311 po
->pressure
= !has_room
;
1312 spin_unlock_bh(&po
->sk
.sk_receive_queue
.lock
);
1317 static void packet_sock_destruct(struct sock
*sk
)
1319 skb_queue_purge(&sk
->sk_error_queue
);
1321 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
1322 WARN_ON(refcount_read(&sk
->sk_wmem_alloc
));
1324 if (!sock_flag(sk
, SOCK_DEAD
)) {
1325 pr_err("Attempt to release alive packet socket: %p\n", sk
);
1329 sk_refcnt_debug_dec(sk
);
1332 static bool fanout_flow_is_huge(struct packet_sock
*po
, struct sk_buff
*skb
)
1337 rxhash
= skb_get_hash(skb
);
1338 for (i
= 0; i
< ROLLOVER_HLEN
; i
++)
1339 if (po
->rollover
->history
[i
] == rxhash
)
1342 po
->rollover
->history
[prandom_u32() % ROLLOVER_HLEN
] = rxhash
;
1343 return count
> (ROLLOVER_HLEN
>> 1);
1346 static unsigned int fanout_demux_hash(struct packet_fanout
*f
,
1347 struct sk_buff
*skb
,
1350 return reciprocal_scale(__skb_get_hash_symmetric(skb
), num
);
1353 static unsigned int fanout_demux_lb(struct packet_fanout
*f
,
1354 struct sk_buff
*skb
,
1357 unsigned int val
= atomic_inc_return(&f
->rr_cur
);
1362 static unsigned int fanout_demux_cpu(struct packet_fanout
*f
,
1363 struct sk_buff
*skb
,
1366 return smp_processor_id() % num
;
1369 static unsigned int fanout_demux_rnd(struct packet_fanout
*f
,
1370 struct sk_buff
*skb
,
1373 return prandom_u32_max(num
);
1376 static unsigned int fanout_demux_rollover(struct packet_fanout
*f
,
1377 struct sk_buff
*skb
,
1378 unsigned int idx
, bool try_self
,
1381 struct packet_sock
*po
, *po_next
, *po_skip
= NULL
;
1382 unsigned int i
, j
, room
= ROOM_NONE
;
1384 po
= pkt_sk(f
->arr
[idx
]);
1387 room
= packet_rcv_has_room(po
, skb
);
1388 if (room
== ROOM_NORMAL
||
1389 (room
== ROOM_LOW
&& !fanout_flow_is_huge(po
, skb
)))
1394 i
= j
= min_t(int, po
->rollover
->sock
, num
- 1);
1396 po_next
= pkt_sk(f
->arr
[i
]);
1397 if (po_next
!= po_skip
&& !po_next
->pressure
&&
1398 packet_rcv_has_room(po_next
, skb
) == ROOM_NORMAL
) {
1400 po
->rollover
->sock
= i
;
1401 atomic_long_inc(&po
->rollover
->num
);
1402 if (room
== ROOM_LOW
)
1403 atomic_long_inc(&po
->rollover
->num_huge
);
1411 atomic_long_inc(&po
->rollover
->num_failed
);
1415 static unsigned int fanout_demux_qm(struct packet_fanout
*f
,
1416 struct sk_buff
*skb
,
1419 return skb_get_queue_mapping(skb
) % num
;
1422 static unsigned int fanout_demux_bpf(struct packet_fanout
*f
,
1423 struct sk_buff
*skb
,
1426 struct bpf_prog
*prog
;
1427 unsigned int ret
= 0;
1430 prog
= rcu_dereference(f
->bpf_prog
);
1432 ret
= bpf_prog_run_clear_cb(prog
, skb
) % num
;
1438 static bool fanout_has_flag(struct packet_fanout
*f
, u16 flag
)
1440 return f
->flags
& (flag
>> 8);
1443 static int packet_rcv_fanout(struct sk_buff
*skb
, struct net_device
*dev
,
1444 struct packet_type
*pt
, struct net_device
*orig_dev
)
1446 struct packet_fanout
*f
= pt
->af_packet_priv
;
1447 unsigned int num
= READ_ONCE(f
->num_members
);
1448 struct net
*net
= read_pnet(&f
->net
);
1449 struct packet_sock
*po
;
1452 if (!net_eq(dev_net(dev
), net
) || !num
) {
1457 if (fanout_has_flag(f
, PACKET_FANOUT_FLAG_DEFRAG
)) {
1458 skb
= ip_check_defrag(net
, skb
, IP_DEFRAG_AF_PACKET
);
1463 case PACKET_FANOUT_HASH
:
1465 idx
= fanout_demux_hash(f
, skb
, num
);
1467 case PACKET_FANOUT_LB
:
1468 idx
= fanout_demux_lb(f
, skb
, num
);
1470 case PACKET_FANOUT_CPU
:
1471 idx
= fanout_demux_cpu(f
, skb
, num
);
1473 case PACKET_FANOUT_RND
:
1474 idx
= fanout_demux_rnd(f
, skb
, num
);
1476 case PACKET_FANOUT_QM
:
1477 idx
= fanout_demux_qm(f
, skb
, num
);
1479 case PACKET_FANOUT_ROLLOVER
:
1480 idx
= fanout_demux_rollover(f
, skb
, 0, false, num
);
1482 case PACKET_FANOUT_CBPF
:
1483 case PACKET_FANOUT_EBPF
:
1484 idx
= fanout_demux_bpf(f
, skb
, num
);
1488 if (fanout_has_flag(f
, PACKET_FANOUT_FLAG_ROLLOVER
))
1489 idx
= fanout_demux_rollover(f
, skb
, idx
, true, num
);
1491 po
= pkt_sk(f
->arr
[idx
]);
1492 return po
->prot_hook
.func(skb
, dev
, &po
->prot_hook
, orig_dev
);
1495 DEFINE_MUTEX(fanout_mutex
);
1496 EXPORT_SYMBOL_GPL(fanout_mutex
);
1497 static LIST_HEAD(fanout_list
);
1498 static u16 fanout_next_id
;
1500 static void __fanout_link(struct sock
*sk
, struct packet_sock
*po
)
1502 struct packet_fanout
*f
= po
->fanout
;
1504 spin_lock(&f
->lock
);
1505 f
->arr
[f
->num_members
] = sk
;
1508 if (f
->num_members
== 1)
1509 dev_add_pack(&f
->prot_hook
);
1510 spin_unlock(&f
->lock
);
1513 static void __fanout_unlink(struct sock
*sk
, struct packet_sock
*po
)
1515 struct packet_fanout
*f
= po
->fanout
;
1518 spin_lock(&f
->lock
);
1519 for (i
= 0; i
< f
->num_members
; i
++) {
1520 if (f
->arr
[i
] == sk
)
1523 BUG_ON(i
>= f
->num_members
);
1524 f
->arr
[i
] = f
->arr
[f
->num_members
- 1];
1526 if (f
->num_members
== 0)
1527 __dev_remove_pack(&f
->prot_hook
);
1528 spin_unlock(&f
->lock
);
1531 static bool match_fanout_group(struct packet_type
*ptype
, struct sock
*sk
)
1533 if (sk
->sk_family
!= PF_PACKET
)
1536 return ptype
->af_packet_priv
== pkt_sk(sk
)->fanout
;
1539 static void fanout_init_data(struct packet_fanout
*f
)
1542 case PACKET_FANOUT_LB
:
1543 atomic_set(&f
->rr_cur
, 0);
1545 case PACKET_FANOUT_CBPF
:
1546 case PACKET_FANOUT_EBPF
:
1547 RCU_INIT_POINTER(f
->bpf_prog
, NULL
);
1552 static void __fanout_set_data_bpf(struct packet_fanout
*f
, struct bpf_prog
*new)
1554 struct bpf_prog
*old
;
1556 spin_lock(&f
->lock
);
1557 old
= rcu_dereference_protected(f
->bpf_prog
, lockdep_is_held(&f
->lock
));
1558 rcu_assign_pointer(f
->bpf_prog
, new);
1559 spin_unlock(&f
->lock
);
1563 bpf_prog_destroy(old
);
1567 static int fanout_set_data_cbpf(struct packet_sock
*po
, char __user
*data
,
1570 struct bpf_prog
*new;
1571 struct sock_fprog fprog
;
1574 if (sock_flag(&po
->sk
, SOCK_FILTER_LOCKED
))
1576 if (len
!= sizeof(fprog
))
1578 if (copy_from_user(&fprog
, data
, len
))
1581 ret
= bpf_prog_create_from_user(&new, &fprog
, NULL
, false);
1585 __fanout_set_data_bpf(po
->fanout
, new);
1589 static int fanout_set_data_ebpf(struct packet_sock
*po
, char __user
*data
,
1592 struct bpf_prog
*new;
1595 if (sock_flag(&po
->sk
, SOCK_FILTER_LOCKED
))
1597 if (len
!= sizeof(fd
))
1599 if (copy_from_user(&fd
, data
, len
))
1602 new = bpf_prog_get_type(fd
, BPF_PROG_TYPE_SOCKET_FILTER
);
1604 return PTR_ERR(new);
1606 __fanout_set_data_bpf(po
->fanout
, new);
1610 static int fanout_set_data(struct packet_sock
*po
, char __user
*data
,
1613 switch (po
->fanout
->type
) {
1614 case PACKET_FANOUT_CBPF
:
1615 return fanout_set_data_cbpf(po
, data
, len
);
1616 case PACKET_FANOUT_EBPF
:
1617 return fanout_set_data_ebpf(po
, data
, len
);
1623 static void fanout_release_data(struct packet_fanout
*f
)
1626 case PACKET_FANOUT_CBPF
:
1627 case PACKET_FANOUT_EBPF
:
1628 __fanout_set_data_bpf(f
, NULL
);
1632 static bool __fanout_id_is_free(struct sock
*sk
, u16 candidate_id
)
1634 struct packet_fanout
*f
;
1636 list_for_each_entry(f
, &fanout_list
, list
) {
1637 if (f
->id
== candidate_id
&&
1638 read_pnet(&f
->net
) == sock_net(sk
)) {
1645 static bool fanout_find_new_id(struct sock
*sk
, u16
*new_id
)
1647 u16 id
= fanout_next_id
;
1650 if (__fanout_id_is_free(sk
, id
)) {
1652 fanout_next_id
= id
+ 1;
1657 } while (id
!= fanout_next_id
);
1662 static int fanout_add(struct sock
*sk
, u16 id
, u16 type_flags
)
1664 struct packet_rollover
*rollover
= NULL
;
1665 struct packet_sock
*po
= pkt_sk(sk
);
1666 struct packet_fanout
*f
, *match
;
1667 u8 type
= type_flags
& 0xff;
1668 u8 flags
= type_flags
>> 8;
1672 case PACKET_FANOUT_ROLLOVER
:
1673 if (type_flags
& PACKET_FANOUT_FLAG_ROLLOVER
)
1675 case PACKET_FANOUT_HASH
:
1676 case PACKET_FANOUT_LB
:
1677 case PACKET_FANOUT_CPU
:
1678 case PACKET_FANOUT_RND
:
1679 case PACKET_FANOUT_QM
:
1680 case PACKET_FANOUT_CBPF
:
1681 case PACKET_FANOUT_EBPF
:
1687 mutex_lock(&fanout_mutex
);
1693 if (type
== PACKET_FANOUT_ROLLOVER
||
1694 (type_flags
& PACKET_FANOUT_FLAG_ROLLOVER
)) {
1696 rollover
= kzalloc(sizeof(*rollover
), GFP_KERNEL
);
1699 atomic_long_set(&rollover
->num
, 0);
1700 atomic_long_set(&rollover
->num_huge
, 0);
1701 atomic_long_set(&rollover
->num_failed
, 0);
1702 po
->rollover
= rollover
;
1705 if (type_flags
& PACKET_FANOUT_FLAG_UNIQUEID
) {
1710 if (!fanout_find_new_id(sk
, &id
)) {
1714 /* ephemeral flag for the first socket in the group: drop it */
1715 flags
&= ~(PACKET_FANOUT_FLAG_UNIQUEID
>> 8);
1719 list_for_each_entry(f
, &fanout_list
, list
) {
1721 read_pnet(&f
->net
) == sock_net(sk
)) {
1727 if (match
&& match
->flags
!= flags
)
1731 match
= kzalloc(sizeof(*match
), GFP_KERNEL
);
1734 write_pnet(&match
->net
, sock_net(sk
));
1737 match
->flags
= flags
;
1738 INIT_LIST_HEAD(&match
->list
);
1739 spin_lock_init(&match
->lock
);
1740 refcount_set(&match
->sk_ref
, 0);
1741 fanout_init_data(match
);
1742 match
->prot_hook
.type
= po
->prot_hook
.type
;
1743 match
->prot_hook
.dev
= po
->prot_hook
.dev
;
1744 match
->prot_hook
.func
= packet_rcv_fanout
;
1745 match
->prot_hook
.af_packet_priv
= match
;
1746 match
->prot_hook
.id_match
= match_fanout_group
;
1747 list_add(&match
->list
, &fanout_list
);
1751 spin_lock(&po
->bind_lock
);
1753 match
->type
== type
&&
1754 match
->prot_hook
.type
== po
->prot_hook
.type
&&
1755 match
->prot_hook
.dev
== po
->prot_hook
.dev
) {
1757 if (refcount_read(&match
->sk_ref
) < PACKET_FANOUT_MAX
) {
1758 __dev_remove_pack(&po
->prot_hook
);
1760 refcount_set(&match
->sk_ref
, refcount_read(&match
->sk_ref
) + 1);
1761 __fanout_link(sk
, po
);
1765 spin_unlock(&po
->bind_lock
);
1767 if (err
&& !refcount_read(&match
->sk_ref
)) {
1768 list_del(&match
->list
);
1773 if (err
&& rollover
) {
1774 kfree_rcu(rollover
, rcu
);
1775 po
->rollover
= NULL
;
1777 mutex_unlock(&fanout_mutex
);
1781 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1782 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1783 * It is the responsibility of the caller to call fanout_release_data() and
1784 * free the returned packet_fanout (after synchronize_net())
1786 static struct packet_fanout
*fanout_release(struct sock
*sk
)
1788 struct packet_sock
*po
= pkt_sk(sk
);
1789 struct packet_fanout
*f
;
1791 mutex_lock(&fanout_mutex
);
1796 if (refcount_dec_and_test(&f
->sk_ref
))
1802 kfree_rcu(po
->rollover
, rcu
);
1803 po
->rollover
= NULL
;
1806 mutex_unlock(&fanout_mutex
);
1811 static bool packet_extra_vlan_len_allowed(const struct net_device
*dev
,
1812 struct sk_buff
*skb
)
1814 /* Earlier code assumed this would be a VLAN pkt, double-check
1815 * this now that we have the actual packet in hand. We can only
1816 * do this check on Ethernet devices.
1818 if (unlikely(dev
->type
!= ARPHRD_ETHER
))
1821 skb_reset_mac_header(skb
);
1822 return likely(eth_hdr(skb
)->h_proto
== htons(ETH_P_8021Q
));
1825 static const struct proto_ops packet_ops
;
1827 static const struct proto_ops packet_ops_spkt
;
1829 static int packet_rcv_spkt(struct sk_buff
*skb
, struct net_device
*dev
,
1830 struct packet_type
*pt
, struct net_device
*orig_dev
)
1833 struct sockaddr_pkt
*spkt
;
1836 * When we registered the protocol we saved the socket in the data
1837 * field for just this event.
1840 sk
= pt
->af_packet_priv
;
1843 * Yank back the headers [hope the device set this
1844 * right or kerboom...]
1846 * Incoming packets have ll header pulled,
1849 * For outgoing ones skb->data == skb_mac_header(skb)
1850 * so that this procedure is noop.
1853 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1856 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1859 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1863 /* drop any routing info */
1866 /* drop conntrack reference */
1869 spkt
= &PACKET_SKB_CB(skb
)->sa
.pkt
;
1871 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1874 * The SOCK_PACKET socket receives _all_ frames.
1877 spkt
->spkt_family
= dev
->type
;
1878 strlcpy(spkt
->spkt_device
, dev
->name
, sizeof(spkt
->spkt_device
));
1879 spkt
->spkt_protocol
= skb
->protocol
;
1882 * Charge the memory to the socket. This is done specifically
1883 * to prevent sockets using all the memory up.
1886 if (sock_queue_rcv_skb(sk
, skb
) == 0)
1897 * Output a raw packet to a device layer. This bypasses all the other
1898 * protocol layers and you must therefore supply it with a complete frame
1901 static int packet_sendmsg_spkt(struct socket
*sock
, struct msghdr
*msg
,
1904 struct sock
*sk
= sock
->sk
;
1905 DECLARE_SOCKADDR(struct sockaddr_pkt
*, saddr
, msg
->msg_name
);
1906 struct sk_buff
*skb
= NULL
;
1907 struct net_device
*dev
;
1908 struct sockcm_cookie sockc
;
1914 * Get and verify the address.
1918 if (msg
->msg_namelen
< sizeof(struct sockaddr
))
1920 if (msg
->msg_namelen
== sizeof(struct sockaddr_pkt
))
1921 proto
= saddr
->spkt_protocol
;
1923 return -ENOTCONN
; /* SOCK_PACKET must be sent giving an address */
1926 * Find the device first to size check it
1929 saddr
->spkt_device
[sizeof(saddr
->spkt_device
) - 1] = 0;
1932 dev
= dev_get_by_name_rcu(sock_net(sk
), saddr
->spkt_device
);
1938 if (!(dev
->flags
& IFF_UP
))
1942 * You may not queue a frame bigger than the mtu. This is the lowest level
1943 * raw protocol and you must do your own fragmentation at this level.
1946 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
1947 if (!netif_supports_nofcs(dev
)) {
1948 err
= -EPROTONOSUPPORT
;
1951 extra_len
= 4; /* We're doing our own CRC */
1955 if (len
> dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
+ extra_len
)
1959 size_t reserved
= LL_RESERVED_SPACE(dev
);
1960 int tlen
= dev
->needed_tailroom
;
1961 unsigned int hhlen
= dev
->header_ops
? dev
->hard_header_len
: 0;
1964 skb
= sock_wmalloc(sk
, len
+ reserved
+ tlen
, 0, GFP_KERNEL
);
1967 /* FIXME: Save some space for broken drivers that write a hard
1968 * header at transmission time by themselves. PPP is the notable
1969 * one here. This should really be fixed at the driver level.
1971 skb_reserve(skb
, reserved
);
1972 skb_reset_network_header(skb
);
1974 /* Try to align data part correctly */
1979 skb_reset_network_header(skb
);
1981 err
= memcpy_from_msg(skb_put(skb
, len
), msg
, len
);
1987 if (!dev_validate_header(dev
, skb
->data
, len
)) {
1991 if (len
> (dev
->mtu
+ dev
->hard_header_len
+ extra_len
) &&
1992 !packet_extra_vlan_len_allowed(dev
, skb
)) {
1997 sockc
.tsflags
= sk
->sk_tsflags
;
1998 if (msg
->msg_controllen
) {
1999 err
= sock_cmsg_send(sk
, msg
, &sockc
);
2004 skb
->protocol
= proto
;
2006 skb
->priority
= sk
->sk_priority
;
2007 skb
->mark
= sk
->sk_mark
;
2009 sock_tx_timestamp(sk
, sockc
.tsflags
, &skb_shinfo(skb
)->tx_flags
);
2011 if (unlikely(extra_len
== 4))
2014 skb_probe_transport_header(skb
, 0);
2016 dev_queue_xmit(skb
);
2027 static unsigned int run_filter(struct sk_buff
*skb
,
2028 const struct sock
*sk
,
2031 struct sk_filter
*filter
;
2034 filter
= rcu_dereference(sk
->sk_filter
);
2036 res
= bpf_prog_run_clear_cb(filter
->prog
, skb
);
2042 static int packet_rcv_vnet(struct msghdr
*msg
, const struct sk_buff
*skb
,
2045 struct virtio_net_hdr vnet_hdr
;
2047 if (*len
< sizeof(vnet_hdr
))
2049 *len
-= sizeof(vnet_hdr
);
2051 if (virtio_net_hdr_from_skb(skb
, &vnet_hdr
, vio_le(), true))
2054 return memcpy_to_msg(msg
, (void *)&vnet_hdr
, sizeof(vnet_hdr
));
2058 * This function makes lazy skb cloning in hope that most of packets
2059 * are discarded by BPF.
2061 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2062 * and skb->cb are mangled. It works because (and until) packets
2063 * falling here are owned by current CPU. Output packets are cloned
2064 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2065 * sequencially, so that if we return skb to original state on exit,
2066 * we will not harm anyone.
2069 static int packet_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
2070 struct packet_type
*pt
, struct net_device
*orig_dev
)
2073 struct sockaddr_ll
*sll
;
2074 struct packet_sock
*po
;
2075 u8
*skb_head
= skb
->data
;
2076 int skb_len
= skb
->len
;
2077 unsigned int snaplen
, res
;
2078 bool is_drop_n_account
= false;
2080 if (skb
->pkt_type
== PACKET_LOOPBACK
)
2083 sk
= pt
->af_packet_priv
;
2086 if (!net_eq(dev_net(dev
), sock_net(sk
)))
2091 if (dev
->header_ops
) {
2092 /* The device has an explicit notion of ll header,
2093 * exported to higher levels.
2095 * Otherwise, the device hides details of its frame
2096 * structure, so that corresponding packet head is
2097 * never delivered to user.
2099 if (sk
->sk_type
!= SOCK_DGRAM
)
2100 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2101 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
2102 /* Special case: outgoing packets have ll header at head */
2103 skb_pull(skb
, skb_network_offset(skb
));
2109 res
= run_filter(skb
, sk
, snaplen
);
2111 goto drop_n_restore
;
2115 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
2118 if (skb_shared(skb
)) {
2119 struct sk_buff
*nskb
= skb_clone(skb
, GFP_ATOMIC
);
2123 if (skb_head
!= skb
->data
) {
2124 skb
->data
= skb_head
;
2131 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb
)) + MAX_ADDR_LEN
- 8);
2133 sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
2134 sll
->sll_hatype
= dev
->type
;
2135 sll
->sll_pkttype
= skb
->pkt_type
;
2136 if (unlikely(po
->origdev
))
2137 sll
->sll_ifindex
= orig_dev
->ifindex
;
2139 sll
->sll_ifindex
= dev
->ifindex
;
2141 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
2143 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2144 * Use their space for storing the original skb length.
2146 PACKET_SKB_CB(skb
)->sa
.origlen
= skb
->len
;
2148 if (pskb_trim(skb
, snaplen
))
2151 skb_set_owner_r(skb
, sk
);
2155 /* drop conntrack reference */
2158 spin_lock(&sk
->sk_receive_queue
.lock
);
2159 po
->stats
.stats1
.tp_packets
++;
2160 sock_skb_set_dropcount(sk
, skb
);
2161 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
2162 spin_unlock(&sk
->sk_receive_queue
.lock
);
2163 sk
->sk_data_ready(sk
);
2167 is_drop_n_account
= true;
2168 spin_lock(&sk
->sk_receive_queue
.lock
);
2169 po
->stats
.stats1
.tp_drops
++;
2170 atomic_inc(&sk
->sk_drops
);
2171 spin_unlock(&sk
->sk_receive_queue
.lock
);
2174 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
2175 skb
->data
= skb_head
;
2179 if (!is_drop_n_account
)
2186 static int tpacket_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
2187 struct packet_type
*pt
, struct net_device
*orig_dev
)
2190 struct packet_sock
*po
;
2191 struct sockaddr_ll
*sll
;
2192 union tpacket_uhdr h
;
2193 u8
*skb_head
= skb
->data
;
2194 int skb_len
= skb
->len
;
2195 unsigned int snaplen
, res
;
2196 unsigned long status
= TP_STATUS_USER
;
2197 unsigned short macoff
, netoff
, hdrlen
;
2198 struct sk_buff
*copy_skb
= NULL
;
2201 bool is_drop_n_account
= false;
2202 bool do_vnet
= false;
2204 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2205 * We may add members to them until current aligned size without forcing
2206 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2208 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h
.h2
)) != 32);
2209 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h
.h3
)) != 48);
2211 if (skb
->pkt_type
== PACKET_LOOPBACK
)
2214 sk
= pt
->af_packet_priv
;
2217 if (!net_eq(dev_net(dev
), sock_net(sk
)))
2220 if (dev
->header_ops
) {
2221 if (sk
->sk_type
!= SOCK_DGRAM
)
2222 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2223 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
2224 /* Special case: outgoing packets have ll header at head */
2225 skb_pull(skb
, skb_network_offset(skb
));
2231 res
= run_filter(skb
, sk
, snaplen
);
2233 goto drop_n_restore
;
2235 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2236 status
|= TP_STATUS_CSUMNOTREADY
;
2237 else if (skb
->pkt_type
!= PACKET_OUTGOING
&&
2238 (skb
->ip_summed
== CHECKSUM_COMPLETE
||
2239 skb_csum_unnecessary(skb
)))
2240 status
|= TP_STATUS_CSUM_VALID
;
2245 if (sk
->sk_type
== SOCK_DGRAM
) {
2246 macoff
= netoff
= TPACKET_ALIGN(po
->tp_hdrlen
) + 16 +
2249 unsigned int maclen
= skb_network_offset(skb
);
2250 netoff
= TPACKET_ALIGN(po
->tp_hdrlen
+
2251 (maclen
< 16 ? 16 : maclen
)) +
2253 if (po
->has_vnet_hdr
) {
2254 netoff
+= sizeof(struct virtio_net_hdr
);
2257 macoff
= netoff
- maclen
;
2259 if (po
->tp_version
<= TPACKET_V2
) {
2260 if (macoff
+ snaplen
> po
->rx_ring
.frame_size
) {
2261 if (po
->copy_thresh
&&
2262 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
2263 if (skb_shared(skb
)) {
2264 copy_skb
= skb_clone(skb
, GFP_ATOMIC
);
2266 copy_skb
= skb_get(skb
);
2267 skb_head
= skb
->data
;
2270 skb_set_owner_r(copy_skb
, sk
);
2272 snaplen
= po
->rx_ring
.frame_size
- macoff
;
2273 if ((int)snaplen
< 0) {
2278 } else if (unlikely(macoff
+ snaplen
>
2279 GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
)) {
2282 nval
= GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
- macoff
;
2283 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2284 snaplen
, nval
, macoff
);
2286 if (unlikely((int)snaplen
< 0)) {
2288 macoff
= GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
;
2292 spin_lock(&sk
->sk_receive_queue
.lock
);
2293 h
.raw
= packet_current_rx_frame(po
, skb
,
2294 TP_STATUS_KERNEL
, (macoff
+snaplen
));
2296 goto drop_n_account
;
2297 if (po
->tp_version
<= TPACKET_V2
) {
2298 packet_increment_rx_head(po
, &po
->rx_ring
);
2300 * LOSING will be reported till you read the stats,
2301 * because it's COR - Clear On Read.
2302 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2305 if (po
->stats
.stats1
.tp_drops
)
2306 status
|= TP_STATUS_LOSING
;
2308 po
->stats
.stats1
.tp_packets
++;
2310 status
|= TP_STATUS_COPY
;
2311 __skb_queue_tail(&sk
->sk_receive_queue
, copy_skb
);
2313 spin_unlock(&sk
->sk_receive_queue
.lock
);
2316 if (virtio_net_hdr_from_skb(skb
, h
.raw
+ macoff
-
2317 sizeof(struct virtio_net_hdr
),
2319 spin_lock(&sk
->sk_receive_queue
.lock
);
2320 goto drop_n_account
;
2324 skb_copy_bits(skb
, 0, h
.raw
+ macoff
, snaplen
);
2326 if (!(ts_status
= tpacket_get_timestamp(skb
, &ts
, po
->tp_tstamp
)))
2327 getnstimeofday(&ts
);
2329 status
|= ts_status
;
2331 switch (po
->tp_version
) {
2333 h
.h1
->tp_len
= skb
->len
;
2334 h
.h1
->tp_snaplen
= snaplen
;
2335 h
.h1
->tp_mac
= macoff
;
2336 h
.h1
->tp_net
= netoff
;
2337 h
.h1
->tp_sec
= ts
.tv_sec
;
2338 h
.h1
->tp_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
2339 hdrlen
= sizeof(*h
.h1
);
2342 h
.h2
->tp_len
= skb
->len
;
2343 h
.h2
->tp_snaplen
= snaplen
;
2344 h
.h2
->tp_mac
= macoff
;
2345 h
.h2
->tp_net
= netoff
;
2346 h
.h2
->tp_sec
= ts
.tv_sec
;
2347 h
.h2
->tp_nsec
= ts
.tv_nsec
;
2348 if (skb_vlan_tag_present(skb
)) {
2349 h
.h2
->tp_vlan_tci
= skb_vlan_tag_get(skb
);
2350 h
.h2
->tp_vlan_tpid
= ntohs(skb
->vlan_proto
);
2351 status
|= TP_STATUS_VLAN_VALID
| TP_STATUS_VLAN_TPID_VALID
;
2353 h
.h2
->tp_vlan_tci
= 0;
2354 h
.h2
->tp_vlan_tpid
= 0;
2356 memset(h
.h2
->tp_padding
, 0, sizeof(h
.h2
->tp_padding
));
2357 hdrlen
= sizeof(*h
.h2
);
2360 /* tp_nxt_offset,vlan are already populated above.
2361 * So DONT clear those fields here
2363 h
.h3
->tp_status
|= status
;
2364 h
.h3
->tp_len
= skb
->len
;
2365 h
.h3
->tp_snaplen
= snaplen
;
2366 h
.h3
->tp_mac
= macoff
;
2367 h
.h3
->tp_net
= netoff
;
2368 h
.h3
->tp_sec
= ts
.tv_sec
;
2369 h
.h3
->tp_nsec
= ts
.tv_nsec
;
2370 memset(h
.h3
->tp_padding
, 0, sizeof(h
.h3
->tp_padding
));
2371 hdrlen
= sizeof(*h
.h3
);
2377 sll
= h
.raw
+ TPACKET_ALIGN(hdrlen
);
2378 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
2379 sll
->sll_family
= AF_PACKET
;
2380 sll
->sll_hatype
= dev
->type
;
2381 sll
->sll_protocol
= skb
->protocol
;
2382 sll
->sll_pkttype
= skb
->pkt_type
;
2383 if (unlikely(po
->origdev
))
2384 sll
->sll_ifindex
= orig_dev
->ifindex
;
2386 sll
->sll_ifindex
= dev
->ifindex
;
2390 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2391 if (po
->tp_version
<= TPACKET_V2
) {
2394 end
= (u8
*) PAGE_ALIGN((unsigned long) h
.raw
+
2397 for (start
= h
.raw
; start
< end
; start
+= PAGE_SIZE
)
2398 flush_dcache_page(pgv_to_page(start
));
2403 if (po
->tp_version
<= TPACKET_V2
) {
2404 __packet_set_status(po
, h
.raw
, status
);
2405 sk
->sk_data_ready(sk
);
2407 prb_clear_blk_fill_status(&po
->rx_ring
);
2411 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
2412 skb
->data
= skb_head
;
2416 if (!is_drop_n_account
)
2423 is_drop_n_account
= true;
2424 po
->stats
.stats1
.tp_drops
++;
2425 spin_unlock(&sk
->sk_receive_queue
.lock
);
2427 sk
->sk_data_ready(sk
);
2428 kfree_skb(copy_skb
);
2429 goto drop_n_restore
;
2432 static void tpacket_destruct_skb(struct sk_buff
*skb
)
2434 struct packet_sock
*po
= pkt_sk(skb
->sk
);
2436 if (likely(po
->tx_ring
.pg_vec
)) {
2440 ph
= skb_shinfo(skb
)->destructor_arg
;
2441 packet_dec_pending(&po
->tx_ring
);
2443 ts
= __packet_set_timestamp(po
, ph
, skb
);
2444 __packet_set_status(po
, ph
, TP_STATUS_AVAILABLE
| ts
);
2450 static void tpacket_set_protocol(const struct net_device
*dev
,
2451 struct sk_buff
*skb
)
2453 if (dev
->type
== ARPHRD_ETHER
) {
2454 skb_reset_mac_header(skb
);
2455 skb
->protocol
= eth_hdr(skb
)->h_proto
;
2459 static int __packet_snd_vnet_parse(struct virtio_net_hdr
*vnet_hdr
, size_t len
)
2461 if ((vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
2462 (__virtio16_to_cpu(vio_le(), vnet_hdr
->csum_start
) +
2463 __virtio16_to_cpu(vio_le(), vnet_hdr
->csum_offset
) + 2 >
2464 __virtio16_to_cpu(vio_le(), vnet_hdr
->hdr_len
)))
2465 vnet_hdr
->hdr_len
= __cpu_to_virtio16(vio_le(),
2466 __virtio16_to_cpu(vio_le(), vnet_hdr
->csum_start
) +
2467 __virtio16_to_cpu(vio_le(), vnet_hdr
->csum_offset
) + 2);
2469 if (__virtio16_to_cpu(vio_le(), vnet_hdr
->hdr_len
) > len
)
2475 static int packet_snd_vnet_parse(struct msghdr
*msg
, size_t *len
,
2476 struct virtio_net_hdr
*vnet_hdr
)
2478 if (*len
< sizeof(*vnet_hdr
))
2480 *len
-= sizeof(*vnet_hdr
);
2482 if (!copy_from_iter_full(vnet_hdr
, sizeof(*vnet_hdr
), &msg
->msg_iter
))
2485 return __packet_snd_vnet_parse(vnet_hdr
, *len
);
2488 static int tpacket_fill_skb(struct packet_sock
*po
, struct sk_buff
*skb
,
2489 void *frame
, struct net_device
*dev
, void *data
, int tp_len
,
2490 __be16 proto
, unsigned char *addr
, int hlen
, int copylen
,
2491 const struct sockcm_cookie
*sockc
)
2493 union tpacket_uhdr ph
;
2494 int to_write
, offset
, len
, nr_frags
, len_max
;
2495 struct socket
*sock
= po
->sk
.sk_socket
;
2501 skb
->protocol
= proto
;
2503 skb
->priority
= po
->sk
.sk_priority
;
2504 skb
->mark
= po
->sk
.sk_mark
;
2505 sock_tx_timestamp(&po
->sk
, sockc
->tsflags
, &skb_shinfo(skb
)->tx_flags
);
2506 skb_shinfo(skb
)->destructor_arg
= ph
.raw
;
2508 skb_reserve(skb
, hlen
);
2509 skb_reset_network_header(skb
);
2513 if (sock
->type
== SOCK_DGRAM
) {
2514 err
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
,
2516 if (unlikely(err
< 0))
2518 } else if (copylen
) {
2519 int hdrlen
= min_t(int, copylen
, tp_len
);
2521 skb_push(skb
, dev
->hard_header_len
);
2522 skb_put(skb
, copylen
- dev
->hard_header_len
);
2523 err
= skb_store_bits(skb
, 0, data
, hdrlen
);
2526 if (!dev_validate_header(dev
, skb
->data
, hdrlen
))
2529 tpacket_set_protocol(dev
, skb
);
2535 offset
= offset_in_page(data
);
2536 len_max
= PAGE_SIZE
- offset
;
2537 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2539 skb
->data_len
= to_write
;
2540 skb
->len
+= to_write
;
2541 skb
->truesize
+= to_write
;
2542 refcount_add(to_write
, &po
->sk
.sk_wmem_alloc
);
2544 while (likely(to_write
)) {
2545 nr_frags
= skb_shinfo(skb
)->nr_frags
;
2547 if (unlikely(nr_frags
>= MAX_SKB_FRAGS
)) {
2548 pr_err("Packet exceed the number of skb frags(%lu)\n",
2553 page
= pgv_to_page(data
);
2555 flush_dcache_page(page
);
2557 skb_fill_page_desc(skb
, nr_frags
, page
, offset
, len
);
2560 len_max
= PAGE_SIZE
;
2561 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2564 skb_probe_transport_header(skb
, 0);
2569 static int tpacket_parse_header(struct packet_sock
*po
, void *frame
,
2570 int size_max
, void **data
)
2572 union tpacket_uhdr ph
;
2577 switch (po
->tp_version
) {
2579 if (ph
.h3
->tp_next_offset
!= 0) {
2580 pr_warn_once("variable sized slot not supported");
2583 tp_len
= ph
.h3
->tp_len
;
2586 tp_len
= ph
.h2
->tp_len
;
2589 tp_len
= ph
.h1
->tp_len
;
2592 if (unlikely(tp_len
> size_max
)) {
2593 pr_err("packet size is too long (%d > %d)\n", tp_len
, size_max
);
2597 if (unlikely(po
->tp_tx_has_off
)) {
2598 int off_min
, off_max
;
2600 off_min
= po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
2601 off_max
= po
->tx_ring
.frame_size
- tp_len
;
2602 if (po
->sk
.sk_type
== SOCK_DGRAM
) {
2603 switch (po
->tp_version
) {
2605 off
= ph
.h3
->tp_net
;
2608 off
= ph
.h2
->tp_net
;
2611 off
= ph
.h1
->tp_net
;
2615 switch (po
->tp_version
) {
2617 off
= ph
.h3
->tp_mac
;
2620 off
= ph
.h2
->tp_mac
;
2623 off
= ph
.h1
->tp_mac
;
2627 if (unlikely((off
< off_min
) || (off_max
< off
)))
2630 off
= po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
2633 *data
= frame
+ off
;
2637 static int tpacket_snd(struct packet_sock
*po
, struct msghdr
*msg
)
2639 struct sk_buff
*skb
;
2640 struct net_device
*dev
;
2641 struct virtio_net_hdr
*vnet_hdr
= NULL
;
2642 struct sockcm_cookie sockc
;
2644 int err
, reserve
= 0;
2646 DECLARE_SOCKADDR(struct sockaddr_ll
*, saddr
, msg
->msg_name
);
2647 bool need_wait
= !(msg
->msg_flags
& MSG_DONTWAIT
);
2648 int tp_len
, size_max
;
2649 unsigned char *addr
;
2652 int status
= TP_STATUS_AVAILABLE
;
2653 int hlen
, tlen
, copylen
= 0;
2655 mutex_lock(&po
->pg_vec_lock
);
2657 if (likely(saddr
== NULL
)) {
2658 dev
= packet_cached_dev_get(po
);
2663 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2665 if (msg
->msg_namelen
< (saddr
->sll_halen
2666 + offsetof(struct sockaddr_ll
,
2669 proto
= saddr
->sll_protocol
;
2670 addr
= saddr
->sll_addr
;
2671 dev
= dev_get_by_index(sock_net(&po
->sk
), saddr
->sll_ifindex
);
2675 if (unlikely(dev
== NULL
))
2678 if (unlikely(!(dev
->flags
& IFF_UP
)))
2681 sockc
.tsflags
= po
->sk
.sk_tsflags
;
2682 if (msg
->msg_controllen
) {
2683 err
= sock_cmsg_send(&po
->sk
, msg
, &sockc
);
2688 if (po
->sk
.sk_socket
->type
== SOCK_RAW
)
2689 reserve
= dev
->hard_header_len
;
2690 size_max
= po
->tx_ring
.frame_size
2691 - (po
->tp_hdrlen
- sizeof(struct sockaddr_ll
));
2693 if ((size_max
> dev
->mtu
+ reserve
+ VLAN_HLEN
) && !po
->has_vnet_hdr
)
2694 size_max
= dev
->mtu
+ reserve
+ VLAN_HLEN
;
2697 ph
= packet_current_frame(po
, &po
->tx_ring
,
2698 TP_STATUS_SEND_REQUEST
);
2699 if (unlikely(ph
== NULL
)) {
2700 if (need_wait
&& need_resched())
2706 tp_len
= tpacket_parse_header(po
, ph
, size_max
, &data
);
2710 status
= TP_STATUS_SEND_REQUEST
;
2711 hlen
= LL_RESERVED_SPACE(dev
);
2712 tlen
= dev
->needed_tailroom
;
2713 if (po
->has_vnet_hdr
) {
2715 data
+= sizeof(*vnet_hdr
);
2716 tp_len
-= sizeof(*vnet_hdr
);
2718 __packet_snd_vnet_parse(vnet_hdr
, tp_len
)) {
2722 copylen
= __virtio16_to_cpu(vio_le(),
2725 copylen
= max_t(int, copylen
, dev
->hard_header_len
);
2726 skb
= sock_alloc_send_skb(&po
->sk
,
2727 hlen
+ tlen
+ sizeof(struct sockaddr_ll
) +
2728 (copylen
- dev
->hard_header_len
),
2731 if (unlikely(skb
== NULL
)) {
2732 /* we assume the socket was initially writeable ... */
2733 if (likely(len_sum
> 0))
2737 tp_len
= tpacket_fill_skb(po
, skb
, ph
, dev
, data
, tp_len
, proto
,
2738 addr
, hlen
, copylen
, &sockc
);
2739 if (likely(tp_len
>= 0) &&
2740 tp_len
> dev
->mtu
+ reserve
&&
2741 !po
->has_vnet_hdr
&&
2742 !packet_extra_vlan_len_allowed(dev
, skb
))
2745 if (unlikely(tp_len
< 0)) {
2748 __packet_set_status(po
, ph
,
2749 TP_STATUS_AVAILABLE
);
2750 packet_increment_head(&po
->tx_ring
);
2754 status
= TP_STATUS_WRONG_FORMAT
;
2760 if (po
->has_vnet_hdr
&& virtio_net_hdr_to_skb(skb
, vnet_hdr
,
2766 skb
->destructor
= tpacket_destruct_skb
;
2767 __packet_set_status(po
, ph
, TP_STATUS_SENDING
);
2768 packet_inc_pending(&po
->tx_ring
);
2770 status
= TP_STATUS_SEND_REQUEST
;
2771 err
= po
->xmit(skb
);
2772 if (unlikely(err
> 0)) {
2773 err
= net_xmit_errno(err
);
2774 if (err
&& __packet_get_status(po
, ph
) ==
2775 TP_STATUS_AVAILABLE
) {
2776 /* skb was destructed already */
2781 * skb was dropped but not destructed yet;
2782 * let's treat it like congestion or err < 0
2786 packet_increment_head(&po
->tx_ring
);
2788 } while (likely((ph
!= NULL
) ||
2789 /* Note: packet_read_pending() might be slow if we have
2790 * to call it as it's per_cpu variable, but in fast-path
2791 * we already short-circuit the loop with the first
2792 * condition, and luckily don't have to go that path
2795 (need_wait
&& packet_read_pending(&po
->tx_ring
))));
2801 __packet_set_status(po
, ph
, status
);
2806 mutex_unlock(&po
->pg_vec_lock
);
2810 static struct sk_buff
*packet_alloc_skb(struct sock
*sk
, size_t prepad
,
2811 size_t reserve
, size_t len
,
2812 size_t linear
, int noblock
,
2815 struct sk_buff
*skb
;
2817 /* Under a page? Don't bother with paged skb. */
2818 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
2821 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
2826 skb_reserve(skb
, reserve
);
2827 skb_put(skb
, linear
);
2828 skb
->data_len
= len
- linear
;
2829 skb
->len
+= len
- linear
;
2834 static int packet_snd(struct socket
*sock
, struct msghdr
*msg
, size_t len
)
2836 struct sock
*sk
= sock
->sk
;
2837 DECLARE_SOCKADDR(struct sockaddr_ll
*, saddr
, msg
->msg_name
);
2838 struct sk_buff
*skb
;
2839 struct net_device
*dev
;
2841 unsigned char *addr
;
2842 int err
, reserve
= 0;
2843 struct sockcm_cookie sockc
;
2844 struct virtio_net_hdr vnet_hdr
= { 0 };
2846 struct packet_sock
*po
= pkt_sk(sk
);
2847 bool has_vnet_hdr
= false;
2848 int hlen
, tlen
, linear
;
2852 * Get and verify the address.
2855 if (likely(saddr
== NULL
)) {
2856 dev
= packet_cached_dev_get(po
);
2861 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2863 if (msg
->msg_namelen
< (saddr
->sll_halen
+ offsetof(struct sockaddr_ll
, sll_addr
)))
2865 proto
= saddr
->sll_protocol
;
2866 addr
= saddr
->sll_addr
;
2867 dev
= dev_get_by_index(sock_net(sk
), saddr
->sll_ifindex
);
2871 if (unlikely(dev
== NULL
))
2874 if (unlikely(!(dev
->flags
& IFF_UP
)))
2877 sockc
.tsflags
= sk
->sk_tsflags
;
2878 sockc
.mark
= sk
->sk_mark
;
2879 if (msg
->msg_controllen
) {
2880 err
= sock_cmsg_send(sk
, msg
, &sockc
);
2885 if (sock
->type
== SOCK_RAW
)
2886 reserve
= dev
->hard_header_len
;
2887 if (po
->has_vnet_hdr
) {
2888 err
= packet_snd_vnet_parse(msg
, &len
, &vnet_hdr
);
2891 has_vnet_hdr
= true;
2894 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
2895 if (!netif_supports_nofcs(dev
)) {
2896 err
= -EPROTONOSUPPORT
;
2899 extra_len
= 4; /* We're doing our own CRC */
2903 if (!vnet_hdr
.gso_type
&&
2904 (len
> dev
->mtu
+ reserve
+ VLAN_HLEN
+ extra_len
))
2908 hlen
= LL_RESERVED_SPACE(dev
);
2909 tlen
= dev
->needed_tailroom
;
2910 linear
= __virtio16_to_cpu(vio_le(), vnet_hdr
.hdr_len
);
2911 linear
= max(linear
, min_t(int, len
, dev
->hard_header_len
));
2912 skb
= packet_alloc_skb(sk
, hlen
+ tlen
, hlen
, len
, linear
,
2913 msg
->msg_flags
& MSG_DONTWAIT
, &err
);
2917 skb_set_network_header(skb
, reserve
);
2920 if (sock
->type
== SOCK_DGRAM
) {
2921 offset
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
, NULL
, len
);
2922 if (unlikely(offset
< 0))
2926 /* Returns -EFAULT on error */
2927 err
= skb_copy_datagram_from_iter(skb
, offset
, &msg
->msg_iter
, len
);
2931 if (sock
->type
== SOCK_RAW
&&
2932 !dev_validate_header(dev
, skb
->data
, len
)) {
2937 sock_tx_timestamp(sk
, sockc
.tsflags
, &skb_shinfo(skb
)->tx_flags
);
2939 if (!vnet_hdr
.gso_type
&& (len
> dev
->mtu
+ reserve
+ extra_len
) &&
2940 !packet_extra_vlan_len_allowed(dev
, skb
)) {
2945 skb
->protocol
= proto
;
2947 skb
->priority
= sk
->sk_priority
;
2948 skb
->mark
= sockc
.mark
;
2951 err
= virtio_net_hdr_to_skb(skb
, &vnet_hdr
, vio_le());
2954 len
+= sizeof(vnet_hdr
);
2957 skb_probe_transport_header(skb
, reserve
);
2959 if (unlikely(extra_len
== 4))
2962 err
= po
->xmit(skb
);
2963 if (err
> 0 && (err
= net_xmit_errno(err
)) != 0)
2979 static int packet_sendmsg(struct socket
*sock
, struct msghdr
*msg
, size_t len
)
2981 struct sock
*sk
= sock
->sk
;
2982 struct packet_sock
*po
= pkt_sk(sk
);
2984 if (po
->tx_ring
.pg_vec
)
2985 return tpacket_snd(po
, msg
);
2987 return packet_snd(sock
, msg
, len
);
2991 * Close a PACKET socket. This is fairly simple. We immediately go
2992 * to 'closed' state and remove our protocol entry in the device list.
2995 static int packet_release(struct socket
*sock
)
2997 struct sock
*sk
= sock
->sk
;
2998 struct packet_sock
*po
;
2999 struct packet_fanout
*f
;
3001 union tpacket_req_u req_u
;
3009 mutex_lock(&net
->packet
.sklist_lock
);
3010 sk_del_node_init_rcu(sk
);
3011 mutex_unlock(&net
->packet
.sklist_lock
);
3014 sock_prot_inuse_add(net
, sk
->sk_prot
, -1);
3017 spin_lock(&po
->bind_lock
);
3018 unregister_prot_hook(sk
, false);
3019 packet_cached_dev_reset(po
);
3021 if (po
->prot_hook
.dev
) {
3022 dev_put(po
->prot_hook
.dev
);
3023 po
->prot_hook
.dev
= NULL
;
3025 spin_unlock(&po
->bind_lock
);
3027 packet_flush_mclist(sk
);
3029 if (po
->rx_ring
.pg_vec
) {
3030 memset(&req_u
, 0, sizeof(req_u
));
3031 packet_set_ring(sk
, &req_u
, 1, 0);
3034 if (po
->tx_ring
.pg_vec
) {
3035 memset(&req_u
, 0, sizeof(req_u
));
3036 packet_set_ring(sk
, &req_u
, 1, 1);
3039 f
= fanout_release(sk
);
3044 fanout_release_data(f
);
3048 * Now the socket is dead. No more input will appear.
3055 skb_queue_purge(&sk
->sk_receive_queue
);
3056 packet_free_pending(po
);
3057 sk_refcnt_debug_release(sk
);
3064 * Attach a packet hook.
3067 static int packet_do_bind(struct sock
*sk
, const char *name
, int ifindex
,
3070 struct packet_sock
*po
= pkt_sk(sk
);
3071 struct net_device
*dev_curr
;
3074 struct net_device
*dev
= NULL
;
3076 bool unlisted
= false;
3079 spin_lock(&po
->bind_lock
);
3088 dev
= dev_get_by_name_rcu(sock_net(sk
), name
);
3093 } else if (ifindex
) {
3094 dev
= dev_get_by_index_rcu(sock_net(sk
), ifindex
);
3104 proto_curr
= po
->prot_hook
.type
;
3105 dev_curr
= po
->prot_hook
.dev
;
3107 need_rehook
= proto_curr
!= proto
|| dev_curr
!= dev
;
3112 __unregister_prot_hook(sk
, true);
3114 dev_curr
= po
->prot_hook
.dev
;
3116 unlisted
= !dev_get_by_index_rcu(sock_net(sk
),
3121 po
->prot_hook
.type
= proto
;
3123 if (unlikely(unlisted
)) {
3125 po
->prot_hook
.dev
= NULL
;
3127 packet_cached_dev_reset(po
);
3129 po
->prot_hook
.dev
= dev
;
3130 po
->ifindex
= dev
? dev
->ifindex
: 0;
3131 packet_cached_dev_assign(po
, dev
);
3137 if (proto
== 0 || !need_rehook
)
3140 if (!unlisted
&& (!dev
|| (dev
->flags
& IFF_UP
))) {
3141 register_prot_hook(sk
);
3143 sk
->sk_err
= ENETDOWN
;
3144 if (!sock_flag(sk
, SOCK_DEAD
))
3145 sk
->sk_error_report(sk
);
3150 spin_unlock(&po
->bind_lock
);
3156 * Bind a packet socket to a device
3159 static int packet_bind_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
3162 struct sock
*sk
= sock
->sk
;
3163 char name
[sizeof(uaddr
->sa_data
) + 1];
3169 if (addr_len
!= sizeof(struct sockaddr
))
3171 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3174 memcpy(name
, uaddr
->sa_data
, sizeof(uaddr
->sa_data
));
3175 name
[sizeof(uaddr
->sa_data
)] = 0;
3177 return packet_do_bind(sk
, name
, 0, pkt_sk(sk
)->num
);
3180 static int packet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
3182 struct sockaddr_ll
*sll
= (struct sockaddr_ll
*)uaddr
;
3183 struct sock
*sk
= sock
->sk
;
3189 if (addr_len
< sizeof(struct sockaddr_ll
))
3191 if (sll
->sll_family
!= AF_PACKET
)
3194 return packet_do_bind(sk
, NULL
, sll
->sll_ifindex
,
3195 sll
->sll_protocol
? : pkt_sk(sk
)->num
);
3198 static struct proto packet_proto
= {
3200 .owner
= THIS_MODULE
,
3201 .obj_size
= sizeof(struct packet_sock
),
3205 * Create a packet of type SOCK_PACKET.
3208 static int packet_create(struct net
*net
, struct socket
*sock
, int protocol
,
3212 struct packet_sock
*po
;
3213 __be16 proto
= (__force __be16
)protocol
; /* weird, but documented */
3216 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
3218 if (sock
->type
!= SOCK_DGRAM
&& sock
->type
!= SOCK_RAW
&&
3219 sock
->type
!= SOCK_PACKET
)
3220 return -ESOCKTNOSUPPORT
;
3222 sock
->state
= SS_UNCONNECTED
;
3225 sk
= sk_alloc(net
, PF_PACKET
, GFP_KERNEL
, &packet_proto
, kern
);
3229 sock
->ops
= &packet_ops
;
3230 if (sock
->type
== SOCK_PACKET
)
3231 sock
->ops
= &packet_ops_spkt
;
3233 sock_init_data(sock
, sk
);
3236 sk
->sk_family
= PF_PACKET
;
3238 po
->xmit
= dev_queue_xmit
;
3240 err
= packet_alloc_pending(po
);
3244 packet_cached_dev_reset(po
);
3246 sk
->sk_destruct
= packet_sock_destruct
;
3247 sk_refcnt_debug_inc(sk
);
3250 * Attach a protocol block
3253 spin_lock_init(&po
->bind_lock
);
3254 mutex_init(&po
->pg_vec_lock
);
3255 po
->rollover
= NULL
;
3256 po
->prot_hook
.func
= packet_rcv
;
3258 if (sock
->type
== SOCK_PACKET
)
3259 po
->prot_hook
.func
= packet_rcv_spkt
;
3261 po
->prot_hook
.af_packet_priv
= sk
;
3264 po
->prot_hook
.type
= proto
;
3265 register_prot_hook(sk
);
3268 mutex_lock(&net
->packet
.sklist_lock
);
3269 sk_add_node_rcu(sk
, &net
->packet
.sklist
);
3270 mutex_unlock(&net
->packet
.sklist_lock
);
3273 sock_prot_inuse_add(net
, &packet_proto
, 1);
3284 * Pull a packet from our receive queue and hand it to the user.
3285 * If necessary we block.
3288 static int packet_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t len
,
3291 struct sock
*sk
= sock
->sk
;
3292 struct sk_buff
*skb
;
3294 int vnet_hdr_len
= 0;
3295 unsigned int origlen
= 0;
3298 if (flags
& ~(MSG_PEEK
|MSG_DONTWAIT
|MSG_TRUNC
|MSG_CMSG_COMPAT
|MSG_ERRQUEUE
))
3302 /* What error should we return now? EUNATTACH? */
3303 if (pkt_sk(sk
)->ifindex
< 0)
3307 if (flags
& MSG_ERRQUEUE
) {
3308 err
= sock_recv_errqueue(sk
, msg
, len
,
3309 SOL_PACKET
, PACKET_TX_TIMESTAMP
);
3314 * Call the generic datagram receiver. This handles all sorts
3315 * of horrible races and re-entrancy so we can forget about it
3316 * in the protocol layers.
3318 * Now it will return ENETDOWN, if device have just gone down,
3319 * but then it will block.
3322 skb
= skb_recv_datagram(sk
, flags
, flags
& MSG_DONTWAIT
, &err
);
3325 * An error occurred so return it. Because skb_recv_datagram()
3326 * handles the blocking we don't see and worry about blocking
3333 if (pkt_sk(sk
)->pressure
)
3334 packet_rcv_has_room(pkt_sk(sk
), NULL
);
3336 if (pkt_sk(sk
)->has_vnet_hdr
) {
3337 err
= packet_rcv_vnet(msg
, skb
, &len
);
3340 vnet_hdr_len
= sizeof(struct virtio_net_hdr
);
3343 /* You lose any data beyond the buffer you gave. If it worries
3344 * a user program they can ask the device for its MTU
3350 msg
->msg_flags
|= MSG_TRUNC
;
3353 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
3357 if (sock
->type
!= SOCK_PACKET
) {
3358 struct sockaddr_ll
*sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
3360 /* Original length was stored in sockaddr_ll fields */
3361 origlen
= PACKET_SKB_CB(skb
)->sa
.origlen
;
3362 sll
->sll_family
= AF_PACKET
;
3363 sll
->sll_protocol
= skb
->protocol
;
3366 sock_recv_ts_and_drops(msg
, sk
, skb
);
3368 if (msg
->msg_name
) {
3369 /* If the address length field is there to be filled
3370 * in, we fill it in now.
3372 if (sock
->type
== SOCK_PACKET
) {
3373 __sockaddr_check_size(sizeof(struct sockaddr_pkt
));
3374 msg
->msg_namelen
= sizeof(struct sockaddr_pkt
);
3376 struct sockaddr_ll
*sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
3378 msg
->msg_namelen
= sll
->sll_halen
+
3379 offsetof(struct sockaddr_ll
, sll_addr
);
3381 memcpy(msg
->msg_name
, &PACKET_SKB_CB(skb
)->sa
,
3385 if (pkt_sk(sk
)->auxdata
) {
3386 struct tpacket_auxdata aux
;
3388 aux
.tp_status
= TP_STATUS_USER
;
3389 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
3390 aux
.tp_status
|= TP_STATUS_CSUMNOTREADY
;
3391 else if (skb
->pkt_type
!= PACKET_OUTGOING
&&
3392 (skb
->ip_summed
== CHECKSUM_COMPLETE
||
3393 skb_csum_unnecessary(skb
)))
3394 aux
.tp_status
|= TP_STATUS_CSUM_VALID
;
3396 aux
.tp_len
= origlen
;
3397 aux
.tp_snaplen
= skb
->len
;
3399 aux
.tp_net
= skb_network_offset(skb
);
3400 if (skb_vlan_tag_present(skb
)) {
3401 aux
.tp_vlan_tci
= skb_vlan_tag_get(skb
);
3402 aux
.tp_vlan_tpid
= ntohs(skb
->vlan_proto
);
3403 aux
.tp_status
|= TP_STATUS_VLAN_VALID
| TP_STATUS_VLAN_TPID_VALID
;
3405 aux
.tp_vlan_tci
= 0;
3406 aux
.tp_vlan_tpid
= 0;
3408 put_cmsg(msg
, SOL_PACKET
, PACKET_AUXDATA
, sizeof(aux
), &aux
);
3412 * Free or return the buffer as appropriate. Again this
3413 * hides all the races and re-entrancy issues from us.
3415 err
= vnet_hdr_len
+ ((flags
&MSG_TRUNC
) ? skb
->len
: copied
);
3418 skb_free_datagram(sk
, skb
);
3423 static int packet_getname_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
3424 int *uaddr_len
, int peer
)
3426 struct net_device
*dev
;
3427 struct sock
*sk
= sock
->sk
;
3432 uaddr
->sa_family
= AF_PACKET
;
3433 memset(uaddr
->sa_data
, 0, sizeof(uaddr
->sa_data
));
3435 dev
= dev_get_by_index_rcu(sock_net(sk
), pkt_sk(sk
)->ifindex
);
3437 strlcpy(uaddr
->sa_data
, dev
->name
, sizeof(uaddr
->sa_data
));
3439 *uaddr_len
= sizeof(*uaddr
);
3444 static int packet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
3445 int *uaddr_len
, int peer
)
3447 struct net_device
*dev
;
3448 struct sock
*sk
= sock
->sk
;
3449 struct packet_sock
*po
= pkt_sk(sk
);
3450 DECLARE_SOCKADDR(struct sockaddr_ll
*, sll
, uaddr
);
3455 sll
->sll_family
= AF_PACKET
;
3456 sll
->sll_ifindex
= po
->ifindex
;
3457 sll
->sll_protocol
= po
->num
;
3458 sll
->sll_pkttype
= 0;
3460 dev
= dev_get_by_index_rcu(sock_net(sk
), po
->ifindex
);
3462 sll
->sll_hatype
= dev
->type
;
3463 sll
->sll_halen
= dev
->addr_len
;
3464 memcpy(sll
->sll_addr
, dev
->dev_addr
, dev
->addr_len
);
3466 sll
->sll_hatype
= 0; /* Bad: we have no ARPHRD_UNSPEC */
3470 *uaddr_len
= offsetof(struct sockaddr_ll
, sll_addr
) + sll
->sll_halen
;
3475 static int packet_dev_mc(struct net_device
*dev
, struct packet_mclist
*i
,
3479 case PACKET_MR_MULTICAST
:
3480 if (i
->alen
!= dev
->addr_len
)
3483 return dev_mc_add(dev
, i
->addr
);
3485 return dev_mc_del(dev
, i
->addr
);
3487 case PACKET_MR_PROMISC
:
3488 return dev_set_promiscuity(dev
, what
);
3489 case PACKET_MR_ALLMULTI
:
3490 return dev_set_allmulti(dev
, what
);
3491 case PACKET_MR_UNICAST
:
3492 if (i
->alen
!= dev
->addr_len
)
3495 return dev_uc_add(dev
, i
->addr
);
3497 return dev_uc_del(dev
, i
->addr
);
3505 static void packet_dev_mclist_delete(struct net_device
*dev
,
3506 struct packet_mclist
**mlp
)
3508 struct packet_mclist
*ml
;
3510 while ((ml
= *mlp
) != NULL
) {
3511 if (ml
->ifindex
== dev
->ifindex
) {
3512 packet_dev_mc(dev
, ml
, -1);
3520 static int packet_mc_add(struct sock
*sk
, struct packet_mreq_max
*mreq
)
3522 struct packet_sock
*po
= pkt_sk(sk
);
3523 struct packet_mclist
*ml
, *i
;
3524 struct net_device
*dev
;
3530 dev
= __dev_get_by_index(sock_net(sk
), mreq
->mr_ifindex
);
3535 if (mreq
->mr_alen
> dev
->addr_len
)
3539 i
= kmalloc(sizeof(*i
), GFP_KERNEL
);
3544 for (ml
= po
->mclist
; ml
; ml
= ml
->next
) {
3545 if (ml
->ifindex
== mreq
->mr_ifindex
&&
3546 ml
->type
== mreq
->mr_type
&&
3547 ml
->alen
== mreq
->mr_alen
&&
3548 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
3550 /* Free the new element ... */
3556 i
->type
= mreq
->mr_type
;
3557 i
->ifindex
= mreq
->mr_ifindex
;
3558 i
->alen
= mreq
->mr_alen
;
3559 memcpy(i
->addr
, mreq
->mr_address
, i
->alen
);
3560 memset(i
->addr
+ i
->alen
, 0, sizeof(i
->addr
) - i
->alen
);
3562 i
->next
= po
->mclist
;
3564 err
= packet_dev_mc(dev
, i
, 1);
3566 po
->mclist
= i
->next
;
3575 static int packet_mc_drop(struct sock
*sk
, struct packet_mreq_max
*mreq
)
3577 struct packet_mclist
*ml
, **mlp
;
3581 for (mlp
= &pkt_sk(sk
)->mclist
; (ml
= *mlp
) != NULL
; mlp
= &ml
->next
) {
3582 if (ml
->ifindex
== mreq
->mr_ifindex
&&
3583 ml
->type
== mreq
->mr_type
&&
3584 ml
->alen
== mreq
->mr_alen
&&
3585 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
3586 if (--ml
->count
== 0) {
3587 struct net_device
*dev
;
3589 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
3591 packet_dev_mc(dev
, ml
, -1);
3601 static void packet_flush_mclist(struct sock
*sk
)
3603 struct packet_sock
*po
= pkt_sk(sk
);
3604 struct packet_mclist
*ml
;
3610 while ((ml
= po
->mclist
) != NULL
) {
3611 struct net_device
*dev
;
3613 po
->mclist
= ml
->next
;
3614 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
3616 packet_dev_mc(dev
, ml
, -1);
3623 packet_setsockopt(struct socket
*sock
, int level
, int optname
, char __user
*optval
, unsigned int optlen
)
3625 struct sock
*sk
= sock
->sk
;
3626 struct packet_sock
*po
= pkt_sk(sk
);
3629 if (level
!= SOL_PACKET
)
3630 return -ENOPROTOOPT
;
3633 case PACKET_ADD_MEMBERSHIP
:
3634 case PACKET_DROP_MEMBERSHIP
:
3636 struct packet_mreq_max mreq
;
3638 memset(&mreq
, 0, sizeof(mreq
));
3639 if (len
< sizeof(struct packet_mreq
))
3641 if (len
> sizeof(mreq
))
3643 if (copy_from_user(&mreq
, optval
, len
))
3645 if (len
< (mreq
.mr_alen
+ offsetof(struct packet_mreq
, mr_address
)))
3647 if (optname
== PACKET_ADD_MEMBERSHIP
)
3648 ret
= packet_mc_add(sk
, &mreq
);
3650 ret
= packet_mc_drop(sk
, &mreq
);
3654 case PACKET_RX_RING
:
3655 case PACKET_TX_RING
:
3657 union tpacket_req_u req_u
;
3660 switch (po
->tp_version
) {
3663 len
= sizeof(req_u
.req
);
3667 len
= sizeof(req_u
.req3
);
3672 if (copy_from_user(&req_u
.req
, optval
, len
))
3674 return packet_set_ring(sk
, &req_u
, 0,
3675 optname
== PACKET_TX_RING
);
3677 case PACKET_COPY_THRESH
:
3681 if (optlen
!= sizeof(val
))
3683 if (copy_from_user(&val
, optval
, sizeof(val
)))
3686 pkt_sk(sk
)->copy_thresh
= val
;
3689 case PACKET_VERSION
:
3693 if (optlen
!= sizeof(val
))
3695 if (copy_from_user(&val
, optval
, sizeof(val
)))
3706 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
) {
3709 po
->tp_version
= val
;
3715 case PACKET_RESERVE
:
3719 if (optlen
!= sizeof(val
))
3721 if (copy_from_user(&val
, optval
, sizeof(val
)))
3726 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
) {
3729 po
->tp_reserve
= val
;
3739 if (optlen
!= sizeof(val
))
3741 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3743 if (copy_from_user(&val
, optval
, sizeof(val
)))
3745 po
->tp_loss
= !!val
;
3748 case PACKET_AUXDATA
:
3752 if (optlen
< sizeof(val
))
3754 if (copy_from_user(&val
, optval
, sizeof(val
)))
3757 po
->auxdata
= !!val
;
3760 case PACKET_ORIGDEV
:
3764 if (optlen
< sizeof(val
))
3766 if (copy_from_user(&val
, optval
, sizeof(val
)))
3769 po
->origdev
= !!val
;
3772 case PACKET_VNET_HDR
:
3776 if (sock
->type
!= SOCK_RAW
)
3778 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3780 if (optlen
< sizeof(val
))
3782 if (copy_from_user(&val
, optval
, sizeof(val
)))
3785 po
->has_vnet_hdr
= !!val
;
3788 case PACKET_TIMESTAMP
:
3792 if (optlen
!= sizeof(val
))
3794 if (copy_from_user(&val
, optval
, sizeof(val
)))
3797 po
->tp_tstamp
= val
;
3804 if (optlen
!= sizeof(val
))
3806 if (copy_from_user(&val
, optval
, sizeof(val
)))
3809 return fanout_add(sk
, val
& 0xffff, val
>> 16);
3811 case PACKET_FANOUT_DATA
:
3816 return fanout_set_data(po
, optval
, optlen
);
3818 case PACKET_TX_HAS_OFF
:
3822 if (optlen
!= sizeof(val
))
3824 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3826 if (copy_from_user(&val
, optval
, sizeof(val
)))
3828 po
->tp_tx_has_off
= !!val
;
3831 case PACKET_QDISC_BYPASS
:
3835 if (optlen
!= sizeof(val
))
3837 if (copy_from_user(&val
, optval
, sizeof(val
)))
3840 po
->xmit
= val
? packet_direct_xmit
: dev_queue_xmit
;
3844 return -ENOPROTOOPT
;
3848 static int packet_getsockopt(struct socket
*sock
, int level
, int optname
,
3849 char __user
*optval
, int __user
*optlen
)
3852 int val
, lv
= sizeof(val
);
3853 struct sock
*sk
= sock
->sk
;
3854 struct packet_sock
*po
= pkt_sk(sk
);
3856 union tpacket_stats_u st
;
3857 struct tpacket_rollover_stats rstats
;
3858 struct packet_rollover
*rollover
;
3860 if (level
!= SOL_PACKET
)
3861 return -ENOPROTOOPT
;
3863 if (get_user(len
, optlen
))
3870 case PACKET_STATISTICS
:
3871 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3872 memcpy(&st
, &po
->stats
, sizeof(st
));
3873 memset(&po
->stats
, 0, sizeof(po
->stats
));
3874 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3876 if (po
->tp_version
== TPACKET_V3
) {
3877 lv
= sizeof(struct tpacket_stats_v3
);
3878 st
.stats3
.tp_packets
+= st
.stats3
.tp_drops
;
3881 lv
= sizeof(struct tpacket_stats
);
3882 st
.stats1
.tp_packets
+= st
.stats1
.tp_drops
;
3887 case PACKET_AUXDATA
:
3890 case PACKET_ORIGDEV
:
3893 case PACKET_VNET_HDR
:
3894 val
= po
->has_vnet_hdr
;
3896 case PACKET_VERSION
:
3897 val
= po
->tp_version
;
3900 if (len
> sizeof(int))
3902 if (len
< sizeof(int))
3904 if (copy_from_user(&val
, optval
, len
))
3908 val
= sizeof(struct tpacket_hdr
);
3911 val
= sizeof(struct tpacket2_hdr
);
3914 val
= sizeof(struct tpacket3_hdr
);
3920 case PACKET_RESERVE
:
3921 val
= po
->tp_reserve
;
3926 case PACKET_TIMESTAMP
:
3927 val
= po
->tp_tstamp
;
3931 ((u32
)po
->fanout
->id
|
3932 ((u32
)po
->fanout
->type
<< 16) |
3933 ((u32
)po
->fanout
->flags
<< 24)) :
3936 case PACKET_ROLLOVER_STATS
:
3938 rollover
= rcu_dereference(po
->rollover
);
3940 rstats
.tp_all
= atomic_long_read(&rollover
->num
);
3941 rstats
.tp_huge
= atomic_long_read(&rollover
->num_huge
);
3942 rstats
.tp_failed
= atomic_long_read(&rollover
->num_failed
);
3944 lv
= sizeof(rstats
);
3950 case PACKET_TX_HAS_OFF
:
3951 val
= po
->tp_tx_has_off
;
3953 case PACKET_QDISC_BYPASS
:
3954 val
= packet_use_direct_xmit(po
);
3957 return -ENOPROTOOPT
;
3962 if (put_user(len
, optlen
))
3964 if (copy_to_user(optval
, data
, len
))
3970 #ifdef CONFIG_COMPAT
3971 static int compat_packet_setsockopt(struct socket
*sock
, int level
, int optname
,
3972 char __user
*optval
, unsigned int optlen
)
3974 struct packet_sock
*po
= pkt_sk(sock
->sk
);
3976 if (level
!= SOL_PACKET
)
3977 return -ENOPROTOOPT
;
3979 if (optname
== PACKET_FANOUT_DATA
&&
3980 po
->fanout
&& po
->fanout
->type
== PACKET_FANOUT_CBPF
) {
3981 optval
= (char __user
*)get_compat_bpf_fprog(optval
);
3984 optlen
= sizeof(struct sock_fprog
);
3987 return packet_setsockopt(sock
, level
, optname
, optval
, optlen
);
3991 static int packet_notifier(struct notifier_block
*this,
3992 unsigned long msg
, void *ptr
)
3995 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
3996 struct net
*net
= dev_net(dev
);
3999 sk_for_each_rcu(sk
, &net
->packet
.sklist
) {
4000 struct packet_sock
*po
= pkt_sk(sk
);
4003 case NETDEV_UNREGISTER
:
4005 packet_dev_mclist_delete(dev
, &po
->mclist
);
4009 if (dev
->ifindex
== po
->ifindex
) {
4010 spin_lock(&po
->bind_lock
);
4012 __unregister_prot_hook(sk
, false);
4013 sk
->sk_err
= ENETDOWN
;
4014 if (!sock_flag(sk
, SOCK_DEAD
))
4015 sk
->sk_error_report(sk
);
4017 if (msg
== NETDEV_UNREGISTER
) {
4018 packet_cached_dev_reset(po
);
4020 if (po
->prot_hook
.dev
)
4021 dev_put(po
->prot_hook
.dev
);
4022 po
->prot_hook
.dev
= NULL
;
4024 spin_unlock(&po
->bind_lock
);
4028 if (dev
->ifindex
== po
->ifindex
) {
4029 spin_lock(&po
->bind_lock
);
4031 register_prot_hook(sk
);
4032 spin_unlock(&po
->bind_lock
);
4042 static int packet_ioctl(struct socket
*sock
, unsigned int cmd
,
4045 struct sock
*sk
= sock
->sk
;
4050 int amount
= sk_wmem_alloc_get(sk
);
4052 return put_user(amount
, (int __user
*)arg
);
4056 struct sk_buff
*skb
;
4059 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
4060 skb
= skb_peek(&sk
->sk_receive_queue
);
4063 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
4064 return put_user(amount
, (int __user
*)arg
);
4067 return sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
4069 return sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
4079 case SIOCGIFBRDADDR
:
4080 case SIOCSIFBRDADDR
:
4081 case SIOCGIFNETMASK
:
4082 case SIOCSIFNETMASK
:
4083 case SIOCGIFDSTADDR
:
4084 case SIOCSIFDSTADDR
:
4086 return inet_dgram_ops
.ioctl(sock
, cmd
, arg
);
4090 return -ENOIOCTLCMD
;
4095 static unsigned int packet_poll(struct file
*file
, struct socket
*sock
,
4098 struct sock
*sk
= sock
->sk
;
4099 struct packet_sock
*po
= pkt_sk(sk
);
4100 unsigned int mask
= datagram_poll(file
, sock
, wait
);
4102 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
4103 if (po
->rx_ring
.pg_vec
) {
4104 if (!packet_previous_rx_frame(po
, &po
->rx_ring
,
4106 mask
|= POLLIN
| POLLRDNORM
;
4108 if (po
->pressure
&& __packet_rcv_has_room(po
, NULL
) == ROOM_NORMAL
)
4110 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
4111 spin_lock_bh(&sk
->sk_write_queue
.lock
);
4112 if (po
->tx_ring
.pg_vec
) {
4113 if (packet_current_frame(po
, &po
->tx_ring
, TP_STATUS_AVAILABLE
))
4114 mask
|= POLLOUT
| POLLWRNORM
;
4116 spin_unlock_bh(&sk
->sk_write_queue
.lock
);
4121 /* Dirty? Well, I still did not learn better way to account
4125 static void packet_mm_open(struct vm_area_struct
*vma
)
4127 struct file
*file
= vma
->vm_file
;
4128 struct socket
*sock
= file
->private_data
;
4129 struct sock
*sk
= sock
->sk
;
4132 atomic_inc(&pkt_sk(sk
)->mapped
);
4135 static void packet_mm_close(struct vm_area_struct
*vma
)
4137 struct file
*file
= vma
->vm_file
;
4138 struct socket
*sock
= file
->private_data
;
4139 struct sock
*sk
= sock
->sk
;
4142 atomic_dec(&pkt_sk(sk
)->mapped
);
4145 static const struct vm_operations_struct packet_mmap_ops
= {
4146 .open
= packet_mm_open
,
4147 .close
= packet_mm_close
,
4150 static void free_pg_vec(struct pgv
*pg_vec
, unsigned int order
,
4155 for (i
= 0; i
< len
; i
++) {
4156 if (likely(pg_vec
[i
].buffer
)) {
4157 if (is_vmalloc_addr(pg_vec
[i
].buffer
))
4158 vfree(pg_vec
[i
].buffer
);
4160 free_pages((unsigned long)pg_vec
[i
].buffer
,
4162 pg_vec
[i
].buffer
= NULL
;
4168 static char *alloc_one_pg_vec_page(unsigned long order
)
4171 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_COMP
|
4172 __GFP_ZERO
| __GFP_NOWARN
| __GFP_NORETRY
;
4174 buffer
= (char *) __get_free_pages(gfp_flags
, order
);
4178 /* __get_free_pages failed, fall back to vmalloc */
4179 buffer
= vzalloc((1 << order
) * PAGE_SIZE
);
4183 /* vmalloc failed, lets dig into swap here */
4184 gfp_flags
&= ~__GFP_NORETRY
;
4185 buffer
= (char *) __get_free_pages(gfp_flags
, order
);
4189 /* complete and utter failure */
4193 static struct pgv
*alloc_pg_vec(struct tpacket_req
*req
, int order
)
4195 unsigned int block_nr
= req
->tp_block_nr
;
4199 pg_vec
= kcalloc(block_nr
, sizeof(struct pgv
), GFP_KERNEL
);
4200 if (unlikely(!pg_vec
))
4203 for (i
= 0; i
< block_nr
; i
++) {
4204 pg_vec
[i
].buffer
= alloc_one_pg_vec_page(order
);
4205 if (unlikely(!pg_vec
[i
].buffer
))
4206 goto out_free_pgvec
;
4213 free_pg_vec(pg_vec
, order
, block_nr
);
4218 static int packet_set_ring(struct sock
*sk
, union tpacket_req_u
*req_u
,
4219 int closing
, int tx_ring
)
4221 struct pgv
*pg_vec
= NULL
;
4222 struct packet_sock
*po
= pkt_sk(sk
);
4223 int was_running
, order
= 0;
4224 struct packet_ring_buffer
*rb
;
4225 struct sk_buff_head
*rb_queue
;
4228 /* Added to avoid minimal code churn */
4229 struct tpacket_req
*req
= &req_u
->req
;
4233 rb
= tx_ring
? &po
->tx_ring
: &po
->rx_ring
;
4234 rb_queue
= tx_ring
? &sk
->sk_write_queue
: &sk
->sk_receive_queue
;
4238 if (atomic_read(&po
->mapped
))
4240 if (packet_read_pending(rb
))
4244 if (req
->tp_block_nr
) {
4245 /* Sanity tests and some calculations */
4247 if (unlikely(rb
->pg_vec
))
4250 switch (po
->tp_version
) {
4252 po
->tp_hdrlen
= TPACKET_HDRLEN
;
4255 po
->tp_hdrlen
= TPACKET2_HDRLEN
;
4258 po
->tp_hdrlen
= TPACKET3_HDRLEN
;
4263 if (unlikely((int)req
->tp_block_size
<= 0))
4265 if (unlikely(!PAGE_ALIGNED(req
->tp_block_size
)))
4267 if (po
->tp_version
>= TPACKET_V3
&&
4268 req
->tp_block_size
<=
4269 BLK_PLUS_PRIV((u64
)req_u
->req3
.tp_sizeof_priv
))
4271 if (unlikely(req
->tp_frame_size
< po
->tp_hdrlen
+
4274 if (unlikely(req
->tp_frame_size
& (TPACKET_ALIGNMENT
- 1)))
4277 rb
->frames_per_block
= req
->tp_block_size
/ req
->tp_frame_size
;
4278 if (unlikely(rb
->frames_per_block
== 0))
4280 if (unlikely(req
->tp_block_size
> UINT_MAX
/ req
->tp_block_nr
))
4282 if (unlikely((rb
->frames_per_block
* req
->tp_block_nr
) !=
4287 order
= get_order(req
->tp_block_size
);
4288 pg_vec
= alloc_pg_vec(req
, order
);
4289 if (unlikely(!pg_vec
))
4291 switch (po
->tp_version
) {
4293 /* Block transmit is not supported yet */
4295 init_prb_bdqc(po
, rb
, pg_vec
, req_u
);
4297 struct tpacket_req3
*req3
= &req_u
->req3
;
4299 if (req3
->tp_retire_blk_tov
||
4300 req3
->tp_sizeof_priv
||
4301 req3
->tp_feature_req_word
) {
4314 if (unlikely(req
->tp_frame_nr
))
4319 /* Detach socket from network */
4320 spin_lock(&po
->bind_lock
);
4321 was_running
= po
->running
;
4325 __unregister_prot_hook(sk
, false);
4327 spin_unlock(&po
->bind_lock
);
4332 mutex_lock(&po
->pg_vec_lock
);
4333 if (closing
|| atomic_read(&po
->mapped
) == 0) {
4335 spin_lock_bh(&rb_queue
->lock
);
4336 swap(rb
->pg_vec
, pg_vec
);
4337 rb
->frame_max
= (req
->tp_frame_nr
- 1);
4339 rb
->frame_size
= req
->tp_frame_size
;
4340 spin_unlock_bh(&rb_queue
->lock
);
4342 swap(rb
->pg_vec_order
, order
);
4343 swap(rb
->pg_vec_len
, req
->tp_block_nr
);
4345 rb
->pg_vec_pages
= req
->tp_block_size
/PAGE_SIZE
;
4346 po
->prot_hook
.func
= (po
->rx_ring
.pg_vec
) ?
4347 tpacket_rcv
: packet_rcv
;
4348 skb_queue_purge(rb_queue
);
4349 if (atomic_read(&po
->mapped
))
4350 pr_err("packet_mmap: vma is busy: %d\n",
4351 atomic_read(&po
->mapped
));
4353 mutex_unlock(&po
->pg_vec_lock
);
4355 spin_lock(&po
->bind_lock
);
4358 register_prot_hook(sk
);
4360 spin_unlock(&po
->bind_lock
);
4361 if (pg_vec
&& (po
->tp_version
> TPACKET_V2
)) {
4362 /* Because we don't support block-based V3 on tx-ring */
4364 prb_shutdown_retire_blk_timer(po
, rb_queue
);
4368 free_pg_vec(pg_vec
, order
, req
->tp_block_nr
);
4374 static int packet_mmap(struct file
*file
, struct socket
*sock
,
4375 struct vm_area_struct
*vma
)
4377 struct sock
*sk
= sock
->sk
;
4378 struct packet_sock
*po
= pkt_sk(sk
);
4379 unsigned long size
, expected_size
;
4380 struct packet_ring_buffer
*rb
;
4381 unsigned long start
;
4388 mutex_lock(&po
->pg_vec_lock
);
4391 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
4393 expected_size
+= rb
->pg_vec_len
4399 if (expected_size
== 0)
4402 size
= vma
->vm_end
- vma
->vm_start
;
4403 if (size
!= expected_size
)
4406 start
= vma
->vm_start
;
4407 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
4408 if (rb
->pg_vec
== NULL
)
4411 for (i
= 0; i
< rb
->pg_vec_len
; i
++) {
4413 void *kaddr
= rb
->pg_vec
[i
].buffer
;
4416 for (pg_num
= 0; pg_num
< rb
->pg_vec_pages
; pg_num
++) {
4417 page
= pgv_to_page(kaddr
);
4418 err
= vm_insert_page(vma
, start
, page
);
4427 atomic_inc(&po
->mapped
);
4428 vma
->vm_ops
= &packet_mmap_ops
;
4432 mutex_unlock(&po
->pg_vec_lock
);
4436 static const struct proto_ops packet_ops_spkt
= {
4437 .family
= PF_PACKET
,
4438 .owner
= THIS_MODULE
,
4439 .release
= packet_release
,
4440 .bind
= packet_bind_spkt
,
4441 .connect
= sock_no_connect
,
4442 .socketpair
= sock_no_socketpair
,
4443 .accept
= sock_no_accept
,
4444 .getname
= packet_getname_spkt
,
4445 .poll
= datagram_poll
,
4446 .ioctl
= packet_ioctl
,
4447 .listen
= sock_no_listen
,
4448 .shutdown
= sock_no_shutdown
,
4449 .setsockopt
= sock_no_setsockopt
,
4450 .getsockopt
= sock_no_getsockopt
,
4451 .sendmsg
= packet_sendmsg_spkt
,
4452 .recvmsg
= packet_recvmsg
,
4453 .mmap
= sock_no_mmap
,
4454 .sendpage
= sock_no_sendpage
,
4457 static const struct proto_ops packet_ops
= {
4458 .family
= PF_PACKET
,
4459 .owner
= THIS_MODULE
,
4460 .release
= packet_release
,
4461 .bind
= packet_bind
,
4462 .connect
= sock_no_connect
,
4463 .socketpair
= sock_no_socketpair
,
4464 .accept
= sock_no_accept
,
4465 .getname
= packet_getname
,
4466 .poll
= packet_poll
,
4467 .ioctl
= packet_ioctl
,
4468 .listen
= sock_no_listen
,
4469 .shutdown
= sock_no_shutdown
,
4470 .setsockopt
= packet_setsockopt
,
4471 .getsockopt
= packet_getsockopt
,
4472 #ifdef CONFIG_COMPAT
4473 .compat_setsockopt
= compat_packet_setsockopt
,
4475 .sendmsg
= packet_sendmsg
,
4476 .recvmsg
= packet_recvmsg
,
4477 .mmap
= packet_mmap
,
4478 .sendpage
= sock_no_sendpage
,
4481 static const struct net_proto_family packet_family_ops
= {
4482 .family
= PF_PACKET
,
4483 .create
= packet_create
,
4484 .owner
= THIS_MODULE
,
4487 static struct notifier_block packet_netdev_notifier
= {
4488 .notifier_call
= packet_notifier
,
4491 #ifdef CONFIG_PROC_FS
4493 static void *packet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4496 struct net
*net
= seq_file_net(seq
);
4499 return seq_hlist_start_head_rcu(&net
->packet
.sklist
, *pos
);
4502 static void *packet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4504 struct net
*net
= seq_file_net(seq
);
4505 return seq_hlist_next_rcu(v
, &net
->packet
.sklist
, pos
);
4508 static void packet_seq_stop(struct seq_file
*seq
, void *v
)
4514 static int packet_seq_show(struct seq_file
*seq
, void *v
)
4516 if (v
== SEQ_START_TOKEN
)
4517 seq_puts(seq
, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4519 struct sock
*s
= sk_entry(v
);
4520 const struct packet_sock
*po
= pkt_sk(s
);
4523 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4525 refcount_read(&s
->sk_refcnt
),
4530 atomic_read(&s
->sk_rmem_alloc
),
4531 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(s
)),
4538 static const struct seq_operations packet_seq_ops
= {
4539 .start
= packet_seq_start
,
4540 .next
= packet_seq_next
,
4541 .stop
= packet_seq_stop
,
4542 .show
= packet_seq_show
,
4545 static int packet_seq_open(struct inode
*inode
, struct file
*file
)
4547 return seq_open_net(inode
, file
, &packet_seq_ops
,
4548 sizeof(struct seq_net_private
));
4551 static const struct file_operations packet_seq_fops
= {
4552 .owner
= THIS_MODULE
,
4553 .open
= packet_seq_open
,
4555 .llseek
= seq_lseek
,
4556 .release
= seq_release_net
,
4561 static int __net_init
packet_net_init(struct net
*net
)
4563 mutex_init(&net
->packet
.sklist_lock
);
4564 INIT_HLIST_HEAD(&net
->packet
.sklist
);
4566 if (!proc_create("packet", 0, net
->proc_net
, &packet_seq_fops
))
4572 static void __net_exit
packet_net_exit(struct net
*net
)
4574 remove_proc_entry("packet", net
->proc_net
);
4577 static struct pernet_operations packet_net_ops
= {
4578 .init
= packet_net_init
,
4579 .exit
= packet_net_exit
,
4583 static void __exit
packet_exit(void)
4585 unregister_netdevice_notifier(&packet_netdev_notifier
);
4586 unregister_pernet_subsys(&packet_net_ops
);
4587 sock_unregister(PF_PACKET
);
4588 proto_unregister(&packet_proto
);
4591 static int __init
packet_init(void)
4593 int rc
= proto_register(&packet_proto
, 0);
4598 sock_register(&packet_family_ops
);
4599 register_pernet_subsys(&packet_net_ops
);
4600 register_netdevice_notifier(&packet_netdev_notifier
);
4605 module_init(packet_init
);
4606 module_exit(packet_exit
);
4607 MODULE_LICENSE("GPL");
4608 MODULE_ALIAS_NETPROTO(PF_PACKET
);