2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
95 #include <linux/bpf.h>
101 - if device has no dev->hard_header routine, it adds and removes ll header
102 inside itself. In this case ll header is invisible outside of device,
103 but higher levels still should reserve dev->hard_header_len.
104 Some devices are enough clever to reallocate skb, when header
105 will not fit to reserved space (tunnel), another ones are silly
107 - packet socket receives packets with pulled ll header,
108 so that SOCK_RAW should push it back.
113 Incoming, dev->hard_header!=NULL
114 mac_header -> ll header
117 Outgoing, dev->hard_header!=NULL
118 mac_header -> ll header
121 Incoming, dev->hard_header==NULL
122 mac_header -> UNKNOWN position. It is very likely, that it points to ll
123 header. PPP makes it, that is wrong, because introduce
124 assymetry between rx and tx paths.
127 Outgoing, dev->hard_header==NULL
128 mac_header -> data. ll header is still not built!
132 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
138 dev->hard_header != NULL
139 mac_header -> ll header
142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
146 We should set nh.raw on output to correct posistion,
147 packet classifier depends on it.
150 /* Private packet socket structures. */
152 /* identical to struct packet_mreq except it has
153 * a longer address field.
155 struct packet_mreq_max
{
157 unsigned short mr_type
;
158 unsigned short mr_alen
;
159 unsigned char mr_address
[MAX_ADDR_LEN
];
163 struct tpacket_hdr
*h1
;
164 struct tpacket2_hdr
*h2
;
165 struct tpacket3_hdr
*h3
;
169 static int packet_set_ring(struct sock
*sk
, union tpacket_req_u
*req_u
,
170 int closing
, int tx_ring
);
172 #define V3_ALIGNMENT (8)
174 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 #define PGV_FROM_VMALLOC 1
181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
182 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
183 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
184 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
185 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
187 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
190 static int tpacket_snd(struct packet_sock
*po
, struct msghdr
*msg
);
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_kbdq_core
*,
199 struct tpacket_block_desc
*);
200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core
*,
201 struct packet_sock
*);
202 static void prb_retire_current_block(struct tpacket_kbdq_core
*,
203 struct packet_sock
*, unsigned int status
);
204 static int prb_queue_frozen(struct tpacket_kbdq_core
*);
205 static void prb_open_block(struct tpacket_kbdq_core
*,
206 struct tpacket_block_desc
*);
207 static void prb_retire_rx_blk_timer_expired(unsigned long);
208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core
*);
209 static void prb_init_blk_timer(struct packet_sock
*,
210 struct tpacket_kbdq_core
*,
211 void (*func
) (unsigned long));
212 static void prb_fill_rxhash(struct tpacket_kbdq_core
*, struct tpacket3_hdr
*);
213 static void prb_clear_rxhash(struct tpacket_kbdq_core
*,
214 struct tpacket3_hdr
*);
215 static void prb_fill_vlan_info(struct tpacket_kbdq_core
*,
216 struct tpacket3_hdr
*);
217 static void packet_flush_mclist(struct sock
*sk
);
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 netdev_features_t features
;
253 struct netdev_queue
*txq
;
254 int ret
= NETDEV_TX_BUSY
;
256 if (unlikely(!netif_running(dev
) ||
257 !netif_carrier_ok(dev
)))
260 features
= netif_skb_features(skb
);
261 if (skb_needs_linearize(skb
, features
) &&
262 __skb_linearize(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
));
413 WARN(1, "TPACKET version not supported.\n");
420 static int __packet_get_status(struct packet_sock
*po
, void *frame
)
422 union tpacket_uhdr h
;
427 switch (po
->tp_version
) {
429 flush_dcache_page(pgv_to_page(&h
.h1
->tp_status
));
430 return h
.h1
->tp_status
;
432 flush_dcache_page(pgv_to_page(&h
.h2
->tp_status
));
433 return h
.h2
->tp_status
;
436 WARN(1, "TPACKET version not supported.\n");
442 static __u32
tpacket_get_timestamp(struct sk_buff
*skb
, struct timespec
*ts
,
445 struct skb_shared_hwtstamps
*shhwtstamps
= skb_hwtstamps(skb
);
448 (flags
& SOF_TIMESTAMPING_RAW_HARDWARE
) &&
449 ktime_to_timespec_cond(shhwtstamps
->hwtstamp
, ts
))
450 return TP_STATUS_TS_RAW_HARDWARE
;
452 if (ktime_to_timespec_cond(skb
->tstamp
, ts
))
453 return TP_STATUS_TS_SOFTWARE
;
458 static __u32
__packet_set_timestamp(struct packet_sock
*po
, void *frame
,
461 union tpacket_uhdr h
;
465 if (!(ts_status
= tpacket_get_timestamp(skb
, &ts
, po
->tp_tstamp
)))
469 switch (po
->tp_version
) {
471 h
.h1
->tp_sec
= ts
.tv_sec
;
472 h
.h1
->tp_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
475 h
.h2
->tp_sec
= ts
.tv_sec
;
476 h
.h2
->tp_nsec
= ts
.tv_nsec
;
480 WARN(1, "TPACKET version not supported.\n");
484 /* one flush is safe, as both fields always lie on the same cacheline */
485 flush_dcache_page(pgv_to_page(&h
.h1
->tp_sec
));
491 static void *packet_lookup_frame(struct packet_sock
*po
,
492 struct packet_ring_buffer
*rb
,
493 unsigned int position
,
496 unsigned int pg_vec_pos
, frame_offset
;
497 union tpacket_uhdr h
;
499 pg_vec_pos
= position
/ rb
->frames_per_block
;
500 frame_offset
= position
% rb
->frames_per_block
;
502 h
.raw
= rb
->pg_vec
[pg_vec_pos
].buffer
+
503 (frame_offset
* rb
->frame_size
);
505 if (status
!= __packet_get_status(po
, h
.raw
))
511 static void *packet_current_frame(struct packet_sock
*po
,
512 struct packet_ring_buffer
*rb
,
515 return packet_lookup_frame(po
, rb
, rb
->head
, status
);
518 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core
*pkc
)
520 del_timer_sync(&pkc
->retire_blk_timer
);
523 static void prb_shutdown_retire_blk_timer(struct packet_sock
*po
,
524 struct sk_buff_head
*rb_queue
)
526 struct tpacket_kbdq_core
*pkc
;
528 pkc
= GET_PBDQC_FROM_RB(&po
->rx_ring
);
530 spin_lock_bh(&rb_queue
->lock
);
531 pkc
->delete_blk_timer
= 1;
532 spin_unlock_bh(&rb_queue
->lock
);
534 prb_del_retire_blk_timer(pkc
);
537 static void prb_init_blk_timer(struct packet_sock
*po
,
538 struct tpacket_kbdq_core
*pkc
,
539 void (*func
) (unsigned long))
541 init_timer(&pkc
->retire_blk_timer
);
542 pkc
->retire_blk_timer
.data
= (long)po
;
543 pkc
->retire_blk_timer
.function
= func
;
544 pkc
->retire_blk_timer
.expires
= jiffies
;
547 static void prb_setup_retire_blk_timer(struct packet_sock
*po
)
549 struct tpacket_kbdq_core
*pkc
;
551 pkc
= GET_PBDQC_FROM_RB(&po
->rx_ring
);
552 prb_init_blk_timer(po
, pkc
, prb_retire_rx_blk_timer_expired
);
555 static int prb_calc_retire_blk_tmo(struct packet_sock
*po
,
556 int blk_size_in_bytes
)
558 struct net_device
*dev
;
559 unsigned int mbits
= 0, msec
= 0, div
= 0, tmo
= 0;
560 struct ethtool_link_ksettings ecmd
;
564 dev
= __dev_get_by_index(sock_net(&po
->sk
), po
->ifindex
);
565 if (unlikely(!dev
)) {
567 return DEFAULT_PRB_RETIRE_TOV
;
569 err
= __ethtool_get_link_ksettings(dev
, &ecmd
);
573 * If the link speed is so slow you don't really
574 * need to worry about perf anyways
576 if (ecmd
.base
.speed
< SPEED_1000
||
577 ecmd
.base
.speed
== SPEED_UNKNOWN
) {
578 return DEFAULT_PRB_RETIRE_TOV
;
581 div
= ecmd
.base
.speed
/ 1000;
585 mbits
= (blk_size_in_bytes
* 8) / (1024 * 1024);
597 static void prb_init_ft_ops(struct tpacket_kbdq_core
*p1
,
598 union tpacket_req_u
*req_u
)
600 p1
->feature_req_word
= req_u
->req3
.tp_feature_req_word
;
603 static void init_prb_bdqc(struct packet_sock
*po
,
604 struct packet_ring_buffer
*rb
,
606 union tpacket_req_u
*req_u
)
608 struct tpacket_kbdq_core
*p1
= GET_PBDQC_FROM_RB(rb
);
609 struct tpacket_block_desc
*pbd
;
611 memset(p1
, 0x0, sizeof(*p1
));
613 p1
->knxt_seq_num
= 1;
615 pbd
= (struct tpacket_block_desc
*)pg_vec
[0].buffer
;
616 p1
->pkblk_start
= pg_vec
[0].buffer
;
617 p1
->kblk_size
= req_u
->req3
.tp_block_size
;
618 p1
->knum_blocks
= req_u
->req3
.tp_block_nr
;
619 p1
->hdrlen
= po
->tp_hdrlen
;
620 p1
->version
= po
->tp_version
;
621 p1
->last_kactive_blk_num
= 0;
622 po
->stats
.stats3
.tp_freeze_q_cnt
= 0;
623 if (req_u
->req3
.tp_retire_blk_tov
)
624 p1
->retire_blk_tov
= req_u
->req3
.tp_retire_blk_tov
;
626 p1
->retire_blk_tov
= prb_calc_retire_blk_tmo(po
,
627 req_u
->req3
.tp_block_size
);
628 p1
->tov_in_jiffies
= msecs_to_jiffies(p1
->retire_blk_tov
);
629 p1
->blk_sizeof_priv
= req_u
->req3
.tp_sizeof_priv
;
631 p1
->max_frame_len
= p1
->kblk_size
- BLK_PLUS_PRIV(p1
->blk_sizeof_priv
);
632 prb_init_ft_ops(p1
, req_u
);
633 prb_setup_retire_blk_timer(po
);
634 prb_open_block(p1
, pbd
);
637 /* Do NOT update the last_blk_num first.
638 * Assumes sk_buff_head lock is held.
640 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core
*pkc
)
642 mod_timer(&pkc
->retire_blk_timer
,
643 jiffies
+ pkc
->tov_in_jiffies
);
644 pkc
->last_kactive_blk_num
= pkc
->kactive_blk_num
;
649 * 1) We refresh the timer only when we open a block.
650 * By doing this we don't waste cycles refreshing the timer
651 * on packet-by-packet basis.
653 * With a 1MB block-size, on a 1Gbps line, it will take
654 * i) ~8 ms to fill a block + ii) memcpy etc.
655 * In this cut we are not accounting for the memcpy time.
657 * So, if the user sets the 'tmo' to 10ms then the timer
658 * will never fire while the block is still getting filled
659 * (which is what we want). However, the user could choose
660 * to close a block early and that's fine.
662 * But when the timer does fire, we check whether or not to refresh it.
663 * Since the tmo granularity is in msecs, it is not too expensive
664 * to refresh the timer, lets say every '8' msecs.
665 * Either the user can set the 'tmo' or we can derive it based on
666 * a) line-speed and b) block-size.
667 * prb_calc_retire_blk_tmo() calculates the tmo.
670 static void prb_retire_rx_blk_timer_expired(unsigned long data
)
672 struct packet_sock
*po
= (struct packet_sock
*)data
;
673 struct tpacket_kbdq_core
*pkc
= GET_PBDQC_FROM_RB(&po
->rx_ring
);
675 struct tpacket_block_desc
*pbd
;
677 spin_lock(&po
->sk
.sk_receive_queue
.lock
);
679 frozen
= prb_queue_frozen(pkc
);
680 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
682 if (unlikely(pkc
->delete_blk_timer
))
685 /* We only need to plug the race when the block is partially filled.
687 * lock(); increment BLOCK_NUM_PKTS; unlock()
688 * copy_bits() is in progress ...
689 * timer fires on other cpu:
690 * we can't retire the current block because copy_bits
694 if (BLOCK_NUM_PKTS(pbd
)) {
695 while (atomic_read(&pkc
->blk_fill_in_prog
)) {
696 /* Waiting for skb_copy_bits to finish... */
701 if (pkc
->last_kactive_blk_num
== pkc
->kactive_blk_num
) {
703 if (!BLOCK_NUM_PKTS(pbd
)) {
704 /* An empty block. Just refresh the timer. */
707 prb_retire_current_block(pkc
, po
, TP_STATUS_BLK_TMO
);
708 if (!prb_dispatch_next_block(pkc
, po
))
713 /* Case 1. Queue was frozen because user-space was
716 if (prb_curr_blk_in_use(pkc
, pbd
)) {
718 * Ok, user-space is still behind.
719 * So just refresh the timer.
723 /* Case 2. queue was frozen,user-space caught up,
724 * now the link went idle && the timer fired.
725 * We don't have a block to close.So we open this
726 * block and restart the timer.
727 * opening a block thaws the queue,restarts timer
728 * Thawing/timer-refresh is a side effect.
730 prb_open_block(pkc
, pbd
);
737 _prb_refresh_rx_retire_blk_timer(pkc
);
740 spin_unlock(&po
->sk
.sk_receive_queue
.lock
);
743 static void prb_flush_block(struct tpacket_kbdq_core
*pkc1
,
744 struct tpacket_block_desc
*pbd1
, __u32 status
)
746 /* Flush everything minus the block header */
748 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
753 /* Skip the block header(we know header WILL fit in 4K) */
756 end
= (u8
*)PAGE_ALIGN((unsigned long)pkc1
->pkblk_end
);
757 for (; start
< end
; start
+= PAGE_SIZE
)
758 flush_dcache_page(pgv_to_page(start
));
763 /* Now update the block status. */
765 BLOCK_STATUS(pbd1
) = status
;
767 /* Flush the block header */
769 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
771 flush_dcache_page(pgv_to_page(start
));
781 * 2) Increment active_blk_num
783 * Note:We DONT refresh the timer on purpose.
784 * Because almost always the next block will be opened.
786 static void prb_close_block(struct tpacket_kbdq_core
*pkc1
,
787 struct tpacket_block_desc
*pbd1
,
788 struct packet_sock
*po
, unsigned int stat
)
790 __u32 status
= TP_STATUS_USER
| stat
;
792 struct tpacket3_hdr
*last_pkt
;
793 struct tpacket_hdr_v1
*h1
= &pbd1
->hdr
.bh1
;
794 struct sock
*sk
= &po
->sk
;
796 if (po
->stats
.stats3
.tp_drops
)
797 status
|= TP_STATUS_LOSING
;
799 last_pkt
= (struct tpacket3_hdr
*)pkc1
->prev
;
800 last_pkt
->tp_next_offset
= 0;
802 /* Get the ts of the last pkt */
803 if (BLOCK_NUM_PKTS(pbd1
)) {
804 h1
->ts_last_pkt
.ts_sec
= last_pkt
->tp_sec
;
805 h1
->ts_last_pkt
.ts_nsec
= last_pkt
->tp_nsec
;
807 /* Ok, we tmo'd - so get the current time.
809 * It shouldn't really happen as we don't close empty
810 * blocks. See prb_retire_rx_blk_timer_expired().
814 h1
->ts_last_pkt
.ts_sec
= ts
.tv_sec
;
815 h1
->ts_last_pkt
.ts_nsec
= ts
.tv_nsec
;
820 /* Flush the block */
821 prb_flush_block(pkc1
, pbd1
, status
);
823 sk
->sk_data_ready(sk
);
825 pkc1
->kactive_blk_num
= GET_NEXT_PRB_BLK_NUM(pkc1
);
828 static void prb_thaw_queue(struct tpacket_kbdq_core
*pkc
)
830 pkc
->reset_pending_on_curr_blk
= 0;
834 * Side effect of opening a block:
836 * 1) prb_queue is thawed.
837 * 2) retire_blk_timer is refreshed.
840 static void prb_open_block(struct tpacket_kbdq_core
*pkc1
,
841 struct tpacket_block_desc
*pbd1
)
844 struct tpacket_hdr_v1
*h1
= &pbd1
->hdr
.bh1
;
848 /* We could have just memset this but we will lose the
849 * flexibility of making the priv area sticky
852 BLOCK_SNUM(pbd1
) = pkc1
->knxt_seq_num
++;
853 BLOCK_NUM_PKTS(pbd1
) = 0;
854 BLOCK_LEN(pbd1
) = BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
858 h1
->ts_first_pkt
.ts_sec
= ts
.tv_sec
;
859 h1
->ts_first_pkt
.ts_nsec
= ts
.tv_nsec
;
861 pkc1
->pkblk_start
= (char *)pbd1
;
862 pkc1
->nxt_offset
= pkc1
->pkblk_start
+ BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
864 BLOCK_O2FP(pbd1
) = (__u32
)BLK_PLUS_PRIV(pkc1
->blk_sizeof_priv
);
865 BLOCK_O2PRIV(pbd1
) = BLK_HDR_LEN
;
867 pbd1
->version
= pkc1
->version
;
868 pkc1
->prev
= pkc1
->nxt_offset
;
869 pkc1
->pkblk_end
= pkc1
->pkblk_start
+ pkc1
->kblk_size
;
871 prb_thaw_queue(pkc1
);
872 _prb_refresh_rx_retire_blk_timer(pkc1
);
878 * Queue freeze logic:
879 * 1) Assume tp_block_nr = 8 blocks.
880 * 2) At time 't0', user opens Rx ring.
881 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
882 * 4) user-space is either sleeping or processing block '0'.
883 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
884 * it will close block-7,loop around and try to fill block '0'.
886 * __packet_lookup_frame_in_block
887 * prb_retire_current_block()
888 * prb_dispatch_next_block()
889 * |->(BLOCK_STATUS == USER) evaluates to true
890 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
891 * 6) Now there are two cases:
892 * 6.1) Link goes idle right after the queue is frozen.
893 * But remember, the last open_block() refreshed the timer.
894 * When this timer expires,it will refresh itself so that we can
895 * re-open block-0 in near future.
896 * 6.2) Link is busy and keeps on receiving packets. This is a simple
897 * case and __packet_lookup_frame_in_block will check if block-0
898 * is free and can now be re-used.
900 static void prb_freeze_queue(struct tpacket_kbdq_core
*pkc
,
901 struct packet_sock
*po
)
903 pkc
->reset_pending_on_curr_blk
= 1;
904 po
->stats
.stats3
.tp_freeze_q_cnt
++;
907 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
910 * If the next block is free then we will dispatch it
911 * and return a good offset.
912 * Else, we will freeze the queue.
913 * So, caller must check the return value.
915 static void *prb_dispatch_next_block(struct tpacket_kbdq_core
*pkc
,
916 struct packet_sock
*po
)
918 struct tpacket_block_desc
*pbd
;
922 /* 1. Get current block num */
923 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
925 /* 2. If this block is currently in_use then freeze the queue */
926 if (TP_STATUS_USER
& BLOCK_STATUS(pbd
)) {
927 prb_freeze_queue(pkc
, po
);
933 * open this block and return the offset where the first packet
934 * needs to get stored.
936 prb_open_block(pkc
, pbd
);
937 return (void *)pkc
->nxt_offset
;
940 static void prb_retire_current_block(struct tpacket_kbdq_core
*pkc
,
941 struct packet_sock
*po
, unsigned int status
)
943 struct tpacket_block_desc
*pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
945 /* retire/close the current block */
946 if (likely(TP_STATUS_KERNEL
== BLOCK_STATUS(pbd
))) {
948 * Plug the case where copy_bits() is in progress on
949 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
950 * have space to copy the pkt in the current block and
951 * called prb_retire_current_block()
953 * We don't need to worry about the TMO case because
954 * the timer-handler already handled this case.
956 if (!(status
& TP_STATUS_BLK_TMO
)) {
957 while (atomic_read(&pkc
->blk_fill_in_prog
)) {
958 /* Waiting for skb_copy_bits to finish... */
962 prb_close_block(pkc
, pbd
, po
, status
);
967 static int prb_curr_blk_in_use(struct tpacket_kbdq_core
*pkc
,
968 struct tpacket_block_desc
*pbd
)
970 return TP_STATUS_USER
& BLOCK_STATUS(pbd
);
973 static int prb_queue_frozen(struct tpacket_kbdq_core
*pkc
)
975 return pkc
->reset_pending_on_curr_blk
;
978 static void prb_clear_blk_fill_status(struct packet_ring_buffer
*rb
)
980 struct tpacket_kbdq_core
*pkc
= GET_PBDQC_FROM_RB(rb
);
981 atomic_dec(&pkc
->blk_fill_in_prog
);
984 static void prb_fill_rxhash(struct tpacket_kbdq_core
*pkc
,
985 struct tpacket3_hdr
*ppd
)
987 ppd
->hv1
.tp_rxhash
= skb_get_hash(pkc
->skb
);
990 static void prb_clear_rxhash(struct tpacket_kbdq_core
*pkc
,
991 struct tpacket3_hdr
*ppd
)
993 ppd
->hv1
.tp_rxhash
= 0;
996 static void prb_fill_vlan_info(struct tpacket_kbdq_core
*pkc
,
997 struct tpacket3_hdr
*ppd
)
999 if (skb_vlan_tag_present(pkc
->skb
)) {
1000 ppd
->hv1
.tp_vlan_tci
= skb_vlan_tag_get(pkc
->skb
);
1001 ppd
->hv1
.tp_vlan_tpid
= ntohs(pkc
->skb
->vlan_proto
);
1002 ppd
->tp_status
= TP_STATUS_VLAN_VALID
| TP_STATUS_VLAN_TPID_VALID
;
1004 ppd
->hv1
.tp_vlan_tci
= 0;
1005 ppd
->hv1
.tp_vlan_tpid
= 0;
1006 ppd
->tp_status
= TP_STATUS_AVAILABLE
;
1010 static void prb_run_all_ft_ops(struct tpacket_kbdq_core
*pkc
,
1011 struct tpacket3_hdr
*ppd
)
1013 ppd
->hv1
.tp_padding
= 0;
1014 prb_fill_vlan_info(pkc
, ppd
);
1016 if (pkc
->feature_req_word
& TP_FT_REQ_FILL_RXHASH
)
1017 prb_fill_rxhash(pkc
, ppd
);
1019 prb_clear_rxhash(pkc
, ppd
);
1022 static void prb_fill_curr_block(char *curr
,
1023 struct tpacket_kbdq_core
*pkc
,
1024 struct tpacket_block_desc
*pbd
,
1027 struct tpacket3_hdr
*ppd
;
1029 ppd
= (struct tpacket3_hdr
*)curr
;
1030 ppd
->tp_next_offset
= TOTAL_PKT_LEN_INCL_ALIGN(len
);
1032 pkc
->nxt_offset
+= TOTAL_PKT_LEN_INCL_ALIGN(len
);
1033 BLOCK_LEN(pbd
) += TOTAL_PKT_LEN_INCL_ALIGN(len
);
1034 BLOCK_NUM_PKTS(pbd
) += 1;
1035 atomic_inc(&pkc
->blk_fill_in_prog
);
1036 prb_run_all_ft_ops(pkc
, ppd
);
1039 /* Assumes caller has the sk->rx_queue.lock */
1040 static void *__packet_lookup_frame_in_block(struct packet_sock
*po
,
1041 struct sk_buff
*skb
,
1046 struct tpacket_kbdq_core
*pkc
;
1047 struct tpacket_block_desc
*pbd
;
1050 pkc
= GET_PBDQC_FROM_RB(&po
->rx_ring
);
1051 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
1053 /* Queue is frozen when user space is lagging behind */
1054 if (prb_queue_frozen(pkc
)) {
1056 * Check if that last block which caused the queue to freeze,
1057 * is still in_use by user-space.
1059 if (prb_curr_blk_in_use(pkc
, pbd
)) {
1060 /* Can't record this packet */
1064 * Ok, the block was released by user-space.
1065 * Now let's open that block.
1066 * opening a block also thaws the queue.
1067 * Thawing is a side effect.
1069 prb_open_block(pkc
, pbd
);
1074 curr
= pkc
->nxt_offset
;
1076 end
= (char *)pbd
+ pkc
->kblk_size
;
1078 /* first try the current block */
1079 if (curr
+TOTAL_PKT_LEN_INCL_ALIGN(len
) < end
) {
1080 prb_fill_curr_block(curr
, pkc
, pbd
, len
);
1081 return (void *)curr
;
1084 /* Ok, close the current block */
1085 prb_retire_current_block(pkc
, po
, 0);
1087 /* Now, try to dispatch the next block */
1088 curr
= (char *)prb_dispatch_next_block(pkc
, po
);
1090 pbd
= GET_CURR_PBLOCK_DESC_FROM_CORE(pkc
);
1091 prb_fill_curr_block(curr
, pkc
, pbd
, len
);
1092 return (void *)curr
;
1096 * No free blocks are available.user_space hasn't caught up yet.
1097 * Queue was just frozen and now this packet will get dropped.
1102 static void *packet_current_rx_frame(struct packet_sock
*po
,
1103 struct sk_buff
*skb
,
1104 int status
, unsigned int len
)
1107 switch (po
->tp_version
) {
1110 curr
= packet_lookup_frame(po
, &po
->rx_ring
,
1111 po
->rx_ring
.head
, status
);
1114 return __packet_lookup_frame_in_block(po
, skb
, status
, len
);
1116 WARN(1, "TPACKET version not supported\n");
1122 static void *prb_lookup_block(struct packet_sock
*po
,
1123 struct packet_ring_buffer
*rb
,
1127 struct tpacket_kbdq_core
*pkc
= GET_PBDQC_FROM_RB(rb
);
1128 struct tpacket_block_desc
*pbd
= GET_PBLOCK_DESC(pkc
, idx
);
1130 if (status
!= BLOCK_STATUS(pbd
))
1135 static int prb_previous_blk_num(struct packet_ring_buffer
*rb
)
1138 if (rb
->prb_bdqc
.kactive_blk_num
)
1139 prev
= rb
->prb_bdqc
.kactive_blk_num
-1;
1141 prev
= rb
->prb_bdqc
.knum_blocks
-1;
1145 /* Assumes caller has held the rx_queue.lock */
1146 static void *__prb_previous_block(struct packet_sock
*po
,
1147 struct packet_ring_buffer
*rb
,
1150 unsigned int previous
= prb_previous_blk_num(rb
);
1151 return prb_lookup_block(po
, rb
, previous
, status
);
1154 static void *packet_previous_rx_frame(struct packet_sock
*po
,
1155 struct packet_ring_buffer
*rb
,
1158 if (po
->tp_version
<= TPACKET_V2
)
1159 return packet_previous_frame(po
, rb
, status
);
1161 return __prb_previous_block(po
, rb
, status
);
1164 static void packet_increment_rx_head(struct packet_sock
*po
,
1165 struct packet_ring_buffer
*rb
)
1167 switch (po
->tp_version
) {
1170 return packet_increment_head(rb
);
1173 WARN(1, "TPACKET version not supported.\n");
1179 static void *packet_previous_frame(struct packet_sock
*po
,
1180 struct packet_ring_buffer
*rb
,
1183 unsigned int previous
= rb
->head
? rb
->head
- 1 : rb
->frame_max
;
1184 return packet_lookup_frame(po
, rb
, previous
, status
);
1187 static void packet_increment_head(struct packet_ring_buffer
*buff
)
1189 buff
->head
= buff
->head
!= buff
->frame_max
? buff
->head
+1 : 0;
1192 static void packet_inc_pending(struct packet_ring_buffer
*rb
)
1194 this_cpu_inc(*rb
->pending_refcnt
);
1197 static void packet_dec_pending(struct packet_ring_buffer
*rb
)
1199 this_cpu_dec(*rb
->pending_refcnt
);
1202 static unsigned int packet_read_pending(const struct packet_ring_buffer
*rb
)
1204 unsigned int refcnt
= 0;
1207 /* We don't use pending refcount in rx_ring. */
1208 if (rb
->pending_refcnt
== NULL
)
1211 for_each_possible_cpu(cpu
)
1212 refcnt
+= *per_cpu_ptr(rb
->pending_refcnt
, cpu
);
1217 static int packet_alloc_pending(struct packet_sock
*po
)
1219 po
->rx_ring
.pending_refcnt
= NULL
;
1221 po
->tx_ring
.pending_refcnt
= alloc_percpu(unsigned int);
1222 if (unlikely(po
->tx_ring
.pending_refcnt
== NULL
))
1228 static void packet_free_pending(struct packet_sock
*po
)
1230 free_percpu(po
->tx_ring
.pending_refcnt
);
1233 #define ROOM_POW_OFF 2
1234 #define ROOM_NONE 0x0
1235 #define ROOM_LOW 0x1
1236 #define ROOM_NORMAL 0x2
1238 static bool __tpacket_has_room(struct packet_sock
*po
, int pow_off
)
1242 len
= po
->rx_ring
.frame_max
+ 1;
1243 idx
= po
->rx_ring
.head
;
1245 idx
+= len
>> pow_off
;
1248 return packet_lookup_frame(po
, &po
->rx_ring
, idx
, TP_STATUS_KERNEL
);
1251 static bool __tpacket_v3_has_room(struct packet_sock
*po
, int pow_off
)
1255 len
= po
->rx_ring
.prb_bdqc
.knum_blocks
;
1256 idx
= po
->rx_ring
.prb_bdqc
.kactive_blk_num
;
1258 idx
+= len
>> pow_off
;
1261 return prb_lookup_block(po
, &po
->rx_ring
, idx
, TP_STATUS_KERNEL
);
1264 static int __packet_rcv_has_room(struct packet_sock
*po
, struct sk_buff
*skb
)
1266 struct sock
*sk
= &po
->sk
;
1267 int ret
= ROOM_NONE
;
1269 if (po
->prot_hook
.func
!= tpacket_rcv
) {
1270 int avail
= sk
->sk_rcvbuf
- atomic_read(&sk
->sk_rmem_alloc
)
1271 - (skb
? skb
->truesize
: 0);
1272 if (avail
> (sk
->sk_rcvbuf
>> ROOM_POW_OFF
))
1280 if (po
->tp_version
== TPACKET_V3
) {
1281 if (__tpacket_v3_has_room(po
, ROOM_POW_OFF
))
1283 else if (__tpacket_v3_has_room(po
, 0))
1286 if (__tpacket_has_room(po
, ROOM_POW_OFF
))
1288 else if (__tpacket_has_room(po
, 0))
1295 static int packet_rcv_has_room(struct packet_sock
*po
, struct sk_buff
*skb
)
1300 spin_lock_bh(&po
->sk
.sk_receive_queue
.lock
);
1301 ret
= __packet_rcv_has_room(po
, skb
);
1302 has_room
= ret
== ROOM_NORMAL
;
1303 if (po
->pressure
== has_room
)
1304 po
->pressure
= !has_room
;
1305 spin_unlock_bh(&po
->sk
.sk_receive_queue
.lock
);
1310 static void packet_sock_destruct(struct sock
*sk
)
1312 skb_queue_purge(&sk
->sk_error_queue
);
1314 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
1315 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
1317 if (!sock_flag(sk
, SOCK_DEAD
)) {
1318 pr_err("Attempt to release alive packet socket: %p\n", sk
);
1322 sk_refcnt_debug_dec(sk
);
1325 static bool fanout_flow_is_huge(struct packet_sock
*po
, struct sk_buff
*skb
)
1330 rxhash
= skb_get_hash(skb
);
1331 for (i
= 0; i
< ROLLOVER_HLEN
; i
++)
1332 if (po
->rollover
->history
[i
] == rxhash
)
1335 po
->rollover
->history
[prandom_u32() % ROLLOVER_HLEN
] = rxhash
;
1336 return count
> (ROLLOVER_HLEN
>> 1);
1339 static unsigned int fanout_demux_hash(struct packet_fanout
*f
,
1340 struct sk_buff
*skb
,
1343 return reciprocal_scale(skb_get_hash(skb
), num
);
1346 static unsigned int fanout_demux_lb(struct packet_fanout
*f
,
1347 struct sk_buff
*skb
,
1350 unsigned int val
= atomic_inc_return(&f
->rr_cur
);
1355 static unsigned int fanout_demux_cpu(struct packet_fanout
*f
,
1356 struct sk_buff
*skb
,
1359 return smp_processor_id() % num
;
1362 static unsigned int fanout_demux_rnd(struct packet_fanout
*f
,
1363 struct sk_buff
*skb
,
1366 return prandom_u32_max(num
);
1369 static unsigned int fanout_demux_rollover(struct packet_fanout
*f
,
1370 struct sk_buff
*skb
,
1371 unsigned int idx
, bool try_self
,
1374 struct packet_sock
*po
, *po_next
, *po_skip
= NULL
;
1375 unsigned int i
, j
, room
= ROOM_NONE
;
1377 po
= pkt_sk(f
->arr
[idx
]);
1380 room
= packet_rcv_has_room(po
, skb
);
1381 if (room
== ROOM_NORMAL
||
1382 (room
== ROOM_LOW
&& !fanout_flow_is_huge(po
, skb
)))
1387 i
= j
= min_t(int, po
->rollover
->sock
, num
- 1);
1389 po_next
= pkt_sk(f
->arr
[i
]);
1390 if (po_next
!= po_skip
&& !po_next
->pressure
&&
1391 packet_rcv_has_room(po_next
, skb
) == ROOM_NORMAL
) {
1393 po
->rollover
->sock
= i
;
1394 atomic_long_inc(&po
->rollover
->num
);
1395 if (room
== ROOM_LOW
)
1396 atomic_long_inc(&po
->rollover
->num_huge
);
1404 atomic_long_inc(&po
->rollover
->num_failed
);
1408 static unsigned int fanout_demux_qm(struct packet_fanout
*f
,
1409 struct sk_buff
*skb
,
1412 return skb_get_queue_mapping(skb
) % num
;
1415 static unsigned int fanout_demux_bpf(struct packet_fanout
*f
,
1416 struct sk_buff
*skb
,
1419 struct bpf_prog
*prog
;
1420 unsigned int ret
= 0;
1423 prog
= rcu_dereference(f
->bpf_prog
);
1425 ret
= bpf_prog_run_clear_cb(prog
, skb
) % num
;
1431 static bool fanout_has_flag(struct packet_fanout
*f
, u16 flag
)
1433 return f
->flags
& (flag
>> 8);
1436 static int packet_rcv_fanout(struct sk_buff
*skb
, struct net_device
*dev
,
1437 struct packet_type
*pt
, struct net_device
*orig_dev
)
1439 struct packet_fanout
*f
= pt
->af_packet_priv
;
1440 unsigned int num
= READ_ONCE(f
->num_members
);
1441 struct net
*net
= read_pnet(&f
->net
);
1442 struct packet_sock
*po
;
1445 if (!net_eq(dev_net(dev
), net
) || !num
) {
1450 if (fanout_has_flag(f
, PACKET_FANOUT_FLAG_DEFRAG
)) {
1451 skb
= ip_check_defrag(net
, skb
, IP_DEFRAG_AF_PACKET
);
1456 case PACKET_FANOUT_HASH
:
1458 idx
= fanout_demux_hash(f
, skb
, num
);
1460 case PACKET_FANOUT_LB
:
1461 idx
= fanout_demux_lb(f
, skb
, num
);
1463 case PACKET_FANOUT_CPU
:
1464 idx
= fanout_demux_cpu(f
, skb
, num
);
1466 case PACKET_FANOUT_RND
:
1467 idx
= fanout_demux_rnd(f
, skb
, num
);
1469 case PACKET_FANOUT_QM
:
1470 idx
= fanout_demux_qm(f
, skb
, num
);
1472 case PACKET_FANOUT_ROLLOVER
:
1473 idx
= fanout_demux_rollover(f
, skb
, 0, false, num
);
1475 case PACKET_FANOUT_CBPF
:
1476 case PACKET_FANOUT_EBPF
:
1477 idx
= fanout_demux_bpf(f
, skb
, num
);
1481 if (fanout_has_flag(f
, PACKET_FANOUT_FLAG_ROLLOVER
))
1482 idx
= fanout_demux_rollover(f
, skb
, idx
, true, num
);
1484 po
= pkt_sk(f
->arr
[idx
]);
1485 return po
->prot_hook
.func(skb
, dev
, &po
->prot_hook
, orig_dev
);
1488 DEFINE_MUTEX(fanout_mutex
);
1489 EXPORT_SYMBOL_GPL(fanout_mutex
);
1490 static LIST_HEAD(fanout_list
);
1492 static void __fanout_link(struct sock
*sk
, struct packet_sock
*po
)
1494 struct packet_fanout
*f
= po
->fanout
;
1496 spin_lock(&f
->lock
);
1497 f
->arr
[f
->num_members
] = sk
;
1500 spin_unlock(&f
->lock
);
1503 static void __fanout_unlink(struct sock
*sk
, struct packet_sock
*po
)
1505 struct packet_fanout
*f
= po
->fanout
;
1508 spin_lock(&f
->lock
);
1509 for (i
= 0; i
< f
->num_members
; i
++) {
1510 if (f
->arr
[i
] == sk
)
1513 BUG_ON(i
>= f
->num_members
);
1514 f
->arr
[i
] = f
->arr
[f
->num_members
- 1];
1516 spin_unlock(&f
->lock
);
1519 static bool match_fanout_group(struct packet_type
*ptype
, struct sock
*sk
)
1521 if (sk
->sk_family
!= PF_PACKET
)
1524 return ptype
->af_packet_priv
== pkt_sk(sk
)->fanout
;
1527 static void fanout_init_data(struct packet_fanout
*f
)
1530 case PACKET_FANOUT_LB
:
1531 atomic_set(&f
->rr_cur
, 0);
1533 case PACKET_FANOUT_CBPF
:
1534 case PACKET_FANOUT_EBPF
:
1535 RCU_INIT_POINTER(f
->bpf_prog
, NULL
);
1540 static void __fanout_set_data_bpf(struct packet_fanout
*f
, struct bpf_prog
*new)
1542 struct bpf_prog
*old
;
1544 spin_lock(&f
->lock
);
1545 old
= rcu_dereference_protected(f
->bpf_prog
, lockdep_is_held(&f
->lock
));
1546 rcu_assign_pointer(f
->bpf_prog
, new);
1547 spin_unlock(&f
->lock
);
1551 bpf_prog_destroy(old
);
1555 static int fanout_set_data_cbpf(struct packet_sock
*po
, char __user
*data
,
1558 struct bpf_prog
*new;
1559 struct sock_fprog fprog
;
1562 if (sock_flag(&po
->sk
, SOCK_FILTER_LOCKED
))
1564 if (len
!= sizeof(fprog
))
1566 if (copy_from_user(&fprog
, data
, len
))
1569 ret
= bpf_prog_create_from_user(&new, &fprog
, NULL
, false);
1573 __fanout_set_data_bpf(po
->fanout
, new);
1577 static int fanout_set_data_ebpf(struct packet_sock
*po
, char __user
*data
,
1580 struct bpf_prog
*new;
1583 if (sock_flag(&po
->sk
, SOCK_FILTER_LOCKED
))
1585 if (len
!= sizeof(fd
))
1587 if (copy_from_user(&fd
, data
, len
))
1590 new = bpf_prog_get(fd
);
1592 return PTR_ERR(new);
1593 if (new->type
!= BPF_PROG_TYPE_SOCKET_FILTER
) {
1598 __fanout_set_data_bpf(po
->fanout
, new);
1602 static int fanout_set_data(struct packet_sock
*po
, char __user
*data
,
1605 switch (po
->fanout
->type
) {
1606 case PACKET_FANOUT_CBPF
:
1607 return fanout_set_data_cbpf(po
, data
, len
);
1608 case PACKET_FANOUT_EBPF
:
1609 return fanout_set_data_ebpf(po
, data
, len
);
1615 static void fanout_release_data(struct packet_fanout
*f
)
1618 case PACKET_FANOUT_CBPF
:
1619 case PACKET_FANOUT_EBPF
:
1620 __fanout_set_data_bpf(f
, NULL
);
1624 static int fanout_add(struct sock
*sk
, u16 id
, u16 type_flags
)
1626 struct packet_sock
*po
= pkt_sk(sk
);
1627 struct packet_fanout
*f
, *match
;
1628 u8 type
= type_flags
& 0xff;
1629 u8 flags
= type_flags
>> 8;
1633 case PACKET_FANOUT_ROLLOVER
:
1634 if (type_flags
& PACKET_FANOUT_FLAG_ROLLOVER
)
1636 case PACKET_FANOUT_HASH
:
1637 case PACKET_FANOUT_LB
:
1638 case PACKET_FANOUT_CPU
:
1639 case PACKET_FANOUT_RND
:
1640 case PACKET_FANOUT_QM
:
1641 case PACKET_FANOUT_CBPF
:
1642 case PACKET_FANOUT_EBPF
:
1654 if (type
== PACKET_FANOUT_ROLLOVER
||
1655 (type_flags
& PACKET_FANOUT_FLAG_ROLLOVER
)) {
1656 po
->rollover
= kzalloc(sizeof(*po
->rollover
), GFP_KERNEL
);
1659 atomic_long_set(&po
->rollover
->num
, 0);
1660 atomic_long_set(&po
->rollover
->num_huge
, 0);
1661 atomic_long_set(&po
->rollover
->num_failed
, 0);
1664 mutex_lock(&fanout_mutex
);
1666 list_for_each_entry(f
, &fanout_list
, list
) {
1668 read_pnet(&f
->net
) == sock_net(sk
)) {
1674 if (match
&& match
->flags
!= flags
)
1678 match
= kzalloc(sizeof(*match
), GFP_KERNEL
);
1681 write_pnet(&match
->net
, sock_net(sk
));
1684 match
->flags
= flags
;
1685 INIT_LIST_HEAD(&match
->list
);
1686 spin_lock_init(&match
->lock
);
1687 atomic_set(&match
->sk_ref
, 0);
1688 fanout_init_data(match
);
1689 match
->prot_hook
.type
= po
->prot_hook
.type
;
1690 match
->prot_hook
.dev
= po
->prot_hook
.dev
;
1691 match
->prot_hook
.func
= packet_rcv_fanout
;
1692 match
->prot_hook
.af_packet_priv
= match
;
1693 match
->prot_hook
.id_match
= match_fanout_group
;
1694 dev_add_pack(&match
->prot_hook
);
1695 list_add(&match
->list
, &fanout_list
);
1698 if (match
->type
== type
&&
1699 match
->prot_hook
.type
== po
->prot_hook
.type
&&
1700 match
->prot_hook
.dev
== po
->prot_hook
.dev
) {
1702 if (atomic_read(&match
->sk_ref
) < PACKET_FANOUT_MAX
) {
1703 __dev_remove_pack(&po
->prot_hook
);
1705 atomic_inc(&match
->sk_ref
);
1706 __fanout_link(sk
, po
);
1711 mutex_unlock(&fanout_mutex
);
1713 kfree(po
->rollover
);
1714 po
->rollover
= NULL
;
1719 static void fanout_release(struct sock
*sk
)
1721 struct packet_sock
*po
= pkt_sk(sk
);
1722 struct packet_fanout
*f
;
1728 mutex_lock(&fanout_mutex
);
1731 if (atomic_dec_and_test(&f
->sk_ref
)) {
1733 dev_remove_pack(&f
->prot_hook
);
1734 fanout_release_data(f
);
1737 mutex_unlock(&fanout_mutex
);
1740 kfree_rcu(po
->rollover
, rcu
);
1743 static bool packet_extra_vlan_len_allowed(const struct net_device
*dev
,
1744 struct sk_buff
*skb
)
1746 /* Earlier code assumed this would be a VLAN pkt, double-check
1747 * this now that we have the actual packet in hand. We can only
1748 * do this check on Ethernet devices.
1750 if (unlikely(dev
->type
!= ARPHRD_ETHER
))
1753 skb_reset_mac_header(skb
);
1754 return likely(eth_hdr(skb
)->h_proto
== htons(ETH_P_8021Q
));
1757 static const struct proto_ops packet_ops
;
1759 static const struct proto_ops packet_ops_spkt
;
1761 static int packet_rcv_spkt(struct sk_buff
*skb
, struct net_device
*dev
,
1762 struct packet_type
*pt
, struct net_device
*orig_dev
)
1765 struct sockaddr_pkt
*spkt
;
1768 * When we registered the protocol we saved the socket in the data
1769 * field for just this event.
1772 sk
= pt
->af_packet_priv
;
1775 * Yank back the headers [hope the device set this
1776 * right or kerboom...]
1778 * Incoming packets have ll header pulled,
1781 * For outgoing ones skb->data == skb_mac_header(skb)
1782 * so that this procedure is noop.
1785 if (skb
->pkt_type
== PACKET_LOOPBACK
)
1788 if (!net_eq(dev_net(dev
), sock_net(sk
)))
1791 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1795 /* drop any routing info */
1798 /* drop conntrack reference */
1801 spkt
= &PACKET_SKB_CB(skb
)->sa
.pkt
;
1803 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1806 * The SOCK_PACKET socket receives _all_ frames.
1809 spkt
->spkt_family
= dev
->type
;
1810 strlcpy(spkt
->spkt_device
, dev
->name
, sizeof(spkt
->spkt_device
));
1811 spkt
->spkt_protocol
= skb
->protocol
;
1814 * Charge the memory to the socket. This is done specifically
1815 * to prevent sockets using all the memory up.
1818 if (sock_queue_rcv_skb(sk
, skb
) == 0)
1829 * Output a raw packet to a device layer. This bypasses all the other
1830 * protocol layers and you must therefore supply it with a complete frame
1833 static int packet_sendmsg_spkt(struct socket
*sock
, struct msghdr
*msg
,
1836 struct sock
*sk
= sock
->sk
;
1837 DECLARE_SOCKADDR(struct sockaddr_pkt
*, saddr
, msg
->msg_name
);
1838 struct sk_buff
*skb
= NULL
;
1839 struct net_device
*dev
;
1845 * Get and verify the address.
1849 if (msg
->msg_namelen
< sizeof(struct sockaddr
))
1851 if (msg
->msg_namelen
== sizeof(struct sockaddr_pkt
))
1852 proto
= saddr
->spkt_protocol
;
1854 return -ENOTCONN
; /* SOCK_PACKET must be sent giving an address */
1857 * Find the device first to size check it
1860 saddr
->spkt_device
[sizeof(saddr
->spkt_device
) - 1] = 0;
1863 dev
= dev_get_by_name_rcu(sock_net(sk
), saddr
->spkt_device
);
1869 if (!(dev
->flags
& IFF_UP
))
1873 * You may not queue a frame bigger than the mtu. This is the lowest level
1874 * raw protocol and you must do your own fragmentation at this level.
1877 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
1878 if (!netif_supports_nofcs(dev
)) {
1879 err
= -EPROTONOSUPPORT
;
1882 extra_len
= 4; /* We're doing our own CRC */
1886 if (len
> dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
+ extra_len
)
1890 size_t reserved
= LL_RESERVED_SPACE(dev
);
1891 int tlen
= dev
->needed_tailroom
;
1892 unsigned int hhlen
= dev
->header_ops
? dev
->hard_header_len
: 0;
1895 skb
= sock_wmalloc(sk
, len
+ reserved
+ tlen
, 0, GFP_KERNEL
);
1898 /* FIXME: Save some space for broken drivers that write a hard
1899 * header at transmission time by themselves. PPP is the notable
1900 * one here. This should really be fixed at the driver level.
1902 skb_reserve(skb
, reserved
);
1903 skb_reset_network_header(skb
);
1905 /* Try to align data part correctly */
1910 skb_reset_network_header(skb
);
1912 err
= memcpy_from_msg(skb_put(skb
, len
), msg
, len
);
1918 if (len
> (dev
->mtu
+ dev
->hard_header_len
+ extra_len
) &&
1919 !packet_extra_vlan_len_allowed(dev
, skb
)) {
1924 skb
->protocol
= proto
;
1926 skb
->priority
= sk
->sk_priority
;
1927 skb
->mark
= sk
->sk_mark
;
1929 sock_tx_timestamp(sk
, &skb_shinfo(skb
)->tx_flags
);
1931 if (unlikely(extra_len
== 4))
1934 skb_probe_transport_header(skb
, 0);
1936 dev_queue_xmit(skb
);
1947 static unsigned int run_filter(struct sk_buff
*skb
,
1948 const struct sock
*sk
,
1951 struct sk_filter
*filter
;
1954 filter
= rcu_dereference(sk
->sk_filter
);
1956 res
= bpf_prog_run_clear_cb(filter
->prog
, skb
);
1962 static int __packet_rcv_vnet(const struct sk_buff
*skb
,
1963 struct virtio_net_hdr
*vnet_hdr
)
1965 *vnet_hdr
= (const struct virtio_net_hdr
) { 0 };
1967 if (skb_is_gso(skb
)) {
1968 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1970 /* This is a hint as to how much should be linear. */
1972 __cpu_to_virtio16(vio_le(), skb_headlen(skb
));
1973 vnet_hdr
->gso_size
=
1974 __cpu_to_virtio16(vio_le(), sinfo
->gso_size
);
1976 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
1977 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
1978 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
1979 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
1980 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
1981 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
1982 else if (sinfo
->gso_type
& SKB_GSO_FCOE
)
1987 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
1988 vnet_hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
1990 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
1992 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1993 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
1994 vnet_hdr
->csum_start
= __cpu_to_virtio16(vio_le(),
1995 skb_checksum_start_offset(skb
));
1996 vnet_hdr
->csum_offset
= __cpu_to_virtio16(vio_le(),
1998 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
1999 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
2000 } /* else everything is zero */
2005 static int packet_rcv_vnet(struct msghdr
*msg
, const struct sk_buff
*skb
,
2008 struct virtio_net_hdr vnet_hdr
;
2010 if (*len
< sizeof(vnet_hdr
))
2012 *len
-= sizeof(vnet_hdr
);
2014 if (__packet_rcv_vnet(skb
, &vnet_hdr
))
2017 return memcpy_to_msg(msg
, (void *)&vnet_hdr
, sizeof(vnet_hdr
));
2021 * This function makes lazy skb cloning in hope that most of packets
2022 * are discarded by BPF.
2024 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2025 * and skb->cb are mangled. It works because (and until) packets
2026 * falling here are owned by current CPU. Output packets are cloned
2027 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2028 * sequencially, so that if we return skb to original state on exit,
2029 * we will not harm anyone.
2032 static int packet_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
2033 struct packet_type
*pt
, struct net_device
*orig_dev
)
2036 struct sockaddr_ll
*sll
;
2037 struct packet_sock
*po
;
2038 u8
*skb_head
= skb
->data
;
2039 int skb_len
= skb
->len
;
2040 unsigned int snaplen
, res
;
2042 if (skb
->pkt_type
== PACKET_LOOPBACK
)
2045 sk
= pt
->af_packet_priv
;
2048 if (!net_eq(dev_net(dev
), sock_net(sk
)))
2053 if (dev
->header_ops
) {
2054 /* The device has an explicit notion of ll header,
2055 * exported to higher levels.
2057 * Otherwise, the device hides details of its frame
2058 * structure, so that corresponding packet head is
2059 * never delivered to user.
2061 if (sk
->sk_type
!= SOCK_DGRAM
)
2062 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2063 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
2064 /* Special case: outgoing packets have ll header at head */
2065 skb_pull(skb
, skb_network_offset(skb
));
2071 res
= run_filter(skb
, sk
, snaplen
);
2073 goto drop_n_restore
;
2077 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
2080 if (skb_shared(skb
)) {
2081 struct sk_buff
*nskb
= skb_clone(skb
, GFP_ATOMIC
);
2085 if (skb_head
!= skb
->data
) {
2086 skb
->data
= skb_head
;
2093 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb
)) + MAX_ADDR_LEN
- 8);
2095 sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
2096 sll
->sll_hatype
= dev
->type
;
2097 sll
->sll_pkttype
= skb
->pkt_type
;
2098 if (unlikely(po
->origdev
))
2099 sll
->sll_ifindex
= orig_dev
->ifindex
;
2101 sll
->sll_ifindex
= dev
->ifindex
;
2103 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
2105 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2106 * Use their space for storing the original skb length.
2108 PACKET_SKB_CB(skb
)->sa
.origlen
= skb
->len
;
2110 if (pskb_trim(skb
, snaplen
))
2113 skb_set_owner_r(skb
, sk
);
2117 /* drop conntrack reference */
2120 spin_lock(&sk
->sk_receive_queue
.lock
);
2121 po
->stats
.stats1
.tp_packets
++;
2122 sock_skb_set_dropcount(sk
, skb
);
2123 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
2124 spin_unlock(&sk
->sk_receive_queue
.lock
);
2125 sk
->sk_data_ready(sk
);
2129 spin_lock(&sk
->sk_receive_queue
.lock
);
2130 po
->stats
.stats1
.tp_drops
++;
2131 atomic_inc(&sk
->sk_drops
);
2132 spin_unlock(&sk
->sk_receive_queue
.lock
);
2135 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
2136 skb
->data
= skb_head
;
2144 static int tpacket_rcv(struct sk_buff
*skb
, struct net_device
*dev
,
2145 struct packet_type
*pt
, struct net_device
*orig_dev
)
2148 struct packet_sock
*po
;
2149 struct sockaddr_ll
*sll
;
2150 union tpacket_uhdr h
;
2151 u8
*skb_head
= skb
->data
;
2152 int skb_len
= skb
->len
;
2153 unsigned int snaplen
, res
;
2154 unsigned long status
= TP_STATUS_USER
;
2155 unsigned short macoff
, netoff
, hdrlen
;
2156 struct sk_buff
*copy_skb
= NULL
;
2160 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2161 * We may add members to them until current aligned size without forcing
2162 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2164 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h
.h2
)) != 32);
2165 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h
.h3
)) != 48);
2167 if (skb
->pkt_type
== PACKET_LOOPBACK
)
2170 sk
= pt
->af_packet_priv
;
2173 if (!net_eq(dev_net(dev
), sock_net(sk
)))
2176 if (dev
->header_ops
) {
2177 if (sk
->sk_type
!= SOCK_DGRAM
)
2178 skb_push(skb
, skb
->data
- skb_mac_header(skb
));
2179 else if (skb
->pkt_type
== PACKET_OUTGOING
) {
2180 /* Special case: outgoing packets have ll header at head */
2181 skb_pull(skb
, skb_network_offset(skb
));
2187 res
= run_filter(skb
, sk
, snaplen
);
2189 goto drop_n_restore
;
2191 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2192 status
|= TP_STATUS_CSUMNOTREADY
;
2193 else if (skb
->pkt_type
!= PACKET_OUTGOING
&&
2194 (skb
->ip_summed
== CHECKSUM_COMPLETE
||
2195 skb_csum_unnecessary(skb
)))
2196 status
|= TP_STATUS_CSUM_VALID
;
2201 if (sk
->sk_type
== SOCK_DGRAM
) {
2202 macoff
= netoff
= TPACKET_ALIGN(po
->tp_hdrlen
) + 16 +
2205 unsigned int maclen
= skb_network_offset(skb
);
2206 netoff
= TPACKET_ALIGN(po
->tp_hdrlen
+
2207 (maclen
< 16 ? 16 : maclen
)) +
2209 if (po
->has_vnet_hdr
)
2210 netoff
+= sizeof(struct virtio_net_hdr
);
2211 macoff
= netoff
- maclen
;
2213 if (po
->tp_version
<= TPACKET_V2
) {
2214 if (macoff
+ snaplen
> po
->rx_ring
.frame_size
) {
2215 if (po
->copy_thresh
&&
2216 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
) {
2217 if (skb_shared(skb
)) {
2218 copy_skb
= skb_clone(skb
, GFP_ATOMIC
);
2220 copy_skb
= skb_get(skb
);
2221 skb_head
= skb
->data
;
2224 skb_set_owner_r(copy_skb
, sk
);
2226 snaplen
= po
->rx_ring
.frame_size
- macoff
;
2227 if ((int)snaplen
< 0)
2230 } else if (unlikely(macoff
+ snaplen
>
2231 GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
)) {
2234 nval
= GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
- macoff
;
2235 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2236 snaplen
, nval
, macoff
);
2238 if (unlikely((int)snaplen
< 0)) {
2240 macoff
= GET_PBDQC_FROM_RB(&po
->rx_ring
)->max_frame_len
;
2243 spin_lock(&sk
->sk_receive_queue
.lock
);
2244 h
.raw
= packet_current_rx_frame(po
, skb
,
2245 TP_STATUS_KERNEL
, (macoff
+snaplen
));
2247 goto drop_n_account
;
2248 if (po
->tp_version
<= TPACKET_V2
) {
2249 packet_increment_rx_head(po
, &po
->rx_ring
);
2251 * LOSING will be reported till you read the stats,
2252 * because it's COR - Clear On Read.
2253 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2256 if (po
->stats
.stats1
.tp_drops
)
2257 status
|= TP_STATUS_LOSING
;
2259 po
->stats
.stats1
.tp_packets
++;
2261 status
|= TP_STATUS_COPY
;
2262 __skb_queue_tail(&sk
->sk_receive_queue
, copy_skb
);
2264 spin_unlock(&sk
->sk_receive_queue
.lock
);
2266 if (po
->has_vnet_hdr
) {
2267 if (__packet_rcv_vnet(skb
, h
.raw
+ macoff
-
2268 sizeof(struct virtio_net_hdr
))) {
2269 spin_lock(&sk
->sk_receive_queue
.lock
);
2270 goto drop_n_account
;
2274 skb_copy_bits(skb
, 0, h
.raw
+ macoff
, snaplen
);
2276 if (!(ts_status
= tpacket_get_timestamp(skb
, &ts
, po
->tp_tstamp
)))
2277 getnstimeofday(&ts
);
2279 status
|= ts_status
;
2281 switch (po
->tp_version
) {
2283 h
.h1
->tp_len
= skb
->len
;
2284 h
.h1
->tp_snaplen
= snaplen
;
2285 h
.h1
->tp_mac
= macoff
;
2286 h
.h1
->tp_net
= netoff
;
2287 h
.h1
->tp_sec
= ts
.tv_sec
;
2288 h
.h1
->tp_usec
= ts
.tv_nsec
/ NSEC_PER_USEC
;
2289 hdrlen
= sizeof(*h
.h1
);
2292 h
.h2
->tp_len
= skb
->len
;
2293 h
.h2
->tp_snaplen
= snaplen
;
2294 h
.h2
->tp_mac
= macoff
;
2295 h
.h2
->tp_net
= netoff
;
2296 h
.h2
->tp_sec
= ts
.tv_sec
;
2297 h
.h2
->tp_nsec
= ts
.tv_nsec
;
2298 if (skb_vlan_tag_present(skb
)) {
2299 h
.h2
->tp_vlan_tci
= skb_vlan_tag_get(skb
);
2300 h
.h2
->tp_vlan_tpid
= ntohs(skb
->vlan_proto
);
2301 status
|= TP_STATUS_VLAN_VALID
| TP_STATUS_VLAN_TPID_VALID
;
2303 h
.h2
->tp_vlan_tci
= 0;
2304 h
.h2
->tp_vlan_tpid
= 0;
2306 memset(h
.h2
->tp_padding
, 0, sizeof(h
.h2
->tp_padding
));
2307 hdrlen
= sizeof(*h
.h2
);
2310 /* tp_nxt_offset,vlan are already populated above.
2311 * So DONT clear those fields here
2313 h
.h3
->tp_status
|= status
;
2314 h
.h3
->tp_len
= skb
->len
;
2315 h
.h3
->tp_snaplen
= snaplen
;
2316 h
.h3
->tp_mac
= macoff
;
2317 h
.h3
->tp_net
= netoff
;
2318 h
.h3
->tp_sec
= ts
.tv_sec
;
2319 h
.h3
->tp_nsec
= ts
.tv_nsec
;
2320 memset(h
.h3
->tp_padding
, 0, sizeof(h
.h3
->tp_padding
));
2321 hdrlen
= sizeof(*h
.h3
);
2327 sll
= h
.raw
+ TPACKET_ALIGN(hdrlen
);
2328 sll
->sll_halen
= dev_parse_header(skb
, sll
->sll_addr
);
2329 sll
->sll_family
= AF_PACKET
;
2330 sll
->sll_hatype
= dev
->type
;
2331 sll
->sll_protocol
= skb
->protocol
;
2332 sll
->sll_pkttype
= skb
->pkt_type
;
2333 if (unlikely(po
->origdev
))
2334 sll
->sll_ifindex
= orig_dev
->ifindex
;
2336 sll
->sll_ifindex
= dev
->ifindex
;
2340 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2341 if (po
->tp_version
<= TPACKET_V2
) {
2344 end
= (u8
*) PAGE_ALIGN((unsigned long) h
.raw
+
2347 for (start
= h
.raw
; start
< end
; start
+= PAGE_SIZE
)
2348 flush_dcache_page(pgv_to_page(start
));
2353 if (po
->tp_version
<= TPACKET_V2
) {
2354 __packet_set_status(po
, h
.raw
, status
);
2355 sk
->sk_data_ready(sk
);
2357 prb_clear_blk_fill_status(&po
->rx_ring
);
2361 if (skb_head
!= skb
->data
&& skb_shared(skb
)) {
2362 skb
->data
= skb_head
;
2370 po
->stats
.stats1
.tp_drops
++;
2371 spin_unlock(&sk
->sk_receive_queue
.lock
);
2373 sk
->sk_data_ready(sk
);
2374 kfree_skb(copy_skb
);
2375 goto drop_n_restore
;
2378 static void tpacket_destruct_skb(struct sk_buff
*skb
)
2380 struct packet_sock
*po
= pkt_sk(skb
->sk
);
2382 if (likely(po
->tx_ring
.pg_vec
)) {
2386 ph
= skb_shinfo(skb
)->destructor_arg
;
2387 packet_dec_pending(&po
->tx_ring
);
2389 ts
= __packet_set_timestamp(po
, ph
, skb
);
2390 __packet_set_status(po
, ph
, TP_STATUS_AVAILABLE
| ts
);
2396 static bool ll_header_truncated(const struct net_device
*dev
, int len
)
2398 /* net device doesn't like empty head */
2399 if (unlikely(len
< dev
->hard_header_len
)) {
2400 net_warn_ratelimited("%s: packet size is too short (%d < %d)\n",
2401 current
->comm
, len
, dev
->hard_header_len
);
2408 static void tpacket_set_protocol(const struct net_device
*dev
,
2409 struct sk_buff
*skb
)
2411 if (dev
->type
== ARPHRD_ETHER
) {
2412 skb_reset_mac_header(skb
);
2413 skb
->protocol
= eth_hdr(skb
)->h_proto
;
2417 static int __packet_snd_vnet_parse(struct virtio_net_hdr
*vnet_hdr
, size_t len
)
2419 unsigned short gso_type
= 0;
2421 if ((vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
2422 (__virtio16_to_cpu(vio_le(), vnet_hdr
->csum_start
) +
2423 __virtio16_to_cpu(vio_le(), vnet_hdr
->csum_offset
) + 2 >
2424 __virtio16_to_cpu(vio_le(), vnet_hdr
->hdr_len
)))
2425 vnet_hdr
->hdr_len
= __cpu_to_virtio16(vio_le(),
2426 __virtio16_to_cpu(vio_le(), vnet_hdr
->csum_start
) +
2427 __virtio16_to_cpu(vio_le(), vnet_hdr
->csum_offset
) + 2);
2429 if (__virtio16_to_cpu(vio_le(), vnet_hdr
->hdr_len
) > len
)
2432 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
2433 switch (vnet_hdr
->gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
2434 case VIRTIO_NET_HDR_GSO_TCPV4
:
2435 gso_type
= SKB_GSO_TCPV4
;
2437 case VIRTIO_NET_HDR_GSO_TCPV6
:
2438 gso_type
= SKB_GSO_TCPV6
;
2440 case VIRTIO_NET_HDR_GSO_UDP
:
2441 gso_type
= SKB_GSO_UDP
;
2447 if (vnet_hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
2448 gso_type
|= SKB_GSO_TCP_ECN
;
2450 if (vnet_hdr
->gso_size
== 0)
2454 vnet_hdr
->gso_type
= gso_type
; /* changes type, temporary storage */
2458 static int packet_snd_vnet_parse(struct msghdr
*msg
, size_t *len
,
2459 struct virtio_net_hdr
*vnet_hdr
)
2463 if (*len
< sizeof(*vnet_hdr
))
2465 *len
-= sizeof(*vnet_hdr
);
2467 n
= copy_from_iter(vnet_hdr
, sizeof(*vnet_hdr
), &msg
->msg_iter
);
2468 if (n
!= sizeof(*vnet_hdr
))
2471 return __packet_snd_vnet_parse(vnet_hdr
, *len
);
2474 static int packet_snd_vnet_gso(struct sk_buff
*skb
,
2475 struct virtio_net_hdr
*vnet_hdr
)
2477 if (vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
2478 u16 s
= __virtio16_to_cpu(vio_le(), vnet_hdr
->csum_start
);
2479 u16 o
= __virtio16_to_cpu(vio_le(), vnet_hdr
->csum_offset
);
2481 if (!skb_partial_csum_set(skb
, s
, o
))
2485 skb_shinfo(skb
)->gso_size
=
2486 __virtio16_to_cpu(vio_le(), vnet_hdr
->gso_size
);
2487 skb_shinfo(skb
)->gso_type
= vnet_hdr
->gso_type
;
2489 /* Header must be checked, and gso_segs computed. */
2490 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
2491 skb_shinfo(skb
)->gso_segs
= 0;
2495 static int tpacket_fill_skb(struct packet_sock
*po
, struct sk_buff
*skb
,
2496 void *frame
, struct net_device
*dev
, void *data
, int tp_len
,
2497 __be16 proto
, unsigned char *addr
, int hlen
, int copylen
)
2499 union tpacket_uhdr ph
;
2500 int to_write
, offset
, len
, nr_frags
, len_max
;
2501 struct socket
*sock
= po
->sk
.sk_socket
;
2507 skb
->protocol
= proto
;
2509 skb
->priority
= po
->sk
.sk_priority
;
2510 skb
->mark
= po
->sk
.sk_mark
;
2511 sock_tx_timestamp(&po
->sk
, &skb_shinfo(skb
)->tx_flags
);
2512 skb_shinfo(skb
)->destructor_arg
= ph
.raw
;
2514 skb_reserve(skb
, hlen
);
2515 skb_reset_network_header(skb
);
2519 if (sock
->type
== SOCK_DGRAM
) {
2520 err
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
,
2522 if (unlikely(err
< 0))
2524 } else if (copylen
) {
2525 skb_push(skb
, dev
->hard_header_len
);
2526 skb_put(skb
, copylen
- dev
->hard_header_len
);
2527 err
= skb_store_bits(skb
, 0, data
, copylen
);
2531 tpacket_set_protocol(dev
, skb
);
2534 to_write
-= copylen
;
2537 offset
= offset_in_page(data
);
2538 len_max
= PAGE_SIZE
- offset
;
2539 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2541 skb
->data_len
= to_write
;
2542 skb
->len
+= to_write
;
2543 skb
->truesize
+= to_write
;
2544 atomic_add(to_write
, &po
->sk
.sk_wmem_alloc
);
2546 while (likely(to_write
)) {
2547 nr_frags
= skb_shinfo(skb
)->nr_frags
;
2549 if (unlikely(nr_frags
>= MAX_SKB_FRAGS
)) {
2550 pr_err("Packet exceed the number of skb frags(%lu)\n",
2555 page
= pgv_to_page(data
);
2557 flush_dcache_page(page
);
2559 skb_fill_page_desc(skb
, nr_frags
, page
, offset
, len
);
2562 len_max
= PAGE_SIZE
;
2563 len
= ((to_write
> len_max
) ? len_max
: to_write
);
2566 skb_probe_transport_header(skb
, 0);
2571 static int tpacket_parse_header(struct packet_sock
*po
, void *frame
,
2572 int size_max
, void **data
)
2574 union tpacket_uhdr ph
;
2579 switch (po
->tp_version
) {
2581 tp_len
= ph
.h2
->tp_len
;
2584 tp_len
= ph
.h1
->tp_len
;
2587 if (unlikely(tp_len
> size_max
)) {
2588 pr_err("packet size is too long (%d > %d)\n", tp_len
, size_max
);
2592 if (unlikely(po
->tp_tx_has_off
)) {
2593 int off_min
, off_max
;
2595 off_min
= po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
2596 off_max
= po
->tx_ring
.frame_size
- tp_len
;
2597 if (po
->sk
.sk_type
== SOCK_DGRAM
) {
2598 switch (po
->tp_version
) {
2600 off
= ph
.h2
->tp_net
;
2603 off
= ph
.h1
->tp_net
;
2607 switch (po
->tp_version
) {
2609 off
= ph
.h2
->tp_mac
;
2612 off
= ph
.h1
->tp_mac
;
2616 if (unlikely((off
< off_min
) || (off_max
< off
)))
2619 off
= po
->tp_hdrlen
- sizeof(struct sockaddr_ll
);
2622 *data
= frame
+ off
;
2626 static int tpacket_snd(struct packet_sock
*po
, struct msghdr
*msg
)
2628 struct sk_buff
*skb
;
2629 struct net_device
*dev
;
2630 struct virtio_net_hdr
*vnet_hdr
= NULL
;
2632 int err
, reserve
= 0;
2634 DECLARE_SOCKADDR(struct sockaddr_ll
*, saddr
, msg
->msg_name
);
2635 bool need_wait
= !(msg
->msg_flags
& MSG_DONTWAIT
);
2636 int tp_len
, size_max
;
2637 unsigned char *addr
;
2640 int status
= TP_STATUS_AVAILABLE
;
2641 int hlen
, tlen
, copylen
= 0;
2643 mutex_lock(&po
->pg_vec_lock
);
2645 if (likely(saddr
== NULL
)) {
2646 dev
= packet_cached_dev_get(po
);
2651 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2653 if (msg
->msg_namelen
< (saddr
->sll_halen
2654 + offsetof(struct sockaddr_ll
,
2657 proto
= saddr
->sll_protocol
;
2658 addr
= saddr
->sll_addr
;
2659 dev
= dev_get_by_index(sock_net(&po
->sk
), saddr
->sll_ifindex
);
2663 if (unlikely(dev
== NULL
))
2666 if (unlikely(!(dev
->flags
& IFF_UP
)))
2669 if (po
->sk
.sk_socket
->type
== SOCK_RAW
)
2670 reserve
= dev
->hard_header_len
;
2671 size_max
= po
->tx_ring
.frame_size
2672 - (po
->tp_hdrlen
- sizeof(struct sockaddr_ll
));
2674 if ((size_max
> dev
->mtu
+ reserve
+ VLAN_HLEN
) && !po
->has_vnet_hdr
)
2675 size_max
= dev
->mtu
+ reserve
+ VLAN_HLEN
;
2678 ph
= packet_current_frame(po
, &po
->tx_ring
,
2679 TP_STATUS_SEND_REQUEST
);
2680 if (unlikely(ph
== NULL
)) {
2681 if (need_wait
&& need_resched())
2687 tp_len
= tpacket_parse_header(po
, ph
, size_max
, &data
);
2691 status
= TP_STATUS_SEND_REQUEST
;
2692 hlen
= LL_RESERVED_SPACE(dev
);
2693 tlen
= dev
->needed_tailroom
;
2694 if (po
->has_vnet_hdr
) {
2696 data
+= sizeof(*vnet_hdr
);
2697 tp_len
-= sizeof(*vnet_hdr
);
2699 __packet_snd_vnet_parse(vnet_hdr
, tp_len
)) {
2703 copylen
= __virtio16_to_cpu(vio_le(),
2706 if (dev
->hard_header_len
) {
2707 if (ll_header_truncated(dev
, tp_len
)) {
2711 copylen
= max_t(int, copylen
, dev
->hard_header_len
);
2713 skb
= sock_alloc_send_skb(&po
->sk
,
2714 hlen
+ tlen
+ sizeof(struct sockaddr_ll
) +
2715 (copylen
- dev
->hard_header_len
),
2718 if (unlikely(skb
== NULL
)) {
2719 /* we assume the socket was initially writeable ... */
2720 if (likely(len_sum
> 0))
2724 tp_len
= tpacket_fill_skb(po
, skb
, ph
, dev
, data
, tp_len
, proto
,
2725 addr
, hlen
, copylen
);
2726 if (likely(tp_len
>= 0) &&
2727 tp_len
> dev
->mtu
+ reserve
&&
2728 !po
->has_vnet_hdr
&&
2729 !packet_extra_vlan_len_allowed(dev
, skb
))
2732 if (unlikely(tp_len
< 0)) {
2735 __packet_set_status(po
, ph
,
2736 TP_STATUS_AVAILABLE
);
2737 packet_increment_head(&po
->tx_ring
);
2741 status
= TP_STATUS_WRONG_FORMAT
;
2747 if (po
->has_vnet_hdr
&& packet_snd_vnet_gso(skb
, vnet_hdr
)) {
2752 packet_pick_tx_queue(dev
, skb
);
2754 skb
->destructor
= tpacket_destruct_skb
;
2755 __packet_set_status(po
, ph
, TP_STATUS_SENDING
);
2756 packet_inc_pending(&po
->tx_ring
);
2758 status
= TP_STATUS_SEND_REQUEST
;
2759 err
= po
->xmit(skb
);
2760 if (unlikely(err
> 0)) {
2761 err
= net_xmit_errno(err
);
2762 if (err
&& __packet_get_status(po
, ph
) ==
2763 TP_STATUS_AVAILABLE
) {
2764 /* skb was destructed already */
2769 * skb was dropped but not destructed yet;
2770 * let's treat it like congestion or err < 0
2774 packet_increment_head(&po
->tx_ring
);
2776 } while (likely((ph
!= NULL
) ||
2777 /* Note: packet_read_pending() might be slow if we have
2778 * to call it as it's per_cpu variable, but in fast-path
2779 * we already short-circuit the loop with the first
2780 * condition, and luckily don't have to go that path
2783 (need_wait
&& packet_read_pending(&po
->tx_ring
))));
2789 __packet_set_status(po
, ph
, status
);
2794 mutex_unlock(&po
->pg_vec_lock
);
2798 static struct sk_buff
*packet_alloc_skb(struct sock
*sk
, size_t prepad
,
2799 size_t reserve
, size_t len
,
2800 size_t linear
, int noblock
,
2803 struct sk_buff
*skb
;
2805 /* Under a page? Don't bother with paged skb. */
2806 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
2809 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
2814 skb_reserve(skb
, reserve
);
2815 skb_put(skb
, linear
);
2816 skb
->data_len
= len
- linear
;
2817 skb
->len
+= len
- linear
;
2822 static int packet_snd(struct socket
*sock
, struct msghdr
*msg
, size_t len
)
2824 struct sock
*sk
= sock
->sk
;
2825 DECLARE_SOCKADDR(struct sockaddr_ll
*, saddr
, msg
->msg_name
);
2826 struct sk_buff
*skb
;
2827 struct net_device
*dev
;
2829 unsigned char *addr
;
2830 int err
, reserve
= 0;
2831 struct sockcm_cookie sockc
;
2832 struct virtio_net_hdr vnet_hdr
= { 0 };
2834 struct packet_sock
*po
= pkt_sk(sk
);
2839 * Get and verify the address.
2842 if (likely(saddr
== NULL
)) {
2843 dev
= packet_cached_dev_get(po
);
2848 if (msg
->msg_namelen
< sizeof(struct sockaddr_ll
))
2850 if (msg
->msg_namelen
< (saddr
->sll_halen
+ offsetof(struct sockaddr_ll
, sll_addr
)))
2852 proto
= saddr
->sll_protocol
;
2853 addr
= saddr
->sll_addr
;
2854 dev
= dev_get_by_index(sock_net(sk
), saddr
->sll_ifindex
);
2858 if (unlikely(dev
== NULL
))
2861 if (unlikely(!(dev
->flags
& IFF_UP
)))
2864 sockc
.mark
= sk
->sk_mark
;
2865 if (msg
->msg_controllen
) {
2866 err
= sock_cmsg_send(sk
, msg
, &sockc
);
2871 if (sock
->type
== SOCK_RAW
)
2872 reserve
= dev
->hard_header_len
;
2873 if (po
->has_vnet_hdr
) {
2874 err
= packet_snd_vnet_parse(msg
, &len
, &vnet_hdr
);
2879 if (unlikely(sock_flag(sk
, SOCK_NOFCS
))) {
2880 if (!netif_supports_nofcs(dev
)) {
2881 err
= -EPROTONOSUPPORT
;
2884 extra_len
= 4; /* We're doing our own CRC */
2888 if (!vnet_hdr
.gso_type
&&
2889 (len
> dev
->mtu
+ reserve
+ VLAN_HLEN
+ extra_len
))
2893 hlen
= LL_RESERVED_SPACE(dev
);
2894 tlen
= dev
->needed_tailroom
;
2895 skb
= packet_alloc_skb(sk
, hlen
+ tlen
, hlen
, len
,
2896 __virtio16_to_cpu(vio_le(), vnet_hdr
.hdr_len
),
2897 msg
->msg_flags
& MSG_DONTWAIT
, &err
);
2901 skb_set_network_header(skb
, reserve
);
2904 if (sock
->type
== SOCK_DGRAM
) {
2905 offset
= dev_hard_header(skb
, dev
, ntohs(proto
), addr
, NULL
, len
);
2906 if (unlikely(offset
< 0))
2909 if (ll_header_truncated(dev
, len
))
2913 /* Returns -EFAULT on error */
2914 err
= skb_copy_datagram_from_iter(skb
, offset
, &msg
->msg_iter
, len
);
2918 sock_tx_timestamp(sk
, &skb_shinfo(skb
)->tx_flags
);
2920 if (!vnet_hdr
.gso_type
&& (len
> dev
->mtu
+ reserve
+ extra_len
) &&
2921 !packet_extra_vlan_len_allowed(dev
, skb
)) {
2926 skb
->protocol
= proto
;
2928 skb
->priority
= sk
->sk_priority
;
2929 skb
->mark
= sockc
.mark
;
2931 packet_pick_tx_queue(dev
, skb
);
2933 if (po
->has_vnet_hdr
) {
2934 err
= packet_snd_vnet_gso(skb
, &vnet_hdr
);
2937 len
+= sizeof(vnet_hdr
);
2940 skb_probe_transport_header(skb
, reserve
);
2942 if (unlikely(extra_len
== 4))
2945 err
= po
->xmit(skb
);
2946 if (err
> 0 && (err
= net_xmit_errno(err
)) != 0)
2962 static int packet_sendmsg(struct socket
*sock
, struct msghdr
*msg
, size_t len
)
2964 struct sock
*sk
= sock
->sk
;
2965 struct packet_sock
*po
= pkt_sk(sk
);
2967 if (po
->tx_ring
.pg_vec
)
2968 return tpacket_snd(po
, msg
);
2970 return packet_snd(sock
, msg
, len
);
2974 * Close a PACKET socket. This is fairly simple. We immediately go
2975 * to 'closed' state and remove our protocol entry in the device list.
2978 static int packet_release(struct socket
*sock
)
2980 struct sock
*sk
= sock
->sk
;
2981 struct packet_sock
*po
;
2983 union tpacket_req_u req_u
;
2991 mutex_lock(&net
->packet
.sklist_lock
);
2992 sk_del_node_init_rcu(sk
);
2993 mutex_unlock(&net
->packet
.sklist_lock
);
2996 sock_prot_inuse_add(net
, sk
->sk_prot
, -1);
2999 spin_lock(&po
->bind_lock
);
3000 unregister_prot_hook(sk
, false);
3001 packet_cached_dev_reset(po
);
3003 if (po
->prot_hook
.dev
) {
3004 dev_put(po
->prot_hook
.dev
);
3005 po
->prot_hook
.dev
= NULL
;
3007 spin_unlock(&po
->bind_lock
);
3009 packet_flush_mclist(sk
);
3011 if (po
->rx_ring
.pg_vec
) {
3012 memset(&req_u
, 0, sizeof(req_u
));
3013 packet_set_ring(sk
, &req_u
, 1, 0);
3016 if (po
->tx_ring
.pg_vec
) {
3017 memset(&req_u
, 0, sizeof(req_u
));
3018 packet_set_ring(sk
, &req_u
, 1, 1);
3025 * Now the socket is dead. No more input will appear.
3032 skb_queue_purge(&sk
->sk_receive_queue
);
3033 packet_free_pending(po
);
3034 sk_refcnt_debug_release(sk
);
3041 * Attach a packet hook.
3044 static int packet_do_bind(struct sock
*sk
, const char *name
, int ifindex
,
3047 struct packet_sock
*po
= pkt_sk(sk
);
3048 struct net_device
*dev_curr
;
3051 struct net_device
*dev
= NULL
;
3053 bool unlisted
= false;
3059 spin_lock(&po
->bind_lock
);
3063 dev
= dev_get_by_name_rcu(sock_net(sk
), name
);
3068 } else if (ifindex
) {
3069 dev
= dev_get_by_index_rcu(sock_net(sk
), ifindex
);
3079 proto_curr
= po
->prot_hook
.type
;
3080 dev_curr
= po
->prot_hook
.dev
;
3082 need_rehook
= proto_curr
!= proto
|| dev_curr
!= dev
;
3087 __unregister_prot_hook(sk
, true);
3089 dev_curr
= po
->prot_hook
.dev
;
3091 unlisted
= !dev_get_by_index_rcu(sock_net(sk
),
3096 po
->prot_hook
.type
= proto
;
3098 if (unlikely(unlisted
)) {
3100 po
->prot_hook
.dev
= NULL
;
3102 packet_cached_dev_reset(po
);
3104 po
->prot_hook
.dev
= dev
;
3105 po
->ifindex
= dev
? dev
->ifindex
: 0;
3106 packet_cached_dev_assign(po
, dev
);
3112 if (proto
== 0 || !need_rehook
)
3115 if (!unlisted
&& (!dev
|| (dev
->flags
& IFF_UP
))) {
3116 register_prot_hook(sk
);
3118 sk
->sk_err
= ENETDOWN
;
3119 if (!sock_flag(sk
, SOCK_DEAD
))
3120 sk
->sk_error_report(sk
);
3125 spin_unlock(&po
->bind_lock
);
3131 * Bind a packet socket to a device
3134 static int packet_bind_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
3137 struct sock
*sk
= sock
->sk
;
3144 if (addr_len
!= sizeof(struct sockaddr
))
3146 strlcpy(name
, uaddr
->sa_data
, sizeof(name
));
3148 return packet_do_bind(sk
, name
, 0, pkt_sk(sk
)->num
);
3151 static int packet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
3153 struct sockaddr_ll
*sll
= (struct sockaddr_ll
*)uaddr
;
3154 struct sock
*sk
= sock
->sk
;
3160 if (addr_len
< sizeof(struct sockaddr_ll
))
3162 if (sll
->sll_family
!= AF_PACKET
)
3165 return packet_do_bind(sk
, NULL
, sll
->sll_ifindex
,
3166 sll
->sll_protocol
? : pkt_sk(sk
)->num
);
3169 static struct proto packet_proto
= {
3171 .owner
= THIS_MODULE
,
3172 .obj_size
= sizeof(struct packet_sock
),
3176 * Create a packet of type SOCK_PACKET.
3179 static int packet_create(struct net
*net
, struct socket
*sock
, int protocol
,
3183 struct packet_sock
*po
;
3184 __be16 proto
= (__force __be16
)protocol
; /* weird, but documented */
3187 if (!ns_capable(net
->user_ns
, CAP_NET_RAW
))
3189 if (sock
->type
!= SOCK_DGRAM
&& sock
->type
!= SOCK_RAW
&&
3190 sock
->type
!= SOCK_PACKET
)
3191 return -ESOCKTNOSUPPORT
;
3193 sock
->state
= SS_UNCONNECTED
;
3196 sk
= sk_alloc(net
, PF_PACKET
, GFP_KERNEL
, &packet_proto
, kern
);
3200 sock
->ops
= &packet_ops
;
3201 if (sock
->type
== SOCK_PACKET
)
3202 sock
->ops
= &packet_ops_spkt
;
3204 sock_init_data(sock
, sk
);
3207 sk
->sk_family
= PF_PACKET
;
3209 po
->xmit
= dev_queue_xmit
;
3211 err
= packet_alloc_pending(po
);
3215 packet_cached_dev_reset(po
);
3217 sk
->sk_destruct
= packet_sock_destruct
;
3218 sk_refcnt_debug_inc(sk
);
3221 * Attach a protocol block
3224 spin_lock_init(&po
->bind_lock
);
3225 mutex_init(&po
->pg_vec_lock
);
3226 po
->rollover
= NULL
;
3227 po
->prot_hook
.func
= packet_rcv
;
3229 if (sock
->type
== SOCK_PACKET
)
3230 po
->prot_hook
.func
= packet_rcv_spkt
;
3232 po
->prot_hook
.af_packet_priv
= sk
;
3235 po
->prot_hook
.type
= proto
;
3236 register_prot_hook(sk
);
3239 mutex_lock(&net
->packet
.sklist_lock
);
3240 sk_add_node_rcu(sk
, &net
->packet
.sklist
);
3241 mutex_unlock(&net
->packet
.sklist_lock
);
3244 sock_prot_inuse_add(net
, &packet_proto
, 1);
3255 * Pull a packet from our receive queue and hand it to the user.
3256 * If necessary we block.
3259 static int packet_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t len
,
3262 struct sock
*sk
= sock
->sk
;
3263 struct sk_buff
*skb
;
3265 int vnet_hdr_len
= 0;
3266 unsigned int origlen
= 0;
3269 if (flags
& ~(MSG_PEEK
|MSG_DONTWAIT
|MSG_TRUNC
|MSG_CMSG_COMPAT
|MSG_ERRQUEUE
))
3273 /* What error should we return now? EUNATTACH? */
3274 if (pkt_sk(sk
)->ifindex
< 0)
3278 if (flags
& MSG_ERRQUEUE
) {
3279 err
= sock_recv_errqueue(sk
, msg
, len
,
3280 SOL_PACKET
, PACKET_TX_TIMESTAMP
);
3285 * Call the generic datagram receiver. This handles all sorts
3286 * of horrible races and re-entrancy so we can forget about it
3287 * in the protocol layers.
3289 * Now it will return ENETDOWN, if device have just gone down,
3290 * but then it will block.
3293 skb
= skb_recv_datagram(sk
, flags
, flags
& MSG_DONTWAIT
, &err
);
3296 * An error occurred so return it. Because skb_recv_datagram()
3297 * handles the blocking we don't see and worry about blocking
3304 if (pkt_sk(sk
)->pressure
)
3305 packet_rcv_has_room(pkt_sk(sk
), NULL
);
3307 if (pkt_sk(sk
)->has_vnet_hdr
) {
3308 err
= packet_rcv_vnet(msg
, skb
, &len
);
3311 vnet_hdr_len
= sizeof(struct virtio_net_hdr
);
3314 /* You lose any data beyond the buffer you gave. If it worries
3315 * a user program they can ask the device for its MTU
3321 msg
->msg_flags
|= MSG_TRUNC
;
3324 err
= skb_copy_datagram_msg(skb
, 0, msg
, copied
);
3328 if (sock
->type
!= SOCK_PACKET
) {
3329 struct sockaddr_ll
*sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
3331 /* Original length was stored in sockaddr_ll fields */
3332 origlen
= PACKET_SKB_CB(skb
)->sa
.origlen
;
3333 sll
->sll_family
= AF_PACKET
;
3334 sll
->sll_protocol
= skb
->protocol
;
3337 sock_recv_ts_and_drops(msg
, sk
, skb
);
3339 if (msg
->msg_name
) {
3340 /* If the address length field is there to be filled
3341 * in, we fill it in now.
3343 if (sock
->type
== SOCK_PACKET
) {
3344 __sockaddr_check_size(sizeof(struct sockaddr_pkt
));
3345 msg
->msg_namelen
= sizeof(struct sockaddr_pkt
);
3347 struct sockaddr_ll
*sll
= &PACKET_SKB_CB(skb
)->sa
.ll
;
3349 msg
->msg_namelen
= sll
->sll_halen
+
3350 offsetof(struct sockaddr_ll
, sll_addr
);
3352 memcpy(msg
->msg_name
, &PACKET_SKB_CB(skb
)->sa
,
3356 if (pkt_sk(sk
)->auxdata
) {
3357 struct tpacket_auxdata aux
;
3359 aux
.tp_status
= TP_STATUS_USER
;
3360 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
3361 aux
.tp_status
|= TP_STATUS_CSUMNOTREADY
;
3362 else if (skb
->pkt_type
!= PACKET_OUTGOING
&&
3363 (skb
->ip_summed
== CHECKSUM_COMPLETE
||
3364 skb_csum_unnecessary(skb
)))
3365 aux
.tp_status
|= TP_STATUS_CSUM_VALID
;
3367 aux
.tp_len
= origlen
;
3368 aux
.tp_snaplen
= skb
->len
;
3370 aux
.tp_net
= skb_network_offset(skb
);
3371 if (skb_vlan_tag_present(skb
)) {
3372 aux
.tp_vlan_tci
= skb_vlan_tag_get(skb
);
3373 aux
.tp_vlan_tpid
= ntohs(skb
->vlan_proto
);
3374 aux
.tp_status
|= TP_STATUS_VLAN_VALID
| TP_STATUS_VLAN_TPID_VALID
;
3376 aux
.tp_vlan_tci
= 0;
3377 aux
.tp_vlan_tpid
= 0;
3379 put_cmsg(msg
, SOL_PACKET
, PACKET_AUXDATA
, sizeof(aux
), &aux
);
3383 * Free or return the buffer as appropriate. Again this
3384 * hides all the races and re-entrancy issues from us.
3386 err
= vnet_hdr_len
+ ((flags
&MSG_TRUNC
) ? skb
->len
: copied
);
3389 skb_free_datagram(sk
, skb
);
3394 static int packet_getname_spkt(struct socket
*sock
, struct sockaddr
*uaddr
,
3395 int *uaddr_len
, int peer
)
3397 struct net_device
*dev
;
3398 struct sock
*sk
= sock
->sk
;
3403 uaddr
->sa_family
= AF_PACKET
;
3404 memset(uaddr
->sa_data
, 0, sizeof(uaddr
->sa_data
));
3406 dev
= dev_get_by_index_rcu(sock_net(sk
), pkt_sk(sk
)->ifindex
);
3408 strlcpy(uaddr
->sa_data
, dev
->name
, sizeof(uaddr
->sa_data
));
3410 *uaddr_len
= sizeof(*uaddr
);
3415 static int packet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
3416 int *uaddr_len
, int peer
)
3418 struct net_device
*dev
;
3419 struct sock
*sk
= sock
->sk
;
3420 struct packet_sock
*po
= pkt_sk(sk
);
3421 DECLARE_SOCKADDR(struct sockaddr_ll
*, sll
, uaddr
);
3426 sll
->sll_family
= AF_PACKET
;
3427 sll
->sll_ifindex
= po
->ifindex
;
3428 sll
->sll_protocol
= po
->num
;
3429 sll
->sll_pkttype
= 0;
3431 dev
= dev_get_by_index_rcu(sock_net(sk
), po
->ifindex
);
3433 sll
->sll_hatype
= dev
->type
;
3434 sll
->sll_halen
= dev
->addr_len
;
3435 memcpy(sll
->sll_addr
, dev
->dev_addr
, dev
->addr_len
);
3437 sll
->sll_hatype
= 0; /* Bad: we have no ARPHRD_UNSPEC */
3441 *uaddr_len
= offsetof(struct sockaddr_ll
, sll_addr
) + sll
->sll_halen
;
3446 static int packet_dev_mc(struct net_device
*dev
, struct packet_mclist
*i
,
3450 case PACKET_MR_MULTICAST
:
3451 if (i
->alen
!= dev
->addr_len
)
3454 return dev_mc_add(dev
, i
->addr
);
3456 return dev_mc_del(dev
, i
->addr
);
3458 case PACKET_MR_PROMISC
:
3459 return dev_set_promiscuity(dev
, what
);
3460 case PACKET_MR_ALLMULTI
:
3461 return dev_set_allmulti(dev
, what
);
3462 case PACKET_MR_UNICAST
:
3463 if (i
->alen
!= dev
->addr_len
)
3466 return dev_uc_add(dev
, i
->addr
);
3468 return dev_uc_del(dev
, i
->addr
);
3476 static void packet_dev_mclist_delete(struct net_device
*dev
,
3477 struct packet_mclist
**mlp
)
3479 struct packet_mclist
*ml
;
3481 while ((ml
= *mlp
) != NULL
) {
3482 if (ml
->ifindex
== dev
->ifindex
) {
3483 packet_dev_mc(dev
, ml
, -1);
3491 static int packet_mc_add(struct sock
*sk
, struct packet_mreq_max
*mreq
)
3493 struct packet_sock
*po
= pkt_sk(sk
);
3494 struct packet_mclist
*ml
, *i
;
3495 struct net_device
*dev
;
3501 dev
= __dev_get_by_index(sock_net(sk
), mreq
->mr_ifindex
);
3506 if (mreq
->mr_alen
> dev
->addr_len
)
3510 i
= kmalloc(sizeof(*i
), GFP_KERNEL
);
3515 for (ml
= po
->mclist
; ml
; ml
= ml
->next
) {
3516 if (ml
->ifindex
== mreq
->mr_ifindex
&&
3517 ml
->type
== mreq
->mr_type
&&
3518 ml
->alen
== mreq
->mr_alen
&&
3519 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
3521 /* Free the new element ... */
3527 i
->type
= mreq
->mr_type
;
3528 i
->ifindex
= mreq
->mr_ifindex
;
3529 i
->alen
= mreq
->mr_alen
;
3530 memcpy(i
->addr
, mreq
->mr_address
, i
->alen
);
3532 i
->next
= po
->mclist
;
3534 err
= packet_dev_mc(dev
, i
, 1);
3536 po
->mclist
= i
->next
;
3545 static int packet_mc_drop(struct sock
*sk
, struct packet_mreq_max
*mreq
)
3547 struct packet_mclist
*ml
, **mlp
;
3551 for (mlp
= &pkt_sk(sk
)->mclist
; (ml
= *mlp
) != NULL
; mlp
= &ml
->next
) {
3552 if (ml
->ifindex
== mreq
->mr_ifindex
&&
3553 ml
->type
== mreq
->mr_type
&&
3554 ml
->alen
== mreq
->mr_alen
&&
3555 memcmp(ml
->addr
, mreq
->mr_address
, ml
->alen
) == 0) {
3556 if (--ml
->count
== 0) {
3557 struct net_device
*dev
;
3559 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
3561 packet_dev_mc(dev
, ml
, -1);
3571 static void packet_flush_mclist(struct sock
*sk
)
3573 struct packet_sock
*po
= pkt_sk(sk
);
3574 struct packet_mclist
*ml
;
3580 while ((ml
= po
->mclist
) != NULL
) {
3581 struct net_device
*dev
;
3583 po
->mclist
= ml
->next
;
3584 dev
= __dev_get_by_index(sock_net(sk
), ml
->ifindex
);
3586 packet_dev_mc(dev
, ml
, -1);
3593 packet_setsockopt(struct socket
*sock
, int level
, int optname
, char __user
*optval
, unsigned int optlen
)
3595 struct sock
*sk
= sock
->sk
;
3596 struct packet_sock
*po
= pkt_sk(sk
);
3599 if (level
!= SOL_PACKET
)
3600 return -ENOPROTOOPT
;
3603 case PACKET_ADD_MEMBERSHIP
:
3604 case PACKET_DROP_MEMBERSHIP
:
3606 struct packet_mreq_max mreq
;
3608 memset(&mreq
, 0, sizeof(mreq
));
3609 if (len
< sizeof(struct packet_mreq
))
3611 if (len
> sizeof(mreq
))
3613 if (copy_from_user(&mreq
, optval
, len
))
3615 if (len
< (mreq
.mr_alen
+ offsetof(struct packet_mreq
, mr_address
)))
3617 if (optname
== PACKET_ADD_MEMBERSHIP
)
3618 ret
= packet_mc_add(sk
, &mreq
);
3620 ret
= packet_mc_drop(sk
, &mreq
);
3624 case PACKET_RX_RING
:
3625 case PACKET_TX_RING
:
3627 union tpacket_req_u req_u
;
3630 switch (po
->tp_version
) {
3633 len
= sizeof(req_u
.req
);
3637 len
= sizeof(req_u
.req3
);
3642 if (copy_from_user(&req_u
.req
, optval
, len
))
3644 return packet_set_ring(sk
, &req_u
, 0,
3645 optname
== PACKET_TX_RING
);
3647 case PACKET_COPY_THRESH
:
3651 if (optlen
!= sizeof(val
))
3653 if (copy_from_user(&val
, optval
, sizeof(val
)))
3656 pkt_sk(sk
)->copy_thresh
= val
;
3659 case PACKET_VERSION
:
3663 if (optlen
!= sizeof(val
))
3665 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3667 if (copy_from_user(&val
, optval
, sizeof(val
)))
3673 po
->tp_version
= val
;
3679 case PACKET_RESERVE
:
3683 if (optlen
!= sizeof(val
))
3685 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3687 if (copy_from_user(&val
, optval
, sizeof(val
)))
3689 po
->tp_reserve
= val
;
3696 if (optlen
!= sizeof(val
))
3698 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3700 if (copy_from_user(&val
, optval
, sizeof(val
)))
3702 po
->tp_loss
= !!val
;
3705 case PACKET_AUXDATA
:
3709 if (optlen
< sizeof(val
))
3711 if (copy_from_user(&val
, optval
, sizeof(val
)))
3714 po
->auxdata
= !!val
;
3717 case PACKET_ORIGDEV
:
3721 if (optlen
< sizeof(val
))
3723 if (copy_from_user(&val
, optval
, sizeof(val
)))
3726 po
->origdev
= !!val
;
3729 case PACKET_VNET_HDR
:
3733 if (sock
->type
!= SOCK_RAW
)
3735 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3737 if (optlen
< sizeof(val
))
3739 if (copy_from_user(&val
, optval
, sizeof(val
)))
3742 po
->has_vnet_hdr
= !!val
;
3745 case PACKET_TIMESTAMP
:
3749 if (optlen
!= sizeof(val
))
3751 if (copy_from_user(&val
, optval
, sizeof(val
)))
3754 po
->tp_tstamp
= val
;
3761 if (optlen
!= sizeof(val
))
3763 if (copy_from_user(&val
, optval
, sizeof(val
)))
3766 return fanout_add(sk
, val
& 0xffff, val
>> 16);
3768 case PACKET_FANOUT_DATA
:
3773 return fanout_set_data(po
, optval
, optlen
);
3775 case PACKET_TX_HAS_OFF
:
3779 if (optlen
!= sizeof(val
))
3781 if (po
->rx_ring
.pg_vec
|| po
->tx_ring
.pg_vec
)
3783 if (copy_from_user(&val
, optval
, sizeof(val
)))
3785 po
->tp_tx_has_off
= !!val
;
3788 case PACKET_QDISC_BYPASS
:
3792 if (optlen
!= sizeof(val
))
3794 if (copy_from_user(&val
, optval
, sizeof(val
)))
3797 po
->xmit
= val
? packet_direct_xmit
: dev_queue_xmit
;
3801 return -ENOPROTOOPT
;
3805 static int packet_getsockopt(struct socket
*sock
, int level
, int optname
,
3806 char __user
*optval
, int __user
*optlen
)
3809 int val
, lv
= sizeof(val
);
3810 struct sock
*sk
= sock
->sk
;
3811 struct packet_sock
*po
= pkt_sk(sk
);
3813 union tpacket_stats_u st
;
3814 struct tpacket_rollover_stats rstats
;
3816 if (level
!= SOL_PACKET
)
3817 return -ENOPROTOOPT
;
3819 if (get_user(len
, optlen
))
3826 case PACKET_STATISTICS
:
3827 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3828 memcpy(&st
, &po
->stats
, sizeof(st
));
3829 memset(&po
->stats
, 0, sizeof(po
->stats
));
3830 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3832 if (po
->tp_version
== TPACKET_V3
) {
3833 lv
= sizeof(struct tpacket_stats_v3
);
3834 st
.stats3
.tp_packets
+= st
.stats3
.tp_drops
;
3837 lv
= sizeof(struct tpacket_stats
);
3838 st
.stats1
.tp_packets
+= st
.stats1
.tp_drops
;
3843 case PACKET_AUXDATA
:
3846 case PACKET_ORIGDEV
:
3849 case PACKET_VNET_HDR
:
3850 val
= po
->has_vnet_hdr
;
3852 case PACKET_VERSION
:
3853 val
= po
->tp_version
;
3856 if (len
> sizeof(int))
3858 if (copy_from_user(&val
, optval
, len
))
3862 val
= sizeof(struct tpacket_hdr
);
3865 val
= sizeof(struct tpacket2_hdr
);
3868 val
= sizeof(struct tpacket3_hdr
);
3874 case PACKET_RESERVE
:
3875 val
= po
->tp_reserve
;
3880 case PACKET_TIMESTAMP
:
3881 val
= po
->tp_tstamp
;
3885 ((u32
)po
->fanout
->id
|
3886 ((u32
)po
->fanout
->type
<< 16) |
3887 ((u32
)po
->fanout
->flags
<< 24)) :
3890 case PACKET_ROLLOVER_STATS
:
3893 rstats
.tp_all
= atomic_long_read(&po
->rollover
->num
);
3894 rstats
.tp_huge
= atomic_long_read(&po
->rollover
->num_huge
);
3895 rstats
.tp_failed
= atomic_long_read(&po
->rollover
->num_failed
);
3897 lv
= sizeof(rstats
);
3899 case PACKET_TX_HAS_OFF
:
3900 val
= po
->tp_tx_has_off
;
3902 case PACKET_QDISC_BYPASS
:
3903 val
= packet_use_direct_xmit(po
);
3906 return -ENOPROTOOPT
;
3911 if (put_user(len
, optlen
))
3913 if (copy_to_user(optval
, data
, len
))
3919 static int packet_notifier(struct notifier_block
*this,
3920 unsigned long msg
, void *ptr
)
3923 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
3924 struct net
*net
= dev_net(dev
);
3927 sk_for_each_rcu(sk
, &net
->packet
.sklist
) {
3928 struct packet_sock
*po
= pkt_sk(sk
);
3931 case NETDEV_UNREGISTER
:
3933 packet_dev_mclist_delete(dev
, &po
->mclist
);
3937 if (dev
->ifindex
== po
->ifindex
) {
3938 spin_lock(&po
->bind_lock
);
3940 __unregister_prot_hook(sk
, false);
3941 sk
->sk_err
= ENETDOWN
;
3942 if (!sock_flag(sk
, SOCK_DEAD
))
3943 sk
->sk_error_report(sk
);
3945 if (msg
== NETDEV_UNREGISTER
) {
3946 packet_cached_dev_reset(po
);
3948 if (po
->prot_hook
.dev
)
3949 dev_put(po
->prot_hook
.dev
);
3950 po
->prot_hook
.dev
= NULL
;
3952 spin_unlock(&po
->bind_lock
);
3956 if (dev
->ifindex
== po
->ifindex
) {
3957 spin_lock(&po
->bind_lock
);
3959 register_prot_hook(sk
);
3960 spin_unlock(&po
->bind_lock
);
3970 static int packet_ioctl(struct socket
*sock
, unsigned int cmd
,
3973 struct sock
*sk
= sock
->sk
;
3978 int amount
= sk_wmem_alloc_get(sk
);
3980 return put_user(amount
, (int __user
*)arg
);
3984 struct sk_buff
*skb
;
3987 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
3988 skb
= skb_peek(&sk
->sk_receive_queue
);
3991 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
3992 return put_user(amount
, (int __user
*)arg
);
3995 return sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
3997 return sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
4007 case SIOCGIFBRDADDR
:
4008 case SIOCSIFBRDADDR
:
4009 case SIOCGIFNETMASK
:
4010 case SIOCSIFNETMASK
:
4011 case SIOCGIFDSTADDR
:
4012 case SIOCSIFDSTADDR
:
4014 return inet_dgram_ops
.ioctl(sock
, cmd
, arg
);
4018 return -ENOIOCTLCMD
;
4023 static unsigned int packet_poll(struct file
*file
, struct socket
*sock
,
4026 struct sock
*sk
= sock
->sk
;
4027 struct packet_sock
*po
= pkt_sk(sk
);
4028 unsigned int mask
= datagram_poll(file
, sock
, wait
);
4030 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
4031 if (po
->rx_ring
.pg_vec
) {
4032 if (!packet_previous_rx_frame(po
, &po
->rx_ring
,
4034 mask
|= POLLIN
| POLLRDNORM
;
4036 if (po
->pressure
&& __packet_rcv_has_room(po
, NULL
) == ROOM_NORMAL
)
4038 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
4039 spin_lock_bh(&sk
->sk_write_queue
.lock
);
4040 if (po
->tx_ring
.pg_vec
) {
4041 if (packet_current_frame(po
, &po
->tx_ring
, TP_STATUS_AVAILABLE
))
4042 mask
|= POLLOUT
| POLLWRNORM
;
4044 spin_unlock_bh(&sk
->sk_write_queue
.lock
);
4049 /* Dirty? Well, I still did not learn better way to account
4053 static void packet_mm_open(struct vm_area_struct
*vma
)
4055 struct file
*file
= vma
->vm_file
;
4056 struct socket
*sock
= file
->private_data
;
4057 struct sock
*sk
= sock
->sk
;
4060 atomic_inc(&pkt_sk(sk
)->mapped
);
4063 static void packet_mm_close(struct vm_area_struct
*vma
)
4065 struct file
*file
= vma
->vm_file
;
4066 struct socket
*sock
= file
->private_data
;
4067 struct sock
*sk
= sock
->sk
;
4070 atomic_dec(&pkt_sk(sk
)->mapped
);
4073 static const struct vm_operations_struct packet_mmap_ops
= {
4074 .open
= packet_mm_open
,
4075 .close
= packet_mm_close
,
4078 static void free_pg_vec(struct pgv
*pg_vec
, unsigned int order
,
4083 for (i
= 0; i
< len
; i
++) {
4084 if (likely(pg_vec
[i
].buffer
)) {
4085 if (is_vmalloc_addr(pg_vec
[i
].buffer
))
4086 vfree(pg_vec
[i
].buffer
);
4088 free_pages((unsigned long)pg_vec
[i
].buffer
,
4090 pg_vec
[i
].buffer
= NULL
;
4096 static char *alloc_one_pg_vec_page(unsigned long order
)
4099 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_COMP
|
4100 __GFP_ZERO
| __GFP_NOWARN
| __GFP_NORETRY
;
4102 buffer
= (char *) __get_free_pages(gfp_flags
, order
);
4106 /* __get_free_pages failed, fall back to vmalloc */
4107 buffer
= vzalloc((1 << order
) * PAGE_SIZE
);
4111 /* vmalloc failed, lets dig into swap here */
4112 gfp_flags
&= ~__GFP_NORETRY
;
4113 buffer
= (char *) __get_free_pages(gfp_flags
, order
);
4117 /* complete and utter failure */
4121 static struct pgv
*alloc_pg_vec(struct tpacket_req
*req
, int order
)
4123 unsigned int block_nr
= req
->tp_block_nr
;
4127 pg_vec
= kcalloc(block_nr
, sizeof(struct pgv
), GFP_KERNEL
);
4128 if (unlikely(!pg_vec
))
4131 for (i
= 0; i
< block_nr
; i
++) {
4132 pg_vec
[i
].buffer
= alloc_one_pg_vec_page(order
);
4133 if (unlikely(!pg_vec
[i
].buffer
))
4134 goto out_free_pgvec
;
4141 free_pg_vec(pg_vec
, order
, block_nr
);
4146 static int packet_set_ring(struct sock
*sk
, union tpacket_req_u
*req_u
,
4147 int closing
, int tx_ring
)
4149 struct pgv
*pg_vec
= NULL
;
4150 struct packet_sock
*po
= pkt_sk(sk
);
4151 int was_running
, order
= 0;
4152 struct packet_ring_buffer
*rb
;
4153 struct sk_buff_head
*rb_queue
;
4156 /* Added to avoid minimal code churn */
4157 struct tpacket_req
*req
= &req_u
->req
;
4159 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4160 if (!closing
&& tx_ring
&& (po
->tp_version
> TPACKET_V2
)) {
4161 WARN(1, "Tx-ring is not supported.\n");
4165 rb
= tx_ring
? &po
->tx_ring
: &po
->rx_ring
;
4166 rb_queue
= tx_ring
? &sk
->sk_write_queue
: &sk
->sk_receive_queue
;
4170 if (atomic_read(&po
->mapped
))
4172 if (packet_read_pending(rb
))
4176 if (req
->tp_block_nr
) {
4177 /* Sanity tests and some calculations */
4179 if (unlikely(rb
->pg_vec
))
4182 switch (po
->tp_version
) {
4184 po
->tp_hdrlen
= TPACKET_HDRLEN
;
4187 po
->tp_hdrlen
= TPACKET2_HDRLEN
;
4190 po
->tp_hdrlen
= TPACKET3_HDRLEN
;
4195 if (unlikely((int)req
->tp_block_size
<= 0))
4197 if (unlikely(!PAGE_ALIGNED(req
->tp_block_size
)))
4199 if (po
->tp_version
>= TPACKET_V3
&&
4200 (int)(req
->tp_block_size
-
4201 BLK_PLUS_PRIV(req_u
->req3
.tp_sizeof_priv
)) <= 0)
4203 if (unlikely(req
->tp_frame_size
< po
->tp_hdrlen
+
4206 if (unlikely(req
->tp_frame_size
& (TPACKET_ALIGNMENT
- 1)))
4209 rb
->frames_per_block
= req
->tp_block_size
/ req
->tp_frame_size
;
4210 if (unlikely(rb
->frames_per_block
== 0))
4212 if (unlikely((rb
->frames_per_block
* req
->tp_block_nr
) !=
4217 order
= get_order(req
->tp_block_size
);
4218 pg_vec
= alloc_pg_vec(req
, order
);
4219 if (unlikely(!pg_vec
))
4221 switch (po
->tp_version
) {
4223 /* Transmit path is not supported. We checked
4224 * it above but just being paranoid
4227 init_prb_bdqc(po
, rb
, pg_vec
, req_u
);
4236 if (unlikely(req
->tp_frame_nr
))
4242 /* Detach socket from network */
4243 spin_lock(&po
->bind_lock
);
4244 was_running
= po
->running
;
4248 __unregister_prot_hook(sk
, false);
4250 spin_unlock(&po
->bind_lock
);
4255 mutex_lock(&po
->pg_vec_lock
);
4256 if (closing
|| atomic_read(&po
->mapped
) == 0) {
4258 spin_lock_bh(&rb_queue
->lock
);
4259 swap(rb
->pg_vec
, pg_vec
);
4260 rb
->frame_max
= (req
->tp_frame_nr
- 1);
4262 rb
->frame_size
= req
->tp_frame_size
;
4263 spin_unlock_bh(&rb_queue
->lock
);
4265 swap(rb
->pg_vec_order
, order
);
4266 swap(rb
->pg_vec_len
, req
->tp_block_nr
);
4268 rb
->pg_vec_pages
= req
->tp_block_size
/PAGE_SIZE
;
4269 po
->prot_hook
.func
= (po
->rx_ring
.pg_vec
) ?
4270 tpacket_rcv
: packet_rcv
;
4271 skb_queue_purge(rb_queue
);
4272 if (atomic_read(&po
->mapped
))
4273 pr_err("packet_mmap: vma is busy: %d\n",
4274 atomic_read(&po
->mapped
));
4276 mutex_unlock(&po
->pg_vec_lock
);
4278 spin_lock(&po
->bind_lock
);
4281 register_prot_hook(sk
);
4283 spin_unlock(&po
->bind_lock
);
4284 if (closing
&& (po
->tp_version
> TPACKET_V2
)) {
4285 /* Because we don't support block-based V3 on tx-ring */
4287 prb_shutdown_retire_blk_timer(po
, rb_queue
);
4292 free_pg_vec(pg_vec
, order
, req
->tp_block_nr
);
4297 static int packet_mmap(struct file
*file
, struct socket
*sock
,
4298 struct vm_area_struct
*vma
)
4300 struct sock
*sk
= sock
->sk
;
4301 struct packet_sock
*po
= pkt_sk(sk
);
4302 unsigned long size
, expected_size
;
4303 struct packet_ring_buffer
*rb
;
4304 unsigned long start
;
4311 mutex_lock(&po
->pg_vec_lock
);
4314 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
4316 expected_size
+= rb
->pg_vec_len
4322 if (expected_size
== 0)
4325 size
= vma
->vm_end
- vma
->vm_start
;
4326 if (size
!= expected_size
)
4329 start
= vma
->vm_start
;
4330 for (rb
= &po
->rx_ring
; rb
<= &po
->tx_ring
; rb
++) {
4331 if (rb
->pg_vec
== NULL
)
4334 for (i
= 0; i
< rb
->pg_vec_len
; i
++) {
4336 void *kaddr
= rb
->pg_vec
[i
].buffer
;
4339 for (pg_num
= 0; pg_num
< rb
->pg_vec_pages
; pg_num
++) {
4340 page
= pgv_to_page(kaddr
);
4341 err
= vm_insert_page(vma
, start
, page
);
4350 atomic_inc(&po
->mapped
);
4351 vma
->vm_ops
= &packet_mmap_ops
;
4355 mutex_unlock(&po
->pg_vec_lock
);
4359 static const struct proto_ops packet_ops_spkt
= {
4360 .family
= PF_PACKET
,
4361 .owner
= THIS_MODULE
,
4362 .release
= packet_release
,
4363 .bind
= packet_bind_spkt
,
4364 .connect
= sock_no_connect
,
4365 .socketpair
= sock_no_socketpair
,
4366 .accept
= sock_no_accept
,
4367 .getname
= packet_getname_spkt
,
4368 .poll
= datagram_poll
,
4369 .ioctl
= packet_ioctl
,
4370 .listen
= sock_no_listen
,
4371 .shutdown
= sock_no_shutdown
,
4372 .setsockopt
= sock_no_setsockopt
,
4373 .getsockopt
= sock_no_getsockopt
,
4374 .sendmsg
= packet_sendmsg_spkt
,
4375 .recvmsg
= packet_recvmsg
,
4376 .mmap
= sock_no_mmap
,
4377 .sendpage
= sock_no_sendpage
,
4380 static const struct proto_ops packet_ops
= {
4381 .family
= PF_PACKET
,
4382 .owner
= THIS_MODULE
,
4383 .release
= packet_release
,
4384 .bind
= packet_bind
,
4385 .connect
= sock_no_connect
,
4386 .socketpair
= sock_no_socketpair
,
4387 .accept
= sock_no_accept
,
4388 .getname
= packet_getname
,
4389 .poll
= packet_poll
,
4390 .ioctl
= packet_ioctl
,
4391 .listen
= sock_no_listen
,
4392 .shutdown
= sock_no_shutdown
,
4393 .setsockopt
= packet_setsockopt
,
4394 .getsockopt
= packet_getsockopt
,
4395 .sendmsg
= packet_sendmsg
,
4396 .recvmsg
= packet_recvmsg
,
4397 .mmap
= packet_mmap
,
4398 .sendpage
= sock_no_sendpage
,
4401 static const struct net_proto_family packet_family_ops
= {
4402 .family
= PF_PACKET
,
4403 .create
= packet_create
,
4404 .owner
= THIS_MODULE
,
4407 static struct notifier_block packet_netdev_notifier
= {
4408 .notifier_call
= packet_notifier
,
4411 #ifdef CONFIG_PROC_FS
4413 static void *packet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4416 struct net
*net
= seq_file_net(seq
);
4419 return seq_hlist_start_head_rcu(&net
->packet
.sklist
, *pos
);
4422 static void *packet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4424 struct net
*net
= seq_file_net(seq
);
4425 return seq_hlist_next_rcu(v
, &net
->packet
.sklist
, pos
);
4428 static void packet_seq_stop(struct seq_file
*seq
, void *v
)
4434 static int packet_seq_show(struct seq_file
*seq
, void *v
)
4436 if (v
== SEQ_START_TOKEN
)
4437 seq_puts(seq
, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4439 struct sock
*s
= sk_entry(v
);
4440 const struct packet_sock
*po
= pkt_sk(s
);
4443 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4445 atomic_read(&s
->sk_refcnt
),
4450 atomic_read(&s
->sk_rmem_alloc
),
4451 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(s
)),
4458 static const struct seq_operations packet_seq_ops
= {
4459 .start
= packet_seq_start
,
4460 .next
= packet_seq_next
,
4461 .stop
= packet_seq_stop
,
4462 .show
= packet_seq_show
,
4465 static int packet_seq_open(struct inode
*inode
, struct file
*file
)
4467 return seq_open_net(inode
, file
, &packet_seq_ops
,
4468 sizeof(struct seq_net_private
));
4471 static const struct file_operations packet_seq_fops
= {
4472 .owner
= THIS_MODULE
,
4473 .open
= packet_seq_open
,
4475 .llseek
= seq_lseek
,
4476 .release
= seq_release_net
,
4481 static int __net_init
packet_net_init(struct net
*net
)
4483 mutex_init(&net
->packet
.sklist_lock
);
4484 INIT_HLIST_HEAD(&net
->packet
.sklist
);
4486 if (!proc_create("packet", 0, net
->proc_net
, &packet_seq_fops
))
4492 static void __net_exit
packet_net_exit(struct net
*net
)
4494 remove_proc_entry("packet", net
->proc_net
);
4497 static struct pernet_operations packet_net_ops
= {
4498 .init
= packet_net_init
,
4499 .exit
= packet_net_exit
,
4503 static void __exit
packet_exit(void)
4505 unregister_netdevice_notifier(&packet_netdev_notifier
);
4506 unregister_pernet_subsys(&packet_net_ops
);
4507 sock_unregister(PF_PACKET
);
4508 proto_unregister(&packet_proto
);
4511 static int __init
packet_init(void)
4513 int rc
= proto_register(&packet_proto
, 0);
4518 sock_register(&packet_family_ops
);
4519 register_pernet_subsys(&packet_net_ops
);
4520 register_netdevice_notifier(&packet_netdev_notifier
);
4525 module_init(packet_init
);
4526 module_exit(packet_exit
);
4527 MODULE_LICENSE("GPL");
4528 MODULE_ALIAS_NETPROTO(PF_PACKET
);