1 /*
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.
5 *
6 * PACKET - implements raw packet sockets.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 *
12 * Fixes:
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
41 * and packet_mreq.
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
44 * layer.
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
46 *
47 *
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.
52 *
53 */
55 #include <linux/types.h>
56 #include <linux/mm.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
60 #include <linux/in.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>
70 #include <net/ip.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <linux/uaccess.h>
77 #include <asm/ioctls.h>
78 #include <asm/page.h>
79 #include <asm/cacheflush.h>
80 #include <asm/io.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>
92 #ifdef CONFIG_INET
93 #include <net/inet_common.h>
94 #endif
95 #include <linux/bpf.h>
96 #include <net/compat.h>
98 #include "internal.h"
100 /*
101 Assumptions:
102 - if device has no dev->hard_header routine, it adds and removes ll header
103 inside itself. In this case ll header is invisible outside of device,
104 but higher levels still should reserve dev->hard_header_len.
105 Some devices are enough clever to reallocate skb, when header
106 will not fit to reserved space (tunnel), another ones are silly
107 (PPP).
108 - packet socket receives packets with pulled ll header,
109 so that SOCK_RAW should push it back.
111 On receive:
112 -----------
114 Incoming, dev->hard_header!=NULL
115 mac_header -> ll header
116 data -> data
118 Outgoing, dev->hard_header!=NULL
119 mac_header -> ll header
120 data -> ll header
122 Incoming, dev->hard_header==NULL
123 mac_header -> UNKNOWN position. It is very likely, that it points to ll
124 header. PPP makes it, that is wrong, because introduce
125 assymetry between rx and tx paths.
126 data -> data
128 Outgoing, dev->hard_header==NULL
129 mac_header -> data. ll header is still not built!
130 data -> data
132 Resume
133 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
136 On transmit:
137 ------------
139 dev->hard_header != NULL
140 mac_header -> ll header
141 data -> ll header
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
144 mac_header -> data
145 data -> data
147 We should set nh.raw on output to correct posistion,
148 packet classifier depends on it.
149 */
151 /* Private packet socket structures. */
153 /* identical to struct packet_mreq except it has
154 * a longer address field.
155 */
156 struct packet_mreq_max {
157 int mr_ifindex;
158 unsigned short mr_type;
159 unsigned short mr_alen;
160 unsigned char mr_address[MAX_ADDR_LEN];
161 };
163 union tpacket_uhdr {
164 struct tpacket_hdr *h1;
165 struct tpacket2_hdr *h2;
166 struct tpacket3_hdr *h3;
167 void *raw;
168 };
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171 int closing, int tx_ring);
173 #define V3_ALIGNMENT (8)
175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
188 struct packet_sock;
189 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
190 struct packet_type *pt, struct net_device *orig_dev);
192 static void *packet_previous_frame(struct packet_sock *po,
193 struct packet_ring_buffer *rb,
194 int status);
195 static void packet_increment_head(struct packet_ring_buffer *buff);
196 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
197 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
198 struct packet_sock *);
199 static void prb_retire_current_block(struct tpacket_kbdq_core *,
200 struct packet_sock *, unsigned int status);
201 static int prb_queue_frozen(struct tpacket_kbdq_core *);
202 static void prb_open_block(struct tpacket_kbdq_core *,
203 struct tpacket_block_desc *);
204 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
205 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
206 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
207 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
208 struct tpacket3_hdr *);
209 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
210 struct tpacket3_hdr *);
211 static void packet_flush_mclist(struct sock *sk);
212 static u16 packet_pick_tx_queue(struct sk_buff *skb);
214 struct packet_skb_cb {
215 union {
216 struct sockaddr_pkt pkt;
217 union {
218 /* Trick: alias skb original length with
219 * ll.sll_family and ll.protocol in order
220 * to save room.
221 */
222 unsigned int origlen;
223 struct sockaddr_ll ll;
224 };
225 } sa;
226 };
228 #define vio_le() virtio_legacy_is_little_endian()
230 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
232 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
233 #define GET_PBLOCK_DESC(x, bid) \
234 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
235 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
237 #define GET_NEXT_PRB_BLK_NUM(x) \
238 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
239 ((x)->kactive_blk_num+1) : 0)
241 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
242 static void __fanout_link(struct sock *sk, struct packet_sock *po);
244 static int packet_direct_xmit(struct sk_buff *skb)
245 {
246 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
247 }
249 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
250 {
251 struct net_device *dev;
253 rcu_read_lock();
254 dev = rcu_dereference(po->cached_dev);
255 if (likely(dev))
256 dev_hold(dev);
257 rcu_read_unlock();
259 return dev;
260 }
262 static void packet_cached_dev_assign(struct packet_sock *po,
263 struct net_device *dev)
264 {
265 rcu_assign_pointer(po->cached_dev, dev);
266 }
268 static void packet_cached_dev_reset(struct packet_sock *po)
269 {
270 RCU_INIT_POINTER(po->cached_dev, NULL);
271 }
273 static bool packet_use_direct_xmit(const struct packet_sock *po)
274 {
275 return po->xmit == packet_direct_xmit;
276 }
278 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb,
279 struct net_device *sb_dev)
280 {
281 return dev_pick_tx_cpu_id(dev, skb, sb_dev, NULL);
282 }
284 static u16 packet_pick_tx_queue(struct sk_buff *skb)
285 {
286 struct net_device *dev = skb->dev;
287 const struct net_device_ops *ops = dev->netdev_ops;
288 u16 queue_index;
290 if (ops->ndo_select_queue) {
291 queue_index = ops->ndo_select_queue(dev, skb, NULL,
292 __packet_pick_tx_queue);
293 queue_index = netdev_cap_txqueue(dev, queue_index);
294 } else {
295 queue_index = __packet_pick_tx_queue(dev, skb, NULL);
296 }
298 return queue_index;
299 }
301 /* __register_prot_hook must be invoked through register_prot_hook
302 * or from a context in which asynchronous accesses to the packet
303 * socket is not possible (packet_create()).
304 */
305 static void __register_prot_hook(struct sock *sk)
306 {
307 struct packet_sock *po = pkt_sk(sk);
309 if (!po->running) {
310 if (po->fanout)
311 __fanout_link(sk, po);
312 else
313 dev_add_pack(&po->prot_hook);
315 sock_hold(sk);
316 po->running = 1;
317 }
318 }
320 static void register_prot_hook(struct sock *sk)
321 {
322 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
323 __register_prot_hook(sk);
324 }
326 /* If the sync parameter is true, we will temporarily drop
327 * the po->bind_lock and do a synchronize_net to make sure no
328 * asynchronous packet processing paths still refer to the elements
329 * of po->prot_hook. If the sync parameter is false, it is the
330 * callers responsibility to take care of this.
331 */
332 static void __unregister_prot_hook(struct sock *sk, bool sync)
333 {
334 struct packet_sock *po = pkt_sk(sk);
336 lockdep_assert_held_once(&po->bind_lock);
338 po->running = 0;
340 if (po->fanout)
341 __fanout_unlink(sk, po);
342 else
343 __dev_remove_pack(&po->prot_hook);
345 __sock_put(sk);
347 if (sync) {
348 spin_unlock(&po->bind_lock);
349 synchronize_net();
350 spin_lock(&po->bind_lock);
351 }
352 }
354 static void unregister_prot_hook(struct sock *sk, bool sync)
355 {
356 struct packet_sock *po = pkt_sk(sk);
358 if (po->running)
359 __unregister_prot_hook(sk, sync);
360 }
362 static inline struct page * __pure pgv_to_page(void *addr)
363 {
364 if (is_vmalloc_addr(addr))
365 return vmalloc_to_page(addr);
366 return virt_to_page(addr);
367 }
369 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
370 {
371 union tpacket_uhdr h;
373 h.raw = frame;
374 switch (po->tp_version) {
375 case TPACKET_V1:
376 h.h1->tp_status = status;
377 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
378 break;
379 case TPACKET_V2:
380 h.h2->tp_status = status;
381 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
382 break;
383 case TPACKET_V3:
384 h.h3->tp_status = status;
385 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
386 break;
387 default:
388 WARN(1, "TPACKET version not supported.\n");
389 BUG();
390 }
392 smp_wmb();
393 }
395 static int __packet_get_status(struct packet_sock *po, void *frame)
396 {
397 union tpacket_uhdr h;
399 smp_rmb();
401 h.raw = frame;
402 switch (po->tp_version) {
403 case TPACKET_V1:
404 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
405 return h.h1->tp_status;
406 case TPACKET_V2:
407 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
408 return h.h2->tp_status;
409 case TPACKET_V3:
410 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
411 return h.h3->tp_status;
412 default:
413 WARN(1, "TPACKET version not supported.\n");
414 BUG();
415 return 0;
416 }
417 }
419 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
420 unsigned int flags)
421 {
422 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
424 if (shhwtstamps &&
425 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
426 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
427 return TP_STATUS_TS_RAW_HARDWARE;
429 if (ktime_to_timespec_cond(skb->tstamp, ts))
430 return TP_STATUS_TS_SOFTWARE;
432 return 0;
433 }
435 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
436 struct sk_buff *skb)
437 {
438 union tpacket_uhdr h;
439 struct timespec ts;
440 __u32 ts_status;
442 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
443 return 0;
445 h.raw = frame;
446 switch (po->tp_version) {
447 case TPACKET_V1:
448 h.h1->tp_sec = ts.tv_sec;
449 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
450 break;
451 case TPACKET_V2:
452 h.h2->tp_sec = ts.tv_sec;
453 h.h2->tp_nsec = ts.tv_nsec;
454 break;
455 case TPACKET_V3:
456 h.h3->tp_sec = ts.tv_sec;
457 h.h3->tp_nsec = ts.tv_nsec;
458 break;
459 default:
460 WARN(1, "TPACKET version not supported.\n");
461 BUG();
462 }
464 /* one flush is safe, as both fields always lie on the same cacheline */
465 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
466 smp_wmb();
468 return ts_status;
469 }
471 static void *packet_lookup_frame(struct packet_sock *po,
472 struct packet_ring_buffer *rb,
473 unsigned int position,
474 int status)
475 {
476 unsigned int pg_vec_pos, frame_offset;
477 union tpacket_uhdr h;
479 pg_vec_pos = position / rb->frames_per_block;
480 frame_offset = position % rb->frames_per_block;
482 h.raw = rb->pg_vec[pg_vec_pos].buffer +
483 (frame_offset * rb->frame_size);
485 if (status != __packet_get_status(po, h.raw))
486 return NULL;
488 return h.raw;
489 }
491 static void *packet_current_frame(struct packet_sock *po,
492 struct packet_ring_buffer *rb,
493 int status)
494 {
495 return packet_lookup_frame(po, rb, rb->head, status);
496 }
498 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
499 {
500 del_timer_sync(&pkc->retire_blk_timer);
501 }
503 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
504 struct sk_buff_head *rb_queue)
505 {
506 struct tpacket_kbdq_core *pkc;
508 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
510 spin_lock_bh(&rb_queue->lock);
511 pkc->delete_blk_timer = 1;
512 spin_unlock_bh(&rb_queue->lock);
514 prb_del_retire_blk_timer(pkc);
515 }
517 static void prb_setup_retire_blk_timer(struct packet_sock *po)
518 {
519 struct tpacket_kbdq_core *pkc;
521 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
522 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
523 0);
524 pkc->retire_blk_timer.expires = jiffies;
525 }
527 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
528 int blk_size_in_bytes)
529 {
530 struct net_device *dev;
531 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
532 struct ethtool_link_ksettings ecmd;
533 int err;
535 rtnl_lock();
536 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
537 if (unlikely(!dev)) {
538 rtnl_unlock();
539 return DEFAULT_PRB_RETIRE_TOV;
540 }
541 err = __ethtool_get_link_ksettings(dev, &ecmd);
542 rtnl_unlock();
543 if (!err) {
544 /*
545 * If the link speed is so slow you don't really
546 * need to worry about perf anyways
547 */
548 if (ecmd.base.speed < SPEED_1000 ||
549 ecmd.base.speed == SPEED_UNKNOWN) {
550 return DEFAULT_PRB_RETIRE_TOV;
551 } else {
552 msec = 1;
553 div = ecmd.base.speed / 1000;
554 }
555 }
557 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
559 if (div)
560 mbits /= div;
562 tmo = mbits * msec;
564 if (div)
565 return tmo+1;
566 return tmo;
567 }
569 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
570 union tpacket_req_u *req_u)
571 {
572 p1->feature_req_word = req_u->req3.tp_feature_req_word;
573 }
575 static void init_prb_bdqc(struct packet_sock *po,
576 struct packet_ring_buffer *rb,
577 struct pgv *pg_vec,
578 union tpacket_req_u *req_u)
579 {
580 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
581 struct tpacket_block_desc *pbd;
583 memset(p1, 0x0, sizeof(*p1));
585 p1->knxt_seq_num = 1;
586 p1->pkbdq = pg_vec;
587 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
588 p1->pkblk_start = pg_vec[0].buffer;
589 p1->kblk_size = req_u->req3.tp_block_size;
590 p1->knum_blocks = req_u->req3.tp_block_nr;
591 p1->hdrlen = po->tp_hdrlen;
592 p1->version = po->tp_version;
593 p1->last_kactive_blk_num = 0;
594 po->stats.stats3.tp_freeze_q_cnt = 0;
595 if (req_u->req3.tp_retire_blk_tov)
596 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
597 else
598 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
599 req_u->req3.tp_block_size);
600 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
601 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
603 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
604 prb_init_ft_ops(p1, req_u);
605 prb_setup_retire_blk_timer(po);
606 prb_open_block(p1, pbd);
607 }
609 /* Do NOT update the last_blk_num first.
610 * Assumes sk_buff_head lock is held.
611 */
612 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
613 {
614 mod_timer(&pkc->retire_blk_timer,
615 jiffies + pkc->tov_in_jiffies);
616 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
617 }
619 /*
620 * Timer logic:
621 * 1) We refresh the timer only when we open a block.
622 * By doing this we don't waste cycles refreshing the timer
623 * on packet-by-packet basis.
624 *
625 * With a 1MB block-size, on a 1Gbps line, it will take
626 * i) ~8 ms to fill a block + ii) memcpy etc.
627 * In this cut we are not accounting for the memcpy time.
628 *
629 * So, if the user sets the 'tmo' to 10ms then the timer
630 * will never fire while the block is still getting filled
631 * (which is what we want). However, the user could choose
632 * to close a block early and that's fine.
633 *
634 * But when the timer does fire, we check whether or not to refresh it.
635 * Since the tmo granularity is in msecs, it is not too expensive
636 * to refresh the timer, lets say every '8' msecs.
637 * Either the user can set the 'tmo' or we can derive it based on
638 * a) line-speed and b) block-size.
639 * prb_calc_retire_blk_tmo() calculates the tmo.
640 *
641 */
642 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
643 {
644 struct packet_sock *po =
645 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
646 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
647 unsigned int frozen;
648 struct tpacket_block_desc *pbd;
650 spin_lock(&po->sk.sk_receive_queue.lock);
652 frozen = prb_queue_frozen(pkc);
653 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
655 if (unlikely(pkc->delete_blk_timer))
656 goto out;
658 /* We only need to plug the race when the block is partially filled.
659 * tpacket_rcv:
660 * lock(); increment BLOCK_NUM_PKTS; unlock()
661 * copy_bits() is in progress ...
662 * timer fires on other cpu:
663 * we can't retire the current block because copy_bits
664 * is in progress.
665 *
666 */
667 if (BLOCK_NUM_PKTS(pbd)) {
668 while (atomic_read(&pkc->blk_fill_in_prog)) {
669 /* Waiting for skb_copy_bits to finish... */
670 cpu_relax();
671 }
672 }
674 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
675 if (!frozen) {
676 if (!BLOCK_NUM_PKTS(pbd)) {
677 /* An empty block. Just refresh the timer. */
678 goto refresh_timer;
679 }
680 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
681 if (!prb_dispatch_next_block(pkc, po))
682 goto refresh_timer;
683 else
684 goto out;
685 } else {
686 /* Case 1. Queue was frozen because user-space was
687 * lagging behind.
688 */
689 if (prb_curr_blk_in_use(pbd)) {
690 /*
691 * Ok, user-space is still behind.
692 * So just refresh the timer.
693 */
694 goto refresh_timer;
695 } else {
696 /* Case 2. queue was frozen,user-space caught up,
697 * now the link went idle && the timer fired.
698 * We don't have a block to close.So we open this
699 * block and restart the timer.
700 * opening a block thaws the queue,restarts timer
701 * Thawing/timer-refresh is a side effect.
702 */
703 prb_open_block(pkc, pbd);
704 goto out;
705 }
706 }
707 }
709 refresh_timer:
710 _prb_refresh_rx_retire_blk_timer(pkc);
712 out:
713 spin_unlock(&po->sk.sk_receive_queue.lock);
714 }
716 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
717 struct tpacket_block_desc *pbd1, __u32 status)
718 {
719 /* Flush everything minus the block header */
721 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
722 u8 *start, *end;
724 start = (u8 *)pbd1;
726 /* Skip the block header(we know header WILL fit in 4K) */
727 start += PAGE_SIZE;
729 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
730 for (; start < end; start += PAGE_SIZE)
731 flush_dcache_page(pgv_to_page(start));
733 smp_wmb();
734 #endif
736 /* Now update the block status. */
738 BLOCK_STATUS(pbd1) = status;
740 /* Flush the block header */
742 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
743 start = (u8 *)pbd1;
744 flush_dcache_page(pgv_to_page(start));
746 smp_wmb();
747 #endif
748 }
750 /*
751 * Side effect:
752 *
753 * 1) flush the block
754 * 2) Increment active_blk_num
755 *
756 * Note:We DONT refresh the timer on purpose.
757 * Because almost always the next block will be opened.
758 */
759 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
760 struct tpacket_block_desc *pbd1,
761 struct packet_sock *po, unsigned int stat)
762 {
763 __u32 status = TP_STATUS_USER | stat;
765 struct tpacket3_hdr *last_pkt;
766 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
767 struct sock *sk = &po->sk;
769 if (po->stats.stats3.tp_drops)
770 status |= TP_STATUS_LOSING;
772 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
773 last_pkt->tp_next_offset = 0;
775 /* Get the ts of the last pkt */
776 if (BLOCK_NUM_PKTS(pbd1)) {
777 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
778 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
779 } else {
780 /* Ok, we tmo'd - so get the current time.
781 *
782 * It shouldn't really happen as we don't close empty
783 * blocks. See prb_retire_rx_blk_timer_expired().
784 */
785 struct timespec ts;
786 getnstimeofday(&ts);
787 h1->ts_last_pkt.ts_sec = ts.tv_sec;
788 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
789 }
791 smp_wmb();
793 /* Flush the block */
794 prb_flush_block(pkc1, pbd1, status);
796 sk->sk_data_ready(sk);
798 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
799 }
801 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
802 {
803 pkc->reset_pending_on_curr_blk = 0;
804 }
806 /*
807 * Side effect of opening a block:
808 *
809 * 1) prb_queue is thawed.
810 * 2) retire_blk_timer is refreshed.
811 *
812 */
813 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
814 struct tpacket_block_desc *pbd1)
815 {
816 struct timespec ts;
817 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
819 smp_rmb();
821 /* We could have just memset this but we will lose the
822 * flexibility of making the priv area sticky
823 */
825 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
826 BLOCK_NUM_PKTS(pbd1) = 0;
827 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
829 getnstimeofday(&ts);
831 h1->ts_first_pkt.ts_sec = ts.tv_sec;
832 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
834 pkc1->pkblk_start = (char *)pbd1;
835 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
837 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
838 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
840 pbd1->version = pkc1->version;
841 pkc1->prev = pkc1->nxt_offset;
842 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
844 prb_thaw_queue(pkc1);
845 _prb_refresh_rx_retire_blk_timer(pkc1);
847 smp_wmb();
848 }
850 /*
851 * Queue freeze logic:
852 * 1) Assume tp_block_nr = 8 blocks.
853 * 2) At time 't0', user opens Rx ring.
854 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
855 * 4) user-space is either sleeping or processing block '0'.
856 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
857 * it will close block-7,loop around and try to fill block '0'.
858 * call-flow:
859 * __packet_lookup_frame_in_block
860 * prb_retire_current_block()
861 * prb_dispatch_next_block()
862 * |->(BLOCK_STATUS == USER) evaluates to true
863 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
864 * 6) Now there are two cases:
865 * 6.1) Link goes idle right after the queue is frozen.
866 * But remember, the last open_block() refreshed the timer.
867 * When this timer expires,it will refresh itself so that we can
868 * re-open block-0 in near future.
869 * 6.2) Link is busy and keeps on receiving packets. This is a simple
870 * case and __packet_lookup_frame_in_block will check if block-0
871 * is free and can now be re-used.
872 */
873 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
874 struct packet_sock *po)
875 {
876 pkc->reset_pending_on_curr_blk = 1;
877 po->stats.stats3.tp_freeze_q_cnt++;
878 }
880 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
882 /*
883 * If the next block is free then we will dispatch it
884 * and return a good offset.
885 * Else, we will freeze the queue.
886 * So, caller must check the return value.
887 */
888 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
889 struct packet_sock *po)
890 {
891 struct tpacket_block_desc *pbd;
893 smp_rmb();
895 /* 1. Get current block num */
896 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
898 /* 2. If this block is currently in_use then freeze the queue */
899 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
900 prb_freeze_queue(pkc, po);
901 return NULL;
902 }
904 /*
905 * 3.
906 * open this block and return the offset where the first packet
907 * needs to get stored.
908 */
909 prb_open_block(pkc, pbd);
910 return (void *)pkc->nxt_offset;
911 }
913 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
914 struct packet_sock *po, unsigned int status)
915 {
916 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
918 /* retire/close the current block */
919 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
920 /*
921 * Plug the case where copy_bits() is in progress on
922 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
923 * have space to copy the pkt in the current block and
924 * called prb_retire_current_block()
925 *
926 * We don't need to worry about the TMO case because
927 * the timer-handler already handled this case.
928 */
929 if (!(status & TP_STATUS_BLK_TMO)) {
930 while (atomic_read(&pkc->blk_fill_in_prog)) {
931 /* Waiting for skb_copy_bits to finish... */
932 cpu_relax();
933 }
934 }
935 prb_close_block(pkc, pbd, po, status);
936 return;
937 }
938 }
940 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
941 {
942 return TP_STATUS_USER & BLOCK_STATUS(pbd);
943 }
945 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
946 {
947 return pkc->reset_pending_on_curr_blk;
948 }
950 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
951 {
952 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
953 atomic_dec(&pkc->blk_fill_in_prog);
954 }
956 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
957 struct tpacket3_hdr *ppd)
958 {
959 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
960 }
962 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
963 struct tpacket3_hdr *ppd)
964 {
965 ppd->hv1.tp_rxhash = 0;
966 }
968 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
969 struct tpacket3_hdr *ppd)
970 {
971 if (skb_vlan_tag_present(pkc->skb)) {
972 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
973 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
974 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
975 } else {
976 ppd->hv1.tp_vlan_tci = 0;
977 ppd->hv1.tp_vlan_tpid = 0;
978 ppd->tp_status = TP_STATUS_AVAILABLE;
979 }
980 }
982 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
983 struct tpacket3_hdr *ppd)
984 {
985 ppd->hv1.tp_padding = 0;
986 prb_fill_vlan_info(pkc, ppd);
988 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
989 prb_fill_rxhash(pkc, ppd);
990 else
991 prb_clear_rxhash(pkc, ppd);
992 }
994 static void prb_fill_curr_block(char *curr,
995 struct tpacket_kbdq_core *pkc,
996 struct tpacket_block_desc *pbd,
997 unsigned int len)
998 {
999 struct tpacket3_hdr *ppd;
1001 ppd = (struct tpacket3_hdr *)curr;
1002 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1003 pkc->prev = curr;
1004 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006 BLOCK_NUM_PKTS(pbd) += 1;
1007 atomic_inc(&pkc->blk_fill_in_prog);
1008 prb_run_all_ft_ops(pkc, ppd);
1009 }
1011 /* Assumes caller has the sk->rx_queue.lock */
1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013 struct sk_buff *skb,
1014 int status,
1015 unsigned int len
1016 )
1017 {
1018 struct tpacket_kbdq_core *pkc;
1019 struct tpacket_block_desc *pbd;
1020 char *curr, *end;
1022 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1023 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1025 /* Queue is frozen when user space is lagging behind */
1026 if (prb_queue_frozen(pkc)) {
1027 /*
1028 * Check if that last block which caused the queue to freeze,
1029 * is still in_use by user-space.
1030 */
1031 if (prb_curr_blk_in_use(pbd)) {
1032 /* Can't record this packet */
1033 return NULL;
1034 } else {
1035 /*
1036 * Ok, the block was released by user-space.
1037 * Now let's open that block.
1038 * opening a block also thaws the queue.
1039 * Thawing is a side effect.
1040 */
1041 prb_open_block(pkc, pbd);
1042 }
1043 }
1045 smp_mb();
1046 curr = pkc->nxt_offset;
1047 pkc->skb = skb;
1048 end = (char *)pbd + pkc->kblk_size;
1050 /* first try the current block */
1051 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052 prb_fill_curr_block(curr, pkc, pbd, len);
1053 return (void *)curr;
1054 }
1056 /* Ok, close the current block */
1057 prb_retire_current_block(pkc, po, 0);
1059 /* Now, try to dispatch the next block */
1060 curr = (char *)prb_dispatch_next_block(pkc, po);
1061 if (curr) {
1062 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063 prb_fill_curr_block(curr, pkc, pbd, len);
1064 return (void *)curr;
1065 }
1067 /*
1068 * No free blocks are available.user_space hasn't caught up yet.
1069 * Queue was just frozen and now this packet will get dropped.
1070 */
1071 return NULL;
1072 }
1074 static void *packet_current_rx_frame(struct packet_sock *po,
1075 struct sk_buff *skb,
1076 int status, unsigned int len)
1077 {
1078 char *curr = NULL;
1079 switch (po->tp_version) {
1080 case TPACKET_V1:
1081 case TPACKET_V2:
1082 curr = packet_lookup_frame(po, &po->rx_ring,
1083 po->rx_ring.head, status);
1084 return curr;
1085 case TPACKET_V3:
1086 return __packet_lookup_frame_in_block(po, skb, status, len);
1087 default:
1088 WARN(1, "TPACKET version not supported\n");
1089 BUG();
1090 return NULL;
1091 }
1092 }
1094 static void *prb_lookup_block(struct packet_sock *po,
1095 struct packet_ring_buffer *rb,
1096 unsigned int idx,
1097 int status)
1098 {
1099 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1100 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1102 if (status != BLOCK_STATUS(pbd))
1103 return NULL;
1104 return pbd;
1105 }
1107 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1108 {
1109 unsigned int prev;
1110 if (rb->prb_bdqc.kactive_blk_num)
1111 prev = rb->prb_bdqc.kactive_blk_num-1;
1112 else
1113 prev = rb->prb_bdqc.knum_blocks-1;
1114 return prev;
1115 }
1117 /* Assumes caller has held the rx_queue.lock */
1118 static void *__prb_previous_block(struct packet_sock *po,
1119 struct packet_ring_buffer *rb,
1120 int status)
1121 {
1122 unsigned int previous = prb_previous_blk_num(rb);
1123 return prb_lookup_block(po, rb, previous, status);
1124 }
1126 static void *packet_previous_rx_frame(struct packet_sock *po,
1127 struct packet_ring_buffer *rb,
1128 int status)
1129 {
1130 if (po->tp_version <= TPACKET_V2)
1131 return packet_previous_frame(po, rb, status);
1133 return __prb_previous_block(po, rb, status);
1134 }
1136 static void packet_increment_rx_head(struct packet_sock *po,
1137 struct packet_ring_buffer *rb)
1138 {
1139 switch (po->tp_version) {
1140 case TPACKET_V1:
1141 case TPACKET_V2:
1142 return packet_increment_head(rb);
1143 case TPACKET_V3:
1144 default:
1145 WARN(1, "TPACKET version not supported.\n");
1146 BUG();
1147 return;
1148 }
1149 }
1151 static void *packet_previous_frame(struct packet_sock *po,
1152 struct packet_ring_buffer *rb,
1153 int status)
1154 {
1155 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156 return packet_lookup_frame(po, rb, previous, status);
1157 }
1159 static void packet_increment_head(struct packet_ring_buffer *buff)
1160 {
1161 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1162 }
1164 static void packet_inc_pending(struct packet_ring_buffer *rb)
1165 {
1166 this_cpu_inc(*rb->pending_refcnt);
1167 }
1169 static void packet_dec_pending(struct packet_ring_buffer *rb)
1170 {
1171 this_cpu_dec(*rb->pending_refcnt);
1172 }
1174 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1175 {
1176 unsigned int refcnt = 0;
1177 int cpu;
1179 /* We don't use pending refcount in rx_ring. */
1180 if (rb->pending_refcnt == NULL)
1181 return 0;
1183 for_each_possible_cpu(cpu)
1184 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1186 return refcnt;
1187 }
1189 static int packet_alloc_pending(struct packet_sock *po)
1190 {
1191 po->rx_ring.pending_refcnt = NULL;
1193 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1194 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1195 return -ENOBUFS;
1197 return 0;
1198 }
1200 static void packet_free_pending(struct packet_sock *po)
1201 {
1202 free_percpu(po->tx_ring.pending_refcnt);
1203 }
1205 #define ROOM_POW_OFF 2
1206 #define ROOM_NONE 0x0
1207 #define ROOM_LOW 0x1
1208 #define ROOM_NORMAL 0x2
1210 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1211 {
1212 int idx, len;
1214 len = po->rx_ring.frame_max + 1;
1215 idx = po->rx_ring.head;
1216 if (pow_off)
1217 idx += len >> pow_off;
1218 if (idx >= len)
1219 idx -= len;
1220 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1221 }
1223 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1224 {
1225 int idx, len;
1227 len = po->rx_ring.prb_bdqc.knum_blocks;
1228 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1229 if (pow_off)
1230 idx += len >> pow_off;
1231 if (idx >= len)
1232 idx -= len;
1233 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1234 }
1236 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1237 {
1238 struct sock *sk = &po->sk;
1239 int ret = ROOM_NONE;
1241 if (po->prot_hook.func != tpacket_rcv) {
1242 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1243 - (skb ? skb->truesize : 0);
1244 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1245 return ROOM_NORMAL;
1246 else if (avail > 0)
1247 return ROOM_LOW;
1248 else
1249 return ROOM_NONE;
1250 }
1252 if (po->tp_version == TPACKET_V3) {
1253 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1254 ret = ROOM_NORMAL;
1255 else if (__tpacket_v3_has_room(po, 0))
1256 ret = ROOM_LOW;
1257 } else {
1258 if (__tpacket_has_room(po, ROOM_POW_OFF))
1259 ret = ROOM_NORMAL;
1260 else if (__tpacket_has_room(po, 0))
1261 ret = ROOM_LOW;
1262 }
1264 return ret;
1265 }
1267 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1268 {
1269 int ret;
1270 bool has_room;
1272 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1273 ret = __packet_rcv_has_room(po, skb);
1274 has_room = ret == ROOM_NORMAL;
1275 if (po->pressure == has_room)
1276 po->pressure = !has_room;
1277 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1279 return ret;
1280 }
1282 static void packet_sock_destruct(struct sock *sk)
1283 {
1284 skb_queue_purge(&sk->sk_error_queue);
1286 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1287 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1289 if (!sock_flag(sk, SOCK_DEAD)) {
1290 pr_err("Attempt to release alive packet socket: %p\n", sk);
1291 return;
1292 }
1294 sk_refcnt_debug_dec(sk);
1295 }
1297 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1298 {
1299 u32 rxhash;
1300 int i, count = 0;
1302 rxhash = skb_get_hash(skb);
1303 for (i = 0; i < ROLLOVER_HLEN; i++)
1304 if (po->rollover->history[i] == rxhash)
1305 count++;
1307 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1308 return count > (ROLLOVER_HLEN >> 1);
1309 }
1311 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1312 struct sk_buff *skb,
1313 unsigned int num)
1314 {
1315 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1316 }
1318 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1319 struct sk_buff *skb,
1320 unsigned int num)
1321 {
1322 unsigned int val = atomic_inc_return(&f->rr_cur);
1324 return val % num;
1325 }
1327 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1328 struct sk_buff *skb,
1329 unsigned int num)
1330 {
1331 return smp_processor_id() % num;
1332 }
1334 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1335 struct sk_buff *skb,
1336 unsigned int num)
1337 {
1338 return prandom_u32_max(num);
1339 }
1341 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1342 struct sk_buff *skb,
1343 unsigned int idx, bool try_self,
1344 unsigned int num)
1345 {
1346 struct packet_sock *po, *po_next, *po_skip = NULL;
1347 unsigned int i, j, room = ROOM_NONE;
1349 po = pkt_sk(f->arr[idx]);
1351 if (try_self) {
1352 room = packet_rcv_has_room(po, skb);
1353 if (room == ROOM_NORMAL ||
1354 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1355 return idx;
1356 po_skip = po;
1357 }
1359 i = j = min_t(int, po->rollover->sock, num - 1);
1360 do {
1361 po_next = pkt_sk(f->arr[i]);
1362 if (po_next != po_skip && !po_next->pressure &&
1363 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1364 if (i != j)
1365 po->rollover->sock = i;
1366 atomic_long_inc(&po->rollover->num);
1367 if (room == ROOM_LOW)
1368 atomic_long_inc(&po->rollover->num_huge);
1369 return i;
1370 }
1372 if (++i == num)
1373 i = 0;
1374 } while (i != j);
1376 atomic_long_inc(&po->rollover->num_failed);
1377 return idx;
1378 }
1380 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1381 struct sk_buff *skb,
1382 unsigned int num)
1383 {
1384 return skb_get_queue_mapping(skb) % num;
1385 }
1387 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1388 struct sk_buff *skb,
1389 unsigned int num)
1390 {
1391 struct bpf_prog *prog;
1392 unsigned int ret = 0;
1394 rcu_read_lock();
1395 prog = rcu_dereference(f->bpf_prog);
1396 if (prog)
1397 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1398 rcu_read_unlock();
1400 return ret;
1401 }
1403 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1404 {
1405 return f->flags & (flag >> 8);
1406 }
1408 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1409 struct packet_type *pt, struct net_device *orig_dev)
1410 {
1411 struct packet_fanout *f = pt->af_packet_priv;
1412 unsigned int num = READ_ONCE(f->num_members);
1413 struct net *net = read_pnet(&f->net);
1414 struct packet_sock *po;
1415 unsigned int idx;
1417 if (!net_eq(dev_net(dev), net) || !num) {
1418 kfree_skb(skb);
1419 return 0;
1420 }
1422 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1423 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1424 if (!skb)
1425 return 0;
1426 }
1427 switch (f->type) {
1428 case PACKET_FANOUT_HASH:
1429 default:
1430 idx = fanout_demux_hash(f, skb, num);
1431 break;
1432 case PACKET_FANOUT_LB:
1433 idx = fanout_demux_lb(f, skb, num);
1434 break;
1435 case PACKET_FANOUT_CPU:
1436 idx = fanout_demux_cpu(f, skb, num);
1437 break;
1438 case PACKET_FANOUT_RND:
1439 idx = fanout_demux_rnd(f, skb, num);
1440 break;
1441 case PACKET_FANOUT_QM:
1442 idx = fanout_demux_qm(f, skb, num);
1443 break;
1444 case PACKET_FANOUT_ROLLOVER:
1445 idx = fanout_demux_rollover(f, skb, 0, false, num);
1446 break;
1447 case PACKET_FANOUT_CBPF:
1448 case PACKET_FANOUT_EBPF:
1449 idx = fanout_demux_bpf(f, skb, num);
1450 break;
1451 }
1453 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1454 idx = fanout_demux_rollover(f, skb, idx, true, num);
1456 po = pkt_sk(f->arr[idx]);
1457 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1458 }
1460 DEFINE_MUTEX(fanout_mutex);
1461 EXPORT_SYMBOL_GPL(fanout_mutex);
1462 static LIST_HEAD(fanout_list);
1463 static u16 fanout_next_id;
1465 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1466 {
1467 struct packet_fanout *f = po->fanout;
1469 spin_lock(&f->lock);
1470 f->arr[f->num_members] = sk;
1471 smp_wmb();
1472 f->num_members++;
1473 if (f->num_members == 1)
1474 dev_add_pack(&f->prot_hook);
1475 spin_unlock(&f->lock);
1476 }
1478 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1479 {
1480 struct packet_fanout *f = po->fanout;
1481 int i;
1483 spin_lock(&f->lock);
1484 for (i = 0; i < f->num_members; i++) {
1485 if (f->arr[i] == sk)
1486 break;
1487 }
1488 BUG_ON(i >= f->num_members);
1489 f->arr[i] = f->arr[f->num_members - 1];
1490 f->num_members--;
1491 if (f->num_members == 0)
1492 __dev_remove_pack(&f->prot_hook);
1493 spin_unlock(&f->lock);
1494 }
1496 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1497 {
1498 if (sk->sk_family != PF_PACKET)
1499 return false;
1501 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1502 }
1504 static void fanout_init_data(struct packet_fanout *f)
1505 {
1506 switch (f->type) {
1507 case PACKET_FANOUT_LB:
1508 atomic_set(&f->rr_cur, 0);
1509 break;
1510 case PACKET_FANOUT_CBPF:
1511 case PACKET_FANOUT_EBPF:
1512 RCU_INIT_POINTER(f->bpf_prog, NULL);
1513 break;
1514 }
1515 }
1517 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1518 {
1519 struct bpf_prog *old;
1521 spin_lock(&f->lock);
1522 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1523 rcu_assign_pointer(f->bpf_prog, new);
1524 spin_unlock(&f->lock);
1526 if (old) {
1527 synchronize_net();
1528 bpf_prog_destroy(old);
1529 }
1530 }
1532 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1533 unsigned int len)
1534 {
1535 struct bpf_prog *new;
1536 struct sock_fprog fprog;
1537 int ret;
1539 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1540 return -EPERM;
1541 if (len != sizeof(fprog))
1542 return -EINVAL;
1543 if (copy_from_user(&fprog, data, len))
1544 return -EFAULT;
1546 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1547 if (ret)
1548 return ret;
1550 __fanout_set_data_bpf(po->fanout, new);
1551 return 0;
1552 }
1554 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1555 unsigned int len)
1556 {
1557 struct bpf_prog *new;
1558 u32 fd;
1560 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1561 return -EPERM;
1562 if (len != sizeof(fd))
1563 return -EINVAL;
1564 if (copy_from_user(&fd, data, len))
1565 return -EFAULT;
1567 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1568 if (IS_ERR(new))
1569 return PTR_ERR(new);
1571 __fanout_set_data_bpf(po->fanout, new);
1572 return 0;
1573 }
1575 static int fanout_set_data(struct packet_sock *po, char __user *data,
1576 unsigned int len)
1577 {
1578 switch (po->fanout->type) {
1579 case PACKET_FANOUT_CBPF:
1580 return fanout_set_data_cbpf(po, data, len);
1581 case PACKET_FANOUT_EBPF:
1582 return fanout_set_data_ebpf(po, data, len);
1583 default:
1584 return -EINVAL;
1585 }
1586 }
1588 static void fanout_release_data(struct packet_fanout *f)
1589 {
1590 switch (f->type) {
1591 case PACKET_FANOUT_CBPF:
1592 case PACKET_FANOUT_EBPF:
1593 __fanout_set_data_bpf(f, NULL);
1594 }
1595 }
1597 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1598 {
1599 struct packet_fanout *f;
1601 list_for_each_entry(f, &fanout_list, list) {
1602 if (f->id == candidate_id &&
1603 read_pnet(&f->net) == sock_net(sk)) {
1604 return false;
1605 }
1606 }
1607 return true;
1608 }
1610 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1611 {
1612 u16 id = fanout_next_id;
1614 do {
1615 if (__fanout_id_is_free(sk, id)) {
1616 *new_id = id;
1617 fanout_next_id = id + 1;
1618 return true;
1619 }
1621 id++;
1622 } while (id != fanout_next_id);
1624 return false;
1625 }
1627 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1628 {
1629 struct packet_rollover *rollover = NULL;
1630 struct packet_sock *po = pkt_sk(sk);
1631 struct packet_fanout *f, *match;
1632 u8 type = type_flags & 0xff;
1633 u8 flags = type_flags >> 8;
1634 int err;
1636 switch (type) {
1637 case PACKET_FANOUT_ROLLOVER:
1638 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1639 return -EINVAL;
1640 case PACKET_FANOUT_HASH:
1641 case PACKET_FANOUT_LB:
1642 case PACKET_FANOUT_CPU:
1643 case PACKET_FANOUT_RND:
1644 case PACKET_FANOUT_QM:
1645 case PACKET_FANOUT_CBPF:
1646 case PACKET_FANOUT_EBPF:
1647 break;
1648 default:
1649 return -EINVAL;
1650 }
1652 mutex_lock(&fanout_mutex);
1654 err = -EALREADY;
1655 if (po->fanout)
1656 goto out;
1658 if (type == PACKET_FANOUT_ROLLOVER ||
1659 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1660 err = -ENOMEM;
1661 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1662 if (!rollover)
1663 goto out;
1664 atomic_long_set(&rollover->num, 0);
1665 atomic_long_set(&rollover->num_huge, 0);
1666 atomic_long_set(&rollover->num_failed, 0);
1667 }
1669 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1670 if (id != 0) {
1671 err = -EINVAL;
1672 goto out;
1673 }
1674 if (!fanout_find_new_id(sk, &id)) {
1675 err = -ENOMEM;
1676 goto out;
1677 }
1678 /* ephemeral flag for the first socket in the group: drop it */
1679 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1680 }
1682 match = NULL;
1683 list_for_each_entry(f, &fanout_list, list) {
1684 if (f->id == id &&
1685 read_pnet(&f->net) == sock_net(sk)) {
1686 match = f;
1687 break;
1688 }
1689 }
1690 err = -EINVAL;
1691 if (match && match->flags != flags)
1692 goto out;
1693 if (!match) {
1694 err = -ENOMEM;
1695 match = kzalloc(sizeof(*match), GFP_KERNEL);
1696 if (!match)
1697 goto out;
1698 write_pnet(&match->net, sock_net(sk));
1699 match->id = id;
1700 match->type = type;
1701 match->flags = flags;
1702 INIT_LIST_HEAD(&match->list);
1703 spin_lock_init(&match->lock);
1704 refcount_set(&match->sk_ref, 0);
1705 fanout_init_data(match);
1706 match->prot_hook.type = po->prot_hook.type;
1707 match->prot_hook.dev = po->prot_hook.dev;
1708 match->prot_hook.func = packet_rcv_fanout;
1709 match->prot_hook.af_packet_priv = match;
1710 match->prot_hook.id_match = match_fanout_group;
1711 list_add(&match->list, &fanout_list);
1712 }
1713 err = -EINVAL;
1715 spin_lock(&po->bind_lock);
1716 if (po->running &&
1717 match->type == type &&
1718 match->prot_hook.type == po->prot_hook.type &&
1719 match->prot_hook.dev == po->prot_hook.dev) {
1720 err = -ENOSPC;
1721 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1722 __dev_remove_pack(&po->prot_hook);
1723 po->fanout = match;
1724 po->rollover = rollover;
1725 rollover = NULL;
1726 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1727 __fanout_link(sk, po);
1728 err = 0;
1729 }
1730 }
1731 spin_unlock(&po->bind_lock);
1733 if (err && !refcount_read(&match->sk_ref)) {
1734 list_del(&match->list);
1735 kfree(match);
1736 }
1738 out:
1739 kfree(rollover);
1740 mutex_unlock(&fanout_mutex);
1741 return err;
1742 }
1744 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1745 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1746 * It is the responsibility of the caller to call fanout_release_data() and
1747 * free the returned packet_fanout (after synchronize_net())
1748 */
1749 static struct packet_fanout *fanout_release(struct sock *sk)
1750 {
1751 struct packet_sock *po = pkt_sk(sk);
1752 struct packet_fanout *f;
1754 mutex_lock(&fanout_mutex);
1755 f = po->fanout;
1756 if (f) {
1757 po->fanout = NULL;
1759 if (refcount_dec_and_test(&f->sk_ref))
1760 list_del(&f->list);
1761 else
1762 f = NULL;
1763 }
1764 mutex_unlock(&fanout_mutex);
1766 return f;
1767 }
1769 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1770 struct sk_buff *skb)
1771 {
1772 /* Earlier code assumed this would be a VLAN pkt, double-check
1773 * this now that we have the actual packet in hand. We can only
1774 * do this check on Ethernet devices.
1775 */
1776 if (unlikely(dev->type != ARPHRD_ETHER))
1777 return false;
1779 skb_reset_mac_header(skb);
1780 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1781 }
1783 static const struct proto_ops packet_ops;
1785 static const struct proto_ops packet_ops_spkt;
1787 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1788 struct packet_type *pt, struct net_device *orig_dev)
1789 {
1790 struct sock *sk;
1791 struct sockaddr_pkt *spkt;
1793 /*
1794 * When we registered the protocol we saved the socket in the data
1795 * field for just this event.
1796 */
1798 sk = pt->af_packet_priv;
1800 /*
1801 * Yank back the headers [hope the device set this
1802 * right or kerboom...]
1803 *
1804 * Incoming packets have ll header pulled,
1805 * push it back.
1806 *
1807 * For outgoing ones skb->data == skb_mac_header(skb)
1808 * so that this procedure is noop.
1809 */
1811 if (skb->pkt_type == PACKET_LOOPBACK)
1812 goto out;
1814 if (!net_eq(dev_net(dev), sock_net(sk)))
1815 goto out;
1817 skb = skb_share_check(skb, GFP_ATOMIC);
1818 if (skb == NULL)
1819 goto oom;
1821 /* drop any routing info */
1822 skb_dst_drop(skb);
1824 /* drop conntrack reference */
1825 nf_reset(skb);
1827 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1829 skb_push(skb, skb->data - skb_mac_header(skb));
1831 /*
1832 * The SOCK_PACKET socket receives _all_ frames.
1833 */
1835 spkt->spkt_family = dev->type;
1836 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1837 spkt->spkt_protocol = skb->protocol;
1839 /*
1840 * Charge the memory to the socket. This is done specifically
1841 * to prevent sockets using all the memory up.
1842 */
1844 if (sock_queue_rcv_skb(sk, skb) == 0)
1845 return 0;
1847 out:
1848 kfree_skb(skb);
1849 oom:
1850 return 0;
1851 }
1854 /*
1855 * Output a raw packet to a device layer. This bypasses all the other
1856 * protocol layers and you must therefore supply it with a complete frame
1857 */
1859 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1860 size_t len)
1861 {
1862 struct sock *sk = sock->sk;
1863 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1864 struct sk_buff *skb = NULL;
1865 struct net_device *dev;
1866 struct sockcm_cookie sockc;
1867 __be16 proto = 0;
1868 int err;
1869 int extra_len = 0;
1871 /*
1872 * Get and verify the address.
1873 */
1875 if (saddr) {
1876 if (msg->msg_namelen < sizeof(struct sockaddr))
1877 return -EINVAL;
1878 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1879 proto = saddr->spkt_protocol;
1880 } else
1881 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1883 /*
1884 * Find the device first to size check it
1885 */
1887 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1888 retry:
1889 rcu_read_lock();
1890 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1891 err = -ENODEV;
1892 if (dev == NULL)
1893 goto out_unlock;
1895 err = -ENETDOWN;
1896 if (!(dev->flags & IFF_UP))
1897 goto out_unlock;
1899 /*
1900 * You may not queue a frame bigger than the mtu. This is the lowest level
1901 * raw protocol and you must do your own fragmentation at this level.
1902 */
1904 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1905 if (!netif_supports_nofcs(dev)) {
1906 err = -EPROTONOSUPPORT;
1907 goto out_unlock;
1908 }
1909 extra_len = 4; /* We're doing our own CRC */
1910 }
1912 err = -EMSGSIZE;
1913 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1914 goto out_unlock;
1916 if (!skb) {
1917 size_t reserved = LL_RESERVED_SPACE(dev);
1918 int tlen = dev->needed_tailroom;
1919 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1921 rcu_read_unlock();
1922 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1923 if (skb == NULL)
1924 return -ENOBUFS;
1925 /* FIXME: Save some space for broken drivers that write a hard
1926 * header at transmission time by themselves. PPP is the notable
1927 * one here. This should really be fixed at the driver level.
1928 */
1929 skb_reserve(skb, reserved);
1930 skb_reset_network_header(skb);
1932 /* Try to align data part correctly */
1933 if (hhlen) {
1934 skb->data -= hhlen;
1935 skb->tail -= hhlen;
1936 if (len < hhlen)
1937 skb_reset_network_header(skb);
1938 }
1939 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1940 if (err)
1941 goto out_free;
1942 goto retry;
1943 }
1945 if (!dev_validate_header(dev, skb->data, len)) {
1946 err = -EINVAL;
1947 goto out_unlock;
1948 }
1949 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1950 !packet_extra_vlan_len_allowed(dev, skb)) {
1951 err = -EMSGSIZE;
1952 goto out_unlock;
1953 }
1955 sockcm_init(&sockc, sk);
1956 if (msg->msg_controllen) {
1957 err = sock_cmsg_send(sk, msg, &sockc);
1958 if (unlikely(err))
1959 goto out_unlock;
1960 }
1962 skb->protocol = proto;
1963 skb->dev = dev;
1964 skb->priority = sk->sk_priority;
1965 skb->mark = sk->sk_mark;
1966 skb->tstamp = sockc.transmit_time;
1968 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1970 if (unlikely(extra_len == 4))
1971 skb->no_fcs = 1;
1973 skb_probe_transport_header(skb, 0);
1975 dev_queue_xmit(skb);
1976 rcu_read_unlock();
1977 return len;
1979 out_unlock:
1980 rcu_read_unlock();
1981 out_free:
1982 kfree_skb(skb);
1983 return err;
1984 }
1986 static unsigned int run_filter(struct sk_buff *skb,
1987 const struct sock *sk,
1988 unsigned int res)
1989 {
1990 struct sk_filter *filter;
1992 rcu_read_lock();
1993 filter = rcu_dereference(sk->sk_filter);
1994 if (filter != NULL)
1995 res = bpf_prog_run_clear_cb(filter->prog, skb);
1996 rcu_read_unlock();
1998 return res;
1999 }
2001 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2002 size_t *len)
2003 {
2004 struct virtio_net_hdr vnet_hdr;
2006 if (*len < sizeof(vnet_hdr))
2007 return -EINVAL;
2008 *len -= sizeof(vnet_hdr);
2010 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2011 return -EINVAL;
2013 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2014 }
2016 /*
2017 * This function makes lazy skb cloning in hope that most of packets
2018 * are discarded by BPF.
2019 *
2020 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2021 * and skb->cb are mangled. It works because (and until) packets
2022 * falling here are owned by current CPU. Output packets are cloned
2023 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2024 * sequencially, so that if we return skb to original state on exit,
2025 * we will not harm anyone.
2026 */
2028 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2029 struct packet_type *pt, struct net_device *orig_dev)
2030 {
2031 struct sock *sk;
2032 struct sockaddr_ll *sll;
2033 struct packet_sock *po;
2034 u8 *skb_head = skb->data;
2035 int skb_len = skb->len;
2036 unsigned int snaplen, res;
2037 bool is_drop_n_account = false;
2039 if (skb->pkt_type == PACKET_LOOPBACK)
2040 goto drop;
2042 sk = pt->af_packet_priv;
2043 po = pkt_sk(sk);
2045 if (!net_eq(dev_net(dev), sock_net(sk)))
2046 goto drop;
2048 skb->dev = dev;
2050 if (dev->header_ops) {
2051 /* The device has an explicit notion of ll header,
2052 * exported to higher levels.
2053 *
2054 * Otherwise, the device hides details of its frame
2055 * structure, so that corresponding packet head is
2056 * never delivered to user.
2057 */
2058 if (sk->sk_type != SOCK_DGRAM)
2059 skb_push(skb, skb->data - skb_mac_header(skb));
2060 else if (skb->pkt_type == PACKET_OUTGOING) {
2061 /* Special case: outgoing packets have ll header at head */
2062 skb_pull(skb, skb_network_offset(skb));
2063 }
2064 }
2066 snaplen = skb->len;
2068 res = run_filter(skb, sk, snaplen);
2069 if (!res)
2070 goto drop_n_restore;
2071 if (snaplen > res)
2072 snaplen = res;
2074 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2075 goto drop_n_acct;
2077 if (skb_shared(skb)) {
2078 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2079 if (nskb == NULL)
2080 goto drop_n_acct;
2082 if (skb_head != skb->data) {
2083 skb->data = skb_head;
2084 skb->len = skb_len;
2085 }
2086 consume_skb(skb);
2087 skb = nskb;
2088 }
2090 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2092 sll = &PACKET_SKB_CB(skb)->sa.ll;
2093 sll->sll_hatype = dev->type;
2094 sll->sll_pkttype = skb->pkt_type;
2095 if (unlikely(po->origdev))
2096 sll->sll_ifindex = orig_dev->ifindex;
2097 else
2098 sll->sll_ifindex = dev->ifindex;
2100 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2102 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2103 * Use their space for storing the original skb length.
2104 */
2105 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2107 if (pskb_trim(skb, snaplen))
2108 goto drop_n_acct;
2110 skb_set_owner_r(skb, sk);
2111 skb->dev = NULL;
2112 skb_dst_drop(skb);
2114 /* drop conntrack reference */
2115 nf_reset(skb);
2117 spin_lock(&sk->sk_receive_queue.lock);
2118 po->stats.stats1.tp_packets++;
2119 sock_skb_set_dropcount(sk, skb);
2120 __skb_queue_tail(&sk->sk_receive_queue, skb);
2121 spin_unlock(&sk->sk_receive_queue.lock);
2122 sk->sk_data_ready(sk);
2123 return 0;
2125 drop_n_acct:
2126 is_drop_n_account = true;
2127 spin_lock(&sk->sk_receive_queue.lock);
2128 po->stats.stats1.tp_drops++;
2129 atomic_inc(&sk->sk_drops);
2130 spin_unlock(&sk->sk_receive_queue.lock);
2132 drop_n_restore:
2133 if (skb_head != skb->data && skb_shared(skb)) {
2134 skb->data = skb_head;
2135 skb->len = skb_len;
2136 }
2137 drop:
2138 if (!is_drop_n_account)
2139 consume_skb(skb);
2140 else
2141 kfree_skb(skb);
2142 return 0;
2143 }
2145 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2146 struct packet_type *pt, struct net_device *orig_dev)
2147 {
2148 struct sock *sk;
2149 struct packet_sock *po;
2150 struct sockaddr_ll *sll;
2151 union tpacket_uhdr h;
2152 u8 *skb_head = skb->data;
2153 int skb_len = skb->len;
2154 unsigned int snaplen, res;
2155 unsigned long status = TP_STATUS_USER;
2156 unsigned short macoff, netoff, hdrlen;
2157 struct sk_buff *copy_skb = NULL;
2158 struct timespec ts;
2159 __u32 ts_status;
2160 bool is_drop_n_account = false;
2161 bool do_vnet = false;
2163 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2164 * We may add members to them until current aligned size without forcing
2165 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2166 */
2167 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2168 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2170 if (skb->pkt_type == PACKET_LOOPBACK)
2171 goto drop;
2173 sk = pt->af_packet_priv;
2174 po = pkt_sk(sk);
2176 if (!net_eq(dev_net(dev), sock_net(sk)))
2177 goto drop;
2179 if (dev->header_ops) {
2180 if (sk->sk_type != SOCK_DGRAM)
2181 skb_push(skb, skb->data - skb_mac_header(skb));
2182 else if (skb->pkt_type == PACKET_OUTGOING) {
2183 /* Special case: outgoing packets have ll header at head */
2184 skb_pull(skb, skb_network_offset(skb));
2185 }
2186 }
2188 snaplen = skb->len;
2190 res = run_filter(skb, sk, snaplen);
2191 if (!res)
2192 goto drop_n_restore;
2194 if (skb->ip_summed == CHECKSUM_PARTIAL)
2195 status |= TP_STATUS_CSUMNOTREADY;
2196 else if (skb->pkt_type != PACKET_OUTGOING &&
2197 (skb->ip_summed == CHECKSUM_COMPLETE ||
2198 skb_csum_unnecessary(skb)))
2199 status |= TP_STATUS_CSUM_VALID;
2201 if (snaplen > res)
2202 snaplen = res;
2204 if (sk->sk_type == SOCK_DGRAM) {
2205 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2206 po->tp_reserve;
2207 } else {
2208 unsigned int maclen = skb_network_offset(skb);
2209 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2210 (maclen < 16 ? 16 : maclen)) +
2211 po->tp_reserve;
2212 if (po->has_vnet_hdr) {
2213 netoff += sizeof(struct virtio_net_hdr);
2214 do_vnet = true;
2215 }
2216 macoff = netoff - maclen;
2217 }
2218 if (po->tp_version <= TPACKET_V2) {
2219 if (macoff + snaplen > po->rx_ring.frame_size) {
2220 if (po->copy_thresh &&
2221 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2222 if (skb_shared(skb)) {
2223 copy_skb = skb_clone(skb, GFP_ATOMIC);
2224 } else {
2225 copy_skb = skb_get(skb);
2226 skb_head = skb->data;
2227 }
2228 if (copy_skb)
2229 skb_set_owner_r(copy_skb, sk);
2230 }
2231 snaplen = po->rx_ring.frame_size - macoff;
2232 if ((int)snaplen < 0) {
2233 snaplen = 0;
2234 do_vnet = false;
2235 }
2236 }
2237 } else if (unlikely(macoff + snaplen >
2238 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2239 u32 nval;
2241 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2242 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2243 snaplen, nval, macoff);
2244 snaplen = nval;
2245 if (unlikely((int)snaplen < 0)) {
2246 snaplen = 0;
2247 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2248 do_vnet = false;
2249 }
2250 }
2251 spin_lock(&sk->sk_receive_queue.lock);
2252 h.raw = packet_current_rx_frame(po, skb,
2253 TP_STATUS_KERNEL, (macoff+snaplen));
2254 if (!h.raw)
2255 goto drop_n_account;
2256 if (po->tp_version <= TPACKET_V2) {
2257 packet_increment_rx_head(po, &po->rx_ring);
2258 /*
2259 * LOSING will be reported till you read the stats,
2260 * because it's COR - Clear On Read.
2261 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2262 * at packet level.
2263 */
2264 if (po->stats.stats1.tp_drops)
2265 status |= TP_STATUS_LOSING;
2266 }
2268 if (do_vnet &&
2269 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2270 sizeof(struct virtio_net_hdr),
2271 vio_le(), true, 0))
2272 goto drop_n_account;
2274 po->stats.stats1.tp_packets++;
2275 if (copy_skb) {
2276 status |= TP_STATUS_COPY;
2277 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2278 }
2279 spin_unlock(&sk->sk_receive_queue.lock);
2281 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2283 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2284 getnstimeofday(&ts);
2286 status |= ts_status;
2288 switch (po->tp_version) {
2289 case TPACKET_V1:
2290 h.h1->tp_len = skb->len;
2291 h.h1->tp_snaplen = snaplen;
2292 h.h1->tp_mac = macoff;
2293 h.h1->tp_net = netoff;
2294 h.h1->tp_sec = ts.tv_sec;
2295 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2296 hdrlen = sizeof(*h.h1);
2297 break;
2298 case TPACKET_V2:
2299 h.h2->tp_len = skb->len;
2300 h.h2->tp_snaplen = snaplen;
2301 h.h2->tp_mac = macoff;
2302 h.h2->tp_net = netoff;
2303 h.h2->tp_sec = ts.tv_sec;
2304 h.h2->tp_nsec = ts.tv_nsec;
2305 if (skb_vlan_tag_present(skb)) {
2306 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2307 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2308 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2309 } else {
2310 h.h2->tp_vlan_tci = 0;
2311 h.h2->tp_vlan_tpid = 0;
2312 }
2313 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2314 hdrlen = sizeof(*h.h2);
2315 break;
2316 case TPACKET_V3:
2317 /* tp_nxt_offset,vlan are already populated above.
2318 * So DONT clear those fields here
2319 */
2320 h.h3->tp_status |= status;
2321 h.h3->tp_len = skb->len;
2322 h.h3->tp_snaplen = snaplen;
2323 h.h3->tp_mac = macoff;
2324 h.h3->tp_net = netoff;
2325 h.h3->tp_sec = ts.tv_sec;
2326 h.h3->tp_nsec = ts.tv_nsec;
2327 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2328 hdrlen = sizeof(*h.h3);
2329 break;
2330 default:
2331 BUG();
2332 }
2334 sll = h.raw + TPACKET_ALIGN(hdrlen);
2335 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2336 sll->sll_family = AF_PACKET;
2337 sll->sll_hatype = dev->type;
2338 sll->sll_protocol = skb->protocol;
2339 sll->sll_pkttype = skb->pkt_type;
2340 if (unlikely(po->origdev))
2341 sll->sll_ifindex = orig_dev->ifindex;
2342 else
2343 sll->sll_ifindex = dev->ifindex;
2345 smp_mb();
2347 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2348 if (po->tp_version <= TPACKET_V2) {
2349 u8 *start, *end;
2351 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2352 macoff + snaplen);
2354 for (start = h.raw; start < end; start += PAGE_SIZE)
2355 flush_dcache_page(pgv_to_page(start));
2356 }
2357 smp_wmb();
2358 #endif
2360 if (po->tp_version <= TPACKET_V2) {
2361 __packet_set_status(po, h.raw, status);
2362 sk->sk_data_ready(sk);
2363 } else {
2364 prb_clear_blk_fill_status(&po->rx_ring);
2365 }
2367 drop_n_restore:
2368 if (skb_head != skb->data && skb_shared(skb)) {
2369 skb->data = skb_head;
2370 skb->len = skb_len;
2371 }
2372 drop:
2373 if (!is_drop_n_account)
2374 consume_skb(skb);
2375 else
2376 kfree_skb(skb);
2377 return 0;
2379 drop_n_account:
2380 is_drop_n_account = true;
2381 po->stats.stats1.tp_drops++;
2382 spin_unlock(&sk->sk_receive_queue.lock);
2384 sk->sk_data_ready(sk);
2385 kfree_skb(copy_skb);
2386 goto drop_n_restore;
2387 }
2389 static void tpacket_destruct_skb(struct sk_buff *skb)
2390 {
2391 struct packet_sock *po = pkt_sk(skb->sk);
2393 if (likely(po->tx_ring.pg_vec)) {
2394 void *ph;
2395 __u32 ts;
2397 ph = skb_shinfo(skb)->destructor_arg;
2398 packet_dec_pending(&po->tx_ring);
2400 ts = __packet_set_timestamp(po, ph, skb);
2401 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2402 }
2404 sock_wfree(skb);
2405 }
2407 static void tpacket_set_protocol(const struct net_device *dev,
2408 struct sk_buff *skb)
2409 {
2410 if (dev->type == ARPHRD_ETHER) {
2411 skb_reset_mac_header(skb);
2412 skb->protocol = eth_hdr(skb)->h_proto;
2413 }
2414 }
2416 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2417 {
2418 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2419 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2420 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2421 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2422 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2423 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2424 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2426 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2427 return -EINVAL;
2429 return 0;
2430 }
2432 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2433 struct virtio_net_hdr *vnet_hdr)
2434 {
2435 if (*len < sizeof(*vnet_hdr))
2436 return -EINVAL;
2437 *len -= sizeof(*vnet_hdr);
2439 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2440 return -EFAULT;
2442 return __packet_snd_vnet_parse(vnet_hdr, *len);
2443 }
2445 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2446 void *frame, struct net_device *dev, void *data, int tp_len,
2447 __be16 proto, unsigned char *addr, int hlen, int copylen,
2448 const struct sockcm_cookie *sockc)
2449 {
2450 union tpacket_uhdr ph;
2451 int to_write, offset, len, nr_frags, len_max;
2452 struct socket *sock = po->sk.sk_socket;
2453 struct page *page;
2454 int err;
2456 ph.raw = frame;
2458 skb->protocol = proto;
2459 skb->dev = dev;
2460 skb->priority = po->sk.sk_priority;
2461 skb->mark = po->sk.sk_mark;
2462 skb->tstamp = sockc->transmit_time;
2463 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2464 skb_shinfo(skb)->destructor_arg = ph.raw;
2466 skb_reserve(skb, hlen);
2467 skb_reset_network_header(skb);
2469 to_write = tp_len;
2471 if (sock->type == SOCK_DGRAM) {
2472 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2473 NULL, tp_len);
2474 if (unlikely(err < 0))
2475 return -EINVAL;
2476 } else if (copylen) {
2477 int hdrlen = min_t(int, copylen, tp_len);
2479 skb_push(skb, dev->hard_header_len);
2480 skb_put(skb, copylen - dev->hard_header_len);
2481 err = skb_store_bits(skb, 0, data, hdrlen);
2482 if (unlikely(err))
2483 return err;
2484 if (!dev_validate_header(dev, skb->data, hdrlen))
2485 return -EINVAL;
2486 if (!skb->protocol)
2487 tpacket_set_protocol(dev, skb);
2489 data += hdrlen;
2490 to_write -= hdrlen;
2491 }
2493 offset = offset_in_page(data);
2494 len_max = PAGE_SIZE - offset;
2495 len = ((to_write > len_max) ? len_max : to_write);
2497 skb->data_len = to_write;
2498 skb->len += to_write;
2499 skb->truesize += to_write;
2500 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2502 while (likely(to_write)) {
2503 nr_frags = skb_shinfo(skb)->nr_frags;
2505 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2506 pr_err("Packet exceed the number of skb frags(%lu)\n",
2507 MAX_SKB_FRAGS);
2508 return -EFAULT;
2509 }
2511 page = pgv_to_page(data);
2512 data += len;
2513 flush_dcache_page(page);
2514 get_page(page);
2515 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2516 to_write -= len;
2517 offset = 0;
2518 len_max = PAGE_SIZE;
2519 len = ((to_write > len_max) ? len_max : to_write);
2520 }
2522 skb_probe_transport_header(skb, 0);
2524 return tp_len;
2525 }
2527 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2528 int size_max, void **data)
2529 {
2530 union tpacket_uhdr ph;
2531 int tp_len, off;
2533 ph.raw = frame;
2535 switch (po->tp_version) {
2536 case TPACKET_V3:
2537 if (ph.h3->tp_next_offset != 0) {
2538 pr_warn_once("variable sized slot not supported");
2539 return -EINVAL;
2540 }
2541 tp_len = ph.h3->tp_len;
2542 break;
2543 case TPACKET_V2:
2544 tp_len = ph.h2->tp_len;
2545 break;
2546 default:
2547 tp_len = ph.h1->tp_len;
2548 break;
2549 }
2550 if (unlikely(tp_len > size_max)) {
2551 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2552 return -EMSGSIZE;
2553 }
2555 if (unlikely(po->tp_tx_has_off)) {
2556 int off_min, off_max;
2558 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2559 off_max = po->tx_ring.frame_size - tp_len;
2560 if (po->sk.sk_type == SOCK_DGRAM) {
2561 switch (po->tp_version) {
2562 case TPACKET_V3:
2563 off = ph.h3->tp_net;
2564 break;
2565 case TPACKET_V2:
2566 off = ph.h2->tp_net;
2567 break;
2568 default:
2569 off = ph.h1->tp_net;
2570 break;
2571 }
2572 } else {
2573 switch (po->tp_version) {
2574 case TPACKET_V3:
2575 off = ph.h3->tp_mac;
2576 break;
2577 case TPACKET_V2:
2578 off = ph.h2->tp_mac;
2579 break;
2580 default:
2581 off = ph.h1->tp_mac;
2582 break;
2583 }
2584 }
2585 if (unlikely((off < off_min) || (off_max < off)))
2586 return -EINVAL;
2587 } else {
2588 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2589 }
2591 *data = frame + off;
2592 return tp_len;
2593 }
2595 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2596 {
2597 struct sk_buff *skb;
2598 struct net_device *dev;
2599 struct virtio_net_hdr *vnet_hdr = NULL;
2600 struct sockcm_cookie sockc;
2601 __be16 proto;
2602 int err, reserve = 0;
2603 void *ph;
2604 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2605 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2606 int tp_len, size_max;
2607 unsigned char *addr;
2608 void *data;
2609 int len_sum = 0;
2610 int status = TP_STATUS_AVAILABLE;
2611 int hlen, tlen, copylen = 0;
2613 mutex_lock(&po->pg_vec_lock);
2615 if (likely(saddr == NULL)) {
2616 dev = packet_cached_dev_get(po);
2617 proto = po->num;
2618 addr = NULL;
2619 } else {
2620 err = -EINVAL;
2621 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2622 goto out;
2623 if (msg->msg_namelen < (saddr->sll_halen
2624 + offsetof(struct sockaddr_ll,
2625 sll_addr)))
2626 goto out;
2627 proto = saddr->sll_protocol;
2628 addr = saddr->sll_addr;
2629 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2630 }
2632 err = -ENXIO;
2633 if (unlikely(dev == NULL))
2634 goto out;
2635 err = -ENETDOWN;
2636 if (unlikely(!(dev->flags & IFF_UP)))
2637 goto out_put;
2639 sockcm_init(&sockc, &po->sk);
2640 if (msg->msg_controllen) {
2641 err = sock_cmsg_send(&po->sk, msg, &sockc);
2642 if (unlikely(err))
2643 goto out_put;
2644 }
2646 if (po->sk.sk_socket->type == SOCK_RAW)
2647 reserve = dev->hard_header_len;
2648 size_max = po->tx_ring.frame_size
2649 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2651 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2652 size_max = dev->mtu + reserve + VLAN_HLEN;
2654 do {
2655 ph = packet_current_frame(po, &po->tx_ring,
2656 TP_STATUS_SEND_REQUEST);
2657 if (unlikely(ph == NULL)) {
2658 if (need_wait && need_resched())
2659 schedule();
2660 continue;
2661 }
2663 skb = NULL;
2664 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2665 if (tp_len < 0)
2666 goto tpacket_error;
2668 status = TP_STATUS_SEND_REQUEST;
2669 hlen = LL_RESERVED_SPACE(dev);
2670 tlen = dev->needed_tailroom;
2671 if (po->has_vnet_hdr) {
2672 vnet_hdr = data;
2673 data += sizeof(*vnet_hdr);
2674 tp_len -= sizeof(*vnet_hdr);
2675 if (tp_len < 0 ||
2676 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2677 tp_len = -EINVAL;
2678 goto tpacket_error;
2679 }
2680 copylen = __virtio16_to_cpu(vio_le(),
2681 vnet_hdr->hdr_len);
2682 }
2683 copylen = max_t(int, copylen, dev->hard_header_len);
2684 skb = sock_alloc_send_skb(&po->sk,
2685 hlen + tlen + sizeof(struct sockaddr_ll) +
2686 (copylen - dev->hard_header_len),
2687 !need_wait, &err);
2689 if (unlikely(skb == NULL)) {
2690 /* we assume the socket was initially writeable ... */
2691 if (likely(len_sum > 0))
2692 err = len_sum;
2693 goto out_status;
2694 }
2695 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2696 addr, hlen, copylen, &sockc);
2697 if (likely(tp_len >= 0) &&
2698 tp_len > dev->mtu + reserve &&
2699 !po->has_vnet_hdr &&
2700 !packet_extra_vlan_len_allowed(dev, skb))
2701 tp_len = -EMSGSIZE;
2703 if (unlikely(tp_len < 0)) {
2704 tpacket_error:
2705 if (po->tp_loss) {
2706 __packet_set_status(po, ph,
2707 TP_STATUS_AVAILABLE);
2708 packet_increment_head(&po->tx_ring);
2709 kfree_skb(skb);
2710 continue;
2711 } else {
2712 status = TP_STATUS_WRONG_FORMAT;
2713 err = tp_len;
2714 goto out_status;
2715 }
2716 }
2718 if (po->has_vnet_hdr) {
2719 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2720 tp_len = -EINVAL;
2721 goto tpacket_error;
2722 }
2723 virtio_net_hdr_set_proto(skb, vnet_hdr);
2724 }
2726 skb->destructor = tpacket_destruct_skb;
2727 __packet_set_status(po, ph, TP_STATUS_SENDING);
2728 packet_inc_pending(&po->tx_ring);
2730 status = TP_STATUS_SEND_REQUEST;
2731 err = po->xmit(skb);
2732 if (unlikely(err > 0)) {
2733 err = net_xmit_errno(err);
2734 if (err && __packet_get_status(po, ph) ==
2735 TP_STATUS_AVAILABLE) {
2736 /* skb was destructed already */
2737 skb = NULL;
2738 goto out_status;
2739 }
2740 /*
2741 * skb was dropped but not destructed yet;
2742 * let's treat it like congestion or err < 0
2743 */
2744 err = 0;
2745 }
2746 packet_increment_head(&po->tx_ring);
2747 len_sum += tp_len;
2748 } while (likely((ph != NULL) ||
2749 /* Note: packet_read_pending() might be slow if we have
2750 * to call it as it's per_cpu variable, but in fast-path
2751 * we already short-circuit the loop with the first
2752 * condition, and luckily don't have to go that path
2753 * anyway.
2754 */
2755 (need_wait && packet_read_pending(&po->tx_ring))));
2757 err = len_sum;
2758 goto out_put;
2760 out_status:
2761 __packet_set_status(po, ph, status);
2762 kfree_skb(skb);
2763 out_put:
2764 dev_put(dev);
2765 out:
2766 mutex_unlock(&po->pg_vec_lock);
2767 return err;
2768 }
2770 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2771 size_t reserve, size_t len,
2772 size_t linear, int noblock,
2773 int *err)
2774 {
2775 struct sk_buff *skb;
2777 /* Under a page? Don't bother with paged skb. */
2778 if (prepad + len < PAGE_SIZE || !linear)
2779 linear = len;
2781 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2782 err, 0);
2783 if (!skb)
2784 return NULL;
2786 skb_reserve(skb, reserve);
2787 skb_put(skb, linear);
2788 skb->data_len = len - linear;
2789 skb->len += len - linear;
2791 return skb;
2792 }
2794 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2795 {
2796 struct sock *sk = sock->sk;
2797 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2798 struct sk_buff *skb;
2799 struct net_device *dev;
2800 __be16 proto;
2801 unsigned char *addr;
2802 int err, reserve = 0;
2803 struct sockcm_cookie sockc;
2804 struct virtio_net_hdr vnet_hdr = { 0 };
2805 int offset = 0;
2806 struct packet_sock *po = pkt_sk(sk);
2807 bool has_vnet_hdr = false;
2808 int hlen, tlen, linear;
2809 int extra_len = 0;
2811 /*
2812 * Get and verify the address.
2813 */
2815 if (likely(saddr == NULL)) {
2816 dev = packet_cached_dev_get(po);
2817 proto = po->num;
2818 addr = NULL;
2819 } else {
2820 err = -EINVAL;
2821 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2822 goto out;
2823 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2824 goto out;
2825 proto = saddr->sll_protocol;
2826 addr = saddr->sll_addr;
2827 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2828 }
2830 err = -ENXIO;
2831 if (unlikely(dev == NULL))
2832 goto out_unlock;
2833 err = -ENETDOWN;
2834 if (unlikely(!(dev->flags & IFF_UP)))
2835 goto out_unlock;
2837 sockcm_init(&sockc, sk);
2838 sockc.mark = sk->sk_mark;
2839 if (msg->msg_controllen) {
2840 err = sock_cmsg_send(sk, msg, &sockc);
2841 if (unlikely(err))
2842 goto out_unlock;
2843 }
2845 if (sock->type == SOCK_RAW)
2846 reserve = dev->hard_header_len;
2847 if (po->has_vnet_hdr) {
2848 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2849 if (err)
2850 goto out_unlock;
2851 has_vnet_hdr = true;
2852 }
2854 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2855 if (!netif_supports_nofcs(dev)) {
2856 err = -EPROTONOSUPPORT;
2857 goto out_unlock;
2858 }
2859 extra_len = 4; /* We're doing our own CRC */
2860 }
2862 err = -EMSGSIZE;
2863 if (!vnet_hdr.gso_type &&
2864 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2865 goto out_unlock;
2867 err = -ENOBUFS;
2868 hlen = LL_RESERVED_SPACE(dev);
2869 tlen = dev->needed_tailroom;
2870 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2871 linear = max(linear, min_t(int, len, dev->hard_header_len));
2872 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2873 msg->msg_flags & MSG_DONTWAIT, &err);
2874 if (skb == NULL)
2875 goto out_unlock;
2877 skb_reset_network_header(skb);
2879 err = -EINVAL;
2880 if (sock->type == SOCK_DGRAM) {
2881 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2882 if (unlikely(offset < 0))
2883 goto out_free;
2884 } else if (reserve) {
2885 skb_reserve(skb, -reserve);
2886 if (len < reserve)
2887 skb_reset_network_header(skb);
2888 }
2890 /* Returns -EFAULT on error */
2891 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2892 if (err)
2893 goto out_free;
2895 if (sock->type == SOCK_RAW &&
2896 !dev_validate_header(dev, skb->data, len)) {
2897 err = -EINVAL;
2898 goto out_free;
2899 }
2901 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2903 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2904 !packet_extra_vlan_len_allowed(dev, skb)) {
2905 err = -EMSGSIZE;
2906 goto out_free;
2907 }
2909 skb->protocol = proto;
2910 skb->dev = dev;
2911 skb->priority = sk->sk_priority;
2912 skb->mark = sockc.mark;
2913 skb->tstamp = sockc.transmit_time;
2915 if (has_vnet_hdr) {
2916 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2917 if (err)
2918 goto out_free;
2919 len += sizeof(vnet_hdr);
2920 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2921 }
2923 skb_probe_transport_header(skb, reserve);
2925 if (unlikely(extra_len == 4))
2926 skb->no_fcs = 1;
2928 err = po->xmit(skb);
2929 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2930 goto out_unlock;
2932 dev_put(dev);
2934 return len;
2936 out_free:
2937 kfree_skb(skb);
2938 out_unlock:
2939 if (dev)
2940 dev_put(dev);
2941 out:
2942 return err;
2943 }
2945 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2946 {
2947 struct sock *sk = sock->sk;
2948 struct packet_sock *po = pkt_sk(sk);
2950 if (po->tx_ring.pg_vec)
2951 return tpacket_snd(po, msg);
2952 else
2953 return packet_snd(sock, msg, len);
2954 }
2956 /*
2957 * Close a PACKET socket. This is fairly simple. We immediately go
2958 * to 'closed' state and remove our protocol entry in the device list.
2959 */
2961 static int packet_release(struct socket *sock)
2962 {
2963 struct sock *sk = sock->sk;
2964 struct packet_sock *po;
2965 struct packet_fanout *f;
2966 struct net *net;
2967 union tpacket_req_u req_u;
2969 if (!sk)
2970 return 0;
2972 net = sock_net(sk);
2973 po = pkt_sk(sk);
2975 mutex_lock(&net->packet.sklist_lock);
2976 sk_del_node_init_rcu(sk);
2977 mutex_unlock(&net->packet.sklist_lock);
2979 preempt_disable();
2980 sock_prot_inuse_add(net, sk->sk_prot, -1);
2981 preempt_enable();
2983 spin_lock(&po->bind_lock);
2984 unregister_prot_hook(sk, false);
2985 packet_cached_dev_reset(po);
2987 if (po->prot_hook.dev) {
2988 dev_put(po->prot_hook.dev);
2989 po->prot_hook.dev = NULL;
2990 }
2991 spin_unlock(&po->bind_lock);
2993 packet_flush_mclist(sk);
2995 lock_sock(sk);
2996 if (po->rx_ring.pg_vec) {
2997 memset(&req_u, 0, sizeof(req_u));
2998 packet_set_ring(sk, &req_u, 1, 0);
2999 }
3001 if (po->tx_ring.pg_vec) {
3002 memset(&req_u, 0, sizeof(req_u));
3003 packet_set_ring(sk, &req_u, 1, 1);
3004 }
3005 release_sock(sk);
3007 f = fanout_release(sk);
3009 synchronize_net();
3011 if (f) {
3012 kfree(po->rollover);
3013 fanout_release_data(f);
3014 kfree(f);
3015 }
3016 /*
3017 * Now the socket is dead. No more input will appear.
3018 */
3019 sock_orphan(sk);
3020 sock->sk = NULL;
3022 /* Purge queues */
3024 skb_queue_purge(&sk->sk_receive_queue);
3025 packet_free_pending(po);
3026 sk_refcnt_debug_release(sk);
3028 sock_put(sk);
3029 return 0;
3030 }
3032 /*
3033 * Attach a packet hook.
3034 */
3036 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3037 __be16 proto)
3038 {
3039 struct packet_sock *po = pkt_sk(sk);
3040 struct net_device *dev_curr;
3041 __be16 proto_curr;
3042 bool need_rehook;
3043 struct net_device *dev = NULL;
3044 int ret = 0;
3045 bool unlisted = false;
3047 lock_sock(sk);
3048 spin_lock(&po->bind_lock);
3049 rcu_read_lock();
3051 if (po->fanout) {
3052 ret = -EINVAL;
3053 goto out_unlock;
3054 }
3056 if (name) {
3057 dev = dev_get_by_name_rcu(sock_net(sk), name);
3058 if (!dev) {
3059 ret = -ENODEV;
3060 goto out_unlock;
3061 }
3062 } else if (ifindex) {
3063 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3064 if (!dev) {
3065 ret = -ENODEV;
3066 goto out_unlock;
3067 }
3068 }
3070 if (dev)
3071 dev_hold(dev);
3073 proto_curr = po->prot_hook.type;
3074 dev_curr = po->prot_hook.dev;
3076 need_rehook = proto_curr != proto || dev_curr != dev;
3078 if (need_rehook) {
3079 if (po->running) {
3080 rcu_read_unlock();
3081 /* prevents packet_notifier() from calling
3082 * register_prot_hook()
3083 */
3084 po->num = 0;
3085 __unregister_prot_hook(sk, true);
3086 rcu_read_lock();
3087 dev_curr = po->prot_hook.dev;
3088 if (dev)
3089 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3090 dev->ifindex);
3091 }
3093 BUG_ON(po->running);
3094 po->num = proto;
3095 po->prot_hook.type = proto;
3097 if (unlikely(unlisted)) {
3098 dev_put(dev);
3099 po->prot_hook.dev = NULL;
3100 po->ifindex = -1;
3101 packet_cached_dev_reset(po);
3102 } else {
3103 po->prot_hook.dev = dev;
3104 po->ifindex = dev ? dev->ifindex : 0;
3105 packet_cached_dev_assign(po, dev);
3106 }
3107 }
3108 if (dev_curr)
3109 dev_put(dev_curr);
3111 if (proto == 0 || !need_rehook)
3112 goto out_unlock;
3114 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3115 register_prot_hook(sk);
3116 } else {
3117 sk->sk_err = ENETDOWN;
3118 if (!sock_flag(sk, SOCK_DEAD))
3119 sk->sk_error_report(sk);
3120 }
3122 out_unlock:
3123 rcu_read_unlock();
3124 spin_unlock(&po->bind_lock);
3125 release_sock(sk);
3126 return ret;
3127 }
3129 /*
3130 * Bind a packet socket to a device
3131 */
3133 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3134 int addr_len)
3135 {
3136 struct sock *sk = sock->sk;
3137 char name[sizeof(uaddr->sa_data) + 1];
3139 /*
3140 * Check legality
3141 */
3143 if (addr_len != sizeof(struct sockaddr))
3144 return -EINVAL;
3145 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3146 * zero-terminated.
3147 */
3148 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3149 name[sizeof(uaddr->sa_data)] = 0;
3151 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3152 }
3154 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3155 {
3156 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3157 struct sock *sk = sock->sk;
3159 /*
3160 * Check legality
3161 */
3163 if (addr_len < sizeof(struct sockaddr_ll))
3164 return -EINVAL;
3165 if (sll->sll_family != AF_PACKET)
3166 return -EINVAL;
3168 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3169 sll->sll_protocol ? : pkt_sk(sk)->num);
3170 }
3172 static struct proto packet_proto = {
3173 .name = "PACKET",
3174 .owner = THIS_MODULE,
3175 .obj_size = sizeof(struct packet_sock),
3176 };
3178 /*
3179 * Create a packet of type SOCK_PACKET.
3180 */
3182 static int packet_create(struct net *net, struct socket *sock, int protocol,
3183 int kern)
3184 {
3185 struct sock *sk;
3186 struct packet_sock *po;
3187 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3188 int err;
3190 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3191 return -EPERM;
3192 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3193 sock->type != SOCK_PACKET)
3194 return -ESOCKTNOSUPPORT;
3196 sock->state = SS_UNCONNECTED;
3198 err = -ENOBUFS;
3199 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3200 if (sk == NULL)
3201 goto out;
3203 sock->ops = &packet_ops;
3204 if (sock->type == SOCK_PACKET)
3205 sock->ops = &packet_ops_spkt;
3207 sock_init_data(sock, sk);
3209 po = pkt_sk(sk);
3210 sk->sk_family = PF_PACKET;
3211 po->num = proto;
3212 po->xmit = dev_queue_xmit;
3214 err = packet_alloc_pending(po);
3215 if (err)
3216 goto out2;
3218 packet_cached_dev_reset(po);
3220 sk->sk_destruct = packet_sock_destruct;
3221 sk_refcnt_debug_inc(sk);
3223 /*
3224 * Attach a protocol block
3225 */
3227 spin_lock_init(&po->bind_lock);
3228 mutex_init(&po->pg_vec_lock);
3229 po->rollover = NULL;
3230 po->prot_hook.func = packet_rcv;
3232 if (sock->type == SOCK_PACKET)
3233 po->prot_hook.func = packet_rcv_spkt;
3235 po->prot_hook.af_packet_priv = sk;
3237 if (proto) {
3238 po->prot_hook.type = proto;
3239 __register_prot_hook(sk);
3240 }
3242 mutex_lock(&net->packet.sklist_lock);
3243 sk_add_node_rcu(sk, &net->packet.sklist);
3244 mutex_unlock(&net->packet.sklist_lock);
3246 preempt_disable();
3247 sock_prot_inuse_add(net, &packet_proto, 1);
3248 preempt_enable();
3250 return 0;
3251 out2:
3252 sk_free(sk);
3253 out:
3254 return err;
3255 }
3257 /*
3258 * Pull a packet from our receive queue and hand it to the user.
3259 * If necessary we block.
3260 */
3262 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3263 int flags)
3264 {
3265 struct sock *sk = sock->sk;
3266 struct sk_buff *skb;
3267 int copied, err;
3268 int vnet_hdr_len = 0;
3269 unsigned int origlen = 0;
3271 err = -EINVAL;
3272 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3273 goto out;
3275 #if 0
3276 /* What error should we return now? EUNATTACH? */
3277 if (pkt_sk(sk)->ifindex < 0)
3278 return -ENODEV;
3279 #endif
3281 if (flags & MSG_ERRQUEUE) {
3282 err = sock_recv_errqueue(sk, msg, len,
3283 SOL_PACKET, PACKET_TX_TIMESTAMP);
3284 goto out;
3285 }
3287 /*
3288 * Call the generic datagram receiver. This handles all sorts
3289 * of horrible races and re-entrancy so we can forget about it
3290 * in the protocol layers.
3291 *
3292 * Now it will return ENETDOWN, if device have just gone down,
3293 * but then it will block.
3294 */
3296 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3298 /*
3299 * An error occurred so return it. Because skb_recv_datagram()
3300 * handles the blocking we don't see and worry about blocking
3301 * retries.
3302 */
3304 if (skb == NULL)
3305 goto out;
3307 if (pkt_sk(sk)->pressure)
3308 packet_rcv_has_room(pkt_sk(sk), NULL);
3310 if (pkt_sk(sk)->has_vnet_hdr) {
3311 err = packet_rcv_vnet(msg, skb, &len);
3312 if (err)
3313 goto out_free;
3314 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3315 }
3317 /* You lose any data beyond the buffer you gave. If it worries
3318 * a user program they can ask the device for its MTU
3319 * anyway.
3320 */
3321 copied = skb->len;
3322 if (copied > len) {
3323 copied = len;
3324 msg->msg_flags |= MSG_TRUNC;
3325 }
3327 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3328 if (err)
3329 goto out_free;
3331 if (sock->type != SOCK_PACKET) {
3332 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3334 /* Original length was stored in sockaddr_ll fields */
3335 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3336 sll->sll_family = AF_PACKET;
3337 sll->sll_protocol = skb->protocol;
3338 }
3340 sock_recv_ts_and_drops(msg, sk, skb);
3342 if (msg->msg_name) {
3343 /* If the address length field is there to be filled
3344 * in, we fill it in now.
3345 */
3346 if (sock->type == SOCK_PACKET) {
3347 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3348 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3349 } else {
3350 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3352 msg->msg_namelen = sll->sll_halen +
3353 offsetof(struct sockaddr_ll, sll_addr);
3354 }
3355 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3356 msg->msg_namelen);
3357 }
3359 if (pkt_sk(sk)->auxdata) {
3360 struct tpacket_auxdata aux;
3362 aux.tp_status = TP_STATUS_USER;
3363 if (skb->ip_summed == CHECKSUM_PARTIAL)
3364 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3365 else if (skb->pkt_type != PACKET_OUTGOING &&
3366 (skb->ip_summed == CHECKSUM_COMPLETE ||
3367 skb_csum_unnecessary(skb)))
3368 aux.tp_status |= TP_STATUS_CSUM_VALID;
3370 aux.tp_len = origlen;
3371 aux.tp_snaplen = skb->len;
3372 aux.tp_mac = 0;
3373 aux.tp_net = skb_network_offset(skb);
3374 if (skb_vlan_tag_present(skb)) {
3375 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3376 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3377 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3378 } else {
3379 aux.tp_vlan_tci = 0;
3380 aux.tp_vlan_tpid = 0;
3381 }
3382 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3383 }
3385 /*
3386 * Free or return the buffer as appropriate. Again this
3387 * hides all the races and re-entrancy issues from us.
3388 */
3389 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3391 out_free:
3392 skb_free_datagram(sk, skb);
3393 out:
3394 return err;
3395 }
3397 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3398 int peer)
3399 {
3400 struct net_device *dev;
3401 struct sock *sk = sock->sk;
3403 if (peer)
3404 return -EOPNOTSUPP;
3406 uaddr->sa_family = AF_PACKET;
3407 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3408 rcu_read_lock();
3409 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3410 if (dev)
3411 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3412 rcu_read_unlock();
3414 return sizeof(*uaddr);
3415 }
3417 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3418 int peer)
3419 {
3420 struct net_device *dev;
3421 struct sock *sk = sock->sk;
3422 struct packet_sock *po = pkt_sk(sk);
3423 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3425 if (peer)
3426 return -EOPNOTSUPP;
3428 sll->sll_family = AF_PACKET;
3429 sll->sll_ifindex = po->ifindex;
3430 sll->sll_protocol = po->num;
3431 sll->sll_pkttype = 0;
3432 rcu_read_lock();
3433 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3434 if (dev) {
3435 sll->sll_hatype = dev->type;
3436 sll->sll_halen = dev->addr_len;
3437 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3438 } else {
3439 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3440 sll->sll_halen = 0;
3441 }
3442 rcu_read_unlock();
3444 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3445 }
3447 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3448 int what)
3449 {
3450 switch (i->type) {
3451 case PACKET_MR_MULTICAST:
3452 if (i->alen != dev->addr_len)
3453 return -EINVAL;
3454 if (what > 0)
3455 return dev_mc_add(dev, i->addr);
3456 else
3457 return dev_mc_del(dev, i->addr);
3458 break;
3459 case PACKET_MR_PROMISC:
3460 return dev_set_promiscuity(dev, what);
3461 case PACKET_MR_ALLMULTI:
3462 return dev_set_allmulti(dev, what);
3463 case PACKET_MR_UNICAST:
3464 if (i->alen != dev->addr_len)
3465 return -EINVAL;
3466 if (what > 0)
3467 return dev_uc_add(dev, i->addr);
3468 else
3469 return dev_uc_del(dev, i->addr);
3470 break;
3471 default:
3472 break;
3473 }
3474 return 0;
3475 }
3477 static void packet_dev_mclist_delete(struct net_device *dev,
3478 struct packet_mclist **mlp)
3479 {
3480 struct packet_mclist *ml;
3482 while ((ml = *mlp) != NULL) {
3483 if (ml->ifindex == dev->ifindex) {
3484 packet_dev_mc(dev, ml, -1);
3485 *mlp = ml->next;
3486 kfree(ml);
3487 } else
3488 mlp = &ml->next;
3489 }
3490 }
3492 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3493 {
3494 struct packet_sock *po = pkt_sk(sk);
3495 struct packet_mclist *ml, *i;
3496 struct net_device *dev;
3497 int err;
3499 rtnl_lock();
3501 err = -ENODEV;
3502 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3503 if (!dev)
3504 goto done;
3506 err = -EINVAL;
3507 if (mreq->mr_alen > dev->addr_len)
3508 goto done;
3510 err = -ENOBUFS;
3511 i = kmalloc(sizeof(*i), GFP_KERNEL);
3512 if (i == NULL)
3513 goto done;
3515 err = 0;
3516 for (ml = po->mclist; ml; ml = ml->next) {
3517 if (ml->ifindex == mreq->mr_ifindex &&
3518 ml->type == mreq->mr_type &&
3519 ml->alen == mreq->mr_alen &&
3520 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3521 ml->count++;
3522 /* Free the new element ... */
3523 kfree(i);
3524 goto done;
3525 }
3526 }
3528 i->type = mreq->mr_type;
3529 i->ifindex = mreq->mr_ifindex;
3530 i->alen = mreq->mr_alen;
3531 memcpy(i->addr, mreq->mr_address, i->alen);
3532 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3533 i->count = 1;
3534 i->next = po->mclist;
3535 po->mclist = i;
3536 err = packet_dev_mc(dev, i, 1);
3537 if (err) {
3538 po->mclist = i->next;
3539 kfree(i);
3540 }
3542 done:
3543 rtnl_unlock();
3544 return err;
3545 }
3547 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3548 {
3549 struct packet_mclist *ml, **mlp;
3551 rtnl_lock();
3553 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3554 if (ml->ifindex == mreq->mr_ifindex &&
3555 ml->type == mreq->mr_type &&
3556 ml->alen == mreq->mr_alen &&
3557 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3558 if (--ml->count == 0) {
3559 struct net_device *dev;
3560 *mlp = ml->next;
3561 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3562 if (dev)
3563 packet_dev_mc(dev, ml, -1);
3564 kfree(ml);
3565 }
3566 break;
3567 }
3568 }
3569 rtnl_unlock();
3570 return 0;
3571 }
3573 static void packet_flush_mclist(struct sock *sk)
3574 {
3575 struct packet_sock *po = pkt_sk(sk);
3576 struct packet_mclist *ml;
3578 if (!po->mclist)
3579 return;
3581 rtnl_lock();
3582 while ((ml = po->mclist) != NULL) {
3583 struct net_device *dev;
3585 po->mclist = ml->next;
3586 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3587 if (dev != NULL)
3588 packet_dev_mc(dev, ml, -1);
3589 kfree(ml);
3590 }
3591 rtnl_unlock();
3592 }
3594 static int
3595 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3596 {
3597 struct sock *sk = sock->sk;
3598 struct packet_sock *po = pkt_sk(sk);
3599 int ret;
3601 if (level != SOL_PACKET)
3602 return -ENOPROTOOPT;
3604 switch (optname) {
3605 case PACKET_ADD_MEMBERSHIP:
3606 case PACKET_DROP_MEMBERSHIP:
3607 {
3608 struct packet_mreq_max mreq;
3609 int len = optlen;
3610 memset(&mreq, 0, sizeof(mreq));
3611 if (len < sizeof(struct packet_mreq))
3612 return -EINVAL;
3613 if (len > sizeof(mreq))
3614 len = sizeof(mreq);
3615 if (copy_from_user(&mreq, optval, len))
3616 return -EFAULT;
3617 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3618 return -EINVAL;
3619 if (optname == PACKET_ADD_MEMBERSHIP)
3620 ret = packet_mc_add(sk, &mreq);
3621 else
3622 ret = packet_mc_drop(sk, &mreq);
3623 return ret;
3624 }
3626 case PACKET_RX_RING:
3627 case PACKET_TX_RING:
3628 {
3629 union tpacket_req_u req_u;
3630 int len;
3632 lock_sock(sk);
3633 switch (po->tp_version) {
3634 case TPACKET_V1:
3635 case TPACKET_V2:
3636 len = sizeof(req_u.req);
3637 break;
3638 case TPACKET_V3:
3639 default:
3640 len = sizeof(req_u.req3);
3641 break;
3642 }
3643 if (optlen < len) {
3644 ret = -EINVAL;
3645 } else {
3646 if (copy_from_user(&req_u.req, optval, len))
3647 ret = -EFAULT;
3648 else
3649 ret = packet_set_ring(sk, &req_u, 0,
3650 optname == PACKET_TX_RING);
3651 }
3652 release_sock(sk);
3653 return ret;
3654 }
3655 case PACKET_COPY_THRESH:
3656 {
3657 int val;
3659 if (optlen != sizeof(val))
3660 return -EINVAL;
3661 if (copy_from_user(&val, optval, sizeof(val)))
3662 return -EFAULT;
3664 pkt_sk(sk)->copy_thresh = val;
3665 return 0;
3666 }
3667 case PACKET_VERSION:
3668 {
3669 int val;
3671 if (optlen != sizeof(val))
3672 return -EINVAL;
3673 if (copy_from_user(&val, optval, sizeof(val)))
3674 return -EFAULT;
3675 switch (val) {
3676 case TPACKET_V1:
3677 case TPACKET_V2:
3678 case TPACKET_V3:
3679 break;
3680 default:
3681 return -EINVAL;
3682 }
3683 lock_sock(sk);
3684 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3685 ret = -EBUSY;
3686 } else {
3687 po->tp_version = val;
3688 ret = 0;
3689 }
3690 release_sock(sk);
3691 return ret;
3692 }
3693 case PACKET_RESERVE:
3694 {
3695 unsigned int val;
3697 if (optlen != sizeof(val))
3698 return -EINVAL;
3699 if (copy_from_user(&val, optval, sizeof(val)))
3700 return -EFAULT;
3701 if (val > INT_MAX)
3702 return -EINVAL;
3703 lock_sock(sk);
3704 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3705 ret = -EBUSY;
3706 } else {
3707 po->tp_reserve = val;
3708 ret = 0;
3709 }
3710 release_sock(sk);
3711 return ret;
3712 }
3713 case PACKET_LOSS:
3714 {
3715 unsigned int val;
3717 if (optlen != sizeof(val))
3718 return -EINVAL;
3719 if (copy_from_user(&val, optval, sizeof(val)))
3720 return -EFAULT;
3722 lock_sock(sk);
3723 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3724 ret = -EBUSY;
3725 } else {
3726 po->tp_loss = !!val;
3727 ret = 0;
3728 }
3729 release_sock(sk);
3730 return ret;
3731 }
3732 case PACKET_AUXDATA:
3733 {
3734 int val;
3736 if (optlen < sizeof(val))
3737 return -EINVAL;
3738 if (copy_from_user(&val, optval, sizeof(val)))
3739 return -EFAULT;
3741 lock_sock(sk);
3742 po->auxdata = !!val;
3743 release_sock(sk);
3744 return 0;
3745 }
3746 case PACKET_ORIGDEV:
3747 {
3748 int val;
3750 if (optlen < sizeof(val))
3751 return -EINVAL;
3752 if (copy_from_user(&val, optval, sizeof(val)))
3753 return -EFAULT;
3755 lock_sock(sk);
3756 po->origdev = !!val;
3757 release_sock(sk);
3758 return 0;
3759 }
3760 case PACKET_VNET_HDR:
3761 {
3762 int val;
3764 if (sock->type != SOCK_RAW)
3765 return -EINVAL;
3766 if (optlen < sizeof(val))
3767 return -EINVAL;
3768 if (copy_from_user(&val, optval, sizeof(val)))
3769 return -EFAULT;
3771 lock_sock(sk);
3772 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3773 ret = -EBUSY;
3774 } else {
3775 po->has_vnet_hdr = !!val;
3776 ret = 0;
3777 }
3778 release_sock(sk);
3779 return ret;
3780 }
3781 case PACKET_TIMESTAMP:
3782 {
3783 int val;
3785 if (optlen != sizeof(val))
3786 return -EINVAL;
3787 if (copy_from_user(&val, optval, sizeof(val)))
3788 return -EFAULT;
3790 po->tp_tstamp = val;
3791 return 0;
3792 }
3793 case PACKET_FANOUT:
3794 {
3795 int val;
3797 if (optlen != sizeof(val))
3798 return -EINVAL;
3799 if (copy_from_user(&val, optval, sizeof(val)))
3800 return -EFAULT;
3802 return fanout_add(sk, val & 0xffff, val >> 16);
3803 }
3804 case PACKET_FANOUT_DATA:
3805 {
3806 if (!po->fanout)
3807 return -EINVAL;
3809 return fanout_set_data(po, optval, optlen);
3810 }
3811 case PACKET_TX_HAS_OFF:
3812 {
3813 unsigned int val;
3815 if (optlen != sizeof(val))
3816 return -EINVAL;
3817 if (copy_from_user(&val, optval, sizeof(val)))
3818 return -EFAULT;
3820 lock_sock(sk);
3821 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3822 ret = -EBUSY;
3823 } else {
3824 po->tp_tx_has_off = !!val;
3825 ret = 0;
3826 }
3827 release_sock(sk);
3828 return 0;
3829 }
3830 case PACKET_QDISC_BYPASS:
3831 {
3832 int val;
3834 if (optlen != sizeof(val))
3835 return -EINVAL;
3836 if (copy_from_user(&val, optval, sizeof(val)))
3837 return -EFAULT;
3839 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3840 return 0;
3841 }
3842 default:
3843 return -ENOPROTOOPT;
3844 }
3845 }
3847 static int packet_getsockopt(struct socket *sock, int level, int optname,
3848 char __user *optval, int __user *optlen)
3849 {
3850 int len;
3851 int val, lv = sizeof(val);
3852 struct sock *sk = sock->sk;
3853 struct packet_sock *po = pkt_sk(sk);
3854 void *data = &val;
3855 union tpacket_stats_u st;
3856 struct tpacket_rollover_stats rstats;
3858 if (level != SOL_PACKET)
3859 return -ENOPROTOOPT;
3861 if (get_user(len, optlen))
3862 return -EFAULT;
3864 if (len < 0)
3865 return -EINVAL;
3867 switch (optname) {
3868 case PACKET_STATISTICS:
3869 spin_lock_bh(&sk->sk_receive_queue.lock);
3870 memcpy(&st, &po->stats, sizeof(st));
3871 memset(&po->stats, 0, sizeof(po->stats));
3872 spin_unlock_bh(&sk->sk_receive_queue.lock);
3874 if (po->tp_version == TPACKET_V3) {
3875 lv = sizeof(struct tpacket_stats_v3);
3876 st.stats3.tp_packets += st.stats3.tp_drops;
3877 data = &st.stats3;
3878 } else {
3879 lv = sizeof(struct tpacket_stats);
3880 st.stats1.tp_packets += st.stats1.tp_drops;
3881 data = &st.stats1;
3882 }
3884 break;
3885 case PACKET_AUXDATA:
3886 val = po->auxdata;
3887 break;
3888 case PACKET_ORIGDEV:
3889 val = po->origdev;
3890 break;
3891 case PACKET_VNET_HDR:
3892 val = po->has_vnet_hdr;
3893 break;
3894 case PACKET_VERSION:
3895 val = po->tp_version;
3896 break;
3897 case PACKET_HDRLEN:
3898 if (len > sizeof(int))
3899 len = sizeof(int);
3900 if (len < sizeof(int))
3901 return -EINVAL;
3902 if (copy_from_user(&val, optval, len))
3903 return -EFAULT;
3904 switch (val) {
3905 case TPACKET_V1:
3906 val = sizeof(struct tpacket_hdr);
3907 break;
3908 case TPACKET_V2:
3909 val = sizeof(struct tpacket2_hdr);
3910 break;
3911 case TPACKET_V3:
3912 val = sizeof(struct tpacket3_hdr);
3913 break;
3914 default:
3915 return -EINVAL;
3916 }
3917 break;
3918 case PACKET_RESERVE:
3919 val = po->tp_reserve;
3920 break;
3921 case PACKET_LOSS:
3922 val = po->tp_loss;
3923 break;
3924 case PACKET_TIMESTAMP:
3925 val = po->tp_tstamp;
3926 break;
3927 case PACKET_FANOUT:
3928 val = (po->fanout ?
3929 ((u32)po->fanout->id |
3930 ((u32)po->fanout->type << 16) |
3931 ((u32)po->fanout->flags << 24)) :
3932 0);
3933 break;
3934 case PACKET_ROLLOVER_STATS:
3935 if (!po->rollover)
3936 return -EINVAL;
3937 rstats.tp_all = atomic_long_read(&po->rollover->num);
3938 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3939 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3940 data = &rstats;
3941 lv = sizeof(rstats);
3942 break;
3943 case PACKET_TX_HAS_OFF:
3944 val = po->tp_tx_has_off;
3945 break;
3946 case PACKET_QDISC_BYPASS:
3947 val = packet_use_direct_xmit(po);
3948 break;
3949 default:
3950 return -ENOPROTOOPT;
3951 }
3953 if (len > lv)
3954 len = lv;
3955 if (put_user(len, optlen))
3956 return -EFAULT;
3957 if (copy_to_user(optval, data, len))
3958 return -EFAULT;
3959 return 0;
3960 }
3963 #ifdef CONFIG_COMPAT
3964 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3965 char __user *optval, unsigned int optlen)
3966 {
3967 struct packet_sock *po = pkt_sk(sock->sk);
3969 if (level != SOL_PACKET)
3970 return -ENOPROTOOPT;
3972 if (optname == PACKET_FANOUT_DATA &&
3973 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3974 optval = (char __user *)get_compat_bpf_fprog(optval);
3975 if (!optval)
3976 return -EFAULT;
3977 optlen = sizeof(struct sock_fprog);
3978 }
3980 return packet_setsockopt(sock, level, optname, optval, optlen);
3981 }
3982 #endif
3984 static int packet_notifier(struct notifier_block *this,
3985 unsigned long msg, void *ptr)
3986 {
3987 struct sock *sk;
3988 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3989 struct net *net = dev_net(dev);
3991 rcu_read_lock();
3992 sk_for_each_rcu(sk, &net->packet.sklist) {
3993 struct packet_sock *po = pkt_sk(sk);
3995 switch (msg) {
3996 case NETDEV_UNREGISTER:
3997 if (po->mclist)
3998 packet_dev_mclist_delete(dev, &po->mclist);
3999 /* fallthrough */
4001 case NETDEV_DOWN:
4002 if (dev->ifindex == po->ifindex) {
4003 spin_lock(&po->bind_lock);
4004 if (po->running) {
4005 __unregister_prot_hook(sk, false);
4006 sk->sk_err = ENETDOWN;
4007 if (!sock_flag(sk, SOCK_DEAD))
4008 sk->sk_error_report(sk);
4009 }
4010 if (msg == NETDEV_UNREGISTER) {
4011 packet_cached_dev_reset(po);
4012 po->ifindex = -1;
4013 if (po->prot_hook.dev)
4014 dev_put(po->prot_hook.dev);
4015 po->prot_hook.dev = NULL;
4016 }
4017 spin_unlock(&po->bind_lock);
4018 }
4019 break;
4020 case NETDEV_UP:
4021 if (dev->ifindex == po->ifindex) {
4022 spin_lock(&po->bind_lock);
4023 if (po->num)
4024 register_prot_hook(sk);
4025 spin_unlock(&po->bind_lock);
4026 }
4027 break;
4028 }
4029 }
4030 rcu_read_unlock();
4031 return NOTIFY_DONE;
4032 }
4035 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4036 unsigned long arg)
4037 {
4038 struct sock *sk = sock->sk;
4040 switch (cmd) {
4041 case SIOCOUTQ:
4042 {
4043 int amount = sk_wmem_alloc_get(sk);
4045 return put_user(amount, (int __user *)arg);
4046 }
4047 case SIOCINQ:
4048 {
4049 struct sk_buff *skb;
4050 int amount = 0;
4052 spin_lock_bh(&sk->sk_receive_queue.lock);
4053 skb = skb_peek(&sk->sk_receive_queue);
4054 if (skb)
4055 amount = skb->len;
4056 spin_unlock_bh(&sk->sk_receive_queue.lock);
4057 return put_user(amount, (int __user *)arg);
4058 }
4059 case SIOCGSTAMP:
4060 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4061 case SIOCGSTAMPNS:
4062 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4064 #ifdef CONFIG_INET
4065 case SIOCADDRT:
4066 case SIOCDELRT:
4067 case SIOCDARP:
4068 case SIOCGARP:
4069 case SIOCSARP:
4070 case SIOCGIFADDR:
4071 case SIOCSIFADDR:
4072 case SIOCGIFBRDADDR:
4073 case SIOCSIFBRDADDR:
4074 case SIOCGIFNETMASK:
4075 case SIOCSIFNETMASK:
4076 case SIOCGIFDSTADDR:
4077 case SIOCSIFDSTADDR:
4078 case SIOCSIFFLAGS:
4079 return inet_dgram_ops.ioctl(sock, cmd, arg);
4080 #endif
4082 default:
4083 return -ENOIOCTLCMD;
4084 }
4085 return 0;
4086 }
4088 static __poll_t packet_poll(struct file *file, struct socket *sock,
4089 poll_table *wait)
4090 {
4091 struct sock *sk = sock->sk;
4092 struct packet_sock *po = pkt_sk(sk);
4093 __poll_t mask = datagram_poll(file, sock, wait);
4095 spin_lock_bh(&sk->sk_receive_queue.lock);
4096 if (po->rx_ring.pg_vec) {
4097 if (!packet_previous_rx_frame(po, &po->rx_ring,
4098 TP_STATUS_KERNEL))
4099 mask |= EPOLLIN | EPOLLRDNORM;
4100 }
4101 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4102 po->pressure = 0;
4103 spin_unlock_bh(&sk->sk_receive_queue.lock);
4104 spin_lock_bh(&sk->sk_write_queue.lock);
4105 if (po->tx_ring.pg_vec) {
4106 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4107 mask |= EPOLLOUT | EPOLLWRNORM;
4108 }
4109 spin_unlock_bh(&sk->sk_write_queue.lock);
4110 return mask;
4111 }
4114 /* Dirty? Well, I still did not learn better way to account
4115 * for user mmaps.
4116 */
4118 static void packet_mm_open(struct vm_area_struct *vma)
4119 {
4120 struct file *file = vma->vm_file;
4121 struct socket *sock = file->private_data;
4122 struct sock *sk = sock->sk;
4124 if (sk)
4125 atomic_inc(&pkt_sk(sk)->mapped);
4126 }
4128 static void packet_mm_close(struct vm_area_struct *vma)
4129 {
4130 struct file *file = vma->vm_file;
4131 struct socket *sock = file->private_data;
4132 struct sock *sk = sock->sk;
4134 if (sk)
4135 atomic_dec(&pkt_sk(sk)->mapped);
4136 }
4138 static const struct vm_operations_struct packet_mmap_ops = {
4139 .open = packet_mm_open,
4140 .close = packet_mm_close,
4141 };
4143 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4144 unsigned int len)
4145 {
4146 int i;
4148 for (i = 0; i < len; i++) {
4149 if (likely(pg_vec[i].buffer)) {
4150 if (is_vmalloc_addr(pg_vec[i].buffer))
4151 vfree(pg_vec[i].buffer);
4152 else
4153 free_pages((unsigned long)pg_vec[i].buffer,
4154 order);
4155 pg_vec[i].buffer = NULL;
4156 }
4157 }
4158 kfree(pg_vec);
4159 }
4161 static char *alloc_one_pg_vec_page(unsigned long order)
4162 {
4163 char *buffer;
4164 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4165 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4167 buffer = (char *) __get_free_pages(gfp_flags, order);
4168 if (buffer)
4169 return buffer;
4171 /* __get_free_pages failed, fall back to vmalloc */
4172 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4173 if (buffer)
4174 return buffer;
4176 /* vmalloc failed, lets dig into swap here */
4177 gfp_flags &= ~__GFP_NORETRY;
4178 buffer = (char *) __get_free_pages(gfp_flags, order);
4179 if (buffer)
4180 return buffer;
4182 /* complete and utter failure */
4183 return NULL;
4184 }
4186 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4187 {
4188 unsigned int block_nr = req->tp_block_nr;
4189 struct pgv *pg_vec;
4190 int i;
4192 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4193 if (unlikely(!pg_vec))
4194 goto out;
4196 for (i = 0; i < block_nr; i++) {
4197 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4198 if (unlikely(!pg_vec[i].buffer))
4199 goto out_free_pgvec;
4200 }
4202 out:
4203 return pg_vec;
4205 out_free_pgvec:
4206 free_pg_vec(pg_vec, order, block_nr);
4207 pg_vec = NULL;
4208 goto out;
4209 }
4211 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4212 int closing, int tx_ring)
4213 {
4214 struct pgv *pg_vec = NULL;
4215 struct packet_sock *po = pkt_sk(sk);
4216 int was_running, order = 0;
4217 struct packet_ring_buffer *rb;
4218 struct sk_buff_head *rb_queue;
4219 __be16 num;
4220 int err = -EINVAL;
4221 /* Added to avoid minimal code churn */
4222 struct tpacket_req *req = &req_u->req;
4224 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4225 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4227 err = -EBUSY;
4228 if (!closing) {
4229 if (atomic_read(&po->mapped))
4230 goto out;
4231 if (packet_read_pending(rb))
4232 goto out;
4233 }
4235 if (req->tp_block_nr) {
4236 unsigned int min_frame_size;
4238 /* Sanity tests and some calculations */
4239 err = -EBUSY;
4240 if (unlikely(rb->pg_vec))
4241 goto out;
4243 switch (po->tp_version) {
4244 case TPACKET_V1:
4245 po->tp_hdrlen = TPACKET_HDRLEN;
4246 break;
4247 case TPACKET_V2:
4248 po->tp_hdrlen = TPACKET2_HDRLEN;
4249 break;
4250 case TPACKET_V3:
4251 po->tp_hdrlen = TPACKET3_HDRLEN;
4252 break;
4253 }
4255 err = -EINVAL;
4256 if (unlikely((int)req->tp_block_size <= 0))
4257 goto out;
4258 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4259 goto out;
4260 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4261 if (po->tp_version >= TPACKET_V3 &&
4262 req->tp_block_size <
4263 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4264 goto out;
4265 if (unlikely(req->tp_frame_size < min_frame_size))
4266 goto out;
4267 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4268 goto out;
4270 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4271 if (unlikely(rb->frames_per_block == 0))
4272 goto out;
4273 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
4274 goto out;
4275 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4276 req->tp_frame_nr))
4277 goto out;
4279 err = -ENOMEM;
4280 order = get_order(req->tp_block_size);
4281 pg_vec = alloc_pg_vec(req, order);
4282 if (unlikely(!pg_vec))
4283 goto out;
4284 switch (po->tp_version) {
4285 case TPACKET_V3:
4286 /* Block transmit is not supported yet */
4287 if (!tx_ring) {
4288 init_prb_bdqc(po, rb, pg_vec, req_u);
4289 } else {
4290 struct tpacket_req3 *req3 = &req_u->req3;
4292 if (req3->tp_retire_blk_tov ||
4293 req3->tp_sizeof_priv ||
4294 req3->tp_feature_req_word) {
4295 err = -EINVAL;
4296 goto out;
4297 }
4298 }
4299 break;
4300 default:
4301 break;
4302 }
4303 }
4304 /* Done */
4305 else {
4306 err = -EINVAL;
4307 if (unlikely(req->tp_frame_nr))
4308 goto out;
4309 }
4312 /* Detach socket from network */
4313 spin_lock(&po->bind_lock);
4314 was_running = po->running;
4315 num = po->num;
4316 if (was_running) {
4317 po->num = 0;
4318 __unregister_prot_hook(sk, false);
4319 }
4320 spin_unlock(&po->bind_lock);
4322 synchronize_net();
4324 err = -EBUSY;
4325 mutex_lock(&po->pg_vec_lock);
4326 if (closing || atomic_read(&po->mapped) == 0) {
4327 err = 0;
4328 spin_lock_bh(&rb_queue->lock);
4329 swap(rb->pg_vec, pg_vec);
4330 rb->frame_max = (req->tp_frame_nr - 1);
4331 rb->head = 0;
4332 rb->frame_size = req->tp_frame_size;
4333 spin_unlock_bh(&rb_queue->lock);
4335 swap(rb->pg_vec_order, order);
4336 swap(rb->pg_vec_len, req->tp_block_nr);
4338 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4339 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4340 tpacket_rcv : packet_rcv;
4341 skb_queue_purge(rb_queue);
4342 if (atomic_read(&po->mapped))
4343 pr_err("packet_mmap: vma is busy: %d\n",
4344 atomic_read(&po->mapped));
4345 }
4346 mutex_unlock(&po->pg_vec_lock);
4348 spin_lock(&po->bind_lock);
4349 if (was_running) {
4350 po->num = num;
4351 register_prot_hook(sk);
4352 }
4353 spin_unlock(&po->bind_lock);
4354 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4355 /* Because we don't support block-based V3 on tx-ring */
4356 if (!tx_ring)
4357 prb_shutdown_retire_blk_timer(po, rb_queue);
4358 }
4360 if (pg_vec)
4361 free_pg_vec(pg_vec, order, req->tp_block_nr);
4362 out:
4363 return err;
4364 }
4366 static int packet_mmap(struct file *file, struct socket *sock,
4367 struct vm_area_struct *vma)
4368 {
4369 struct sock *sk = sock->sk;
4370 struct packet_sock *po = pkt_sk(sk);
4371 unsigned long size, expected_size;
4372 struct packet_ring_buffer *rb;
4373 unsigned long start;
4374 int err = -EINVAL;
4375 int i;
4377 if (vma->vm_pgoff)
4378 return -EINVAL;
4380 mutex_lock(&po->pg_vec_lock);
4382 expected_size = 0;
4383 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4384 if (rb->pg_vec) {
4385 expected_size += rb->pg_vec_len
4386 * rb->pg_vec_pages
4387 * PAGE_SIZE;
4388 }
4389 }
4391 if (expected_size == 0)
4392 goto out;
4394 size = vma->vm_end - vma->vm_start;
4395 if (size != expected_size)
4396 goto out;
4398 start = vma->vm_start;
4399 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4400 if (rb->pg_vec == NULL)
4401 continue;
4403 for (i = 0; i < rb->pg_vec_len; i++) {
4404 struct page *page;
4405 void *kaddr = rb->pg_vec[i].buffer;
4406 int pg_num;
4408 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4409 page = pgv_to_page(kaddr);
4410 err = vm_insert_page(vma, start, page);
4411 if (unlikely(err))
4412 goto out;
4413 start += PAGE_SIZE;
4414 kaddr += PAGE_SIZE;
4415 }
4416 }
4417 }
4419 atomic_inc(&po->mapped);
4420 vma->vm_ops = &packet_mmap_ops;
4421 err = 0;
4423 out:
4424 mutex_unlock(&po->pg_vec_lock);
4425 return err;
4426 }
4428 static const struct proto_ops packet_ops_spkt = {
4429 .family = PF_PACKET,
4430 .owner = THIS_MODULE,
4431 .release = packet_release,
4432 .bind = packet_bind_spkt,
4433 .connect = sock_no_connect,
4434 .socketpair = sock_no_socketpair,
4435 .accept = sock_no_accept,
4436 .getname = packet_getname_spkt,
4437 .poll = datagram_poll,
4438 .ioctl = packet_ioctl,
4439 .listen = sock_no_listen,
4440 .shutdown = sock_no_shutdown,
4441 .setsockopt = sock_no_setsockopt,
4442 .getsockopt = sock_no_getsockopt,
4443 .sendmsg = packet_sendmsg_spkt,
4444 .recvmsg = packet_recvmsg,
4445 .mmap = sock_no_mmap,
4446 .sendpage = sock_no_sendpage,
4447 };
4449 static const struct proto_ops packet_ops = {
4450 .family = PF_PACKET,
4451 .owner = THIS_MODULE,
4452 .release = packet_release,
4453 .bind = packet_bind,
4454 .connect = sock_no_connect,
4455 .socketpair = sock_no_socketpair,
4456 .accept = sock_no_accept,
4457 .getname = packet_getname,
4458 .poll = packet_poll,
4459 .ioctl = packet_ioctl,
4460 .listen = sock_no_listen,
4461 .shutdown = sock_no_shutdown,
4462 .setsockopt = packet_setsockopt,
4463 .getsockopt = packet_getsockopt,
4464 #ifdef CONFIG_COMPAT
4465 .compat_setsockopt = compat_packet_setsockopt,
4466 #endif
4467 .sendmsg = packet_sendmsg,
4468 .recvmsg = packet_recvmsg,
4469 .mmap = packet_mmap,
4470 .sendpage = sock_no_sendpage,
4471 };
4473 static const struct net_proto_family packet_family_ops = {
4474 .family = PF_PACKET,
4475 .create = packet_create,
4476 .owner = THIS_MODULE,
4477 };
4479 static struct notifier_block packet_netdev_notifier = {
4480 .notifier_call = packet_notifier,
4481 };
4483 #ifdef CONFIG_PROC_FS
4485 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4486 __acquires(RCU)
4487 {
4488 struct net *net = seq_file_net(seq);
4490 rcu_read_lock();
4491 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4492 }
4494 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4495 {
4496 struct net *net = seq_file_net(seq);
4497 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4498 }
4500 static void packet_seq_stop(struct seq_file *seq, void *v)
4501 __releases(RCU)
4502 {
4503 rcu_read_unlock();
4504 }
4506 static int packet_seq_show(struct seq_file *seq, void *v)
4507 {
4508 if (v == SEQ_START_TOKEN)
4509 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4510 else {
4511 struct sock *s = sk_entry(v);
4512 const struct packet_sock *po = pkt_sk(s);
4514 seq_printf(seq,
4515 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4516 s,
4517 refcount_read(&s->sk_refcnt),
4518 s->sk_type,
4519 ntohs(po->num),
4520 po->ifindex,
4521 po->running,
4522 atomic_read(&s->sk_rmem_alloc),
4523 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4524 sock_i_ino(s));
4525 }
4527 return 0;
4528 }
4530 static const struct seq_operations packet_seq_ops = {
4531 .start = packet_seq_start,
4532 .next = packet_seq_next,
4533 .stop = packet_seq_stop,
4534 .show = packet_seq_show,
4535 };
4536 #endif
4538 static int __net_init packet_net_init(struct net *net)
4539 {
4540 mutex_init(&net->packet.sklist_lock);
4541 INIT_HLIST_HEAD(&net->packet.sklist);
4543 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4544 sizeof(struct seq_net_private)))
4545 return -ENOMEM;
4547 return 0;
4548 }
4550 static void __net_exit packet_net_exit(struct net *net)
4551 {
4552 remove_proc_entry("packet", net->proc_net);
4553 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4554 }
4556 static struct pernet_operations packet_net_ops = {
4557 .init = packet_net_init,
4558 .exit = packet_net_exit,
4559 };
4562 static void __exit packet_exit(void)
4563 {
4564 unregister_netdevice_notifier(&packet_netdev_notifier);
4565 unregister_pernet_subsys(&packet_net_ops);
4566 sock_unregister(PF_PACKET);
4567 proto_unregister(&packet_proto);
4568 }
4570 static int __init packet_init(void)
4571 {
4572 int rc = proto_register(&packet_proto, 0);
4574 if (rc != 0)
4575 goto out;
4577 sock_register(&packet_family_ops);
4578 register_pernet_subsys(&packet_net_ops);
4579 register_netdevice_notifier(&packet_netdev_notifier);
4580 out:
4581 return rc;
4582 }
4584 module_init(packet_init);
4585 module_exit(packet_exit);
4586 MODULE_LICENSE("GPL");
4587 MODULE_ALIAS_NETPROTO(PF_PACKET);