Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * PACKET - implements raw packet sockets.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 *
13 * Fixes:
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
42 * and packet_mreq.
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * layer.
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
47 */
48
49#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
50
51#include <linux/ethtool.h>
52#include <linux/filter.h>
53#include <linux/types.h>
54#include <linux/mm.h>
55#include <linux/capability.h>
56#include <linux/fcntl.h>
57#include <linux/socket.h>
58#include <linux/in.h>
59#include <linux/inet.h>
60#include <linux/netdevice.h>
61#include <linux/if_packet.h>
62#include <linux/wireless.h>
63#include <linux/kernel.h>
64#include <linux/kmod.h>
65#include <linux/slab.h>
66#include <linux/vmalloc.h>
67#include <net/net_namespace.h>
68#include <net/ip.h>
69#include <net/protocol.h>
70#include <linux/skbuff.h>
71#include <net/sock.h>
72#include <linux/errno.h>
73#include <linux/timer.h>
74#include <linux/uaccess.h>
75#include <asm/ioctls.h>
76#include <asm/page.h>
77#include <asm/cacheflush.h>
78#include <asm/io.h>
79#include <linux/proc_fs.h>
80#include <linux/seq_file.h>
81#include <linux/poll.h>
82#include <linux/module.h>
83#include <linux/init.h>
84#include <linux/mutex.h>
85#include <linux/if_vlan.h>
86#include <linux/virtio_net.h>
87#include <linux/errqueue.h>
88#include <linux/net_tstamp.h>
89#include <linux/percpu.h>
90#ifdef CONFIG_INET
91#include <net/inet_common.h>
92#endif
93#include <linux/bpf.h>
94#include <net/compat.h>
95#include <linux/netfilter_netdev.h>
96
97#include "internal.h"
98
99/*
100 Assumptions:
101 - If the device has no dev->header_ops->create, there is no LL header
102 visible above the device. In this case, its hard_header_len should be 0.
103 The device may prepend its own header internally. In this case, its
104 needed_headroom should be set to the space needed for it to add its
105 internal header.
106 For example, a WiFi driver pretending to be an Ethernet driver should
107 set its hard_header_len to be the Ethernet header length, and set its
108 needed_headroom to be (the real WiFi header length - the fake Ethernet
109 header length).
110 - packet socket receives packets with pulled ll header,
111 so that SOCK_RAW should push it back.
112
113On receive:
114-----------
115
116Incoming, dev_has_header(dev) == true
117 mac_header -> ll header
118 data -> data
119
120Outgoing, dev_has_header(dev) == true
121 mac_header -> ll header
122 data -> ll header
123
124Incoming, dev_has_header(dev) == false
125 mac_header -> data
126 However drivers often make it point to the ll header.
127 This is incorrect because the ll header should be invisible to us.
128 data -> data
129
130Outgoing, dev_has_header(dev) == false
131 mac_header -> data. ll header is invisible to us.
132 data -> data
133
134Resume
135 If dev_has_header(dev) == false we are unable to restore the ll header,
136 because it is invisible to us.
137
138
139On transmit:
140------------
141
142dev_has_header(dev) == true
143 mac_header -> ll header
144 data -> ll header
145
146dev_has_header(dev) == false (ll header is invisible to us)
147 mac_header -> data
148 data -> data
149
150 We should set network_header on output to the correct position,
151 packet classifier depends on it.
152 */
153
154/* Private packet socket structures. */
155
156/* identical to struct packet_mreq except it has
157 * a longer address field.
158 */
159struct packet_mreq_max {
160 int mr_ifindex;
161 unsigned short mr_type;
162 unsigned short mr_alen;
163 unsigned char mr_address[MAX_ADDR_LEN];
164};
165
166union tpacket_uhdr {
167 struct tpacket_hdr *h1;
168 struct tpacket2_hdr *h2;
169 struct tpacket3_hdr *h3;
170 void *raw;
171};
172
173static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
174 int closing, int tx_ring);
175
176#define V3_ALIGNMENT (8)
177
178#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
179
180#define BLK_PLUS_PRIV(sz_of_priv) \
181 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
182
183#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
184#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
185#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
186#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
187#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
188#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
189
190struct packet_sock;
191static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 struct packet_type *pt, struct net_device *orig_dev);
193
194static void *packet_previous_frame(struct packet_sock *po,
195 struct packet_ring_buffer *rb,
196 int status);
197static void packet_increment_head(struct packet_ring_buffer *buff);
198static int prb_curr_blk_in_use(struct tpacket_block_desc *);
199static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203static int prb_queue_frozen(struct tpacket_kbdq_core *);
204static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206static void prb_retire_rx_blk_timer_expired(struct timer_list *);
207static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
209static void prb_clear_rxhash(struct tpacket_kbdq_core *,
210 struct tpacket3_hdr *);
211static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
212 struct tpacket3_hdr *);
213static void packet_flush_mclist(struct sock *sk);
214static u16 packet_pick_tx_queue(struct sk_buff *skb);
215
216struct packet_skb_cb {
217 union {
218 struct sockaddr_pkt pkt;
219 union {
220 /* Trick: alias skb original length with
221 * ll.sll_family and ll.protocol in order
222 * to save room.
223 */
224 unsigned int origlen;
225 struct sockaddr_ll ll;
226 };
227 } sa;
228};
229
230#define vio_le() virtio_legacy_is_little_endian()
231
232#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
233
234#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235#define GET_PBLOCK_DESC(x, bid) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239#define GET_NEXT_PRB_BLK_NUM(x) \
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
242
243static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244static void __fanout_link(struct sock *sk, struct packet_sock *po);
245
246#ifdef CONFIG_NETFILTER_EGRESS
247static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb)
248{
249 struct sk_buff *next, *head = NULL, *tail;
250 int rc;
251
252 rcu_read_lock();
253 for (; skb != NULL; skb = next) {
254 next = skb->next;
255 skb_mark_not_on_list(skb);
256
257 if (!nf_hook_egress(skb, &rc, skb->dev))
258 continue;
259
260 if (!head)
261 head = skb;
262 else
263 tail->next = skb;
264
265 tail = skb;
266 }
267 rcu_read_unlock();
268
269 return head;
270}
271#endif
272
273static int packet_direct_xmit(struct sk_buff *skb)
274{
275#ifdef CONFIG_NETFILTER_EGRESS
276 if (nf_hook_egress_active()) {
277 skb = nf_hook_direct_egress(skb);
278 if (!skb)
279 return NET_XMIT_DROP;
280 }
281#endif
282 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
283}
284
285static struct net_device *packet_cached_dev_get(struct packet_sock *po)
286{
287 struct net_device *dev;
288
289 rcu_read_lock();
290 dev = rcu_dereference(po->cached_dev);
291 dev_hold(dev);
292 rcu_read_unlock();
293
294 return dev;
295}
296
297static void packet_cached_dev_assign(struct packet_sock *po,
298 struct net_device *dev)
299{
300 rcu_assign_pointer(po->cached_dev, dev);
301}
302
303static void packet_cached_dev_reset(struct packet_sock *po)
304{
305 RCU_INIT_POINTER(po->cached_dev, NULL);
306}
307
308static bool packet_use_direct_xmit(const struct packet_sock *po)
309{
310 return po->xmit == packet_direct_xmit;
311}
312
313static u16 packet_pick_tx_queue(struct sk_buff *skb)
314{
315 struct net_device *dev = skb->dev;
316 const struct net_device_ops *ops = dev->netdev_ops;
317 int cpu = raw_smp_processor_id();
318 u16 queue_index;
319
320#ifdef CONFIG_XPS
321 skb->sender_cpu = cpu + 1;
322#endif
323 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
324 if (ops->ndo_select_queue) {
325 queue_index = ops->ndo_select_queue(dev, skb, NULL);
326 queue_index = netdev_cap_txqueue(dev, queue_index);
327 } else {
328 queue_index = netdev_pick_tx(dev, skb, NULL);
329 }
330
331 return queue_index;
332}
333
334/* __register_prot_hook must be invoked through register_prot_hook
335 * or from a context in which asynchronous accesses to the packet
336 * socket is not possible (packet_create()).
337 */
338static void __register_prot_hook(struct sock *sk)
339{
340 struct packet_sock *po = pkt_sk(sk);
341
342 if (!po->running) {
343 if (po->fanout)
344 __fanout_link(sk, po);
345 else
346 dev_add_pack(&po->prot_hook);
347
348 sock_hold(sk);
349 po->running = 1;
350 }
351}
352
353static void register_prot_hook(struct sock *sk)
354{
355 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
356 __register_prot_hook(sk);
357}
358
359/* If the sync parameter is true, we will temporarily drop
360 * the po->bind_lock and do a synchronize_net to make sure no
361 * asynchronous packet processing paths still refer to the elements
362 * of po->prot_hook. If the sync parameter is false, it is the
363 * callers responsibility to take care of this.
364 */
365static void __unregister_prot_hook(struct sock *sk, bool sync)
366{
367 struct packet_sock *po = pkt_sk(sk);
368
369 lockdep_assert_held_once(&po->bind_lock);
370
371 po->running = 0;
372
373 if (po->fanout)
374 __fanout_unlink(sk, po);
375 else
376 __dev_remove_pack(&po->prot_hook);
377
378 __sock_put(sk);
379
380 if (sync) {
381 spin_unlock(&po->bind_lock);
382 synchronize_net();
383 spin_lock(&po->bind_lock);
384 }
385}
386
387static void unregister_prot_hook(struct sock *sk, bool sync)
388{
389 struct packet_sock *po = pkt_sk(sk);
390
391 if (po->running)
392 __unregister_prot_hook(sk, sync);
393}
394
395static inline struct page * __pure pgv_to_page(void *addr)
396{
397 if (is_vmalloc_addr(addr))
398 return vmalloc_to_page(addr);
399 return virt_to_page(addr);
400}
401
402static void __packet_set_status(struct packet_sock *po, void *frame, int status)
403{
404 union tpacket_uhdr h;
405
406 h.raw = frame;
407 switch (po->tp_version) {
408 case TPACKET_V1:
409 h.h1->tp_status = status;
410 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
411 break;
412 case TPACKET_V2:
413 h.h2->tp_status = status;
414 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
415 break;
416 case TPACKET_V3:
417 h.h3->tp_status = status;
418 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
419 break;
420 default:
421 WARN(1, "TPACKET version not supported.\n");
422 BUG();
423 }
424
425 smp_wmb();
426}
427
428static int __packet_get_status(const struct packet_sock *po, void *frame)
429{
430 union tpacket_uhdr h;
431
432 smp_rmb();
433
434 h.raw = frame;
435 switch (po->tp_version) {
436 case TPACKET_V1:
437 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
438 return h.h1->tp_status;
439 case TPACKET_V2:
440 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
441 return h.h2->tp_status;
442 case TPACKET_V3:
443 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
444 return h.h3->tp_status;
445 default:
446 WARN(1, "TPACKET version not supported.\n");
447 BUG();
448 return 0;
449 }
450}
451
452static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
453 unsigned int flags)
454{
455 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
456
457 if (shhwtstamps &&
458 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
459 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
460 return TP_STATUS_TS_RAW_HARDWARE;
461
462 if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
463 ktime_to_timespec64_cond(skb_tstamp(skb), ts))
464 return TP_STATUS_TS_SOFTWARE;
465
466 return 0;
467}
468
469static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
470 struct sk_buff *skb)
471{
472 union tpacket_uhdr h;
473 struct timespec64 ts;
474 __u32 ts_status;
475
476 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
477 return 0;
478
479 h.raw = frame;
480 /*
481 * versions 1 through 3 overflow the timestamps in y2106, since they
482 * all store the seconds in a 32-bit unsigned integer.
483 * If we create a version 4, that should have a 64-bit timestamp,
484 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
485 * nanoseconds.
486 */
487 switch (po->tp_version) {
488 case TPACKET_V1:
489 h.h1->tp_sec = ts.tv_sec;
490 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
491 break;
492 case TPACKET_V2:
493 h.h2->tp_sec = ts.tv_sec;
494 h.h2->tp_nsec = ts.tv_nsec;
495 break;
496 case TPACKET_V3:
497 h.h3->tp_sec = ts.tv_sec;
498 h.h3->tp_nsec = ts.tv_nsec;
499 break;
500 default:
501 WARN(1, "TPACKET version not supported.\n");
502 BUG();
503 }
504
505 /* one flush is safe, as both fields always lie on the same cacheline */
506 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
507 smp_wmb();
508
509 return ts_status;
510}
511
512static void *packet_lookup_frame(const struct packet_sock *po,
513 const struct packet_ring_buffer *rb,
514 unsigned int position,
515 int status)
516{
517 unsigned int pg_vec_pos, frame_offset;
518 union tpacket_uhdr h;
519
520 pg_vec_pos = position / rb->frames_per_block;
521 frame_offset = position % rb->frames_per_block;
522
523 h.raw = rb->pg_vec[pg_vec_pos].buffer +
524 (frame_offset * rb->frame_size);
525
526 if (status != __packet_get_status(po, h.raw))
527 return NULL;
528
529 return h.raw;
530}
531
532static void *packet_current_frame(struct packet_sock *po,
533 struct packet_ring_buffer *rb,
534 int status)
535{
536 return packet_lookup_frame(po, rb, rb->head, status);
537}
538
539static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
540{
541 del_timer_sync(&pkc->retire_blk_timer);
542}
543
544static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
545 struct sk_buff_head *rb_queue)
546{
547 struct tpacket_kbdq_core *pkc;
548
549 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
550
551 spin_lock_bh(&rb_queue->lock);
552 pkc->delete_blk_timer = 1;
553 spin_unlock_bh(&rb_queue->lock);
554
555 prb_del_retire_blk_timer(pkc);
556}
557
558static void prb_setup_retire_blk_timer(struct packet_sock *po)
559{
560 struct tpacket_kbdq_core *pkc;
561
562 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
563 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
564 0);
565 pkc->retire_blk_timer.expires = jiffies;
566}
567
568static int prb_calc_retire_blk_tmo(struct packet_sock *po,
569 int blk_size_in_bytes)
570{
571 struct net_device *dev;
572 unsigned int mbits, div;
573 struct ethtool_link_ksettings ecmd;
574 int err;
575
576 rtnl_lock();
577 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
578 if (unlikely(!dev)) {
579 rtnl_unlock();
580 return DEFAULT_PRB_RETIRE_TOV;
581 }
582 err = __ethtool_get_link_ksettings(dev, &ecmd);
583 rtnl_unlock();
584 if (err)
585 return DEFAULT_PRB_RETIRE_TOV;
586
587 /* If the link speed is so slow you don't really
588 * need to worry about perf anyways
589 */
590 if (ecmd.base.speed < SPEED_1000 ||
591 ecmd.base.speed == SPEED_UNKNOWN)
592 return DEFAULT_PRB_RETIRE_TOV;
593
594 div = ecmd.base.speed / 1000;
595 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
596
597 if (div)
598 mbits /= div;
599
600 if (div)
601 return mbits + 1;
602 return mbits;
603}
604
605static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
606 union tpacket_req_u *req_u)
607{
608 p1->feature_req_word = req_u->req3.tp_feature_req_word;
609}
610
611static void init_prb_bdqc(struct packet_sock *po,
612 struct packet_ring_buffer *rb,
613 struct pgv *pg_vec,
614 union tpacket_req_u *req_u)
615{
616 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
617 struct tpacket_block_desc *pbd;
618
619 memset(p1, 0x0, sizeof(*p1));
620
621 p1->knxt_seq_num = 1;
622 p1->pkbdq = pg_vec;
623 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
624 p1->pkblk_start = pg_vec[0].buffer;
625 p1->kblk_size = req_u->req3.tp_block_size;
626 p1->knum_blocks = req_u->req3.tp_block_nr;
627 p1->hdrlen = po->tp_hdrlen;
628 p1->version = po->tp_version;
629 p1->last_kactive_blk_num = 0;
630 po->stats.stats3.tp_freeze_q_cnt = 0;
631 if (req_u->req3.tp_retire_blk_tov)
632 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
633 else
634 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
635 req_u->req3.tp_block_size);
636 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
637 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
638 rwlock_init(&p1->blk_fill_in_prog_lock);
639
640 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
641 prb_init_ft_ops(p1, req_u);
642 prb_setup_retire_blk_timer(po);
643 prb_open_block(p1, pbd);
644}
645
646/* Do NOT update the last_blk_num first.
647 * Assumes sk_buff_head lock is held.
648 */
649static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
650{
651 mod_timer(&pkc->retire_blk_timer,
652 jiffies + pkc->tov_in_jiffies);
653 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
654}
655
656/*
657 * Timer logic:
658 * 1) We refresh the timer only when we open a block.
659 * By doing this we don't waste cycles refreshing the timer
660 * on packet-by-packet basis.
661 *
662 * With a 1MB block-size, on a 1Gbps line, it will take
663 * i) ~8 ms to fill a block + ii) memcpy etc.
664 * In this cut we are not accounting for the memcpy time.
665 *
666 * So, if the user sets the 'tmo' to 10ms then the timer
667 * will never fire while the block is still getting filled
668 * (which is what we want). However, the user could choose
669 * to close a block early and that's fine.
670 *
671 * But when the timer does fire, we check whether or not to refresh it.
672 * Since the tmo granularity is in msecs, it is not too expensive
673 * to refresh the timer, lets say every '8' msecs.
674 * Either the user can set the 'tmo' or we can derive it based on
675 * a) line-speed and b) block-size.
676 * prb_calc_retire_blk_tmo() calculates the tmo.
677 *
678 */
679static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
680{
681 struct packet_sock *po =
682 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
683 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
684 unsigned int frozen;
685 struct tpacket_block_desc *pbd;
686
687 spin_lock(&po->sk.sk_receive_queue.lock);
688
689 frozen = prb_queue_frozen(pkc);
690 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
691
692 if (unlikely(pkc->delete_blk_timer))
693 goto out;
694
695 /* We only need to plug the race when the block is partially filled.
696 * tpacket_rcv:
697 * lock(); increment BLOCK_NUM_PKTS; unlock()
698 * copy_bits() is in progress ...
699 * timer fires on other cpu:
700 * we can't retire the current block because copy_bits
701 * is in progress.
702 *
703 */
704 if (BLOCK_NUM_PKTS(pbd)) {
705 /* Waiting for skb_copy_bits to finish... */
706 write_lock(&pkc->blk_fill_in_prog_lock);
707 write_unlock(&pkc->blk_fill_in_prog_lock);
708 }
709
710 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
711 if (!frozen) {
712 if (!BLOCK_NUM_PKTS(pbd)) {
713 /* An empty block. Just refresh the timer. */
714 goto refresh_timer;
715 }
716 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
717 if (!prb_dispatch_next_block(pkc, po))
718 goto refresh_timer;
719 else
720 goto out;
721 } else {
722 /* Case 1. Queue was frozen because user-space was
723 * lagging behind.
724 */
725 if (prb_curr_blk_in_use(pbd)) {
726 /*
727 * Ok, user-space is still behind.
728 * So just refresh the timer.
729 */
730 goto refresh_timer;
731 } else {
732 /* Case 2. queue was frozen,user-space caught up,
733 * now the link went idle && the timer fired.
734 * We don't have a block to close.So we open this
735 * block and restart the timer.
736 * opening a block thaws the queue,restarts timer
737 * Thawing/timer-refresh is a side effect.
738 */
739 prb_open_block(pkc, pbd);
740 goto out;
741 }
742 }
743 }
744
745refresh_timer:
746 _prb_refresh_rx_retire_blk_timer(pkc);
747
748out:
749 spin_unlock(&po->sk.sk_receive_queue.lock);
750}
751
752static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
753 struct tpacket_block_desc *pbd1, __u32 status)
754{
755 /* Flush everything minus the block header */
756
757#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
758 u8 *start, *end;
759
760 start = (u8 *)pbd1;
761
762 /* Skip the block header(we know header WILL fit in 4K) */
763 start += PAGE_SIZE;
764
765 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
766 for (; start < end; start += PAGE_SIZE)
767 flush_dcache_page(pgv_to_page(start));
768
769 smp_wmb();
770#endif
771
772 /* Now update the block status. */
773
774 BLOCK_STATUS(pbd1) = status;
775
776 /* Flush the block header */
777
778#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
779 start = (u8 *)pbd1;
780 flush_dcache_page(pgv_to_page(start));
781
782 smp_wmb();
783#endif
784}
785
786/*
787 * Side effect:
788 *
789 * 1) flush the block
790 * 2) Increment active_blk_num
791 *
792 * Note:We DONT refresh the timer on purpose.
793 * Because almost always the next block will be opened.
794 */
795static void prb_close_block(struct tpacket_kbdq_core *pkc1,
796 struct tpacket_block_desc *pbd1,
797 struct packet_sock *po, unsigned int stat)
798{
799 __u32 status = TP_STATUS_USER | stat;
800
801 struct tpacket3_hdr *last_pkt;
802 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
803 struct sock *sk = &po->sk;
804
805 if (atomic_read(&po->tp_drops))
806 status |= TP_STATUS_LOSING;
807
808 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
809 last_pkt->tp_next_offset = 0;
810
811 /* Get the ts of the last pkt */
812 if (BLOCK_NUM_PKTS(pbd1)) {
813 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
814 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
815 } else {
816 /* Ok, we tmo'd - so get the current time.
817 *
818 * It shouldn't really happen as we don't close empty
819 * blocks. See prb_retire_rx_blk_timer_expired().
820 */
821 struct timespec64 ts;
822 ktime_get_real_ts64(&ts);
823 h1->ts_last_pkt.ts_sec = ts.tv_sec;
824 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
825 }
826
827 smp_wmb();
828
829 /* Flush the block */
830 prb_flush_block(pkc1, pbd1, status);
831
832 sk->sk_data_ready(sk);
833
834 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
835}
836
837static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
838{
839 pkc->reset_pending_on_curr_blk = 0;
840}
841
842/*
843 * Side effect of opening a block:
844 *
845 * 1) prb_queue is thawed.
846 * 2) retire_blk_timer is refreshed.
847 *
848 */
849static void prb_open_block(struct tpacket_kbdq_core *pkc1,
850 struct tpacket_block_desc *pbd1)
851{
852 struct timespec64 ts;
853 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
854
855 smp_rmb();
856
857 /* We could have just memset this but we will lose the
858 * flexibility of making the priv area sticky
859 */
860
861 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
862 BLOCK_NUM_PKTS(pbd1) = 0;
863 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
864
865 ktime_get_real_ts64(&ts);
866
867 h1->ts_first_pkt.ts_sec = ts.tv_sec;
868 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
869
870 pkc1->pkblk_start = (char *)pbd1;
871 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
872
873 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
874 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
875
876 pbd1->version = pkc1->version;
877 pkc1->prev = pkc1->nxt_offset;
878 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
879
880 prb_thaw_queue(pkc1);
881 _prb_refresh_rx_retire_blk_timer(pkc1);
882
883 smp_wmb();
884}
885
886/*
887 * Queue freeze logic:
888 * 1) Assume tp_block_nr = 8 blocks.
889 * 2) At time 't0', user opens Rx ring.
890 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
891 * 4) user-space is either sleeping or processing block '0'.
892 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
893 * it will close block-7,loop around and try to fill block '0'.
894 * call-flow:
895 * __packet_lookup_frame_in_block
896 * prb_retire_current_block()
897 * prb_dispatch_next_block()
898 * |->(BLOCK_STATUS == USER) evaluates to true
899 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
900 * 6) Now there are two cases:
901 * 6.1) Link goes idle right after the queue is frozen.
902 * But remember, the last open_block() refreshed the timer.
903 * When this timer expires,it will refresh itself so that we can
904 * re-open block-0 in near future.
905 * 6.2) Link is busy and keeps on receiving packets. This is a simple
906 * case and __packet_lookup_frame_in_block will check if block-0
907 * is free and can now be re-used.
908 */
909static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
910 struct packet_sock *po)
911{
912 pkc->reset_pending_on_curr_blk = 1;
913 po->stats.stats3.tp_freeze_q_cnt++;
914}
915
916#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
917
918/*
919 * If the next block is free then we will dispatch it
920 * and return a good offset.
921 * Else, we will freeze the queue.
922 * So, caller must check the return value.
923 */
924static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
925 struct packet_sock *po)
926{
927 struct tpacket_block_desc *pbd;
928
929 smp_rmb();
930
931 /* 1. Get current block num */
932 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
933
934 /* 2. If this block is currently in_use then freeze the queue */
935 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
936 prb_freeze_queue(pkc, po);
937 return NULL;
938 }
939
940 /*
941 * 3.
942 * open this block and return the offset where the first packet
943 * needs to get stored.
944 */
945 prb_open_block(pkc, pbd);
946 return (void *)pkc->nxt_offset;
947}
948
949static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
950 struct packet_sock *po, unsigned int status)
951{
952 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
953
954 /* retire/close the current block */
955 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
956 /*
957 * Plug the case where copy_bits() is in progress on
958 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
959 * have space to copy the pkt in the current block and
960 * called prb_retire_current_block()
961 *
962 * We don't need to worry about the TMO case because
963 * the timer-handler already handled this case.
964 */
965 if (!(status & TP_STATUS_BLK_TMO)) {
966 /* Waiting for skb_copy_bits to finish... */
967 write_lock(&pkc->blk_fill_in_prog_lock);
968 write_unlock(&pkc->blk_fill_in_prog_lock);
969 }
970 prb_close_block(pkc, pbd, po, status);
971 return;
972 }
973}
974
975static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
976{
977 return TP_STATUS_USER & BLOCK_STATUS(pbd);
978}
979
980static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
981{
982 return pkc->reset_pending_on_curr_blk;
983}
984
985static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
986 __releases(&pkc->blk_fill_in_prog_lock)
987{
988 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
989
990 read_unlock(&pkc->blk_fill_in_prog_lock);
991}
992
993static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
994 struct tpacket3_hdr *ppd)
995{
996 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
997}
998
999static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1000 struct tpacket3_hdr *ppd)
1001{
1002 ppd->hv1.tp_rxhash = 0;
1003}
1004
1005static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1006 struct tpacket3_hdr *ppd)
1007{
1008 if (skb_vlan_tag_present(pkc->skb)) {
1009 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1010 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1011 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1012 } else {
1013 ppd->hv1.tp_vlan_tci = 0;
1014 ppd->hv1.tp_vlan_tpid = 0;
1015 ppd->tp_status = TP_STATUS_AVAILABLE;
1016 }
1017}
1018
1019static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1020 struct tpacket3_hdr *ppd)
1021{
1022 ppd->hv1.tp_padding = 0;
1023 prb_fill_vlan_info(pkc, ppd);
1024
1025 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1026 prb_fill_rxhash(pkc, ppd);
1027 else
1028 prb_clear_rxhash(pkc, ppd);
1029}
1030
1031static void prb_fill_curr_block(char *curr,
1032 struct tpacket_kbdq_core *pkc,
1033 struct tpacket_block_desc *pbd,
1034 unsigned int len)
1035 __acquires(&pkc->blk_fill_in_prog_lock)
1036{
1037 struct tpacket3_hdr *ppd;
1038
1039 ppd = (struct tpacket3_hdr *)curr;
1040 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1041 pkc->prev = curr;
1042 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1043 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1044 BLOCK_NUM_PKTS(pbd) += 1;
1045 read_lock(&pkc->blk_fill_in_prog_lock);
1046 prb_run_all_ft_ops(pkc, ppd);
1047}
1048
1049/* Assumes caller has the sk->rx_queue.lock */
1050static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1051 struct sk_buff *skb,
1052 unsigned int len
1053 )
1054{
1055 struct tpacket_kbdq_core *pkc;
1056 struct tpacket_block_desc *pbd;
1057 char *curr, *end;
1058
1059 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1060 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1061
1062 /* Queue is frozen when user space is lagging behind */
1063 if (prb_queue_frozen(pkc)) {
1064 /*
1065 * Check if that last block which caused the queue to freeze,
1066 * is still in_use by user-space.
1067 */
1068 if (prb_curr_blk_in_use(pbd)) {
1069 /* Can't record this packet */
1070 return NULL;
1071 } else {
1072 /*
1073 * Ok, the block was released by user-space.
1074 * Now let's open that block.
1075 * opening a block also thaws the queue.
1076 * Thawing is a side effect.
1077 */
1078 prb_open_block(pkc, pbd);
1079 }
1080 }
1081
1082 smp_mb();
1083 curr = pkc->nxt_offset;
1084 pkc->skb = skb;
1085 end = (char *)pbd + pkc->kblk_size;
1086
1087 /* first try the current block */
1088 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1089 prb_fill_curr_block(curr, pkc, pbd, len);
1090 return (void *)curr;
1091 }
1092
1093 /* Ok, close the current block */
1094 prb_retire_current_block(pkc, po, 0);
1095
1096 /* Now, try to dispatch the next block */
1097 curr = (char *)prb_dispatch_next_block(pkc, po);
1098 if (curr) {
1099 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1100 prb_fill_curr_block(curr, pkc, pbd, len);
1101 return (void *)curr;
1102 }
1103
1104 /*
1105 * No free blocks are available.user_space hasn't caught up yet.
1106 * Queue was just frozen and now this packet will get dropped.
1107 */
1108 return NULL;
1109}
1110
1111static void *packet_current_rx_frame(struct packet_sock *po,
1112 struct sk_buff *skb,
1113 int status, unsigned int len)
1114{
1115 char *curr = NULL;
1116 switch (po->tp_version) {
1117 case TPACKET_V1:
1118 case TPACKET_V2:
1119 curr = packet_lookup_frame(po, &po->rx_ring,
1120 po->rx_ring.head, status);
1121 return curr;
1122 case TPACKET_V3:
1123 return __packet_lookup_frame_in_block(po, skb, len);
1124 default:
1125 WARN(1, "TPACKET version not supported\n");
1126 BUG();
1127 return NULL;
1128 }
1129}
1130
1131static void *prb_lookup_block(const struct packet_sock *po,
1132 const struct packet_ring_buffer *rb,
1133 unsigned int idx,
1134 int status)
1135{
1136 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1137 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1138
1139 if (status != BLOCK_STATUS(pbd))
1140 return NULL;
1141 return pbd;
1142}
1143
1144static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1145{
1146 unsigned int prev;
1147 if (rb->prb_bdqc.kactive_blk_num)
1148 prev = rb->prb_bdqc.kactive_blk_num-1;
1149 else
1150 prev = rb->prb_bdqc.knum_blocks-1;
1151 return prev;
1152}
1153
1154/* Assumes caller has held the rx_queue.lock */
1155static void *__prb_previous_block(struct packet_sock *po,
1156 struct packet_ring_buffer *rb,
1157 int status)
1158{
1159 unsigned int previous = prb_previous_blk_num(rb);
1160 return prb_lookup_block(po, rb, previous, status);
1161}
1162
1163static void *packet_previous_rx_frame(struct packet_sock *po,
1164 struct packet_ring_buffer *rb,
1165 int status)
1166{
1167 if (po->tp_version <= TPACKET_V2)
1168 return packet_previous_frame(po, rb, status);
1169
1170 return __prb_previous_block(po, rb, status);
1171}
1172
1173static void packet_increment_rx_head(struct packet_sock *po,
1174 struct packet_ring_buffer *rb)
1175{
1176 switch (po->tp_version) {
1177 case TPACKET_V1:
1178 case TPACKET_V2:
1179 return packet_increment_head(rb);
1180 case TPACKET_V3:
1181 default:
1182 WARN(1, "TPACKET version not supported.\n");
1183 BUG();
1184 return;
1185 }
1186}
1187
1188static void *packet_previous_frame(struct packet_sock *po,
1189 struct packet_ring_buffer *rb,
1190 int status)
1191{
1192 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1193 return packet_lookup_frame(po, rb, previous, status);
1194}
1195
1196static void packet_increment_head(struct packet_ring_buffer *buff)
1197{
1198 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1199}
1200
1201static void packet_inc_pending(struct packet_ring_buffer *rb)
1202{
1203 this_cpu_inc(*rb->pending_refcnt);
1204}
1205
1206static void packet_dec_pending(struct packet_ring_buffer *rb)
1207{
1208 this_cpu_dec(*rb->pending_refcnt);
1209}
1210
1211static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1212{
1213 unsigned int refcnt = 0;
1214 int cpu;
1215
1216 /* We don't use pending refcount in rx_ring. */
1217 if (rb->pending_refcnt == NULL)
1218 return 0;
1219
1220 for_each_possible_cpu(cpu)
1221 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1222
1223 return refcnt;
1224}
1225
1226static int packet_alloc_pending(struct packet_sock *po)
1227{
1228 po->rx_ring.pending_refcnt = NULL;
1229
1230 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1231 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1232 return -ENOBUFS;
1233
1234 return 0;
1235}
1236
1237static void packet_free_pending(struct packet_sock *po)
1238{
1239 free_percpu(po->tx_ring.pending_refcnt);
1240}
1241
1242#define ROOM_POW_OFF 2
1243#define ROOM_NONE 0x0
1244#define ROOM_LOW 0x1
1245#define ROOM_NORMAL 0x2
1246
1247static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1248{
1249 int idx, len;
1250
1251 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1252 idx = READ_ONCE(po->rx_ring.head);
1253 if (pow_off)
1254 idx += len >> pow_off;
1255 if (idx >= len)
1256 idx -= len;
1257 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1258}
1259
1260static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1261{
1262 int idx, len;
1263
1264 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1265 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1266 if (pow_off)
1267 idx += len >> pow_off;
1268 if (idx >= len)
1269 idx -= len;
1270 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1271}
1272
1273static int __packet_rcv_has_room(const struct packet_sock *po,
1274 const struct sk_buff *skb)
1275{
1276 const struct sock *sk = &po->sk;
1277 int ret = ROOM_NONE;
1278
1279 if (po->prot_hook.func != tpacket_rcv) {
1280 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1281 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1282 - (skb ? skb->truesize : 0);
1283
1284 if (avail > (rcvbuf >> ROOM_POW_OFF))
1285 return ROOM_NORMAL;
1286 else if (avail > 0)
1287 return ROOM_LOW;
1288 else
1289 return ROOM_NONE;
1290 }
1291
1292 if (po->tp_version == TPACKET_V3) {
1293 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1294 ret = ROOM_NORMAL;
1295 else if (__tpacket_v3_has_room(po, 0))
1296 ret = ROOM_LOW;
1297 } else {
1298 if (__tpacket_has_room(po, ROOM_POW_OFF))
1299 ret = ROOM_NORMAL;
1300 else if (__tpacket_has_room(po, 0))
1301 ret = ROOM_LOW;
1302 }
1303
1304 return ret;
1305}
1306
1307static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1308{
1309 int pressure, ret;
1310
1311 ret = __packet_rcv_has_room(po, skb);
1312 pressure = ret != ROOM_NORMAL;
1313
1314 if (READ_ONCE(po->pressure) != pressure)
1315 WRITE_ONCE(po->pressure, pressure);
1316
1317 return ret;
1318}
1319
1320static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1321{
1322 if (READ_ONCE(po->pressure) &&
1323 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1324 WRITE_ONCE(po->pressure, 0);
1325}
1326
1327static void packet_sock_destruct(struct sock *sk)
1328{
1329 skb_queue_purge(&sk->sk_error_queue);
1330
1331 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1332 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1333
1334 if (!sock_flag(sk, SOCK_DEAD)) {
1335 pr_err("Attempt to release alive packet socket: %p\n", sk);
1336 return;
1337 }
1338
1339 sk_refcnt_debug_dec(sk);
1340}
1341
1342static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1343{
1344 u32 *history = po->rollover->history;
1345 u32 victim, rxhash;
1346 int i, count = 0;
1347
1348 rxhash = skb_get_hash(skb);
1349 for (i = 0; i < ROLLOVER_HLEN; i++)
1350 if (READ_ONCE(history[i]) == rxhash)
1351 count++;
1352
1353 victim = get_random_u32_below(ROLLOVER_HLEN);
1354
1355 /* Avoid dirtying the cache line if possible */
1356 if (READ_ONCE(history[victim]) != rxhash)
1357 WRITE_ONCE(history[victim], rxhash);
1358
1359 return count > (ROLLOVER_HLEN >> 1);
1360}
1361
1362static unsigned int fanout_demux_hash(struct packet_fanout *f,
1363 struct sk_buff *skb,
1364 unsigned int num)
1365{
1366 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1367}
1368
1369static unsigned int fanout_demux_lb(struct packet_fanout *f,
1370 struct sk_buff *skb,
1371 unsigned int num)
1372{
1373 unsigned int val = atomic_inc_return(&f->rr_cur);
1374
1375 return val % num;
1376}
1377
1378static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1379 struct sk_buff *skb,
1380 unsigned int num)
1381{
1382 return smp_processor_id() % num;
1383}
1384
1385static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1386 struct sk_buff *skb,
1387 unsigned int num)
1388{
1389 return get_random_u32_below(num);
1390}
1391
1392static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1393 struct sk_buff *skb,
1394 unsigned int idx, bool try_self,
1395 unsigned int num)
1396{
1397 struct packet_sock *po, *po_next, *po_skip = NULL;
1398 unsigned int i, j, room = ROOM_NONE;
1399
1400 po = pkt_sk(rcu_dereference(f->arr[idx]));
1401
1402 if (try_self) {
1403 room = packet_rcv_has_room(po, skb);
1404 if (room == ROOM_NORMAL ||
1405 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1406 return idx;
1407 po_skip = po;
1408 }
1409
1410 i = j = min_t(int, po->rollover->sock, num - 1);
1411 do {
1412 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1413 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1414 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1415 if (i != j)
1416 po->rollover->sock = i;
1417 atomic_long_inc(&po->rollover->num);
1418 if (room == ROOM_LOW)
1419 atomic_long_inc(&po->rollover->num_huge);
1420 return i;
1421 }
1422
1423 if (++i == num)
1424 i = 0;
1425 } while (i != j);
1426
1427 atomic_long_inc(&po->rollover->num_failed);
1428 return idx;
1429}
1430
1431static unsigned int fanout_demux_qm(struct packet_fanout *f,
1432 struct sk_buff *skb,
1433 unsigned int num)
1434{
1435 return skb_get_queue_mapping(skb) % num;
1436}
1437
1438static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1439 struct sk_buff *skb,
1440 unsigned int num)
1441{
1442 struct bpf_prog *prog;
1443 unsigned int ret = 0;
1444
1445 rcu_read_lock();
1446 prog = rcu_dereference(f->bpf_prog);
1447 if (prog)
1448 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1449 rcu_read_unlock();
1450
1451 return ret;
1452}
1453
1454static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1455{
1456 return f->flags & (flag >> 8);
1457}
1458
1459static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1460 struct packet_type *pt, struct net_device *orig_dev)
1461{
1462 struct packet_fanout *f = pt->af_packet_priv;
1463 unsigned int num = READ_ONCE(f->num_members);
1464 struct net *net = read_pnet(&f->net);
1465 struct packet_sock *po;
1466 unsigned int idx;
1467
1468 if (!net_eq(dev_net(dev), net) || !num) {
1469 kfree_skb(skb);
1470 return 0;
1471 }
1472
1473 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1474 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1475 if (!skb)
1476 return 0;
1477 }
1478 switch (f->type) {
1479 case PACKET_FANOUT_HASH:
1480 default:
1481 idx = fanout_demux_hash(f, skb, num);
1482 break;
1483 case PACKET_FANOUT_LB:
1484 idx = fanout_demux_lb(f, skb, num);
1485 break;
1486 case PACKET_FANOUT_CPU:
1487 idx = fanout_demux_cpu(f, skb, num);
1488 break;
1489 case PACKET_FANOUT_RND:
1490 idx = fanout_demux_rnd(f, skb, num);
1491 break;
1492 case PACKET_FANOUT_QM:
1493 idx = fanout_demux_qm(f, skb, num);
1494 break;
1495 case PACKET_FANOUT_ROLLOVER:
1496 idx = fanout_demux_rollover(f, skb, 0, false, num);
1497 break;
1498 case PACKET_FANOUT_CBPF:
1499 case PACKET_FANOUT_EBPF:
1500 idx = fanout_demux_bpf(f, skb, num);
1501 break;
1502 }
1503
1504 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1505 idx = fanout_demux_rollover(f, skb, idx, true, num);
1506
1507 po = pkt_sk(rcu_dereference(f->arr[idx]));
1508 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1509}
1510
1511DEFINE_MUTEX(fanout_mutex);
1512EXPORT_SYMBOL_GPL(fanout_mutex);
1513static LIST_HEAD(fanout_list);
1514static u16 fanout_next_id;
1515
1516static void __fanout_link(struct sock *sk, struct packet_sock *po)
1517{
1518 struct packet_fanout *f = po->fanout;
1519
1520 spin_lock(&f->lock);
1521 rcu_assign_pointer(f->arr[f->num_members], sk);
1522 smp_wmb();
1523 f->num_members++;
1524 if (f->num_members == 1)
1525 dev_add_pack(&f->prot_hook);
1526 spin_unlock(&f->lock);
1527}
1528
1529static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1530{
1531 struct packet_fanout *f = po->fanout;
1532 int i;
1533
1534 spin_lock(&f->lock);
1535 for (i = 0; i < f->num_members; i++) {
1536 if (rcu_dereference_protected(f->arr[i],
1537 lockdep_is_held(&f->lock)) == sk)
1538 break;
1539 }
1540 BUG_ON(i >= f->num_members);
1541 rcu_assign_pointer(f->arr[i],
1542 rcu_dereference_protected(f->arr[f->num_members - 1],
1543 lockdep_is_held(&f->lock)));
1544 f->num_members--;
1545 if (f->num_members == 0)
1546 __dev_remove_pack(&f->prot_hook);
1547 spin_unlock(&f->lock);
1548}
1549
1550static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1551{
1552 if (sk->sk_family != PF_PACKET)
1553 return false;
1554
1555 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1556}
1557
1558static void fanout_init_data(struct packet_fanout *f)
1559{
1560 switch (f->type) {
1561 case PACKET_FANOUT_LB:
1562 atomic_set(&f->rr_cur, 0);
1563 break;
1564 case PACKET_FANOUT_CBPF:
1565 case PACKET_FANOUT_EBPF:
1566 RCU_INIT_POINTER(f->bpf_prog, NULL);
1567 break;
1568 }
1569}
1570
1571static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1572{
1573 struct bpf_prog *old;
1574
1575 spin_lock(&f->lock);
1576 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1577 rcu_assign_pointer(f->bpf_prog, new);
1578 spin_unlock(&f->lock);
1579
1580 if (old) {
1581 synchronize_net();
1582 bpf_prog_destroy(old);
1583 }
1584}
1585
1586static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1587 unsigned int len)
1588{
1589 struct bpf_prog *new;
1590 struct sock_fprog fprog;
1591 int ret;
1592
1593 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1594 return -EPERM;
1595
1596 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1597 if (ret)
1598 return ret;
1599
1600 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1601 if (ret)
1602 return ret;
1603
1604 __fanout_set_data_bpf(po->fanout, new);
1605 return 0;
1606}
1607
1608static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1609 unsigned int len)
1610{
1611 struct bpf_prog *new;
1612 u32 fd;
1613
1614 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1615 return -EPERM;
1616 if (len != sizeof(fd))
1617 return -EINVAL;
1618 if (copy_from_sockptr(&fd, data, len))
1619 return -EFAULT;
1620
1621 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1622 if (IS_ERR(new))
1623 return PTR_ERR(new);
1624
1625 __fanout_set_data_bpf(po->fanout, new);
1626 return 0;
1627}
1628
1629static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1630 unsigned int len)
1631{
1632 switch (po->fanout->type) {
1633 case PACKET_FANOUT_CBPF:
1634 return fanout_set_data_cbpf(po, data, len);
1635 case PACKET_FANOUT_EBPF:
1636 return fanout_set_data_ebpf(po, data, len);
1637 default:
1638 return -EINVAL;
1639 }
1640}
1641
1642static void fanout_release_data(struct packet_fanout *f)
1643{
1644 switch (f->type) {
1645 case PACKET_FANOUT_CBPF:
1646 case PACKET_FANOUT_EBPF:
1647 __fanout_set_data_bpf(f, NULL);
1648 }
1649}
1650
1651static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1652{
1653 struct packet_fanout *f;
1654
1655 list_for_each_entry(f, &fanout_list, list) {
1656 if (f->id == candidate_id &&
1657 read_pnet(&f->net) == sock_net(sk)) {
1658 return false;
1659 }
1660 }
1661 return true;
1662}
1663
1664static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1665{
1666 u16 id = fanout_next_id;
1667
1668 do {
1669 if (__fanout_id_is_free(sk, id)) {
1670 *new_id = id;
1671 fanout_next_id = id + 1;
1672 return true;
1673 }
1674
1675 id++;
1676 } while (id != fanout_next_id);
1677
1678 return false;
1679}
1680
1681static int fanout_add(struct sock *sk, struct fanout_args *args)
1682{
1683 struct packet_rollover *rollover = NULL;
1684 struct packet_sock *po = pkt_sk(sk);
1685 u16 type_flags = args->type_flags;
1686 struct packet_fanout *f, *match;
1687 u8 type = type_flags & 0xff;
1688 u8 flags = type_flags >> 8;
1689 u16 id = args->id;
1690 int err;
1691
1692 switch (type) {
1693 case PACKET_FANOUT_ROLLOVER:
1694 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1695 return -EINVAL;
1696 break;
1697 case PACKET_FANOUT_HASH:
1698 case PACKET_FANOUT_LB:
1699 case PACKET_FANOUT_CPU:
1700 case PACKET_FANOUT_RND:
1701 case PACKET_FANOUT_QM:
1702 case PACKET_FANOUT_CBPF:
1703 case PACKET_FANOUT_EBPF:
1704 break;
1705 default:
1706 return -EINVAL;
1707 }
1708
1709 mutex_lock(&fanout_mutex);
1710
1711 err = -EALREADY;
1712 if (po->fanout)
1713 goto out;
1714
1715 if (type == PACKET_FANOUT_ROLLOVER ||
1716 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1717 err = -ENOMEM;
1718 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1719 if (!rollover)
1720 goto out;
1721 atomic_long_set(&rollover->num, 0);
1722 atomic_long_set(&rollover->num_huge, 0);
1723 atomic_long_set(&rollover->num_failed, 0);
1724 }
1725
1726 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1727 if (id != 0) {
1728 err = -EINVAL;
1729 goto out;
1730 }
1731 if (!fanout_find_new_id(sk, &id)) {
1732 err = -ENOMEM;
1733 goto out;
1734 }
1735 /* ephemeral flag for the first socket in the group: drop it */
1736 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1737 }
1738
1739 match = NULL;
1740 list_for_each_entry(f, &fanout_list, list) {
1741 if (f->id == id &&
1742 read_pnet(&f->net) == sock_net(sk)) {
1743 match = f;
1744 break;
1745 }
1746 }
1747 err = -EINVAL;
1748 if (match) {
1749 if (match->flags != flags)
1750 goto out;
1751 if (args->max_num_members &&
1752 args->max_num_members != match->max_num_members)
1753 goto out;
1754 } else {
1755 if (args->max_num_members > PACKET_FANOUT_MAX)
1756 goto out;
1757 if (!args->max_num_members)
1758 /* legacy PACKET_FANOUT_MAX */
1759 args->max_num_members = 256;
1760 err = -ENOMEM;
1761 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1762 GFP_KERNEL);
1763 if (!match)
1764 goto out;
1765 write_pnet(&match->net, sock_net(sk));
1766 match->id = id;
1767 match->type = type;
1768 match->flags = flags;
1769 INIT_LIST_HEAD(&match->list);
1770 spin_lock_init(&match->lock);
1771 refcount_set(&match->sk_ref, 0);
1772 fanout_init_data(match);
1773 match->prot_hook.type = po->prot_hook.type;
1774 match->prot_hook.dev = po->prot_hook.dev;
1775 match->prot_hook.func = packet_rcv_fanout;
1776 match->prot_hook.af_packet_priv = match;
1777 match->prot_hook.af_packet_net = read_pnet(&match->net);
1778 match->prot_hook.id_match = match_fanout_group;
1779 match->max_num_members = args->max_num_members;
1780 match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1781 list_add(&match->list, &fanout_list);
1782 }
1783 err = -EINVAL;
1784
1785 spin_lock(&po->bind_lock);
1786 if (po->running &&
1787 match->type == type &&
1788 match->prot_hook.type == po->prot_hook.type &&
1789 match->prot_hook.dev == po->prot_hook.dev) {
1790 err = -ENOSPC;
1791 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1792 __dev_remove_pack(&po->prot_hook);
1793
1794 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1795 WRITE_ONCE(po->fanout, match);
1796
1797 po->rollover = rollover;
1798 rollover = NULL;
1799 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1800 __fanout_link(sk, po);
1801 err = 0;
1802 }
1803 }
1804 spin_unlock(&po->bind_lock);
1805
1806 if (err && !refcount_read(&match->sk_ref)) {
1807 list_del(&match->list);
1808 kvfree(match);
1809 }
1810
1811out:
1812 kfree(rollover);
1813 mutex_unlock(&fanout_mutex);
1814 return err;
1815}
1816
1817/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1818 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1819 * It is the responsibility of the caller to call fanout_release_data() and
1820 * free the returned packet_fanout (after synchronize_net())
1821 */
1822static struct packet_fanout *fanout_release(struct sock *sk)
1823{
1824 struct packet_sock *po = pkt_sk(sk);
1825 struct packet_fanout *f;
1826
1827 mutex_lock(&fanout_mutex);
1828 f = po->fanout;
1829 if (f) {
1830 po->fanout = NULL;
1831
1832 if (refcount_dec_and_test(&f->sk_ref))
1833 list_del(&f->list);
1834 else
1835 f = NULL;
1836 }
1837 mutex_unlock(&fanout_mutex);
1838
1839 return f;
1840}
1841
1842static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1843 struct sk_buff *skb)
1844{
1845 /* Earlier code assumed this would be a VLAN pkt, double-check
1846 * this now that we have the actual packet in hand. We can only
1847 * do this check on Ethernet devices.
1848 */
1849 if (unlikely(dev->type != ARPHRD_ETHER))
1850 return false;
1851
1852 skb_reset_mac_header(skb);
1853 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1854}
1855
1856static const struct proto_ops packet_ops;
1857
1858static const struct proto_ops packet_ops_spkt;
1859
1860static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1861 struct packet_type *pt, struct net_device *orig_dev)
1862{
1863 struct sock *sk;
1864 struct sockaddr_pkt *spkt;
1865
1866 /*
1867 * When we registered the protocol we saved the socket in the data
1868 * field for just this event.
1869 */
1870
1871 sk = pt->af_packet_priv;
1872
1873 /*
1874 * Yank back the headers [hope the device set this
1875 * right or kerboom...]
1876 *
1877 * Incoming packets have ll header pulled,
1878 * push it back.
1879 *
1880 * For outgoing ones skb->data == skb_mac_header(skb)
1881 * so that this procedure is noop.
1882 */
1883
1884 if (skb->pkt_type == PACKET_LOOPBACK)
1885 goto out;
1886
1887 if (!net_eq(dev_net(dev), sock_net(sk)))
1888 goto out;
1889
1890 skb = skb_share_check(skb, GFP_ATOMIC);
1891 if (skb == NULL)
1892 goto oom;
1893
1894 /* drop any routing info */
1895 skb_dst_drop(skb);
1896
1897 /* drop conntrack reference */
1898 nf_reset_ct(skb);
1899
1900 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1901
1902 skb_push(skb, skb->data - skb_mac_header(skb));
1903
1904 /*
1905 * The SOCK_PACKET socket receives _all_ frames.
1906 */
1907
1908 spkt->spkt_family = dev->type;
1909 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1910 spkt->spkt_protocol = skb->protocol;
1911
1912 /*
1913 * Charge the memory to the socket. This is done specifically
1914 * to prevent sockets using all the memory up.
1915 */
1916
1917 if (sock_queue_rcv_skb(sk, skb) == 0)
1918 return 0;
1919
1920out:
1921 kfree_skb(skb);
1922oom:
1923 return 0;
1924}
1925
1926static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1927{
1928 int depth;
1929
1930 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1931 sock->type == SOCK_RAW) {
1932 skb_reset_mac_header(skb);
1933 skb->protocol = dev_parse_header_protocol(skb);
1934 }
1935
1936 /* Move network header to the right position for VLAN tagged packets */
1937 if (likely(skb->dev->type == ARPHRD_ETHER) &&
1938 eth_type_vlan(skb->protocol) &&
1939 __vlan_get_protocol(skb, skb->protocol, &depth) != 0) {
1940 if (pskb_may_pull(skb, depth))
1941 skb_set_network_header(skb, depth);
1942 }
1943
1944 skb_probe_transport_header(skb);
1945}
1946
1947/*
1948 * Output a raw packet to a device layer. This bypasses all the other
1949 * protocol layers and you must therefore supply it with a complete frame
1950 */
1951
1952static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1953 size_t len)
1954{
1955 struct sock *sk = sock->sk;
1956 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1957 struct sk_buff *skb = NULL;
1958 struct net_device *dev;
1959 struct sockcm_cookie sockc;
1960 __be16 proto = 0;
1961 int err;
1962 int extra_len = 0;
1963
1964 /*
1965 * Get and verify the address.
1966 */
1967
1968 if (saddr) {
1969 if (msg->msg_namelen < sizeof(struct sockaddr))
1970 return -EINVAL;
1971 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1972 proto = saddr->spkt_protocol;
1973 } else
1974 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1975
1976 /*
1977 * Find the device first to size check it
1978 */
1979
1980 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1981retry:
1982 rcu_read_lock();
1983 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1984 err = -ENODEV;
1985 if (dev == NULL)
1986 goto out_unlock;
1987
1988 err = -ENETDOWN;
1989 if (!(dev->flags & IFF_UP))
1990 goto out_unlock;
1991
1992 /*
1993 * You may not queue a frame bigger than the mtu. This is the lowest level
1994 * raw protocol and you must do your own fragmentation at this level.
1995 */
1996
1997 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1998 if (!netif_supports_nofcs(dev)) {
1999 err = -EPROTONOSUPPORT;
2000 goto out_unlock;
2001 }
2002 extra_len = 4; /* We're doing our own CRC */
2003 }
2004
2005 err = -EMSGSIZE;
2006 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2007 goto out_unlock;
2008
2009 if (!skb) {
2010 size_t reserved = LL_RESERVED_SPACE(dev);
2011 int tlen = dev->needed_tailroom;
2012 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2013
2014 rcu_read_unlock();
2015 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2016 if (skb == NULL)
2017 return -ENOBUFS;
2018 /* FIXME: Save some space for broken drivers that write a hard
2019 * header at transmission time by themselves. PPP is the notable
2020 * one here. This should really be fixed at the driver level.
2021 */
2022 skb_reserve(skb, reserved);
2023 skb_reset_network_header(skb);
2024
2025 /* Try to align data part correctly */
2026 if (hhlen) {
2027 skb->data -= hhlen;
2028 skb->tail -= hhlen;
2029 if (len < hhlen)
2030 skb_reset_network_header(skb);
2031 }
2032 err = memcpy_from_msg(skb_put(skb, len), msg, len);
2033 if (err)
2034 goto out_free;
2035 goto retry;
2036 }
2037
2038 if (!dev_validate_header(dev, skb->data, len)) {
2039 err = -EINVAL;
2040 goto out_unlock;
2041 }
2042 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2043 !packet_extra_vlan_len_allowed(dev, skb)) {
2044 err = -EMSGSIZE;
2045 goto out_unlock;
2046 }
2047
2048 sockcm_init(&sockc, sk);
2049 if (msg->msg_controllen) {
2050 err = sock_cmsg_send(sk, msg, &sockc);
2051 if (unlikely(err))
2052 goto out_unlock;
2053 }
2054
2055 skb->protocol = proto;
2056 skb->dev = dev;
2057 skb->priority = sk->sk_priority;
2058 skb->mark = sk->sk_mark;
2059 skb->tstamp = sockc.transmit_time;
2060
2061 skb_setup_tx_timestamp(skb, sockc.tsflags);
2062
2063 if (unlikely(extra_len == 4))
2064 skb->no_fcs = 1;
2065
2066 packet_parse_headers(skb, sock);
2067
2068 dev_queue_xmit(skb);
2069 rcu_read_unlock();
2070 return len;
2071
2072out_unlock:
2073 rcu_read_unlock();
2074out_free:
2075 kfree_skb(skb);
2076 return err;
2077}
2078
2079static unsigned int run_filter(struct sk_buff *skb,
2080 const struct sock *sk,
2081 unsigned int res)
2082{
2083 struct sk_filter *filter;
2084
2085 rcu_read_lock();
2086 filter = rcu_dereference(sk->sk_filter);
2087 if (filter != NULL)
2088 res = bpf_prog_run_clear_cb(filter->prog, skb);
2089 rcu_read_unlock();
2090
2091 return res;
2092}
2093
2094static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2095 size_t *len)
2096{
2097 struct virtio_net_hdr vnet_hdr;
2098
2099 if (*len < sizeof(vnet_hdr))
2100 return -EINVAL;
2101 *len -= sizeof(vnet_hdr);
2102
2103 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2104 return -EINVAL;
2105
2106 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2107}
2108
2109/*
2110 * This function makes lazy skb cloning in hope that most of packets
2111 * are discarded by BPF.
2112 *
2113 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2114 * and skb->cb are mangled. It works because (and until) packets
2115 * falling here are owned by current CPU. Output packets are cloned
2116 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2117 * sequentially, so that if we return skb to original state on exit,
2118 * we will not harm anyone.
2119 */
2120
2121static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2122 struct packet_type *pt, struct net_device *orig_dev)
2123{
2124 struct sock *sk;
2125 struct sockaddr_ll *sll;
2126 struct packet_sock *po;
2127 u8 *skb_head = skb->data;
2128 int skb_len = skb->len;
2129 unsigned int snaplen, res;
2130 bool is_drop_n_account = false;
2131
2132 if (skb->pkt_type == PACKET_LOOPBACK)
2133 goto drop;
2134
2135 sk = pt->af_packet_priv;
2136 po = pkt_sk(sk);
2137
2138 if (!net_eq(dev_net(dev), sock_net(sk)))
2139 goto drop;
2140
2141 skb->dev = dev;
2142
2143 if (dev_has_header(dev)) {
2144 /* The device has an explicit notion of ll header,
2145 * exported to higher levels.
2146 *
2147 * Otherwise, the device hides details of its frame
2148 * structure, so that corresponding packet head is
2149 * never delivered to user.
2150 */
2151 if (sk->sk_type != SOCK_DGRAM)
2152 skb_push(skb, skb->data - skb_mac_header(skb));
2153 else if (skb->pkt_type == PACKET_OUTGOING) {
2154 /* Special case: outgoing packets have ll header at head */
2155 skb_pull(skb, skb_network_offset(skb));
2156 }
2157 }
2158
2159 snaplen = skb->len;
2160
2161 res = run_filter(skb, sk, snaplen);
2162 if (!res)
2163 goto drop_n_restore;
2164 if (snaplen > res)
2165 snaplen = res;
2166
2167 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2168 goto drop_n_acct;
2169
2170 if (skb_shared(skb)) {
2171 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2172 if (nskb == NULL)
2173 goto drop_n_acct;
2174
2175 if (skb_head != skb->data) {
2176 skb->data = skb_head;
2177 skb->len = skb_len;
2178 }
2179 consume_skb(skb);
2180 skb = nskb;
2181 }
2182
2183 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2184
2185 sll = &PACKET_SKB_CB(skb)->sa.ll;
2186 sll->sll_hatype = dev->type;
2187 sll->sll_pkttype = skb->pkt_type;
2188 if (unlikely(po->origdev))
2189 sll->sll_ifindex = orig_dev->ifindex;
2190 else
2191 sll->sll_ifindex = dev->ifindex;
2192
2193 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2194
2195 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2196 * Use their space for storing the original skb length.
2197 */
2198 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2199
2200 if (pskb_trim(skb, snaplen))
2201 goto drop_n_acct;
2202
2203 skb_set_owner_r(skb, sk);
2204 skb->dev = NULL;
2205 skb_dst_drop(skb);
2206
2207 /* drop conntrack reference */
2208 nf_reset_ct(skb);
2209
2210 spin_lock(&sk->sk_receive_queue.lock);
2211 po->stats.stats1.tp_packets++;
2212 sock_skb_set_dropcount(sk, skb);
2213 skb_clear_delivery_time(skb);
2214 __skb_queue_tail(&sk->sk_receive_queue, skb);
2215 spin_unlock(&sk->sk_receive_queue.lock);
2216 sk->sk_data_ready(sk);
2217 return 0;
2218
2219drop_n_acct:
2220 is_drop_n_account = true;
2221 atomic_inc(&po->tp_drops);
2222 atomic_inc(&sk->sk_drops);
2223
2224drop_n_restore:
2225 if (skb_head != skb->data && skb_shared(skb)) {
2226 skb->data = skb_head;
2227 skb->len = skb_len;
2228 }
2229drop:
2230 if (!is_drop_n_account)
2231 consume_skb(skb);
2232 else
2233 kfree_skb(skb);
2234 return 0;
2235}
2236
2237static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2238 struct packet_type *pt, struct net_device *orig_dev)
2239{
2240 struct sock *sk;
2241 struct packet_sock *po;
2242 struct sockaddr_ll *sll;
2243 union tpacket_uhdr h;
2244 u8 *skb_head = skb->data;
2245 int skb_len = skb->len;
2246 unsigned int snaplen, res;
2247 unsigned long status = TP_STATUS_USER;
2248 unsigned short macoff, hdrlen;
2249 unsigned int netoff;
2250 struct sk_buff *copy_skb = NULL;
2251 struct timespec64 ts;
2252 __u32 ts_status;
2253 bool is_drop_n_account = false;
2254 unsigned int slot_id = 0;
2255 bool do_vnet = false;
2256
2257 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2258 * We may add members to them until current aligned size without forcing
2259 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2260 */
2261 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2262 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2263
2264 if (skb->pkt_type == PACKET_LOOPBACK)
2265 goto drop;
2266
2267 sk = pt->af_packet_priv;
2268 po = pkt_sk(sk);
2269
2270 if (!net_eq(dev_net(dev), sock_net(sk)))
2271 goto drop;
2272
2273 if (dev_has_header(dev)) {
2274 if (sk->sk_type != SOCK_DGRAM)
2275 skb_push(skb, skb->data - skb_mac_header(skb));
2276 else if (skb->pkt_type == PACKET_OUTGOING) {
2277 /* Special case: outgoing packets have ll header at head */
2278 skb_pull(skb, skb_network_offset(skb));
2279 }
2280 }
2281
2282 snaplen = skb->len;
2283
2284 res = run_filter(skb, sk, snaplen);
2285 if (!res)
2286 goto drop_n_restore;
2287
2288 /* If we are flooded, just give up */
2289 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2290 atomic_inc(&po->tp_drops);
2291 goto drop_n_restore;
2292 }
2293
2294 if (skb->ip_summed == CHECKSUM_PARTIAL)
2295 status |= TP_STATUS_CSUMNOTREADY;
2296 else if (skb->pkt_type != PACKET_OUTGOING &&
2297 skb_csum_unnecessary(skb))
2298 status |= TP_STATUS_CSUM_VALID;
2299
2300 if (snaplen > res)
2301 snaplen = res;
2302
2303 if (sk->sk_type == SOCK_DGRAM) {
2304 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2305 po->tp_reserve;
2306 } else {
2307 unsigned int maclen = skb_network_offset(skb);
2308 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2309 (maclen < 16 ? 16 : maclen)) +
2310 po->tp_reserve;
2311 if (po->has_vnet_hdr) {
2312 netoff += sizeof(struct virtio_net_hdr);
2313 do_vnet = true;
2314 }
2315 macoff = netoff - maclen;
2316 }
2317 if (netoff > USHRT_MAX) {
2318 atomic_inc(&po->tp_drops);
2319 goto drop_n_restore;
2320 }
2321 if (po->tp_version <= TPACKET_V2) {
2322 if (macoff + snaplen > po->rx_ring.frame_size) {
2323 if (po->copy_thresh &&
2324 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2325 if (skb_shared(skb)) {
2326 copy_skb = skb_clone(skb, GFP_ATOMIC);
2327 } else {
2328 copy_skb = skb_get(skb);
2329 skb_head = skb->data;
2330 }
2331 if (copy_skb) {
2332 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2333 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2334 skb_set_owner_r(copy_skb, sk);
2335 }
2336 }
2337 snaplen = po->rx_ring.frame_size - macoff;
2338 if ((int)snaplen < 0) {
2339 snaplen = 0;
2340 do_vnet = false;
2341 }
2342 }
2343 } else if (unlikely(macoff + snaplen >
2344 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2345 u32 nval;
2346
2347 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2348 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2349 snaplen, nval, macoff);
2350 snaplen = nval;
2351 if (unlikely((int)snaplen < 0)) {
2352 snaplen = 0;
2353 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2354 do_vnet = false;
2355 }
2356 }
2357 spin_lock(&sk->sk_receive_queue.lock);
2358 h.raw = packet_current_rx_frame(po, skb,
2359 TP_STATUS_KERNEL, (macoff+snaplen));
2360 if (!h.raw)
2361 goto drop_n_account;
2362
2363 if (po->tp_version <= TPACKET_V2) {
2364 slot_id = po->rx_ring.head;
2365 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2366 goto drop_n_account;
2367 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2368 }
2369
2370 if (do_vnet &&
2371 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2372 sizeof(struct virtio_net_hdr),
2373 vio_le(), true, 0)) {
2374 if (po->tp_version == TPACKET_V3)
2375 prb_clear_blk_fill_status(&po->rx_ring);
2376 goto drop_n_account;
2377 }
2378
2379 if (po->tp_version <= TPACKET_V2) {
2380 packet_increment_rx_head(po, &po->rx_ring);
2381 /*
2382 * LOSING will be reported till you read the stats,
2383 * because it's COR - Clear On Read.
2384 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2385 * at packet level.
2386 */
2387 if (atomic_read(&po->tp_drops))
2388 status |= TP_STATUS_LOSING;
2389 }
2390
2391 po->stats.stats1.tp_packets++;
2392 if (copy_skb) {
2393 status |= TP_STATUS_COPY;
2394 skb_clear_delivery_time(copy_skb);
2395 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2396 }
2397 spin_unlock(&sk->sk_receive_queue.lock);
2398
2399 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2400
2401 /* Always timestamp; prefer an existing software timestamp taken
2402 * closer to the time of capture.
2403 */
2404 ts_status = tpacket_get_timestamp(skb, &ts,
2405 po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2406 if (!ts_status)
2407 ktime_get_real_ts64(&ts);
2408
2409 status |= ts_status;
2410
2411 switch (po->tp_version) {
2412 case TPACKET_V1:
2413 h.h1->tp_len = skb->len;
2414 h.h1->tp_snaplen = snaplen;
2415 h.h1->tp_mac = macoff;
2416 h.h1->tp_net = netoff;
2417 h.h1->tp_sec = ts.tv_sec;
2418 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2419 hdrlen = sizeof(*h.h1);
2420 break;
2421 case TPACKET_V2:
2422 h.h2->tp_len = skb->len;
2423 h.h2->tp_snaplen = snaplen;
2424 h.h2->tp_mac = macoff;
2425 h.h2->tp_net = netoff;
2426 h.h2->tp_sec = ts.tv_sec;
2427 h.h2->tp_nsec = ts.tv_nsec;
2428 if (skb_vlan_tag_present(skb)) {
2429 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2430 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2431 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2432 } else {
2433 h.h2->tp_vlan_tci = 0;
2434 h.h2->tp_vlan_tpid = 0;
2435 }
2436 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2437 hdrlen = sizeof(*h.h2);
2438 break;
2439 case TPACKET_V3:
2440 /* tp_nxt_offset,vlan are already populated above.
2441 * So DONT clear those fields here
2442 */
2443 h.h3->tp_status |= status;
2444 h.h3->tp_len = skb->len;
2445 h.h3->tp_snaplen = snaplen;
2446 h.h3->tp_mac = macoff;
2447 h.h3->tp_net = netoff;
2448 h.h3->tp_sec = ts.tv_sec;
2449 h.h3->tp_nsec = ts.tv_nsec;
2450 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2451 hdrlen = sizeof(*h.h3);
2452 break;
2453 default:
2454 BUG();
2455 }
2456
2457 sll = h.raw + TPACKET_ALIGN(hdrlen);
2458 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2459 sll->sll_family = AF_PACKET;
2460 sll->sll_hatype = dev->type;
2461 sll->sll_protocol = skb->protocol;
2462 sll->sll_pkttype = skb->pkt_type;
2463 if (unlikely(po->origdev))
2464 sll->sll_ifindex = orig_dev->ifindex;
2465 else
2466 sll->sll_ifindex = dev->ifindex;
2467
2468 smp_mb();
2469
2470#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2471 if (po->tp_version <= TPACKET_V2) {
2472 u8 *start, *end;
2473
2474 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2475 macoff + snaplen);
2476
2477 for (start = h.raw; start < end; start += PAGE_SIZE)
2478 flush_dcache_page(pgv_to_page(start));
2479 }
2480 smp_wmb();
2481#endif
2482
2483 if (po->tp_version <= TPACKET_V2) {
2484 spin_lock(&sk->sk_receive_queue.lock);
2485 __packet_set_status(po, h.raw, status);
2486 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2487 spin_unlock(&sk->sk_receive_queue.lock);
2488 sk->sk_data_ready(sk);
2489 } else if (po->tp_version == TPACKET_V3) {
2490 prb_clear_blk_fill_status(&po->rx_ring);
2491 }
2492
2493drop_n_restore:
2494 if (skb_head != skb->data && skb_shared(skb)) {
2495 skb->data = skb_head;
2496 skb->len = skb_len;
2497 }
2498drop:
2499 if (!is_drop_n_account)
2500 consume_skb(skb);
2501 else
2502 kfree_skb(skb);
2503 return 0;
2504
2505drop_n_account:
2506 spin_unlock(&sk->sk_receive_queue.lock);
2507 atomic_inc(&po->tp_drops);
2508 is_drop_n_account = true;
2509
2510 sk->sk_data_ready(sk);
2511 kfree_skb(copy_skb);
2512 goto drop_n_restore;
2513}
2514
2515static void tpacket_destruct_skb(struct sk_buff *skb)
2516{
2517 struct packet_sock *po = pkt_sk(skb->sk);
2518
2519 if (likely(po->tx_ring.pg_vec)) {
2520 void *ph;
2521 __u32 ts;
2522
2523 ph = skb_zcopy_get_nouarg(skb);
2524 packet_dec_pending(&po->tx_ring);
2525
2526 ts = __packet_set_timestamp(po, ph, skb);
2527 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2528
2529 if (!packet_read_pending(&po->tx_ring))
2530 complete(&po->skb_completion);
2531 }
2532
2533 sock_wfree(skb);
2534}
2535
2536static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2537{
2538 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2539 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2540 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2541 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2542 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2543 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2544 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2545
2546 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2547 return -EINVAL;
2548
2549 return 0;
2550}
2551
2552static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2553 struct virtio_net_hdr *vnet_hdr)
2554{
2555 if (*len < sizeof(*vnet_hdr))
2556 return -EINVAL;
2557 *len -= sizeof(*vnet_hdr);
2558
2559 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2560 return -EFAULT;
2561
2562 return __packet_snd_vnet_parse(vnet_hdr, *len);
2563}
2564
2565static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2566 void *frame, struct net_device *dev, void *data, int tp_len,
2567 __be16 proto, unsigned char *addr, int hlen, int copylen,
2568 const struct sockcm_cookie *sockc)
2569{
2570 union tpacket_uhdr ph;
2571 int to_write, offset, len, nr_frags, len_max;
2572 struct socket *sock = po->sk.sk_socket;
2573 struct page *page;
2574 int err;
2575
2576 ph.raw = frame;
2577
2578 skb->protocol = proto;
2579 skb->dev = dev;
2580 skb->priority = po->sk.sk_priority;
2581 skb->mark = po->sk.sk_mark;
2582 skb->tstamp = sockc->transmit_time;
2583 skb_setup_tx_timestamp(skb, sockc->tsflags);
2584 skb_zcopy_set_nouarg(skb, ph.raw);
2585
2586 skb_reserve(skb, hlen);
2587 skb_reset_network_header(skb);
2588
2589 to_write = tp_len;
2590
2591 if (sock->type == SOCK_DGRAM) {
2592 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2593 NULL, tp_len);
2594 if (unlikely(err < 0))
2595 return -EINVAL;
2596 } else if (copylen) {
2597 int hdrlen = min_t(int, copylen, tp_len);
2598
2599 skb_push(skb, dev->hard_header_len);
2600 skb_put(skb, copylen - dev->hard_header_len);
2601 err = skb_store_bits(skb, 0, data, hdrlen);
2602 if (unlikely(err))
2603 return err;
2604 if (!dev_validate_header(dev, skb->data, hdrlen))
2605 return -EINVAL;
2606
2607 data += hdrlen;
2608 to_write -= hdrlen;
2609 }
2610
2611 offset = offset_in_page(data);
2612 len_max = PAGE_SIZE - offset;
2613 len = ((to_write > len_max) ? len_max : to_write);
2614
2615 skb->data_len = to_write;
2616 skb->len += to_write;
2617 skb->truesize += to_write;
2618 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2619
2620 while (likely(to_write)) {
2621 nr_frags = skb_shinfo(skb)->nr_frags;
2622
2623 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2624 pr_err("Packet exceed the number of skb frags(%lu)\n",
2625 MAX_SKB_FRAGS);
2626 return -EFAULT;
2627 }
2628
2629 page = pgv_to_page(data);
2630 data += len;
2631 flush_dcache_page(page);
2632 get_page(page);
2633 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2634 to_write -= len;
2635 offset = 0;
2636 len_max = PAGE_SIZE;
2637 len = ((to_write > len_max) ? len_max : to_write);
2638 }
2639
2640 packet_parse_headers(skb, sock);
2641
2642 return tp_len;
2643}
2644
2645static int tpacket_parse_header(struct packet_sock *po, void *frame,
2646 int size_max, void **data)
2647{
2648 union tpacket_uhdr ph;
2649 int tp_len, off;
2650
2651 ph.raw = frame;
2652
2653 switch (po->tp_version) {
2654 case TPACKET_V3:
2655 if (ph.h3->tp_next_offset != 0) {
2656 pr_warn_once("variable sized slot not supported");
2657 return -EINVAL;
2658 }
2659 tp_len = ph.h3->tp_len;
2660 break;
2661 case TPACKET_V2:
2662 tp_len = ph.h2->tp_len;
2663 break;
2664 default:
2665 tp_len = ph.h1->tp_len;
2666 break;
2667 }
2668 if (unlikely(tp_len > size_max)) {
2669 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2670 return -EMSGSIZE;
2671 }
2672
2673 if (unlikely(po->tp_tx_has_off)) {
2674 int off_min, off_max;
2675
2676 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2677 off_max = po->tx_ring.frame_size - tp_len;
2678 if (po->sk.sk_type == SOCK_DGRAM) {
2679 switch (po->tp_version) {
2680 case TPACKET_V3:
2681 off = ph.h3->tp_net;
2682 break;
2683 case TPACKET_V2:
2684 off = ph.h2->tp_net;
2685 break;
2686 default:
2687 off = ph.h1->tp_net;
2688 break;
2689 }
2690 } else {
2691 switch (po->tp_version) {
2692 case TPACKET_V3:
2693 off = ph.h3->tp_mac;
2694 break;
2695 case TPACKET_V2:
2696 off = ph.h2->tp_mac;
2697 break;
2698 default:
2699 off = ph.h1->tp_mac;
2700 break;
2701 }
2702 }
2703 if (unlikely((off < off_min) || (off_max < off)))
2704 return -EINVAL;
2705 } else {
2706 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2707 }
2708
2709 *data = frame + off;
2710 return tp_len;
2711}
2712
2713static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2714{
2715 struct sk_buff *skb = NULL;
2716 struct net_device *dev;
2717 struct virtio_net_hdr *vnet_hdr = NULL;
2718 struct sockcm_cookie sockc;
2719 __be16 proto;
2720 int err, reserve = 0;
2721 void *ph;
2722 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2723 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2724 unsigned char *addr = NULL;
2725 int tp_len, size_max;
2726 void *data;
2727 int len_sum = 0;
2728 int status = TP_STATUS_AVAILABLE;
2729 int hlen, tlen, copylen = 0;
2730 long timeo = 0;
2731
2732 mutex_lock(&po->pg_vec_lock);
2733
2734 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2735 * we need to confirm it under protection of pg_vec_lock.
2736 */
2737 if (unlikely(!po->tx_ring.pg_vec)) {
2738 err = -EBUSY;
2739 goto out;
2740 }
2741 if (likely(saddr == NULL)) {
2742 dev = packet_cached_dev_get(po);
2743 proto = READ_ONCE(po->num);
2744 } else {
2745 err = -EINVAL;
2746 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2747 goto out;
2748 if (msg->msg_namelen < (saddr->sll_halen
2749 + offsetof(struct sockaddr_ll,
2750 sll_addr)))
2751 goto out;
2752 proto = saddr->sll_protocol;
2753 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2754 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2755 if (dev && msg->msg_namelen < dev->addr_len +
2756 offsetof(struct sockaddr_ll, sll_addr))
2757 goto out_put;
2758 addr = saddr->sll_addr;
2759 }
2760 }
2761
2762 err = -ENXIO;
2763 if (unlikely(dev == NULL))
2764 goto out;
2765 err = -ENETDOWN;
2766 if (unlikely(!(dev->flags & IFF_UP)))
2767 goto out_put;
2768
2769 sockcm_init(&sockc, &po->sk);
2770 if (msg->msg_controllen) {
2771 err = sock_cmsg_send(&po->sk, msg, &sockc);
2772 if (unlikely(err))
2773 goto out_put;
2774 }
2775
2776 if (po->sk.sk_socket->type == SOCK_RAW)
2777 reserve = dev->hard_header_len;
2778 size_max = po->tx_ring.frame_size
2779 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2780
2781 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2782 size_max = dev->mtu + reserve + VLAN_HLEN;
2783
2784 reinit_completion(&po->skb_completion);
2785
2786 do {
2787 ph = packet_current_frame(po, &po->tx_ring,
2788 TP_STATUS_SEND_REQUEST);
2789 if (unlikely(ph == NULL)) {
2790 if (need_wait && skb) {
2791 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2792 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2793 if (timeo <= 0) {
2794 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2795 goto out_put;
2796 }
2797 }
2798 /* check for additional frames */
2799 continue;
2800 }
2801
2802 skb = NULL;
2803 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2804 if (tp_len < 0)
2805 goto tpacket_error;
2806
2807 status = TP_STATUS_SEND_REQUEST;
2808 hlen = LL_RESERVED_SPACE(dev);
2809 tlen = dev->needed_tailroom;
2810 if (po->has_vnet_hdr) {
2811 vnet_hdr = data;
2812 data += sizeof(*vnet_hdr);
2813 tp_len -= sizeof(*vnet_hdr);
2814 if (tp_len < 0 ||
2815 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2816 tp_len = -EINVAL;
2817 goto tpacket_error;
2818 }
2819 copylen = __virtio16_to_cpu(vio_le(),
2820 vnet_hdr->hdr_len);
2821 }
2822 copylen = max_t(int, copylen, dev->hard_header_len);
2823 skb = sock_alloc_send_skb(&po->sk,
2824 hlen + tlen + sizeof(struct sockaddr_ll) +
2825 (copylen - dev->hard_header_len),
2826 !need_wait, &err);
2827
2828 if (unlikely(skb == NULL)) {
2829 /* we assume the socket was initially writeable ... */
2830 if (likely(len_sum > 0))
2831 err = len_sum;
2832 goto out_status;
2833 }
2834 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2835 addr, hlen, copylen, &sockc);
2836 if (likely(tp_len >= 0) &&
2837 tp_len > dev->mtu + reserve &&
2838 !po->has_vnet_hdr &&
2839 !packet_extra_vlan_len_allowed(dev, skb))
2840 tp_len = -EMSGSIZE;
2841
2842 if (unlikely(tp_len < 0)) {
2843tpacket_error:
2844 if (po->tp_loss) {
2845 __packet_set_status(po, ph,
2846 TP_STATUS_AVAILABLE);
2847 packet_increment_head(&po->tx_ring);
2848 kfree_skb(skb);
2849 continue;
2850 } else {
2851 status = TP_STATUS_WRONG_FORMAT;
2852 err = tp_len;
2853 goto out_status;
2854 }
2855 }
2856
2857 if (po->has_vnet_hdr) {
2858 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2859 tp_len = -EINVAL;
2860 goto tpacket_error;
2861 }
2862 virtio_net_hdr_set_proto(skb, vnet_hdr);
2863 }
2864
2865 skb->destructor = tpacket_destruct_skb;
2866 __packet_set_status(po, ph, TP_STATUS_SENDING);
2867 packet_inc_pending(&po->tx_ring);
2868
2869 status = TP_STATUS_SEND_REQUEST;
2870 err = po->xmit(skb);
2871 if (unlikely(err != 0)) {
2872 if (err > 0)
2873 err = net_xmit_errno(err);
2874 if (err && __packet_get_status(po, ph) ==
2875 TP_STATUS_AVAILABLE) {
2876 /* skb was destructed already */
2877 skb = NULL;
2878 goto out_status;
2879 }
2880 /*
2881 * skb was dropped but not destructed yet;
2882 * let's treat it like congestion or err < 0
2883 */
2884 err = 0;
2885 }
2886 packet_increment_head(&po->tx_ring);
2887 len_sum += tp_len;
2888 } while (likely((ph != NULL) ||
2889 /* Note: packet_read_pending() might be slow if we have
2890 * to call it as it's per_cpu variable, but in fast-path
2891 * we already short-circuit the loop with the first
2892 * condition, and luckily don't have to go that path
2893 * anyway.
2894 */
2895 (need_wait && packet_read_pending(&po->tx_ring))));
2896
2897 err = len_sum;
2898 goto out_put;
2899
2900out_status:
2901 __packet_set_status(po, ph, status);
2902 kfree_skb(skb);
2903out_put:
2904 dev_put(dev);
2905out:
2906 mutex_unlock(&po->pg_vec_lock);
2907 return err;
2908}
2909
2910static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2911 size_t reserve, size_t len,
2912 size_t linear, int noblock,
2913 int *err)
2914{
2915 struct sk_buff *skb;
2916
2917 /* Under a page? Don't bother with paged skb. */
2918 if (prepad + len < PAGE_SIZE || !linear)
2919 linear = len;
2920
2921 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2922 err, 0);
2923 if (!skb)
2924 return NULL;
2925
2926 skb_reserve(skb, reserve);
2927 skb_put(skb, linear);
2928 skb->data_len = len - linear;
2929 skb->len += len - linear;
2930
2931 return skb;
2932}
2933
2934static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2935{
2936 struct sock *sk = sock->sk;
2937 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2938 struct sk_buff *skb;
2939 struct net_device *dev;
2940 __be16 proto;
2941 unsigned char *addr = NULL;
2942 int err, reserve = 0;
2943 struct sockcm_cookie sockc;
2944 struct virtio_net_hdr vnet_hdr = { 0 };
2945 int offset = 0;
2946 struct packet_sock *po = pkt_sk(sk);
2947 bool has_vnet_hdr = false;
2948 int hlen, tlen, linear;
2949 int extra_len = 0;
2950
2951 /*
2952 * Get and verify the address.
2953 */
2954
2955 if (likely(saddr == NULL)) {
2956 dev = packet_cached_dev_get(po);
2957 proto = READ_ONCE(po->num);
2958 } else {
2959 err = -EINVAL;
2960 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2961 goto out;
2962 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2963 goto out;
2964 proto = saddr->sll_protocol;
2965 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2966 if (sock->type == SOCK_DGRAM) {
2967 if (dev && msg->msg_namelen < dev->addr_len +
2968 offsetof(struct sockaddr_ll, sll_addr))
2969 goto out_unlock;
2970 addr = saddr->sll_addr;
2971 }
2972 }
2973
2974 err = -ENXIO;
2975 if (unlikely(dev == NULL))
2976 goto out_unlock;
2977 err = -ENETDOWN;
2978 if (unlikely(!(dev->flags & IFF_UP)))
2979 goto out_unlock;
2980
2981 sockcm_init(&sockc, sk);
2982 sockc.mark = sk->sk_mark;
2983 if (msg->msg_controllen) {
2984 err = sock_cmsg_send(sk, msg, &sockc);
2985 if (unlikely(err))
2986 goto out_unlock;
2987 }
2988
2989 if (sock->type == SOCK_RAW)
2990 reserve = dev->hard_header_len;
2991 if (po->has_vnet_hdr) {
2992 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2993 if (err)
2994 goto out_unlock;
2995 has_vnet_hdr = true;
2996 }
2997
2998 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2999 if (!netif_supports_nofcs(dev)) {
3000 err = -EPROTONOSUPPORT;
3001 goto out_unlock;
3002 }
3003 extra_len = 4; /* We're doing our own CRC */
3004 }
3005
3006 err = -EMSGSIZE;
3007 if (!vnet_hdr.gso_type &&
3008 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3009 goto out_unlock;
3010
3011 err = -ENOBUFS;
3012 hlen = LL_RESERVED_SPACE(dev);
3013 tlen = dev->needed_tailroom;
3014 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3015 linear = max(linear, min_t(int, len, dev->hard_header_len));
3016 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3017 msg->msg_flags & MSG_DONTWAIT, &err);
3018 if (skb == NULL)
3019 goto out_unlock;
3020
3021 skb_reset_network_header(skb);
3022
3023 err = -EINVAL;
3024 if (sock->type == SOCK_DGRAM) {
3025 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3026 if (unlikely(offset < 0))
3027 goto out_free;
3028 } else if (reserve) {
3029 skb_reserve(skb, -reserve);
3030 if (len < reserve + sizeof(struct ipv6hdr) &&
3031 dev->min_header_len != dev->hard_header_len)
3032 skb_reset_network_header(skb);
3033 }
3034
3035 /* Returns -EFAULT on error */
3036 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3037 if (err)
3038 goto out_free;
3039
3040 if ((sock->type == SOCK_RAW &&
3041 !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3042 err = -EINVAL;
3043 goto out_free;
3044 }
3045
3046 skb_setup_tx_timestamp(skb, sockc.tsflags);
3047
3048 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3049 !packet_extra_vlan_len_allowed(dev, skb)) {
3050 err = -EMSGSIZE;
3051 goto out_free;
3052 }
3053
3054 skb->protocol = proto;
3055 skb->dev = dev;
3056 skb->priority = sk->sk_priority;
3057 skb->mark = sockc.mark;
3058 skb->tstamp = sockc.transmit_time;
3059
3060 if (unlikely(extra_len == 4))
3061 skb->no_fcs = 1;
3062
3063 packet_parse_headers(skb, sock);
3064
3065 if (has_vnet_hdr) {
3066 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3067 if (err)
3068 goto out_free;
3069 len += sizeof(vnet_hdr);
3070 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3071 }
3072
3073 err = po->xmit(skb);
3074 if (unlikely(err != 0)) {
3075 if (err > 0)
3076 err = net_xmit_errno(err);
3077 if (err)
3078 goto out_unlock;
3079 }
3080
3081 dev_put(dev);
3082
3083 return len;
3084
3085out_free:
3086 kfree_skb(skb);
3087out_unlock:
3088 dev_put(dev);
3089out:
3090 return err;
3091}
3092
3093static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3094{
3095 struct sock *sk = sock->sk;
3096 struct packet_sock *po = pkt_sk(sk);
3097
3098 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3099 * tpacket_snd() will redo the check safely.
3100 */
3101 if (data_race(po->tx_ring.pg_vec))
3102 return tpacket_snd(po, msg);
3103
3104 return packet_snd(sock, msg, len);
3105}
3106
3107/*
3108 * Close a PACKET socket. This is fairly simple. We immediately go
3109 * to 'closed' state and remove our protocol entry in the device list.
3110 */
3111
3112static int packet_release(struct socket *sock)
3113{
3114 struct sock *sk = sock->sk;
3115 struct packet_sock *po;
3116 struct packet_fanout *f;
3117 struct net *net;
3118 union tpacket_req_u req_u;
3119
3120 if (!sk)
3121 return 0;
3122
3123 net = sock_net(sk);
3124 po = pkt_sk(sk);
3125
3126 mutex_lock(&net->packet.sklist_lock);
3127 sk_del_node_init_rcu(sk);
3128 mutex_unlock(&net->packet.sklist_lock);
3129
3130 sock_prot_inuse_add(net, sk->sk_prot, -1);
3131
3132 spin_lock(&po->bind_lock);
3133 unregister_prot_hook(sk, false);
3134 packet_cached_dev_reset(po);
3135
3136 if (po->prot_hook.dev) {
3137 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3138 po->prot_hook.dev = NULL;
3139 }
3140 spin_unlock(&po->bind_lock);
3141
3142 packet_flush_mclist(sk);
3143
3144 lock_sock(sk);
3145 if (po->rx_ring.pg_vec) {
3146 memset(&req_u, 0, sizeof(req_u));
3147 packet_set_ring(sk, &req_u, 1, 0);
3148 }
3149
3150 if (po->tx_ring.pg_vec) {
3151 memset(&req_u, 0, sizeof(req_u));
3152 packet_set_ring(sk, &req_u, 1, 1);
3153 }
3154 release_sock(sk);
3155
3156 f = fanout_release(sk);
3157
3158 synchronize_net();
3159
3160 kfree(po->rollover);
3161 if (f) {
3162 fanout_release_data(f);
3163 kvfree(f);
3164 }
3165 /*
3166 * Now the socket is dead. No more input will appear.
3167 */
3168 sock_orphan(sk);
3169 sock->sk = NULL;
3170
3171 /* Purge queues */
3172
3173 skb_queue_purge(&sk->sk_receive_queue);
3174 packet_free_pending(po);
3175 sk_refcnt_debug_release(sk);
3176
3177 sock_put(sk);
3178 return 0;
3179}
3180
3181/*
3182 * Attach a packet hook.
3183 */
3184
3185static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3186 __be16 proto)
3187{
3188 struct packet_sock *po = pkt_sk(sk);
3189 struct net_device *dev = NULL;
3190 bool unlisted = false;
3191 bool need_rehook;
3192 int ret = 0;
3193
3194 lock_sock(sk);
3195 spin_lock(&po->bind_lock);
3196 rcu_read_lock();
3197
3198 if (po->fanout) {
3199 ret = -EINVAL;
3200 goto out_unlock;
3201 }
3202
3203 if (name) {
3204 dev = dev_get_by_name_rcu(sock_net(sk), name);
3205 if (!dev) {
3206 ret = -ENODEV;
3207 goto out_unlock;
3208 }
3209 } else if (ifindex) {
3210 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3211 if (!dev) {
3212 ret = -ENODEV;
3213 goto out_unlock;
3214 }
3215 }
3216
3217 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3218
3219 if (need_rehook) {
3220 dev_hold(dev);
3221 if (po->running) {
3222 rcu_read_unlock();
3223 /* prevents packet_notifier() from calling
3224 * register_prot_hook()
3225 */
3226 WRITE_ONCE(po->num, 0);
3227 __unregister_prot_hook(sk, true);
3228 rcu_read_lock();
3229 if (dev)
3230 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3231 dev->ifindex);
3232 }
3233
3234 BUG_ON(po->running);
3235 WRITE_ONCE(po->num, proto);
3236 po->prot_hook.type = proto;
3237
3238 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3239
3240 if (unlikely(unlisted)) {
3241 po->prot_hook.dev = NULL;
3242 WRITE_ONCE(po->ifindex, -1);
3243 packet_cached_dev_reset(po);
3244 } else {
3245 netdev_hold(dev, &po->prot_hook.dev_tracker,
3246 GFP_ATOMIC);
3247 po->prot_hook.dev = dev;
3248 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3249 packet_cached_dev_assign(po, dev);
3250 }
3251 dev_put(dev);
3252 }
3253
3254 if (proto == 0 || !need_rehook)
3255 goto out_unlock;
3256
3257 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3258 register_prot_hook(sk);
3259 } else {
3260 sk->sk_err = ENETDOWN;
3261 if (!sock_flag(sk, SOCK_DEAD))
3262 sk_error_report(sk);
3263 }
3264
3265out_unlock:
3266 rcu_read_unlock();
3267 spin_unlock(&po->bind_lock);
3268 release_sock(sk);
3269 return ret;
3270}
3271
3272/*
3273 * Bind a packet socket to a device
3274 */
3275
3276static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3277 int addr_len)
3278{
3279 struct sock *sk = sock->sk;
3280 char name[sizeof(uaddr->sa_data_min) + 1];
3281
3282 /*
3283 * Check legality
3284 */
3285
3286 if (addr_len != sizeof(struct sockaddr))
3287 return -EINVAL;
3288 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3289 * zero-terminated.
3290 */
3291 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3292 name[sizeof(uaddr->sa_data_min)] = 0;
3293
3294 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3295}
3296
3297static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3298{
3299 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3300 struct sock *sk = sock->sk;
3301
3302 /*
3303 * Check legality
3304 */
3305
3306 if (addr_len < sizeof(struct sockaddr_ll))
3307 return -EINVAL;
3308 if (sll->sll_family != AF_PACKET)
3309 return -EINVAL;
3310
3311 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3312 sll->sll_protocol ? : pkt_sk(sk)->num);
3313}
3314
3315static struct proto packet_proto = {
3316 .name = "PACKET",
3317 .owner = THIS_MODULE,
3318 .obj_size = sizeof(struct packet_sock),
3319};
3320
3321/*
3322 * Create a packet of type SOCK_PACKET.
3323 */
3324
3325static int packet_create(struct net *net, struct socket *sock, int protocol,
3326 int kern)
3327{
3328 struct sock *sk;
3329 struct packet_sock *po;
3330 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3331 int err;
3332
3333 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3334 return -EPERM;
3335 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3336 sock->type != SOCK_PACKET)
3337 return -ESOCKTNOSUPPORT;
3338
3339 sock->state = SS_UNCONNECTED;
3340
3341 err = -ENOBUFS;
3342 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3343 if (sk == NULL)
3344 goto out;
3345
3346 sock->ops = &packet_ops;
3347 if (sock->type == SOCK_PACKET)
3348 sock->ops = &packet_ops_spkt;
3349
3350 sock_init_data(sock, sk);
3351
3352 po = pkt_sk(sk);
3353 init_completion(&po->skb_completion);
3354 sk->sk_family = PF_PACKET;
3355 po->num = proto;
3356 po->xmit = dev_queue_xmit;
3357
3358 err = packet_alloc_pending(po);
3359 if (err)
3360 goto out2;
3361
3362 packet_cached_dev_reset(po);
3363
3364 sk->sk_destruct = packet_sock_destruct;
3365 sk_refcnt_debug_inc(sk);
3366
3367 /*
3368 * Attach a protocol block
3369 */
3370
3371 spin_lock_init(&po->bind_lock);
3372 mutex_init(&po->pg_vec_lock);
3373 po->rollover = NULL;
3374 po->prot_hook.func = packet_rcv;
3375
3376 if (sock->type == SOCK_PACKET)
3377 po->prot_hook.func = packet_rcv_spkt;
3378
3379 po->prot_hook.af_packet_priv = sk;
3380 po->prot_hook.af_packet_net = sock_net(sk);
3381
3382 if (proto) {
3383 po->prot_hook.type = proto;
3384 __register_prot_hook(sk);
3385 }
3386
3387 mutex_lock(&net->packet.sklist_lock);
3388 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3389 mutex_unlock(&net->packet.sklist_lock);
3390
3391 sock_prot_inuse_add(net, &packet_proto, 1);
3392
3393 return 0;
3394out2:
3395 sk_free(sk);
3396out:
3397 return err;
3398}
3399
3400/*
3401 * Pull a packet from our receive queue and hand it to the user.
3402 * If necessary we block.
3403 */
3404
3405static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3406 int flags)
3407{
3408 struct sock *sk = sock->sk;
3409 struct sk_buff *skb;
3410 int copied, err;
3411 int vnet_hdr_len = 0;
3412 unsigned int origlen = 0;
3413
3414 err = -EINVAL;
3415 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3416 goto out;
3417
3418#if 0
3419 /* What error should we return now? EUNATTACH? */
3420 if (pkt_sk(sk)->ifindex < 0)
3421 return -ENODEV;
3422#endif
3423
3424 if (flags & MSG_ERRQUEUE) {
3425 err = sock_recv_errqueue(sk, msg, len,
3426 SOL_PACKET, PACKET_TX_TIMESTAMP);
3427 goto out;
3428 }
3429
3430 /*
3431 * Call the generic datagram receiver. This handles all sorts
3432 * of horrible races and re-entrancy so we can forget about it
3433 * in the protocol layers.
3434 *
3435 * Now it will return ENETDOWN, if device have just gone down,
3436 * but then it will block.
3437 */
3438
3439 skb = skb_recv_datagram(sk, flags, &err);
3440
3441 /*
3442 * An error occurred so return it. Because skb_recv_datagram()
3443 * handles the blocking we don't see and worry about blocking
3444 * retries.
3445 */
3446
3447 if (skb == NULL)
3448 goto out;
3449
3450 packet_rcv_try_clear_pressure(pkt_sk(sk));
3451
3452 if (pkt_sk(sk)->has_vnet_hdr) {
3453 err = packet_rcv_vnet(msg, skb, &len);
3454 if (err)
3455 goto out_free;
3456 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3457 }
3458
3459 /* You lose any data beyond the buffer you gave. If it worries
3460 * a user program they can ask the device for its MTU
3461 * anyway.
3462 */
3463 copied = skb->len;
3464 if (copied > len) {
3465 copied = len;
3466 msg->msg_flags |= MSG_TRUNC;
3467 }
3468
3469 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3470 if (err)
3471 goto out_free;
3472
3473 if (sock->type != SOCK_PACKET) {
3474 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3475
3476 /* Original length was stored in sockaddr_ll fields */
3477 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3478 sll->sll_family = AF_PACKET;
3479 sll->sll_protocol = skb->protocol;
3480 }
3481
3482 sock_recv_cmsgs(msg, sk, skb);
3483
3484 if (msg->msg_name) {
3485 const size_t max_len = min(sizeof(skb->cb),
3486 sizeof(struct sockaddr_storage));
3487 int copy_len;
3488
3489 /* If the address length field is there to be filled
3490 * in, we fill it in now.
3491 */
3492 if (sock->type == SOCK_PACKET) {
3493 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3494 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3495 copy_len = msg->msg_namelen;
3496 } else {
3497 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3498
3499 msg->msg_namelen = sll->sll_halen +
3500 offsetof(struct sockaddr_ll, sll_addr);
3501 copy_len = msg->msg_namelen;
3502 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3503 memset(msg->msg_name +
3504 offsetof(struct sockaddr_ll, sll_addr),
3505 0, sizeof(sll->sll_addr));
3506 msg->msg_namelen = sizeof(struct sockaddr_ll);
3507 }
3508 }
3509 if (WARN_ON_ONCE(copy_len > max_len)) {
3510 copy_len = max_len;
3511 msg->msg_namelen = copy_len;
3512 }
3513 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3514 }
3515
3516 if (pkt_sk(sk)->auxdata) {
3517 struct tpacket_auxdata aux;
3518
3519 aux.tp_status = TP_STATUS_USER;
3520 if (skb->ip_summed == CHECKSUM_PARTIAL)
3521 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3522 else if (skb->pkt_type != PACKET_OUTGOING &&
3523 skb_csum_unnecessary(skb))
3524 aux.tp_status |= TP_STATUS_CSUM_VALID;
3525
3526 aux.tp_len = origlen;
3527 aux.tp_snaplen = skb->len;
3528 aux.tp_mac = 0;
3529 aux.tp_net = skb_network_offset(skb);
3530 if (skb_vlan_tag_present(skb)) {
3531 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3532 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3533 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3534 } else {
3535 aux.tp_vlan_tci = 0;
3536 aux.tp_vlan_tpid = 0;
3537 }
3538 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3539 }
3540
3541 /*
3542 * Free or return the buffer as appropriate. Again this
3543 * hides all the races and re-entrancy issues from us.
3544 */
3545 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3546
3547out_free:
3548 skb_free_datagram(sk, skb);
3549out:
3550 return err;
3551}
3552
3553static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3554 int peer)
3555{
3556 struct net_device *dev;
3557 struct sock *sk = sock->sk;
3558
3559 if (peer)
3560 return -EOPNOTSUPP;
3561
3562 uaddr->sa_family = AF_PACKET;
3563 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3564 rcu_read_lock();
3565 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3566 if (dev)
3567 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3568 rcu_read_unlock();
3569
3570 return sizeof(*uaddr);
3571}
3572
3573static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3574 int peer)
3575{
3576 struct net_device *dev;
3577 struct sock *sk = sock->sk;
3578 struct packet_sock *po = pkt_sk(sk);
3579 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3580 int ifindex;
3581
3582 if (peer)
3583 return -EOPNOTSUPP;
3584
3585 ifindex = READ_ONCE(po->ifindex);
3586 sll->sll_family = AF_PACKET;
3587 sll->sll_ifindex = ifindex;
3588 sll->sll_protocol = READ_ONCE(po->num);
3589 sll->sll_pkttype = 0;
3590 rcu_read_lock();
3591 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3592 if (dev) {
3593 sll->sll_hatype = dev->type;
3594 sll->sll_halen = dev->addr_len;
3595 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3596 } else {
3597 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3598 sll->sll_halen = 0;
3599 }
3600 rcu_read_unlock();
3601
3602 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3603}
3604
3605static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3606 int what)
3607{
3608 switch (i->type) {
3609 case PACKET_MR_MULTICAST:
3610 if (i->alen != dev->addr_len)
3611 return -EINVAL;
3612 if (what > 0)
3613 return dev_mc_add(dev, i->addr);
3614 else
3615 return dev_mc_del(dev, i->addr);
3616 break;
3617 case PACKET_MR_PROMISC:
3618 return dev_set_promiscuity(dev, what);
3619 case PACKET_MR_ALLMULTI:
3620 return dev_set_allmulti(dev, what);
3621 case PACKET_MR_UNICAST:
3622 if (i->alen != dev->addr_len)
3623 return -EINVAL;
3624 if (what > 0)
3625 return dev_uc_add(dev, i->addr);
3626 else
3627 return dev_uc_del(dev, i->addr);
3628 break;
3629 default:
3630 break;
3631 }
3632 return 0;
3633}
3634
3635static void packet_dev_mclist_delete(struct net_device *dev,
3636 struct packet_mclist **mlp)
3637{
3638 struct packet_mclist *ml;
3639
3640 while ((ml = *mlp) != NULL) {
3641 if (ml->ifindex == dev->ifindex) {
3642 packet_dev_mc(dev, ml, -1);
3643 *mlp = ml->next;
3644 kfree(ml);
3645 } else
3646 mlp = &ml->next;
3647 }
3648}
3649
3650static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3651{
3652 struct packet_sock *po = pkt_sk(sk);
3653 struct packet_mclist *ml, *i;
3654 struct net_device *dev;
3655 int err;
3656
3657 rtnl_lock();
3658
3659 err = -ENODEV;
3660 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3661 if (!dev)
3662 goto done;
3663
3664 err = -EINVAL;
3665 if (mreq->mr_alen > dev->addr_len)
3666 goto done;
3667
3668 err = -ENOBUFS;
3669 i = kmalloc(sizeof(*i), GFP_KERNEL);
3670 if (i == NULL)
3671 goto done;
3672
3673 err = 0;
3674 for (ml = po->mclist; ml; ml = ml->next) {
3675 if (ml->ifindex == mreq->mr_ifindex &&
3676 ml->type == mreq->mr_type &&
3677 ml->alen == mreq->mr_alen &&
3678 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3679 ml->count++;
3680 /* Free the new element ... */
3681 kfree(i);
3682 goto done;
3683 }
3684 }
3685
3686 i->type = mreq->mr_type;
3687 i->ifindex = mreq->mr_ifindex;
3688 i->alen = mreq->mr_alen;
3689 memcpy(i->addr, mreq->mr_address, i->alen);
3690 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3691 i->count = 1;
3692 i->next = po->mclist;
3693 po->mclist = i;
3694 err = packet_dev_mc(dev, i, 1);
3695 if (err) {
3696 po->mclist = i->next;
3697 kfree(i);
3698 }
3699
3700done:
3701 rtnl_unlock();
3702 return err;
3703}
3704
3705static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3706{
3707 struct packet_mclist *ml, **mlp;
3708
3709 rtnl_lock();
3710
3711 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3712 if (ml->ifindex == mreq->mr_ifindex &&
3713 ml->type == mreq->mr_type &&
3714 ml->alen == mreq->mr_alen &&
3715 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3716 if (--ml->count == 0) {
3717 struct net_device *dev;
3718 *mlp = ml->next;
3719 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3720 if (dev)
3721 packet_dev_mc(dev, ml, -1);
3722 kfree(ml);
3723 }
3724 break;
3725 }
3726 }
3727 rtnl_unlock();
3728 return 0;
3729}
3730
3731static void packet_flush_mclist(struct sock *sk)
3732{
3733 struct packet_sock *po = pkt_sk(sk);
3734 struct packet_mclist *ml;
3735
3736 if (!po->mclist)
3737 return;
3738
3739 rtnl_lock();
3740 while ((ml = po->mclist) != NULL) {
3741 struct net_device *dev;
3742
3743 po->mclist = ml->next;
3744 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3745 if (dev != NULL)
3746 packet_dev_mc(dev, ml, -1);
3747 kfree(ml);
3748 }
3749 rtnl_unlock();
3750}
3751
3752static int
3753packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3754 unsigned int optlen)
3755{
3756 struct sock *sk = sock->sk;
3757 struct packet_sock *po = pkt_sk(sk);
3758 int ret;
3759
3760 if (level != SOL_PACKET)
3761 return -ENOPROTOOPT;
3762
3763 switch (optname) {
3764 case PACKET_ADD_MEMBERSHIP:
3765 case PACKET_DROP_MEMBERSHIP:
3766 {
3767 struct packet_mreq_max mreq;
3768 int len = optlen;
3769 memset(&mreq, 0, sizeof(mreq));
3770 if (len < sizeof(struct packet_mreq))
3771 return -EINVAL;
3772 if (len > sizeof(mreq))
3773 len = sizeof(mreq);
3774 if (copy_from_sockptr(&mreq, optval, len))
3775 return -EFAULT;
3776 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3777 return -EINVAL;
3778 if (optname == PACKET_ADD_MEMBERSHIP)
3779 ret = packet_mc_add(sk, &mreq);
3780 else
3781 ret = packet_mc_drop(sk, &mreq);
3782 return ret;
3783 }
3784
3785 case PACKET_RX_RING:
3786 case PACKET_TX_RING:
3787 {
3788 union tpacket_req_u req_u;
3789 int len;
3790
3791 lock_sock(sk);
3792 switch (po->tp_version) {
3793 case TPACKET_V1:
3794 case TPACKET_V2:
3795 len = sizeof(req_u.req);
3796 break;
3797 case TPACKET_V3:
3798 default:
3799 len = sizeof(req_u.req3);
3800 break;
3801 }
3802 if (optlen < len) {
3803 ret = -EINVAL;
3804 } else {
3805 if (copy_from_sockptr(&req_u.req, optval, len))
3806 ret = -EFAULT;
3807 else
3808 ret = packet_set_ring(sk, &req_u, 0,
3809 optname == PACKET_TX_RING);
3810 }
3811 release_sock(sk);
3812 return ret;
3813 }
3814 case PACKET_COPY_THRESH:
3815 {
3816 int val;
3817
3818 if (optlen != sizeof(val))
3819 return -EINVAL;
3820 if (copy_from_sockptr(&val, optval, sizeof(val)))
3821 return -EFAULT;
3822
3823 pkt_sk(sk)->copy_thresh = val;
3824 return 0;
3825 }
3826 case PACKET_VERSION:
3827 {
3828 int val;
3829
3830 if (optlen != sizeof(val))
3831 return -EINVAL;
3832 if (copy_from_sockptr(&val, optval, sizeof(val)))
3833 return -EFAULT;
3834 switch (val) {
3835 case TPACKET_V1:
3836 case TPACKET_V2:
3837 case TPACKET_V3:
3838 break;
3839 default:
3840 return -EINVAL;
3841 }
3842 lock_sock(sk);
3843 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3844 ret = -EBUSY;
3845 } else {
3846 po->tp_version = val;
3847 ret = 0;
3848 }
3849 release_sock(sk);
3850 return ret;
3851 }
3852 case PACKET_RESERVE:
3853 {
3854 unsigned int val;
3855
3856 if (optlen != sizeof(val))
3857 return -EINVAL;
3858 if (copy_from_sockptr(&val, optval, sizeof(val)))
3859 return -EFAULT;
3860 if (val > INT_MAX)
3861 return -EINVAL;
3862 lock_sock(sk);
3863 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3864 ret = -EBUSY;
3865 } else {
3866 po->tp_reserve = val;
3867 ret = 0;
3868 }
3869 release_sock(sk);
3870 return ret;
3871 }
3872 case PACKET_LOSS:
3873 {
3874 unsigned int val;
3875
3876 if (optlen != sizeof(val))
3877 return -EINVAL;
3878 if (copy_from_sockptr(&val, optval, sizeof(val)))
3879 return -EFAULT;
3880
3881 lock_sock(sk);
3882 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3883 ret = -EBUSY;
3884 } else {
3885 po->tp_loss = !!val;
3886 ret = 0;
3887 }
3888 release_sock(sk);
3889 return ret;
3890 }
3891 case PACKET_AUXDATA:
3892 {
3893 int val;
3894
3895 if (optlen < sizeof(val))
3896 return -EINVAL;
3897 if (copy_from_sockptr(&val, optval, sizeof(val)))
3898 return -EFAULT;
3899
3900 lock_sock(sk);
3901 po->auxdata = !!val;
3902 release_sock(sk);
3903 return 0;
3904 }
3905 case PACKET_ORIGDEV:
3906 {
3907 int val;
3908
3909 if (optlen < sizeof(val))
3910 return -EINVAL;
3911 if (copy_from_sockptr(&val, optval, sizeof(val)))
3912 return -EFAULT;
3913
3914 lock_sock(sk);
3915 po->origdev = !!val;
3916 release_sock(sk);
3917 return 0;
3918 }
3919 case PACKET_VNET_HDR:
3920 {
3921 int val;
3922
3923 if (sock->type != SOCK_RAW)
3924 return -EINVAL;
3925 if (optlen < sizeof(val))
3926 return -EINVAL;
3927 if (copy_from_sockptr(&val, optval, sizeof(val)))
3928 return -EFAULT;
3929
3930 lock_sock(sk);
3931 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3932 ret = -EBUSY;
3933 } else {
3934 po->has_vnet_hdr = !!val;
3935 ret = 0;
3936 }
3937 release_sock(sk);
3938 return ret;
3939 }
3940 case PACKET_TIMESTAMP:
3941 {
3942 int val;
3943
3944 if (optlen != sizeof(val))
3945 return -EINVAL;
3946 if (copy_from_sockptr(&val, optval, sizeof(val)))
3947 return -EFAULT;
3948
3949 po->tp_tstamp = val;
3950 return 0;
3951 }
3952 case PACKET_FANOUT:
3953 {
3954 struct fanout_args args = { 0 };
3955
3956 if (optlen != sizeof(int) && optlen != sizeof(args))
3957 return -EINVAL;
3958 if (copy_from_sockptr(&args, optval, optlen))
3959 return -EFAULT;
3960
3961 return fanout_add(sk, &args);
3962 }
3963 case PACKET_FANOUT_DATA:
3964 {
3965 /* Paired with the WRITE_ONCE() in fanout_add() */
3966 if (!READ_ONCE(po->fanout))
3967 return -EINVAL;
3968
3969 return fanout_set_data(po, optval, optlen);
3970 }
3971 case PACKET_IGNORE_OUTGOING:
3972 {
3973 int val;
3974
3975 if (optlen != sizeof(val))
3976 return -EINVAL;
3977 if (copy_from_sockptr(&val, optval, sizeof(val)))
3978 return -EFAULT;
3979 if (val < 0 || val > 1)
3980 return -EINVAL;
3981
3982 po->prot_hook.ignore_outgoing = !!val;
3983 return 0;
3984 }
3985 case PACKET_TX_HAS_OFF:
3986 {
3987 unsigned int val;
3988
3989 if (optlen != sizeof(val))
3990 return -EINVAL;
3991 if (copy_from_sockptr(&val, optval, sizeof(val)))
3992 return -EFAULT;
3993
3994 lock_sock(sk);
3995 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
3996 po->tp_tx_has_off = !!val;
3997
3998 release_sock(sk);
3999 return 0;
4000 }
4001 case PACKET_QDISC_BYPASS:
4002 {
4003 int val;
4004
4005 if (optlen != sizeof(val))
4006 return -EINVAL;
4007 if (copy_from_sockptr(&val, optval, sizeof(val)))
4008 return -EFAULT;
4009
4010 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
4011 return 0;
4012 }
4013 default:
4014 return -ENOPROTOOPT;
4015 }
4016}
4017
4018static int packet_getsockopt(struct socket *sock, int level, int optname,
4019 char __user *optval, int __user *optlen)
4020{
4021 int len;
4022 int val, lv = sizeof(val);
4023 struct sock *sk = sock->sk;
4024 struct packet_sock *po = pkt_sk(sk);
4025 void *data = &val;
4026 union tpacket_stats_u st;
4027 struct tpacket_rollover_stats rstats;
4028 int drops;
4029
4030 if (level != SOL_PACKET)
4031 return -ENOPROTOOPT;
4032
4033 if (get_user(len, optlen))
4034 return -EFAULT;
4035
4036 if (len < 0)
4037 return -EINVAL;
4038
4039 switch (optname) {
4040 case PACKET_STATISTICS:
4041 spin_lock_bh(&sk->sk_receive_queue.lock);
4042 memcpy(&st, &po->stats, sizeof(st));
4043 memset(&po->stats, 0, sizeof(po->stats));
4044 spin_unlock_bh(&sk->sk_receive_queue.lock);
4045 drops = atomic_xchg(&po->tp_drops, 0);
4046
4047 if (po->tp_version == TPACKET_V3) {
4048 lv = sizeof(struct tpacket_stats_v3);
4049 st.stats3.tp_drops = drops;
4050 st.stats3.tp_packets += drops;
4051 data = &st.stats3;
4052 } else {
4053 lv = sizeof(struct tpacket_stats);
4054 st.stats1.tp_drops = drops;
4055 st.stats1.tp_packets += drops;
4056 data = &st.stats1;
4057 }
4058
4059 break;
4060 case PACKET_AUXDATA:
4061 val = po->auxdata;
4062 break;
4063 case PACKET_ORIGDEV:
4064 val = po->origdev;
4065 break;
4066 case PACKET_VNET_HDR:
4067 val = po->has_vnet_hdr;
4068 break;
4069 case PACKET_VERSION:
4070 val = po->tp_version;
4071 break;
4072 case PACKET_HDRLEN:
4073 if (len > sizeof(int))
4074 len = sizeof(int);
4075 if (len < sizeof(int))
4076 return -EINVAL;
4077 if (copy_from_user(&val, optval, len))
4078 return -EFAULT;
4079 switch (val) {
4080 case TPACKET_V1:
4081 val = sizeof(struct tpacket_hdr);
4082 break;
4083 case TPACKET_V2:
4084 val = sizeof(struct tpacket2_hdr);
4085 break;
4086 case TPACKET_V3:
4087 val = sizeof(struct tpacket3_hdr);
4088 break;
4089 default:
4090 return -EINVAL;
4091 }
4092 break;
4093 case PACKET_RESERVE:
4094 val = po->tp_reserve;
4095 break;
4096 case PACKET_LOSS:
4097 val = po->tp_loss;
4098 break;
4099 case PACKET_TIMESTAMP:
4100 val = po->tp_tstamp;
4101 break;
4102 case PACKET_FANOUT:
4103 val = (po->fanout ?
4104 ((u32)po->fanout->id |
4105 ((u32)po->fanout->type << 16) |
4106 ((u32)po->fanout->flags << 24)) :
4107 0);
4108 break;
4109 case PACKET_IGNORE_OUTGOING:
4110 val = po->prot_hook.ignore_outgoing;
4111 break;
4112 case PACKET_ROLLOVER_STATS:
4113 if (!po->rollover)
4114 return -EINVAL;
4115 rstats.tp_all = atomic_long_read(&po->rollover->num);
4116 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4117 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4118 data = &rstats;
4119 lv = sizeof(rstats);
4120 break;
4121 case PACKET_TX_HAS_OFF:
4122 val = po->tp_tx_has_off;
4123 break;
4124 case PACKET_QDISC_BYPASS:
4125 val = packet_use_direct_xmit(po);
4126 break;
4127 default:
4128 return -ENOPROTOOPT;
4129 }
4130
4131 if (len > lv)
4132 len = lv;
4133 if (put_user(len, optlen))
4134 return -EFAULT;
4135 if (copy_to_user(optval, data, len))
4136 return -EFAULT;
4137 return 0;
4138}
4139
4140static int packet_notifier(struct notifier_block *this,
4141 unsigned long msg, void *ptr)
4142{
4143 struct sock *sk;
4144 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4145 struct net *net = dev_net(dev);
4146
4147 rcu_read_lock();
4148 sk_for_each_rcu(sk, &net->packet.sklist) {
4149 struct packet_sock *po = pkt_sk(sk);
4150
4151 switch (msg) {
4152 case NETDEV_UNREGISTER:
4153 if (po->mclist)
4154 packet_dev_mclist_delete(dev, &po->mclist);
4155 fallthrough;
4156
4157 case NETDEV_DOWN:
4158 if (dev->ifindex == po->ifindex) {
4159 spin_lock(&po->bind_lock);
4160 if (po->running) {
4161 __unregister_prot_hook(sk, false);
4162 sk->sk_err = ENETDOWN;
4163 if (!sock_flag(sk, SOCK_DEAD))
4164 sk_error_report(sk);
4165 }
4166 if (msg == NETDEV_UNREGISTER) {
4167 packet_cached_dev_reset(po);
4168 WRITE_ONCE(po->ifindex, -1);
4169 netdev_put(po->prot_hook.dev,
4170 &po->prot_hook.dev_tracker);
4171 po->prot_hook.dev = NULL;
4172 }
4173 spin_unlock(&po->bind_lock);
4174 }
4175 break;
4176 case NETDEV_UP:
4177 if (dev->ifindex == po->ifindex) {
4178 spin_lock(&po->bind_lock);
4179 if (po->num)
4180 register_prot_hook(sk);
4181 spin_unlock(&po->bind_lock);
4182 }
4183 break;
4184 }
4185 }
4186 rcu_read_unlock();
4187 return NOTIFY_DONE;
4188}
4189
4190
4191static int packet_ioctl(struct socket *sock, unsigned int cmd,
4192 unsigned long arg)
4193{
4194 struct sock *sk = sock->sk;
4195
4196 switch (cmd) {
4197 case SIOCOUTQ:
4198 {
4199 int amount = sk_wmem_alloc_get(sk);
4200
4201 return put_user(amount, (int __user *)arg);
4202 }
4203 case SIOCINQ:
4204 {
4205 struct sk_buff *skb;
4206 int amount = 0;
4207
4208 spin_lock_bh(&sk->sk_receive_queue.lock);
4209 skb = skb_peek(&sk->sk_receive_queue);
4210 if (skb)
4211 amount = skb->len;
4212 spin_unlock_bh(&sk->sk_receive_queue.lock);
4213 return put_user(amount, (int __user *)arg);
4214 }
4215#ifdef CONFIG_INET
4216 case SIOCADDRT:
4217 case SIOCDELRT:
4218 case SIOCDARP:
4219 case SIOCGARP:
4220 case SIOCSARP:
4221 case SIOCGIFADDR:
4222 case SIOCSIFADDR:
4223 case SIOCGIFBRDADDR:
4224 case SIOCSIFBRDADDR:
4225 case SIOCGIFNETMASK:
4226 case SIOCSIFNETMASK:
4227 case SIOCGIFDSTADDR:
4228 case SIOCSIFDSTADDR:
4229 case SIOCSIFFLAGS:
4230 return inet_dgram_ops.ioctl(sock, cmd, arg);
4231#endif
4232
4233 default:
4234 return -ENOIOCTLCMD;
4235 }
4236 return 0;
4237}
4238
4239static __poll_t packet_poll(struct file *file, struct socket *sock,
4240 poll_table *wait)
4241{
4242 struct sock *sk = sock->sk;
4243 struct packet_sock *po = pkt_sk(sk);
4244 __poll_t mask = datagram_poll(file, sock, wait);
4245
4246 spin_lock_bh(&sk->sk_receive_queue.lock);
4247 if (po->rx_ring.pg_vec) {
4248 if (!packet_previous_rx_frame(po, &po->rx_ring,
4249 TP_STATUS_KERNEL))
4250 mask |= EPOLLIN | EPOLLRDNORM;
4251 }
4252 packet_rcv_try_clear_pressure(po);
4253 spin_unlock_bh(&sk->sk_receive_queue.lock);
4254 spin_lock_bh(&sk->sk_write_queue.lock);
4255 if (po->tx_ring.pg_vec) {
4256 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4257 mask |= EPOLLOUT | EPOLLWRNORM;
4258 }
4259 spin_unlock_bh(&sk->sk_write_queue.lock);
4260 return mask;
4261}
4262
4263
4264/* Dirty? Well, I still did not learn better way to account
4265 * for user mmaps.
4266 */
4267
4268static void packet_mm_open(struct vm_area_struct *vma)
4269{
4270 struct file *file = vma->vm_file;
4271 struct socket *sock = file->private_data;
4272 struct sock *sk = sock->sk;
4273
4274 if (sk)
4275 atomic_inc(&pkt_sk(sk)->mapped);
4276}
4277
4278static void packet_mm_close(struct vm_area_struct *vma)
4279{
4280 struct file *file = vma->vm_file;
4281 struct socket *sock = file->private_data;
4282 struct sock *sk = sock->sk;
4283
4284 if (sk)
4285 atomic_dec(&pkt_sk(sk)->mapped);
4286}
4287
4288static const struct vm_operations_struct packet_mmap_ops = {
4289 .open = packet_mm_open,
4290 .close = packet_mm_close,
4291};
4292
4293static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4294 unsigned int len)
4295{
4296 int i;
4297
4298 for (i = 0; i < len; i++) {
4299 if (likely(pg_vec[i].buffer)) {
4300 if (is_vmalloc_addr(pg_vec[i].buffer))
4301 vfree(pg_vec[i].buffer);
4302 else
4303 free_pages((unsigned long)pg_vec[i].buffer,
4304 order);
4305 pg_vec[i].buffer = NULL;
4306 }
4307 }
4308 kfree(pg_vec);
4309}
4310
4311static char *alloc_one_pg_vec_page(unsigned long order)
4312{
4313 char *buffer;
4314 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4315 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4316
4317 buffer = (char *) __get_free_pages(gfp_flags, order);
4318 if (buffer)
4319 return buffer;
4320
4321 /* __get_free_pages failed, fall back to vmalloc */
4322 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4323 if (buffer)
4324 return buffer;
4325
4326 /* vmalloc failed, lets dig into swap here */
4327 gfp_flags &= ~__GFP_NORETRY;
4328 buffer = (char *) __get_free_pages(gfp_flags, order);
4329 if (buffer)
4330 return buffer;
4331
4332 /* complete and utter failure */
4333 return NULL;
4334}
4335
4336static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4337{
4338 unsigned int block_nr = req->tp_block_nr;
4339 struct pgv *pg_vec;
4340 int i;
4341
4342 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4343 if (unlikely(!pg_vec))
4344 goto out;
4345
4346 for (i = 0; i < block_nr; i++) {
4347 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4348 if (unlikely(!pg_vec[i].buffer))
4349 goto out_free_pgvec;
4350 }
4351
4352out:
4353 return pg_vec;
4354
4355out_free_pgvec:
4356 free_pg_vec(pg_vec, order, block_nr);
4357 pg_vec = NULL;
4358 goto out;
4359}
4360
4361static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4362 int closing, int tx_ring)
4363{
4364 struct pgv *pg_vec = NULL;
4365 struct packet_sock *po = pkt_sk(sk);
4366 unsigned long *rx_owner_map = NULL;
4367 int was_running, order = 0;
4368 struct packet_ring_buffer *rb;
4369 struct sk_buff_head *rb_queue;
4370 __be16 num;
4371 int err;
4372 /* Added to avoid minimal code churn */
4373 struct tpacket_req *req = &req_u->req;
4374
4375 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4376 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4377
4378 err = -EBUSY;
4379 if (!closing) {
4380 if (atomic_read(&po->mapped))
4381 goto out;
4382 if (packet_read_pending(rb))
4383 goto out;
4384 }
4385
4386 if (req->tp_block_nr) {
4387 unsigned int min_frame_size;
4388
4389 /* Sanity tests and some calculations */
4390 err = -EBUSY;
4391 if (unlikely(rb->pg_vec))
4392 goto out;
4393
4394 switch (po->tp_version) {
4395 case TPACKET_V1:
4396 po->tp_hdrlen = TPACKET_HDRLEN;
4397 break;
4398 case TPACKET_V2:
4399 po->tp_hdrlen = TPACKET2_HDRLEN;
4400 break;
4401 case TPACKET_V3:
4402 po->tp_hdrlen = TPACKET3_HDRLEN;
4403 break;
4404 }
4405
4406 err = -EINVAL;
4407 if (unlikely((int)req->tp_block_size <= 0))
4408 goto out;
4409 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4410 goto out;
4411 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4412 if (po->tp_version >= TPACKET_V3 &&
4413 req->tp_block_size <
4414 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4415 goto out;
4416 if (unlikely(req->tp_frame_size < min_frame_size))
4417 goto out;
4418 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4419 goto out;
4420
4421 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4422 if (unlikely(rb->frames_per_block == 0))
4423 goto out;
4424 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4425 goto out;
4426 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4427 req->tp_frame_nr))
4428 goto out;
4429
4430 err = -ENOMEM;
4431 order = get_order(req->tp_block_size);
4432 pg_vec = alloc_pg_vec(req, order);
4433 if (unlikely(!pg_vec))
4434 goto out;
4435 switch (po->tp_version) {
4436 case TPACKET_V3:
4437 /* Block transmit is not supported yet */
4438 if (!tx_ring) {
4439 init_prb_bdqc(po, rb, pg_vec, req_u);
4440 } else {
4441 struct tpacket_req3 *req3 = &req_u->req3;
4442
4443 if (req3->tp_retire_blk_tov ||
4444 req3->tp_sizeof_priv ||
4445 req3->tp_feature_req_word) {
4446 err = -EINVAL;
4447 goto out_free_pg_vec;
4448 }
4449 }
4450 break;
4451 default:
4452 if (!tx_ring) {
4453 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4454 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4455 if (!rx_owner_map)
4456 goto out_free_pg_vec;
4457 }
4458 break;
4459 }
4460 }
4461 /* Done */
4462 else {
4463 err = -EINVAL;
4464 if (unlikely(req->tp_frame_nr))
4465 goto out;
4466 }
4467
4468
4469 /* Detach socket from network */
4470 spin_lock(&po->bind_lock);
4471 was_running = po->running;
4472 num = po->num;
4473 if (was_running) {
4474 WRITE_ONCE(po->num, 0);
4475 __unregister_prot_hook(sk, false);
4476 }
4477 spin_unlock(&po->bind_lock);
4478
4479 synchronize_net();
4480
4481 err = -EBUSY;
4482 mutex_lock(&po->pg_vec_lock);
4483 if (closing || atomic_read(&po->mapped) == 0) {
4484 err = 0;
4485 spin_lock_bh(&rb_queue->lock);
4486 swap(rb->pg_vec, pg_vec);
4487 if (po->tp_version <= TPACKET_V2)
4488 swap(rb->rx_owner_map, rx_owner_map);
4489 rb->frame_max = (req->tp_frame_nr - 1);
4490 rb->head = 0;
4491 rb->frame_size = req->tp_frame_size;
4492 spin_unlock_bh(&rb_queue->lock);
4493
4494 swap(rb->pg_vec_order, order);
4495 swap(rb->pg_vec_len, req->tp_block_nr);
4496
4497 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4498 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4499 tpacket_rcv : packet_rcv;
4500 skb_queue_purge(rb_queue);
4501 if (atomic_read(&po->mapped))
4502 pr_err("packet_mmap: vma is busy: %d\n",
4503 atomic_read(&po->mapped));
4504 }
4505 mutex_unlock(&po->pg_vec_lock);
4506
4507 spin_lock(&po->bind_lock);
4508 if (was_running) {
4509 WRITE_ONCE(po->num, num);
4510 register_prot_hook(sk);
4511 }
4512 spin_unlock(&po->bind_lock);
4513 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4514 /* Because we don't support block-based V3 on tx-ring */
4515 if (!tx_ring)
4516 prb_shutdown_retire_blk_timer(po, rb_queue);
4517 }
4518
4519out_free_pg_vec:
4520 if (pg_vec) {
4521 bitmap_free(rx_owner_map);
4522 free_pg_vec(pg_vec, order, req->tp_block_nr);
4523 }
4524out:
4525 return err;
4526}
4527
4528static int packet_mmap(struct file *file, struct socket *sock,
4529 struct vm_area_struct *vma)
4530{
4531 struct sock *sk = sock->sk;
4532 struct packet_sock *po = pkt_sk(sk);
4533 unsigned long size, expected_size;
4534 struct packet_ring_buffer *rb;
4535 unsigned long start;
4536 int err = -EINVAL;
4537 int i;
4538
4539 if (vma->vm_pgoff)
4540 return -EINVAL;
4541
4542 mutex_lock(&po->pg_vec_lock);
4543
4544 expected_size = 0;
4545 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4546 if (rb->pg_vec) {
4547 expected_size += rb->pg_vec_len
4548 * rb->pg_vec_pages
4549 * PAGE_SIZE;
4550 }
4551 }
4552
4553 if (expected_size == 0)
4554 goto out;
4555
4556 size = vma->vm_end - vma->vm_start;
4557 if (size != expected_size)
4558 goto out;
4559
4560 start = vma->vm_start;
4561 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4562 if (rb->pg_vec == NULL)
4563 continue;
4564
4565 for (i = 0; i < rb->pg_vec_len; i++) {
4566 struct page *page;
4567 void *kaddr = rb->pg_vec[i].buffer;
4568 int pg_num;
4569
4570 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4571 page = pgv_to_page(kaddr);
4572 err = vm_insert_page(vma, start, page);
4573 if (unlikely(err))
4574 goto out;
4575 start += PAGE_SIZE;
4576 kaddr += PAGE_SIZE;
4577 }
4578 }
4579 }
4580
4581 atomic_inc(&po->mapped);
4582 vma->vm_ops = &packet_mmap_ops;
4583 err = 0;
4584
4585out:
4586 mutex_unlock(&po->pg_vec_lock);
4587 return err;
4588}
4589
4590static const struct proto_ops packet_ops_spkt = {
4591 .family = PF_PACKET,
4592 .owner = THIS_MODULE,
4593 .release = packet_release,
4594 .bind = packet_bind_spkt,
4595 .connect = sock_no_connect,
4596 .socketpair = sock_no_socketpair,
4597 .accept = sock_no_accept,
4598 .getname = packet_getname_spkt,
4599 .poll = datagram_poll,
4600 .ioctl = packet_ioctl,
4601 .gettstamp = sock_gettstamp,
4602 .listen = sock_no_listen,
4603 .shutdown = sock_no_shutdown,
4604 .sendmsg = packet_sendmsg_spkt,
4605 .recvmsg = packet_recvmsg,
4606 .mmap = sock_no_mmap,
4607 .sendpage = sock_no_sendpage,
4608};
4609
4610static const struct proto_ops packet_ops = {
4611 .family = PF_PACKET,
4612 .owner = THIS_MODULE,
4613 .release = packet_release,
4614 .bind = packet_bind,
4615 .connect = sock_no_connect,
4616 .socketpair = sock_no_socketpair,
4617 .accept = sock_no_accept,
4618 .getname = packet_getname,
4619 .poll = packet_poll,
4620 .ioctl = packet_ioctl,
4621 .gettstamp = sock_gettstamp,
4622 .listen = sock_no_listen,
4623 .shutdown = sock_no_shutdown,
4624 .setsockopt = packet_setsockopt,
4625 .getsockopt = packet_getsockopt,
4626 .sendmsg = packet_sendmsg,
4627 .recvmsg = packet_recvmsg,
4628 .mmap = packet_mmap,
4629 .sendpage = sock_no_sendpage,
4630};
4631
4632static const struct net_proto_family packet_family_ops = {
4633 .family = PF_PACKET,
4634 .create = packet_create,
4635 .owner = THIS_MODULE,
4636};
4637
4638static struct notifier_block packet_netdev_notifier = {
4639 .notifier_call = packet_notifier,
4640};
4641
4642#ifdef CONFIG_PROC_FS
4643
4644static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4645 __acquires(RCU)
4646{
4647 struct net *net = seq_file_net(seq);
4648
4649 rcu_read_lock();
4650 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4651}
4652
4653static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4654{
4655 struct net *net = seq_file_net(seq);
4656 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4657}
4658
4659static void packet_seq_stop(struct seq_file *seq, void *v)
4660 __releases(RCU)
4661{
4662 rcu_read_unlock();
4663}
4664
4665static int packet_seq_show(struct seq_file *seq, void *v)
4666{
4667 if (v == SEQ_START_TOKEN)
4668 seq_printf(seq,
4669 "%*sRefCnt Type Proto Iface R Rmem User Inode\n",
4670 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4671 else {
4672 struct sock *s = sk_entry(v);
4673 const struct packet_sock *po = pkt_sk(s);
4674
4675 seq_printf(seq,
4676 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4677 s,
4678 refcount_read(&s->sk_refcnt),
4679 s->sk_type,
4680 ntohs(READ_ONCE(po->num)),
4681 READ_ONCE(po->ifindex),
4682 po->running,
4683 atomic_read(&s->sk_rmem_alloc),
4684 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4685 sock_i_ino(s));
4686 }
4687
4688 return 0;
4689}
4690
4691static const struct seq_operations packet_seq_ops = {
4692 .start = packet_seq_start,
4693 .next = packet_seq_next,
4694 .stop = packet_seq_stop,
4695 .show = packet_seq_show,
4696};
4697#endif
4698
4699static int __net_init packet_net_init(struct net *net)
4700{
4701 mutex_init(&net->packet.sklist_lock);
4702 INIT_HLIST_HEAD(&net->packet.sklist);
4703
4704#ifdef CONFIG_PROC_FS
4705 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4706 sizeof(struct seq_net_private)))
4707 return -ENOMEM;
4708#endif /* CONFIG_PROC_FS */
4709
4710 return 0;
4711}
4712
4713static void __net_exit packet_net_exit(struct net *net)
4714{
4715 remove_proc_entry("packet", net->proc_net);
4716 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4717}
4718
4719static struct pernet_operations packet_net_ops = {
4720 .init = packet_net_init,
4721 .exit = packet_net_exit,
4722};
4723
4724
4725static void __exit packet_exit(void)
4726{
4727 sock_unregister(PF_PACKET);
4728 proto_unregister(&packet_proto);
4729 unregister_netdevice_notifier(&packet_netdev_notifier);
4730 unregister_pernet_subsys(&packet_net_ops);
4731}
4732
4733static int __init packet_init(void)
4734{
4735 int rc;
4736
4737 rc = register_pernet_subsys(&packet_net_ops);
4738 if (rc)
4739 goto out;
4740 rc = register_netdevice_notifier(&packet_netdev_notifier);
4741 if (rc)
4742 goto out_pernet;
4743 rc = proto_register(&packet_proto, 0);
4744 if (rc)
4745 goto out_notifier;
4746 rc = sock_register(&packet_family_ops);
4747 if (rc)
4748 goto out_proto;
4749
4750 return 0;
4751
4752out_proto:
4753 proto_unregister(&packet_proto);
4754out_notifier:
4755 unregister_netdevice_notifier(&packet_netdev_notifier);
4756out_pernet:
4757 unregister_pernet_subsys(&packet_net_ops);
4758out:
4759 return rc;
4760}
4761
4762module_init(packet_init);
4763module_exit(packet_exit);
4764MODULE_LICENSE("GPL");
4765MODULE_ALIAS_NETPROTO(PF_PACKET);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * PACKET - implements raw packet sockets.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 *
13 * Fixes:
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
42 * and packet_mreq.
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * layer.
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
47 */
48
49#include <linux/ethtool.h>
50#include <linux/types.h>
51#include <linux/mm.h>
52#include <linux/capability.h>
53#include <linux/fcntl.h>
54#include <linux/socket.h>
55#include <linux/in.h>
56#include <linux/inet.h>
57#include <linux/netdevice.h>
58#include <linux/if_packet.h>
59#include <linux/wireless.h>
60#include <linux/kernel.h>
61#include <linux/kmod.h>
62#include <linux/slab.h>
63#include <linux/vmalloc.h>
64#include <net/net_namespace.h>
65#include <net/ip.h>
66#include <net/protocol.h>
67#include <linux/skbuff.h>
68#include <net/sock.h>
69#include <linux/errno.h>
70#include <linux/timer.h>
71#include <linux/uaccess.h>
72#include <asm/ioctls.h>
73#include <asm/page.h>
74#include <asm/cacheflush.h>
75#include <asm/io.h>
76#include <linux/proc_fs.h>
77#include <linux/seq_file.h>
78#include <linux/poll.h>
79#include <linux/module.h>
80#include <linux/init.h>
81#include <linux/mutex.h>
82#include <linux/if_vlan.h>
83#include <linux/virtio_net.h>
84#include <linux/errqueue.h>
85#include <linux/net_tstamp.h>
86#include <linux/percpu.h>
87#ifdef CONFIG_INET
88#include <net/inet_common.h>
89#endif
90#include <linux/bpf.h>
91#include <net/compat.h>
92
93#include "internal.h"
94
95/*
96 Assumptions:
97 - If the device has no dev->header_ops->create, there is no LL header
98 visible above the device. In this case, its hard_header_len should be 0.
99 The device may prepend its own header internally. In this case, its
100 needed_headroom should be set to the space needed for it to add its
101 internal header.
102 For example, a WiFi driver pretending to be an Ethernet driver should
103 set its hard_header_len to be the Ethernet header length, and set its
104 needed_headroom to be (the real WiFi header length - the fake Ethernet
105 header length).
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
108
109On receive:
110-----------
111
112Incoming, dev_has_header(dev) == true
113 mac_header -> ll header
114 data -> data
115
116Outgoing, dev_has_header(dev) == true
117 mac_header -> ll header
118 data -> ll header
119
120Incoming, dev_has_header(dev) == false
121 mac_header -> data
122 However drivers often make it point to the ll header.
123 This is incorrect because the ll header should be invisible to us.
124 data -> data
125
126Outgoing, dev_has_header(dev) == false
127 mac_header -> data. ll header is invisible to us.
128 data -> data
129
130Resume
131 If dev_has_header(dev) == false we are unable to restore the ll header,
132 because it is invisible to us.
133
134
135On transmit:
136------------
137
138dev_has_header(dev) == true
139 mac_header -> ll header
140 data -> ll header
141
142dev_has_header(dev) == false (ll header is invisible to us)
143 mac_header -> data
144 data -> data
145
146 We should set network_header on output to the correct position,
147 packet classifier depends on it.
148 */
149
150/* Private packet socket structures. */
151
152/* identical to struct packet_mreq except it has
153 * a longer address field.
154 */
155struct packet_mreq_max {
156 int mr_ifindex;
157 unsigned short mr_type;
158 unsigned short mr_alen;
159 unsigned char mr_address[MAX_ADDR_LEN];
160};
161
162union tpacket_uhdr {
163 struct tpacket_hdr *h1;
164 struct tpacket2_hdr *h2;
165 struct tpacket3_hdr *h3;
166 void *raw;
167};
168
169static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
170 int closing, int tx_ring);
171
172#define V3_ALIGNMENT (8)
173
174#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175
176#define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178
179#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
180#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
181#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
182#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
183#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
184#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
185
186struct packet_sock;
187static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
188 struct packet_type *pt, struct net_device *orig_dev);
189
190static void *packet_previous_frame(struct packet_sock *po,
191 struct packet_ring_buffer *rb,
192 int status);
193static void packet_increment_head(struct packet_ring_buffer *buff);
194static int prb_curr_blk_in_use(struct tpacket_block_desc *);
195static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
196 struct packet_sock *);
197static void prb_retire_current_block(struct tpacket_kbdq_core *,
198 struct packet_sock *, unsigned int status);
199static int prb_queue_frozen(struct tpacket_kbdq_core *);
200static void prb_open_block(struct tpacket_kbdq_core *,
201 struct tpacket_block_desc *);
202static void prb_retire_rx_blk_timer_expired(struct timer_list *);
203static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
204static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
205static void prb_clear_rxhash(struct tpacket_kbdq_core *,
206 struct tpacket3_hdr *);
207static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
208 struct tpacket3_hdr *);
209static void packet_flush_mclist(struct sock *sk);
210static u16 packet_pick_tx_queue(struct sk_buff *skb);
211
212struct packet_skb_cb {
213 union {
214 struct sockaddr_pkt pkt;
215 union {
216 /* Trick: alias skb original length with
217 * ll.sll_family and ll.protocol in order
218 * to save room.
219 */
220 unsigned int origlen;
221 struct sockaddr_ll ll;
222 };
223 } sa;
224};
225
226#define vio_le() virtio_legacy_is_little_endian()
227
228#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
229
230#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
231#define GET_PBLOCK_DESC(x, bid) \
232 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
233#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
234 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
235#define GET_NEXT_PRB_BLK_NUM(x) \
236 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
237 ((x)->kactive_blk_num+1) : 0)
238
239static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
240static void __fanout_link(struct sock *sk, struct packet_sock *po);
241
242static int packet_direct_xmit(struct sk_buff *skb)
243{
244 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
245}
246
247static struct net_device *packet_cached_dev_get(struct packet_sock *po)
248{
249 struct net_device *dev;
250
251 rcu_read_lock();
252 dev = rcu_dereference(po->cached_dev);
253 if (likely(dev))
254 dev_hold(dev);
255 rcu_read_unlock();
256
257 return dev;
258}
259
260static void packet_cached_dev_assign(struct packet_sock *po,
261 struct net_device *dev)
262{
263 rcu_assign_pointer(po->cached_dev, dev);
264}
265
266static void packet_cached_dev_reset(struct packet_sock *po)
267{
268 RCU_INIT_POINTER(po->cached_dev, NULL);
269}
270
271static bool packet_use_direct_xmit(const struct packet_sock *po)
272{
273 return po->xmit == packet_direct_xmit;
274}
275
276static u16 packet_pick_tx_queue(struct sk_buff *skb)
277{
278 struct net_device *dev = skb->dev;
279 const struct net_device_ops *ops = dev->netdev_ops;
280 int cpu = raw_smp_processor_id();
281 u16 queue_index;
282
283#ifdef CONFIG_XPS
284 skb->sender_cpu = cpu + 1;
285#endif
286 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
287 if (ops->ndo_select_queue) {
288 queue_index = ops->ndo_select_queue(dev, skb, NULL);
289 queue_index = netdev_cap_txqueue(dev, queue_index);
290 } else {
291 queue_index = netdev_pick_tx(dev, skb, NULL);
292 }
293
294 return queue_index;
295}
296
297/* __register_prot_hook must be invoked through register_prot_hook
298 * or from a context in which asynchronous accesses to the packet
299 * socket is not possible (packet_create()).
300 */
301static void __register_prot_hook(struct sock *sk)
302{
303 struct packet_sock *po = pkt_sk(sk);
304
305 if (!po->running) {
306 if (po->fanout)
307 __fanout_link(sk, po);
308 else
309 dev_add_pack(&po->prot_hook);
310
311 sock_hold(sk);
312 po->running = 1;
313 }
314}
315
316static void register_prot_hook(struct sock *sk)
317{
318 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
319 __register_prot_hook(sk);
320}
321
322/* If the sync parameter is true, we will temporarily drop
323 * the po->bind_lock and do a synchronize_net to make sure no
324 * asynchronous packet processing paths still refer to the elements
325 * of po->prot_hook. If the sync parameter is false, it is the
326 * callers responsibility to take care of this.
327 */
328static void __unregister_prot_hook(struct sock *sk, bool sync)
329{
330 struct packet_sock *po = pkt_sk(sk);
331
332 lockdep_assert_held_once(&po->bind_lock);
333
334 po->running = 0;
335
336 if (po->fanout)
337 __fanout_unlink(sk, po);
338 else
339 __dev_remove_pack(&po->prot_hook);
340
341 __sock_put(sk);
342
343 if (sync) {
344 spin_unlock(&po->bind_lock);
345 synchronize_net();
346 spin_lock(&po->bind_lock);
347 }
348}
349
350static void unregister_prot_hook(struct sock *sk, bool sync)
351{
352 struct packet_sock *po = pkt_sk(sk);
353
354 if (po->running)
355 __unregister_prot_hook(sk, sync);
356}
357
358static inline struct page * __pure pgv_to_page(void *addr)
359{
360 if (is_vmalloc_addr(addr))
361 return vmalloc_to_page(addr);
362 return virt_to_page(addr);
363}
364
365static void __packet_set_status(struct packet_sock *po, void *frame, int status)
366{
367 union tpacket_uhdr h;
368
369 h.raw = frame;
370 switch (po->tp_version) {
371 case TPACKET_V1:
372 h.h1->tp_status = status;
373 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
374 break;
375 case TPACKET_V2:
376 h.h2->tp_status = status;
377 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
378 break;
379 case TPACKET_V3:
380 h.h3->tp_status = status;
381 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
382 break;
383 default:
384 WARN(1, "TPACKET version not supported.\n");
385 BUG();
386 }
387
388 smp_wmb();
389}
390
391static int __packet_get_status(const struct packet_sock *po, void *frame)
392{
393 union tpacket_uhdr h;
394
395 smp_rmb();
396
397 h.raw = frame;
398 switch (po->tp_version) {
399 case TPACKET_V1:
400 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
401 return h.h1->tp_status;
402 case TPACKET_V2:
403 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
404 return h.h2->tp_status;
405 case TPACKET_V3:
406 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
407 return h.h3->tp_status;
408 default:
409 WARN(1, "TPACKET version not supported.\n");
410 BUG();
411 return 0;
412 }
413}
414
415static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
416 unsigned int flags)
417{
418 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
419
420 if (shhwtstamps &&
421 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
422 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
423 return TP_STATUS_TS_RAW_HARDWARE;
424
425 if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
426 ktime_to_timespec64_cond(skb->tstamp, ts))
427 return TP_STATUS_TS_SOFTWARE;
428
429 return 0;
430}
431
432static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
433 struct sk_buff *skb)
434{
435 union tpacket_uhdr h;
436 struct timespec64 ts;
437 __u32 ts_status;
438
439 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
440 return 0;
441
442 h.raw = frame;
443 /*
444 * versions 1 through 3 overflow the timestamps in y2106, since they
445 * all store the seconds in a 32-bit unsigned integer.
446 * If we create a version 4, that should have a 64-bit timestamp,
447 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
448 * nanoseconds.
449 */
450 switch (po->tp_version) {
451 case TPACKET_V1:
452 h.h1->tp_sec = ts.tv_sec;
453 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
454 break;
455 case TPACKET_V2:
456 h.h2->tp_sec = ts.tv_sec;
457 h.h2->tp_nsec = ts.tv_nsec;
458 break;
459 case TPACKET_V3:
460 h.h3->tp_sec = ts.tv_sec;
461 h.h3->tp_nsec = ts.tv_nsec;
462 break;
463 default:
464 WARN(1, "TPACKET version not supported.\n");
465 BUG();
466 }
467
468 /* one flush is safe, as both fields always lie on the same cacheline */
469 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
470 smp_wmb();
471
472 return ts_status;
473}
474
475static void *packet_lookup_frame(const struct packet_sock *po,
476 const struct packet_ring_buffer *rb,
477 unsigned int position,
478 int status)
479{
480 unsigned int pg_vec_pos, frame_offset;
481 union tpacket_uhdr h;
482
483 pg_vec_pos = position / rb->frames_per_block;
484 frame_offset = position % rb->frames_per_block;
485
486 h.raw = rb->pg_vec[pg_vec_pos].buffer +
487 (frame_offset * rb->frame_size);
488
489 if (status != __packet_get_status(po, h.raw))
490 return NULL;
491
492 return h.raw;
493}
494
495static void *packet_current_frame(struct packet_sock *po,
496 struct packet_ring_buffer *rb,
497 int status)
498{
499 return packet_lookup_frame(po, rb, rb->head, status);
500}
501
502static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
503{
504 del_timer_sync(&pkc->retire_blk_timer);
505}
506
507static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
508 struct sk_buff_head *rb_queue)
509{
510 struct tpacket_kbdq_core *pkc;
511
512 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
513
514 spin_lock_bh(&rb_queue->lock);
515 pkc->delete_blk_timer = 1;
516 spin_unlock_bh(&rb_queue->lock);
517
518 prb_del_retire_blk_timer(pkc);
519}
520
521static void prb_setup_retire_blk_timer(struct packet_sock *po)
522{
523 struct tpacket_kbdq_core *pkc;
524
525 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
526 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
527 0);
528 pkc->retire_blk_timer.expires = jiffies;
529}
530
531static int prb_calc_retire_blk_tmo(struct packet_sock *po,
532 int blk_size_in_bytes)
533{
534 struct net_device *dev;
535 unsigned int mbits, div;
536 struct ethtool_link_ksettings ecmd;
537 int err;
538
539 rtnl_lock();
540 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
541 if (unlikely(!dev)) {
542 rtnl_unlock();
543 return DEFAULT_PRB_RETIRE_TOV;
544 }
545 err = __ethtool_get_link_ksettings(dev, &ecmd);
546 rtnl_unlock();
547 if (err)
548 return DEFAULT_PRB_RETIRE_TOV;
549
550 /* If the link speed is so slow you don't really
551 * need to worry about perf anyways
552 */
553 if (ecmd.base.speed < SPEED_1000 ||
554 ecmd.base.speed == SPEED_UNKNOWN)
555 return DEFAULT_PRB_RETIRE_TOV;
556
557 div = ecmd.base.speed / 1000;
558 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
559
560 if (div)
561 mbits /= div;
562
563 if (div)
564 return mbits + 1;
565 return mbits;
566}
567
568static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
569 union tpacket_req_u *req_u)
570{
571 p1->feature_req_word = req_u->req3.tp_feature_req_word;
572}
573
574static void init_prb_bdqc(struct packet_sock *po,
575 struct packet_ring_buffer *rb,
576 struct pgv *pg_vec,
577 union tpacket_req_u *req_u)
578{
579 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
580 struct tpacket_block_desc *pbd;
581
582 memset(p1, 0x0, sizeof(*p1));
583
584 p1->knxt_seq_num = 1;
585 p1->pkbdq = pg_vec;
586 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
587 p1->pkblk_start = pg_vec[0].buffer;
588 p1->kblk_size = req_u->req3.tp_block_size;
589 p1->knum_blocks = req_u->req3.tp_block_nr;
590 p1->hdrlen = po->tp_hdrlen;
591 p1->version = po->tp_version;
592 p1->last_kactive_blk_num = 0;
593 po->stats.stats3.tp_freeze_q_cnt = 0;
594 if (req_u->req3.tp_retire_blk_tov)
595 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
596 else
597 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
598 req_u->req3.tp_block_size);
599 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
600 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
601 rwlock_init(&p1->blk_fill_in_prog_lock);
602
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}
608
609/* Do NOT update the last_blk_num first.
610 * Assumes sk_buff_head lock is held.
611 */
612static 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}
618
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 */
642static 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;
649
650 spin_lock(&po->sk.sk_receive_queue.lock);
651
652 frozen = prb_queue_frozen(pkc);
653 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
654
655 if (unlikely(pkc->delete_blk_timer))
656 goto out;
657
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 /* Waiting for skb_copy_bits to finish... */
669 write_lock(&pkc->blk_fill_in_prog_lock);
670 write_unlock(&pkc->blk_fill_in_prog_lock);
671 }
672
673 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
674 if (!frozen) {
675 if (!BLOCK_NUM_PKTS(pbd)) {
676 /* An empty block. Just refresh the timer. */
677 goto refresh_timer;
678 }
679 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
680 if (!prb_dispatch_next_block(pkc, po))
681 goto refresh_timer;
682 else
683 goto out;
684 } else {
685 /* Case 1. Queue was frozen because user-space was
686 * lagging behind.
687 */
688 if (prb_curr_blk_in_use(pbd)) {
689 /*
690 * Ok, user-space is still behind.
691 * So just refresh the timer.
692 */
693 goto refresh_timer;
694 } else {
695 /* Case 2. queue was frozen,user-space caught up,
696 * now the link went idle && the timer fired.
697 * We don't have a block to close.So we open this
698 * block and restart the timer.
699 * opening a block thaws the queue,restarts timer
700 * Thawing/timer-refresh is a side effect.
701 */
702 prb_open_block(pkc, pbd);
703 goto out;
704 }
705 }
706 }
707
708refresh_timer:
709 _prb_refresh_rx_retire_blk_timer(pkc);
710
711out:
712 spin_unlock(&po->sk.sk_receive_queue.lock);
713}
714
715static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
716 struct tpacket_block_desc *pbd1, __u32 status)
717{
718 /* Flush everything minus the block header */
719
720#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
721 u8 *start, *end;
722
723 start = (u8 *)pbd1;
724
725 /* Skip the block header(we know header WILL fit in 4K) */
726 start += PAGE_SIZE;
727
728 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
729 for (; start < end; start += PAGE_SIZE)
730 flush_dcache_page(pgv_to_page(start));
731
732 smp_wmb();
733#endif
734
735 /* Now update the block status. */
736
737 BLOCK_STATUS(pbd1) = status;
738
739 /* Flush the block header */
740
741#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
742 start = (u8 *)pbd1;
743 flush_dcache_page(pgv_to_page(start));
744
745 smp_wmb();
746#endif
747}
748
749/*
750 * Side effect:
751 *
752 * 1) flush the block
753 * 2) Increment active_blk_num
754 *
755 * Note:We DONT refresh the timer on purpose.
756 * Because almost always the next block will be opened.
757 */
758static void prb_close_block(struct tpacket_kbdq_core *pkc1,
759 struct tpacket_block_desc *pbd1,
760 struct packet_sock *po, unsigned int stat)
761{
762 __u32 status = TP_STATUS_USER | stat;
763
764 struct tpacket3_hdr *last_pkt;
765 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
766 struct sock *sk = &po->sk;
767
768 if (atomic_read(&po->tp_drops))
769 status |= TP_STATUS_LOSING;
770
771 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
772 last_pkt->tp_next_offset = 0;
773
774 /* Get the ts of the last pkt */
775 if (BLOCK_NUM_PKTS(pbd1)) {
776 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
777 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
778 } else {
779 /* Ok, we tmo'd - so get the current time.
780 *
781 * It shouldn't really happen as we don't close empty
782 * blocks. See prb_retire_rx_blk_timer_expired().
783 */
784 struct timespec64 ts;
785 ktime_get_real_ts64(&ts);
786 h1->ts_last_pkt.ts_sec = ts.tv_sec;
787 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
788 }
789
790 smp_wmb();
791
792 /* Flush the block */
793 prb_flush_block(pkc1, pbd1, status);
794
795 sk->sk_data_ready(sk);
796
797 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
798}
799
800static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
801{
802 pkc->reset_pending_on_curr_blk = 0;
803}
804
805/*
806 * Side effect of opening a block:
807 *
808 * 1) prb_queue is thawed.
809 * 2) retire_blk_timer is refreshed.
810 *
811 */
812static void prb_open_block(struct tpacket_kbdq_core *pkc1,
813 struct tpacket_block_desc *pbd1)
814{
815 struct timespec64 ts;
816 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
817
818 smp_rmb();
819
820 /* We could have just memset this but we will lose the
821 * flexibility of making the priv area sticky
822 */
823
824 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
825 BLOCK_NUM_PKTS(pbd1) = 0;
826 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
827
828 ktime_get_real_ts64(&ts);
829
830 h1->ts_first_pkt.ts_sec = ts.tv_sec;
831 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
832
833 pkc1->pkblk_start = (char *)pbd1;
834 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
835
836 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
837 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
838
839 pbd1->version = pkc1->version;
840 pkc1->prev = pkc1->nxt_offset;
841 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
842
843 prb_thaw_queue(pkc1);
844 _prb_refresh_rx_retire_blk_timer(pkc1);
845
846 smp_wmb();
847}
848
849/*
850 * Queue freeze logic:
851 * 1) Assume tp_block_nr = 8 blocks.
852 * 2) At time 't0', user opens Rx ring.
853 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854 * 4) user-space is either sleeping or processing block '0'.
855 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856 * it will close block-7,loop around and try to fill block '0'.
857 * call-flow:
858 * __packet_lookup_frame_in_block
859 * prb_retire_current_block()
860 * prb_dispatch_next_block()
861 * |->(BLOCK_STATUS == USER) evaluates to true
862 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
863 * 6) Now there are two cases:
864 * 6.1) Link goes idle right after the queue is frozen.
865 * But remember, the last open_block() refreshed the timer.
866 * When this timer expires,it will refresh itself so that we can
867 * re-open block-0 in near future.
868 * 6.2) Link is busy and keeps on receiving packets. This is a simple
869 * case and __packet_lookup_frame_in_block will check if block-0
870 * is free and can now be re-used.
871 */
872static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873 struct packet_sock *po)
874{
875 pkc->reset_pending_on_curr_blk = 1;
876 po->stats.stats3.tp_freeze_q_cnt++;
877}
878
879#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
880
881/*
882 * If the next block is free then we will dispatch it
883 * and return a good offset.
884 * Else, we will freeze the queue.
885 * So, caller must check the return value.
886 */
887static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888 struct packet_sock *po)
889{
890 struct tpacket_block_desc *pbd;
891
892 smp_rmb();
893
894 /* 1. Get current block num */
895 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
896
897 /* 2. If this block is currently in_use then freeze the queue */
898 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899 prb_freeze_queue(pkc, po);
900 return NULL;
901 }
902
903 /*
904 * 3.
905 * open this block and return the offset where the first packet
906 * needs to get stored.
907 */
908 prb_open_block(pkc, pbd);
909 return (void *)pkc->nxt_offset;
910}
911
912static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913 struct packet_sock *po, unsigned int status)
914{
915 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
916
917 /* retire/close the current block */
918 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
919 /*
920 * Plug the case where copy_bits() is in progress on
921 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922 * have space to copy the pkt in the current block and
923 * called prb_retire_current_block()
924 *
925 * We don't need to worry about the TMO case because
926 * the timer-handler already handled this case.
927 */
928 if (!(status & TP_STATUS_BLK_TMO)) {
929 /* Waiting for skb_copy_bits to finish... */
930 write_lock(&pkc->blk_fill_in_prog_lock);
931 write_unlock(&pkc->blk_fill_in_prog_lock);
932 }
933 prb_close_block(pkc, pbd, po, status);
934 return;
935 }
936}
937
938static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
939{
940 return TP_STATUS_USER & BLOCK_STATUS(pbd);
941}
942
943static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
944{
945 return pkc->reset_pending_on_curr_blk;
946}
947
948static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
949 __releases(&pkc->blk_fill_in_prog_lock)
950{
951 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
952
953 read_unlock(&pkc->blk_fill_in_prog_lock);
954}
955
956static 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}
961
962static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
963 struct tpacket3_hdr *ppd)
964{
965 ppd->hv1.tp_rxhash = 0;
966}
967
968static 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}
981
982static 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);
987
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}
993
994static void prb_fill_curr_block(char *curr,
995 struct tpacket_kbdq_core *pkc,
996 struct tpacket_block_desc *pbd,
997 unsigned int len)
998 __acquires(&pkc->blk_fill_in_prog_lock)
999{
1000 struct tpacket3_hdr *ppd;
1001
1002 ppd = (struct tpacket3_hdr *)curr;
1003 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1004 pkc->prev = curr;
1005 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1007 BLOCK_NUM_PKTS(pbd) += 1;
1008 read_lock(&pkc->blk_fill_in_prog_lock);
1009 prb_run_all_ft_ops(pkc, ppd);
1010}
1011
1012/* Assumes caller has the sk->rx_queue.lock */
1013static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1014 struct sk_buff *skb,
1015 unsigned int len
1016 )
1017{
1018 struct tpacket_kbdq_core *pkc;
1019 struct tpacket_block_desc *pbd;
1020 char *curr, *end;
1021
1022 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1023 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024
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 }
1044
1045 smp_mb();
1046 curr = pkc->nxt_offset;
1047 pkc->skb = skb;
1048 end = (char *)pbd + pkc->kblk_size;
1049
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 }
1055
1056 /* Ok, close the current block */
1057 prb_retire_current_block(pkc, po, 0);
1058
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 }
1066
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}
1073
1074static 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, len);
1087 default:
1088 WARN(1, "TPACKET version not supported\n");
1089 BUG();
1090 return NULL;
1091 }
1092}
1093
1094static void *prb_lookup_block(const struct packet_sock *po,
1095 const 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);
1101
1102 if (status != BLOCK_STATUS(pbd))
1103 return NULL;
1104 return pbd;
1105}
1106
1107static 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}
1116
1117/* Assumes caller has held the rx_queue.lock */
1118static 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}
1125
1126static 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);
1132
1133 return __prb_previous_block(po, rb, status);
1134}
1135
1136static 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}
1150
1151static 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}
1158
1159static void packet_increment_head(struct packet_ring_buffer *buff)
1160{
1161 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1162}
1163
1164static void packet_inc_pending(struct packet_ring_buffer *rb)
1165{
1166 this_cpu_inc(*rb->pending_refcnt);
1167}
1168
1169static void packet_dec_pending(struct packet_ring_buffer *rb)
1170{
1171 this_cpu_dec(*rb->pending_refcnt);
1172}
1173
1174static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1175{
1176 unsigned int refcnt = 0;
1177 int cpu;
1178
1179 /* We don't use pending refcount in rx_ring. */
1180 if (rb->pending_refcnt == NULL)
1181 return 0;
1182
1183 for_each_possible_cpu(cpu)
1184 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1185
1186 return refcnt;
1187}
1188
1189static int packet_alloc_pending(struct packet_sock *po)
1190{
1191 po->rx_ring.pending_refcnt = NULL;
1192
1193 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1194 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1195 return -ENOBUFS;
1196
1197 return 0;
1198}
1199
1200static void packet_free_pending(struct packet_sock *po)
1201{
1202 free_percpu(po->tx_ring.pending_refcnt);
1203}
1204
1205#define ROOM_POW_OFF 2
1206#define ROOM_NONE 0x0
1207#define ROOM_LOW 0x1
1208#define ROOM_NORMAL 0x2
1209
1210static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1211{
1212 int idx, len;
1213
1214 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1215 idx = READ_ONCE(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}
1222
1223static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1224{
1225 int idx, len;
1226
1227 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1228 idx = READ_ONCE(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}
1235
1236static int __packet_rcv_has_room(const struct packet_sock *po,
1237 const struct sk_buff *skb)
1238{
1239 const struct sock *sk = &po->sk;
1240 int ret = ROOM_NONE;
1241
1242 if (po->prot_hook.func != tpacket_rcv) {
1243 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1244 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1245 - (skb ? skb->truesize : 0);
1246
1247 if (avail > (rcvbuf >> ROOM_POW_OFF))
1248 return ROOM_NORMAL;
1249 else if (avail > 0)
1250 return ROOM_LOW;
1251 else
1252 return ROOM_NONE;
1253 }
1254
1255 if (po->tp_version == TPACKET_V3) {
1256 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1257 ret = ROOM_NORMAL;
1258 else if (__tpacket_v3_has_room(po, 0))
1259 ret = ROOM_LOW;
1260 } else {
1261 if (__tpacket_has_room(po, ROOM_POW_OFF))
1262 ret = ROOM_NORMAL;
1263 else if (__tpacket_has_room(po, 0))
1264 ret = ROOM_LOW;
1265 }
1266
1267 return ret;
1268}
1269
1270static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1271{
1272 int pressure, ret;
1273
1274 ret = __packet_rcv_has_room(po, skb);
1275 pressure = ret != ROOM_NORMAL;
1276
1277 if (READ_ONCE(po->pressure) != pressure)
1278 WRITE_ONCE(po->pressure, pressure);
1279
1280 return ret;
1281}
1282
1283static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1284{
1285 if (READ_ONCE(po->pressure) &&
1286 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1287 WRITE_ONCE(po->pressure, 0);
1288}
1289
1290static void packet_sock_destruct(struct sock *sk)
1291{
1292 skb_queue_purge(&sk->sk_error_queue);
1293
1294 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1295 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1296
1297 if (!sock_flag(sk, SOCK_DEAD)) {
1298 pr_err("Attempt to release alive packet socket: %p\n", sk);
1299 return;
1300 }
1301
1302 sk_refcnt_debug_dec(sk);
1303}
1304
1305static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1306{
1307 u32 *history = po->rollover->history;
1308 u32 victim, rxhash;
1309 int i, count = 0;
1310
1311 rxhash = skb_get_hash(skb);
1312 for (i = 0; i < ROLLOVER_HLEN; i++)
1313 if (READ_ONCE(history[i]) == rxhash)
1314 count++;
1315
1316 victim = prandom_u32() % ROLLOVER_HLEN;
1317
1318 /* Avoid dirtying the cache line if possible */
1319 if (READ_ONCE(history[victim]) != rxhash)
1320 WRITE_ONCE(history[victim], rxhash);
1321
1322 return count > (ROLLOVER_HLEN >> 1);
1323}
1324
1325static unsigned int fanout_demux_hash(struct packet_fanout *f,
1326 struct sk_buff *skb,
1327 unsigned int num)
1328{
1329 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1330}
1331
1332static unsigned int fanout_demux_lb(struct packet_fanout *f,
1333 struct sk_buff *skb,
1334 unsigned int num)
1335{
1336 unsigned int val = atomic_inc_return(&f->rr_cur);
1337
1338 return val % num;
1339}
1340
1341static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1342 struct sk_buff *skb,
1343 unsigned int num)
1344{
1345 return smp_processor_id() % num;
1346}
1347
1348static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1349 struct sk_buff *skb,
1350 unsigned int num)
1351{
1352 return prandom_u32_max(num);
1353}
1354
1355static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1356 struct sk_buff *skb,
1357 unsigned int idx, bool try_self,
1358 unsigned int num)
1359{
1360 struct packet_sock *po, *po_next, *po_skip = NULL;
1361 unsigned int i, j, room = ROOM_NONE;
1362
1363 po = pkt_sk(rcu_dereference(f->arr[idx]));
1364
1365 if (try_self) {
1366 room = packet_rcv_has_room(po, skb);
1367 if (room == ROOM_NORMAL ||
1368 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1369 return idx;
1370 po_skip = po;
1371 }
1372
1373 i = j = min_t(int, po->rollover->sock, num - 1);
1374 do {
1375 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1376 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1377 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1378 if (i != j)
1379 po->rollover->sock = i;
1380 atomic_long_inc(&po->rollover->num);
1381 if (room == ROOM_LOW)
1382 atomic_long_inc(&po->rollover->num_huge);
1383 return i;
1384 }
1385
1386 if (++i == num)
1387 i = 0;
1388 } while (i != j);
1389
1390 atomic_long_inc(&po->rollover->num_failed);
1391 return idx;
1392}
1393
1394static unsigned int fanout_demux_qm(struct packet_fanout *f,
1395 struct sk_buff *skb,
1396 unsigned int num)
1397{
1398 return skb_get_queue_mapping(skb) % num;
1399}
1400
1401static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1402 struct sk_buff *skb,
1403 unsigned int num)
1404{
1405 struct bpf_prog *prog;
1406 unsigned int ret = 0;
1407
1408 rcu_read_lock();
1409 prog = rcu_dereference(f->bpf_prog);
1410 if (prog)
1411 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1412 rcu_read_unlock();
1413
1414 return ret;
1415}
1416
1417static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1418{
1419 return f->flags & (flag >> 8);
1420}
1421
1422static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1423 struct packet_type *pt, struct net_device *orig_dev)
1424{
1425 struct packet_fanout *f = pt->af_packet_priv;
1426 unsigned int num = READ_ONCE(f->num_members);
1427 struct net *net = read_pnet(&f->net);
1428 struct packet_sock *po;
1429 unsigned int idx;
1430
1431 if (!net_eq(dev_net(dev), net) || !num) {
1432 kfree_skb(skb);
1433 return 0;
1434 }
1435
1436 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1437 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1438 if (!skb)
1439 return 0;
1440 }
1441 switch (f->type) {
1442 case PACKET_FANOUT_HASH:
1443 default:
1444 idx = fanout_demux_hash(f, skb, num);
1445 break;
1446 case PACKET_FANOUT_LB:
1447 idx = fanout_demux_lb(f, skb, num);
1448 break;
1449 case PACKET_FANOUT_CPU:
1450 idx = fanout_demux_cpu(f, skb, num);
1451 break;
1452 case PACKET_FANOUT_RND:
1453 idx = fanout_demux_rnd(f, skb, num);
1454 break;
1455 case PACKET_FANOUT_QM:
1456 idx = fanout_demux_qm(f, skb, num);
1457 break;
1458 case PACKET_FANOUT_ROLLOVER:
1459 idx = fanout_demux_rollover(f, skb, 0, false, num);
1460 break;
1461 case PACKET_FANOUT_CBPF:
1462 case PACKET_FANOUT_EBPF:
1463 idx = fanout_demux_bpf(f, skb, num);
1464 break;
1465 }
1466
1467 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1468 idx = fanout_demux_rollover(f, skb, idx, true, num);
1469
1470 po = pkt_sk(rcu_dereference(f->arr[idx]));
1471 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1472}
1473
1474DEFINE_MUTEX(fanout_mutex);
1475EXPORT_SYMBOL_GPL(fanout_mutex);
1476static LIST_HEAD(fanout_list);
1477static u16 fanout_next_id;
1478
1479static void __fanout_link(struct sock *sk, struct packet_sock *po)
1480{
1481 struct packet_fanout *f = po->fanout;
1482
1483 spin_lock(&f->lock);
1484 rcu_assign_pointer(f->arr[f->num_members], sk);
1485 smp_wmb();
1486 f->num_members++;
1487 if (f->num_members == 1)
1488 dev_add_pack(&f->prot_hook);
1489 spin_unlock(&f->lock);
1490}
1491
1492static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1493{
1494 struct packet_fanout *f = po->fanout;
1495 int i;
1496
1497 spin_lock(&f->lock);
1498 for (i = 0; i < f->num_members; i++) {
1499 if (rcu_dereference_protected(f->arr[i],
1500 lockdep_is_held(&f->lock)) == sk)
1501 break;
1502 }
1503 BUG_ON(i >= f->num_members);
1504 rcu_assign_pointer(f->arr[i],
1505 rcu_dereference_protected(f->arr[f->num_members - 1],
1506 lockdep_is_held(&f->lock)));
1507 f->num_members--;
1508 if (f->num_members == 0)
1509 __dev_remove_pack(&f->prot_hook);
1510 spin_unlock(&f->lock);
1511}
1512
1513static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1514{
1515 if (sk->sk_family != PF_PACKET)
1516 return false;
1517
1518 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1519}
1520
1521static void fanout_init_data(struct packet_fanout *f)
1522{
1523 switch (f->type) {
1524 case PACKET_FANOUT_LB:
1525 atomic_set(&f->rr_cur, 0);
1526 break;
1527 case PACKET_FANOUT_CBPF:
1528 case PACKET_FANOUT_EBPF:
1529 RCU_INIT_POINTER(f->bpf_prog, NULL);
1530 break;
1531 }
1532}
1533
1534static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1535{
1536 struct bpf_prog *old;
1537
1538 spin_lock(&f->lock);
1539 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1540 rcu_assign_pointer(f->bpf_prog, new);
1541 spin_unlock(&f->lock);
1542
1543 if (old) {
1544 synchronize_net();
1545 bpf_prog_destroy(old);
1546 }
1547}
1548
1549static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1550 unsigned int len)
1551{
1552 struct bpf_prog *new;
1553 struct sock_fprog fprog;
1554 int ret;
1555
1556 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1557 return -EPERM;
1558
1559 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1560 if (ret)
1561 return ret;
1562
1563 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1564 if (ret)
1565 return ret;
1566
1567 __fanout_set_data_bpf(po->fanout, new);
1568 return 0;
1569}
1570
1571static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1572 unsigned int len)
1573{
1574 struct bpf_prog *new;
1575 u32 fd;
1576
1577 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1578 return -EPERM;
1579 if (len != sizeof(fd))
1580 return -EINVAL;
1581 if (copy_from_sockptr(&fd, data, len))
1582 return -EFAULT;
1583
1584 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1585 if (IS_ERR(new))
1586 return PTR_ERR(new);
1587
1588 __fanout_set_data_bpf(po->fanout, new);
1589 return 0;
1590}
1591
1592static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1593 unsigned int len)
1594{
1595 switch (po->fanout->type) {
1596 case PACKET_FANOUT_CBPF:
1597 return fanout_set_data_cbpf(po, data, len);
1598 case PACKET_FANOUT_EBPF:
1599 return fanout_set_data_ebpf(po, data, len);
1600 default:
1601 return -EINVAL;
1602 }
1603}
1604
1605static void fanout_release_data(struct packet_fanout *f)
1606{
1607 switch (f->type) {
1608 case PACKET_FANOUT_CBPF:
1609 case PACKET_FANOUT_EBPF:
1610 __fanout_set_data_bpf(f, NULL);
1611 }
1612}
1613
1614static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1615{
1616 struct packet_fanout *f;
1617
1618 list_for_each_entry(f, &fanout_list, list) {
1619 if (f->id == candidate_id &&
1620 read_pnet(&f->net) == sock_net(sk)) {
1621 return false;
1622 }
1623 }
1624 return true;
1625}
1626
1627static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1628{
1629 u16 id = fanout_next_id;
1630
1631 do {
1632 if (__fanout_id_is_free(sk, id)) {
1633 *new_id = id;
1634 fanout_next_id = id + 1;
1635 return true;
1636 }
1637
1638 id++;
1639 } while (id != fanout_next_id);
1640
1641 return false;
1642}
1643
1644static int fanout_add(struct sock *sk, struct fanout_args *args)
1645{
1646 struct packet_rollover *rollover = NULL;
1647 struct packet_sock *po = pkt_sk(sk);
1648 u16 type_flags = args->type_flags;
1649 struct packet_fanout *f, *match;
1650 u8 type = type_flags & 0xff;
1651 u8 flags = type_flags >> 8;
1652 u16 id = args->id;
1653 int err;
1654
1655 switch (type) {
1656 case PACKET_FANOUT_ROLLOVER:
1657 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1658 return -EINVAL;
1659 break;
1660 case PACKET_FANOUT_HASH:
1661 case PACKET_FANOUT_LB:
1662 case PACKET_FANOUT_CPU:
1663 case PACKET_FANOUT_RND:
1664 case PACKET_FANOUT_QM:
1665 case PACKET_FANOUT_CBPF:
1666 case PACKET_FANOUT_EBPF:
1667 break;
1668 default:
1669 return -EINVAL;
1670 }
1671
1672 mutex_lock(&fanout_mutex);
1673
1674 err = -EALREADY;
1675 if (po->fanout)
1676 goto out;
1677
1678 if (type == PACKET_FANOUT_ROLLOVER ||
1679 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1680 err = -ENOMEM;
1681 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1682 if (!rollover)
1683 goto out;
1684 atomic_long_set(&rollover->num, 0);
1685 atomic_long_set(&rollover->num_huge, 0);
1686 atomic_long_set(&rollover->num_failed, 0);
1687 }
1688
1689 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1690 if (id != 0) {
1691 err = -EINVAL;
1692 goto out;
1693 }
1694 if (!fanout_find_new_id(sk, &id)) {
1695 err = -ENOMEM;
1696 goto out;
1697 }
1698 /* ephemeral flag for the first socket in the group: drop it */
1699 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1700 }
1701
1702 match = NULL;
1703 list_for_each_entry(f, &fanout_list, list) {
1704 if (f->id == id &&
1705 read_pnet(&f->net) == sock_net(sk)) {
1706 match = f;
1707 break;
1708 }
1709 }
1710 err = -EINVAL;
1711 if (match) {
1712 if (match->flags != flags)
1713 goto out;
1714 if (args->max_num_members &&
1715 args->max_num_members != match->max_num_members)
1716 goto out;
1717 } else {
1718 if (args->max_num_members > PACKET_FANOUT_MAX)
1719 goto out;
1720 if (!args->max_num_members)
1721 /* legacy PACKET_FANOUT_MAX */
1722 args->max_num_members = 256;
1723 err = -ENOMEM;
1724 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1725 GFP_KERNEL);
1726 if (!match)
1727 goto out;
1728 write_pnet(&match->net, sock_net(sk));
1729 match->id = id;
1730 match->type = type;
1731 match->flags = flags;
1732 INIT_LIST_HEAD(&match->list);
1733 spin_lock_init(&match->lock);
1734 refcount_set(&match->sk_ref, 0);
1735 fanout_init_data(match);
1736 match->prot_hook.type = po->prot_hook.type;
1737 match->prot_hook.dev = po->prot_hook.dev;
1738 match->prot_hook.func = packet_rcv_fanout;
1739 match->prot_hook.af_packet_priv = match;
1740 match->prot_hook.id_match = match_fanout_group;
1741 match->max_num_members = args->max_num_members;
1742 list_add(&match->list, &fanout_list);
1743 }
1744 err = -EINVAL;
1745
1746 spin_lock(&po->bind_lock);
1747 if (po->running &&
1748 match->type == type &&
1749 match->prot_hook.type == po->prot_hook.type &&
1750 match->prot_hook.dev == po->prot_hook.dev) {
1751 err = -ENOSPC;
1752 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1753 __dev_remove_pack(&po->prot_hook);
1754 po->fanout = match;
1755 po->rollover = rollover;
1756 rollover = NULL;
1757 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1758 __fanout_link(sk, po);
1759 err = 0;
1760 }
1761 }
1762 spin_unlock(&po->bind_lock);
1763
1764 if (err && !refcount_read(&match->sk_ref)) {
1765 list_del(&match->list);
1766 kvfree(match);
1767 }
1768
1769out:
1770 kfree(rollover);
1771 mutex_unlock(&fanout_mutex);
1772 return err;
1773}
1774
1775/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1776 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1777 * It is the responsibility of the caller to call fanout_release_data() and
1778 * free the returned packet_fanout (after synchronize_net())
1779 */
1780static struct packet_fanout *fanout_release(struct sock *sk)
1781{
1782 struct packet_sock *po = pkt_sk(sk);
1783 struct packet_fanout *f;
1784
1785 mutex_lock(&fanout_mutex);
1786 f = po->fanout;
1787 if (f) {
1788 po->fanout = NULL;
1789
1790 if (refcount_dec_and_test(&f->sk_ref))
1791 list_del(&f->list);
1792 else
1793 f = NULL;
1794 }
1795 mutex_unlock(&fanout_mutex);
1796
1797 return f;
1798}
1799
1800static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1801 struct sk_buff *skb)
1802{
1803 /* Earlier code assumed this would be a VLAN pkt, double-check
1804 * this now that we have the actual packet in hand. We can only
1805 * do this check on Ethernet devices.
1806 */
1807 if (unlikely(dev->type != ARPHRD_ETHER))
1808 return false;
1809
1810 skb_reset_mac_header(skb);
1811 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1812}
1813
1814static const struct proto_ops packet_ops;
1815
1816static const struct proto_ops packet_ops_spkt;
1817
1818static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1819 struct packet_type *pt, struct net_device *orig_dev)
1820{
1821 struct sock *sk;
1822 struct sockaddr_pkt *spkt;
1823
1824 /*
1825 * When we registered the protocol we saved the socket in the data
1826 * field for just this event.
1827 */
1828
1829 sk = pt->af_packet_priv;
1830
1831 /*
1832 * Yank back the headers [hope the device set this
1833 * right or kerboom...]
1834 *
1835 * Incoming packets have ll header pulled,
1836 * push it back.
1837 *
1838 * For outgoing ones skb->data == skb_mac_header(skb)
1839 * so that this procedure is noop.
1840 */
1841
1842 if (skb->pkt_type == PACKET_LOOPBACK)
1843 goto out;
1844
1845 if (!net_eq(dev_net(dev), sock_net(sk)))
1846 goto out;
1847
1848 skb = skb_share_check(skb, GFP_ATOMIC);
1849 if (skb == NULL)
1850 goto oom;
1851
1852 /* drop any routing info */
1853 skb_dst_drop(skb);
1854
1855 /* drop conntrack reference */
1856 nf_reset_ct(skb);
1857
1858 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1859
1860 skb_push(skb, skb->data - skb_mac_header(skb));
1861
1862 /*
1863 * The SOCK_PACKET socket receives _all_ frames.
1864 */
1865
1866 spkt->spkt_family = dev->type;
1867 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1868 spkt->spkt_protocol = skb->protocol;
1869
1870 /*
1871 * Charge the memory to the socket. This is done specifically
1872 * to prevent sockets using all the memory up.
1873 */
1874
1875 if (sock_queue_rcv_skb(sk, skb) == 0)
1876 return 0;
1877
1878out:
1879 kfree_skb(skb);
1880oom:
1881 return 0;
1882}
1883
1884static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1885{
1886 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1887 sock->type == SOCK_RAW) {
1888 skb_reset_mac_header(skb);
1889 skb->protocol = dev_parse_header_protocol(skb);
1890 }
1891
1892 skb_probe_transport_header(skb);
1893}
1894
1895/*
1896 * Output a raw packet to a device layer. This bypasses all the other
1897 * protocol layers and you must therefore supply it with a complete frame
1898 */
1899
1900static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1901 size_t len)
1902{
1903 struct sock *sk = sock->sk;
1904 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1905 struct sk_buff *skb = NULL;
1906 struct net_device *dev;
1907 struct sockcm_cookie sockc;
1908 __be16 proto = 0;
1909 int err;
1910 int extra_len = 0;
1911
1912 /*
1913 * Get and verify the address.
1914 */
1915
1916 if (saddr) {
1917 if (msg->msg_namelen < sizeof(struct sockaddr))
1918 return -EINVAL;
1919 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1920 proto = saddr->spkt_protocol;
1921 } else
1922 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1923
1924 /*
1925 * Find the device first to size check it
1926 */
1927
1928 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1929retry:
1930 rcu_read_lock();
1931 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1932 err = -ENODEV;
1933 if (dev == NULL)
1934 goto out_unlock;
1935
1936 err = -ENETDOWN;
1937 if (!(dev->flags & IFF_UP))
1938 goto out_unlock;
1939
1940 /*
1941 * You may not queue a frame bigger than the mtu. This is the lowest level
1942 * raw protocol and you must do your own fragmentation at this level.
1943 */
1944
1945 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1946 if (!netif_supports_nofcs(dev)) {
1947 err = -EPROTONOSUPPORT;
1948 goto out_unlock;
1949 }
1950 extra_len = 4; /* We're doing our own CRC */
1951 }
1952
1953 err = -EMSGSIZE;
1954 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1955 goto out_unlock;
1956
1957 if (!skb) {
1958 size_t reserved = LL_RESERVED_SPACE(dev);
1959 int tlen = dev->needed_tailroom;
1960 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1961
1962 rcu_read_unlock();
1963 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1964 if (skb == NULL)
1965 return -ENOBUFS;
1966 /* FIXME: Save some space for broken drivers that write a hard
1967 * header at transmission time by themselves. PPP is the notable
1968 * one here. This should really be fixed at the driver level.
1969 */
1970 skb_reserve(skb, reserved);
1971 skb_reset_network_header(skb);
1972
1973 /* Try to align data part correctly */
1974 if (hhlen) {
1975 skb->data -= hhlen;
1976 skb->tail -= hhlen;
1977 if (len < hhlen)
1978 skb_reset_network_header(skb);
1979 }
1980 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1981 if (err)
1982 goto out_free;
1983 goto retry;
1984 }
1985
1986 if (!dev_validate_header(dev, skb->data, len)) {
1987 err = -EINVAL;
1988 goto out_unlock;
1989 }
1990 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1991 !packet_extra_vlan_len_allowed(dev, skb)) {
1992 err = -EMSGSIZE;
1993 goto out_unlock;
1994 }
1995
1996 sockcm_init(&sockc, sk);
1997 if (msg->msg_controllen) {
1998 err = sock_cmsg_send(sk, msg, &sockc);
1999 if (unlikely(err))
2000 goto out_unlock;
2001 }
2002
2003 skb->protocol = proto;
2004 skb->dev = dev;
2005 skb->priority = sk->sk_priority;
2006 skb->mark = sk->sk_mark;
2007 skb->tstamp = sockc.transmit_time;
2008
2009 skb_setup_tx_timestamp(skb, sockc.tsflags);
2010
2011 if (unlikely(extra_len == 4))
2012 skb->no_fcs = 1;
2013
2014 packet_parse_headers(skb, sock);
2015
2016 dev_queue_xmit(skb);
2017 rcu_read_unlock();
2018 return len;
2019
2020out_unlock:
2021 rcu_read_unlock();
2022out_free:
2023 kfree_skb(skb);
2024 return err;
2025}
2026
2027static unsigned int run_filter(struct sk_buff *skb,
2028 const struct sock *sk,
2029 unsigned int res)
2030{
2031 struct sk_filter *filter;
2032
2033 rcu_read_lock();
2034 filter = rcu_dereference(sk->sk_filter);
2035 if (filter != NULL)
2036 res = bpf_prog_run_clear_cb(filter->prog, skb);
2037 rcu_read_unlock();
2038
2039 return res;
2040}
2041
2042static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2043 size_t *len)
2044{
2045 struct virtio_net_hdr vnet_hdr;
2046
2047 if (*len < sizeof(vnet_hdr))
2048 return -EINVAL;
2049 *len -= sizeof(vnet_hdr);
2050
2051 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2052 return -EINVAL;
2053
2054 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2055}
2056
2057/*
2058 * This function makes lazy skb cloning in hope that most of packets
2059 * are discarded by BPF.
2060 *
2061 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2062 * and skb->cb are mangled. It works because (and until) packets
2063 * falling here are owned by current CPU. Output packets are cloned
2064 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2065 * sequentially, so that if we return skb to original state on exit,
2066 * we will not harm anyone.
2067 */
2068
2069static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2070 struct packet_type *pt, struct net_device *orig_dev)
2071{
2072 struct sock *sk;
2073 struct sockaddr_ll *sll;
2074 struct packet_sock *po;
2075 u8 *skb_head = skb->data;
2076 int skb_len = skb->len;
2077 unsigned int snaplen, res;
2078 bool is_drop_n_account = false;
2079
2080 if (skb->pkt_type == PACKET_LOOPBACK)
2081 goto drop;
2082
2083 sk = pt->af_packet_priv;
2084 po = pkt_sk(sk);
2085
2086 if (!net_eq(dev_net(dev), sock_net(sk)))
2087 goto drop;
2088
2089 skb->dev = dev;
2090
2091 if (dev_has_header(dev)) {
2092 /* The device has an explicit notion of ll header,
2093 * exported to higher levels.
2094 *
2095 * Otherwise, the device hides details of its frame
2096 * structure, so that corresponding packet head is
2097 * never delivered to user.
2098 */
2099 if (sk->sk_type != SOCK_DGRAM)
2100 skb_push(skb, skb->data - skb_mac_header(skb));
2101 else if (skb->pkt_type == PACKET_OUTGOING) {
2102 /* Special case: outgoing packets have ll header at head */
2103 skb_pull(skb, skb_network_offset(skb));
2104 }
2105 }
2106
2107 snaplen = skb->len;
2108
2109 res = run_filter(skb, sk, snaplen);
2110 if (!res)
2111 goto drop_n_restore;
2112 if (snaplen > res)
2113 snaplen = res;
2114
2115 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2116 goto drop_n_acct;
2117
2118 if (skb_shared(skb)) {
2119 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2120 if (nskb == NULL)
2121 goto drop_n_acct;
2122
2123 if (skb_head != skb->data) {
2124 skb->data = skb_head;
2125 skb->len = skb_len;
2126 }
2127 consume_skb(skb);
2128 skb = nskb;
2129 }
2130
2131 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2132
2133 sll = &PACKET_SKB_CB(skb)->sa.ll;
2134 sll->sll_hatype = dev->type;
2135 sll->sll_pkttype = skb->pkt_type;
2136 if (unlikely(po->origdev))
2137 sll->sll_ifindex = orig_dev->ifindex;
2138 else
2139 sll->sll_ifindex = dev->ifindex;
2140
2141 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2142
2143 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2144 * Use their space for storing the original skb length.
2145 */
2146 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2147
2148 if (pskb_trim(skb, snaplen))
2149 goto drop_n_acct;
2150
2151 skb_set_owner_r(skb, sk);
2152 skb->dev = NULL;
2153 skb_dst_drop(skb);
2154
2155 /* drop conntrack reference */
2156 nf_reset_ct(skb);
2157
2158 spin_lock(&sk->sk_receive_queue.lock);
2159 po->stats.stats1.tp_packets++;
2160 sock_skb_set_dropcount(sk, skb);
2161 __skb_queue_tail(&sk->sk_receive_queue, skb);
2162 spin_unlock(&sk->sk_receive_queue.lock);
2163 sk->sk_data_ready(sk);
2164 return 0;
2165
2166drop_n_acct:
2167 is_drop_n_account = true;
2168 atomic_inc(&po->tp_drops);
2169 atomic_inc(&sk->sk_drops);
2170
2171drop_n_restore:
2172 if (skb_head != skb->data && skb_shared(skb)) {
2173 skb->data = skb_head;
2174 skb->len = skb_len;
2175 }
2176drop:
2177 if (!is_drop_n_account)
2178 consume_skb(skb);
2179 else
2180 kfree_skb(skb);
2181 return 0;
2182}
2183
2184static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2185 struct packet_type *pt, struct net_device *orig_dev)
2186{
2187 struct sock *sk;
2188 struct packet_sock *po;
2189 struct sockaddr_ll *sll;
2190 union tpacket_uhdr h;
2191 u8 *skb_head = skb->data;
2192 int skb_len = skb->len;
2193 unsigned int snaplen, res;
2194 unsigned long status = TP_STATUS_USER;
2195 unsigned short macoff, hdrlen;
2196 unsigned int netoff;
2197 struct sk_buff *copy_skb = NULL;
2198 struct timespec64 ts;
2199 __u32 ts_status;
2200 bool is_drop_n_account = false;
2201 unsigned int slot_id = 0;
2202 bool do_vnet = false;
2203
2204 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2205 * We may add members to them until current aligned size without forcing
2206 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2207 */
2208 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2209 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2210
2211 if (skb->pkt_type == PACKET_LOOPBACK)
2212 goto drop;
2213
2214 sk = pt->af_packet_priv;
2215 po = pkt_sk(sk);
2216
2217 if (!net_eq(dev_net(dev), sock_net(sk)))
2218 goto drop;
2219
2220 if (dev_has_header(dev)) {
2221 if (sk->sk_type != SOCK_DGRAM)
2222 skb_push(skb, skb->data - skb_mac_header(skb));
2223 else if (skb->pkt_type == PACKET_OUTGOING) {
2224 /* Special case: outgoing packets have ll header at head */
2225 skb_pull(skb, skb_network_offset(skb));
2226 }
2227 }
2228
2229 snaplen = skb->len;
2230
2231 res = run_filter(skb, sk, snaplen);
2232 if (!res)
2233 goto drop_n_restore;
2234
2235 /* If we are flooded, just give up */
2236 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2237 atomic_inc(&po->tp_drops);
2238 goto drop_n_restore;
2239 }
2240
2241 if (skb->ip_summed == CHECKSUM_PARTIAL)
2242 status |= TP_STATUS_CSUMNOTREADY;
2243 else if (skb->pkt_type != PACKET_OUTGOING &&
2244 (skb->ip_summed == CHECKSUM_COMPLETE ||
2245 skb_csum_unnecessary(skb)))
2246 status |= TP_STATUS_CSUM_VALID;
2247
2248 if (snaplen > res)
2249 snaplen = res;
2250
2251 if (sk->sk_type == SOCK_DGRAM) {
2252 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2253 po->tp_reserve;
2254 } else {
2255 unsigned int maclen = skb_network_offset(skb);
2256 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2257 (maclen < 16 ? 16 : maclen)) +
2258 po->tp_reserve;
2259 if (po->has_vnet_hdr) {
2260 netoff += sizeof(struct virtio_net_hdr);
2261 do_vnet = true;
2262 }
2263 macoff = netoff - maclen;
2264 }
2265 if (netoff > USHRT_MAX) {
2266 atomic_inc(&po->tp_drops);
2267 goto drop_n_restore;
2268 }
2269 if (po->tp_version <= TPACKET_V2) {
2270 if (macoff + snaplen > po->rx_ring.frame_size) {
2271 if (po->copy_thresh &&
2272 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2273 if (skb_shared(skb)) {
2274 copy_skb = skb_clone(skb, GFP_ATOMIC);
2275 } else {
2276 copy_skb = skb_get(skb);
2277 skb_head = skb->data;
2278 }
2279 if (copy_skb)
2280 skb_set_owner_r(copy_skb, sk);
2281 }
2282 snaplen = po->rx_ring.frame_size - macoff;
2283 if ((int)snaplen < 0) {
2284 snaplen = 0;
2285 do_vnet = false;
2286 }
2287 }
2288 } else if (unlikely(macoff + snaplen >
2289 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2290 u32 nval;
2291
2292 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2293 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2294 snaplen, nval, macoff);
2295 snaplen = nval;
2296 if (unlikely((int)snaplen < 0)) {
2297 snaplen = 0;
2298 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2299 do_vnet = false;
2300 }
2301 }
2302 spin_lock(&sk->sk_receive_queue.lock);
2303 h.raw = packet_current_rx_frame(po, skb,
2304 TP_STATUS_KERNEL, (macoff+snaplen));
2305 if (!h.raw)
2306 goto drop_n_account;
2307
2308 if (po->tp_version <= TPACKET_V2) {
2309 slot_id = po->rx_ring.head;
2310 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2311 goto drop_n_account;
2312 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2313 }
2314
2315 if (do_vnet &&
2316 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2317 sizeof(struct virtio_net_hdr),
2318 vio_le(), true, 0)) {
2319 if (po->tp_version == TPACKET_V3)
2320 prb_clear_blk_fill_status(&po->rx_ring);
2321 goto drop_n_account;
2322 }
2323
2324 if (po->tp_version <= TPACKET_V2) {
2325 packet_increment_rx_head(po, &po->rx_ring);
2326 /*
2327 * LOSING will be reported till you read the stats,
2328 * because it's COR - Clear On Read.
2329 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2330 * at packet level.
2331 */
2332 if (atomic_read(&po->tp_drops))
2333 status |= TP_STATUS_LOSING;
2334 }
2335
2336 po->stats.stats1.tp_packets++;
2337 if (copy_skb) {
2338 status |= TP_STATUS_COPY;
2339 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2340 }
2341 spin_unlock(&sk->sk_receive_queue.lock);
2342
2343 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2344
2345 /* Always timestamp; prefer an existing software timestamp taken
2346 * closer to the time of capture.
2347 */
2348 ts_status = tpacket_get_timestamp(skb, &ts,
2349 po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2350 if (!ts_status)
2351 ktime_get_real_ts64(&ts);
2352
2353 status |= ts_status;
2354
2355 switch (po->tp_version) {
2356 case TPACKET_V1:
2357 h.h1->tp_len = skb->len;
2358 h.h1->tp_snaplen = snaplen;
2359 h.h1->tp_mac = macoff;
2360 h.h1->tp_net = netoff;
2361 h.h1->tp_sec = ts.tv_sec;
2362 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2363 hdrlen = sizeof(*h.h1);
2364 break;
2365 case TPACKET_V2:
2366 h.h2->tp_len = skb->len;
2367 h.h2->tp_snaplen = snaplen;
2368 h.h2->tp_mac = macoff;
2369 h.h2->tp_net = netoff;
2370 h.h2->tp_sec = ts.tv_sec;
2371 h.h2->tp_nsec = ts.tv_nsec;
2372 if (skb_vlan_tag_present(skb)) {
2373 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2374 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2375 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2376 } else {
2377 h.h2->tp_vlan_tci = 0;
2378 h.h2->tp_vlan_tpid = 0;
2379 }
2380 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2381 hdrlen = sizeof(*h.h2);
2382 break;
2383 case TPACKET_V3:
2384 /* tp_nxt_offset,vlan are already populated above.
2385 * So DONT clear those fields here
2386 */
2387 h.h3->tp_status |= status;
2388 h.h3->tp_len = skb->len;
2389 h.h3->tp_snaplen = snaplen;
2390 h.h3->tp_mac = macoff;
2391 h.h3->tp_net = netoff;
2392 h.h3->tp_sec = ts.tv_sec;
2393 h.h3->tp_nsec = ts.tv_nsec;
2394 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2395 hdrlen = sizeof(*h.h3);
2396 break;
2397 default:
2398 BUG();
2399 }
2400
2401 sll = h.raw + TPACKET_ALIGN(hdrlen);
2402 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2403 sll->sll_family = AF_PACKET;
2404 sll->sll_hatype = dev->type;
2405 sll->sll_protocol = skb->protocol;
2406 sll->sll_pkttype = skb->pkt_type;
2407 if (unlikely(po->origdev))
2408 sll->sll_ifindex = orig_dev->ifindex;
2409 else
2410 sll->sll_ifindex = dev->ifindex;
2411
2412 smp_mb();
2413
2414#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2415 if (po->tp_version <= TPACKET_V2) {
2416 u8 *start, *end;
2417
2418 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2419 macoff + snaplen);
2420
2421 for (start = h.raw; start < end; start += PAGE_SIZE)
2422 flush_dcache_page(pgv_to_page(start));
2423 }
2424 smp_wmb();
2425#endif
2426
2427 if (po->tp_version <= TPACKET_V2) {
2428 spin_lock(&sk->sk_receive_queue.lock);
2429 __packet_set_status(po, h.raw, status);
2430 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2431 spin_unlock(&sk->sk_receive_queue.lock);
2432 sk->sk_data_ready(sk);
2433 } else if (po->tp_version == TPACKET_V3) {
2434 prb_clear_blk_fill_status(&po->rx_ring);
2435 }
2436
2437drop_n_restore:
2438 if (skb_head != skb->data && skb_shared(skb)) {
2439 skb->data = skb_head;
2440 skb->len = skb_len;
2441 }
2442drop:
2443 if (!is_drop_n_account)
2444 consume_skb(skb);
2445 else
2446 kfree_skb(skb);
2447 return 0;
2448
2449drop_n_account:
2450 spin_unlock(&sk->sk_receive_queue.lock);
2451 atomic_inc(&po->tp_drops);
2452 is_drop_n_account = true;
2453
2454 sk->sk_data_ready(sk);
2455 kfree_skb(copy_skb);
2456 goto drop_n_restore;
2457}
2458
2459static void tpacket_destruct_skb(struct sk_buff *skb)
2460{
2461 struct packet_sock *po = pkt_sk(skb->sk);
2462
2463 if (likely(po->tx_ring.pg_vec)) {
2464 void *ph;
2465 __u32 ts;
2466
2467 ph = skb_zcopy_get_nouarg(skb);
2468 packet_dec_pending(&po->tx_ring);
2469
2470 ts = __packet_set_timestamp(po, ph, skb);
2471 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2472
2473 if (!packet_read_pending(&po->tx_ring))
2474 complete(&po->skb_completion);
2475 }
2476
2477 sock_wfree(skb);
2478}
2479
2480static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2481{
2482 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2483 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2484 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2485 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2486 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2487 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2488 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2489
2490 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2491 return -EINVAL;
2492
2493 return 0;
2494}
2495
2496static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2497 struct virtio_net_hdr *vnet_hdr)
2498{
2499 if (*len < sizeof(*vnet_hdr))
2500 return -EINVAL;
2501 *len -= sizeof(*vnet_hdr);
2502
2503 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2504 return -EFAULT;
2505
2506 return __packet_snd_vnet_parse(vnet_hdr, *len);
2507}
2508
2509static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2510 void *frame, struct net_device *dev, void *data, int tp_len,
2511 __be16 proto, unsigned char *addr, int hlen, int copylen,
2512 const struct sockcm_cookie *sockc)
2513{
2514 union tpacket_uhdr ph;
2515 int to_write, offset, len, nr_frags, len_max;
2516 struct socket *sock = po->sk.sk_socket;
2517 struct page *page;
2518 int err;
2519
2520 ph.raw = frame;
2521
2522 skb->protocol = proto;
2523 skb->dev = dev;
2524 skb->priority = po->sk.sk_priority;
2525 skb->mark = po->sk.sk_mark;
2526 skb->tstamp = sockc->transmit_time;
2527 skb_setup_tx_timestamp(skb, sockc->tsflags);
2528 skb_zcopy_set_nouarg(skb, ph.raw);
2529
2530 skb_reserve(skb, hlen);
2531 skb_reset_network_header(skb);
2532
2533 to_write = tp_len;
2534
2535 if (sock->type == SOCK_DGRAM) {
2536 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2537 NULL, tp_len);
2538 if (unlikely(err < 0))
2539 return -EINVAL;
2540 } else if (copylen) {
2541 int hdrlen = min_t(int, copylen, tp_len);
2542
2543 skb_push(skb, dev->hard_header_len);
2544 skb_put(skb, copylen - dev->hard_header_len);
2545 err = skb_store_bits(skb, 0, data, hdrlen);
2546 if (unlikely(err))
2547 return err;
2548 if (!dev_validate_header(dev, skb->data, hdrlen))
2549 return -EINVAL;
2550
2551 data += hdrlen;
2552 to_write -= hdrlen;
2553 }
2554
2555 offset = offset_in_page(data);
2556 len_max = PAGE_SIZE - offset;
2557 len = ((to_write > len_max) ? len_max : to_write);
2558
2559 skb->data_len = to_write;
2560 skb->len += to_write;
2561 skb->truesize += to_write;
2562 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2563
2564 while (likely(to_write)) {
2565 nr_frags = skb_shinfo(skb)->nr_frags;
2566
2567 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2568 pr_err("Packet exceed the number of skb frags(%lu)\n",
2569 MAX_SKB_FRAGS);
2570 return -EFAULT;
2571 }
2572
2573 page = pgv_to_page(data);
2574 data += len;
2575 flush_dcache_page(page);
2576 get_page(page);
2577 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2578 to_write -= len;
2579 offset = 0;
2580 len_max = PAGE_SIZE;
2581 len = ((to_write > len_max) ? len_max : to_write);
2582 }
2583
2584 packet_parse_headers(skb, sock);
2585
2586 return tp_len;
2587}
2588
2589static int tpacket_parse_header(struct packet_sock *po, void *frame,
2590 int size_max, void **data)
2591{
2592 union tpacket_uhdr ph;
2593 int tp_len, off;
2594
2595 ph.raw = frame;
2596
2597 switch (po->tp_version) {
2598 case TPACKET_V3:
2599 if (ph.h3->tp_next_offset != 0) {
2600 pr_warn_once("variable sized slot not supported");
2601 return -EINVAL;
2602 }
2603 tp_len = ph.h3->tp_len;
2604 break;
2605 case TPACKET_V2:
2606 tp_len = ph.h2->tp_len;
2607 break;
2608 default:
2609 tp_len = ph.h1->tp_len;
2610 break;
2611 }
2612 if (unlikely(tp_len > size_max)) {
2613 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2614 return -EMSGSIZE;
2615 }
2616
2617 if (unlikely(po->tp_tx_has_off)) {
2618 int off_min, off_max;
2619
2620 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2621 off_max = po->tx_ring.frame_size - tp_len;
2622 if (po->sk.sk_type == SOCK_DGRAM) {
2623 switch (po->tp_version) {
2624 case TPACKET_V3:
2625 off = ph.h3->tp_net;
2626 break;
2627 case TPACKET_V2:
2628 off = ph.h2->tp_net;
2629 break;
2630 default:
2631 off = ph.h1->tp_net;
2632 break;
2633 }
2634 } else {
2635 switch (po->tp_version) {
2636 case TPACKET_V3:
2637 off = ph.h3->tp_mac;
2638 break;
2639 case TPACKET_V2:
2640 off = ph.h2->tp_mac;
2641 break;
2642 default:
2643 off = ph.h1->tp_mac;
2644 break;
2645 }
2646 }
2647 if (unlikely((off < off_min) || (off_max < off)))
2648 return -EINVAL;
2649 } else {
2650 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2651 }
2652
2653 *data = frame + off;
2654 return tp_len;
2655}
2656
2657static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2658{
2659 struct sk_buff *skb = NULL;
2660 struct net_device *dev;
2661 struct virtio_net_hdr *vnet_hdr = NULL;
2662 struct sockcm_cookie sockc;
2663 __be16 proto;
2664 int err, reserve = 0;
2665 void *ph;
2666 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2667 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2668 unsigned char *addr = NULL;
2669 int tp_len, size_max;
2670 void *data;
2671 int len_sum = 0;
2672 int status = TP_STATUS_AVAILABLE;
2673 int hlen, tlen, copylen = 0;
2674 long timeo = 0;
2675
2676 mutex_lock(&po->pg_vec_lock);
2677
2678 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2679 * we need to confirm it under protection of pg_vec_lock.
2680 */
2681 if (unlikely(!po->tx_ring.pg_vec)) {
2682 err = -EBUSY;
2683 goto out;
2684 }
2685 if (likely(saddr == NULL)) {
2686 dev = packet_cached_dev_get(po);
2687 proto = READ_ONCE(po->num);
2688 } else {
2689 err = -EINVAL;
2690 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2691 goto out;
2692 if (msg->msg_namelen < (saddr->sll_halen
2693 + offsetof(struct sockaddr_ll,
2694 sll_addr)))
2695 goto out;
2696 proto = saddr->sll_protocol;
2697 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2698 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2699 if (dev && msg->msg_namelen < dev->addr_len +
2700 offsetof(struct sockaddr_ll, sll_addr))
2701 goto out_put;
2702 addr = saddr->sll_addr;
2703 }
2704 }
2705
2706 err = -ENXIO;
2707 if (unlikely(dev == NULL))
2708 goto out;
2709 err = -ENETDOWN;
2710 if (unlikely(!(dev->flags & IFF_UP)))
2711 goto out_put;
2712
2713 sockcm_init(&sockc, &po->sk);
2714 if (msg->msg_controllen) {
2715 err = sock_cmsg_send(&po->sk, msg, &sockc);
2716 if (unlikely(err))
2717 goto out_put;
2718 }
2719
2720 if (po->sk.sk_socket->type == SOCK_RAW)
2721 reserve = dev->hard_header_len;
2722 size_max = po->tx_ring.frame_size
2723 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2724
2725 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2726 size_max = dev->mtu + reserve + VLAN_HLEN;
2727
2728 reinit_completion(&po->skb_completion);
2729
2730 do {
2731 ph = packet_current_frame(po, &po->tx_ring,
2732 TP_STATUS_SEND_REQUEST);
2733 if (unlikely(ph == NULL)) {
2734 if (need_wait && skb) {
2735 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2736 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2737 if (timeo <= 0) {
2738 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2739 goto out_put;
2740 }
2741 }
2742 /* check for additional frames */
2743 continue;
2744 }
2745
2746 skb = NULL;
2747 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2748 if (tp_len < 0)
2749 goto tpacket_error;
2750
2751 status = TP_STATUS_SEND_REQUEST;
2752 hlen = LL_RESERVED_SPACE(dev);
2753 tlen = dev->needed_tailroom;
2754 if (po->has_vnet_hdr) {
2755 vnet_hdr = data;
2756 data += sizeof(*vnet_hdr);
2757 tp_len -= sizeof(*vnet_hdr);
2758 if (tp_len < 0 ||
2759 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2760 tp_len = -EINVAL;
2761 goto tpacket_error;
2762 }
2763 copylen = __virtio16_to_cpu(vio_le(),
2764 vnet_hdr->hdr_len);
2765 }
2766 copylen = max_t(int, copylen, dev->hard_header_len);
2767 skb = sock_alloc_send_skb(&po->sk,
2768 hlen + tlen + sizeof(struct sockaddr_ll) +
2769 (copylen - dev->hard_header_len),
2770 !need_wait, &err);
2771
2772 if (unlikely(skb == NULL)) {
2773 /* we assume the socket was initially writeable ... */
2774 if (likely(len_sum > 0))
2775 err = len_sum;
2776 goto out_status;
2777 }
2778 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2779 addr, hlen, copylen, &sockc);
2780 if (likely(tp_len >= 0) &&
2781 tp_len > dev->mtu + reserve &&
2782 !po->has_vnet_hdr &&
2783 !packet_extra_vlan_len_allowed(dev, skb))
2784 tp_len = -EMSGSIZE;
2785
2786 if (unlikely(tp_len < 0)) {
2787tpacket_error:
2788 if (po->tp_loss) {
2789 __packet_set_status(po, ph,
2790 TP_STATUS_AVAILABLE);
2791 packet_increment_head(&po->tx_ring);
2792 kfree_skb(skb);
2793 continue;
2794 } else {
2795 status = TP_STATUS_WRONG_FORMAT;
2796 err = tp_len;
2797 goto out_status;
2798 }
2799 }
2800
2801 if (po->has_vnet_hdr) {
2802 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2803 tp_len = -EINVAL;
2804 goto tpacket_error;
2805 }
2806 virtio_net_hdr_set_proto(skb, vnet_hdr);
2807 }
2808
2809 skb->destructor = tpacket_destruct_skb;
2810 __packet_set_status(po, ph, TP_STATUS_SENDING);
2811 packet_inc_pending(&po->tx_ring);
2812
2813 status = TP_STATUS_SEND_REQUEST;
2814 err = po->xmit(skb);
2815 if (unlikely(err > 0)) {
2816 err = net_xmit_errno(err);
2817 if (err && __packet_get_status(po, ph) ==
2818 TP_STATUS_AVAILABLE) {
2819 /* skb was destructed already */
2820 skb = NULL;
2821 goto out_status;
2822 }
2823 /*
2824 * skb was dropped but not destructed yet;
2825 * let's treat it like congestion or err < 0
2826 */
2827 err = 0;
2828 }
2829 packet_increment_head(&po->tx_ring);
2830 len_sum += tp_len;
2831 } while (likely((ph != NULL) ||
2832 /* Note: packet_read_pending() might be slow if we have
2833 * to call it as it's per_cpu variable, but in fast-path
2834 * we already short-circuit the loop with the first
2835 * condition, and luckily don't have to go that path
2836 * anyway.
2837 */
2838 (need_wait && packet_read_pending(&po->tx_ring))));
2839
2840 err = len_sum;
2841 goto out_put;
2842
2843out_status:
2844 __packet_set_status(po, ph, status);
2845 kfree_skb(skb);
2846out_put:
2847 dev_put(dev);
2848out:
2849 mutex_unlock(&po->pg_vec_lock);
2850 return err;
2851}
2852
2853static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2854 size_t reserve, size_t len,
2855 size_t linear, int noblock,
2856 int *err)
2857{
2858 struct sk_buff *skb;
2859
2860 /* Under a page? Don't bother with paged skb. */
2861 if (prepad + len < PAGE_SIZE || !linear)
2862 linear = len;
2863
2864 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2865 err, 0);
2866 if (!skb)
2867 return NULL;
2868
2869 skb_reserve(skb, reserve);
2870 skb_put(skb, linear);
2871 skb->data_len = len - linear;
2872 skb->len += len - linear;
2873
2874 return skb;
2875}
2876
2877static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2878{
2879 struct sock *sk = sock->sk;
2880 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2881 struct sk_buff *skb;
2882 struct net_device *dev;
2883 __be16 proto;
2884 unsigned char *addr = NULL;
2885 int err, reserve = 0;
2886 struct sockcm_cookie sockc;
2887 struct virtio_net_hdr vnet_hdr = { 0 };
2888 int offset = 0;
2889 struct packet_sock *po = pkt_sk(sk);
2890 bool has_vnet_hdr = false;
2891 int hlen, tlen, linear;
2892 int extra_len = 0;
2893
2894 /*
2895 * Get and verify the address.
2896 */
2897
2898 if (likely(saddr == NULL)) {
2899 dev = packet_cached_dev_get(po);
2900 proto = READ_ONCE(po->num);
2901 } else {
2902 err = -EINVAL;
2903 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2904 goto out;
2905 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2906 goto out;
2907 proto = saddr->sll_protocol;
2908 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2909 if (sock->type == SOCK_DGRAM) {
2910 if (dev && msg->msg_namelen < dev->addr_len +
2911 offsetof(struct sockaddr_ll, sll_addr))
2912 goto out_unlock;
2913 addr = saddr->sll_addr;
2914 }
2915 }
2916
2917 err = -ENXIO;
2918 if (unlikely(dev == NULL))
2919 goto out_unlock;
2920 err = -ENETDOWN;
2921 if (unlikely(!(dev->flags & IFF_UP)))
2922 goto out_unlock;
2923
2924 sockcm_init(&sockc, sk);
2925 sockc.mark = sk->sk_mark;
2926 if (msg->msg_controllen) {
2927 err = sock_cmsg_send(sk, msg, &sockc);
2928 if (unlikely(err))
2929 goto out_unlock;
2930 }
2931
2932 if (sock->type == SOCK_RAW)
2933 reserve = dev->hard_header_len;
2934 if (po->has_vnet_hdr) {
2935 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2936 if (err)
2937 goto out_unlock;
2938 has_vnet_hdr = true;
2939 }
2940
2941 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2942 if (!netif_supports_nofcs(dev)) {
2943 err = -EPROTONOSUPPORT;
2944 goto out_unlock;
2945 }
2946 extra_len = 4; /* We're doing our own CRC */
2947 }
2948
2949 err = -EMSGSIZE;
2950 if (!vnet_hdr.gso_type &&
2951 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2952 goto out_unlock;
2953
2954 err = -ENOBUFS;
2955 hlen = LL_RESERVED_SPACE(dev);
2956 tlen = dev->needed_tailroom;
2957 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2958 linear = max(linear, min_t(int, len, dev->hard_header_len));
2959 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2960 msg->msg_flags & MSG_DONTWAIT, &err);
2961 if (skb == NULL)
2962 goto out_unlock;
2963
2964 skb_reset_network_header(skb);
2965
2966 err = -EINVAL;
2967 if (sock->type == SOCK_DGRAM) {
2968 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2969 if (unlikely(offset < 0))
2970 goto out_free;
2971 } else if (reserve) {
2972 skb_reserve(skb, -reserve);
2973 if (len < reserve + sizeof(struct ipv6hdr) &&
2974 dev->min_header_len != dev->hard_header_len)
2975 skb_reset_network_header(skb);
2976 }
2977
2978 /* Returns -EFAULT on error */
2979 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2980 if (err)
2981 goto out_free;
2982
2983 if (sock->type == SOCK_RAW &&
2984 !dev_validate_header(dev, skb->data, len)) {
2985 err = -EINVAL;
2986 goto out_free;
2987 }
2988
2989 skb_setup_tx_timestamp(skb, sockc.tsflags);
2990
2991 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2992 !packet_extra_vlan_len_allowed(dev, skb)) {
2993 err = -EMSGSIZE;
2994 goto out_free;
2995 }
2996
2997 skb->protocol = proto;
2998 skb->dev = dev;
2999 skb->priority = sk->sk_priority;
3000 skb->mark = sockc.mark;
3001 skb->tstamp = sockc.transmit_time;
3002
3003 if (has_vnet_hdr) {
3004 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3005 if (err)
3006 goto out_free;
3007 len += sizeof(vnet_hdr);
3008 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3009 }
3010
3011 packet_parse_headers(skb, sock);
3012
3013 if (unlikely(extra_len == 4))
3014 skb->no_fcs = 1;
3015
3016 err = po->xmit(skb);
3017 if (err > 0 && (err = net_xmit_errno(err)) != 0)
3018 goto out_unlock;
3019
3020 dev_put(dev);
3021
3022 return len;
3023
3024out_free:
3025 kfree_skb(skb);
3026out_unlock:
3027 if (dev)
3028 dev_put(dev);
3029out:
3030 return err;
3031}
3032
3033static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3034{
3035 struct sock *sk = sock->sk;
3036 struct packet_sock *po = pkt_sk(sk);
3037
3038 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3039 * tpacket_snd() will redo the check safely.
3040 */
3041 if (data_race(po->tx_ring.pg_vec))
3042 return tpacket_snd(po, msg);
3043
3044 return packet_snd(sock, msg, len);
3045}
3046
3047/*
3048 * Close a PACKET socket. This is fairly simple. We immediately go
3049 * to 'closed' state and remove our protocol entry in the device list.
3050 */
3051
3052static int packet_release(struct socket *sock)
3053{
3054 struct sock *sk = sock->sk;
3055 struct packet_sock *po;
3056 struct packet_fanout *f;
3057 struct net *net;
3058 union tpacket_req_u req_u;
3059
3060 if (!sk)
3061 return 0;
3062
3063 net = sock_net(sk);
3064 po = pkt_sk(sk);
3065
3066 mutex_lock(&net->packet.sklist_lock);
3067 sk_del_node_init_rcu(sk);
3068 mutex_unlock(&net->packet.sklist_lock);
3069
3070 preempt_disable();
3071 sock_prot_inuse_add(net, sk->sk_prot, -1);
3072 preempt_enable();
3073
3074 spin_lock(&po->bind_lock);
3075 unregister_prot_hook(sk, false);
3076 packet_cached_dev_reset(po);
3077
3078 if (po->prot_hook.dev) {
3079 dev_put(po->prot_hook.dev);
3080 po->prot_hook.dev = NULL;
3081 }
3082 spin_unlock(&po->bind_lock);
3083
3084 packet_flush_mclist(sk);
3085
3086 lock_sock(sk);
3087 if (po->rx_ring.pg_vec) {
3088 memset(&req_u, 0, sizeof(req_u));
3089 packet_set_ring(sk, &req_u, 1, 0);
3090 }
3091
3092 if (po->tx_ring.pg_vec) {
3093 memset(&req_u, 0, sizeof(req_u));
3094 packet_set_ring(sk, &req_u, 1, 1);
3095 }
3096 release_sock(sk);
3097
3098 f = fanout_release(sk);
3099
3100 synchronize_net();
3101
3102 kfree(po->rollover);
3103 if (f) {
3104 fanout_release_data(f);
3105 kvfree(f);
3106 }
3107 /*
3108 * Now the socket is dead. No more input will appear.
3109 */
3110 sock_orphan(sk);
3111 sock->sk = NULL;
3112
3113 /* Purge queues */
3114
3115 skb_queue_purge(&sk->sk_receive_queue);
3116 packet_free_pending(po);
3117 sk_refcnt_debug_release(sk);
3118
3119 sock_put(sk);
3120 return 0;
3121}
3122
3123/*
3124 * Attach a packet hook.
3125 */
3126
3127static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3128 __be16 proto)
3129{
3130 struct packet_sock *po = pkt_sk(sk);
3131 struct net_device *dev_curr;
3132 __be16 proto_curr;
3133 bool need_rehook;
3134 struct net_device *dev = NULL;
3135 int ret = 0;
3136 bool unlisted = false;
3137
3138 lock_sock(sk);
3139 spin_lock(&po->bind_lock);
3140 rcu_read_lock();
3141
3142 if (po->fanout) {
3143 ret = -EINVAL;
3144 goto out_unlock;
3145 }
3146
3147 if (name) {
3148 dev = dev_get_by_name_rcu(sock_net(sk), name);
3149 if (!dev) {
3150 ret = -ENODEV;
3151 goto out_unlock;
3152 }
3153 } else if (ifindex) {
3154 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3155 if (!dev) {
3156 ret = -ENODEV;
3157 goto out_unlock;
3158 }
3159 }
3160
3161 if (dev)
3162 dev_hold(dev);
3163
3164 proto_curr = po->prot_hook.type;
3165 dev_curr = po->prot_hook.dev;
3166
3167 need_rehook = proto_curr != proto || dev_curr != dev;
3168
3169 if (need_rehook) {
3170 if (po->running) {
3171 rcu_read_unlock();
3172 /* prevents packet_notifier() from calling
3173 * register_prot_hook()
3174 */
3175 WRITE_ONCE(po->num, 0);
3176 __unregister_prot_hook(sk, true);
3177 rcu_read_lock();
3178 dev_curr = po->prot_hook.dev;
3179 if (dev)
3180 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3181 dev->ifindex);
3182 }
3183
3184 BUG_ON(po->running);
3185 WRITE_ONCE(po->num, proto);
3186 po->prot_hook.type = proto;
3187
3188 if (unlikely(unlisted)) {
3189 dev_put(dev);
3190 po->prot_hook.dev = NULL;
3191 WRITE_ONCE(po->ifindex, -1);
3192 packet_cached_dev_reset(po);
3193 } else {
3194 po->prot_hook.dev = dev;
3195 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3196 packet_cached_dev_assign(po, dev);
3197 }
3198 }
3199 if (dev_curr)
3200 dev_put(dev_curr);
3201
3202 if (proto == 0 || !need_rehook)
3203 goto out_unlock;
3204
3205 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3206 register_prot_hook(sk);
3207 } else {
3208 sk->sk_err = ENETDOWN;
3209 if (!sock_flag(sk, SOCK_DEAD))
3210 sk_error_report(sk);
3211 }
3212
3213out_unlock:
3214 rcu_read_unlock();
3215 spin_unlock(&po->bind_lock);
3216 release_sock(sk);
3217 return ret;
3218}
3219
3220/*
3221 * Bind a packet socket to a device
3222 */
3223
3224static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3225 int addr_len)
3226{
3227 struct sock *sk = sock->sk;
3228 char name[sizeof(uaddr->sa_data) + 1];
3229
3230 /*
3231 * Check legality
3232 */
3233
3234 if (addr_len != sizeof(struct sockaddr))
3235 return -EINVAL;
3236 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3237 * zero-terminated.
3238 */
3239 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3240 name[sizeof(uaddr->sa_data)] = 0;
3241
3242 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3243}
3244
3245static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3246{
3247 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3248 struct sock *sk = sock->sk;
3249
3250 /*
3251 * Check legality
3252 */
3253
3254 if (addr_len < sizeof(struct sockaddr_ll))
3255 return -EINVAL;
3256 if (sll->sll_family != AF_PACKET)
3257 return -EINVAL;
3258
3259 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3260 sll->sll_protocol ? : pkt_sk(sk)->num);
3261}
3262
3263static struct proto packet_proto = {
3264 .name = "PACKET",
3265 .owner = THIS_MODULE,
3266 .obj_size = sizeof(struct packet_sock),
3267};
3268
3269/*
3270 * Create a packet of type SOCK_PACKET.
3271 */
3272
3273static int packet_create(struct net *net, struct socket *sock, int protocol,
3274 int kern)
3275{
3276 struct sock *sk;
3277 struct packet_sock *po;
3278 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3279 int err;
3280
3281 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3282 return -EPERM;
3283 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3284 sock->type != SOCK_PACKET)
3285 return -ESOCKTNOSUPPORT;
3286
3287 sock->state = SS_UNCONNECTED;
3288
3289 err = -ENOBUFS;
3290 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3291 if (sk == NULL)
3292 goto out;
3293
3294 sock->ops = &packet_ops;
3295 if (sock->type == SOCK_PACKET)
3296 sock->ops = &packet_ops_spkt;
3297
3298 sock_init_data(sock, sk);
3299
3300 po = pkt_sk(sk);
3301 init_completion(&po->skb_completion);
3302 sk->sk_family = PF_PACKET;
3303 po->num = proto;
3304 po->xmit = dev_queue_xmit;
3305
3306 err = packet_alloc_pending(po);
3307 if (err)
3308 goto out2;
3309
3310 packet_cached_dev_reset(po);
3311
3312 sk->sk_destruct = packet_sock_destruct;
3313 sk_refcnt_debug_inc(sk);
3314
3315 /*
3316 * Attach a protocol block
3317 */
3318
3319 spin_lock_init(&po->bind_lock);
3320 mutex_init(&po->pg_vec_lock);
3321 po->rollover = NULL;
3322 po->prot_hook.func = packet_rcv;
3323
3324 if (sock->type == SOCK_PACKET)
3325 po->prot_hook.func = packet_rcv_spkt;
3326
3327 po->prot_hook.af_packet_priv = sk;
3328
3329 if (proto) {
3330 po->prot_hook.type = proto;
3331 __register_prot_hook(sk);
3332 }
3333
3334 mutex_lock(&net->packet.sklist_lock);
3335 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3336 mutex_unlock(&net->packet.sklist_lock);
3337
3338 preempt_disable();
3339 sock_prot_inuse_add(net, &packet_proto, 1);
3340 preempt_enable();
3341
3342 return 0;
3343out2:
3344 sk_free(sk);
3345out:
3346 return err;
3347}
3348
3349/*
3350 * Pull a packet from our receive queue and hand it to the user.
3351 * If necessary we block.
3352 */
3353
3354static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3355 int flags)
3356{
3357 struct sock *sk = sock->sk;
3358 struct sk_buff *skb;
3359 int copied, err;
3360 int vnet_hdr_len = 0;
3361 unsigned int origlen = 0;
3362
3363 err = -EINVAL;
3364 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3365 goto out;
3366
3367#if 0
3368 /* What error should we return now? EUNATTACH? */
3369 if (pkt_sk(sk)->ifindex < 0)
3370 return -ENODEV;
3371#endif
3372
3373 if (flags & MSG_ERRQUEUE) {
3374 err = sock_recv_errqueue(sk, msg, len,
3375 SOL_PACKET, PACKET_TX_TIMESTAMP);
3376 goto out;
3377 }
3378
3379 /*
3380 * Call the generic datagram receiver. This handles all sorts
3381 * of horrible races and re-entrancy so we can forget about it
3382 * in the protocol layers.
3383 *
3384 * Now it will return ENETDOWN, if device have just gone down,
3385 * but then it will block.
3386 */
3387
3388 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3389
3390 /*
3391 * An error occurred so return it. Because skb_recv_datagram()
3392 * handles the blocking we don't see and worry about blocking
3393 * retries.
3394 */
3395
3396 if (skb == NULL)
3397 goto out;
3398
3399 packet_rcv_try_clear_pressure(pkt_sk(sk));
3400
3401 if (pkt_sk(sk)->has_vnet_hdr) {
3402 err = packet_rcv_vnet(msg, skb, &len);
3403 if (err)
3404 goto out_free;
3405 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3406 }
3407
3408 /* You lose any data beyond the buffer you gave. If it worries
3409 * a user program they can ask the device for its MTU
3410 * anyway.
3411 */
3412 copied = skb->len;
3413 if (copied > len) {
3414 copied = len;
3415 msg->msg_flags |= MSG_TRUNC;
3416 }
3417
3418 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3419 if (err)
3420 goto out_free;
3421
3422 if (sock->type != SOCK_PACKET) {
3423 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3424
3425 /* Original length was stored in sockaddr_ll fields */
3426 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3427 sll->sll_family = AF_PACKET;
3428 sll->sll_protocol = skb->protocol;
3429 }
3430
3431 sock_recv_ts_and_drops(msg, sk, skb);
3432
3433 if (msg->msg_name) {
3434 int copy_len;
3435
3436 /* If the address length field is there to be filled
3437 * in, we fill it in now.
3438 */
3439 if (sock->type == SOCK_PACKET) {
3440 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3441 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3442 copy_len = msg->msg_namelen;
3443 } else {
3444 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3445
3446 msg->msg_namelen = sll->sll_halen +
3447 offsetof(struct sockaddr_ll, sll_addr);
3448 copy_len = msg->msg_namelen;
3449 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3450 memset(msg->msg_name +
3451 offsetof(struct sockaddr_ll, sll_addr),
3452 0, sizeof(sll->sll_addr));
3453 msg->msg_namelen = sizeof(struct sockaddr_ll);
3454 }
3455 }
3456 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3457 }
3458
3459 if (pkt_sk(sk)->auxdata) {
3460 struct tpacket_auxdata aux;
3461
3462 aux.tp_status = TP_STATUS_USER;
3463 if (skb->ip_summed == CHECKSUM_PARTIAL)
3464 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3465 else if (skb->pkt_type != PACKET_OUTGOING &&
3466 (skb->ip_summed == CHECKSUM_COMPLETE ||
3467 skb_csum_unnecessary(skb)))
3468 aux.tp_status |= TP_STATUS_CSUM_VALID;
3469
3470 aux.tp_len = origlen;
3471 aux.tp_snaplen = skb->len;
3472 aux.tp_mac = 0;
3473 aux.tp_net = skb_network_offset(skb);
3474 if (skb_vlan_tag_present(skb)) {
3475 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3476 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3477 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3478 } else {
3479 aux.tp_vlan_tci = 0;
3480 aux.tp_vlan_tpid = 0;
3481 }
3482 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3483 }
3484
3485 /*
3486 * Free or return the buffer as appropriate. Again this
3487 * hides all the races and re-entrancy issues from us.
3488 */
3489 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3490
3491out_free:
3492 skb_free_datagram(sk, skb);
3493out:
3494 return err;
3495}
3496
3497static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3498 int peer)
3499{
3500 struct net_device *dev;
3501 struct sock *sk = sock->sk;
3502
3503 if (peer)
3504 return -EOPNOTSUPP;
3505
3506 uaddr->sa_family = AF_PACKET;
3507 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3508 rcu_read_lock();
3509 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3510 if (dev)
3511 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3512 rcu_read_unlock();
3513
3514 return sizeof(*uaddr);
3515}
3516
3517static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3518 int peer)
3519{
3520 struct net_device *dev;
3521 struct sock *sk = sock->sk;
3522 struct packet_sock *po = pkt_sk(sk);
3523 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3524 int ifindex;
3525
3526 if (peer)
3527 return -EOPNOTSUPP;
3528
3529 ifindex = READ_ONCE(po->ifindex);
3530 sll->sll_family = AF_PACKET;
3531 sll->sll_ifindex = ifindex;
3532 sll->sll_protocol = READ_ONCE(po->num);
3533 sll->sll_pkttype = 0;
3534 rcu_read_lock();
3535 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3536 if (dev) {
3537 sll->sll_hatype = dev->type;
3538 sll->sll_halen = dev->addr_len;
3539 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3540 } else {
3541 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3542 sll->sll_halen = 0;
3543 }
3544 rcu_read_unlock();
3545
3546 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3547}
3548
3549static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3550 int what)
3551{
3552 switch (i->type) {
3553 case PACKET_MR_MULTICAST:
3554 if (i->alen != dev->addr_len)
3555 return -EINVAL;
3556 if (what > 0)
3557 return dev_mc_add(dev, i->addr);
3558 else
3559 return dev_mc_del(dev, i->addr);
3560 break;
3561 case PACKET_MR_PROMISC:
3562 return dev_set_promiscuity(dev, what);
3563 case PACKET_MR_ALLMULTI:
3564 return dev_set_allmulti(dev, what);
3565 case PACKET_MR_UNICAST:
3566 if (i->alen != dev->addr_len)
3567 return -EINVAL;
3568 if (what > 0)
3569 return dev_uc_add(dev, i->addr);
3570 else
3571 return dev_uc_del(dev, i->addr);
3572 break;
3573 default:
3574 break;
3575 }
3576 return 0;
3577}
3578
3579static void packet_dev_mclist_delete(struct net_device *dev,
3580 struct packet_mclist **mlp)
3581{
3582 struct packet_mclist *ml;
3583
3584 while ((ml = *mlp) != NULL) {
3585 if (ml->ifindex == dev->ifindex) {
3586 packet_dev_mc(dev, ml, -1);
3587 *mlp = ml->next;
3588 kfree(ml);
3589 } else
3590 mlp = &ml->next;
3591 }
3592}
3593
3594static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3595{
3596 struct packet_sock *po = pkt_sk(sk);
3597 struct packet_mclist *ml, *i;
3598 struct net_device *dev;
3599 int err;
3600
3601 rtnl_lock();
3602
3603 err = -ENODEV;
3604 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3605 if (!dev)
3606 goto done;
3607
3608 err = -EINVAL;
3609 if (mreq->mr_alen > dev->addr_len)
3610 goto done;
3611
3612 err = -ENOBUFS;
3613 i = kmalloc(sizeof(*i), GFP_KERNEL);
3614 if (i == NULL)
3615 goto done;
3616
3617 err = 0;
3618 for (ml = po->mclist; ml; ml = ml->next) {
3619 if (ml->ifindex == mreq->mr_ifindex &&
3620 ml->type == mreq->mr_type &&
3621 ml->alen == mreq->mr_alen &&
3622 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3623 ml->count++;
3624 /* Free the new element ... */
3625 kfree(i);
3626 goto done;
3627 }
3628 }
3629
3630 i->type = mreq->mr_type;
3631 i->ifindex = mreq->mr_ifindex;
3632 i->alen = mreq->mr_alen;
3633 memcpy(i->addr, mreq->mr_address, i->alen);
3634 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3635 i->count = 1;
3636 i->next = po->mclist;
3637 po->mclist = i;
3638 err = packet_dev_mc(dev, i, 1);
3639 if (err) {
3640 po->mclist = i->next;
3641 kfree(i);
3642 }
3643
3644done:
3645 rtnl_unlock();
3646 return err;
3647}
3648
3649static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3650{
3651 struct packet_mclist *ml, **mlp;
3652
3653 rtnl_lock();
3654
3655 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3656 if (ml->ifindex == mreq->mr_ifindex &&
3657 ml->type == mreq->mr_type &&
3658 ml->alen == mreq->mr_alen &&
3659 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3660 if (--ml->count == 0) {
3661 struct net_device *dev;
3662 *mlp = ml->next;
3663 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3664 if (dev)
3665 packet_dev_mc(dev, ml, -1);
3666 kfree(ml);
3667 }
3668 break;
3669 }
3670 }
3671 rtnl_unlock();
3672 return 0;
3673}
3674
3675static void packet_flush_mclist(struct sock *sk)
3676{
3677 struct packet_sock *po = pkt_sk(sk);
3678 struct packet_mclist *ml;
3679
3680 if (!po->mclist)
3681 return;
3682
3683 rtnl_lock();
3684 while ((ml = po->mclist) != NULL) {
3685 struct net_device *dev;
3686
3687 po->mclist = ml->next;
3688 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3689 if (dev != NULL)
3690 packet_dev_mc(dev, ml, -1);
3691 kfree(ml);
3692 }
3693 rtnl_unlock();
3694}
3695
3696static int
3697packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3698 unsigned int optlen)
3699{
3700 struct sock *sk = sock->sk;
3701 struct packet_sock *po = pkt_sk(sk);
3702 int ret;
3703
3704 if (level != SOL_PACKET)
3705 return -ENOPROTOOPT;
3706
3707 switch (optname) {
3708 case PACKET_ADD_MEMBERSHIP:
3709 case PACKET_DROP_MEMBERSHIP:
3710 {
3711 struct packet_mreq_max mreq;
3712 int len = optlen;
3713 memset(&mreq, 0, sizeof(mreq));
3714 if (len < sizeof(struct packet_mreq))
3715 return -EINVAL;
3716 if (len > sizeof(mreq))
3717 len = sizeof(mreq);
3718 if (copy_from_sockptr(&mreq, optval, len))
3719 return -EFAULT;
3720 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3721 return -EINVAL;
3722 if (optname == PACKET_ADD_MEMBERSHIP)
3723 ret = packet_mc_add(sk, &mreq);
3724 else
3725 ret = packet_mc_drop(sk, &mreq);
3726 return ret;
3727 }
3728
3729 case PACKET_RX_RING:
3730 case PACKET_TX_RING:
3731 {
3732 union tpacket_req_u req_u;
3733 int len;
3734
3735 lock_sock(sk);
3736 switch (po->tp_version) {
3737 case TPACKET_V1:
3738 case TPACKET_V2:
3739 len = sizeof(req_u.req);
3740 break;
3741 case TPACKET_V3:
3742 default:
3743 len = sizeof(req_u.req3);
3744 break;
3745 }
3746 if (optlen < len) {
3747 ret = -EINVAL;
3748 } else {
3749 if (copy_from_sockptr(&req_u.req, optval, len))
3750 ret = -EFAULT;
3751 else
3752 ret = packet_set_ring(sk, &req_u, 0,
3753 optname == PACKET_TX_RING);
3754 }
3755 release_sock(sk);
3756 return ret;
3757 }
3758 case PACKET_COPY_THRESH:
3759 {
3760 int val;
3761
3762 if (optlen != sizeof(val))
3763 return -EINVAL;
3764 if (copy_from_sockptr(&val, optval, sizeof(val)))
3765 return -EFAULT;
3766
3767 pkt_sk(sk)->copy_thresh = val;
3768 return 0;
3769 }
3770 case PACKET_VERSION:
3771 {
3772 int val;
3773
3774 if (optlen != sizeof(val))
3775 return -EINVAL;
3776 if (copy_from_sockptr(&val, optval, sizeof(val)))
3777 return -EFAULT;
3778 switch (val) {
3779 case TPACKET_V1:
3780 case TPACKET_V2:
3781 case TPACKET_V3:
3782 break;
3783 default:
3784 return -EINVAL;
3785 }
3786 lock_sock(sk);
3787 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3788 ret = -EBUSY;
3789 } else {
3790 po->tp_version = val;
3791 ret = 0;
3792 }
3793 release_sock(sk);
3794 return ret;
3795 }
3796 case PACKET_RESERVE:
3797 {
3798 unsigned int val;
3799
3800 if (optlen != sizeof(val))
3801 return -EINVAL;
3802 if (copy_from_sockptr(&val, optval, sizeof(val)))
3803 return -EFAULT;
3804 if (val > INT_MAX)
3805 return -EINVAL;
3806 lock_sock(sk);
3807 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3808 ret = -EBUSY;
3809 } else {
3810 po->tp_reserve = val;
3811 ret = 0;
3812 }
3813 release_sock(sk);
3814 return ret;
3815 }
3816 case PACKET_LOSS:
3817 {
3818 unsigned int val;
3819
3820 if (optlen != sizeof(val))
3821 return -EINVAL;
3822 if (copy_from_sockptr(&val, optval, sizeof(val)))
3823 return -EFAULT;
3824
3825 lock_sock(sk);
3826 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3827 ret = -EBUSY;
3828 } else {
3829 po->tp_loss = !!val;
3830 ret = 0;
3831 }
3832 release_sock(sk);
3833 return ret;
3834 }
3835 case PACKET_AUXDATA:
3836 {
3837 int val;
3838
3839 if (optlen < sizeof(val))
3840 return -EINVAL;
3841 if (copy_from_sockptr(&val, optval, sizeof(val)))
3842 return -EFAULT;
3843
3844 lock_sock(sk);
3845 po->auxdata = !!val;
3846 release_sock(sk);
3847 return 0;
3848 }
3849 case PACKET_ORIGDEV:
3850 {
3851 int val;
3852
3853 if (optlen < sizeof(val))
3854 return -EINVAL;
3855 if (copy_from_sockptr(&val, optval, sizeof(val)))
3856 return -EFAULT;
3857
3858 lock_sock(sk);
3859 po->origdev = !!val;
3860 release_sock(sk);
3861 return 0;
3862 }
3863 case PACKET_VNET_HDR:
3864 {
3865 int val;
3866
3867 if (sock->type != SOCK_RAW)
3868 return -EINVAL;
3869 if (optlen < sizeof(val))
3870 return -EINVAL;
3871 if (copy_from_sockptr(&val, optval, sizeof(val)))
3872 return -EFAULT;
3873
3874 lock_sock(sk);
3875 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3876 ret = -EBUSY;
3877 } else {
3878 po->has_vnet_hdr = !!val;
3879 ret = 0;
3880 }
3881 release_sock(sk);
3882 return ret;
3883 }
3884 case PACKET_TIMESTAMP:
3885 {
3886 int val;
3887
3888 if (optlen != sizeof(val))
3889 return -EINVAL;
3890 if (copy_from_sockptr(&val, optval, sizeof(val)))
3891 return -EFAULT;
3892
3893 po->tp_tstamp = val;
3894 return 0;
3895 }
3896 case PACKET_FANOUT:
3897 {
3898 struct fanout_args args = { 0 };
3899
3900 if (optlen != sizeof(int) && optlen != sizeof(args))
3901 return -EINVAL;
3902 if (copy_from_sockptr(&args, optval, optlen))
3903 return -EFAULT;
3904
3905 return fanout_add(sk, &args);
3906 }
3907 case PACKET_FANOUT_DATA:
3908 {
3909 if (!po->fanout)
3910 return -EINVAL;
3911
3912 return fanout_set_data(po, optval, optlen);
3913 }
3914 case PACKET_IGNORE_OUTGOING:
3915 {
3916 int val;
3917
3918 if (optlen != sizeof(val))
3919 return -EINVAL;
3920 if (copy_from_sockptr(&val, optval, sizeof(val)))
3921 return -EFAULT;
3922 if (val < 0 || val > 1)
3923 return -EINVAL;
3924
3925 po->prot_hook.ignore_outgoing = !!val;
3926 return 0;
3927 }
3928 case PACKET_TX_HAS_OFF:
3929 {
3930 unsigned int val;
3931
3932 if (optlen != sizeof(val))
3933 return -EINVAL;
3934 if (copy_from_sockptr(&val, optval, sizeof(val)))
3935 return -EFAULT;
3936
3937 lock_sock(sk);
3938 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
3939 po->tp_tx_has_off = !!val;
3940
3941 release_sock(sk);
3942 return 0;
3943 }
3944 case PACKET_QDISC_BYPASS:
3945 {
3946 int val;
3947
3948 if (optlen != sizeof(val))
3949 return -EINVAL;
3950 if (copy_from_sockptr(&val, optval, sizeof(val)))
3951 return -EFAULT;
3952
3953 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3954 return 0;
3955 }
3956 default:
3957 return -ENOPROTOOPT;
3958 }
3959}
3960
3961static int packet_getsockopt(struct socket *sock, int level, int optname,
3962 char __user *optval, int __user *optlen)
3963{
3964 int len;
3965 int val, lv = sizeof(val);
3966 struct sock *sk = sock->sk;
3967 struct packet_sock *po = pkt_sk(sk);
3968 void *data = &val;
3969 union tpacket_stats_u st;
3970 struct tpacket_rollover_stats rstats;
3971 int drops;
3972
3973 if (level != SOL_PACKET)
3974 return -ENOPROTOOPT;
3975
3976 if (get_user(len, optlen))
3977 return -EFAULT;
3978
3979 if (len < 0)
3980 return -EINVAL;
3981
3982 switch (optname) {
3983 case PACKET_STATISTICS:
3984 spin_lock_bh(&sk->sk_receive_queue.lock);
3985 memcpy(&st, &po->stats, sizeof(st));
3986 memset(&po->stats, 0, sizeof(po->stats));
3987 spin_unlock_bh(&sk->sk_receive_queue.lock);
3988 drops = atomic_xchg(&po->tp_drops, 0);
3989
3990 if (po->tp_version == TPACKET_V3) {
3991 lv = sizeof(struct tpacket_stats_v3);
3992 st.stats3.tp_drops = drops;
3993 st.stats3.tp_packets += drops;
3994 data = &st.stats3;
3995 } else {
3996 lv = sizeof(struct tpacket_stats);
3997 st.stats1.tp_drops = drops;
3998 st.stats1.tp_packets += drops;
3999 data = &st.stats1;
4000 }
4001
4002 break;
4003 case PACKET_AUXDATA:
4004 val = po->auxdata;
4005 break;
4006 case PACKET_ORIGDEV:
4007 val = po->origdev;
4008 break;
4009 case PACKET_VNET_HDR:
4010 val = po->has_vnet_hdr;
4011 break;
4012 case PACKET_VERSION:
4013 val = po->tp_version;
4014 break;
4015 case PACKET_HDRLEN:
4016 if (len > sizeof(int))
4017 len = sizeof(int);
4018 if (len < sizeof(int))
4019 return -EINVAL;
4020 if (copy_from_user(&val, optval, len))
4021 return -EFAULT;
4022 switch (val) {
4023 case TPACKET_V1:
4024 val = sizeof(struct tpacket_hdr);
4025 break;
4026 case TPACKET_V2:
4027 val = sizeof(struct tpacket2_hdr);
4028 break;
4029 case TPACKET_V3:
4030 val = sizeof(struct tpacket3_hdr);
4031 break;
4032 default:
4033 return -EINVAL;
4034 }
4035 break;
4036 case PACKET_RESERVE:
4037 val = po->tp_reserve;
4038 break;
4039 case PACKET_LOSS:
4040 val = po->tp_loss;
4041 break;
4042 case PACKET_TIMESTAMP:
4043 val = po->tp_tstamp;
4044 break;
4045 case PACKET_FANOUT:
4046 val = (po->fanout ?
4047 ((u32)po->fanout->id |
4048 ((u32)po->fanout->type << 16) |
4049 ((u32)po->fanout->flags << 24)) :
4050 0);
4051 break;
4052 case PACKET_IGNORE_OUTGOING:
4053 val = po->prot_hook.ignore_outgoing;
4054 break;
4055 case PACKET_ROLLOVER_STATS:
4056 if (!po->rollover)
4057 return -EINVAL;
4058 rstats.tp_all = atomic_long_read(&po->rollover->num);
4059 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4060 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4061 data = &rstats;
4062 lv = sizeof(rstats);
4063 break;
4064 case PACKET_TX_HAS_OFF:
4065 val = po->tp_tx_has_off;
4066 break;
4067 case PACKET_QDISC_BYPASS:
4068 val = packet_use_direct_xmit(po);
4069 break;
4070 default:
4071 return -ENOPROTOOPT;
4072 }
4073
4074 if (len > lv)
4075 len = lv;
4076 if (put_user(len, optlen))
4077 return -EFAULT;
4078 if (copy_to_user(optval, data, len))
4079 return -EFAULT;
4080 return 0;
4081}
4082
4083static int packet_notifier(struct notifier_block *this,
4084 unsigned long msg, void *ptr)
4085{
4086 struct sock *sk;
4087 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4088 struct net *net = dev_net(dev);
4089
4090 rcu_read_lock();
4091 sk_for_each_rcu(sk, &net->packet.sklist) {
4092 struct packet_sock *po = pkt_sk(sk);
4093
4094 switch (msg) {
4095 case NETDEV_UNREGISTER:
4096 if (po->mclist)
4097 packet_dev_mclist_delete(dev, &po->mclist);
4098 fallthrough;
4099
4100 case NETDEV_DOWN:
4101 if (dev->ifindex == po->ifindex) {
4102 spin_lock(&po->bind_lock);
4103 if (po->running) {
4104 __unregister_prot_hook(sk, false);
4105 sk->sk_err = ENETDOWN;
4106 if (!sock_flag(sk, SOCK_DEAD))
4107 sk_error_report(sk);
4108 }
4109 if (msg == NETDEV_UNREGISTER) {
4110 packet_cached_dev_reset(po);
4111 WRITE_ONCE(po->ifindex, -1);
4112 if (po->prot_hook.dev)
4113 dev_put(po->prot_hook.dev);
4114 po->prot_hook.dev = NULL;
4115 }
4116 spin_unlock(&po->bind_lock);
4117 }
4118 break;
4119 case NETDEV_UP:
4120 if (dev->ifindex == po->ifindex) {
4121 spin_lock(&po->bind_lock);
4122 if (po->num)
4123 register_prot_hook(sk);
4124 spin_unlock(&po->bind_lock);
4125 }
4126 break;
4127 }
4128 }
4129 rcu_read_unlock();
4130 return NOTIFY_DONE;
4131}
4132
4133
4134static int packet_ioctl(struct socket *sock, unsigned int cmd,
4135 unsigned long arg)
4136{
4137 struct sock *sk = sock->sk;
4138
4139 switch (cmd) {
4140 case SIOCOUTQ:
4141 {
4142 int amount = sk_wmem_alloc_get(sk);
4143
4144 return put_user(amount, (int __user *)arg);
4145 }
4146 case SIOCINQ:
4147 {
4148 struct sk_buff *skb;
4149 int amount = 0;
4150
4151 spin_lock_bh(&sk->sk_receive_queue.lock);
4152 skb = skb_peek(&sk->sk_receive_queue);
4153 if (skb)
4154 amount = skb->len;
4155 spin_unlock_bh(&sk->sk_receive_queue.lock);
4156 return put_user(amount, (int __user *)arg);
4157 }
4158#ifdef CONFIG_INET
4159 case SIOCADDRT:
4160 case SIOCDELRT:
4161 case SIOCDARP:
4162 case SIOCGARP:
4163 case SIOCSARP:
4164 case SIOCGIFADDR:
4165 case SIOCSIFADDR:
4166 case SIOCGIFBRDADDR:
4167 case SIOCSIFBRDADDR:
4168 case SIOCGIFNETMASK:
4169 case SIOCSIFNETMASK:
4170 case SIOCGIFDSTADDR:
4171 case SIOCSIFDSTADDR:
4172 case SIOCSIFFLAGS:
4173 return inet_dgram_ops.ioctl(sock, cmd, arg);
4174#endif
4175
4176 default:
4177 return -ENOIOCTLCMD;
4178 }
4179 return 0;
4180}
4181
4182static __poll_t packet_poll(struct file *file, struct socket *sock,
4183 poll_table *wait)
4184{
4185 struct sock *sk = sock->sk;
4186 struct packet_sock *po = pkt_sk(sk);
4187 __poll_t mask = datagram_poll(file, sock, wait);
4188
4189 spin_lock_bh(&sk->sk_receive_queue.lock);
4190 if (po->rx_ring.pg_vec) {
4191 if (!packet_previous_rx_frame(po, &po->rx_ring,
4192 TP_STATUS_KERNEL))
4193 mask |= EPOLLIN | EPOLLRDNORM;
4194 }
4195 packet_rcv_try_clear_pressure(po);
4196 spin_unlock_bh(&sk->sk_receive_queue.lock);
4197 spin_lock_bh(&sk->sk_write_queue.lock);
4198 if (po->tx_ring.pg_vec) {
4199 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4200 mask |= EPOLLOUT | EPOLLWRNORM;
4201 }
4202 spin_unlock_bh(&sk->sk_write_queue.lock);
4203 return mask;
4204}
4205
4206
4207/* Dirty? Well, I still did not learn better way to account
4208 * for user mmaps.
4209 */
4210
4211static void packet_mm_open(struct vm_area_struct *vma)
4212{
4213 struct file *file = vma->vm_file;
4214 struct socket *sock = file->private_data;
4215 struct sock *sk = sock->sk;
4216
4217 if (sk)
4218 atomic_inc(&pkt_sk(sk)->mapped);
4219}
4220
4221static void packet_mm_close(struct vm_area_struct *vma)
4222{
4223 struct file *file = vma->vm_file;
4224 struct socket *sock = file->private_data;
4225 struct sock *sk = sock->sk;
4226
4227 if (sk)
4228 atomic_dec(&pkt_sk(sk)->mapped);
4229}
4230
4231static const struct vm_operations_struct packet_mmap_ops = {
4232 .open = packet_mm_open,
4233 .close = packet_mm_close,
4234};
4235
4236static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4237 unsigned int len)
4238{
4239 int i;
4240
4241 for (i = 0; i < len; i++) {
4242 if (likely(pg_vec[i].buffer)) {
4243 if (is_vmalloc_addr(pg_vec[i].buffer))
4244 vfree(pg_vec[i].buffer);
4245 else
4246 free_pages((unsigned long)pg_vec[i].buffer,
4247 order);
4248 pg_vec[i].buffer = NULL;
4249 }
4250 }
4251 kfree(pg_vec);
4252}
4253
4254static char *alloc_one_pg_vec_page(unsigned long order)
4255{
4256 char *buffer;
4257 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4258 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4259
4260 buffer = (char *) __get_free_pages(gfp_flags, order);
4261 if (buffer)
4262 return buffer;
4263
4264 /* __get_free_pages failed, fall back to vmalloc */
4265 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4266 if (buffer)
4267 return buffer;
4268
4269 /* vmalloc failed, lets dig into swap here */
4270 gfp_flags &= ~__GFP_NORETRY;
4271 buffer = (char *) __get_free_pages(gfp_flags, order);
4272 if (buffer)
4273 return buffer;
4274
4275 /* complete and utter failure */
4276 return NULL;
4277}
4278
4279static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4280{
4281 unsigned int block_nr = req->tp_block_nr;
4282 struct pgv *pg_vec;
4283 int i;
4284
4285 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4286 if (unlikely(!pg_vec))
4287 goto out;
4288
4289 for (i = 0; i < block_nr; i++) {
4290 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4291 if (unlikely(!pg_vec[i].buffer))
4292 goto out_free_pgvec;
4293 }
4294
4295out:
4296 return pg_vec;
4297
4298out_free_pgvec:
4299 free_pg_vec(pg_vec, order, block_nr);
4300 pg_vec = NULL;
4301 goto out;
4302}
4303
4304static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4305 int closing, int tx_ring)
4306{
4307 struct pgv *pg_vec = NULL;
4308 struct packet_sock *po = pkt_sk(sk);
4309 unsigned long *rx_owner_map = NULL;
4310 int was_running, order = 0;
4311 struct packet_ring_buffer *rb;
4312 struct sk_buff_head *rb_queue;
4313 __be16 num;
4314 int err;
4315 /* Added to avoid minimal code churn */
4316 struct tpacket_req *req = &req_u->req;
4317
4318 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4319 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4320
4321 err = -EBUSY;
4322 if (!closing) {
4323 if (atomic_read(&po->mapped))
4324 goto out;
4325 if (packet_read_pending(rb))
4326 goto out;
4327 }
4328
4329 if (req->tp_block_nr) {
4330 unsigned int min_frame_size;
4331
4332 /* Sanity tests and some calculations */
4333 err = -EBUSY;
4334 if (unlikely(rb->pg_vec))
4335 goto out;
4336
4337 switch (po->tp_version) {
4338 case TPACKET_V1:
4339 po->tp_hdrlen = TPACKET_HDRLEN;
4340 break;
4341 case TPACKET_V2:
4342 po->tp_hdrlen = TPACKET2_HDRLEN;
4343 break;
4344 case TPACKET_V3:
4345 po->tp_hdrlen = TPACKET3_HDRLEN;
4346 break;
4347 }
4348
4349 err = -EINVAL;
4350 if (unlikely((int)req->tp_block_size <= 0))
4351 goto out;
4352 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4353 goto out;
4354 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4355 if (po->tp_version >= TPACKET_V3 &&
4356 req->tp_block_size <
4357 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4358 goto out;
4359 if (unlikely(req->tp_frame_size < min_frame_size))
4360 goto out;
4361 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4362 goto out;
4363
4364 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4365 if (unlikely(rb->frames_per_block == 0))
4366 goto out;
4367 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4368 goto out;
4369 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4370 req->tp_frame_nr))
4371 goto out;
4372
4373 err = -ENOMEM;
4374 order = get_order(req->tp_block_size);
4375 pg_vec = alloc_pg_vec(req, order);
4376 if (unlikely(!pg_vec))
4377 goto out;
4378 switch (po->tp_version) {
4379 case TPACKET_V3:
4380 /* Block transmit is not supported yet */
4381 if (!tx_ring) {
4382 init_prb_bdqc(po, rb, pg_vec, req_u);
4383 } else {
4384 struct tpacket_req3 *req3 = &req_u->req3;
4385
4386 if (req3->tp_retire_blk_tov ||
4387 req3->tp_sizeof_priv ||
4388 req3->tp_feature_req_word) {
4389 err = -EINVAL;
4390 goto out_free_pg_vec;
4391 }
4392 }
4393 break;
4394 default:
4395 if (!tx_ring) {
4396 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4397 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4398 if (!rx_owner_map)
4399 goto out_free_pg_vec;
4400 }
4401 break;
4402 }
4403 }
4404 /* Done */
4405 else {
4406 err = -EINVAL;
4407 if (unlikely(req->tp_frame_nr))
4408 goto out;
4409 }
4410
4411
4412 /* Detach socket from network */
4413 spin_lock(&po->bind_lock);
4414 was_running = po->running;
4415 num = po->num;
4416 if (was_running) {
4417 WRITE_ONCE(po->num, 0);
4418 __unregister_prot_hook(sk, false);
4419 }
4420 spin_unlock(&po->bind_lock);
4421
4422 synchronize_net();
4423
4424 err = -EBUSY;
4425 mutex_lock(&po->pg_vec_lock);
4426 if (closing || atomic_read(&po->mapped) == 0) {
4427 err = 0;
4428 spin_lock_bh(&rb_queue->lock);
4429 swap(rb->pg_vec, pg_vec);
4430 if (po->tp_version <= TPACKET_V2)
4431 swap(rb->rx_owner_map, rx_owner_map);
4432 rb->frame_max = (req->tp_frame_nr - 1);
4433 rb->head = 0;
4434 rb->frame_size = req->tp_frame_size;
4435 spin_unlock_bh(&rb_queue->lock);
4436
4437 swap(rb->pg_vec_order, order);
4438 swap(rb->pg_vec_len, req->tp_block_nr);
4439
4440 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4441 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4442 tpacket_rcv : packet_rcv;
4443 skb_queue_purge(rb_queue);
4444 if (atomic_read(&po->mapped))
4445 pr_err("packet_mmap: vma is busy: %d\n",
4446 atomic_read(&po->mapped));
4447 }
4448 mutex_unlock(&po->pg_vec_lock);
4449
4450 spin_lock(&po->bind_lock);
4451 if (was_running) {
4452 WRITE_ONCE(po->num, num);
4453 register_prot_hook(sk);
4454 }
4455 spin_unlock(&po->bind_lock);
4456 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4457 /* Because we don't support block-based V3 on tx-ring */
4458 if (!tx_ring)
4459 prb_shutdown_retire_blk_timer(po, rb_queue);
4460 }
4461
4462out_free_pg_vec:
4463 bitmap_free(rx_owner_map);
4464 if (pg_vec)
4465 free_pg_vec(pg_vec, order, req->tp_block_nr);
4466out:
4467 return err;
4468}
4469
4470static int packet_mmap(struct file *file, struct socket *sock,
4471 struct vm_area_struct *vma)
4472{
4473 struct sock *sk = sock->sk;
4474 struct packet_sock *po = pkt_sk(sk);
4475 unsigned long size, expected_size;
4476 struct packet_ring_buffer *rb;
4477 unsigned long start;
4478 int err = -EINVAL;
4479 int i;
4480
4481 if (vma->vm_pgoff)
4482 return -EINVAL;
4483
4484 mutex_lock(&po->pg_vec_lock);
4485
4486 expected_size = 0;
4487 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4488 if (rb->pg_vec) {
4489 expected_size += rb->pg_vec_len
4490 * rb->pg_vec_pages
4491 * PAGE_SIZE;
4492 }
4493 }
4494
4495 if (expected_size == 0)
4496 goto out;
4497
4498 size = vma->vm_end - vma->vm_start;
4499 if (size != expected_size)
4500 goto out;
4501
4502 start = vma->vm_start;
4503 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4504 if (rb->pg_vec == NULL)
4505 continue;
4506
4507 for (i = 0; i < rb->pg_vec_len; i++) {
4508 struct page *page;
4509 void *kaddr = rb->pg_vec[i].buffer;
4510 int pg_num;
4511
4512 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4513 page = pgv_to_page(kaddr);
4514 err = vm_insert_page(vma, start, page);
4515 if (unlikely(err))
4516 goto out;
4517 start += PAGE_SIZE;
4518 kaddr += PAGE_SIZE;
4519 }
4520 }
4521 }
4522
4523 atomic_inc(&po->mapped);
4524 vma->vm_ops = &packet_mmap_ops;
4525 err = 0;
4526
4527out:
4528 mutex_unlock(&po->pg_vec_lock);
4529 return err;
4530}
4531
4532static const struct proto_ops packet_ops_spkt = {
4533 .family = PF_PACKET,
4534 .owner = THIS_MODULE,
4535 .release = packet_release,
4536 .bind = packet_bind_spkt,
4537 .connect = sock_no_connect,
4538 .socketpair = sock_no_socketpair,
4539 .accept = sock_no_accept,
4540 .getname = packet_getname_spkt,
4541 .poll = datagram_poll,
4542 .ioctl = packet_ioctl,
4543 .gettstamp = sock_gettstamp,
4544 .listen = sock_no_listen,
4545 .shutdown = sock_no_shutdown,
4546 .sendmsg = packet_sendmsg_spkt,
4547 .recvmsg = packet_recvmsg,
4548 .mmap = sock_no_mmap,
4549 .sendpage = sock_no_sendpage,
4550};
4551
4552static const struct proto_ops packet_ops = {
4553 .family = PF_PACKET,
4554 .owner = THIS_MODULE,
4555 .release = packet_release,
4556 .bind = packet_bind,
4557 .connect = sock_no_connect,
4558 .socketpair = sock_no_socketpair,
4559 .accept = sock_no_accept,
4560 .getname = packet_getname,
4561 .poll = packet_poll,
4562 .ioctl = packet_ioctl,
4563 .gettstamp = sock_gettstamp,
4564 .listen = sock_no_listen,
4565 .shutdown = sock_no_shutdown,
4566 .setsockopt = packet_setsockopt,
4567 .getsockopt = packet_getsockopt,
4568 .sendmsg = packet_sendmsg,
4569 .recvmsg = packet_recvmsg,
4570 .mmap = packet_mmap,
4571 .sendpage = sock_no_sendpage,
4572};
4573
4574static const struct net_proto_family packet_family_ops = {
4575 .family = PF_PACKET,
4576 .create = packet_create,
4577 .owner = THIS_MODULE,
4578};
4579
4580static struct notifier_block packet_netdev_notifier = {
4581 .notifier_call = packet_notifier,
4582};
4583
4584#ifdef CONFIG_PROC_FS
4585
4586static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4587 __acquires(RCU)
4588{
4589 struct net *net = seq_file_net(seq);
4590
4591 rcu_read_lock();
4592 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4593}
4594
4595static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4596{
4597 struct net *net = seq_file_net(seq);
4598 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4599}
4600
4601static void packet_seq_stop(struct seq_file *seq, void *v)
4602 __releases(RCU)
4603{
4604 rcu_read_unlock();
4605}
4606
4607static int packet_seq_show(struct seq_file *seq, void *v)
4608{
4609 if (v == SEQ_START_TOKEN)
4610 seq_printf(seq,
4611 "%*sRefCnt Type Proto Iface R Rmem User Inode\n",
4612 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4613 else {
4614 struct sock *s = sk_entry(v);
4615 const struct packet_sock *po = pkt_sk(s);
4616
4617 seq_printf(seq,
4618 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4619 s,
4620 refcount_read(&s->sk_refcnt),
4621 s->sk_type,
4622 ntohs(READ_ONCE(po->num)),
4623 READ_ONCE(po->ifindex),
4624 po->running,
4625 atomic_read(&s->sk_rmem_alloc),
4626 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4627 sock_i_ino(s));
4628 }
4629
4630 return 0;
4631}
4632
4633static const struct seq_operations packet_seq_ops = {
4634 .start = packet_seq_start,
4635 .next = packet_seq_next,
4636 .stop = packet_seq_stop,
4637 .show = packet_seq_show,
4638};
4639#endif
4640
4641static int __net_init packet_net_init(struct net *net)
4642{
4643 mutex_init(&net->packet.sklist_lock);
4644 INIT_HLIST_HEAD(&net->packet.sklist);
4645
4646#ifdef CONFIG_PROC_FS
4647 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4648 sizeof(struct seq_net_private)))
4649 return -ENOMEM;
4650#endif /* CONFIG_PROC_FS */
4651
4652 return 0;
4653}
4654
4655static void __net_exit packet_net_exit(struct net *net)
4656{
4657 remove_proc_entry("packet", net->proc_net);
4658 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4659}
4660
4661static struct pernet_operations packet_net_ops = {
4662 .init = packet_net_init,
4663 .exit = packet_net_exit,
4664};
4665
4666
4667static void __exit packet_exit(void)
4668{
4669 unregister_netdevice_notifier(&packet_netdev_notifier);
4670 unregister_pernet_subsys(&packet_net_ops);
4671 sock_unregister(PF_PACKET);
4672 proto_unregister(&packet_proto);
4673}
4674
4675static int __init packet_init(void)
4676{
4677 int rc;
4678
4679 rc = proto_register(&packet_proto, 0);
4680 if (rc)
4681 goto out;
4682 rc = sock_register(&packet_family_ops);
4683 if (rc)
4684 goto out_proto;
4685 rc = register_pernet_subsys(&packet_net_ops);
4686 if (rc)
4687 goto out_sock;
4688 rc = register_netdevice_notifier(&packet_netdev_notifier);
4689 if (rc)
4690 goto out_pernet;
4691
4692 return 0;
4693
4694out_pernet:
4695 unregister_pernet_subsys(&packet_net_ops);
4696out_sock:
4697 sock_unregister(PF_PACKET);
4698out_proto:
4699 proto_unregister(&packet_proto);
4700out:
4701 return rc;
4702}
4703
4704module_init(packet_init);
4705module_exit(packet_exit);
4706MODULE_LICENSE("GPL");
4707MODULE_ALIAS_NETPROTO(PF_PACKET);