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_xmit(const struct packet_sock *po, struct sk_buff *skb)
274{
275 if (!packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS))
276 return dev_queue_xmit(skb);
277
278#ifdef CONFIG_NETFILTER_EGRESS
279 if (nf_hook_egress_active()) {
280 skb = nf_hook_direct_egress(skb);
281 if (!skb)
282 return NET_XMIT_DROP;
283 }
284#endif
285 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
286}
287
288static struct net_device *packet_cached_dev_get(struct packet_sock *po)
289{
290 struct net_device *dev;
291
292 rcu_read_lock();
293 dev = rcu_dereference(po->cached_dev);
294 dev_hold(dev);
295 rcu_read_unlock();
296
297 return dev;
298}
299
300static void packet_cached_dev_assign(struct packet_sock *po,
301 struct net_device *dev)
302{
303 rcu_assign_pointer(po->cached_dev, dev);
304}
305
306static void packet_cached_dev_reset(struct packet_sock *po)
307{
308 RCU_INIT_POINTER(po->cached_dev, NULL);
309}
310
311static u16 packet_pick_tx_queue(struct sk_buff *skb)
312{
313 struct net_device *dev = skb->dev;
314 const struct net_device_ops *ops = dev->netdev_ops;
315 int cpu = raw_smp_processor_id();
316 u16 queue_index;
317
318#ifdef CONFIG_XPS
319 skb->sender_cpu = cpu + 1;
320#endif
321 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
322 if (ops->ndo_select_queue) {
323 queue_index = ops->ndo_select_queue(dev, skb, NULL);
324 queue_index = netdev_cap_txqueue(dev, queue_index);
325 } else {
326 queue_index = netdev_pick_tx(dev, skb, NULL);
327 }
328
329 return queue_index;
330}
331
332/* __register_prot_hook must be invoked through register_prot_hook
333 * or from a context in which asynchronous accesses to the packet
334 * socket is not possible (packet_create()).
335 */
336static void __register_prot_hook(struct sock *sk)
337{
338 struct packet_sock *po = pkt_sk(sk);
339
340 if (!packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
341 if (po->fanout)
342 __fanout_link(sk, po);
343 else
344 dev_add_pack(&po->prot_hook);
345
346 sock_hold(sk);
347 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 1);
348 }
349}
350
351static void register_prot_hook(struct sock *sk)
352{
353 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
354 __register_prot_hook(sk);
355}
356
357/* If the sync parameter is true, we will temporarily drop
358 * the po->bind_lock and do a synchronize_net to make sure no
359 * asynchronous packet processing paths still refer to the elements
360 * of po->prot_hook. If the sync parameter is false, it is the
361 * callers responsibility to take care of this.
362 */
363static void __unregister_prot_hook(struct sock *sk, bool sync)
364{
365 struct packet_sock *po = pkt_sk(sk);
366
367 lockdep_assert_held_once(&po->bind_lock);
368
369 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 0);
370
371 if (po->fanout)
372 __fanout_unlink(sk, po);
373 else
374 __dev_remove_pack(&po->prot_hook);
375
376 __sock_put(sk);
377
378 if (sync) {
379 spin_unlock(&po->bind_lock);
380 synchronize_net();
381 spin_lock(&po->bind_lock);
382 }
383}
384
385static void unregister_prot_hook(struct sock *sk, bool sync)
386{
387 struct packet_sock *po = pkt_sk(sk);
388
389 if (packet_sock_flag(po, PACKET_SOCK_RUNNING))
390 __unregister_prot_hook(sk, sync);
391}
392
393static inline struct page * __pure pgv_to_page(void *addr)
394{
395 if (is_vmalloc_addr(addr))
396 return vmalloc_to_page(addr);
397 return virt_to_page(addr);
398}
399
400static void __packet_set_status(struct packet_sock *po, void *frame, int status)
401{
402 union tpacket_uhdr h;
403
404 /* WRITE_ONCE() are paired with READ_ONCE() in __packet_get_status */
405
406 h.raw = frame;
407 switch (po->tp_version) {
408 case TPACKET_V1:
409 WRITE_ONCE(h.h1->tp_status, status);
410 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
411 break;
412 case TPACKET_V2:
413 WRITE_ONCE(h.h2->tp_status, status);
414 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
415 break;
416 case TPACKET_V3:
417 WRITE_ONCE(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 /* READ_ONCE() are paired with WRITE_ONCE() in __packet_set_status */
435
436 h.raw = frame;
437 switch (po->tp_version) {
438 case TPACKET_V1:
439 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
440 return READ_ONCE(h.h1->tp_status);
441 case TPACKET_V2:
442 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
443 return READ_ONCE(h.h2->tp_status);
444 case TPACKET_V3:
445 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
446 return READ_ONCE(h.h3->tp_status);
447 default:
448 WARN(1, "TPACKET version not supported.\n");
449 BUG();
450 return 0;
451 }
452}
453
454static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
455 unsigned int flags)
456{
457 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
458
459 if (shhwtstamps &&
460 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
461 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
462 return TP_STATUS_TS_RAW_HARDWARE;
463
464 if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
465 ktime_to_timespec64_cond(skb_tstamp(skb), ts))
466 return TP_STATUS_TS_SOFTWARE;
467
468 return 0;
469}
470
471static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
472 struct sk_buff *skb)
473{
474 union tpacket_uhdr h;
475 struct timespec64 ts;
476 __u32 ts_status;
477
478 if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp))))
479 return 0;
480
481 h.raw = frame;
482 /*
483 * versions 1 through 3 overflow the timestamps in y2106, since they
484 * all store the seconds in a 32-bit unsigned integer.
485 * If we create a version 4, that should have a 64-bit timestamp,
486 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
487 * nanoseconds.
488 */
489 switch (po->tp_version) {
490 case TPACKET_V1:
491 h.h1->tp_sec = ts.tv_sec;
492 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
493 break;
494 case TPACKET_V2:
495 h.h2->tp_sec = ts.tv_sec;
496 h.h2->tp_nsec = ts.tv_nsec;
497 break;
498 case TPACKET_V3:
499 h.h3->tp_sec = ts.tv_sec;
500 h.h3->tp_nsec = ts.tv_nsec;
501 break;
502 default:
503 WARN(1, "TPACKET version not supported.\n");
504 BUG();
505 }
506
507 /* one flush is safe, as both fields always lie on the same cacheline */
508 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
509 smp_wmb();
510
511 return ts_status;
512}
513
514static void *packet_lookup_frame(const struct packet_sock *po,
515 const struct packet_ring_buffer *rb,
516 unsigned int position,
517 int status)
518{
519 unsigned int pg_vec_pos, frame_offset;
520 union tpacket_uhdr h;
521
522 pg_vec_pos = position / rb->frames_per_block;
523 frame_offset = position % rb->frames_per_block;
524
525 h.raw = rb->pg_vec[pg_vec_pos].buffer +
526 (frame_offset * rb->frame_size);
527
528 if (status != __packet_get_status(po, h.raw))
529 return NULL;
530
531 return h.raw;
532}
533
534static void *packet_current_frame(struct packet_sock *po,
535 struct packet_ring_buffer *rb,
536 int status)
537{
538 return packet_lookup_frame(po, rb, rb->head, status);
539}
540
541static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
542{
543 del_timer_sync(&pkc->retire_blk_timer);
544}
545
546static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
547 struct sk_buff_head *rb_queue)
548{
549 struct tpacket_kbdq_core *pkc;
550
551 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
552
553 spin_lock_bh(&rb_queue->lock);
554 pkc->delete_blk_timer = 1;
555 spin_unlock_bh(&rb_queue->lock);
556
557 prb_del_retire_blk_timer(pkc);
558}
559
560static void prb_setup_retire_blk_timer(struct packet_sock *po)
561{
562 struct tpacket_kbdq_core *pkc;
563
564 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
565 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
566 0);
567 pkc->retire_blk_timer.expires = jiffies;
568}
569
570static int prb_calc_retire_blk_tmo(struct packet_sock *po,
571 int blk_size_in_bytes)
572{
573 struct net_device *dev;
574 unsigned int mbits, div;
575 struct ethtool_link_ksettings ecmd;
576 int err;
577
578 rtnl_lock();
579 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
580 if (unlikely(!dev)) {
581 rtnl_unlock();
582 return DEFAULT_PRB_RETIRE_TOV;
583 }
584 err = __ethtool_get_link_ksettings(dev, &ecmd);
585 rtnl_unlock();
586 if (err)
587 return DEFAULT_PRB_RETIRE_TOV;
588
589 /* If the link speed is so slow you don't really
590 * need to worry about perf anyways
591 */
592 if (ecmd.base.speed < SPEED_1000 ||
593 ecmd.base.speed == SPEED_UNKNOWN)
594 return DEFAULT_PRB_RETIRE_TOV;
595
596 div = ecmd.base.speed / 1000;
597 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
598
599 if (div)
600 mbits /= div;
601
602 if (div)
603 return mbits + 1;
604 return mbits;
605}
606
607static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
608 union tpacket_req_u *req_u)
609{
610 p1->feature_req_word = req_u->req3.tp_feature_req_word;
611}
612
613static void init_prb_bdqc(struct packet_sock *po,
614 struct packet_ring_buffer *rb,
615 struct pgv *pg_vec,
616 union tpacket_req_u *req_u)
617{
618 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
619 struct tpacket_block_desc *pbd;
620
621 memset(p1, 0x0, sizeof(*p1));
622
623 p1->knxt_seq_num = 1;
624 p1->pkbdq = pg_vec;
625 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
626 p1->pkblk_start = pg_vec[0].buffer;
627 p1->kblk_size = req_u->req3.tp_block_size;
628 p1->knum_blocks = req_u->req3.tp_block_nr;
629 p1->hdrlen = po->tp_hdrlen;
630 p1->version = po->tp_version;
631 p1->last_kactive_blk_num = 0;
632 po->stats.stats3.tp_freeze_q_cnt = 0;
633 if (req_u->req3.tp_retire_blk_tov)
634 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
635 else
636 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
637 req_u->req3.tp_block_size);
638 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
639 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
640 rwlock_init(&p1->blk_fill_in_prog_lock);
641
642 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
643 prb_init_ft_ops(p1, req_u);
644 prb_setup_retire_blk_timer(po);
645 prb_open_block(p1, pbd);
646}
647
648/* Do NOT update the last_blk_num first.
649 * Assumes sk_buff_head lock is held.
650 */
651static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
652{
653 mod_timer(&pkc->retire_blk_timer,
654 jiffies + pkc->tov_in_jiffies);
655 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
656}
657
658/*
659 * Timer logic:
660 * 1) We refresh the timer only when we open a block.
661 * By doing this we don't waste cycles refreshing the timer
662 * on packet-by-packet basis.
663 *
664 * With a 1MB block-size, on a 1Gbps line, it will take
665 * i) ~8 ms to fill a block + ii) memcpy etc.
666 * In this cut we are not accounting for the memcpy time.
667 *
668 * So, if the user sets the 'tmo' to 10ms then the timer
669 * will never fire while the block is still getting filled
670 * (which is what we want). However, the user could choose
671 * to close a block early and that's fine.
672 *
673 * But when the timer does fire, we check whether or not to refresh it.
674 * Since the tmo granularity is in msecs, it is not too expensive
675 * to refresh the timer, lets say every '8' msecs.
676 * Either the user can set the 'tmo' or we can derive it based on
677 * a) line-speed and b) block-size.
678 * prb_calc_retire_blk_tmo() calculates the tmo.
679 *
680 */
681static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
682{
683 struct packet_sock *po =
684 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
685 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
686 unsigned int frozen;
687 struct tpacket_block_desc *pbd;
688
689 spin_lock(&po->sk.sk_receive_queue.lock);
690
691 frozen = prb_queue_frozen(pkc);
692 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
693
694 if (unlikely(pkc->delete_blk_timer))
695 goto out;
696
697 /* We only need to plug the race when the block is partially filled.
698 * tpacket_rcv:
699 * lock(); increment BLOCK_NUM_PKTS; unlock()
700 * copy_bits() is in progress ...
701 * timer fires on other cpu:
702 * we can't retire the current block because copy_bits
703 * is in progress.
704 *
705 */
706 if (BLOCK_NUM_PKTS(pbd)) {
707 /* Waiting for skb_copy_bits to finish... */
708 write_lock(&pkc->blk_fill_in_prog_lock);
709 write_unlock(&pkc->blk_fill_in_prog_lock);
710 }
711
712 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
713 if (!frozen) {
714 if (!BLOCK_NUM_PKTS(pbd)) {
715 /* An empty block. Just refresh the timer. */
716 goto refresh_timer;
717 }
718 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
719 if (!prb_dispatch_next_block(pkc, po))
720 goto refresh_timer;
721 else
722 goto out;
723 } else {
724 /* Case 1. Queue was frozen because user-space was
725 * lagging behind.
726 */
727 if (prb_curr_blk_in_use(pbd)) {
728 /*
729 * Ok, user-space is still behind.
730 * So just refresh the timer.
731 */
732 goto refresh_timer;
733 } else {
734 /* Case 2. queue was frozen,user-space caught up,
735 * now the link went idle && the timer fired.
736 * We don't have a block to close.So we open this
737 * block and restart the timer.
738 * opening a block thaws the queue,restarts timer
739 * Thawing/timer-refresh is a side effect.
740 */
741 prb_open_block(pkc, pbd);
742 goto out;
743 }
744 }
745 }
746
747refresh_timer:
748 _prb_refresh_rx_retire_blk_timer(pkc);
749
750out:
751 spin_unlock(&po->sk.sk_receive_queue.lock);
752}
753
754static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
755 struct tpacket_block_desc *pbd1, __u32 status)
756{
757 /* Flush everything minus the block header */
758
759#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
760 u8 *start, *end;
761
762 start = (u8 *)pbd1;
763
764 /* Skip the block header(we know header WILL fit in 4K) */
765 start += PAGE_SIZE;
766
767 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
768 for (; start < end; start += PAGE_SIZE)
769 flush_dcache_page(pgv_to_page(start));
770
771 smp_wmb();
772#endif
773
774 /* Now update the block status. */
775
776 BLOCK_STATUS(pbd1) = status;
777
778 /* Flush the block header */
779
780#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
781 start = (u8 *)pbd1;
782 flush_dcache_page(pgv_to_page(start));
783
784 smp_wmb();
785#endif
786}
787
788/*
789 * Side effect:
790 *
791 * 1) flush the block
792 * 2) Increment active_blk_num
793 *
794 * Note:We DONT refresh the timer on purpose.
795 * Because almost always the next block will be opened.
796 */
797static void prb_close_block(struct tpacket_kbdq_core *pkc1,
798 struct tpacket_block_desc *pbd1,
799 struct packet_sock *po, unsigned int stat)
800{
801 __u32 status = TP_STATUS_USER | stat;
802
803 struct tpacket3_hdr *last_pkt;
804 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
805 struct sock *sk = &po->sk;
806
807 if (atomic_read(&po->tp_drops))
808 status |= TP_STATUS_LOSING;
809
810 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
811 last_pkt->tp_next_offset = 0;
812
813 /* Get the ts of the last pkt */
814 if (BLOCK_NUM_PKTS(pbd1)) {
815 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
816 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
817 } else {
818 /* Ok, we tmo'd - so get the current time.
819 *
820 * It shouldn't really happen as we don't close empty
821 * blocks. See prb_retire_rx_blk_timer_expired().
822 */
823 struct timespec64 ts;
824 ktime_get_real_ts64(&ts);
825 h1->ts_last_pkt.ts_sec = ts.tv_sec;
826 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
827 }
828
829 smp_wmb();
830
831 /* Flush the block */
832 prb_flush_block(pkc1, pbd1, status);
833
834 sk->sk_data_ready(sk);
835
836 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
837}
838
839static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
840{
841 pkc->reset_pending_on_curr_blk = 0;
842}
843
844/*
845 * Side effect of opening a block:
846 *
847 * 1) prb_queue is thawed.
848 * 2) retire_blk_timer is refreshed.
849 *
850 */
851static void prb_open_block(struct tpacket_kbdq_core *pkc1,
852 struct tpacket_block_desc *pbd1)
853{
854 struct timespec64 ts;
855 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
856
857 smp_rmb();
858
859 /* We could have just memset this but we will lose the
860 * flexibility of making the priv area sticky
861 */
862
863 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
864 BLOCK_NUM_PKTS(pbd1) = 0;
865 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866
867 ktime_get_real_ts64(&ts);
868
869 h1->ts_first_pkt.ts_sec = ts.tv_sec;
870 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
871
872 pkc1->pkblk_start = (char *)pbd1;
873 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
874
875 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
876 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
877
878 pbd1->version = pkc1->version;
879 pkc1->prev = pkc1->nxt_offset;
880 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
881
882 prb_thaw_queue(pkc1);
883 _prb_refresh_rx_retire_blk_timer(pkc1);
884
885 smp_wmb();
886}
887
888/*
889 * Queue freeze logic:
890 * 1) Assume tp_block_nr = 8 blocks.
891 * 2) At time 't0', user opens Rx ring.
892 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
893 * 4) user-space is either sleeping or processing block '0'.
894 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
895 * it will close block-7,loop around and try to fill block '0'.
896 * call-flow:
897 * __packet_lookup_frame_in_block
898 * prb_retire_current_block()
899 * prb_dispatch_next_block()
900 * |->(BLOCK_STATUS == USER) evaluates to true
901 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
902 * 6) Now there are two cases:
903 * 6.1) Link goes idle right after the queue is frozen.
904 * But remember, the last open_block() refreshed the timer.
905 * When this timer expires,it will refresh itself so that we can
906 * re-open block-0 in near future.
907 * 6.2) Link is busy and keeps on receiving packets. This is a simple
908 * case and __packet_lookup_frame_in_block will check if block-0
909 * is free and can now be re-used.
910 */
911static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
912 struct packet_sock *po)
913{
914 pkc->reset_pending_on_curr_blk = 1;
915 po->stats.stats3.tp_freeze_q_cnt++;
916}
917
918#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
919
920/*
921 * If the next block is free then we will dispatch it
922 * and return a good offset.
923 * Else, we will freeze the queue.
924 * So, caller must check the return value.
925 */
926static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
927 struct packet_sock *po)
928{
929 struct tpacket_block_desc *pbd;
930
931 smp_rmb();
932
933 /* 1. Get current block num */
934 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
935
936 /* 2. If this block is currently in_use then freeze the queue */
937 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
938 prb_freeze_queue(pkc, po);
939 return NULL;
940 }
941
942 /*
943 * 3.
944 * open this block and return the offset where the first packet
945 * needs to get stored.
946 */
947 prb_open_block(pkc, pbd);
948 return (void *)pkc->nxt_offset;
949}
950
951static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
952 struct packet_sock *po, unsigned int status)
953{
954 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
955
956 /* retire/close the current block */
957 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
958 /*
959 * Plug the case where copy_bits() is in progress on
960 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
961 * have space to copy the pkt in the current block and
962 * called prb_retire_current_block()
963 *
964 * We don't need to worry about the TMO case because
965 * the timer-handler already handled this case.
966 */
967 if (!(status & TP_STATUS_BLK_TMO)) {
968 /* Waiting for skb_copy_bits to finish... */
969 write_lock(&pkc->blk_fill_in_prog_lock);
970 write_unlock(&pkc->blk_fill_in_prog_lock);
971 }
972 prb_close_block(pkc, pbd, po, status);
973 return;
974 }
975}
976
977static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
978{
979 return TP_STATUS_USER & BLOCK_STATUS(pbd);
980}
981
982static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
983{
984 return pkc->reset_pending_on_curr_blk;
985}
986
987static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
988 __releases(&pkc->blk_fill_in_prog_lock)
989{
990 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
991
992 read_unlock(&pkc->blk_fill_in_prog_lock);
993}
994
995static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
996 struct tpacket3_hdr *ppd)
997{
998 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
999}
1000
1001static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1002 struct tpacket3_hdr *ppd)
1003{
1004 ppd->hv1.tp_rxhash = 0;
1005}
1006
1007static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1008 struct tpacket3_hdr *ppd)
1009{
1010 if (skb_vlan_tag_present(pkc->skb)) {
1011 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1012 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1013 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1014 } else {
1015 ppd->hv1.tp_vlan_tci = 0;
1016 ppd->hv1.tp_vlan_tpid = 0;
1017 ppd->tp_status = TP_STATUS_AVAILABLE;
1018 }
1019}
1020
1021static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1022 struct tpacket3_hdr *ppd)
1023{
1024 ppd->hv1.tp_padding = 0;
1025 prb_fill_vlan_info(pkc, ppd);
1026
1027 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1028 prb_fill_rxhash(pkc, ppd);
1029 else
1030 prb_clear_rxhash(pkc, ppd);
1031}
1032
1033static void prb_fill_curr_block(char *curr,
1034 struct tpacket_kbdq_core *pkc,
1035 struct tpacket_block_desc *pbd,
1036 unsigned int len)
1037 __acquires(&pkc->blk_fill_in_prog_lock)
1038{
1039 struct tpacket3_hdr *ppd;
1040
1041 ppd = (struct tpacket3_hdr *)curr;
1042 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1043 pkc->prev = curr;
1044 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1045 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1046 BLOCK_NUM_PKTS(pbd) += 1;
1047 read_lock(&pkc->blk_fill_in_prog_lock);
1048 prb_run_all_ft_ops(pkc, ppd);
1049}
1050
1051/* Assumes caller has the sk->rx_queue.lock */
1052static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1053 struct sk_buff *skb,
1054 unsigned int len
1055 )
1056{
1057 struct tpacket_kbdq_core *pkc;
1058 struct tpacket_block_desc *pbd;
1059 char *curr, *end;
1060
1061 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1062 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063
1064 /* Queue is frozen when user space is lagging behind */
1065 if (prb_queue_frozen(pkc)) {
1066 /*
1067 * Check if that last block which caused the queue to freeze,
1068 * is still in_use by user-space.
1069 */
1070 if (prb_curr_blk_in_use(pbd)) {
1071 /* Can't record this packet */
1072 return NULL;
1073 } else {
1074 /*
1075 * Ok, the block was released by user-space.
1076 * Now let's open that block.
1077 * opening a block also thaws the queue.
1078 * Thawing is a side effect.
1079 */
1080 prb_open_block(pkc, pbd);
1081 }
1082 }
1083
1084 smp_mb();
1085 curr = pkc->nxt_offset;
1086 pkc->skb = skb;
1087 end = (char *)pbd + pkc->kblk_size;
1088
1089 /* first try the current block */
1090 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1091 prb_fill_curr_block(curr, pkc, pbd, len);
1092 return (void *)curr;
1093 }
1094
1095 /* Ok, close the current block */
1096 prb_retire_current_block(pkc, po, 0);
1097
1098 /* Now, try to dispatch the next block */
1099 curr = (char *)prb_dispatch_next_block(pkc, po);
1100 if (curr) {
1101 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1102 prb_fill_curr_block(curr, pkc, pbd, len);
1103 return (void *)curr;
1104 }
1105
1106 /*
1107 * No free blocks are available.user_space hasn't caught up yet.
1108 * Queue was just frozen and now this packet will get dropped.
1109 */
1110 return NULL;
1111}
1112
1113static void *packet_current_rx_frame(struct packet_sock *po,
1114 struct sk_buff *skb,
1115 int status, unsigned int len)
1116{
1117 char *curr = NULL;
1118 switch (po->tp_version) {
1119 case TPACKET_V1:
1120 case TPACKET_V2:
1121 curr = packet_lookup_frame(po, &po->rx_ring,
1122 po->rx_ring.head, status);
1123 return curr;
1124 case TPACKET_V3:
1125 return __packet_lookup_frame_in_block(po, skb, len);
1126 default:
1127 WARN(1, "TPACKET version not supported\n");
1128 BUG();
1129 return NULL;
1130 }
1131}
1132
1133static void *prb_lookup_block(const struct packet_sock *po,
1134 const struct packet_ring_buffer *rb,
1135 unsigned int idx,
1136 int status)
1137{
1138 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1139 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1140
1141 if (status != BLOCK_STATUS(pbd))
1142 return NULL;
1143 return pbd;
1144}
1145
1146static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1147{
1148 unsigned int prev;
1149 if (rb->prb_bdqc.kactive_blk_num)
1150 prev = rb->prb_bdqc.kactive_blk_num-1;
1151 else
1152 prev = rb->prb_bdqc.knum_blocks-1;
1153 return prev;
1154}
1155
1156/* Assumes caller has held the rx_queue.lock */
1157static void *__prb_previous_block(struct packet_sock *po,
1158 struct packet_ring_buffer *rb,
1159 int status)
1160{
1161 unsigned int previous = prb_previous_blk_num(rb);
1162 return prb_lookup_block(po, rb, previous, status);
1163}
1164
1165static void *packet_previous_rx_frame(struct packet_sock *po,
1166 struct packet_ring_buffer *rb,
1167 int status)
1168{
1169 if (po->tp_version <= TPACKET_V2)
1170 return packet_previous_frame(po, rb, status);
1171
1172 return __prb_previous_block(po, rb, status);
1173}
1174
1175static void packet_increment_rx_head(struct packet_sock *po,
1176 struct packet_ring_buffer *rb)
1177{
1178 switch (po->tp_version) {
1179 case TPACKET_V1:
1180 case TPACKET_V2:
1181 return packet_increment_head(rb);
1182 case TPACKET_V3:
1183 default:
1184 WARN(1, "TPACKET version not supported.\n");
1185 BUG();
1186 return;
1187 }
1188}
1189
1190static void *packet_previous_frame(struct packet_sock *po,
1191 struct packet_ring_buffer *rb,
1192 int status)
1193{
1194 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1195 return packet_lookup_frame(po, rb, previous, status);
1196}
1197
1198static void packet_increment_head(struct packet_ring_buffer *buff)
1199{
1200 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1201}
1202
1203static void packet_inc_pending(struct packet_ring_buffer *rb)
1204{
1205 this_cpu_inc(*rb->pending_refcnt);
1206}
1207
1208static void packet_dec_pending(struct packet_ring_buffer *rb)
1209{
1210 this_cpu_dec(*rb->pending_refcnt);
1211}
1212
1213static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1214{
1215 unsigned int refcnt = 0;
1216 int cpu;
1217
1218 /* We don't use pending refcount in rx_ring. */
1219 if (rb->pending_refcnt == NULL)
1220 return 0;
1221
1222 for_each_possible_cpu(cpu)
1223 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1224
1225 return refcnt;
1226}
1227
1228static int packet_alloc_pending(struct packet_sock *po)
1229{
1230 po->rx_ring.pending_refcnt = NULL;
1231
1232 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1233 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1234 return -ENOBUFS;
1235
1236 return 0;
1237}
1238
1239static void packet_free_pending(struct packet_sock *po)
1240{
1241 free_percpu(po->tx_ring.pending_refcnt);
1242}
1243
1244#define ROOM_POW_OFF 2
1245#define ROOM_NONE 0x0
1246#define ROOM_LOW 0x1
1247#define ROOM_NORMAL 0x2
1248
1249static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1250{
1251 int idx, len;
1252
1253 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1254 idx = READ_ONCE(po->rx_ring.head);
1255 if (pow_off)
1256 idx += len >> pow_off;
1257 if (idx >= len)
1258 idx -= len;
1259 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1260}
1261
1262static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1263{
1264 int idx, len;
1265
1266 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1267 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1268 if (pow_off)
1269 idx += len >> pow_off;
1270 if (idx >= len)
1271 idx -= len;
1272 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1273}
1274
1275static int __packet_rcv_has_room(const struct packet_sock *po,
1276 const struct sk_buff *skb)
1277{
1278 const struct sock *sk = &po->sk;
1279 int ret = ROOM_NONE;
1280
1281 if (po->prot_hook.func != tpacket_rcv) {
1282 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1283 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1284 - (skb ? skb->truesize : 0);
1285
1286 if (avail > (rcvbuf >> ROOM_POW_OFF))
1287 return ROOM_NORMAL;
1288 else if (avail > 0)
1289 return ROOM_LOW;
1290 else
1291 return ROOM_NONE;
1292 }
1293
1294 if (po->tp_version == TPACKET_V3) {
1295 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1296 ret = ROOM_NORMAL;
1297 else if (__tpacket_v3_has_room(po, 0))
1298 ret = ROOM_LOW;
1299 } else {
1300 if (__tpacket_has_room(po, ROOM_POW_OFF))
1301 ret = ROOM_NORMAL;
1302 else if (__tpacket_has_room(po, 0))
1303 ret = ROOM_LOW;
1304 }
1305
1306 return ret;
1307}
1308
1309static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1310{
1311 bool pressure;
1312 int ret;
1313
1314 ret = __packet_rcv_has_room(po, skb);
1315 pressure = ret != ROOM_NORMAL;
1316
1317 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure)
1318 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure);
1319
1320 return ret;
1321}
1322
1323static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1324{
1325 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) &&
1326 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1327 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false);
1328}
1329
1330static void packet_sock_destruct(struct sock *sk)
1331{
1332 skb_queue_purge(&sk->sk_error_queue);
1333
1334 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1335 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1336
1337 if (!sock_flag(sk, SOCK_DEAD)) {
1338 pr_err("Attempt to release alive packet socket: %p\n", sk);
1339 return;
1340 }
1341}
1342
1343static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1344{
1345 u32 *history = po->rollover->history;
1346 u32 victim, rxhash;
1347 int i, count = 0;
1348
1349 rxhash = skb_get_hash(skb);
1350 for (i = 0; i < ROLLOVER_HLEN; i++)
1351 if (READ_ONCE(history[i]) == rxhash)
1352 count++;
1353
1354 victim = get_random_u32_below(ROLLOVER_HLEN);
1355
1356 /* Avoid dirtying the cache line if possible */
1357 if (READ_ONCE(history[victim]) != rxhash)
1358 WRITE_ONCE(history[victim], rxhash);
1359
1360 return count > (ROLLOVER_HLEN >> 1);
1361}
1362
1363static unsigned int fanout_demux_hash(struct packet_fanout *f,
1364 struct sk_buff *skb,
1365 unsigned int num)
1366{
1367 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1368}
1369
1370static unsigned int fanout_demux_lb(struct packet_fanout *f,
1371 struct sk_buff *skb,
1372 unsigned int num)
1373{
1374 unsigned int val = atomic_inc_return(&f->rr_cur);
1375
1376 return val % num;
1377}
1378
1379static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1380 struct sk_buff *skb,
1381 unsigned int num)
1382{
1383 return smp_processor_id() % num;
1384}
1385
1386static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1387 struct sk_buff *skb,
1388 unsigned int num)
1389{
1390 return get_random_u32_below(num);
1391}
1392
1393static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1394 struct sk_buff *skb,
1395 unsigned int idx, bool try_self,
1396 unsigned int num)
1397{
1398 struct packet_sock *po, *po_next, *po_skip = NULL;
1399 unsigned int i, j, room = ROOM_NONE;
1400
1401 po = pkt_sk(rcu_dereference(f->arr[idx]));
1402
1403 if (try_self) {
1404 room = packet_rcv_has_room(po, skb);
1405 if (room == ROOM_NORMAL ||
1406 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1407 return idx;
1408 po_skip = po;
1409 }
1410
1411 i = j = min_t(int, po->rollover->sock, num - 1);
1412 do {
1413 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1414 if (po_next != po_skip &&
1415 !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1416 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1417 if (i != j)
1418 po->rollover->sock = i;
1419 atomic_long_inc(&po->rollover->num);
1420 if (room == ROOM_LOW)
1421 atomic_long_inc(&po->rollover->num_huge);
1422 return i;
1423 }
1424
1425 if (++i == num)
1426 i = 0;
1427 } while (i != j);
1428
1429 atomic_long_inc(&po->rollover->num_failed);
1430 return idx;
1431}
1432
1433static unsigned int fanout_demux_qm(struct packet_fanout *f,
1434 struct sk_buff *skb,
1435 unsigned int num)
1436{
1437 return skb_get_queue_mapping(skb) % num;
1438}
1439
1440static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1441 struct sk_buff *skb,
1442 unsigned int num)
1443{
1444 struct bpf_prog *prog;
1445 unsigned int ret = 0;
1446
1447 rcu_read_lock();
1448 prog = rcu_dereference(f->bpf_prog);
1449 if (prog)
1450 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1451 rcu_read_unlock();
1452
1453 return ret;
1454}
1455
1456static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1457{
1458 return f->flags & (flag >> 8);
1459}
1460
1461static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1462 struct packet_type *pt, struct net_device *orig_dev)
1463{
1464 struct packet_fanout *f = pt->af_packet_priv;
1465 unsigned int num = READ_ONCE(f->num_members);
1466 struct net *net = read_pnet(&f->net);
1467 struct packet_sock *po;
1468 unsigned int idx;
1469
1470 if (!net_eq(dev_net(dev), net) || !num) {
1471 kfree_skb(skb);
1472 return 0;
1473 }
1474
1475 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1476 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1477 if (!skb)
1478 return 0;
1479 }
1480 switch (f->type) {
1481 case PACKET_FANOUT_HASH:
1482 default:
1483 idx = fanout_demux_hash(f, skb, num);
1484 break;
1485 case PACKET_FANOUT_LB:
1486 idx = fanout_demux_lb(f, skb, num);
1487 break;
1488 case PACKET_FANOUT_CPU:
1489 idx = fanout_demux_cpu(f, skb, num);
1490 break;
1491 case PACKET_FANOUT_RND:
1492 idx = fanout_demux_rnd(f, skb, num);
1493 break;
1494 case PACKET_FANOUT_QM:
1495 idx = fanout_demux_qm(f, skb, num);
1496 break;
1497 case PACKET_FANOUT_ROLLOVER:
1498 idx = fanout_demux_rollover(f, skb, 0, false, num);
1499 break;
1500 case PACKET_FANOUT_CBPF:
1501 case PACKET_FANOUT_EBPF:
1502 idx = fanout_demux_bpf(f, skb, num);
1503 break;
1504 }
1505
1506 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1507 idx = fanout_demux_rollover(f, skb, idx, true, num);
1508
1509 po = pkt_sk(rcu_dereference(f->arr[idx]));
1510 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1511}
1512
1513DEFINE_MUTEX(fanout_mutex);
1514EXPORT_SYMBOL_GPL(fanout_mutex);
1515static LIST_HEAD(fanout_list);
1516static u16 fanout_next_id;
1517
1518static void __fanout_link(struct sock *sk, struct packet_sock *po)
1519{
1520 struct packet_fanout *f = po->fanout;
1521
1522 spin_lock(&f->lock);
1523 rcu_assign_pointer(f->arr[f->num_members], sk);
1524 smp_wmb();
1525 f->num_members++;
1526 if (f->num_members == 1)
1527 dev_add_pack(&f->prot_hook);
1528 spin_unlock(&f->lock);
1529}
1530
1531static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1532{
1533 struct packet_fanout *f = po->fanout;
1534 int i;
1535
1536 spin_lock(&f->lock);
1537 for (i = 0; i < f->num_members; i++) {
1538 if (rcu_dereference_protected(f->arr[i],
1539 lockdep_is_held(&f->lock)) == sk)
1540 break;
1541 }
1542 BUG_ON(i >= f->num_members);
1543 rcu_assign_pointer(f->arr[i],
1544 rcu_dereference_protected(f->arr[f->num_members - 1],
1545 lockdep_is_held(&f->lock)));
1546 f->num_members--;
1547 if (f->num_members == 0)
1548 __dev_remove_pack(&f->prot_hook);
1549 spin_unlock(&f->lock);
1550}
1551
1552static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1553{
1554 if (sk->sk_family != PF_PACKET)
1555 return false;
1556
1557 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1558}
1559
1560static void fanout_init_data(struct packet_fanout *f)
1561{
1562 switch (f->type) {
1563 case PACKET_FANOUT_LB:
1564 atomic_set(&f->rr_cur, 0);
1565 break;
1566 case PACKET_FANOUT_CBPF:
1567 case PACKET_FANOUT_EBPF:
1568 RCU_INIT_POINTER(f->bpf_prog, NULL);
1569 break;
1570 }
1571}
1572
1573static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1574{
1575 struct bpf_prog *old;
1576
1577 spin_lock(&f->lock);
1578 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1579 rcu_assign_pointer(f->bpf_prog, new);
1580 spin_unlock(&f->lock);
1581
1582 if (old) {
1583 synchronize_net();
1584 bpf_prog_destroy(old);
1585 }
1586}
1587
1588static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1589 unsigned int len)
1590{
1591 struct bpf_prog *new;
1592 struct sock_fprog fprog;
1593 int ret;
1594
1595 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1596 return -EPERM;
1597
1598 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1599 if (ret)
1600 return ret;
1601
1602 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1603 if (ret)
1604 return ret;
1605
1606 __fanout_set_data_bpf(po->fanout, new);
1607 return 0;
1608}
1609
1610static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1611 unsigned int len)
1612{
1613 struct bpf_prog *new;
1614 u32 fd;
1615
1616 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1617 return -EPERM;
1618 if (len != sizeof(fd))
1619 return -EINVAL;
1620 if (copy_from_sockptr(&fd, data, len))
1621 return -EFAULT;
1622
1623 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1624 if (IS_ERR(new))
1625 return PTR_ERR(new);
1626
1627 __fanout_set_data_bpf(po->fanout, new);
1628 return 0;
1629}
1630
1631static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1632 unsigned int len)
1633{
1634 switch (po->fanout->type) {
1635 case PACKET_FANOUT_CBPF:
1636 return fanout_set_data_cbpf(po, data, len);
1637 case PACKET_FANOUT_EBPF:
1638 return fanout_set_data_ebpf(po, data, len);
1639 default:
1640 return -EINVAL;
1641 }
1642}
1643
1644static void fanout_release_data(struct packet_fanout *f)
1645{
1646 switch (f->type) {
1647 case PACKET_FANOUT_CBPF:
1648 case PACKET_FANOUT_EBPF:
1649 __fanout_set_data_bpf(f, NULL);
1650 }
1651}
1652
1653static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1654{
1655 struct packet_fanout *f;
1656
1657 list_for_each_entry(f, &fanout_list, list) {
1658 if (f->id == candidate_id &&
1659 read_pnet(&f->net) == sock_net(sk)) {
1660 return false;
1661 }
1662 }
1663 return true;
1664}
1665
1666static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1667{
1668 u16 id = fanout_next_id;
1669
1670 do {
1671 if (__fanout_id_is_free(sk, id)) {
1672 *new_id = id;
1673 fanout_next_id = id + 1;
1674 return true;
1675 }
1676
1677 id++;
1678 } while (id != fanout_next_id);
1679
1680 return false;
1681}
1682
1683static int fanout_add(struct sock *sk, struct fanout_args *args)
1684{
1685 struct packet_rollover *rollover = NULL;
1686 struct packet_sock *po = pkt_sk(sk);
1687 u16 type_flags = args->type_flags;
1688 struct packet_fanout *f, *match;
1689 u8 type = type_flags & 0xff;
1690 u8 flags = type_flags >> 8;
1691 u16 id = args->id;
1692 int err;
1693
1694 switch (type) {
1695 case PACKET_FANOUT_ROLLOVER:
1696 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1697 return -EINVAL;
1698 break;
1699 case PACKET_FANOUT_HASH:
1700 case PACKET_FANOUT_LB:
1701 case PACKET_FANOUT_CPU:
1702 case PACKET_FANOUT_RND:
1703 case PACKET_FANOUT_QM:
1704 case PACKET_FANOUT_CBPF:
1705 case PACKET_FANOUT_EBPF:
1706 break;
1707 default:
1708 return -EINVAL;
1709 }
1710
1711 mutex_lock(&fanout_mutex);
1712
1713 err = -EALREADY;
1714 if (po->fanout)
1715 goto out;
1716
1717 if (type == PACKET_FANOUT_ROLLOVER ||
1718 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1719 err = -ENOMEM;
1720 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1721 if (!rollover)
1722 goto out;
1723 atomic_long_set(&rollover->num, 0);
1724 atomic_long_set(&rollover->num_huge, 0);
1725 atomic_long_set(&rollover->num_failed, 0);
1726 }
1727
1728 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1729 if (id != 0) {
1730 err = -EINVAL;
1731 goto out;
1732 }
1733 if (!fanout_find_new_id(sk, &id)) {
1734 err = -ENOMEM;
1735 goto out;
1736 }
1737 /* ephemeral flag for the first socket in the group: drop it */
1738 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1739 }
1740
1741 match = NULL;
1742 list_for_each_entry(f, &fanout_list, list) {
1743 if (f->id == id &&
1744 read_pnet(&f->net) == sock_net(sk)) {
1745 match = f;
1746 break;
1747 }
1748 }
1749 err = -EINVAL;
1750 if (match) {
1751 if (match->flags != flags)
1752 goto out;
1753 if (args->max_num_members &&
1754 args->max_num_members != match->max_num_members)
1755 goto out;
1756 } else {
1757 if (args->max_num_members > PACKET_FANOUT_MAX)
1758 goto out;
1759 if (!args->max_num_members)
1760 /* legacy PACKET_FANOUT_MAX */
1761 args->max_num_members = 256;
1762 err = -ENOMEM;
1763 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1764 GFP_KERNEL);
1765 if (!match)
1766 goto out;
1767 write_pnet(&match->net, sock_net(sk));
1768 match->id = id;
1769 match->type = type;
1770 match->flags = flags;
1771 INIT_LIST_HEAD(&match->list);
1772 spin_lock_init(&match->lock);
1773 refcount_set(&match->sk_ref, 0);
1774 fanout_init_data(match);
1775 match->prot_hook.type = po->prot_hook.type;
1776 match->prot_hook.dev = po->prot_hook.dev;
1777 match->prot_hook.func = packet_rcv_fanout;
1778 match->prot_hook.af_packet_priv = match;
1779 match->prot_hook.af_packet_net = read_pnet(&match->net);
1780 match->prot_hook.id_match = match_fanout_group;
1781 match->max_num_members = args->max_num_members;
1782 match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1783 list_add(&match->list, &fanout_list);
1784 }
1785 err = -EINVAL;
1786
1787 spin_lock(&po->bind_lock);
1788 if (packet_sock_flag(po, PACKET_SOCK_RUNNING) &&
1789 match->type == type &&
1790 match->prot_hook.type == po->prot_hook.type &&
1791 match->prot_hook.dev == po->prot_hook.dev) {
1792 err = -ENOSPC;
1793 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1794 __dev_remove_pack(&po->prot_hook);
1795
1796 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1797 WRITE_ONCE(po->fanout, match);
1798
1799 po->rollover = rollover;
1800 rollover = NULL;
1801 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1802 __fanout_link(sk, po);
1803 err = 0;
1804 }
1805 }
1806 spin_unlock(&po->bind_lock);
1807
1808 if (err && !refcount_read(&match->sk_ref)) {
1809 list_del(&match->list);
1810 kvfree(match);
1811 }
1812
1813out:
1814 kfree(rollover);
1815 mutex_unlock(&fanout_mutex);
1816 return err;
1817}
1818
1819/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1820 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1821 * It is the responsibility of the caller to call fanout_release_data() and
1822 * free the returned packet_fanout (after synchronize_net())
1823 */
1824static struct packet_fanout *fanout_release(struct sock *sk)
1825{
1826 struct packet_sock *po = pkt_sk(sk);
1827 struct packet_fanout *f;
1828
1829 mutex_lock(&fanout_mutex);
1830 f = po->fanout;
1831 if (f) {
1832 po->fanout = NULL;
1833
1834 if (refcount_dec_and_test(&f->sk_ref))
1835 list_del(&f->list);
1836 else
1837 f = NULL;
1838 }
1839 mutex_unlock(&fanout_mutex);
1840
1841 return f;
1842}
1843
1844static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1845 struct sk_buff *skb)
1846{
1847 /* Earlier code assumed this would be a VLAN pkt, double-check
1848 * this now that we have the actual packet in hand. We can only
1849 * do this check on Ethernet devices.
1850 */
1851 if (unlikely(dev->type != ARPHRD_ETHER))
1852 return false;
1853
1854 skb_reset_mac_header(skb);
1855 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1856}
1857
1858static const struct proto_ops packet_ops;
1859
1860static const struct proto_ops packet_ops_spkt;
1861
1862static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1863 struct packet_type *pt, struct net_device *orig_dev)
1864{
1865 struct sock *sk;
1866 struct sockaddr_pkt *spkt;
1867
1868 /*
1869 * When we registered the protocol we saved the socket in the data
1870 * field for just this event.
1871 */
1872
1873 sk = pt->af_packet_priv;
1874
1875 /*
1876 * Yank back the headers [hope the device set this
1877 * right or kerboom...]
1878 *
1879 * Incoming packets have ll header pulled,
1880 * push it back.
1881 *
1882 * For outgoing ones skb->data == skb_mac_header(skb)
1883 * so that this procedure is noop.
1884 */
1885
1886 if (skb->pkt_type == PACKET_LOOPBACK)
1887 goto out;
1888
1889 if (!net_eq(dev_net(dev), sock_net(sk)))
1890 goto out;
1891
1892 skb = skb_share_check(skb, GFP_ATOMIC);
1893 if (skb == NULL)
1894 goto oom;
1895
1896 /* drop any routing info */
1897 skb_dst_drop(skb);
1898
1899 /* drop conntrack reference */
1900 nf_reset_ct(skb);
1901
1902 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1903
1904 skb_push(skb, skb->data - skb_mac_header(skb));
1905
1906 /*
1907 * The SOCK_PACKET socket receives _all_ frames.
1908 */
1909
1910 spkt->spkt_family = dev->type;
1911 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1912 spkt->spkt_protocol = skb->protocol;
1913
1914 /*
1915 * Charge the memory to the socket. This is done specifically
1916 * to prevent sockets using all the memory up.
1917 */
1918
1919 if (sock_queue_rcv_skb(sk, skb) == 0)
1920 return 0;
1921
1922out:
1923 kfree_skb(skb);
1924oom:
1925 return 0;
1926}
1927
1928static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1929{
1930 int depth;
1931
1932 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1933 sock->type == SOCK_RAW) {
1934 skb_reset_mac_header(skb);
1935 skb->protocol = dev_parse_header_protocol(skb);
1936 }
1937
1938 /* Move network header to the right position for VLAN tagged packets */
1939 if (likely(skb->dev->type == ARPHRD_ETHER) &&
1940 eth_type_vlan(skb->protocol) &&
1941 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1942 skb_set_network_header(skb, depth);
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) || !skb->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 = READ_ONCE(sk->sk_priority);
2058 skb->mark = READ_ONCE(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, int vnet_hdr_sz)
2096{
2097 struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2098
2099 if (*len < vnet_hdr_sz)
2100 return -EINVAL;
2101 *len -= vnet_hdr_sz;
2102
2103 if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2104 return -EINVAL;
2105
2106 return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
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 enum skb_drop_reason drop_reason = SKB_CONSUMED;
2125 struct sock *sk;
2126 struct sockaddr_ll *sll;
2127 struct packet_sock *po;
2128 u8 *skb_head = skb->data;
2129 int skb_len = skb->len;
2130 unsigned int snaplen, res;
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(packet_sock_flag(po, PACKET_SOCK_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 atomic_inc(&po->tp_drops);
2221 atomic_inc(&sk->sk_drops);
2222 drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
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 kfree_skb_reason(skb, drop_reason);
2231 return 0;
2232}
2233
2234static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2235 struct packet_type *pt, struct net_device *orig_dev)
2236{
2237 enum skb_drop_reason drop_reason = SKB_CONSUMED;
2238 struct sock *sk;
2239 struct packet_sock *po;
2240 struct sockaddr_ll *sll;
2241 union tpacket_uhdr h;
2242 u8 *skb_head = skb->data;
2243 int skb_len = skb->len;
2244 unsigned int snaplen, res;
2245 unsigned long status = TP_STATUS_USER;
2246 unsigned short macoff, hdrlen;
2247 unsigned int netoff;
2248 struct sk_buff *copy_skb = NULL;
2249 struct timespec64 ts;
2250 __u32 ts_status;
2251 unsigned int slot_id = 0;
2252 int vnet_hdr_sz = 0;
2253
2254 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2255 * We may add members to them until current aligned size without forcing
2256 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2257 */
2258 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2259 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2260
2261 if (skb->pkt_type == PACKET_LOOPBACK)
2262 goto drop;
2263
2264 sk = pt->af_packet_priv;
2265 po = pkt_sk(sk);
2266
2267 if (!net_eq(dev_net(dev), sock_net(sk)))
2268 goto drop;
2269
2270 if (dev_has_header(dev)) {
2271 if (sk->sk_type != SOCK_DGRAM)
2272 skb_push(skb, skb->data - skb_mac_header(skb));
2273 else if (skb->pkt_type == PACKET_OUTGOING) {
2274 /* Special case: outgoing packets have ll header at head */
2275 skb_pull(skb, skb_network_offset(skb));
2276 }
2277 }
2278
2279 snaplen = skb->len;
2280
2281 res = run_filter(skb, sk, snaplen);
2282 if (!res)
2283 goto drop_n_restore;
2284
2285 /* If we are flooded, just give up */
2286 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2287 atomic_inc(&po->tp_drops);
2288 goto drop_n_restore;
2289 }
2290
2291 if (skb->ip_summed == CHECKSUM_PARTIAL)
2292 status |= TP_STATUS_CSUMNOTREADY;
2293 else if (skb->pkt_type != PACKET_OUTGOING &&
2294 skb_csum_unnecessary(skb))
2295 status |= TP_STATUS_CSUM_VALID;
2296 if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2297 status |= TP_STATUS_GSO_TCP;
2298
2299 if (snaplen > res)
2300 snaplen = res;
2301
2302 if (sk->sk_type == SOCK_DGRAM) {
2303 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2304 po->tp_reserve;
2305 } else {
2306 unsigned int maclen = skb_network_offset(skb);
2307 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2308 (maclen < 16 ? 16 : maclen)) +
2309 po->tp_reserve;
2310 vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2311 if (vnet_hdr_sz)
2312 netoff += vnet_hdr_sz;
2313 macoff = netoff - maclen;
2314 }
2315 if (netoff > USHRT_MAX) {
2316 atomic_inc(&po->tp_drops);
2317 goto drop_n_restore;
2318 }
2319 if (po->tp_version <= TPACKET_V2) {
2320 if (macoff + snaplen > po->rx_ring.frame_size) {
2321 if (po->copy_thresh &&
2322 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2323 if (skb_shared(skb)) {
2324 copy_skb = skb_clone(skb, GFP_ATOMIC);
2325 } else {
2326 copy_skb = skb_get(skb);
2327 skb_head = skb->data;
2328 }
2329 if (copy_skb) {
2330 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2331 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2332 skb_set_owner_r(copy_skb, sk);
2333 }
2334 }
2335 snaplen = po->rx_ring.frame_size - macoff;
2336 if ((int)snaplen < 0) {
2337 snaplen = 0;
2338 vnet_hdr_sz = 0;
2339 }
2340 }
2341 } else if (unlikely(macoff + snaplen >
2342 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2343 u32 nval;
2344
2345 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2346 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2347 snaplen, nval, macoff);
2348 snaplen = nval;
2349 if (unlikely((int)snaplen < 0)) {
2350 snaplen = 0;
2351 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2352 vnet_hdr_sz = 0;
2353 }
2354 }
2355 spin_lock(&sk->sk_receive_queue.lock);
2356 h.raw = packet_current_rx_frame(po, skb,
2357 TP_STATUS_KERNEL, (macoff+snaplen));
2358 if (!h.raw)
2359 goto drop_n_account;
2360
2361 if (po->tp_version <= TPACKET_V2) {
2362 slot_id = po->rx_ring.head;
2363 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2364 goto drop_n_account;
2365 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2366 }
2367
2368 if (vnet_hdr_sz &&
2369 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2370 sizeof(struct virtio_net_hdr),
2371 vio_le(), true, 0)) {
2372 if (po->tp_version == TPACKET_V3)
2373 prb_clear_blk_fill_status(&po->rx_ring);
2374 goto drop_n_account;
2375 }
2376
2377 if (po->tp_version <= TPACKET_V2) {
2378 packet_increment_rx_head(po, &po->rx_ring);
2379 /*
2380 * LOSING will be reported till you read the stats,
2381 * because it's COR - Clear On Read.
2382 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2383 * at packet level.
2384 */
2385 if (atomic_read(&po->tp_drops))
2386 status |= TP_STATUS_LOSING;
2387 }
2388
2389 po->stats.stats1.tp_packets++;
2390 if (copy_skb) {
2391 status |= TP_STATUS_COPY;
2392 skb_clear_delivery_time(copy_skb);
2393 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2394 }
2395 spin_unlock(&sk->sk_receive_queue.lock);
2396
2397 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2398
2399 /* Always timestamp; prefer an existing software timestamp taken
2400 * closer to the time of capture.
2401 */
2402 ts_status = tpacket_get_timestamp(skb, &ts,
2403 READ_ONCE(po->tp_tstamp) |
2404 SOF_TIMESTAMPING_SOFTWARE);
2405 if (!ts_status)
2406 ktime_get_real_ts64(&ts);
2407
2408 status |= ts_status;
2409
2410 switch (po->tp_version) {
2411 case TPACKET_V1:
2412 h.h1->tp_len = skb->len;
2413 h.h1->tp_snaplen = snaplen;
2414 h.h1->tp_mac = macoff;
2415 h.h1->tp_net = netoff;
2416 h.h1->tp_sec = ts.tv_sec;
2417 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2418 hdrlen = sizeof(*h.h1);
2419 break;
2420 case TPACKET_V2:
2421 h.h2->tp_len = skb->len;
2422 h.h2->tp_snaplen = snaplen;
2423 h.h2->tp_mac = macoff;
2424 h.h2->tp_net = netoff;
2425 h.h2->tp_sec = ts.tv_sec;
2426 h.h2->tp_nsec = ts.tv_nsec;
2427 if (skb_vlan_tag_present(skb)) {
2428 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2429 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2430 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2431 } else {
2432 h.h2->tp_vlan_tci = 0;
2433 h.h2->tp_vlan_tpid = 0;
2434 }
2435 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2436 hdrlen = sizeof(*h.h2);
2437 break;
2438 case TPACKET_V3:
2439 /* tp_nxt_offset,vlan are already populated above.
2440 * So DONT clear those fields here
2441 */
2442 h.h3->tp_status |= status;
2443 h.h3->tp_len = skb->len;
2444 h.h3->tp_snaplen = snaplen;
2445 h.h3->tp_mac = macoff;
2446 h.h3->tp_net = netoff;
2447 h.h3->tp_sec = ts.tv_sec;
2448 h.h3->tp_nsec = ts.tv_nsec;
2449 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2450 hdrlen = sizeof(*h.h3);
2451 break;
2452 default:
2453 BUG();
2454 }
2455
2456 sll = h.raw + TPACKET_ALIGN(hdrlen);
2457 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2458 sll->sll_family = AF_PACKET;
2459 sll->sll_hatype = dev->type;
2460 sll->sll_protocol = skb->protocol;
2461 sll->sll_pkttype = skb->pkt_type;
2462 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2463 sll->sll_ifindex = orig_dev->ifindex;
2464 else
2465 sll->sll_ifindex = dev->ifindex;
2466
2467 smp_mb();
2468
2469#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2470 if (po->tp_version <= TPACKET_V2) {
2471 u8 *start, *end;
2472
2473 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2474 macoff + snaplen);
2475
2476 for (start = h.raw; start < end; start += PAGE_SIZE)
2477 flush_dcache_page(pgv_to_page(start));
2478 }
2479 smp_wmb();
2480#endif
2481
2482 if (po->tp_version <= TPACKET_V2) {
2483 spin_lock(&sk->sk_receive_queue.lock);
2484 __packet_set_status(po, h.raw, status);
2485 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2486 spin_unlock(&sk->sk_receive_queue.lock);
2487 sk->sk_data_ready(sk);
2488 } else if (po->tp_version == TPACKET_V3) {
2489 prb_clear_blk_fill_status(&po->rx_ring);
2490 }
2491
2492drop_n_restore:
2493 if (skb_head != skb->data && skb_shared(skb)) {
2494 skb->data = skb_head;
2495 skb->len = skb_len;
2496 }
2497drop:
2498 kfree_skb_reason(skb, drop_reason);
2499 return 0;
2500
2501drop_n_account:
2502 spin_unlock(&sk->sk_receive_queue.lock);
2503 atomic_inc(&po->tp_drops);
2504 drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2505
2506 sk->sk_data_ready(sk);
2507 kfree_skb_reason(copy_skb, drop_reason);
2508 goto drop_n_restore;
2509}
2510
2511static void tpacket_destruct_skb(struct sk_buff *skb)
2512{
2513 struct packet_sock *po = pkt_sk(skb->sk);
2514
2515 if (likely(po->tx_ring.pg_vec)) {
2516 void *ph;
2517 __u32 ts;
2518
2519 ph = skb_zcopy_get_nouarg(skb);
2520 packet_dec_pending(&po->tx_ring);
2521
2522 ts = __packet_set_timestamp(po, ph, skb);
2523 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2524
2525 if (!packet_read_pending(&po->tx_ring))
2526 complete(&po->skb_completion);
2527 }
2528
2529 sock_wfree(skb);
2530}
2531
2532static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2533{
2534 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2535 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2536 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2537 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2538 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2539 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2540 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2541
2542 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2543 return -EINVAL;
2544
2545 return 0;
2546}
2547
2548static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2549 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz)
2550{
2551 int ret;
2552
2553 if (*len < vnet_hdr_sz)
2554 return -EINVAL;
2555 *len -= vnet_hdr_sz;
2556
2557 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2558 return -EFAULT;
2559
2560 ret = __packet_snd_vnet_parse(vnet_hdr, *len);
2561 if (ret)
2562 return ret;
2563
2564 /* move iter to point to the start of mac header */
2565 if (vnet_hdr_sz != sizeof(struct virtio_net_hdr))
2566 iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr));
2567
2568 return 0;
2569}
2570
2571static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2572 void *frame, struct net_device *dev, void *data, int tp_len,
2573 __be16 proto, unsigned char *addr, int hlen, int copylen,
2574 const struct sockcm_cookie *sockc)
2575{
2576 union tpacket_uhdr ph;
2577 int to_write, offset, len, nr_frags, len_max;
2578 struct socket *sock = po->sk.sk_socket;
2579 struct page *page;
2580 int err;
2581
2582 ph.raw = frame;
2583
2584 skb->protocol = proto;
2585 skb->dev = dev;
2586 skb->priority = READ_ONCE(po->sk.sk_priority);
2587 skb->mark = READ_ONCE(po->sk.sk_mark);
2588 skb->tstamp = sockc->transmit_time;
2589 skb_setup_tx_timestamp(skb, sockc->tsflags);
2590 skb_zcopy_set_nouarg(skb, ph.raw);
2591
2592 skb_reserve(skb, hlen);
2593 skb_reset_network_header(skb);
2594
2595 to_write = tp_len;
2596
2597 if (sock->type == SOCK_DGRAM) {
2598 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2599 NULL, tp_len);
2600 if (unlikely(err < 0))
2601 return -EINVAL;
2602 } else if (copylen) {
2603 int hdrlen = min_t(int, copylen, tp_len);
2604
2605 skb_push(skb, dev->hard_header_len);
2606 skb_put(skb, copylen - dev->hard_header_len);
2607 err = skb_store_bits(skb, 0, data, hdrlen);
2608 if (unlikely(err))
2609 return err;
2610 if (!dev_validate_header(dev, skb->data, hdrlen))
2611 return -EINVAL;
2612
2613 data += hdrlen;
2614 to_write -= hdrlen;
2615 }
2616
2617 offset = offset_in_page(data);
2618 len_max = PAGE_SIZE - offset;
2619 len = ((to_write > len_max) ? len_max : to_write);
2620
2621 skb->data_len = to_write;
2622 skb->len += to_write;
2623 skb->truesize += to_write;
2624 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2625
2626 while (likely(to_write)) {
2627 nr_frags = skb_shinfo(skb)->nr_frags;
2628
2629 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2630 pr_err("Packet exceed the number of skb frags(%u)\n",
2631 (unsigned int)MAX_SKB_FRAGS);
2632 return -EFAULT;
2633 }
2634
2635 page = pgv_to_page(data);
2636 data += len;
2637 flush_dcache_page(page);
2638 get_page(page);
2639 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2640 to_write -= len;
2641 offset = 0;
2642 len_max = PAGE_SIZE;
2643 len = ((to_write > len_max) ? len_max : to_write);
2644 }
2645
2646 packet_parse_headers(skb, sock);
2647
2648 return tp_len;
2649}
2650
2651static int tpacket_parse_header(struct packet_sock *po, void *frame,
2652 int size_max, void **data)
2653{
2654 union tpacket_uhdr ph;
2655 int tp_len, off;
2656
2657 ph.raw = frame;
2658
2659 switch (po->tp_version) {
2660 case TPACKET_V3:
2661 if (ph.h3->tp_next_offset != 0) {
2662 pr_warn_once("variable sized slot not supported");
2663 return -EINVAL;
2664 }
2665 tp_len = ph.h3->tp_len;
2666 break;
2667 case TPACKET_V2:
2668 tp_len = ph.h2->tp_len;
2669 break;
2670 default:
2671 tp_len = ph.h1->tp_len;
2672 break;
2673 }
2674 if (unlikely(tp_len > size_max)) {
2675 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2676 return -EMSGSIZE;
2677 }
2678
2679 if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) {
2680 int off_min, off_max;
2681
2682 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2683 off_max = po->tx_ring.frame_size - tp_len;
2684 if (po->sk.sk_type == SOCK_DGRAM) {
2685 switch (po->tp_version) {
2686 case TPACKET_V3:
2687 off = ph.h3->tp_net;
2688 break;
2689 case TPACKET_V2:
2690 off = ph.h2->tp_net;
2691 break;
2692 default:
2693 off = ph.h1->tp_net;
2694 break;
2695 }
2696 } else {
2697 switch (po->tp_version) {
2698 case TPACKET_V3:
2699 off = ph.h3->tp_mac;
2700 break;
2701 case TPACKET_V2:
2702 off = ph.h2->tp_mac;
2703 break;
2704 default:
2705 off = ph.h1->tp_mac;
2706 break;
2707 }
2708 }
2709 if (unlikely((off < off_min) || (off_max < off)))
2710 return -EINVAL;
2711 } else {
2712 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2713 }
2714
2715 *data = frame + off;
2716 return tp_len;
2717}
2718
2719static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2720{
2721 struct sk_buff *skb = NULL;
2722 struct net_device *dev;
2723 struct virtio_net_hdr *vnet_hdr = NULL;
2724 struct sockcm_cookie sockc;
2725 __be16 proto;
2726 int err, reserve = 0;
2727 void *ph;
2728 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2729 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2730 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2731 unsigned char *addr = NULL;
2732 int tp_len, size_max;
2733 void *data;
2734 int len_sum = 0;
2735 int status = TP_STATUS_AVAILABLE;
2736 int hlen, tlen, copylen = 0;
2737 long timeo = 0;
2738
2739 mutex_lock(&po->pg_vec_lock);
2740
2741 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2742 * we need to confirm it under protection of pg_vec_lock.
2743 */
2744 if (unlikely(!po->tx_ring.pg_vec)) {
2745 err = -EBUSY;
2746 goto out;
2747 }
2748 if (likely(saddr == NULL)) {
2749 dev = packet_cached_dev_get(po);
2750 proto = READ_ONCE(po->num);
2751 } else {
2752 err = -EINVAL;
2753 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2754 goto out;
2755 if (msg->msg_namelen < (saddr->sll_halen
2756 + offsetof(struct sockaddr_ll,
2757 sll_addr)))
2758 goto out;
2759 proto = saddr->sll_protocol;
2760 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2761 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2762 if (dev && msg->msg_namelen < dev->addr_len +
2763 offsetof(struct sockaddr_ll, sll_addr))
2764 goto out_put;
2765 addr = saddr->sll_addr;
2766 }
2767 }
2768
2769 err = -ENXIO;
2770 if (unlikely(dev == NULL))
2771 goto out;
2772 err = -ENETDOWN;
2773 if (unlikely(!(dev->flags & IFF_UP)))
2774 goto out_put;
2775
2776 sockcm_init(&sockc, &po->sk);
2777 if (msg->msg_controllen) {
2778 err = sock_cmsg_send(&po->sk, msg, &sockc);
2779 if (unlikely(err))
2780 goto out_put;
2781 }
2782
2783 if (po->sk.sk_socket->type == SOCK_RAW)
2784 reserve = dev->hard_header_len;
2785 size_max = po->tx_ring.frame_size
2786 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2787
2788 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2789 size_max = dev->mtu + reserve + VLAN_HLEN;
2790
2791 reinit_completion(&po->skb_completion);
2792
2793 do {
2794 ph = packet_current_frame(po, &po->tx_ring,
2795 TP_STATUS_SEND_REQUEST);
2796 if (unlikely(ph == NULL)) {
2797 if (need_wait && skb) {
2798 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2799 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2800 if (timeo <= 0) {
2801 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2802 goto out_put;
2803 }
2804 }
2805 /* check for additional frames */
2806 continue;
2807 }
2808
2809 skb = NULL;
2810 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2811 if (tp_len < 0)
2812 goto tpacket_error;
2813
2814 status = TP_STATUS_SEND_REQUEST;
2815 hlen = LL_RESERVED_SPACE(dev);
2816 tlen = dev->needed_tailroom;
2817 if (vnet_hdr_sz) {
2818 vnet_hdr = data;
2819 data += vnet_hdr_sz;
2820 tp_len -= vnet_hdr_sz;
2821 if (tp_len < 0 ||
2822 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2823 tp_len = -EINVAL;
2824 goto tpacket_error;
2825 }
2826 copylen = __virtio16_to_cpu(vio_le(),
2827 vnet_hdr->hdr_len);
2828 }
2829 copylen = max_t(int, copylen, dev->hard_header_len);
2830 skb = sock_alloc_send_skb(&po->sk,
2831 hlen + tlen + sizeof(struct sockaddr_ll) +
2832 (copylen - dev->hard_header_len),
2833 !need_wait, &err);
2834
2835 if (unlikely(skb == NULL)) {
2836 /* we assume the socket was initially writeable ... */
2837 if (likely(len_sum > 0))
2838 err = len_sum;
2839 goto out_status;
2840 }
2841 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2842 addr, hlen, copylen, &sockc);
2843 if (likely(tp_len >= 0) &&
2844 tp_len > dev->mtu + reserve &&
2845 !vnet_hdr_sz &&
2846 !packet_extra_vlan_len_allowed(dev, skb))
2847 tp_len = -EMSGSIZE;
2848
2849 if (unlikely(tp_len < 0)) {
2850tpacket_error:
2851 if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2852 __packet_set_status(po, ph,
2853 TP_STATUS_AVAILABLE);
2854 packet_increment_head(&po->tx_ring);
2855 kfree_skb(skb);
2856 continue;
2857 } else {
2858 status = TP_STATUS_WRONG_FORMAT;
2859 err = tp_len;
2860 goto out_status;
2861 }
2862 }
2863
2864 if (vnet_hdr_sz) {
2865 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2866 tp_len = -EINVAL;
2867 goto tpacket_error;
2868 }
2869 virtio_net_hdr_set_proto(skb, vnet_hdr);
2870 }
2871
2872 skb->destructor = tpacket_destruct_skb;
2873 __packet_set_status(po, ph, TP_STATUS_SENDING);
2874 packet_inc_pending(&po->tx_ring);
2875
2876 status = TP_STATUS_SEND_REQUEST;
2877 err = packet_xmit(po, skb);
2878 if (unlikely(err != 0)) {
2879 if (err > 0)
2880 err = net_xmit_errno(err);
2881 if (err && __packet_get_status(po, ph) ==
2882 TP_STATUS_AVAILABLE) {
2883 /* skb was destructed already */
2884 skb = NULL;
2885 goto out_status;
2886 }
2887 /*
2888 * skb was dropped but not destructed yet;
2889 * let's treat it like congestion or err < 0
2890 */
2891 err = 0;
2892 }
2893 packet_increment_head(&po->tx_ring);
2894 len_sum += tp_len;
2895 } while (likely((ph != NULL) ||
2896 /* Note: packet_read_pending() might be slow if we have
2897 * to call it as it's per_cpu variable, but in fast-path
2898 * we already short-circuit the loop with the first
2899 * condition, and luckily don't have to go that path
2900 * anyway.
2901 */
2902 (need_wait && packet_read_pending(&po->tx_ring))));
2903
2904 err = len_sum;
2905 goto out_put;
2906
2907out_status:
2908 __packet_set_status(po, ph, status);
2909 kfree_skb(skb);
2910out_put:
2911 dev_put(dev);
2912out:
2913 mutex_unlock(&po->pg_vec_lock);
2914 return err;
2915}
2916
2917static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2918 size_t reserve, size_t len,
2919 size_t linear, int noblock,
2920 int *err)
2921{
2922 struct sk_buff *skb;
2923
2924 /* Under a page? Don't bother with paged skb. */
2925 if (prepad + len < PAGE_SIZE || !linear)
2926 linear = len;
2927
2928 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
2929 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
2930 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2931 err, PAGE_ALLOC_COSTLY_ORDER);
2932 if (!skb)
2933 return NULL;
2934
2935 skb_reserve(skb, reserve);
2936 skb_put(skb, linear);
2937 skb->data_len = len - linear;
2938 skb->len += len - linear;
2939
2940 return skb;
2941}
2942
2943static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2944{
2945 struct sock *sk = sock->sk;
2946 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2947 struct sk_buff *skb;
2948 struct net_device *dev;
2949 __be16 proto;
2950 unsigned char *addr = NULL;
2951 int err, reserve = 0;
2952 struct sockcm_cookie sockc;
2953 struct virtio_net_hdr vnet_hdr = { 0 };
2954 int offset = 0;
2955 struct packet_sock *po = pkt_sk(sk);
2956 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2957 int hlen, tlen, linear;
2958 int extra_len = 0;
2959
2960 /*
2961 * Get and verify the address.
2962 */
2963
2964 if (likely(saddr == NULL)) {
2965 dev = packet_cached_dev_get(po);
2966 proto = READ_ONCE(po->num);
2967 } else {
2968 err = -EINVAL;
2969 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2970 goto out;
2971 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2972 goto out;
2973 proto = saddr->sll_protocol;
2974 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2975 if (sock->type == SOCK_DGRAM) {
2976 if (dev && msg->msg_namelen < dev->addr_len +
2977 offsetof(struct sockaddr_ll, sll_addr))
2978 goto out_unlock;
2979 addr = saddr->sll_addr;
2980 }
2981 }
2982
2983 err = -ENXIO;
2984 if (unlikely(dev == NULL))
2985 goto out_unlock;
2986 err = -ENETDOWN;
2987 if (unlikely(!(dev->flags & IFF_UP)))
2988 goto out_unlock;
2989
2990 sockcm_init(&sockc, sk);
2991 sockc.mark = READ_ONCE(sk->sk_mark);
2992 if (msg->msg_controllen) {
2993 err = sock_cmsg_send(sk, msg, &sockc);
2994 if (unlikely(err))
2995 goto out_unlock;
2996 }
2997
2998 if (sock->type == SOCK_RAW)
2999 reserve = dev->hard_header_len;
3000 if (vnet_hdr_sz) {
3001 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3002 if (err)
3003 goto out_unlock;
3004 }
3005
3006 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3007 if (!netif_supports_nofcs(dev)) {
3008 err = -EPROTONOSUPPORT;
3009 goto out_unlock;
3010 }
3011 extra_len = 4; /* We're doing our own CRC */
3012 }
3013
3014 err = -EMSGSIZE;
3015 if (!vnet_hdr.gso_type &&
3016 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3017 goto out_unlock;
3018
3019 err = -ENOBUFS;
3020 hlen = LL_RESERVED_SPACE(dev);
3021 tlen = dev->needed_tailroom;
3022 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3023 linear = max(linear, min_t(int, len, dev->hard_header_len));
3024 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3025 msg->msg_flags & MSG_DONTWAIT, &err);
3026 if (skb == NULL)
3027 goto out_unlock;
3028
3029 skb_reset_network_header(skb);
3030
3031 err = -EINVAL;
3032 if (sock->type == SOCK_DGRAM) {
3033 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3034 if (unlikely(offset < 0))
3035 goto out_free;
3036 } else if (reserve) {
3037 skb_reserve(skb, -reserve);
3038 if (len < reserve + sizeof(struct ipv6hdr) &&
3039 dev->min_header_len != dev->hard_header_len)
3040 skb_reset_network_header(skb);
3041 }
3042
3043 /* Returns -EFAULT on error */
3044 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3045 if (err)
3046 goto out_free;
3047
3048 if ((sock->type == SOCK_RAW &&
3049 !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3050 err = -EINVAL;
3051 goto out_free;
3052 }
3053
3054 skb_setup_tx_timestamp(skb, sockc.tsflags);
3055
3056 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3057 !packet_extra_vlan_len_allowed(dev, skb)) {
3058 err = -EMSGSIZE;
3059 goto out_free;
3060 }
3061
3062 skb->protocol = proto;
3063 skb->dev = dev;
3064 skb->priority = READ_ONCE(sk->sk_priority);
3065 skb->mark = sockc.mark;
3066 skb->tstamp = sockc.transmit_time;
3067
3068 if (unlikely(extra_len == 4))
3069 skb->no_fcs = 1;
3070
3071 packet_parse_headers(skb, sock);
3072
3073 if (vnet_hdr_sz) {
3074 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3075 if (err)
3076 goto out_free;
3077 len += vnet_hdr_sz;
3078 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3079 }
3080
3081 err = packet_xmit(po, skb);
3082
3083 if (unlikely(err != 0)) {
3084 if (err > 0)
3085 err = net_xmit_errno(err);
3086 if (err)
3087 goto out_unlock;
3088 }
3089
3090 dev_put(dev);
3091
3092 return len;
3093
3094out_free:
3095 kfree_skb(skb);
3096out_unlock:
3097 dev_put(dev);
3098out:
3099 return err;
3100}
3101
3102static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3103{
3104 struct sock *sk = sock->sk;
3105 struct packet_sock *po = pkt_sk(sk);
3106
3107 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3108 * tpacket_snd() will redo the check safely.
3109 */
3110 if (data_race(po->tx_ring.pg_vec))
3111 return tpacket_snd(po, msg);
3112
3113 return packet_snd(sock, msg, len);
3114}
3115
3116/*
3117 * Close a PACKET socket. This is fairly simple. We immediately go
3118 * to 'closed' state and remove our protocol entry in the device list.
3119 */
3120
3121static int packet_release(struct socket *sock)
3122{
3123 struct sock *sk = sock->sk;
3124 struct packet_sock *po;
3125 struct packet_fanout *f;
3126 struct net *net;
3127 union tpacket_req_u req_u;
3128
3129 if (!sk)
3130 return 0;
3131
3132 net = sock_net(sk);
3133 po = pkt_sk(sk);
3134
3135 mutex_lock(&net->packet.sklist_lock);
3136 sk_del_node_init_rcu(sk);
3137 mutex_unlock(&net->packet.sklist_lock);
3138
3139 sock_prot_inuse_add(net, sk->sk_prot, -1);
3140
3141 spin_lock(&po->bind_lock);
3142 unregister_prot_hook(sk, false);
3143 packet_cached_dev_reset(po);
3144
3145 if (po->prot_hook.dev) {
3146 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3147 po->prot_hook.dev = NULL;
3148 }
3149 spin_unlock(&po->bind_lock);
3150
3151 packet_flush_mclist(sk);
3152
3153 lock_sock(sk);
3154 if (po->rx_ring.pg_vec) {
3155 memset(&req_u, 0, sizeof(req_u));
3156 packet_set_ring(sk, &req_u, 1, 0);
3157 }
3158
3159 if (po->tx_ring.pg_vec) {
3160 memset(&req_u, 0, sizeof(req_u));
3161 packet_set_ring(sk, &req_u, 1, 1);
3162 }
3163 release_sock(sk);
3164
3165 f = fanout_release(sk);
3166
3167 synchronize_net();
3168
3169 kfree(po->rollover);
3170 if (f) {
3171 fanout_release_data(f);
3172 kvfree(f);
3173 }
3174 /*
3175 * Now the socket is dead. No more input will appear.
3176 */
3177 sock_orphan(sk);
3178 sock->sk = NULL;
3179
3180 /* Purge queues */
3181
3182 skb_queue_purge(&sk->sk_receive_queue);
3183 packet_free_pending(po);
3184
3185 sock_put(sk);
3186 return 0;
3187}
3188
3189/*
3190 * Attach a packet hook.
3191 */
3192
3193static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3194 __be16 proto)
3195{
3196 struct packet_sock *po = pkt_sk(sk);
3197 struct net_device *dev = NULL;
3198 bool unlisted = false;
3199 bool need_rehook;
3200 int ret = 0;
3201
3202 lock_sock(sk);
3203 spin_lock(&po->bind_lock);
3204 if (!proto)
3205 proto = po->num;
3206
3207 rcu_read_lock();
3208
3209 if (po->fanout) {
3210 ret = -EINVAL;
3211 goto out_unlock;
3212 }
3213
3214 if (name) {
3215 dev = dev_get_by_name_rcu(sock_net(sk), name);
3216 if (!dev) {
3217 ret = -ENODEV;
3218 goto out_unlock;
3219 }
3220 } else if (ifindex) {
3221 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3222 if (!dev) {
3223 ret = -ENODEV;
3224 goto out_unlock;
3225 }
3226 }
3227
3228 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3229
3230 if (need_rehook) {
3231 dev_hold(dev);
3232 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3233 rcu_read_unlock();
3234 /* prevents packet_notifier() from calling
3235 * register_prot_hook()
3236 */
3237 WRITE_ONCE(po->num, 0);
3238 __unregister_prot_hook(sk, true);
3239 rcu_read_lock();
3240 if (dev)
3241 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3242 dev->ifindex);
3243 }
3244
3245 BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3246 WRITE_ONCE(po->num, proto);
3247 po->prot_hook.type = proto;
3248
3249 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3250
3251 if (unlikely(unlisted)) {
3252 po->prot_hook.dev = NULL;
3253 WRITE_ONCE(po->ifindex, -1);
3254 packet_cached_dev_reset(po);
3255 } else {
3256 netdev_hold(dev, &po->prot_hook.dev_tracker,
3257 GFP_ATOMIC);
3258 po->prot_hook.dev = dev;
3259 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3260 packet_cached_dev_assign(po, dev);
3261 }
3262 dev_put(dev);
3263 }
3264
3265 if (proto == 0 || !need_rehook)
3266 goto out_unlock;
3267
3268 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3269 register_prot_hook(sk);
3270 } else {
3271 sk->sk_err = ENETDOWN;
3272 if (!sock_flag(sk, SOCK_DEAD))
3273 sk_error_report(sk);
3274 }
3275
3276out_unlock:
3277 rcu_read_unlock();
3278 spin_unlock(&po->bind_lock);
3279 release_sock(sk);
3280 return ret;
3281}
3282
3283/*
3284 * Bind a packet socket to a device
3285 */
3286
3287static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3288 int addr_len)
3289{
3290 struct sock *sk = sock->sk;
3291 char name[sizeof(uaddr->sa_data_min) + 1];
3292
3293 /*
3294 * Check legality
3295 */
3296
3297 if (addr_len != sizeof(struct sockaddr))
3298 return -EINVAL;
3299 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3300 * zero-terminated.
3301 */
3302 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3303 name[sizeof(uaddr->sa_data_min)] = 0;
3304
3305 return packet_do_bind(sk, name, 0, 0);
3306}
3307
3308static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3309{
3310 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3311 struct sock *sk = sock->sk;
3312
3313 /*
3314 * Check legality
3315 */
3316
3317 if (addr_len < sizeof(struct sockaddr_ll))
3318 return -EINVAL;
3319 if (sll->sll_family != AF_PACKET)
3320 return -EINVAL;
3321
3322 return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3323}
3324
3325static struct proto packet_proto = {
3326 .name = "PACKET",
3327 .owner = THIS_MODULE,
3328 .obj_size = sizeof(struct packet_sock),
3329};
3330
3331/*
3332 * Create a packet of type SOCK_PACKET.
3333 */
3334
3335static int packet_create(struct net *net, struct socket *sock, int protocol,
3336 int kern)
3337{
3338 struct sock *sk;
3339 struct packet_sock *po;
3340 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3341 int err;
3342
3343 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3344 return -EPERM;
3345 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3346 sock->type != SOCK_PACKET)
3347 return -ESOCKTNOSUPPORT;
3348
3349 sock->state = SS_UNCONNECTED;
3350
3351 err = -ENOBUFS;
3352 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3353 if (sk == NULL)
3354 goto out;
3355
3356 sock->ops = &packet_ops;
3357 if (sock->type == SOCK_PACKET)
3358 sock->ops = &packet_ops_spkt;
3359
3360 sock_init_data(sock, sk);
3361
3362 po = pkt_sk(sk);
3363 init_completion(&po->skb_completion);
3364 sk->sk_family = PF_PACKET;
3365 po->num = proto;
3366
3367 err = packet_alloc_pending(po);
3368 if (err)
3369 goto out2;
3370
3371 packet_cached_dev_reset(po);
3372
3373 sk->sk_destruct = packet_sock_destruct;
3374
3375 /*
3376 * Attach a protocol block
3377 */
3378
3379 spin_lock_init(&po->bind_lock);
3380 mutex_init(&po->pg_vec_lock);
3381 po->rollover = NULL;
3382 po->prot_hook.func = packet_rcv;
3383
3384 if (sock->type == SOCK_PACKET)
3385 po->prot_hook.func = packet_rcv_spkt;
3386
3387 po->prot_hook.af_packet_priv = sk;
3388 po->prot_hook.af_packet_net = sock_net(sk);
3389
3390 if (proto) {
3391 po->prot_hook.type = proto;
3392 __register_prot_hook(sk);
3393 }
3394
3395 mutex_lock(&net->packet.sklist_lock);
3396 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3397 mutex_unlock(&net->packet.sklist_lock);
3398
3399 sock_prot_inuse_add(net, &packet_proto, 1);
3400
3401 return 0;
3402out2:
3403 sk_free(sk);
3404out:
3405 return err;
3406}
3407
3408/*
3409 * Pull a packet from our receive queue and hand it to the user.
3410 * If necessary we block.
3411 */
3412
3413static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3414 int flags)
3415{
3416 struct sock *sk = sock->sk;
3417 struct sk_buff *skb;
3418 int copied, err;
3419 int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3420 unsigned int origlen = 0;
3421
3422 err = -EINVAL;
3423 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3424 goto out;
3425
3426#if 0
3427 /* What error should we return now? EUNATTACH? */
3428 if (pkt_sk(sk)->ifindex < 0)
3429 return -ENODEV;
3430#endif
3431
3432 if (flags & MSG_ERRQUEUE) {
3433 err = sock_recv_errqueue(sk, msg, len,
3434 SOL_PACKET, PACKET_TX_TIMESTAMP);
3435 goto out;
3436 }
3437
3438 /*
3439 * Call the generic datagram receiver. This handles all sorts
3440 * of horrible races and re-entrancy so we can forget about it
3441 * in the protocol layers.
3442 *
3443 * Now it will return ENETDOWN, if device have just gone down,
3444 * but then it will block.
3445 */
3446
3447 skb = skb_recv_datagram(sk, flags, &err);
3448
3449 /*
3450 * An error occurred so return it. Because skb_recv_datagram()
3451 * handles the blocking we don't see and worry about blocking
3452 * retries.
3453 */
3454
3455 if (skb == NULL)
3456 goto out;
3457
3458 packet_rcv_try_clear_pressure(pkt_sk(sk));
3459
3460 if (vnet_hdr_len) {
3461 err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3462 if (err)
3463 goto out_free;
3464 }
3465
3466 /* You lose any data beyond the buffer you gave. If it worries
3467 * a user program they can ask the device for its MTU
3468 * anyway.
3469 */
3470 copied = skb->len;
3471 if (copied > len) {
3472 copied = len;
3473 msg->msg_flags |= MSG_TRUNC;
3474 }
3475
3476 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3477 if (err)
3478 goto out_free;
3479
3480 if (sock->type != SOCK_PACKET) {
3481 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3482
3483 /* Original length was stored in sockaddr_ll fields */
3484 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3485 sll->sll_family = AF_PACKET;
3486 sll->sll_protocol = skb->protocol;
3487 }
3488
3489 sock_recv_cmsgs(msg, sk, skb);
3490
3491 if (msg->msg_name) {
3492 const size_t max_len = min(sizeof(skb->cb),
3493 sizeof(struct sockaddr_storage));
3494 int copy_len;
3495
3496 /* If the address length field is there to be filled
3497 * in, we fill it in now.
3498 */
3499 if (sock->type == SOCK_PACKET) {
3500 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3501 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3502 copy_len = msg->msg_namelen;
3503 } else {
3504 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3505
3506 msg->msg_namelen = sll->sll_halen +
3507 offsetof(struct sockaddr_ll, sll_addr);
3508 copy_len = msg->msg_namelen;
3509 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3510 memset(msg->msg_name +
3511 offsetof(struct sockaddr_ll, sll_addr),
3512 0, sizeof(sll->sll_addr));
3513 msg->msg_namelen = sizeof(struct sockaddr_ll);
3514 }
3515 }
3516 if (WARN_ON_ONCE(copy_len > max_len)) {
3517 copy_len = max_len;
3518 msg->msg_namelen = copy_len;
3519 }
3520 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3521 }
3522
3523 if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3524 struct tpacket_auxdata aux;
3525
3526 aux.tp_status = TP_STATUS_USER;
3527 if (skb->ip_summed == CHECKSUM_PARTIAL)
3528 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3529 else if (skb->pkt_type != PACKET_OUTGOING &&
3530 skb_csum_unnecessary(skb))
3531 aux.tp_status |= TP_STATUS_CSUM_VALID;
3532 if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3533 aux.tp_status |= TP_STATUS_GSO_TCP;
3534
3535 aux.tp_len = origlen;
3536 aux.tp_snaplen = skb->len;
3537 aux.tp_mac = 0;
3538 aux.tp_net = skb_network_offset(skb);
3539 if (skb_vlan_tag_present(skb)) {
3540 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3541 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3542 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3543 } else {
3544 aux.tp_vlan_tci = 0;
3545 aux.tp_vlan_tpid = 0;
3546 }
3547 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3548 }
3549
3550 /*
3551 * Free or return the buffer as appropriate. Again this
3552 * hides all the races and re-entrancy issues from us.
3553 */
3554 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3555
3556out_free:
3557 skb_free_datagram(sk, skb);
3558out:
3559 return err;
3560}
3561
3562static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3563 int peer)
3564{
3565 struct net_device *dev;
3566 struct sock *sk = sock->sk;
3567
3568 if (peer)
3569 return -EOPNOTSUPP;
3570
3571 uaddr->sa_family = AF_PACKET;
3572 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3573 rcu_read_lock();
3574 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3575 if (dev)
3576 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3577 rcu_read_unlock();
3578
3579 return sizeof(*uaddr);
3580}
3581
3582static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3583 int peer)
3584{
3585 struct net_device *dev;
3586 struct sock *sk = sock->sk;
3587 struct packet_sock *po = pkt_sk(sk);
3588 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3589 int ifindex;
3590
3591 if (peer)
3592 return -EOPNOTSUPP;
3593
3594 ifindex = READ_ONCE(po->ifindex);
3595 sll->sll_family = AF_PACKET;
3596 sll->sll_ifindex = ifindex;
3597 sll->sll_protocol = READ_ONCE(po->num);
3598 sll->sll_pkttype = 0;
3599 rcu_read_lock();
3600 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3601 if (dev) {
3602 sll->sll_hatype = dev->type;
3603 sll->sll_halen = dev->addr_len;
3604
3605 /* Let __fortify_memcpy_chk() know the actual buffer size. */
3606 memcpy(((struct sockaddr_storage *)sll)->__data +
3607 offsetof(struct sockaddr_ll, sll_addr) -
3608 offsetofend(struct sockaddr_ll, sll_family),
3609 dev->dev_addr, dev->addr_len);
3610 } else {
3611 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3612 sll->sll_halen = 0;
3613 }
3614 rcu_read_unlock();
3615
3616 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3617}
3618
3619static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3620 int what)
3621{
3622 switch (i->type) {
3623 case PACKET_MR_MULTICAST:
3624 if (i->alen != dev->addr_len)
3625 return -EINVAL;
3626 if (what > 0)
3627 return dev_mc_add(dev, i->addr);
3628 else
3629 return dev_mc_del(dev, i->addr);
3630 break;
3631 case PACKET_MR_PROMISC:
3632 return dev_set_promiscuity(dev, what);
3633 case PACKET_MR_ALLMULTI:
3634 return dev_set_allmulti(dev, what);
3635 case PACKET_MR_UNICAST:
3636 if (i->alen != dev->addr_len)
3637 return -EINVAL;
3638 if (what > 0)
3639 return dev_uc_add(dev, i->addr);
3640 else
3641 return dev_uc_del(dev, i->addr);
3642 break;
3643 default:
3644 break;
3645 }
3646 return 0;
3647}
3648
3649static void packet_dev_mclist_delete(struct net_device *dev,
3650 struct packet_mclist **mlp)
3651{
3652 struct packet_mclist *ml;
3653
3654 while ((ml = *mlp) != NULL) {
3655 if (ml->ifindex == dev->ifindex) {
3656 packet_dev_mc(dev, ml, -1);
3657 *mlp = ml->next;
3658 kfree(ml);
3659 } else
3660 mlp = &ml->next;
3661 }
3662}
3663
3664static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3665{
3666 struct packet_sock *po = pkt_sk(sk);
3667 struct packet_mclist *ml, *i;
3668 struct net_device *dev;
3669 int err;
3670
3671 rtnl_lock();
3672
3673 err = -ENODEV;
3674 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3675 if (!dev)
3676 goto done;
3677
3678 err = -EINVAL;
3679 if (mreq->mr_alen > dev->addr_len)
3680 goto done;
3681
3682 err = -ENOBUFS;
3683 i = kmalloc(sizeof(*i), GFP_KERNEL);
3684 if (i == NULL)
3685 goto done;
3686
3687 err = 0;
3688 for (ml = po->mclist; ml; ml = ml->next) {
3689 if (ml->ifindex == mreq->mr_ifindex &&
3690 ml->type == mreq->mr_type &&
3691 ml->alen == mreq->mr_alen &&
3692 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3693 ml->count++;
3694 /* Free the new element ... */
3695 kfree(i);
3696 goto done;
3697 }
3698 }
3699
3700 i->type = mreq->mr_type;
3701 i->ifindex = mreq->mr_ifindex;
3702 i->alen = mreq->mr_alen;
3703 memcpy(i->addr, mreq->mr_address, i->alen);
3704 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3705 i->count = 1;
3706 i->next = po->mclist;
3707 po->mclist = i;
3708 err = packet_dev_mc(dev, i, 1);
3709 if (err) {
3710 po->mclist = i->next;
3711 kfree(i);
3712 }
3713
3714done:
3715 rtnl_unlock();
3716 return err;
3717}
3718
3719static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3720{
3721 struct packet_mclist *ml, **mlp;
3722
3723 rtnl_lock();
3724
3725 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3726 if (ml->ifindex == mreq->mr_ifindex &&
3727 ml->type == mreq->mr_type &&
3728 ml->alen == mreq->mr_alen &&
3729 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3730 if (--ml->count == 0) {
3731 struct net_device *dev;
3732 *mlp = ml->next;
3733 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3734 if (dev)
3735 packet_dev_mc(dev, ml, -1);
3736 kfree(ml);
3737 }
3738 break;
3739 }
3740 }
3741 rtnl_unlock();
3742 return 0;
3743}
3744
3745static void packet_flush_mclist(struct sock *sk)
3746{
3747 struct packet_sock *po = pkt_sk(sk);
3748 struct packet_mclist *ml;
3749
3750 if (!po->mclist)
3751 return;
3752
3753 rtnl_lock();
3754 while ((ml = po->mclist) != NULL) {
3755 struct net_device *dev;
3756
3757 po->mclist = ml->next;
3758 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3759 if (dev != NULL)
3760 packet_dev_mc(dev, ml, -1);
3761 kfree(ml);
3762 }
3763 rtnl_unlock();
3764}
3765
3766static int
3767packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3768 unsigned int optlen)
3769{
3770 struct sock *sk = sock->sk;
3771 struct packet_sock *po = pkt_sk(sk);
3772 int ret;
3773
3774 if (level != SOL_PACKET)
3775 return -ENOPROTOOPT;
3776
3777 switch (optname) {
3778 case PACKET_ADD_MEMBERSHIP:
3779 case PACKET_DROP_MEMBERSHIP:
3780 {
3781 struct packet_mreq_max mreq;
3782 int len = optlen;
3783 memset(&mreq, 0, sizeof(mreq));
3784 if (len < sizeof(struct packet_mreq))
3785 return -EINVAL;
3786 if (len > sizeof(mreq))
3787 len = sizeof(mreq);
3788 if (copy_from_sockptr(&mreq, optval, len))
3789 return -EFAULT;
3790 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3791 return -EINVAL;
3792 if (optname == PACKET_ADD_MEMBERSHIP)
3793 ret = packet_mc_add(sk, &mreq);
3794 else
3795 ret = packet_mc_drop(sk, &mreq);
3796 return ret;
3797 }
3798
3799 case PACKET_RX_RING:
3800 case PACKET_TX_RING:
3801 {
3802 union tpacket_req_u req_u;
3803 int len;
3804
3805 lock_sock(sk);
3806 switch (po->tp_version) {
3807 case TPACKET_V1:
3808 case TPACKET_V2:
3809 len = sizeof(req_u.req);
3810 break;
3811 case TPACKET_V3:
3812 default:
3813 len = sizeof(req_u.req3);
3814 break;
3815 }
3816 if (optlen < len) {
3817 ret = -EINVAL;
3818 } else {
3819 if (copy_from_sockptr(&req_u.req, optval, len))
3820 ret = -EFAULT;
3821 else
3822 ret = packet_set_ring(sk, &req_u, 0,
3823 optname == PACKET_TX_RING);
3824 }
3825 release_sock(sk);
3826 return ret;
3827 }
3828 case PACKET_COPY_THRESH:
3829 {
3830 int val;
3831
3832 if (optlen != sizeof(val))
3833 return -EINVAL;
3834 if (copy_from_sockptr(&val, optval, sizeof(val)))
3835 return -EFAULT;
3836
3837 pkt_sk(sk)->copy_thresh = val;
3838 return 0;
3839 }
3840 case PACKET_VERSION:
3841 {
3842 int val;
3843
3844 if (optlen != sizeof(val))
3845 return -EINVAL;
3846 if (copy_from_sockptr(&val, optval, sizeof(val)))
3847 return -EFAULT;
3848 switch (val) {
3849 case TPACKET_V1:
3850 case TPACKET_V2:
3851 case TPACKET_V3:
3852 break;
3853 default:
3854 return -EINVAL;
3855 }
3856 lock_sock(sk);
3857 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3858 ret = -EBUSY;
3859 } else {
3860 po->tp_version = val;
3861 ret = 0;
3862 }
3863 release_sock(sk);
3864 return ret;
3865 }
3866 case PACKET_RESERVE:
3867 {
3868 unsigned int val;
3869
3870 if (optlen != sizeof(val))
3871 return -EINVAL;
3872 if (copy_from_sockptr(&val, optval, sizeof(val)))
3873 return -EFAULT;
3874 if (val > INT_MAX)
3875 return -EINVAL;
3876 lock_sock(sk);
3877 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3878 ret = -EBUSY;
3879 } else {
3880 po->tp_reserve = val;
3881 ret = 0;
3882 }
3883 release_sock(sk);
3884 return ret;
3885 }
3886 case PACKET_LOSS:
3887 {
3888 unsigned int val;
3889
3890 if (optlen != sizeof(val))
3891 return -EINVAL;
3892 if (copy_from_sockptr(&val, optval, sizeof(val)))
3893 return -EFAULT;
3894
3895 lock_sock(sk);
3896 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3897 ret = -EBUSY;
3898 } else {
3899 packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3900 ret = 0;
3901 }
3902 release_sock(sk);
3903 return ret;
3904 }
3905 case PACKET_AUXDATA:
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 packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3915 return 0;
3916 }
3917 case PACKET_ORIGDEV:
3918 {
3919 int val;
3920
3921 if (optlen < sizeof(val))
3922 return -EINVAL;
3923 if (copy_from_sockptr(&val, optval, sizeof(val)))
3924 return -EFAULT;
3925
3926 packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3927 return 0;
3928 }
3929 case PACKET_VNET_HDR:
3930 case PACKET_VNET_HDR_SZ:
3931 {
3932 int val, hdr_len;
3933
3934 if (sock->type != SOCK_RAW)
3935 return -EINVAL;
3936 if (optlen < sizeof(val))
3937 return -EINVAL;
3938 if (copy_from_sockptr(&val, optval, sizeof(val)))
3939 return -EFAULT;
3940
3941 if (optname == PACKET_VNET_HDR_SZ) {
3942 if (val && val != sizeof(struct virtio_net_hdr) &&
3943 val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
3944 return -EINVAL;
3945 hdr_len = val;
3946 } else {
3947 hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
3948 }
3949 lock_sock(sk);
3950 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3951 ret = -EBUSY;
3952 } else {
3953 WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
3954 ret = 0;
3955 }
3956 release_sock(sk);
3957 return ret;
3958 }
3959 case PACKET_TIMESTAMP:
3960 {
3961 int val;
3962
3963 if (optlen != sizeof(val))
3964 return -EINVAL;
3965 if (copy_from_sockptr(&val, optval, sizeof(val)))
3966 return -EFAULT;
3967
3968 WRITE_ONCE(po->tp_tstamp, val);
3969 return 0;
3970 }
3971 case PACKET_FANOUT:
3972 {
3973 struct fanout_args args = { 0 };
3974
3975 if (optlen != sizeof(int) && optlen != sizeof(args))
3976 return -EINVAL;
3977 if (copy_from_sockptr(&args, optval, optlen))
3978 return -EFAULT;
3979
3980 return fanout_add(sk, &args);
3981 }
3982 case PACKET_FANOUT_DATA:
3983 {
3984 /* Paired with the WRITE_ONCE() in fanout_add() */
3985 if (!READ_ONCE(po->fanout))
3986 return -EINVAL;
3987
3988 return fanout_set_data(po, optval, optlen);
3989 }
3990 case PACKET_IGNORE_OUTGOING:
3991 {
3992 int val;
3993
3994 if (optlen != sizeof(val))
3995 return -EINVAL;
3996 if (copy_from_sockptr(&val, optval, sizeof(val)))
3997 return -EFAULT;
3998 if (val < 0 || val > 1)
3999 return -EINVAL;
4000
4001 po->prot_hook.ignore_outgoing = !!val;
4002 return 0;
4003 }
4004 case PACKET_TX_HAS_OFF:
4005 {
4006 unsigned int val;
4007
4008 if (optlen != sizeof(val))
4009 return -EINVAL;
4010 if (copy_from_sockptr(&val, optval, sizeof(val)))
4011 return -EFAULT;
4012
4013 lock_sock(sk);
4014 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4015 packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4016
4017 release_sock(sk);
4018 return 0;
4019 }
4020 case PACKET_QDISC_BYPASS:
4021 {
4022 int val;
4023
4024 if (optlen != sizeof(val))
4025 return -EINVAL;
4026 if (copy_from_sockptr(&val, optval, sizeof(val)))
4027 return -EFAULT;
4028
4029 packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4030 return 0;
4031 }
4032 default:
4033 return -ENOPROTOOPT;
4034 }
4035}
4036
4037static int packet_getsockopt(struct socket *sock, int level, int optname,
4038 char __user *optval, int __user *optlen)
4039{
4040 int len;
4041 int val, lv = sizeof(val);
4042 struct sock *sk = sock->sk;
4043 struct packet_sock *po = pkt_sk(sk);
4044 void *data = &val;
4045 union tpacket_stats_u st;
4046 struct tpacket_rollover_stats rstats;
4047 int drops;
4048
4049 if (level != SOL_PACKET)
4050 return -ENOPROTOOPT;
4051
4052 if (get_user(len, optlen))
4053 return -EFAULT;
4054
4055 if (len < 0)
4056 return -EINVAL;
4057
4058 switch (optname) {
4059 case PACKET_STATISTICS:
4060 spin_lock_bh(&sk->sk_receive_queue.lock);
4061 memcpy(&st, &po->stats, sizeof(st));
4062 memset(&po->stats, 0, sizeof(po->stats));
4063 spin_unlock_bh(&sk->sk_receive_queue.lock);
4064 drops = atomic_xchg(&po->tp_drops, 0);
4065
4066 if (po->tp_version == TPACKET_V3) {
4067 lv = sizeof(struct tpacket_stats_v3);
4068 st.stats3.tp_drops = drops;
4069 st.stats3.tp_packets += drops;
4070 data = &st.stats3;
4071 } else {
4072 lv = sizeof(struct tpacket_stats);
4073 st.stats1.tp_drops = drops;
4074 st.stats1.tp_packets += drops;
4075 data = &st.stats1;
4076 }
4077
4078 break;
4079 case PACKET_AUXDATA:
4080 val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4081 break;
4082 case PACKET_ORIGDEV:
4083 val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4084 break;
4085 case PACKET_VNET_HDR:
4086 val = !!READ_ONCE(po->vnet_hdr_sz);
4087 break;
4088 case PACKET_VNET_HDR_SZ:
4089 val = READ_ONCE(po->vnet_hdr_sz);
4090 break;
4091 case PACKET_VERSION:
4092 val = po->tp_version;
4093 break;
4094 case PACKET_HDRLEN:
4095 if (len > sizeof(int))
4096 len = sizeof(int);
4097 if (len < sizeof(int))
4098 return -EINVAL;
4099 if (copy_from_user(&val, optval, len))
4100 return -EFAULT;
4101 switch (val) {
4102 case TPACKET_V1:
4103 val = sizeof(struct tpacket_hdr);
4104 break;
4105 case TPACKET_V2:
4106 val = sizeof(struct tpacket2_hdr);
4107 break;
4108 case TPACKET_V3:
4109 val = sizeof(struct tpacket3_hdr);
4110 break;
4111 default:
4112 return -EINVAL;
4113 }
4114 break;
4115 case PACKET_RESERVE:
4116 val = po->tp_reserve;
4117 break;
4118 case PACKET_LOSS:
4119 val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4120 break;
4121 case PACKET_TIMESTAMP:
4122 val = READ_ONCE(po->tp_tstamp);
4123 break;
4124 case PACKET_FANOUT:
4125 val = (po->fanout ?
4126 ((u32)po->fanout->id |
4127 ((u32)po->fanout->type << 16) |
4128 ((u32)po->fanout->flags << 24)) :
4129 0);
4130 break;
4131 case PACKET_IGNORE_OUTGOING:
4132 val = po->prot_hook.ignore_outgoing;
4133 break;
4134 case PACKET_ROLLOVER_STATS:
4135 if (!po->rollover)
4136 return -EINVAL;
4137 rstats.tp_all = atomic_long_read(&po->rollover->num);
4138 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4139 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4140 data = &rstats;
4141 lv = sizeof(rstats);
4142 break;
4143 case PACKET_TX_HAS_OFF:
4144 val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4145 break;
4146 case PACKET_QDISC_BYPASS:
4147 val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4148 break;
4149 default:
4150 return -ENOPROTOOPT;
4151 }
4152
4153 if (len > lv)
4154 len = lv;
4155 if (put_user(len, optlen))
4156 return -EFAULT;
4157 if (copy_to_user(optval, data, len))
4158 return -EFAULT;
4159 return 0;
4160}
4161
4162static int packet_notifier(struct notifier_block *this,
4163 unsigned long msg, void *ptr)
4164{
4165 struct sock *sk;
4166 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4167 struct net *net = dev_net(dev);
4168
4169 rcu_read_lock();
4170 sk_for_each_rcu(sk, &net->packet.sklist) {
4171 struct packet_sock *po = pkt_sk(sk);
4172
4173 switch (msg) {
4174 case NETDEV_UNREGISTER:
4175 if (po->mclist)
4176 packet_dev_mclist_delete(dev, &po->mclist);
4177 fallthrough;
4178
4179 case NETDEV_DOWN:
4180 if (dev->ifindex == po->ifindex) {
4181 spin_lock(&po->bind_lock);
4182 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4183 __unregister_prot_hook(sk, false);
4184 sk->sk_err = ENETDOWN;
4185 if (!sock_flag(sk, SOCK_DEAD))
4186 sk_error_report(sk);
4187 }
4188 if (msg == NETDEV_UNREGISTER) {
4189 packet_cached_dev_reset(po);
4190 WRITE_ONCE(po->ifindex, -1);
4191 netdev_put(po->prot_hook.dev,
4192 &po->prot_hook.dev_tracker);
4193 po->prot_hook.dev = NULL;
4194 }
4195 spin_unlock(&po->bind_lock);
4196 }
4197 break;
4198 case NETDEV_UP:
4199 if (dev->ifindex == po->ifindex) {
4200 spin_lock(&po->bind_lock);
4201 if (po->num)
4202 register_prot_hook(sk);
4203 spin_unlock(&po->bind_lock);
4204 }
4205 break;
4206 }
4207 }
4208 rcu_read_unlock();
4209 return NOTIFY_DONE;
4210}
4211
4212
4213static int packet_ioctl(struct socket *sock, unsigned int cmd,
4214 unsigned long arg)
4215{
4216 struct sock *sk = sock->sk;
4217
4218 switch (cmd) {
4219 case SIOCOUTQ:
4220 {
4221 int amount = sk_wmem_alloc_get(sk);
4222
4223 return put_user(amount, (int __user *)arg);
4224 }
4225 case SIOCINQ:
4226 {
4227 struct sk_buff *skb;
4228 int amount = 0;
4229
4230 spin_lock_bh(&sk->sk_receive_queue.lock);
4231 skb = skb_peek(&sk->sk_receive_queue);
4232 if (skb)
4233 amount = skb->len;
4234 spin_unlock_bh(&sk->sk_receive_queue.lock);
4235 return put_user(amount, (int __user *)arg);
4236 }
4237#ifdef CONFIG_INET
4238 case SIOCADDRT:
4239 case SIOCDELRT:
4240 case SIOCDARP:
4241 case SIOCGARP:
4242 case SIOCSARP:
4243 case SIOCGIFADDR:
4244 case SIOCSIFADDR:
4245 case SIOCGIFBRDADDR:
4246 case SIOCSIFBRDADDR:
4247 case SIOCGIFNETMASK:
4248 case SIOCSIFNETMASK:
4249 case SIOCGIFDSTADDR:
4250 case SIOCSIFDSTADDR:
4251 case SIOCSIFFLAGS:
4252 return inet_dgram_ops.ioctl(sock, cmd, arg);
4253#endif
4254
4255 default:
4256 return -ENOIOCTLCMD;
4257 }
4258 return 0;
4259}
4260
4261static __poll_t packet_poll(struct file *file, struct socket *sock,
4262 poll_table *wait)
4263{
4264 struct sock *sk = sock->sk;
4265 struct packet_sock *po = pkt_sk(sk);
4266 __poll_t mask = datagram_poll(file, sock, wait);
4267
4268 spin_lock_bh(&sk->sk_receive_queue.lock);
4269 if (po->rx_ring.pg_vec) {
4270 if (!packet_previous_rx_frame(po, &po->rx_ring,
4271 TP_STATUS_KERNEL))
4272 mask |= EPOLLIN | EPOLLRDNORM;
4273 }
4274 packet_rcv_try_clear_pressure(po);
4275 spin_unlock_bh(&sk->sk_receive_queue.lock);
4276 spin_lock_bh(&sk->sk_write_queue.lock);
4277 if (po->tx_ring.pg_vec) {
4278 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4279 mask |= EPOLLOUT | EPOLLWRNORM;
4280 }
4281 spin_unlock_bh(&sk->sk_write_queue.lock);
4282 return mask;
4283}
4284
4285
4286/* Dirty? Well, I still did not learn better way to account
4287 * for user mmaps.
4288 */
4289
4290static void packet_mm_open(struct vm_area_struct *vma)
4291{
4292 struct file *file = vma->vm_file;
4293 struct socket *sock = file->private_data;
4294 struct sock *sk = sock->sk;
4295
4296 if (sk)
4297 atomic_long_inc(&pkt_sk(sk)->mapped);
4298}
4299
4300static void packet_mm_close(struct vm_area_struct *vma)
4301{
4302 struct file *file = vma->vm_file;
4303 struct socket *sock = file->private_data;
4304 struct sock *sk = sock->sk;
4305
4306 if (sk)
4307 atomic_long_dec(&pkt_sk(sk)->mapped);
4308}
4309
4310static const struct vm_operations_struct packet_mmap_ops = {
4311 .open = packet_mm_open,
4312 .close = packet_mm_close,
4313};
4314
4315static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4316 unsigned int len)
4317{
4318 int i;
4319
4320 for (i = 0; i < len; i++) {
4321 if (likely(pg_vec[i].buffer)) {
4322 if (is_vmalloc_addr(pg_vec[i].buffer))
4323 vfree(pg_vec[i].buffer);
4324 else
4325 free_pages((unsigned long)pg_vec[i].buffer,
4326 order);
4327 pg_vec[i].buffer = NULL;
4328 }
4329 }
4330 kfree(pg_vec);
4331}
4332
4333static char *alloc_one_pg_vec_page(unsigned long order)
4334{
4335 char *buffer;
4336 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4337 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4338
4339 buffer = (char *) __get_free_pages(gfp_flags, order);
4340 if (buffer)
4341 return buffer;
4342
4343 /* __get_free_pages failed, fall back to vmalloc */
4344 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4345 if (buffer)
4346 return buffer;
4347
4348 /* vmalloc failed, lets dig into swap here */
4349 gfp_flags &= ~__GFP_NORETRY;
4350 buffer = (char *) __get_free_pages(gfp_flags, order);
4351 if (buffer)
4352 return buffer;
4353
4354 /* complete and utter failure */
4355 return NULL;
4356}
4357
4358static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4359{
4360 unsigned int block_nr = req->tp_block_nr;
4361 struct pgv *pg_vec;
4362 int i;
4363
4364 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4365 if (unlikely(!pg_vec))
4366 goto out;
4367
4368 for (i = 0; i < block_nr; i++) {
4369 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4370 if (unlikely(!pg_vec[i].buffer))
4371 goto out_free_pgvec;
4372 }
4373
4374out:
4375 return pg_vec;
4376
4377out_free_pgvec:
4378 free_pg_vec(pg_vec, order, block_nr);
4379 pg_vec = NULL;
4380 goto out;
4381}
4382
4383static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4384 int closing, int tx_ring)
4385{
4386 struct pgv *pg_vec = NULL;
4387 struct packet_sock *po = pkt_sk(sk);
4388 unsigned long *rx_owner_map = NULL;
4389 int was_running, order = 0;
4390 struct packet_ring_buffer *rb;
4391 struct sk_buff_head *rb_queue;
4392 __be16 num;
4393 int err;
4394 /* Added to avoid minimal code churn */
4395 struct tpacket_req *req = &req_u->req;
4396
4397 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4398 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4399
4400 err = -EBUSY;
4401 if (!closing) {
4402 if (atomic_long_read(&po->mapped))
4403 goto out;
4404 if (packet_read_pending(rb))
4405 goto out;
4406 }
4407
4408 if (req->tp_block_nr) {
4409 unsigned int min_frame_size;
4410
4411 /* Sanity tests and some calculations */
4412 err = -EBUSY;
4413 if (unlikely(rb->pg_vec))
4414 goto out;
4415
4416 switch (po->tp_version) {
4417 case TPACKET_V1:
4418 po->tp_hdrlen = TPACKET_HDRLEN;
4419 break;
4420 case TPACKET_V2:
4421 po->tp_hdrlen = TPACKET2_HDRLEN;
4422 break;
4423 case TPACKET_V3:
4424 po->tp_hdrlen = TPACKET3_HDRLEN;
4425 break;
4426 }
4427
4428 err = -EINVAL;
4429 if (unlikely((int)req->tp_block_size <= 0))
4430 goto out;
4431 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4432 goto out;
4433 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4434 if (po->tp_version >= TPACKET_V3 &&
4435 req->tp_block_size <
4436 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4437 goto out;
4438 if (unlikely(req->tp_frame_size < min_frame_size))
4439 goto out;
4440 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4441 goto out;
4442
4443 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4444 if (unlikely(rb->frames_per_block == 0))
4445 goto out;
4446 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4447 goto out;
4448 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4449 req->tp_frame_nr))
4450 goto out;
4451
4452 err = -ENOMEM;
4453 order = get_order(req->tp_block_size);
4454 pg_vec = alloc_pg_vec(req, order);
4455 if (unlikely(!pg_vec))
4456 goto out;
4457 switch (po->tp_version) {
4458 case TPACKET_V3:
4459 /* Block transmit is not supported yet */
4460 if (!tx_ring) {
4461 init_prb_bdqc(po, rb, pg_vec, req_u);
4462 } else {
4463 struct tpacket_req3 *req3 = &req_u->req3;
4464
4465 if (req3->tp_retire_blk_tov ||
4466 req3->tp_sizeof_priv ||
4467 req3->tp_feature_req_word) {
4468 err = -EINVAL;
4469 goto out_free_pg_vec;
4470 }
4471 }
4472 break;
4473 default:
4474 if (!tx_ring) {
4475 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4476 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4477 if (!rx_owner_map)
4478 goto out_free_pg_vec;
4479 }
4480 break;
4481 }
4482 }
4483 /* Done */
4484 else {
4485 err = -EINVAL;
4486 if (unlikely(req->tp_frame_nr))
4487 goto out;
4488 }
4489
4490
4491 /* Detach socket from network */
4492 spin_lock(&po->bind_lock);
4493 was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4494 num = po->num;
4495 if (was_running) {
4496 WRITE_ONCE(po->num, 0);
4497 __unregister_prot_hook(sk, false);
4498 }
4499 spin_unlock(&po->bind_lock);
4500
4501 synchronize_net();
4502
4503 err = -EBUSY;
4504 mutex_lock(&po->pg_vec_lock);
4505 if (closing || atomic_long_read(&po->mapped) == 0) {
4506 err = 0;
4507 spin_lock_bh(&rb_queue->lock);
4508 swap(rb->pg_vec, pg_vec);
4509 if (po->tp_version <= TPACKET_V2)
4510 swap(rb->rx_owner_map, rx_owner_map);
4511 rb->frame_max = (req->tp_frame_nr - 1);
4512 rb->head = 0;
4513 rb->frame_size = req->tp_frame_size;
4514 spin_unlock_bh(&rb_queue->lock);
4515
4516 swap(rb->pg_vec_order, order);
4517 swap(rb->pg_vec_len, req->tp_block_nr);
4518
4519 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4520 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4521 tpacket_rcv : packet_rcv;
4522 skb_queue_purge(rb_queue);
4523 if (atomic_long_read(&po->mapped))
4524 pr_err("packet_mmap: vma is busy: %ld\n",
4525 atomic_long_read(&po->mapped));
4526 }
4527 mutex_unlock(&po->pg_vec_lock);
4528
4529 spin_lock(&po->bind_lock);
4530 if (was_running) {
4531 WRITE_ONCE(po->num, num);
4532 register_prot_hook(sk);
4533 }
4534 spin_unlock(&po->bind_lock);
4535 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4536 /* Because we don't support block-based V3 on tx-ring */
4537 if (!tx_ring)
4538 prb_shutdown_retire_blk_timer(po, rb_queue);
4539 }
4540
4541out_free_pg_vec:
4542 if (pg_vec) {
4543 bitmap_free(rx_owner_map);
4544 free_pg_vec(pg_vec, order, req->tp_block_nr);
4545 }
4546out:
4547 return err;
4548}
4549
4550static int packet_mmap(struct file *file, struct socket *sock,
4551 struct vm_area_struct *vma)
4552{
4553 struct sock *sk = sock->sk;
4554 struct packet_sock *po = pkt_sk(sk);
4555 unsigned long size, expected_size;
4556 struct packet_ring_buffer *rb;
4557 unsigned long start;
4558 int err = -EINVAL;
4559 int i;
4560
4561 if (vma->vm_pgoff)
4562 return -EINVAL;
4563
4564 mutex_lock(&po->pg_vec_lock);
4565
4566 expected_size = 0;
4567 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4568 if (rb->pg_vec) {
4569 expected_size += rb->pg_vec_len
4570 * rb->pg_vec_pages
4571 * PAGE_SIZE;
4572 }
4573 }
4574
4575 if (expected_size == 0)
4576 goto out;
4577
4578 size = vma->vm_end - vma->vm_start;
4579 if (size != expected_size)
4580 goto out;
4581
4582 start = vma->vm_start;
4583 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4584 if (rb->pg_vec == NULL)
4585 continue;
4586
4587 for (i = 0; i < rb->pg_vec_len; i++) {
4588 struct page *page;
4589 void *kaddr = rb->pg_vec[i].buffer;
4590 int pg_num;
4591
4592 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4593 page = pgv_to_page(kaddr);
4594 err = vm_insert_page(vma, start, page);
4595 if (unlikely(err))
4596 goto out;
4597 start += PAGE_SIZE;
4598 kaddr += PAGE_SIZE;
4599 }
4600 }
4601 }
4602
4603 atomic_long_inc(&po->mapped);
4604 vma->vm_ops = &packet_mmap_ops;
4605 err = 0;
4606
4607out:
4608 mutex_unlock(&po->pg_vec_lock);
4609 return err;
4610}
4611
4612static const struct proto_ops packet_ops_spkt = {
4613 .family = PF_PACKET,
4614 .owner = THIS_MODULE,
4615 .release = packet_release,
4616 .bind = packet_bind_spkt,
4617 .connect = sock_no_connect,
4618 .socketpair = sock_no_socketpair,
4619 .accept = sock_no_accept,
4620 .getname = packet_getname_spkt,
4621 .poll = datagram_poll,
4622 .ioctl = packet_ioctl,
4623 .gettstamp = sock_gettstamp,
4624 .listen = sock_no_listen,
4625 .shutdown = sock_no_shutdown,
4626 .sendmsg = packet_sendmsg_spkt,
4627 .recvmsg = packet_recvmsg,
4628 .mmap = sock_no_mmap,
4629};
4630
4631static const struct proto_ops packet_ops = {
4632 .family = PF_PACKET,
4633 .owner = THIS_MODULE,
4634 .release = packet_release,
4635 .bind = packet_bind,
4636 .connect = sock_no_connect,
4637 .socketpair = sock_no_socketpair,
4638 .accept = sock_no_accept,
4639 .getname = packet_getname,
4640 .poll = packet_poll,
4641 .ioctl = packet_ioctl,
4642 .gettstamp = sock_gettstamp,
4643 .listen = sock_no_listen,
4644 .shutdown = sock_no_shutdown,
4645 .setsockopt = packet_setsockopt,
4646 .getsockopt = packet_getsockopt,
4647 .sendmsg = packet_sendmsg,
4648 .recvmsg = packet_recvmsg,
4649 .mmap = packet_mmap,
4650};
4651
4652static const struct net_proto_family packet_family_ops = {
4653 .family = PF_PACKET,
4654 .create = packet_create,
4655 .owner = THIS_MODULE,
4656};
4657
4658static struct notifier_block packet_netdev_notifier = {
4659 .notifier_call = packet_notifier,
4660};
4661
4662#ifdef CONFIG_PROC_FS
4663
4664static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4665 __acquires(RCU)
4666{
4667 struct net *net = seq_file_net(seq);
4668
4669 rcu_read_lock();
4670 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4671}
4672
4673static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4674{
4675 struct net *net = seq_file_net(seq);
4676 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4677}
4678
4679static void packet_seq_stop(struct seq_file *seq, void *v)
4680 __releases(RCU)
4681{
4682 rcu_read_unlock();
4683}
4684
4685static int packet_seq_show(struct seq_file *seq, void *v)
4686{
4687 if (v == SEQ_START_TOKEN)
4688 seq_printf(seq,
4689 "%*sRefCnt Type Proto Iface R Rmem User Inode\n",
4690 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4691 else {
4692 struct sock *s = sk_entry(v);
4693 const struct packet_sock *po = pkt_sk(s);
4694
4695 seq_printf(seq,
4696 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4697 s,
4698 refcount_read(&s->sk_refcnt),
4699 s->sk_type,
4700 ntohs(READ_ONCE(po->num)),
4701 READ_ONCE(po->ifindex),
4702 packet_sock_flag(po, PACKET_SOCK_RUNNING),
4703 atomic_read(&s->sk_rmem_alloc),
4704 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4705 sock_i_ino(s));
4706 }
4707
4708 return 0;
4709}
4710
4711static const struct seq_operations packet_seq_ops = {
4712 .start = packet_seq_start,
4713 .next = packet_seq_next,
4714 .stop = packet_seq_stop,
4715 .show = packet_seq_show,
4716};
4717#endif
4718
4719static int __net_init packet_net_init(struct net *net)
4720{
4721 mutex_init(&net->packet.sklist_lock);
4722 INIT_HLIST_HEAD(&net->packet.sklist);
4723
4724#ifdef CONFIG_PROC_FS
4725 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4726 sizeof(struct seq_net_private)))
4727 return -ENOMEM;
4728#endif /* CONFIG_PROC_FS */
4729
4730 return 0;
4731}
4732
4733static void __net_exit packet_net_exit(struct net *net)
4734{
4735 remove_proc_entry("packet", net->proc_net);
4736 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4737}
4738
4739static struct pernet_operations packet_net_ops = {
4740 .init = packet_net_init,
4741 .exit = packet_net_exit,
4742};
4743
4744
4745static void __exit packet_exit(void)
4746{
4747 sock_unregister(PF_PACKET);
4748 proto_unregister(&packet_proto);
4749 unregister_netdevice_notifier(&packet_netdev_notifier);
4750 unregister_pernet_subsys(&packet_net_ops);
4751}
4752
4753static int __init packet_init(void)
4754{
4755 int rc;
4756
4757 rc = register_pernet_subsys(&packet_net_ops);
4758 if (rc)
4759 goto out;
4760 rc = register_netdevice_notifier(&packet_netdev_notifier);
4761 if (rc)
4762 goto out_pernet;
4763 rc = proto_register(&packet_proto, 0);
4764 if (rc)
4765 goto out_notifier;
4766 rc = sock_register(&packet_family_ops);
4767 if (rc)
4768 goto out_proto;
4769
4770 return 0;
4771
4772out_proto:
4773 proto_unregister(&packet_proto);
4774out_notifier:
4775 unregister_netdevice_notifier(&packet_netdev_notifier);
4776out_pernet:
4777 unregister_pernet_subsys(&packet_net_ops);
4778out:
4779 return rc;
4780}
4781
4782module_init(packet_init);
4783module_exit(packet_exit);
4784MODULE_DESCRIPTION("Packet socket support (AF_PACKET)");
4785MODULE_LICENSE("GPL");
4786MODULE_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#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);