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 u16 vlan_get_tci(const struct sk_buff *skb, struct net_device *dev)
542{
543 struct vlan_hdr vhdr, *vh;
544 unsigned int header_len;
545
546 if (!dev)
547 return 0;
548
549 /* In the SOCK_DGRAM scenario, skb data starts at the network
550 * protocol, which is after the VLAN headers. The outer VLAN
551 * header is at the hard_header_len offset in non-variable
552 * length link layer headers. If it's a VLAN device, the
553 * min_header_len should be used to exclude the VLAN header
554 * size.
555 */
556 if (dev->min_header_len == dev->hard_header_len)
557 header_len = dev->hard_header_len;
558 else if (is_vlan_dev(dev))
559 header_len = dev->min_header_len;
560 else
561 return 0;
562
563 vh = skb_header_pointer(skb, skb_mac_offset(skb) + header_len,
564 sizeof(vhdr), &vhdr);
565 if (unlikely(!vh))
566 return 0;
567
568 return ntohs(vh->h_vlan_TCI);
569}
570
571static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb)
572{
573 __be16 proto = skb->protocol;
574
575 if (unlikely(eth_type_vlan(proto)))
576 proto = __vlan_get_protocol_offset(skb, proto,
577 skb_mac_offset(skb), NULL);
578
579 return proto;
580}
581
582static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
583{
584 del_timer_sync(&pkc->retire_blk_timer);
585}
586
587static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
588 struct sk_buff_head *rb_queue)
589{
590 struct tpacket_kbdq_core *pkc;
591
592 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
593
594 spin_lock_bh(&rb_queue->lock);
595 pkc->delete_blk_timer = 1;
596 spin_unlock_bh(&rb_queue->lock);
597
598 prb_del_retire_blk_timer(pkc);
599}
600
601static void prb_setup_retire_blk_timer(struct packet_sock *po)
602{
603 struct tpacket_kbdq_core *pkc;
604
605 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
606 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
607 0);
608 pkc->retire_blk_timer.expires = jiffies;
609}
610
611static int prb_calc_retire_blk_tmo(struct packet_sock *po,
612 int blk_size_in_bytes)
613{
614 struct net_device *dev;
615 unsigned int mbits, div;
616 struct ethtool_link_ksettings ecmd;
617 int err;
618
619 rtnl_lock();
620 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
621 if (unlikely(!dev)) {
622 rtnl_unlock();
623 return DEFAULT_PRB_RETIRE_TOV;
624 }
625 err = __ethtool_get_link_ksettings(dev, &ecmd);
626 rtnl_unlock();
627 if (err)
628 return DEFAULT_PRB_RETIRE_TOV;
629
630 /* If the link speed is so slow you don't really
631 * need to worry about perf anyways
632 */
633 if (ecmd.base.speed < SPEED_1000 ||
634 ecmd.base.speed == SPEED_UNKNOWN)
635 return DEFAULT_PRB_RETIRE_TOV;
636
637 div = ecmd.base.speed / 1000;
638 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
639
640 if (div)
641 mbits /= div;
642
643 if (div)
644 return mbits + 1;
645 return mbits;
646}
647
648static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
649 union tpacket_req_u *req_u)
650{
651 p1->feature_req_word = req_u->req3.tp_feature_req_word;
652}
653
654static void init_prb_bdqc(struct packet_sock *po,
655 struct packet_ring_buffer *rb,
656 struct pgv *pg_vec,
657 union tpacket_req_u *req_u)
658{
659 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
660 struct tpacket_block_desc *pbd;
661
662 memset(p1, 0x0, sizeof(*p1));
663
664 p1->knxt_seq_num = 1;
665 p1->pkbdq = pg_vec;
666 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
667 p1->pkblk_start = pg_vec[0].buffer;
668 p1->kblk_size = req_u->req3.tp_block_size;
669 p1->knum_blocks = req_u->req3.tp_block_nr;
670 p1->hdrlen = po->tp_hdrlen;
671 p1->version = po->tp_version;
672 p1->last_kactive_blk_num = 0;
673 po->stats.stats3.tp_freeze_q_cnt = 0;
674 if (req_u->req3.tp_retire_blk_tov)
675 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
676 else
677 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
678 req_u->req3.tp_block_size);
679 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
680 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
681 rwlock_init(&p1->blk_fill_in_prog_lock);
682
683 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
684 prb_init_ft_ops(p1, req_u);
685 prb_setup_retire_blk_timer(po);
686 prb_open_block(p1, pbd);
687}
688
689/* Do NOT update the last_blk_num first.
690 * Assumes sk_buff_head lock is held.
691 */
692static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
693{
694 mod_timer(&pkc->retire_blk_timer,
695 jiffies + pkc->tov_in_jiffies);
696 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
697}
698
699/*
700 * Timer logic:
701 * 1) We refresh the timer only when we open a block.
702 * By doing this we don't waste cycles refreshing the timer
703 * on packet-by-packet basis.
704 *
705 * With a 1MB block-size, on a 1Gbps line, it will take
706 * i) ~8 ms to fill a block + ii) memcpy etc.
707 * In this cut we are not accounting for the memcpy time.
708 *
709 * So, if the user sets the 'tmo' to 10ms then the timer
710 * will never fire while the block is still getting filled
711 * (which is what we want). However, the user could choose
712 * to close a block early and that's fine.
713 *
714 * But when the timer does fire, we check whether or not to refresh it.
715 * Since the tmo granularity is in msecs, it is not too expensive
716 * to refresh the timer, lets say every '8' msecs.
717 * Either the user can set the 'tmo' or we can derive it based on
718 * a) line-speed and b) block-size.
719 * prb_calc_retire_blk_tmo() calculates the tmo.
720 *
721 */
722static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
723{
724 struct packet_sock *po =
725 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
726 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
727 unsigned int frozen;
728 struct tpacket_block_desc *pbd;
729
730 spin_lock(&po->sk.sk_receive_queue.lock);
731
732 frozen = prb_queue_frozen(pkc);
733 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
734
735 if (unlikely(pkc->delete_blk_timer))
736 goto out;
737
738 /* We only need to plug the race when the block is partially filled.
739 * tpacket_rcv:
740 * lock(); increment BLOCK_NUM_PKTS; unlock()
741 * copy_bits() is in progress ...
742 * timer fires on other cpu:
743 * we can't retire the current block because copy_bits
744 * is in progress.
745 *
746 */
747 if (BLOCK_NUM_PKTS(pbd)) {
748 /* Waiting for skb_copy_bits to finish... */
749 write_lock(&pkc->blk_fill_in_prog_lock);
750 write_unlock(&pkc->blk_fill_in_prog_lock);
751 }
752
753 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
754 if (!frozen) {
755 if (!BLOCK_NUM_PKTS(pbd)) {
756 /* An empty block. Just refresh the timer. */
757 goto refresh_timer;
758 }
759 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
760 if (!prb_dispatch_next_block(pkc, po))
761 goto refresh_timer;
762 else
763 goto out;
764 } else {
765 /* Case 1. Queue was frozen because user-space was
766 * lagging behind.
767 */
768 if (prb_curr_blk_in_use(pbd)) {
769 /*
770 * Ok, user-space is still behind.
771 * So just refresh the timer.
772 */
773 goto refresh_timer;
774 } else {
775 /* Case 2. queue was frozen,user-space caught up,
776 * now the link went idle && the timer fired.
777 * We don't have a block to close.So we open this
778 * block and restart the timer.
779 * opening a block thaws the queue,restarts timer
780 * Thawing/timer-refresh is a side effect.
781 */
782 prb_open_block(pkc, pbd);
783 goto out;
784 }
785 }
786 }
787
788refresh_timer:
789 _prb_refresh_rx_retire_blk_timer(pkc);
790
791out:
792 spin_unlock(&po->sk.sk_receive_queue.lock);
793}
794
795static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
796 struct tpacket_block_desc *pbd1, __u32 status)
797{
798 /* Flush everything minus the block header */
799
800#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
801 u8 *start, *end;
802
803 start = (u8 *)pbd1;
804
805 /* Skip the block header(we know header WILL fit in 4K) */
806 start += PAGE_SIZE;
807
808 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
809 for (; start < end; start += PAGE_SIZE)
810 flush_dcache_page(pgv_to_page(start));
811
812 smp_wmb();
813#endif
814
815 /* Now update the block status. */
816
817 BLOCK_STATUS(pbd1) = status;
818
819 /* Flush the block header */
820
821#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
822 start = (u8 *)pbd1;
823 flush_dcache_page(pgv_to_page(start));
824
825 smp_wmb();
826#endif
827}
828
829/*
830 * Side effect:
831 *
832 * 1) flush the block
833 * 2) Increment active_blk_num
834 *
835 * Note:We DONT refresh the timer on purpose.
836 * Because almost always the next block will be opened.
837 */
838static void prb_close_block(struct tpacket_kbdq_core *pkc1,
839 struct tpacket_block_desc *pbd1,
840 struct packet_sock *po, unsigned int stat)
841{
842 __u32 status = TP_STATUS_USER | stat;
843
844 struct tpacket3_hdr *last_pkt;
845 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
846 struct sock *sk = &po->sk;
847
848 if (atomic_read(&po->tp_drops))
849 status |= TP_STATUS_LOSING;
850
851 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
852 last_pkt->tp_next_offset = 0;
853
854 /* Get the ts of the last pkt */
855 if (BLOCK_NUM_PKTS(pbd1)) {
856 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
857 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
858 } else {
859 /* Ok, we tmo'd - so get the current time.
860 *
861 * It shouldn't really happen as we don't close empty
862 * blocks. See prb_retire_rx_blk_timer_expired().
863 */
864 struct timespec64 ts;
865 ktime_get_real_ts64(&ts);
866 h1->ts_last_pkt.ts_sec = ts.tv_sec;
867 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
868 }
869
870 smp_wmb();
871
872 /* Flush the block */
873 prb_flush_block(pkc1, pbd1, status);
874
875 sk->sk_data_ready(sk);
876
877 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
878}
879
880static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
881{
882 pkc->reset_pending_on_curr_blk = 0;
883}
884
885/*
886 * Side effect of opening a block:
887 *
888 * 1) prb_queue is thawed.
889 * 2) retire_blk_timer is refreshed.
890 *
891 */
892static void prb_open_block(struct tpacket_kbdq_core *pkc1,
893 struct tpacket_block_desc *pbd1)
894{
895 struct timespec64 ts;
896 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
897
898 smp_rmb();
899
900 /* We could have just memset this but we will lose the
901 * flexibility of making the priv area sticky
902 */
903
904 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
905 BLOCK_NUM_PKTS(pbd1) = 0;
906 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
907
908 ktime_get_real_ts64(&ts);
909
910 h1->ts_first_pkt.ts_sec = ts.tv_sec;
911 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
912
913 pkc1->pkblk_start = (char *)pbd1;
914 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
915
916 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
917 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
918
919 pbd1->version = pkc1->version;
920 pkc1->prev = pkc1->nxt_offset;
921 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
922
923 prb_thaw_queue(pkc1);
924 _prb_refresh_rx_retire_blk_timer(pkc1);
925
926 smp_wmb();
927}
928
929/*
930 * Queue freeze logic:
931 * 1) Assume tp_block_nr = 8 blocks.
932 * 2) At time 't0', user opens Rx ring.
933 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
934 * 4) user-space is either sleeping or processing block '0'.
935 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
936 * it will close block-7,loop around and try to fill block '0'.
937 * call-flow:
938 * __packet_lookup_frame_in_block
939 * prb_retire_current_block()
940 * prb_dispatch_next_block()
941 * |->(BLOCK_STATUS == USER) evaluates to true
942 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
943 * 6) Now there are two cases:
944 * 6.1) Link goes idle right after the queue is frozen.
945 * But remember, the last open_block() refreshed the timer.
946 * When this timer expires,it will refresh itself so that we can
947 * re-open block-0 in near future.
948 * 6.2) Link is busy and keeps on receiving packets. This is a simple
949 * case and __packet_lookup_frame_in_block will check if block-0
950 * is free and can now be re-used.
951 */
952static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
953 struct packet_sock *po)
954{
955 pkc->reset_pending_on_curr_blk = 1;
956 po->stats.stats3.tp_freeze_q_cnt++;
957}
958
959#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
960
961/*
962 * If the next block is free then we will dispatch it
963 * and return a good offset.
964 * Else, we will freeze the queue.
965 * So, caller must check the return value.
966 */
967static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
968 struct packet_sock *po)
969{
970 struct tpacket_block_desc *pbd;
971
972 smp_rmb();
973
974 /* 1. Get current block num */
975 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
976
977 /* 2. If this block is currently in_use then freeze the queue */
978 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
979 prb_freeze_queue(pkc, po);
980 return NULL;
981 }
982
983 /*
984 * 3.
985 * open this block and return the offset where the first packet
986 * needs to get stored.
987 */
988 prb_open_block(pkc, pbd);
989 return (void *)pkc->nxt_offset;
990}
991
992static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
993 struct packet_sock *po, unsigned int status)
994{
995 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
996
997 /* retire/close the current block */
998 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
999 /*
1000 * Plug the case where copy_bits() is in progress on
1001 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
1002 * have space to copy the pkt in the current block and
1003 * called prb_retire_current_block()
1004 *
1005 * We don't need to worry about the TMO case because
1006 * the timer-handler already handled this case.
1007 */
1008 if (!(status & TP_STATUS_BLK_TMO)) {
1009 /* Waiting for skb_copy_bits to finish... */
1010 write_lock(&pkc->blk_fill_in_prog_lock);
1011 write_unlock(&pkc->blk_fill_in_prog_lock);
1012 }
1013 prb_close_block(pkc, pbd, po, status);
1014 return;
1015 }
1016}
1017
1018static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
1019{
1020 return TP_STATUS_USER & BLOCK_STATUS(pbd);
1021}
1022
1023static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
1024{
1025 return pkc->reset_pending_on_curr_blk;
1026}
1027
1028static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
1029 __releases(&pkc->blk_fill_in_prog_lock)
1030{
1031 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1032
1033 read_unlock(&pkc->blk_fill_in_prog_lock);
1034}
1035
1036static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
1037 struct tpacket3_hdr *ppd)
1038{
1039 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
1040}
1041
1042static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1043 struct tpacket3_hdr *ppd)
1044{
1045 ppd->hv1.tp_rxhash = 0;
1046}
1047
1048static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1049 struct tpacket3_hdr *ppd)
1050{
1051 struct packet_sock *po = container_of(pkc, struct packet_sock, rx_ring.prb_bdqc);
1052
1053 if (skb_vlan_tag_present(pkc->skb)) {
1054 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1055 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1056 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1057 } else if (unlikely(po->sk.sk_type == SOCK_DGRAM && eth_type_vlan(pkc->skb->protocol))) {
1058 ppd->hv1.tp_vlan_tci = vlan_get_tci(pkc->skb, pkc->skb->dev);
1059 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->protocol);
1060 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1061 } else {
1062 ppd->hv1.tp_vlan_tci = 0;
1063 ppd->hv1.tp_vlan_tpid = 0;
1064 ppd->tp_status = TP_STATUS_AVAILABLE;
1065 }
1066}
1067
1068static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1069 struct tpacket3_hdr *ppd)
1070{
1071 ppd->hv1.tp_padding = 0;
1072 prb_fill_vlan_info(pkc, ppd);
1073
1074 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1075 prb_fill_rxhash(pkc, ppd);
1076 else
1077 prb_clear_rxhash(pkc, ppd);
1078}
1079
1080static void prb_fill_curr_block(char *curr,
1081 struct tpacket_kbdq_core *pkc,
1082 struct tpacket_block_desc *pbd,
1083 unsigned int len)
1084 __acquires(&pkc->blk_fill_in_prog_lock)
1085{
1086 struct tpacket3_hdr *ppd;
1087
1088 ppd = (struct tpacket3_hdr *)curr;
1089 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1090 pkc->prev = curr;
1091 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1092 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1093 BLOCK_NUM_PKTS(pbd) += 1;
1094 read_lock(&pkc->blk_fill_in_prog_lock);
1095 prb_run_all_ft_ops(pkc, ppd);
1096}
1097
1098/* Assumes caller has the sk->rx_queue.lock */
1099static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1100 struct sk_buff *skb,
1101 unsigned int len
1102 )
1103{
1104 struct tpacket_kbdq_core *pkc;
1105 struct tpacket_block_desc *pbd;
1106 char *curr, *end;
1107
1108 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1109 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1110
1111 /* Queue is frozen when user space is lagging behind */
1112 if (prb_queue_frozen(pkc)) {
1113 /*
1114 * Check if that last block which caused the queue to freeze,
1115 * is still in_use by user-space.
1116 */
1117 if (prb_curr_blk_in_use(pbd)) {
1118 /* Can't record this packet */
1119 return NULL;
1120 } else {
1121 /*
1122 * Ok, the block was released by user-space.
1123 * Now let's open that block.
1124 * opening a block also thaws the queue.
1125 * Thawing is a side effect.
1126 */
1127 prb_open_block(pkc, pbd);
1128 }
1129 }
1130
1131 smp_mb();
1132 curr = pkc->nxt_offset;
1133 pkc->skb = skb;
1134 end = (char *)pbd + pkc->kblk_size;
1135
1136 /* first try the current block */
1137 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1138 prb_fill_curr_block(curr, pkc, pbd, len);
1139 return (void *)curr;
1140 }
1141
1142 /* Ok, close the current block */
1143 prb_retire_current_block(pkc, po, 0);
1144
1145 /* Now, try to dispatch the next block */
1146 curr = (char *)prb_dispatch_next_block(pkc, po);
1147 if (curr) {
1148 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1149 prb_fill_curr_block(curr, pkc, pbd, len);
1150 return (void *)curr;
1151 }
1152
1153 /*
1154 * No free blocks are available.user_space hasn't caught up yet.
1155 * Queue was just frozen and now this packet will get dropped.
1156 */
1157 return NULL;
1158}
1159
1160static void *packet_current_rx_frame(struct packet_sock *po,
1161 struct sk_buff *skb,
1162 int status, unsigned int len)
1163{
1164 char *curr = NULL;
1165 switch (po->tp_version) {
1166 case TPACKET_V1:
1167 case TPACKET_V2:
1168 curr = packet_lookup_frame(po, &po->rx_ring,
1169 po->rx_ring.head, status);
1170 return curr;
1171 case TPACKET_V3:
1172 return __packet_lookup_frame_in_block(po, skb, len);
1173 default:
1174 WARN(1, "TPACKET version not supported\n");
1175 BUG();
1176 return NULL;
1177 }
1178}
1179
1180static void *prb_lookup_block(const struct packet_sock *po,
1181 const struct packet_ring_buffer *rb,
1182 unsigned int idx,
1183 int status)
1184{
1185 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1186 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1187
1188 if (status != BLOCK_STATUS(pbd))
1189 return NULL;
1190 return pbd;
1191}
1192
1193static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1194{
1195 unsigned int prev;
1196 if (rb->prb_bdqc.kactive_blk_num)
1197 prev = rb->prb_bdqc.kactive_blk_num-1;
1198 else
1199 prev = rb->prb_bdqc.knum_blocks-1;
1200 return prev;
1201}
1202
1203/* Assumes caller has held the rx_queue.lock */
1204static void *__prb_previous_block(struct packet_sock *po,
1205 struct packet_ring_buffer *rb,
1206 int status)
1207{
1208 unsigned int previous = prb_previous_blk_num(rb);
1209 return prb_lookup_block(po, rb, previous, status);
1210}
1211
1212static void *packet_previous_rx_frame(struct packet_sock *po,
1213 struct packet_ring_buffer *rb,
1214 int status)
1215{
1216 if (po->tp_version <= TPACKET_V2)
1217 return packet_previous_frame(po, rb, status);
1218
1219 return __prb_previous_block(po, rb, status);
1220}
1221
1222static void packet_increment_rx_head(struct packet_sock *po,
1223 struct packet_ring_buffer *rb)
1224{
1225 switch (po->tp_version) {
1226 case TPACKET_V1:
1227 case TPACKET_V2:
1228 return packet_increment_head(rb);
1229 case TPACKET_V3:
1230 default:
1231 WARN(1, "TPACKET version not supported.\n");
1232 BUG();
1233 return;
1234 }
1235}
1236
1237static void *packet_previous_frame(struct packet_sock *po,
1238 struct packet_ring_buffer *rb,
1239 int status)
1240{
1241 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1242 return packet_lookup_frame(po, rb, previous, status);
1243}
1244
1245static void packet_increment_head(struct packet_ring_buffer *buff)
1246{
1247 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1248}
1249
1250static void packet_inc_pending(struct packet_ring_buffer *rb)
1251{
1252 this_cpu_inc(*rb->pending_refcnt);
1253}
1254
1255static void packet_dec_pending(struct packet_ring_buffer *rb)
1256{
1257 this_cpu_dec(*rb->pending_refcnt);
1258}
1259
1260static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1261{
1262 unsigned int refcnt = 0;
1263 int cpu;
1264
1265 /* We don't use pending refcount in rx_ring. */
1266 if (rb->pending_refcnt == NULL)
1267 return 0;
1268
1269 for_each_possible_cpu(cpu)
1270 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1271
1272 return refcnt;
1273}
1274
1275static int packet_alloc_pending(struct packet_sock *po)
1276{
1277 po->rx_ring.pending_refcnt = NULL;
1278
1279 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1280 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1281 return -ENOBUFS;
1282
1283 return 0;
1284}
1285
1286static void packet_free_pending(struct packet_sock *po)
1287{
1288 free_percpu(po->tx_ring.pending_refcnt);
1289}
1290
1291#define ROOM_POW_OFF 2
1292#define ROOM_NONE 0x0
1293#define ROOM_LOW 0x1
1294#define ROOM_NORMAL 0x2
1295
1296static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1297{
1298 int idx, len;
1299
1300 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1301 idx = READ_ONCE(po->rx_ring.head);
1302 if (pow_off)
1303 idx += len >> pow_off;
1304 if (idx >= len)
1305 idx -= len;
1306 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1307}
1308
1309static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1310{
1311 int idx, len;
1312
1313 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1314 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1315 if (pow_off)
1316 idx += len >> pow_off;
1317 if (idx >= len)
1318 idx -= len;
1319 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1320}
1321
1322static int __packet_rcv_has_room(const struct packet_sock *po,
1323 const struct sk_buff *skb)
1324{
1325 const struct sock *sk = &po->sk;
1326 int ret = ROOM_NONE;
1327
1328 if (po->prot_hook.func != tpacket_rcv) {
1329 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1330 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1331 - (skb ? skb->truesize : 0);
1332
1333 if (avail > (rcvbuf >> ROOM_POW_OFF))
1334 return ROOM_NORMAL;
1335 else if (avail > 0)
1336 return ROOM_LOW;
1337 else
1338 return ROOM_NONE;
1339 }
1340
1341 if (po->tp_version == TPACKET_V3) {
1342 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1343 ret = ROOM_NORMAL;
1344 else if (__tpacket_v3_has_room(po, 0))
1345 ret = ROOM_LOW;
1346 } else {
1347 if (__tpacket_has_room(po, ROOM_POW_OFF))
1348 ret = ROOM_NORMAL;
1349 else if (__tpacket_has_room(po, 0))
1350 ret = ROOM_LOW;
1351 }
1352
1353 return ret;
1354}
1355
1356static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1357{
1358 bool pressure;
1359 int ret;
1360
1361 ret = __packet_rcv_has_room(po, skb);
1362 pressure = ret != ROOM_NORMAL;
1363
1364 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure)
1365 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure);
1366
1367 return ret;
1368}
1369
1370static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1371{
1372 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) &&
1373 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1374 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false);
1375}
1376
1377static void packet_sock_destruct(struct sock *sk)
1378{
1379 skb_queue_purge(&sk->sk_error_queue);
1380
1381 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1382 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1383
1384 if (!sock_flag(sk, SOCK_DEAD)) {
1385 pr_err("Attempt to release alive packet socket: %p\n", sk);
1386 return;
1387 }
1388}
1389
1390static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1391{
1392 u32 *history = po->rollover->history;
1393 u32 victim, rxhash;
1394 int i, count = 0;
1395
1396 rxhash = skb_get_hash(skb);
1397 for (i = 0; i < ROLLOVER_HLEN; i++)
1398 if (READ_ONCE(history[i]) == rxhash)
1399 count++;
1400
1401 victim = get_random_u32_below(ROLLOVER_HLEN);
1402
1403 /* Avoid dirtying the cache line if possible */
1404 if (READ_ONCE(history[victim]) != rxhash)
1405 WRITE_ONCE(history[victim], rxhash);
1406
1407 return count > (ROLLOVER_HLEN >> 1);
1408}
1409
1410static unsigned int fanout_demux_hash(struct packet_fanout *f,
1411 struct sk_buff *skb,
1412 unsigned int num)
1413{
1414 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1415}
1416
1417static unsigned int fanout_demux_lb(struct packet_fanout *f,
1418 struct sk_buff *skb,
1419 unsigned int num)
1420{
1421 unsigned int val = atomic_inc_return(&f->rr_cur);
1422
1423 return val % num;
1424}
1425
1426static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1427 struct sk_buff *skb,
1428 unsigned int num)
1429{
1430 return smp_processor_id() % num;
1431}
1432
1433static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1434 struct sk_buff *skb,
1435 unsigned int num)
1436{
1437 return get_random_u32_below(num);
1438}
1439
1440static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1441 struct sk_buff *skb,
1442 unsigned int idx, bool try_self,
1443 unsigned int num)
1444{
1445 struct packet_sock *po, *po_next, *po_skip = NULL;
1446 unsigned int i, j, room = ROOM_NONE;
1447
1448 po = pkt_sk(rcu_dereference(f->arr[idx]));
1449
1450 if (try_self) {
1451 room = packet_rcv_has_room(po, skb);
1452 if (room == ROOM_NORMAL ||
1453 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1454 return idx;
1455 po_skip = po;
1456 }
1457
1458 i = j = min_t(int, po->rollover->sock, num - 1);
1459 do {
1460 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1461 if (po_next != po_skip &&
1462 !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1463 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1464 if (i != j)
1465 po->rollover->sock = i;
1466 atomic_long_inc(&po->rollover->num);
1467 if (room == ROOM_LOW)
1468 atomic_long_inc(&po->rollover->num_huge);
1469 return i;
1470 }
1471
1472 if (++i == num)
1473 i = 0;
1474 } while (i != j);
1475
1476 atomic_long_inc(&po->rollover->num_failed);
1477 return idx;
1478}
1479
1480static unsigned int fanout_demux_qm(struct packet_fanout *f,
1481 struct sk_buff *skb,
1482 unsigned int num)
1483{
1484 return skb_get_queue_mapping(skb) % num;
1485}
1486
1487static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1488 struct sk_buff *skb,
1489 unsigned int num)
1490{
1491 struct bpf_prog *prog;
1492 unsigned int ret = 0;
1493
1494 rcu_read_lock();
1495 prog = rcu_dereference(f->bpf_prog);
1496 if (prog)
1497 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1498 rcu_read_unlock();
1499
1500 return ret;
1501}
1502
1503static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1504{
1505 return f->flags & (flag >> 8);
1506}
1507
1508static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1509 struct packet_type *pt, struct net_device *orig_dev)
1510{
1511 struct packet_fanout *f = pt->af_packet_priv;
1512 unsigned int num = READ_ONCE(f->num_members);
1513 struct net *net = read_pnet(&f->net);
1514 struct packet_sock *po;
1515 unsigned int idx;
1516
1517 if (!net_eq(dev_net(dev), net) || !num) {
1518 kfree_skb(skb);
1519 return 0;
1520 }
1521
1522 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1523 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1524 if (!skb)
1525 return 0;
1526 }
1527 switch (f->type) {
1528 case PACKET_FANOUT_HASH:
1529 default:
1530 idx = fanout_demux_hash(f, skb, num);
1531 break;
1532 case PACKET_FANOUT_LB:
1533 idx = fanout_demux_lb(f, skb, num);
1534 break;
1535 case PACKET_FANOUT_CPU:
1536 idx = fanout_demux_cpu(f, skb, num);
1537 break;
1538 case PACKET_FANOUT_RND:
1539 idx = fanout_demux_rnd(f, skb, num);
1540 break;
1541 case PACKET_FANOUT_QM:
1542 idx = fanout_demux_qm(f, skb, num);
1543 break;
1544 case PACKET_FANOUT_ROLLOVER:
1545 idx = fanout_demux_rollover(f, skb, 0, false, num);
1546 break;
1547 case PACKET_FANOUT_CBPF:
1548 case PACKET_FANOUT_EBPF:
1549 idx = fanout_demux_bpf(f, skb, num);
1550 break;
1551 }
1552
1553 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1554 idx = fanout_demux_rollover(f, skb, idx, true, num);
1555
1556 po = pkt_sk(rcu_dereference(f->arr[idx]));
1557 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1558}
1559
1560DEFINE_MUTEX(fanout_mutex);
1561EXPORT_SYMBOL_GPL(fanout_mutex);
1562static LIST_HEAD(fanout_list);
1563static u16 fanout_next_id;
1564
1565static void __fanout_link(struct sock *sk, struct packet_sock *po)
1566{
1567 struct packet_fanout *f = po->fanout;
1568
1569 spin_lock(&f->lock);
1570 rcu_assign_pointer(f->arr[f->num_members], sk);
1571 smp_wmb();
1572 f->num_members++;
1573 if (f->num_members == 1)
1574 dev_add_pack(&f->prot_hook);
1575 spin_unlock(&f->lock);
1576}
1577
1578static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1579{
1580 struct packet_fanout *f = po->fanout;
1581 int i;
1582
1583 spin_lock(&f->lock);
1584 for (i = 0; i < f->num_members; i++) {
1585 if (rcu_dereference_protected(f->arr[i],
1586 lockdep_is_held(&f->lock)) == sk)
1587 break;
1588 }
1589 BUG_ON(i >= f->num_members);
1590 rcu_assign_pointer(f->arr[i],
1591 rcu_dereference_protected(f->arr[f->num_members - 1],
1592 lockdep_is_held(&f->lock)));
1593 f->num_members--;
1594 if (f->num_members == 0)
1595 __dev_remove_pack(&f->prot_hook);
1596 spin_unlock(&f->lock);
1597}
1598
1599static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1600{
1601 if (sk->sk_family != PF_PACKET)
1602 return false;
1603
1604 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1605}
1606
1607static void fanout_init_data(struct packet_fanout *f)
1608{
1609 switch (f->type) {
1610 case PACKET_FANOUT_LB:
1611 atomic_set(&f->rr_cur, 0);
1612 break;
1613 case PACKET_FANOUT_CBPF:
1614 case PACKET_FANOUT_EBPF:
1615 RCU_INIT_POINTER(f->bpf_prog, NULL);
1616 break;
1617 }
1618}
1619
1620static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1621{
1622 struct bpf_prog *old;
1623
1624 spin_lock(&f->lock);
1625 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1626 rcu_assign_pointer(f->bpf_prog, new);
1627 spin_unlock(&f->lock);
1628
1629 if (old) {
1630 synchronize_net();
1631 bpf_prog_destroy(old);
1632 }
1633}
1634
1635static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1636 unsigned int len)
1637{
1638 struct bpf_prog *new;
1639 struct sock_fprog fprog;
1640 int ret;
1641
1642 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1643 return -EPERM;
1644
1645 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1646 if (ret)
1647 return ret;
1648
1649 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1650 if (ret)
1651 return ret;
1652
1653 __fanout_set_data_bpf(po->fanout, new);
1654 return 0;
1655}
1656
1657static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1658 unsigned int len)
1659{
1660 struct bpf_prog *new;
1661 u32 fd;
1662
1663 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1664 return -EPERM;
1665 if (len != sizeof(fd))
1666 return -EINVAL;
1667 if (copy_from_sockptr(&fd, data, len))
1668 return -EFAULT;
1669
1670 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1671 if (IS_ERR(new))
1672 return PTR_ERR(new);
1673
1674 __fanout_set_data_bpf(po->fanout, new);
1675 return 0;
1676}
1677
1678static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1679 unsigned int len)
1680{
1681 switch (po->fanout->type) {
1682 case PACKET_FANOUT_CBPF:
1683 return fanout_set_data_cbpf(po, data, len);
1684 case PACKET_FANOUT_EBPF:
1685 return fanout_set_data_ebpf(po, data, len);
1686 default:
1687 return -EINVAL;
1688 }
1689}
1690
1691static void fanout_release_data(struct packet_fanout *f)
1692{
1693 switch (f->type) {
1694 case PACKET_FANOUT_CBPF:
1695 case PACKET_FANOUT_EBPF:
1696 __fanout_set_data_bpf(f, NULL);
1697 }
1698}
1699
1700static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1701{
1702 struct packet_fanout *f;
1703
1704 list_for_each_entry(f, &fanout_list, list) {
1705 if (f->id == candidate_id &&
1706 read_pnet(&f->net) == sock_net(sk)) {
1707 return false;
1708 }
1709 }
1710 return true;
1711}
1712
1713static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1714{
1715 u16 id = fanout_next_id;
1716
1717 do {
1718 if (__fanout_id_is_free(sk, id)) {
1719 *new_id = id;
1720 fanout_next_id = id + 1;
1721 return true;
1722 }
1723
1724 id++;
1725 } while (id != fanout_next_id);
1726
1727 return false;
1728}
1729
1730static int fanout_add(struct sock *sk, struct fanout_args *args)
1731{
1732 struct packet_rollover *rollover = NULL;
1733 struct packet_sock *po = pkt_sk(sk);
1734 u16 type_flags = args->type_flags;
1735 struct packet_fanout *f, *match;
1736 u8 type = type_flags & 0xff;
1737 u8 flags = type_flags >> 8;
1738 u16 id = args->id;
1739 int err;
1740
1741 switch (type) {
1742 case PACKET_FANOUT_ROLLOVER:
1743 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1744 return -EINVAL;
1745 break;
1746 case PACKET_FANOUT_HASH:
1747 case PACKET_FANOUT_LB:
1748 case PACKET_FANOUT_CPU:
1749 case PACKET_FANOUT_RND:
1750 case PACKET_FANOUT_QM:
1751 case PACKET_FANOUT_CBPF:
1752 case PACKET_FANOUT_EBPF:
1753 break;
1754 default:
1755 return -EINVAL;
1756 }
1757
1758 mutex_lock(&fanout_mutex);
1759
1760 err = -EALREADY;
1761 if (po->fanout)
1762 goto out;
1763
1764 if (type == PACKET_FANOUT_ROLLOVER ||
1765 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1766 err = -ENOMEM;
1767 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1768 if (!rollover)
1769 goto out;
1770 atomic_long_set(&rollover->num, 0);
1771 atomic_long_set(&rollover->num_huge, 0);
1772 atomic_long_set(&rollover->num_failed, 0);
1773 }
1774
1775 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1776 if (id != 0) {
1777 err = -EINVAL;
1778 goto out;
1779 }
1780 if (!fanout_find_new_id(sk, &id)) {
1781 err = -ENOMEM;
1782 goto out;
1783 }
1784 /* ephemeral flag for the first socket in the group: drop it */
1785 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1786 }
1787
1788 match = NULL;
1789 list_for_each_entry(f, &fanout_list, list) {
1790 if (f->id == id &&
1791 read_pnet(&f->net) == sock_net(sk)) {
1792 match = f;
1793 break;
1794 }
1795 }
1796 err = -EINVAL;
1797 if (match) {
1798 if (match->flags != flags)
1799 goto out;
1800 if (args->max_num_members &&
1801 args->max_num_members != match->max_num_members)
1802 goto out;
1803 } else {
1804 if (args->max_num_members > PACKET_FANOUT_MAX)
1805 goto out;
1806 if (!args->max_num_members)
1807 /* legacy PACKET_FANOUT_MAX */
1808 args->max_num_members = 256;
1809 err = -ENOMEM;
1810 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1811 GFP_KERNEL);
1812 if (!match)
1813 goto out;
1814 write_pnet(&match->net, sock_net(sk));
1815 match->id = id;
1816 match->type = type;
1817 match->flags = flags;
1818 INIT_LIST_HEAD(&match->list);
1819 spin_lock_init(&match->lock);
1820 refcount_set(&match->sk_ref, 0);
1821 fanout_init_data(match);
1822 match->prot_hook.type = po->prot_hook.type;
1823 match->prot_hook.dev = po->prot_hook.dev;
1824 match->prot_hook.func = packet_rcv_fanout;
1825 match->prot_hook.af_packet_priv = match;
1826 match->prot_hook.af_packet_net = read_pnet(&match->net);
1827 match->prot_hook.id_match = match_fanout_group;
1828 match->max_num_members = args->max_num_members;
1829 match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1830 list_add(&match->list, &fanout_list);
1831 }
1832 err = -EINVAL;
1833
1834 spin_lock(&po->bind_lock);
1835 if (po->num &&
1836 match->type == type &&
1837 match->prot_hook.type == po->prot_hook.type &&
1838 match->prot_hook.dev == po->prot_hook.dev) {
1839 err = -ENOSPC;
1840 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1841 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1842 WRITE_ONCE(po->fanout, match);
1843
1844 po->rollover = rollover;
1845 rollover = NULL;
1846 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1847 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
1848 __dev_remove_pack(&po->prot_hook);
1849 __fanout_link(sk, po);
1850 }
1851 err = 0;
1852 }
1853 }
1854 spin_unlock(&po->bind_lock);
1855
1856 if (err && !refcount_read(&match->sk_ref)) {
1857 list_del(&match->list);
1858 kvfree(match);
1859 }
1860
1861out:
1862 kfree(rollover);
1863 mutex_unlock(&fanout_mutex);
1864 return err;
1865}
1866
1867/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1868 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1869 * It is the responsibility of the caller to call fanout_release_data() and
1870 * free the returned packet_fanout (after synchronize_net())
1871 */
1872static struct packet_fanout *fanout_release(struct sock *sk)
1873{
1874 struct packet_sock *po = pkt_sk(sk);
1875 struct packet_fanout *f;
1876
1877 mutex_lock(&fanout_mutex);
1878 f = po->fanout;
1879 if (f) {
1880 po->fanout = NULL;
1881
1882 if (refcount_dec_and_test(&f->sk_ref))
1883 list_del(&f->list);
1884 else
1885 f = NULL;
1886 }
1887 mutex_unlock(&fanout_mutex);
1888
1889 return f;
1890}
1891
1892static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1893 struct sk_buff *skb)
1894{
1895 /* Earlier code assumed this would be a VLAN pkt, double-check
1896 * this now that we have the actual packet in hand. We can only
1897 * do this check on Ethernet devices.
1898 */
1899 if (unlikely(dev->type != ARPHRD_ETHER))
1900 return false;
1901
1902 skb_reset_mac_header(skb);
1903 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1904}
1905
1906static const struct proto_ops packet_ops;
1907
1908static const struct proto_ops packet_ops_spkt;
1909
1910static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1911 struct packet_type *pt, struct net_device *orig_dev)
1912{
1913 struct sock *sk;
1914 struct sockaddr_pkt *spkt;
1915
1916 /*
1917 * When we registered the protocol we saved the socket in the data
1918 * field for just this event.
1919 */
1920
1921 sk = pt->af_packet_priv;
1922
1923 /*
1924 * Yank back the headers [hope the device set this
1925 * right or kerboom...]
1926 *
1927 * Incoming packets have ll header pulled,
1928 * push it back.
1929 *
1930 * For outgoing ones skb->data == skb_mac_header(skb)
1931 * so that this procedure is noop.
1932 */
1933
1934 if (skb->pkt_type == PACKET_LOOPBACK)
1935 goto out;
1936
1937 if (!net_eq(dev_net(dev), sock_net(sk)))
1938 goto out;
1939
1940 skb = skb_share_check(skb, GFP_ATOMIC);
1941 if (skb == NULL)
1942 goto oom;
1943
1944 /* drop any routing info */
1945 skb_dst_drop(skb);
1946
1947 /* drop conntrack reference */
1948 nf_reset_ct(skb);
1949
1950 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1951
1952 skb_push(skb, skb->data - skb_mac_header(skb));
1953
1954 /*
1955 * The SOCK_PACKET socket receives _all_ frames.
1956 */
1957
1958 spkt->spkt_family = dev->type;
1959 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1960 spkt->spkt_protocol = skb->protocol;
1961
1962 /*
1963 * Charge the memory to the socket. This is done specifically
1964 * to prevent sockets using all the memory up.
1965 */
1966
1967 if (sock_queue_rcv_skb(sk, skb) == 0)
1968 return 0;
1969
1970out:
1971 kfree_skb(skb);
1972oom:
1973 return 0;
1974}
1975
1976static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1977{
1978 int depth;
1979
1980 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1981 sock->type == SOCK_RAW) {
1982 skb_reset_mac_header(skb);
1983 skb->protocol = dev_parse_header_protocol(skb);
1984 }
1985
1986 /* Move network header to the right position for VLAN tagged packets */
1987 if (likely(skb->dev->type == ARPHRD_ETHER) &&
1988 eth_type_vlan(skb->protocol) &&
1989 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1990 skb_set_network_header(skb, depth);
1991
1992 skb_probe_transport_header(skb);
1993}
1994
1995/*
1996 * Output a raw packet to a device layer. This bypasses all the other
1997 * protocol layers and you must therefore supply it with a complete frame
1998 */
1999
2000static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
2001 size_t len)
2002{
2003 struct sock *sk = sock->sk;
2004 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
2005 struct sk_buff *skb = NULL;
2006 struct net_device *dev;
2007 struct sockcm_cookie sockc;
2008 __be16 proto = 0;
2009 int err;
2010 int extra_len = 0;
2011
2012 /*
2013 * Get and verify the address.
2014 */
2015
2016 if (saddr) {
2017 if (msg->msg_namelen < sizeof(struct sockaddr))
2018 return -EINVAL;
2019 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
2020 proto = saddr->spkt_protocol;
2021 } else
2022 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
2023
2024 /*
2025 * Find the device first to size check it
2026 */
2027
2028 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
2029retry:
2030 rcu_read_lock();
2031 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
2032 err = -ENODEV;
2033 if (dev == NULL)
2034 goto out_unlock;
2035
2036 err = -ENETDOWN;
2037 if (!(dev->flags & IFF_UP))
2038 goto out_unlock;
2039
2040 /*
2041 * You may not queue a frame bigger than the mtu. This is the lowest level
2042 * raw protocol and you must do your own fragmentation at this level.
2043 */
2044
2045 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2046 if (!netif_supports_nofcs(dev)) {
2047 err = -EPROTONOSUPPORT;
2048 goto out_unlock;
2049 }
2050 extra_len = 4; /* We're doing our own CRC */
2051 }
2052
2053 err = -EMSGSIZE;
2054 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2055 goto out_unlock;
2056
2057 if (!skb) {
2058 size_t reserved = LL_RESERVED_SPACE(dev);
2059 int tlen = dev->needed_tailroom;
2060 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2061
2062 rcu_read_unlock();
2063 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2064 if (skb == NULL)
2065 return -ENOBUFS;
2066 /* FIXME: Save some space for broken drivers that write a hard
2067 * header at transmission time by themselves. PPP is the notable
2068 * one here. This should really be fixed at the driver level.
2069 */
2070 skb_reserve(skb, reserved);
2071 skb_reset_network_header(skb);
2072
2073 /* Try to align data part correctly */
2074 if (hhlen) {
2075 skb->data -= hhlen;
2076 skb->tail -= hhlen;
2077 if (len < hhlen)
2078 skb_reset_network_header(skb);
2079 }
2080 err = memcpy_from_msg(skb_put(skb, len), msg, len);
2081 if (err)
2082 goto out_free;
2083 goto retry;
2084 }
2085
2086 if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2087 err = -EINVAL;
2088 goto out_unlock;
2089 }
2090 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2091 !packet_extra_vlan_len_allowed(dev, skb)) {
2092 err = -EMSGSIZE;
2093 goto out_unlock;
2094 }
2095
2096 sockcm_init(&sockc, sk);
2097 if (msg->msg_controllen) {
2098 err = sock_cmsg_send(sk, msg, &sockc);
2099 if (unlikely(err))
2100 goto out_unlock;
2101 }
2102
2103 skb->protocol = proto;
2104 skb->dev = dev;
2105 skb->priority = READ_ONCE(sk->sk_priority);
2106 skb->mark = READ_ONCE(sk->sk_mark);
2107 skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
2108 skb_setup_tx_timestamp(skb, &sockc);
2109
2110 if (unlikely(extra_len == 4))
2111 skb->no_fcs = 1;
2112
2113 packet_parse_headers(skb, sock);
2114
2115 dev_queue_xmit(skb);
2116 rcu_read_unlock();
2117 return len;
2118
2119out_unlock:
2120 rcu_read_unlock();
2121out_free:
2122 kfree_skb(skb);
2123 return err;
2124}
2125
2126static unsigned int run_filter(struct sk_buff *skb,
2127 const struct sock *sk,
2128 unsigned int res)
2129{
2130 struct sk_filter *filter;
2131
2132 rcu_read_lock();
2133 filter = rcu_dereference(sk->sk_filter);
2134 if (filter != NULL)
2135 res = bpf_prog_run_clear_cb(filter->prog, skb);
2136 rcu_read_unlock();
2137
2138 return res;
2139}
2140
2141static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2142 size_t *len, int vnet_hdr_sz)
2143{
2144 struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2145
2146 if (*len < vnet_hdr_sz)
2147 return -EINVAL;
2148 *len -= vnet_hdr_sz;
2149
2150 if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2151 return -EINVAL;
2152
2153 return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
2154}
2155
2156/*
2157 * This function makes lazy skb cloning in hope that most of packets
2158 * are discarded by BPF.
2159 *
2160 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2161 * and skb->cb are mangled. It works because (and until) packets
2162 * falling here are owned by current CPU. Output packets are cloned
2163 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2164 * sequentially, so that if we return skb to original state on exit,
2165 * we will not harm anyone.
2166 */
2167
2168static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2169 struct packet_type *pt, struct net_device *orig_dev)
2170{
2171 enum skb_drop_reason drop_reason = SKB_CONSUMED;
2172 struct sock *sk = NULL;
2173 struct sockaddr_ll *sll;
2174 struct packet_sock *po;
2175 u8 *skb_head = skb->data;
2176 int skb_len = skb->len;
2177 unsigned int snaplen, res;
2178
2179 if (skb->pkt_type == PACKET_LOOPBACK)
2180 goto drop;
2181
2182 sk = pt->af_packet_priv;
2183 po = pkt_sk(sk);
2184
2185 if (!net_eq(dev_net(dev), sock_net(sk)))
2186 goto drop;
2187
2188 skb->dev = dev;
2189
2190 if (dev_has_header(dev)) {
2191 /* The device has an explicit notion of ll header,
2192 * exported to higher levels.
2193 *
2194 * Otherwise, the device hides details of its frame
2195 * structure, so that corresponding packet head is
2196 * never delivered to user.
2197 */
2198 if (sk->sk_type != SOCK_DGRAM)
2199 skb_push(skb, skb->data - skb_mac_header(skb));
2200 else if (skb->pkt_type == PACKET_OUTGOING) {
2201 /* Special case: outgoing packets have ll header at head */
2202 skb_pull(skb, skb_network_offset(skb));
2203 }
2204 }
2205
2206 snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2207
2208 res = run_filter(skb, sk, snaplen);
2209 if (!res)
2210 goto drop_n_restore;
2211 if (snaplen > res)
2212 snaplen = res;
2213
2214 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2215 goto drop_n_acct;
2216
2217 if (skb_shared(skb)) {
2218 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2219 if (nskb == NULL)
2220 goto drop_n_acct;
2221
2222 if (skb_head != skb->data) {
2223 skb->data = skb_head;
2224 skb->len = skb_len;
2225 }
2226 consume_skb(skb);
2227 skb = nskb;
2228 }
2229
2230 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2231
2232 sll = &PACKET_SKB_CB(skb)->sa.ll;
2233 sll->sll_hatype = dev->type;
2234 sll->sll_pkttype = skb->pkt_type;
2235 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2236 sll->sll_ifindex = orig_dev->ifindex;
2237 else
2238 sll->sll_ifindex = dev->ifindex;
2239
2240 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2241
2242 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2243 * Use their space for storing the original skb length.
2244 */
2245 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2246
2247 if (pskb_trim(skb, snaplen))
2248 goto drop_n_acct;
2249
2250 skb_set_owner_r(skb, sk);
2251 skb->dev = NULL;
2252 skb_dst_drop(skb);
2253
2254 /* drop conntrack reference */
2255 nf_reset_ct(skb);
2256
2257 spin_lock(&sk->sk_receive_queue.lock);
2258 po->stats.stats1.tp_packets++;
2259 sock_skb_set_dropcount(sk, skb);
2260 skb_clear_delivery_time(skb);
2261 __skb_queue_tail(&sk->sk_receive_queue, skb);
2262 spin_unlock(&sk->sk_receive_queue.lock);
2263 sk->sk_data_ready(sk);
2264 return 0;
2265
2266drop_n_acct:
2267 atomic_inc(&po->tp_drops);
2268 atomic_inc(&sk->sk_drops);
2269 drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2270
2271drop_n_restore:
2272 if (skb_head != skb->data && skb_shared(skb)) {
2273 skb->data = skb_head;
2274 skb->len = skb_len;
2275 }
2276drop:
2277 sk_skb_reason_drop(sk, skb, drop_reason);
2278 return 0;
2279}
2280
2281static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2282 struct packet_type *pt, struct net_device *orig_dev)
2283{
2284 enum skb_drop_reason drop_reason = SKB_CONSUMED;
2285 struct sock *sk = NULL;
2286 struct packet_sock *po;
2287 struct sockaddr_ll *sll;
2288 union tpacket_uhdr h;
2289 u8 *skb_head = skb->data;
2290 int skb_len = skb->len;
2291 unsigned int snaplen, res;
2292 unsigned long status = TP_STATUS_USER;
2293 unsigned short macoff, hdrlen;
2294 unsigned int netoff;
2295 struct sk_buff *copy_skb = NULL;
2296 struct timespec64 ts;
2297 __u32 ts_status;
2298 unsigned int slot_id = 0;
2299 int vnet_hdr_sz = 0;
2300
2301 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2302 * We may add members to them until current aligned size without forcing
2303 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2304 */
2305 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2306 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2307
2308 if (skb->pkt_type == PACKET_LOOPBACK)
2309 goto drop;
2310
2311 sk = pt->af_packet_priv;
2312 po = pkt_sk(sk);
2313
2314 if (!net_eq(dev_net(dev), sock_net(sk)))
2315 goto drop;
2316
2317 if (dev_has_header(dev)) {
2318 if (sk->sk_type != SOCK_DGRAM)
2319 skb_push(skb, skb->data - skb_mac_header(skb));
2320 else if (skb->pkt_type == PACKET_OUTGOING) {
2321 /* Special case: outgoing packets have ll header at head */
2322 skb_pull(skb, skb_network_offset(skb));
2323 }
2324 }
2325
2326 snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2327
2328 res = run_filter(skb, sk, snaplen);
2329 if (!res)
2330 goto drop_n_restore;
2331
2332 /* If we are flooded, just give up */
2333 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2334 atomic_inc(&po->tp_drops);
2335 goto drop_n_restore;
2336 }
2337
2338 if (skb->ip_summed == CHECKSUM_PARTIAL)
2339 status |= TP_STATUS_CSUMNOTREADY;
2340 else if (skb->pkt_type != PACKET_OUTGOING &&
2341 skb_csum_unnecessary(skb))
2342 status |= TP_STATUS_CSUM_VALID;
2343 if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2344 status |= TP_STATUS_GSO_TCP;
2345
2346 if (snaplen > res)
2347 snaplen = res;
2348
2349 if (sk->sk_type == SOCK_DGRAM) {
2350 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2351 po->tp_reserve;
2352 } else {
2353 unsigned int maclen = skb_network_offset(skb);
2354 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2355 (maclen < 16 ? 16 : maclen)) +
2356 po->tp_reserve;
2357 vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2358 if (vnet_hdr_sz)
2359 netoff += vnet_hdr_sz;
2360 macoff = netoff - maclen;
2361 }
2362 if (netoff > USHRT_MAX) {
2363 atomic_inc(&po->tp_drops);
2364 goto drop_n_restore;
2365 }
2366 if (po->tp_version <= TPACKET_V2) {
2367 if (macoff + snaplen > po->rx_ring.frame_size) {
2368 if (READ_ONCE(po->copy_thresh) &&
2369 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2370 if (skb_shared(skb)) {
2371 copy_skb = skb_clone(skb, GFP_ATOMIC);
2372 } else {
2373 copy_skb = skb_get(skb);
2374 skb_head = skb->data;
2375 }
2376 if (copy_skb) {
2377 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2378 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2379 skb_set_owner_r(copy_skb, sk);
2380 }
2381 }
2382 snaplen = po->rx_ring.frame_size - macoff;
2383 if ((int)snaplen < 0) {
2384 snaplen = 0;
2385 vnet_hdr_sz = 0;
2386 }
2387 }
2388 } else if (unlikely(macoff + snaplen >
2389 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2390 u32 nval;
2391
2392 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2393 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2394 snaplen, nval, macoff);
2395 snaplen = nval;
2396 if (unlikely((int)snaplen < 0)) {
2397 snaplen = 0;
2398 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2399 vnet_hdr_sz = 0;
2400 }
2401 }
2402 spin_lock(&sk->sk_receive_queue.lock);
2403 h.raw = packet_current_rx_frame(po, skb,
2404 TP_STATUS_KERNEL, (macoff+snaplen));
2405 if (!h.raw)
2406 goto drop_n_account;
2407
2408 if (po->tp_version <= TPACKET_V2) {
2409 slot_id = po->rx_ring.head;
2410 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2411 goto drop_n_account;
2412 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2413 }
2414
2415 if (vnet_hdr_sz &&
2416 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2417 sizeof(struct virtio_net_hdr),
2418 vio_le(), true, 0)) {
2419 if (po->tp_version == TPACKET_V3)
2420 prb_clear_blk_fill_status(&po->rx_ring);
2421 goto drop_n_account;
2422 }
2423
2424 if (po->tp_version <= TPACKET_V2) {
2425 packet_increment_rx_head(po, &po->rx_ring);
2426 /*
2427 * LOSING will be reported till you read the stats,
2428 * because it's COR - Clear On Read.
2429 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2430 * at packet level.
2431 */
2432 if (atomic_read(&po->tp_drops))
2433 status |= TP_STATUS_LOSING;
2434 }
2435
2436 po->stats.stats1.tp_packets++;
2437 if (copy_skb) {
2438 status |= TP_STATUS_COPY;
2439 skb_clear_delivery_time(copy_skb);
2440 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2441 }
2442 spin_unlock(&sk->sk_receive_queue.lock);
2443
2444 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2445
2446 /* Always timestamp; prefer an existing software timestamp taken
2447 * closer to the time of capture.
2448 */
2449 ts_status = tpacket_get_timestamp(skb, &ts,
2450 READ_ONCE(po->tp_tstamp) |
2451 SOF_TIMESTAMPING_SOFTWARE);
2452 if (!ts_status)
2453 ktime_get_real_ts64(&ts);
2454
2455 status |= ts_status;
2456
2457 switch (po->tp_version) {
2458 case TPACKET_V1:
2459 h.h1->tp_len = skb->len;
2460 h.h1->tp_snaplen = snaplen;
2461 h.h1->tp_mac = macoff;
2462 h.h1->tp_net = netoff;
2463 h.h1->tp_sec = ts.tv_sec;
2464 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2465 hdrlen = sizeof(*h.h1);
2466 break;
2467 case TPACKET_V2:
2468 h.h2->tp_len = skb->len;
2469 h.h2->tp_snaplen = snaplen;
2470 h.h2->tp_mac = macoff;
2471 h.h2->tp_net = netoff;
2472 h.h2->tp_sec = ts.tv_sec;
2473 h.h2->tp_nsec = ts.tv_nsec;
2474 if (skb_vlan_tag_present(skb)) {
2475 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2476 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2477 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2478 } else if (unlikely(sk->sk_type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
2479 h.h2->tp_vlan_tci = vlan_get_tci(skb, skb->dev);
2480 h.h2->tp_vlan_tpid = ntohs(skb->protocol);
2481 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2482 } else {
2483 h.h2->tp_vlan_tci = 0;
2484 h.h2->tp_vlan_tpid = 0;
2485 }
2486 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2487 hdrlen = sizeof(*h.h2);
2488 break;
2489 case TPACKET_V3:
2490 /* tp_nxt_offset,vlan are already populated above.
2491 * So DONT clear those fields here
2492 */
2493 h.h3->tp_status |= status;
2494 h.h3->tp_len = skb->len;
2495 h.h3->tp_snaplen = snaplen;
2496 h.h3->tp_mac = macoff;
2497 h.h3->tp_net = netoff;
2498 h.h3->tp_sec = ts.tv_sec;
2499 h.h3->tp_nsec = ts.tv_nsec;
2500 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2501 hdrlen = sizeof(*h.h3);
2502 break;
2503 default:
2504 BUG();
2505 }
2506
2507 sll = h.raw + TPACKET_ALIGN(hdrlen);
2508 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2509 sll->sll_family = AF_PACKET;
2510 sll->sll_hatype = dev->type;
2511 sll->sll_protocol = (sk->sk_type == SOCK_DGRAM) ?
2512 vlan_get_protocol_dgram(skb) : skb->protocol;
2513 sll->sll_pkttype = skb->pkt_type;
2514 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2515 sll->sll_ifindex = orig_dev->ifindex;
2516 else
2517 sll->sll_ifindex = dev->ifindex;
2518
2519 smp_mb();
2520
2521#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2522 if (po->tp_version <= TPACKET_V2) {
2523 u8 *start, *end;
2524
2525 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2526 macoff + snaplen);
2527
2528 for (start = h.raw; start < end; start += PAGE_SIZE)
2529 flush_dcache_page(pgv_to_page(start));
2530 }
2531 smp_wmb();
2532#endif
2533
2534 if (po->tp_version <= TPACKET_V2) {
2535 spin_lock(&sk->sk_receive_queue.lock);
2536 __packet_set_status(po, h.raw, status);
2537 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2538 spin_unlock(&sk->sk_receive_queue.lock);
2539 sk->sk_data_ready(sk);
2540 } else if (po->tp_version == TPACKET_V3) {
2541 prb_clear_blk_fill_status(&po->rx_ring);
2542 }
2543
2544drop_n_restore:
2545 if (skb_head != skb->data && skb_shared(skb)) {
2546 skb->data = skb_head;
2547 skb->len = skb_len;
2548 }
2549drop:
2550 sk_skb_reason_drop(sk, skb, drop_reason);
2551 return 0;
2552
2553drop_n_account:
2554 spin_unlock(&sk->sk_receive_queue.lock);
2555 atomic_inc(&po->tp_drops);
2556 drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2557
2558 sk->sk_data_ready(sk);
2559 sk_skb_reason_drop(sk, copy_skb, drop_reason);
2560 goto drop_n_restore;
2561}
2562
2563static void tpacket_destruct_skb(struct sk_buff *skb)
2564{
2565 struct packet_sock *po = pkt_sk(skb->sk);
2566
2567 if (likely(po->tx_ring.pg_vec)) {
2568 void *ph;
2569 __u32 ts;
2570
2571 ph = skb_zcopy_get_nouarg(skb);
2572 packet_dec_pending(&po->tx_ring);
2573
2574 ts = __packet_set_timestamp(po, ph, skb);
2575 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2576
2577 complete(&po->skb_completion);
2578 }
2579
2580 sock_wfree(skb);
2581}
2582
2583static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2584{
2585 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2586 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2587 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2588 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2589 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2590 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2591 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2592
2593 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2594 return -EINVAL;
2595
2596 return 0;
2597}
2598
2599static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2600 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz)
2601{
2602 int ret;
2603
2604 if (*len < vnet_hdr_sz)
2605 return -EINVAL;
2606 *len -= vnet_hdr_sz;
2607
2608 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2609 return -EFAULT;
2610
2611 ret = __packet_snd_vnet_parse(vnet_hdr, *len);
2612 if (ret)
2613 return ret;
2614
2615 /* move iter to point to the start of mac header */
2616 if (vnet_hdr_sz != sizeof(struct virtio_net_hdr))
2617 iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr));
2618
2619 return 0;
2620}
2621
2622static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2623 void *frame, struct net_device *dev, void *data, int tp_len,
2624 __be16 proto, unsigned char *addr, int hlen, int copylen,
2625 const struct sockcm_cookie *sockc)
2626{
2627 union tpacket_uhdr ph;
2628 int to_write, offset, len, nr_frags, len_max;
2629 struct socket *sock = po->sk.sk_socket;
2630 struct page *page;
2631 int err;
2632
2633 ph.raw = frame;
2634
2635 skb->protocol = proto;
2636 skb->dev = dev;
2637 skb->priority = READ_ONCE(po->sk.sk_priority);
2638 skb->mark = READ_ONCE(po->sk.sk_mark);
2639 skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, po->sk.sk_clockid);
2640 skb_setup_tx_timestamp(skb, sockc);
2641 skb_zcopy_set_nouarg(skb, ph.raw);
2642
2643 skb_reserve(skb, hlen);
2644 skb_reset_network_header(skb);
2645
2646 to_write = tp_len;
2647
2648 if (sock->type == SOCK_DGRAM) {
2649 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2650 NULL, tp_len);
2651 if (unlikely(err < 0))
2652 return -EINVAL;
2653 } else if (copylen) {
2654 int hdrlen = min_t(int, copylen, tp_len);
2655
2656 skb_push(skb, dev->hard_header_len);
2657 skb_put(skb, copylen - dev->hard_header_len);
2658 err = skb_store_bits(skb, 0, data, hdrlen);
2659 if (unlikely(err))
2660 return err;
2661 if (!dev_validate_header(dev, skb->data, hdrlen))
2662 return -EINVAL;
2663
2664 data += hdrlen;
2665 to_write -= hdrlen;
2666 }
2667
2668 offset = offset_in_page(data);
2669 len_max = PAGE_SIZE - offset;
2670 len = ((to_write > len_max) ? len_max : to_write);
2671
2672 skb->data_len = to_write;
2673 skb->len += to_write;
2674 skb->truesize += to_write;
2675 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2676
2677 while (likely(to_write)) {
2678 nr_frags = skb_shinfo(skb)->nr_frags;
2679
2680 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2681 pr_err("Packet exceed the number of skb frags(%u)\n",
2682 (unsigned int)MAX_SKB_FRAGS);
2683 return -EFAULT;
2684 }
2685
2686 page = pgv_to_page(data);
2687 data += len;
2688 flush_dcache_page(page);
2689 get_page(page);
2690 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2691 to_write -= len;
2692 offset = 0;
2693 len_max = PAGE_SIZE;
2694 len = ((to_write > len_max) ? len_max : to_write);
2695 }
2696
2697 packet_parse_headers(skb, sock);
2698
2699 return tp_len;
2700}
2701
2702static int tpacket_parse_header(struct packet_sock *po, void *frame,
2703 int size_max, void **data)
2704{
2705 union tpacket_uhdr ph;
2706 int tp_len, off;
2707
2708 ph.raw = frame;
2709
2710 switch (po->tp_version) {
2711 case TPACKET_V3:
2712 if (ph.h3->tp_next_offset != 0) {
2713 pr_warn_once("variable sized slot not supported");
2714 return -EINVAL;
2715 }
2716 tp_len = ph.h3->tp_len;
2717 break;
2718 case TPACKET_V2:
2719 tp_len = ph.h2->tp_len;
2720 break;
2721 default:
2722 tp_len = ph.h1->tp_len;
2723 break;
2724 }
2725 if (unlikely(tp_len > size_max)) {
2726 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2727 return -EMSGSIZE;
2728 }
2729
2730 if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) {
2731 int off_min, off_max;
2732
2733 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2734 off_max = po->tx_ring.frame_size - tp_len;
2735 if (po->sk.sk_type == SOCK_DGRAM) {
2736 switch (po->tp_version) {
2737 case TPACKET_V3:
2738 off = ph.h3->tp_net;
2739 break;
2740 case TPACKET_V2:
2741 off = ph.h2->tp_net;
2742 break;
2743 default:
2744 off = ph.h1->tp_net;
2745 break;
2746 }
2747 } else {
2748 switch (po->tp_version) {
2749 case TPACKET_V3:
2750 off = ph.h3->tp_mac;
2751 break;
2752 case TPACKET_V2:
2753 off = ph.h2->tp_mac;
2754 break;
2755 default:
2756 off = ph.h1->tp_mac;
2757 break;
2758 }
2759 }
2760 if (unlikely((off < off_min) || (off_max < off)))
2761 return -EINVAL;
2762 } else {
2763 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2764 }
2765
2766 *data = frame + off;
2767 return tp_len;
2768}
2769
2770static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2771{
2772 struct sk_buff *skb = NULL;
2773 struct net_device *dev;
2774 struct virtio_net_hdr *vnet_hdr = NULL;
2775 struct sockcm_cookie sockc;
2776 __be16 proto;
2777 int err, reserve = 0;
2778 void *ph;
2779 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2780 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2781 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2782 unsigned char *addr = NULL;
2783 int tp_len, size_max;
2784 void *data;
2785 int len_sum = 0;
2786 int status = TP_STATUS_AVAILABLE;
2787 int hlen, tlen, copylen = 0;
2788 long timeo = 0;
2789
2790 mutex_lock(&po->pg_vec_lock);
2791
2792 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2793 * we need to confirm it under protection of pg_vec_lock.
2794 */
2795 if (unlikely(!po->tx_ring.pg_vec)) {
2796 err = -EBUSY;
2797 goto out;
2798 }
2799 if (likely(saddr == NULL)) {
2800 dev = packet_cached_dev_get(po);
2801 proto = READ_ONCE(po->num);
2802 } else {
2803 err = -EINVAL;
2804 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2805 goto out;
2806 if (msg->msg_namelen < (saddr->sll_halen
2807 + offsetof(struct sockaddr_ll,
2808 sll_addr)))
2809 goto out;
2810 proto = saddr->sll_protocol;
2811 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2812 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2813 if (dev && msg->msg_namelen < dev->addr_len +
2814 offsetof(struct sockaddr_ll, sll_addr))
2815 goto out_put;
2816 addr = saddr->sll_addr;
2817 }
2818 }
2819
2820 err = -ENXIO;
2821 if (unlikely(dev == NULL))
2822 goto out;
2823 err = -ENETDOWN;
2824 if (unlikely(!(dev->flags & IFF_UP)))
2825 goto out_put;
2826
2827 sockcm_init(&sockc, &po->sk);
2828 if (msg->msg_controllen) {
2829 err = sock_cmsg_send(&po->sk, msg, &sockc);
2830 if (unlikely(err))
2831 goto out_put;
2832 }
2833
2834 if (po->sk.sk_socket->type == SOCK_RAW)
2835 reserve = dev->hard_header_len;
2836 size_max = po->tx_ring.frame_size
2837 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2838
2839 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2840 size_max = dev->mtu + reserve + VLAN_HLEN;
2841
2842 reinit_completion(&po->skb_completion);
2843
2844 do {
2845 ph = packet_current_frame(po, &po->tx_ring,
2846 TP_STATUS_SEND_REQUEST);
2847 if (unlikely(ph == NULL)) {
2848 if (need_wait && skb) {
2849 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2850 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2851 if (timeo <= 0) {
2852 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2853 goto out_put;
2854 }
2855 }
2856 /* check for additional frames */
2857 continue;
2858 }
2859
2860 skb = NULL;
2861 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2862 if (tp_len < 0)
2863 goto tpacket_error;
2864
2865 status = TP_STATUS_SEND_REQUEST;
2866 hlen = LL_RESERVED_SPACE(dev);
2867 tlen = dev->needed_tailroom;
2868 if (vnet_hdr_sz) {
2869 vnet_hdr = data;
2870 data += vnet_hdr_sz;
2871 tp_len -= vnet_hdr_sz;
2872 if (tp_len < 0 ||
2873 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2874 tp_len = -EINVAL;
2875 goto tpacket_error;
2876 }
2877 copylen = __virtio16_to_cpu(vio_le(),
2878 vnet_hdr->hdr_len);
2879 }
2880 copylen = max_t(int, copylen, dev->hard_header_len);
2881 skb = sock_alloc_send_skb(&po->sk,
2882 hlen + tlen + sizeof(struct sockaddr_ll) +
2883 (copylen - dev->hard_header_len),
2884 !need_wait, &err);
2885
2886 if (unlikely(skb == NULL)) {
2887 /* we assume the socket was initially writeable ... */
2888 if (likely(len_sum > 0))
2889 err = len_sum;
2890 goto out_status;
2891 }
2892 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2893 addr, hlen, copylen, &sockc);
2894 if (likely(tp_len >= 0) &&
2895 tp_len > dev->mtu + reserve &&
2896 !vnet_hdr_sz &&
2897 !packet_extra_vlan_len_allowed(dev, skb))
2898 tp_len = -EMSGSIZE;
2899
2900 if (unlikely(tp_len < 0)) {
2901tpacket_error:
2902 if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2903 __packet_set_status(po, ph,
2904 TP_STATUS_AVAILABLE);
2905 packet_increment_head(&po->tx_ring);
2906 kfree_skb(skb);
2907 continue;
2908 } else {
2909 status = TP_STATUS_WRONG_FORMAT;
2910 err = tp_len;
2911 goto out_status;
2912 }
2913 }
2914
2915 if (vnet_hdr_sz) {
2916 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2917 tp_len = -EINVAL;
2918 goto tpacket_error;
2919 }
2920 virtio_net_hdr_set_proto(skb, vnet_hdr);
2921 }
2922
2923 skb->destructor = tpacket_destruct_skb;
2924 __packet_set_status(po, ph, TP_STATUS_SENDING);
2925 packet_inc_pending(&po->tx_ring);
2926
2927 status = TP_STATUS_SEND_REQUEST;
2928 err = packet_xmit(po, skb);
2929 if (unlikely(err != 0)) {
2930 if (err > 0)
2931 err = net_xmit_errno(err);
2932 if (err && __packet_get_status(po, ph) ==
2933 TP_STATUS_AVAILABLE) {
2934 /* skb was destructed already */
2935 skb = NULL;
2936 goto out_status;
2937 }
2938 /*
2939 * skb was dropped but not destructed yet;
2940 * let's treat it like congestion or err < 0
2941 */
2942 err = 0;
2943 }
2944 packet_increment_head(&po->tx_ring);
2945 len_sum += tp_len;
2946 } while (likely((ph != NULL) ||
2947 /* Note: packet_read_pending() might be slow if we have
2948 * to call it as it's per_cpu variable, but in fast-path
2949 * we already short-circuit the loop with the first
2950 * condition, and luckily don't have to go that path
2951 * anyway.
2952 */
2953 (need_wait && packet_read_pending(&po->tx_ring))));
2954
2955 err = len_sum;
2956 goto out_put;
2957
2958out_status:
2959 __packet_set_status(po, ph, status);
2960 kfree_skb(skb);
2961out_put:
2962 dev_put(dev);
2963out:
2964 mutex_unlock(&po->pg_vec_lock);
2965 return err;
2966}
2967
2968static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2969 size_t reserve, size_t len,
2970 size_t linear, int noblock,
2971 int *err)
2972{
2973 struct sk_buff *skb;
2974
2975 /* Under a page? Don't bother with paged skb. */
2976 if (prepad + len < PAGE_SIZE || !linear)
2977 linear = len;
2978
2979 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
2980 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
2981 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2982 err, PAGE_ALLOC_COSTLY_ORDER);
2983 if (!skb)
2984 return NULL;
2985
2986 skb_reserve(skb, reserve);
2987 skb_put(skb, linear);
2988 skb->data_len = len - linear;
2989 skb->len += len - linear;
2990
2991 return skb;
2992}
2993
2994static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2995{
2996 struct sock *sk = sock->sk;
2997 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2998 struct sk_buff *skb;
2999 struct net_device *dev;
3000 __be16 proto;
3001 unsigned char *addr = NULL;
3002 int err, reserve = 0;
3003 struct sockcm_cookie sockc;
3004 struct virtio_net_hdr vnet_hdr = { 0 };
3005 int offset = 0;
3006 struct packet_sock *po = pkt_sk(sk);
3007 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
3008 int hlen, tlen, linear;
3009 int extra_len = 0;
3010
3011 /*
3012 * Get and verify the address.
3013 */
3014
3015 if (likely(saddr == NULL)) {
3016 dev = packet_cached_dev_get(po);
3017 proto = READ_ONCE(po->num);
3018 } else {
3019 err = -EINVAL;
3020 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
3021 goto out;
3022 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
3023 goto out;
3024 proto = saddr->sll_protocol;
3025 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
3026 if (sock->type == SOCK_DGRAM) {
3027 if (dev && msg->msg_namelen < dev->addr_len +
3028 offsetof(struct sockaddr_ll, sll_addr))
3029 goto out_unlock;
3030 addr = saddr->sll_addr;
3031 }
3032 }
3033
3034 err = -ENXIO;
3035 if (unlikely(dev == NULL))
3036 goto out_unlock;
3037 err = -ENETDOWN;
3038 if (unlikely(!(dev->flags & IFF_UP)))
3039 goto out_unlock;
3040
3041 sockcm_init(&sockc, sk);
3042 sockc.mark = READ_ONCE(sk->sk_mark);
3043 if (msg->msg_controllen) {
3044 err = sock_cmsg_send(sk, msg, &sockc);
3045 if (unlikely(err))
3046 goto out_unlock;
3047 }
3048
3049 if (sock->type == SOCK_RAW)
3050 reserve = dev->hard_header_len;
3051 if (vnet_hdr_sz) {
3052 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3053 if (err)
3054 goto out_unlock;
3055 }
3056
3057 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3058 if (!netif_supports_nofcs(dev)) {
3059 err = -EPROTONOSUPPORT;
3060 goto out_unlock;
3061 }
3062 extra_len = 4; /* We're doing our own CRC */
3063 }
3064
3065 err = -EMSGSIZE;
3066 if (!vnet_hdr.gso_type &&
3067 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3068 goto out_unlock;
3069
3070 err = -ENOBUFS;
3071 hlen = LL_RESERVED_SPACE(dev);
3072 tlen = dev->needed_tailroom;
3073 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3074 linear = max(linear, min_t(int, len, dev->hard_header_len));
3075 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3076 msg->msg_flags & MSG_DONTWAIT, &err);
3077 if (skb == NULL)
3078 goto out_unlock;
3079
3080 skb_reset_network_header(skb);
3081
3082 err = -EINVAL;
3083 if (sock->type == SOCK_DGRAM) {
3084 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3085 if (unlikely(offset < 0))
3086 goto out_free;
3087 } else if (reserve) {
3088 skb_reserve(skb, -reserve);
3089 if (len < reserve + sizeof(struct ipv6hdr) &&
3090 dev->min_header_len != dev->hard_header_len)
3091 skb_reset_network_header(skb);
3092 }
3093
3094 /* Returns -EFAULT on error */
3095 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3096 if (err)
3097 goto out_free;
3098
3099 if ((sock->type == SOCK_RAW &&
3100 !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3101 err = -EINVAL;
3102 goto out_free;
3103 }
3104
3105 skb_setup_tx_timestamp(skb, &sockc);
3106
3107 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3108 !packet_extra_vlan_len_allowed(dev, skb)) {
3109 err = -EMSGSIZE;
3110 goto out_free;
3111 }
3112
3113 skb->protocol = proto;
3114 skb->dev = dev;
3115 skb->priority = READ_ONCE(sk->sk_priority);
3116 skb->mark = sockc.mark;
3117 skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
3118
3119 if (unlikely(extra_len == 4))
3120 skb->no_fcs = 1;
3121
3122 packet_parse_headers(skb, sock);
3123
3124 if (vnet_hdr_sz) {
3125 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3126 if (err)
3127 goto out_free;
3128 len += vnet_hdr_sz;
3129 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3130 }
3131
3132 err = packet_xmit(po, skb);
3133
3134 if (unlikely(err != 0)) {
3135 if (err > 0)
3136 err = net_xmit_errno(err);
3137 if (err)
3138 goto out_unlock;
3139 }
3140
3141 dev_put(dev);
3142
3143 return len;
3144
3145out_free:
3146 kfree_skb(skb);
3147out_unlock:
3148 dev_put(dev);
3149out:
3150 return err;
3151}
3152
3153static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3154{
3155 struct sock *sk = sock->sk;
3156 struct packet_sock *po = pkt_sk(sk);
3157
3158 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3159 * tpacket_snd() will redo the check safely.
3160 */
3161 if (data_race(po->tx_ring.pg_vec))
3162 return tpacket_snd(po, msg);
3163
3164 return packet_snd(sock, msg, len);
3165}
3166
3167/*
3168 * Close a PACKET socket. This is fairly simple. We immediately go
3169 * to 'closed' state and remove our protocol entry in the device list.
3170 */
3171
3172static int packet_release(struct socket *sock)
3173{
3174 struct sock *sk = sock->sk;
3175 struct packet_sock *po;
3176 struct packet_fanout *f;
3177 struct net *net;
3178 union tpacket_req_u req_u;
3179
3180 if (!sk)
3181 return 0;
3182
3183 net = sock_net(sk);
3184 po = pkt_sk(sk);
3185
3186 mutex_lock(&net->packet.sklist_lock);
3187 sk_del_node_init_rcu(sk);
3188 mutex_unlock(&net->packet.sklist_lock);
3189
3190 sock_prot_inuse_add(net, sk->sk_prot, -1);
3191
3192 spin_lock(&po->bind_lock);
3193 unregister_prot_hook(sk, false);
3194 packet_cached_dev_reset(po);
3195
3196 if (po->prot_hook.dev) {
3197 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3198 po->prot_hook.dev = NULL;
3199 }
3200 spin_unlock(&po->bind_lock);
3201
3202 packet_flush_mclist(sk);
3203
3204 lock_sock(sk);
3205 if (po->rx_ring.pg_vec) {
3206 memset(&req_u, 0, sizeof(req_u));
3207 packet_set_ring(sk, &req_u, 1, 0);
3208 }
3209
3210 if (po->tx_ring.pg_vec) {
3211 memset(&req_u, 0, sizeof(req_u));
3212 packet_set_ring(sk, &req_u, 1, 1);
3213 }
3214 release_sock(sk);
3215
3216 f = fanout_release(sk);
3217
3218 synchronize_net();
3219
3220 kfree(po->rollover);
3221 if (f) {
3222 fanout_release_data(f);
3223 kvfree(f);
3224 }
3225 /*
3226 * Now the socket is dead. No more input will appear.
3227 */
3228 sock_orphan(sk);
3229 sock->sk = NULL;
3230
3231 /* Purge queues */
3232
3233 skb_queue_purge(&sk->sk_receive_queue);
3234 packet_free_pending(po);
3235
3236 sock_put(sk);
3237 return 0;
3238}
3239
3240/*
3241 * Attach a packet hook.
3242 */
3243
3244static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3245 __be16 proto)
3246{
3247 struct packet_sock *po = pkt_sk(sk);
3248 struct net_device *dev = NULL;
3249 bool unlisted = false;
3250 bool need_rehook;
3251 int ret = 0;
3252
3253 lock_sock(sk);
3254 spin_lock(&po->bind_lock);
3255 if (!proto)
3256 proto = po->num;
3257
3258 rcu_read_lock();
3259
3260 if (po->fanout) {
3261 ret = -EINVAL;
3262 goto out_unlock;
3263 }
3264
3265 if (name) {
3266 dev = dev_get_by_name_rcu(sock_net(sk), name);
3267 if (!dev) {
3268 ret = -ENODEV;
3269 goto out_unlock;
3270 }
3271 } else if (ifindex) {
3272 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3273 if (!dev) {
3274 ret = -ENODEV;
3275 goto out_unlock;
3276 }
3277 }
3278
3279 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3280
3281 if (need_rehook) {
3282 dev_hold(dev);
3283 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3284 rcu_read_unlock();
3285 /* prevents packet_notifier() from calling
3286 * register_prot_hook()
3287 */
3288 WRITE_ONCE(po->num, 0);
3289 __unregister_prot_hook(sk, true);
3290 rcu_read_lock();
3291 if (dev)
3292 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3293 dev->ifindex);
3294 }
3295
3296 BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3297 WRITE_ONCE(po->num, proto);
3298 po->prot_hook.type = proto;
3299
3300 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3301
3302 if (unlikely(unlisted)) {
3303 po->prot_hook.dev = NULL;
3304 WRITE_ONCE(po->ifindex, -1);
3305 packet_cached_dev_reset(po);
3306 } else {
3307 netdev_hold(dev, &po->prot_hook.dev_tracker,
3308 GFP_ATOMIC);
3309 po->prot_hook.dev = dev;
3310 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3311 packet_cached_dev_assign(po, dev);
3312 }
3313 dev_put(dev);
3314 }
3315
3316 if (proto == 0 || !need_rehook)
3317 goto out_unlock;
3318
3319 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3320 register_prot_hook(sk);
3321 } else {
3322 sk->sk_err = ENETDOWN;
3323 if (!sock_flag(sk, SOCK_DEAD))
3324 sk_error_report(sk);
3325 }
3326
3327out_unlock:
3328 rcu_read_unlock();
3329 spin_unlock(&po->bind_lock);
3330 release_sock(sk);
3331 return ret;
3332}
3333
3334/*
3335 * Bind a packet socket to a device
3336 */
3337
3338static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3339 int addr_len)
3340{
3341 struct sock *sk = sock->sk;
3342 char name[sizeof(uaddr->sa_data_min) + 1];
3343
3344 /*
3345 * Check legality
3346 */
3347
3348 if (addr_len != sizeof(struct sockaddr))
3349 return -EINVAL;
3350 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3351 * zero-terminated.
3352 */
3353 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3354 name[sizeof(uaddr->sa_data_min)] = 0;
3355
3356 return packet_do_bind(sk, name, 0, 0);
3357}
3358
3359static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3360{
3361 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3362 struct sock *sk = sock->sk;
3363
3364 /*
3365 * Check legality
3366 */
3367
3368 if (addr_len < sizeof(struct sockaddr_ll))
3369 return -EINVAL;
3370 if (sll->sll_family != AF_PACKET)
3371 return -EINVAL;
3372
3373 return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3374}
3375
3376static struct proto packet_proto = {
3377 .name = "PACKET",
3378 .owner = THIS_MODULE,
3379 .obj_size = sizeof(struct packet_sock),
3380};
3381
3382/*
3383 * Create a packet of type SOCK_PACKET.
3384 */
3385
3386static int packet_create(struct net *net, struct socket *sock, int protocol,
3387 int kern)
3388{
3389 struct sock *sk;
3390 struct packet_sock *po;
3391 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3392 int err;
3393
3394 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3395 return -EPERM;
3396 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3397 sock->type != SOCK_PACKET)
3398 return -ESOCKTNOSUPPORT;
3399
3400 sock->state = SS_UNCONNECTED;
3401
3402 err = -ENOBUFS;
3403 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3404 if (sk == NULL)
3405 goto out;
3406
3407 sock->ops = &packet_ops;
3408 if (sock->type == SOCK_PACKET)
3409 sock->ops = &packet_ops_spkt;
3410
3411 po = pkt_sk(sk);
3412 err = packet_alloc_pending(po);
3413 if (err)
3414 goto out_sk_free;
3415
3416 sock_init_data(sock, sk);
3417
3418 init_completion(&po->skb_completion);
3419 sk->sk_family = PF_PACKET;
3420 po->num = proto;
3421
3422 packet_cached_dev_reset(po);
3423
3424 sk->sk_destruct = packet_sock_destruct;
3425
3426 /*
3427 * Attach a protocol block
3428 */
3429
3430 spin_lock_init(&po->bind_lock);
3431 mutex_init(&po->pg_vec_lock);
3432 po->rollover = NULL;
3433 po->prot_hook.func = packet_rcv;
3434
3435 if (sock->type == SOCK_PACKET)
3436 po->prot_hook.func = packet_rcv_spkt;
3437
3438 po->prot_hook.af_packet_priv = sk;
3439 po->prot_hook.af_packet_net = sock_net(sk);
3440
3441 if (proto) {
3442 po->prot_hook.type = proto;
3443 __register_prot_hook(sk);
3444 }
3445
3446 mutex_lock(&net->packet.sklist_lock);
3447 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3448 mutex_unlock(&net->packet.sklist_lock);
3449
3450 sock_prot_inuse_add(net, &packet_proto, 1);
3451
3452 return 0;
3453out_sk_free:
3454 sk_free(sk);
3455out:
3456 return err;
3457}
3458
3459/*
3460 * Pull a packet from our receive queue and hand it to the user.
3461 * If necessary we block.
3462 */
3463
3464static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3465 int flags)
3466{
3467 struct sock *sk = sock->sk;
3468 struct sk_buff *skb;
3469 int copied, err;
3470 int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3471 unsigned int origlen = 0;
3472
3473 err = -EINVAL;
3474 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3475 goto out;
3476
3477#if 0
3478 /* What error should we return now? EUNATTACH? */
3479 if (pkt_sk(sk)->ifindex < 0)
3480 return -ENODEV;
3481#endif
3482
3483 if (flags & MSG_ERRQUEUE) {
3484 err = sock_recv_errqueue(sk, msg, len,
3485 SOL_PACKET, PACKET_TX_TIMESTAMP);
3486 goto out;
3487 }
3488
3489 /*
3490 * Call the generic datagram receiver. This handles all sorts
3491 * of horrible races and re-entrancy so we can forget about it
3492 * in the protocol layers.
3493 *
3494 * Now it will return ENETDOWN, if device have just gone down,
3495 * but then it will block.
3496 */
3497
3498 skb = skb_recv_datagram(sk, flags, &err);
3499
3500 /*
3501 * An error occurred so return it. Because skb_recv_datagram()
3502 * handles the blocking we don't see and worry about blocking
3503 * retries.
3504 */
3505
3506 if (skb == NULL)
3507 goto out;
3508
3509 packet_rcv_try_clear_pressure(pkt_sk(sk));
3510
3511 if (vnet_hdr_len) {
3512 err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3513 if (err)
3514 goto out_free;
3515 }
3516
3517 /* You lose any data beyond the buffer you gave. If it worries
3518 * a user program they can ask the device for its MTU
3519 * anyway.
3520 */
3521 copied = skb->len;
3522 if (copied > len) {
3523 copied = len;
3524 msg->msg_flags |= MSG_TRUNC;
3525 }
3526
3527 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3528 if (err)
3529 goto out_free;
3530
3531 if (sock->type != SOCK_PACKET) {
3532 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3533
3534 /* Original length was stored in sockaddr_ll fields */
3535 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3536 sll->sll_family = AF_PACKET;
3537 sll->sll_protocol = (sock->type == SOCK_DGRAM) ?
3538 vlan_get_protocol_dgram(skb) : skb->protocol;
3539 }
3540
3541 sock_recv_cmsgs(msg, sk, skb);
3542
3543 if (msg->msg_name) {
3544 const size_t max_len = min(sizeof(skb->cb),
3545 sizeof(struct sockaddr_storage));
3546 int copy_len;
3547
3548 /* If the address length field is there to be filled
3549 * in, we fill it in now.
3550 */
3551 if (sock->type == SOCK_PACKET) {
3552 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3553 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3554 copy_len = msg->msg_namelen;
3555 } else {
3556 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3557
3558 msg->msg_namelen = sll->sll_halen +
3559 offsetof(struct sockaddr_ll, sll_addr);
3560 copy_len = msg->msg_namelen;
3561 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3562 memset(msg->msg_name +
3563 offsetof(struct sockaddr_ll, sll_addr),
3564 0, sizeof(sll->sll_addr));
3565 msg->msg_namelen = sizeof(struct sockaddr_ll);
3566 }
3567 }
3568 if (WARN_ON_ONCE(copy_len > max_len)) {
3569 copy_len = max_len;
3570 msg->msg_namelen = copy_len;
3571 }
3572 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3573 }
3574
3575 if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3576 struct tpacket_auxdata aux;
3577
3578 aux.tp_status = TP_STATUS_USER;
3579 if (skb->ip_summed == CHECKSUM_PARTIAL)
3580 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3581 else if (skb->pkt_type != PACKET_OUTGOING &&
3582 skb_csum_unnecessary(skb))
3583 aux.tp_status |= TP_STATUS_CSUM_VALID;
3584 if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3585 aux.tp_status |= TP_STATUS_GSO_TCP;
3586
3587 aux.tp_len = origlen;
3588 aux.tp_snaplen = skb->len;
3589 aux.tp_mac = 0;
3590 aux.tp_net = skb_network_offset(skb);
3591 if (skb_vlan_tag_present(skb)) {
3592 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3593 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3594 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3595 } else if (unlikely(sock->type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
3596 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3597 struct net_device *dev;
3598
3599 rcu_read_lock();
3600 dev = dev_get_by_index_rcu(sock_net(sk), sll->sll_ifindex);
3601 if (dev) {
3602 aux.tp_vlan_tci = vlan_get_tci(skb, dev);
3603 aux.tp_vlan_tpid = ntohs(skb->protocol);
3604 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3605 } else {
3606 aux.tp_vlan_tci = 0;
3607 aux.tp_vlan_tpid = 0;
3608 }
3609 rcu_read_unlock();
3610 } else {
3611 aux.tp_vlan_tci = 0;
3612 aux.tp_vlan_tpid = 0;
3613 }
3614 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3615 }
3616
3617 /*
3618 * Free or return the buffer as appropriate. Again this
3619 * hides all the races and re-entrancy issues from us.
3620 */
3621 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3622
3623out_free:
3624 skb_free_datagram(sk, skb);
3625out:
3626 return err;
3627}
3628
3629static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3630 int peer)
3631{
3632 struct net_device *dev;
3633 struct sock *sk = sock->sk;
3634
3635 if (peer)
3636 return -EOPNOTSUPP;
3637
3638 uaddr->sa_family = AF_PACKET;
3639 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3640 rcu_read_lock();
3641 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3642 if (dev)
3643 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3644 rcu_read_unlock();
3645
3646 return sizeof(*uaddr);
3647}
3648
3649static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3650 int peer)
3651{
3652 struct net_device *dev;
3653 struct sock *sk = sock->sk;
3654 struct packet_sock *po = pkt_sk(sk);
3655 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3656 int ifindex;
3657
3658 if (peer)
3659 return -EOPNOTSUPP;
3660
3661 ifindex = READ_ONCE(po->ifindex);
3662 sll->sll_family = AF_PACKET;
3663 sll->sll_ifindex = ifindex;
3664 sll->sll_protocol = READ_ONCE(po->num);
3665 sll->sll_pkttype = 0;
3666 rcu_read_lock();
3667 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3668 if (dev) {
3669 sll->sll_hatype = dev->type;
3670 sll->sll_halen = dev->addr_len;
3671
3672 /* Let __fortify_memcpy_chk() know the actual buffer size. */
3673 memcpy(((struct sockaddr_storage *)sll)->__data +
3674 offsetof(struct sockaddr_ll, sll_addr) -
3675 offsetofend(struct sockaddr_ll, sll_family),
3676 dev->dev_addr, dev->addr_len);
3677 } else {
3678 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3679 sll->sll_halen = 0;
3680 }
3681 rcu_read_unlock();
3682
3683 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3684}
3685
3686static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3687 int what)
3688{
3689 switch (i->type) {
3690 case PACKET_MR_MULTICAST:
3691 if (i->alen != dev->addr_len)
3692 return -EINVAL;
3693 if (what > 0)
3694 return dev_mc_add(dev, i->addr);
3695 else
3696 return dev_mc_del(dev, i->addr);
3697 break;
3698 case PACKET_MR_PROMISC:
3699 return dev_set_promiscuity(dev, what);
3700 case PACKET_MR_ALLMULTI:
3701 return dev_set_allmulti(dev, what);
3702 case PACKET_MR_UNICAST:
3703 if (i->alen != dev->addr_len)
3704 return -EINVAL;
3705 if (what > 0)
3706 return dev_uc_add(dev, i->addr);
3707 else
3708 return dev_uc_del(dev, i->addr);
3709 break;
3710 default:
3711 break;
3712 }
3713 return 0;
3714}
3715
3716static void packet_dev_mclist_delete(struct net_device *dev,
3717 struct packet_mclist **mlp)
3718{
3719 struct packet_mclist *ml;
3720
3721 while ((ml = *mlp) != NULL) {
3722 if (ml->ifindex == dev->ifindex) {
3723 packet_dev_mc(dev, ml, -1);
3724 *mlp = ml->next;
3725 kfree(ml);
3726 } else
3727 mlp = &ml->next;
3728 }
3729}
3730
3731static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3732{
3733 struct packet_sock *po = pkt_sk(sk);
3734 struct packet_mclist *ml, *i;
3735 struct net_device *dev;
3736 int err;
3737
3738 rtnl_lock();
3739
3740 err = -ENODEV;
3741 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3742 if (!dev)
3743 goto done;
3744
3745 err = -EINVAL;
3746 if (mreq->mr_alen > dev->addr_len)
3747 goto done;
3748
3749 err = -ENOBUFS;
3750 i = kmalloc(sizeof(*i), GFP_KERNEL);
3751 if (i == NULL)
3752 goto done;
3753
3754 err = 0;
3755 for (ml = po->mclist; ml; ml = ml->next) {
3756 if (ml->ifindex == mreq->mr_ifindex &&
3757 ml->type == mreq->mr_type &&
3758 ml->alen == mreq->mr_alen &&
3759 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3760 ml->count++;
3761 /* Free the new element ... */
3762 kfree(i);
3763 goto done;
3764 }
3765 }
3766
3767 i->type = mreq->mr_type;
3768 i->ifindex = mreq->mr_ifindex;
3769 i->alen = mreq->mr_alen;
3770 memcpy(i->addr, mreq->mr_address, i->alen);
3771 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3772 i->count = 1;
3773 i->next = po->mclist;
3774 po->mclist = i;
3775 err = packet_dev_mc(dev, i, 1);
3776 if (err) {
3777 po->mclist = i->next;
3778 kfree(i);
3779 }
3780
3781done:
3782 rtnl_unlock();
3783 return err;
3784}
3785
3786static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3787{
3788 struct packet_mclist *ml, **mlp;
3789
3790 rtnl_lock();
3791
3792 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3793 if (ml->ifindex == mreq->mr_ifindex &&
3794 ml->type == mreq->mr_type &&
3795 ml->alen == mreq->mr_alen &&
3796 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3797 if (--ml->count == 0) {
3798 struct net_device *dev;
3799 *mlp = ml->next;
3800 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3801 if (dev)
3802 packet_dev_mc(dev, ml, -1);
3803 kfree(ml);
3804 }
3805 break;
3806 }
3807 }
3808 rtnl_unlock();
3809 return 0;
3810}
3811
3812static void packet_flush_mclist(struct sock *sk)
3813{
3814 struct packet_sock *po = pkt_sk(sk);
3815 struct packet_mclist *ml;
3816
3817 if (!po->mclist)
3818 return;
3819
3820 rtnl_lock();
3821 while ((ml = po->mclist) != NULL) {
3822 struct net_device *dev;
3823
3824 po->mclist = ml->next;
3825 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3826 if (dev != NULL)
3827 packet_dev_mc(dev, ml, -1);
3828 kfree(ml);
3829 }
3830 rtnl_unlock();
3831}
3832
3833static int
3834packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3835 unsigned int optlen)
3836{
3837 struct sock *sk = sock->sk;
3838 struct packet_sock *po = pkt_sk(sk);
3839 int ret;
3840
3841 if (level != SOL_PACKET)
3842 return -ENOPROTOOPT;
3843
3844 switch (optname) {
3845 case PACKET_ADD_MEMBERSHIP:
3846 case PACKET_DROP_MEMBERSHIP:
3847 {
3848 struct packet_mreq_max mreq;
3849 int len = optlen;
3850 memset(&mreq, 0, sizeof(mreq));
3851 if (len < sizeof(struct packet_mreq))
3852 return -EINVAL;
3853 if (len > sizeof(mreq))
3854 len = sizeof(mreq);
3855 if (copy_from_sockptr(&mreq, optval, len))
3856 return -EFAULT;
3857 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3858 return -EINVAL;
3859 if (optname == PACKET_ADD_MEMBERSHIP)
3860 ret = packet_mc_add(sk, &mreq);
3861 else
3862 ret = packet_mc_drop(sk, &mreq);
3863 return ret;
3864 }
3865
3866 case PACKET_RX_RING:
3867 case PACKET_TX_RING:
3868 {
3869 union tpacket_req_u req_u;
3870
3871 ret = -EINVAL;
3872 lock_sock(sk);
3873 switch (po->tp_version) {
3874 case TPACKET_V1:
3875 case TPACKET_V2:
3876 if (optlen < sizeof(req_u.req))
3877 break;
3878 ret = copy_from_sockptr(&req_u.req, optval,
3879 sizeof(req_u.req)) ?
3880 -EINVAL : 0;
3881 break;
3882 case TPACKET_V3:
3883 default:
3884 if (optlen < sizeof(req_u.req3))
3885 break;
3886 ret = copy_from_sockptr(&req_u.req3, optval,
3887 sizeof(req_u.req3)) ?
3888 -EINVAL : 0;
3889 break;
3890 }
3891 if (!ret)
3892 ret = packet_set_ring(sk, &req_u, 0,
3893 optname == PACKET_TX_RING);
3894 release_sock(sk);
3895 return ret;
3896 }
3897 case PACKET_COPY_THRESH:
3898 {
3899 int val;
3900
3901 if (optlen != sizeof(val))
3902 return -EINVAL;
3903 if (copy_from_sockptr(&val, optval, sizeof(val)))
3904 return -EFAULT;
3905
3906 WRITE_ONCE(pkt_sk(sk)->copy_thresh, val);
3907 return 0;
3908 }
3909 case PACKET_VERSION:
3910 {
3911 int val;
3912
3913 if (optlen != sizeof(val))
3914 return -EINVAL;
3915 if (copy_from_sockptr(&val, optval, sizeof(val)))
3916 return -EFAULT;
3917 switch (val) {
3918 case TPACKET_V1:
3919 case TPACKET_V2:
3920 case TPACKET_V3:
3921 break;
3922 default:
3923 return -EINVAL;
3924 }
3925 lock_sock(sk);
3926 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3927 ret = -EBUSY;
3928 } else {
3929 po->tp_version = val;
3930 ret = 0;
3931 }
3932 release_sock(sk);
3933 return ret;
3934 }
3935 case PACKET_RESERVE:
3936 {
3937 unsigned int val;
3938
3939 if (optlen != sizeof(val))
3940 return -EINVAL;
3941 if (copy_from_sockptr(&val, optval, sizeof(val)))
3942 return -EFAULT;
3943 if (val > INT_MAX)
3944 return -EINVAL;
3945 lock_sock(sk);
3946 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3947 ret = -EBUSY;
3948 } else {
3949 po->tp_reserve = val;
3950 ret = 0;
3951 }
3952 release_sock(sk);
3953 return ret;
3954 }
3955 case PACKET_LOSS:
3956 {
3957 unsigned int val;
3958
3959 if (optlen != sizeof(val))
3960 return -EINVAL;
3961 if (copy_from_sockptr(&val, optval, sizeof(val)))
3962 return -EFAULT;
3963
3964 lock_sock(sk);
3965 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3966 ret = -EBUSY;
3967 } else {
3968 packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3969 ret = 0;
3970 }
3971 release_sock(sk);
3972 return ret;
3973 }
3974 case PACKET_AUXDATA:
3975 {
3976 int val;
3977
3978 if (optlen < sizeof(val))
3979 return -EINVAL;
3980 if (copy_from_sockptr(&val, optval, sizeof(val)))
3981 return -EFAULT;
3982
3983 packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3984 return 0;
3985 }
3986 case PACKET_ORIGDEV:
3987 {
3988 int val;
3989
3990 if (optlen < sizeof(val))
3991 return -EINVAL;
3992 if (copy_from_sockptr(&val, optval, sizeof(val)))
3993 return -EFAULT;
3994
3995 packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3996 return 0;
3997 }
3998 case PACKET_VNET_HDR:
3999 case PACKET_VNET_HDR_SZ:
4000 {
4001 int val, hdr_len;
4002
4003 if (sock->type != SOCK_RAW)
4004 return -EINVAL;
4005 if (optlen < sizeof(val))
4006 return -EINVAL;
4007 if (copy_from_sockptr(&val, optval, sizeof(val)))
4008 return -EFAULT;
4009
4010 if (optname == PACKET_VNET_HDR_SZ) {
4011 if (val && val != sizeof(struct virtio_net_hdr) &&
4012 val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
4013 return -EINVAL;
4014 hdr_len = val;
4015 } else {
4016 hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
4017 }
4018 lock_sock(sk);
4019 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
4020 ret = -EBUSY;
4021 } else {
4022 WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
4023 ret = 0;
4024 }
4025 release_sock(sk);
4026 return ret;
4027 }
4028 case PACKET_TIMESTAMP:
4029 {
4030 int val;
4031
4032 if (optlen != sizeof(val))
4033 return -EINVAL;
4034 if (copy_from_sockptr(&val, optval, sizeof(val)))
4035 return -EFAULT;
4036
4037 WRITE_ONCE(po->tp_tstamp, val);
4038 return 0;
4039 }
4040 case PACKET_FANOUT:
4041 {
4042 struct fanout_args args = { 0 };
4043
4044 if (optlen != sizeof(int) && optlen != sizeof(args))
4045 return -EINVAL;
4046 if (copy_from_sockptr(&args, optval, optlen))
4047 return -EFAULT;
4048
4049 return fanout_add(sk, &args);
4050 }
4051 case PACKET_FANOUT_DATA:
4052 {
4053 /* Paired with the WRITE_ONCE() in fanout_add() */
4054 if (!READ_ONCE(po->fanout))
4055 return -EINVAL;
4056
4057 return fanout_set_data(po, optval, optlen);
4058 }
4059 case PACKET_IGNORE_OUTGOING:
4060 {
4061 int val;
4062
4063 if (optlen != sizeof(val))
4064 return -EINVAL;
4065 if (copy_from_sockptr(&val, optval, sizeof(val)))
4066 return -EFAULT;
4067 if (val < 0 || val > 1)
4068 return -EINVAL;
4069
4070 WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val);
4071 return 0;
4072 }
4073 case PACKET_TX_HAS_OFF:
4074 {
4075 unsigned int val;
4076
4077 if (optlen != sizeof(val))
4078 return -EINVAL;
4079 if (copy_from_sockptr(&val, optval, sizeof(val)))
4080 return -EFAULT;
4081
4082 lock_sock(sk);
4083 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4084 packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4085
4086 release_sock(sk);
4087 return 0;
4088 }
4089 case PACKET_QDISC_BYPASS:
4090 {
4091 int val;
4092
4093 if (optlen != sizeof(val))
4094 return -EINVAL;
4095 if (copy_from_sockptr(&val, optval, sizeof(val)))
4096 return -EFAULT;
4097
4098 packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4099 return 0;
4100 }
4101 default:
4102 return -ENOPROTOOPT;
4103 }
4104}
4105
4106static int packet_getsockopt(struct socket *sock, int level, int optname,
4107 char __user *optval, int __user *optlen)
4108{
4109 int len;
4110 int val, lv = sizeof(val);
4111 struct sock *sk = sock->sk;
4112 struct packet_sock *po = pkt_sk(sk);
4113 void *data = &val;
4114 union tpacket_stats_u st;
4115 struct tpacket_rollover_stats rstats;
4116 int drops;
4117
4118 if (level != SOL_PACKET)
4119 return -ENOPROTOOPT;
4120
4121 if (get_user(len, optlen))
4122 return -EFAULT;
4123
4124 if (len < 0)
4125 return -EINVAL;
4126
4127 switch (optname) {
4128 case PACKET_STATISTICS:
4129 spin_lock_bh(&sk->sk_receive_queue.lock);
4130 memcpy(&st, &po->stats, sizeof(st));
4131 memset(&po->stats, 0, sizeof(po->stats));
4132 spin_unlock_bh(&sk->sk_receive_queue.lock);
4133 drops = atomic_xchg(&po->tp_drops, 0);
4134
4135 if (po->tp_version == TPACKET_V3) {
4136 lv = sizeof(struct tpacket_stats_v3);
4137 st.stats3.tp_drops = drops;
4138 st.stats3.tp_packets += drops;
4139 data = &st.stats3;
4140 } else {
4141 lv = sizeof(struct tpacket_stats);
4142 st.stats1.tp_drops = drops;
4143 st.stats1.tp_packets += drops;
4144 data = &st.stats1;
4145 }
4146
4147 break;
4148 case PACKET_AUXDATA:
4149 val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4150 break;
4151 case PACKET_ORIGDEV:
4152 val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4153 break;
4154 case PACKET_VNET_HDR:
4155 val = !!READ_ONCE(po->vnet_hdr_sz);
4156 break;
4157 case PACKET_VNET_HDR_SZ:
4158 val = READ_ONCE(po->vnet_hdr_sz);
4159 break;
4160 case PACKET_COPY_THRESH:
4161 val = READ_ONCE(pkt_sk(sk)->copy_thresh);
4162 break;
4163 case PACKET_VERSION:
4164 val = po->tp_version;
4165 break;
4166 case PACKET_HDRLEN:
4167 if (len > sizeof(int))
4168 len = sizeof(int);
4169 if (len < sizeof(int))
4170 return -EINVAL;
4171 if (copy_from_user(&val, optval, len))
4172 return -EFAULT;
4173 switch (val) {
4174 case TPACKET_V1:
4175 val = sizeof(struct tpacket_hdr);
4176 break;
4177 case TPACKET_V2:
4178 val = sizeof(struct tpacket2_hdr);
4179 break;
4180 case TPACKET_V3:
4181 val = sizeof(struct tpacket3_hdr);
4182 break;
4183 default:
4184 return -EINVAL;
4185 }
4186 break;
4187 case PACKET_RESERVE:
4188 val = po->tp_reserve;
4189 break;
4190 case PACKET_LOSS:
4191 val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4192 break;
4193 case PACKET_TIMESTAMP:
4194 val = READ_ONCE(po->tp_tstamp);
4195 break;
4196 case PACKET_FANOUT:
4197 val = (po->fanout ?
4198 ((u32)po->fanout->id |
4199 ((u32)po->fanout->type << 16) |
4200 ((u32)po->fanout->flags << 24)) :
4201 0);
4202 break;
4203 case PACKET_IGNORE_OUTGOING:
4204 val = READ_ONCE(po->prot_hook.ignore_outgoing);
4205 break;
4206 case PACKET_ROLLOVER_STATS:
4207 if (!po->rollover)
4208 return -EINVAL;
4209 rstats.tp_all = atomic_long_read(&po->rollover->num);
4210 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4211 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4212 data = &rstats;
4213 lv = sizeof(rstats);
4214 break;
4215 case PACKET_TX_HAS_OFF:
4216 val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4217 break;
4218 case PACKET_QDISC_BYPASS:
4219 val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4220 break;
4221 default:
4222 return -ENOPROTOOPT;
4223 }
4224
4225 if (len > lv)
4226 len = lv;
4227 if (put_user(len, optlen))
4228 return -EFAULT;
4229 if (copy_to_user(optval, data, len))
4230 return -EFAULT;
4231 return 0;
4232}
4233
4234static int packet_notifier(struct notifier_block *this,
4235 unsigned long msg, void *ptr)
4236{
4237 struct sock *sk;
4238 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4239 struct net *net = dev_net(dev);
4240
4241 rcu_read_lock();
4242 sk_for_each_rcu(sk, &net->packet.sklist) {
4243 struct packet_sock *po = pkt_sk(sk);
4244
4245 switch (msg) {
4246 case NETDEV_UNREGISTER:
4247 if (po->mclist)
4248 packet_dev_mclist_delete(dev, &po->mclist);
4249 fallthrough;
4250
4251 case NETDEV_DOWN:
4252 if (dev->ifindex == po->ifindex) {
4253 spin_lock(&po->bind_lock);
4254 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4255 __unregister_prot_hook(sk, false);
4256 sk->sk_err = ENETDOWN;
4257 if (!sock_flag(sk, SOCK_DEAD))
4258 sk_error_report(sk);
4259 }
4260 if (msg == NETDEV_UNREGISTER) {
4261 packet_cached_dev_reset(po);
4262 WRITE_ONCE(po->ifindex, -1);
4263 netdev_put(po->prot_hook.dev,
4264 &po->prot_hook.dev_tracker);
4265 po->prot_hook.dev = NULL;
4266 }
4267 spin_unlock(&po->bind_lock);
4268 }
4269 break;
4270 case NETDEV_UP:
4271 if (dev->ifindex == po->ifindex) {
4272 spin_lock(&po->bind_lock);
4273 if (po->num)
4274 register_prot_hook(sk);
4275 spin_unlock(&po->bind_lock);
4276 }
4277 break;
4278 }
4279 }
4280 rcu_read_unlock();
4281 return NOTIFY_DONE;
4282}
4283
4284
4285static int packet_ioctl(struct socket *sock, unsigned int cmd,
4286 unsigned long arg)
4287{
4288 struct sock *sk = sock->sk;
4289
4290 switch (cmd) {
4291 case SIOCOUTQ:
4292 {
4293 int amount = sk_wmem_alloc_get(sk);
4294
4295 return put_user(amount, (int __user *)arg);
4296 }
4297 case SIOCINQ:
4298 {
4299 struct sk_buff *skb;
4300 int amount = 0;
4301
4302 spin_lock_bh(&sk->sk_receive_queue.lock);
4303 skb = skb_peek(&sk->sk_receive_queue);
4304 if (skb)
4305 amount = skb->len;
4306 spin_unlock_bh(&sk->sk_receive_queue.lock);
4307 return put_user(amount, (int __user *)arg);
4308 }
4309#ifdef CONFIG_INET
4310 case SIOCADDRT:
4311 case SIOCDELRT:
4312 case SIOCDARP:
4313 case SIOCGARP:
4314 case SIOCSARP:
4315 case SIOCGIFADDR:
4316 case SIOCSIFADDR:
4317 case SIOCGIFBRDADDR:
4318 case SIOCSIFBRDADDR:
4319 case SIOCGIFNETMASK:
4320 case SIOCSIFNETMASK:
4321 case SIOCGIFDSTADDR:
4322 case SIOCSIFDSTADDR:
4323 case SIOCSIFFLAGS:
4324 return inet_dgram_ops.ioctl(sock, cmd, arg);
4325#endif
4326
4327 default:
4328 return -ENOIOCTLCMD;
4329 }
4330 return 0;
4331}
4332
4333static __poll_t packet_poll(struct file *file, struct socket *sock,
4334 poll_table *wait)
4335{
4336 struct sock *sk = sock->sk;
4337 struct packet_sock *po = pkt_sk(sk);
4338 __poll_t mask = datagram_poll(file, sock, wait);
4339
4340 spin_lock_bh(&sk->sk_receive_queue.lock);
4341 if (po->rx_ring.pg_vec) {
4342 if (!packet_previous_rx_frame(po, &po->rx_ring,
4343 TP_STATUS_KERNEL))
4344 mask |= EPOLLIN | EPOLLRDNORM;
4345 }
4346 packet_rcv_try_clear_pressure(po);
4347 spin_unlock_bh(&sk->sk_receive_queue.lock);
4348 spin_lock_bh(&sk->sk_write_queue.lock);
4349 if (po->tx_ring.pg_vec) {
4350 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4351 mask |= EPOLLOUT | EPOLLWRNORM;
4352 }
4353 spin_unlock_bh(&sk->sk_write_queue.lock);
4354 return mask;
4355}
4356
4357
4358/* Dirty? Well, I still did not learn better way to account
4359 * for user mmaps.
4360 */
4361
4362static void packet_mm_open(struct vm_area_struct *vma)
4363{
4364 struct file *file = vma->vm_file;
4365 struct socket *sock = file->private_data;
4366 struct sock *sk = sock->sk;
4367
4368 if (sk)
4369 atomic_long_inc(&pkt_sk(sk)->mapped);
4370}
4371
4372static void packet_mm_close(struct vm_area_struct *vma)
4373{
4374 struct file *file = vma->vm_file;
4375 struct socket *sock = file->private_data;
4376 struct sock *sk = sock->sk;
4377
4378 if (sk)
4379 atomic_long_dec(&pkt_sk(sk)->mapped);
4380}
4381
4382static const struct vm_operations_struct packet_mmap_ops = {
4383 .open = packet_mm_open,
4384 .close = packet_mm_close,
4385};
4386
4387static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4388 unsigned int len)
4389{
4390 int i;
4391
4392 for (i = 0; i < len; i++) {
4393 if (likely(pg_vec[i].buffer)) {
4394 if (is_vmalloc_addr(pg_vec[i].buffer))
4395 vfree(pg_vec[i].buffer);
4396 else
4397 free_pages((unsigned long)pg_vec[i].buffer,
4398 order);
4399 pg_vec[i].buffer = NULL;
4400 }
4401 }
4402 kfree(pg_vec);
4403}
4404
4405static char *alloc_one_pg_vec_page(unsigned long order)
4406{
4407 char *buffer;
4408 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4409 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4410
4411 buffer = (char *) __get_free_pages(gfp_flags, order);
4412 if (buffer)
4413 return buffer;
4414
4415 /* __get_free_pages failed, fall back to vmalloc */
4416 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4417 if (buffer)
4418 return buffer;
4419
4420 /* vmalloc failed, lets dig into swap here */
4421 gfp_flags &= ~__GFP_NORETRY;
4422 buffer = (char *) __get_free_pages(gfp_flags, order);
4423 if (buffer)
4424 return buffer;
4425
4426 /* complete and utter failure */
4427 return NULL;
4428}
4429
4430static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4431{
4432 unsigned int block_nr = req->tp_block_nr;
4433 struct pgv *pg_vec;
4434 int i;
4435
4436 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4437 if (unlikely(!pg_vec))
4438 goto out;
4439
4440 for (i = 0; i < block_nr; i++) {
4441 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4442 if (unlikely(!pg_vec[i].buffer))
4443 goto out_free_pgvec;
4444 }
4445
4446out:
4447 return pg_vec;
4448
4449out_free_pgvec:
4450 free_pg_vec(pg_vec, order, block_nr);
4451 pg_vec = NULL;
4452 goto out;
4453}
4454
4455static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4456 int closing, int tx_ring)
4457{
4458 struct pgv *pg_vec = NULL;
4459 struct packet_sock *po = pkt_sk(sk);
4460 unsigned long *rx_owner_map = NULL;
4461 int was_running, order = 0;
4462 struct packet_ring_buffer *rb;
4463 struct sk_buff_head *rb_queue;
4464 __be16 num;
4465 int err;
4466 /* Added to avoid minimal code churn */
4467 struct tpacket_req *req = &req_u->req;
4468
4469 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4470 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4471
4472 err = -EBUSY;
4473 if (!closing) {
4474 if (atomic_long_read(&po->mapped))
4475 goto out;
4476 if (packet_read_pending(rb))
4477 goto out;
4478 }
4479
4480 if (req->tp_block_nr) {
4481 unsigned int min_frame_size;
4482
4483 /* Sanity tests and some calculations */
4484 err = -EBUSY;
4485 if (unlikely(rb->pg_vec))
4486 goto out;
4487
4488 switch (po->tp_version) {
4489 case TPACKET_V1:
4490 po->tp_hdrlen = TPACKET_HDRLEN;
4491 break;
4492 case TPACKET_V2:
4493 po->tp_hdrlen = TPACKET2_HDRLEN;
4494 break;
4495 case TPACKET_V3:
4496 po->tp_hdrlen = TPACKET3_HDRLEN;
4497 break;
4498 }
4499
4500 err = -EINVAL;
4501 if (unlikely((int)req->tp_block_size <= 0))
4502 goto out;
4503 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4504 goto out;
4505 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4506 if (po->tp_version >= TPACKET_V3 &&
4507 req->tp_block_size <
4508 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4509 goto out;
4510 if (unlikely(req->tp_frame_size < min_frame_size))
4511 goto out;
4512 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4513 goto out;
4514
4515 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4516 if (unlikely(rb->frames_per_block == 0))
4517 goto out;
4518 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4519 goto out;
4520 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4521 req->tp_frame_nr))
4522 goto out;
4523
4524 err = -ENOMEM;
4525 order = get_order(req->tp_block_size);
4526 pg_vec = alloc_pg_vec(req, order);
4527 if (unlikely(!pg_vec))
4528 goto out;
4529 switch (po->tp_version) {
4530 case TPACKET_V3:
4531 /* Block transmit is not supported yet */
4532 if (!tx_ring) {
4533 init_prb_bdqc(po, rb, pg_vec, req_u);
4534 } else {
4535 struct tpacket_req3 *req3 = &req_u->req3;
4536
4537 if (req3->tp_retire_blk_tov ||
4538 req3->tp_sizeof_priv ||
4539 req3->tp_feature_req_word) {
4540 err = -EINVAL;
4541 goto out_free_pg_vec;
4542 }
4543 }
4544 break;
4545 default:
4546 if (!tx_ring) {
4547 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4548 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4549 if (!rx_owner_map)
4550 goto out_free_pg_vec;
4551 }
4552 break;
4553 }
4554 }
4555 /* Done */
4556 else {
4557 err = -EINVAL;
4558 if (unlikely(req->tp_frame_nr))
4559 goto out;
4560 }
4561
4562
4563 /* Detach socket from network */
4564 spin_lock(&po->bind_lock);
4565 was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4566 num = po->num;
4567 if (was_running) {
4568 WRITE_ONCE(po->num, 0);
4569 __unregister_prot_hook(sk, false);
4570 }
4571 spin_unlock(&po->bind_lock);
4572
4573 synchronize_net();
4574
4575 err = -EBUSY;
4576 mutex_lock(&po->pg_vec_lock);
4577 if (closing || atomic_long_read(&po->mapped) == 0) {
4578 err = 0;
4579 spin_lock_bh(&rb_queue->lock);
4580 swap(rb->pg_vec, pg_vec);
4581 if (po->tp_version <= TPACKET_V2)
4582 swap(rb->rx_owner_map, rx_owner_map);
4583 rb->frame_max = (req->tp_frame_nr - 1);
4584 rb->head = 0;
4585 rb->frame_size = req->tp_frame_size;
4586 spin_unlock_bh(&rb_queue->lock);
4587
4588 swap(rb->pg_vec_order, order);
4589 swap(rb->pg_vec_len, req->tp_block_nr);
4590
4591 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4592 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4593 tpacket_rcv : packet_rcv;
4594 skb_queue_purge(rb_queue);
4595 if (atomic_long_read(&po->mapped))
4596 pr_err("packet_mmap: vma is busy: %ld\n",
4597 atomic_long_read(&po->mapped));
4598 }
4599 mutex_unlock(&po->pg_vec_lock);
4600
4601 spin_lock(&po->bind_lock);
4602 if (was_running) {
4603 WRITE_ONCE(po->num, num);
4604 register_prot_hook(sk);
4605 }
4606 spin_unlock(&po->bind_lock);
4607 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4608 /* Because we don't support block-based V3 on tx-ring */
4609 if (!tx_ring)
4610 prb_shutdown_retire_blk_timer(po, rb_queue);
4611 }
4612
4613out_free_pg_vec:
4614 if (pg_vec) {
4615 bitmap_free(rx_owner_map);
4616 free_pg_vec(pg_vec, order, req->tp_block_nr);
4617 }
4618out:
4619 return err;
4620}
4621
4622static int packet_mmap(struct file *file, struct socket *sock,
4623 struct vm_area_struct *vma)
4624{
4625 struct sock *sk = sock->sk;
4626 struct packet_sock *po = pkt_sk(sk);
4627 unsigned long size, expected_size;
4628 struct packet_ring_buffer *rb;
4629 unsigned long start;
4630 int err = -EINVAL;
4631 int i;
4632
4633 if (vma->vm_pgoff)
4634 return -EINVAL;
4635
4636 mutex_lock(&po->pg_vec_lock);
4637
4638 expected_size = 0;
4639 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4640 if (rb->pg_vec) {
4641 expected_size += rb->pg_vec_len
4642 * rb->pg_vec_pages
4643 * PAGE_SIZE;
4644 }
4645 }
4646
4647 if (expected_size == 0)
4648 goto out;
4649
4650 size = vma->vm_end - vma->vm_start;
4651 if (size != expected_size)
4652 goto out;
4653
4654 start = vma->vm_start;
4655 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4656 if (rb->pg_vec == NULL)
4657 continue;
4658
4659 for (i = 0; i < rb->pg_vec_len; i++) {
4660 struct page *page;
4661 void *kaddr = rb->pg_vec[i].buffer;
4662 int pg_num;
4663
4664 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4665 page = pgv_to_page(kaddr);
4666 err = vm_insert_page(vma, start, page);
4667 if (unlikely(err))
4668 goto out;
4669 start += PAGE_SIZE;
4670 kaddr += PAGE_SIZE;
4671 }
4672 }
4673 }
4674
4675 atomic_long_inc(&po->mapped);
4676 vma->vm_ops = &packet_mmap_ops;
4677 err = 0;
4678
4679out:
4680 mutex_unlock(&po->pg_vec_lock);
4681 return err;
4682}
4683
4684static const struct proto_ops packet_ops_spkt = {
4685 .family = PF_PACKET,
4686 .owner = THIS_MODULE,
4687 .release = packet_release,
4688 .bind = packet_bind_spkt,
4689 .connect = sock_no_connect,
4690 .socketpair = sock_no_socketpair,
4691 .accept = sock_no_accept,
4692 .getname = packet_getname_spkt,
4693 .poll = datagram_poll,
4694 .ioctl = packet_ioctl,
4695 .gettstamp = sock_gettstamp,
4696 .listen = sock_no_listen,
4697 .shutdown = sock_no_shutdown,
4698 .sendmsg = packet_sendmsg_spkt,
4699 .recvmsg = packet_recvmsg,
4700 .mmap = sock_no_mmap,
4701};
4702
4703static const struct proto_ops packet_ops = {
4704 .family = PF_PACKET,
4705 .owner = THIS_MODULE,
4706 .release = packet_release,
4707 .bind = packet_bind,
4708 .connect = sock_no_connect,
4709 .socketpair = sock_no_socketpair,
4710 .accept = sock_no_accept,
4711 .getname = packet_getname,
4712 .poll = packet_poll,
4713 .ioctl = packet_ioctl,
4714 .gettstamp = sock_gettstamp,
4715 .listen = sock_no_listen,
4716 .shutdown = sock_no_shutdown,
4717 .setsockopt = packet_setsockopt,
4718 .getsockopt = packet_getsockopt,
4719 .sendmsg = packet_sendmsg,
4720 .recvmsg = packet_recvmsg,
4721 .mmap = packet_mmap,
4722};
4723
4724static const struct net_proto_family packet_family_ops = {
4725 .family = PF_PACKET,
4726 .create = packet_create,
4727 .owner = THIS_MODULE,
4728};
4729
4730static struct notifier_block packet_netdev_notifier = {
4731 .notifier_call = packet_notifier,
4732};
4733
4734#ifdef CONFIG_PROC_FS
4735
4736static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4737 __acquires(RCU)
4738{
4739 struct net *net = seq_file_net(seq);
4740
4741 rcu_read_lock();
4742 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4743}
4744
4745static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4746{
4747 struct net *net = seq_file_net(seq);
4748 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4749}
4750
4751static void packet_seq_stop(struct seq_file *seq, void *v)
4752 __releases(RCU)
4753{
4754 rcu_read_unlock();
4755}
4756
4757static int packet_seq_show(struct seq_file *seq, void *v)
4758{
4759 if (v == SEQ_START_TOKEN)
4760 seq_printf(seq,
4761 "%*sRefCnt Type Proto Iface R Rmem User Inode\n",
4762 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4763 else {
4764 struct sock *s = sk_entry(v);
4765 const struct packet_sock *po = pkt_sk(s);
4766
4767 seq_printf(seq,
4768 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4769 s,
4770 refcount_read(&s->sk_refcnt),
4771 s->sk_type,
4772 ntohs(READ_ONCE(po->num)),
4773 READ_ONCE(po->ifindex),
4774 packet_sock_flag(po, PACKET_SOCK_RUNNING),
4775 atomic_read(&s->sk_rmem_alloc),
4776 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4777 sock_i_ino(s));
4778 }
4779
4780 return 0;
4781}
4782
4783static const struct seq_operations packet_seq_ops = {
4784 .start = packet_seq_start,
4785 .next = packet_seq_next,
4786 .stop = packet_seq_stop,
4787 .show = packet_seq_show,
4788};
4789#endif
4790
4791static int __net_init packet_net_init(struct net *net)
4792{
4793 mutex_init(&net->packet.sklist_lock);
4794 INIT_HLIST_HEAD(&net->packet.sklist);
4795
4796#ifdef CONFIG_PROC_FS
4797 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4798 sizeof(struct seq_net_private)))
4799 return -ENOMEM;
4800#endif /* CONFIG_PROC_FS */
4801
4802 return 0;
4803}
4804
4805static void __net_exit packet_net_exit(struct net *net)
4806{
4807 remove_proc_entry("packet", net->proc_net);
4808 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4809}
4810
4811static struct pernet_operations packet_net_ops = {
4812 .init = packet_net_init,
4813 .exit = packet_net_exit,
4814};
4815
4816
4817static void __exit packet_exit(void)
4818{
4819 sock_unregister(PF_PACKET);
4820 proto_unregister(&packet_proto);
4821 unregister_netdevice_notifier(&packet_netdev_notifier);
4822 unregister_pernet_subsys(&packet_net_ops);
4823}
4824
4825static int __init packet_init(void)
4826{
4827 int rc;
4828
4829 rc = register_pernet_subsys(&packet_net_ops);
4830 if (rc)
4831 goto out;
4832 rc = register_netdevice_notifier(&packet_netdev_notifier);
4833 if (rc)
4834 goto out_pernet;
4835 rc = proto_register(&packet_proto, 0);
4836 if (rc)
4837 goto out_notifier;
4838 rc = sock_register(&packet_family_ops);
4839 if (rc)
4840 goto out_proto;
4841
4842 return 0;
4843
4844out_proto:
4845 proto_unregister(&packet_proto);
4846out_notifier:
4847 unregister_netdevice_notifier(&packet_netdev_notifier);
4848out_pernet:
4849 unregister_pernet_subsys(&packet_net_ops);
4850out:
4851 return rc;
4852}
4853
4854module_init(packet_init);
4855module_exit(packet_exit);
4856MODULE_DESCRIPTION("Packet socket support (AF_PACKET)");
4857MODULE_LICENSE("GPL");
4858MODULE_ALIAS_NETPROTO(PF_PACKET);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * PACKET - implements raw packet sockets.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 *
13 * Fixes:
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
42 * and packet_mreq.
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * layer.
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
47 */
48
49#include <linux/types.h>
50#include <linux/mm.h>
51#include <linux/capability.h>
52#include <linux/fcntl.h>
53#include <linux/socket.h>
54#include <linux/in.h>
55#include <linux/inet.h>
56#include <linux/netdevice.h>
57#include <linux/if_packet.h>
58#include <linux/wireless.h>
59#include <linux/kernel.h>
60#include <linux/kmod.h>
61#include <linux/slab.h>
62#include <linux/vmalloc.h>
63#include <net/net_namespace.h>
64#include <net/ip.h>
65#include <net/protocol.h>
66#include <linux/skbuff.h>
67#include <net/sock.h>
68#include <linux/errno.h>
69#include <linux/timer.h>
70#include <linux/uaccess.h>
71#include <asm/ioctls.h>
72#include <asm/page.h>
73#include <asm/cacheflush.h>
74#include <asm/io.h>
75#include <linux/proc_fs.h>
76#include <linux/seq_file.h>
77#include <linux/poll.h>
78#include <linux/module.h>
79#include <linux/init.h>
80#include <linux/mutex.h>
81#include <linux/if_vlan.h>
82#include <linux/virtio_net.h>
83#include <linux/errqueue.h>
84#include <linux/net_tstamp.h>
85#include <linux/percpu.h>
86#ifdef CONFIG_INET
87#include <net/inet_common.h>
88#endif
89#include <linux/bpf.h>
90#include <net/compat.h>
91
92#include "internal.h"
93
94/*
95 Assumptions:
96 - if device has no dev->hard_header routine, it adds and removes ll header
97 inside itself. In this case ll header is invisible outside of device,
98 but higher levels still should reserve dev->hard_header_len.
99 Some devices are enough clever to reallocate skb, when header
100 will not fit to reserved space (tunnel), another ones are silly
101 (PPP).
102 - packet socket receives packets with pulled ll header,
103 so that SOCK_RAW should push it back.
104
105On receive:
106-----------
107
108Incoming, dev->hard_header!=NULL
109 mac_header -> ll header
110 data -> data
111
112Outgoing, dev->hard_header!=NULL
113 mac_header -> ll header
114 data -> ll header
115
116Incoming, dev->hard_header==NULL
117 mac_header -> UNKNOWN position. It is very likely, that it points to ll
118 header. PPP makes it, that is wrong, because introduce
119 assymetry between rx and tx paths.
120 data -> data
121
122Outgoing, dev->hard_header==NULL
123 mac_header -> data. ll header is still not built!
124 data -> data
125
126Resume
127 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
128
129
130On transmit:
131------------
132
133dev->hard_header != NULL
134 mac_header -> ll header
135 data -> ll header
136
137dev->hard_header == NULL (ll header is added by device, we cannot control it)
138 mac_header -> data
139 data -> data
140
141 We should set nh.raw on output to correct posistion,
142 packet classifier depends on it.
143 */
144
145/* Private packet socket structures. */
146
147/* identical to struct packet_mreq except it has
148 * a longer address field.
149 */
150struct packet_mreq_max {
151 int mr_ifindex;
152 unsigned short mr_type;
153 unsigned short mr_alen;
154 unsigned char mr_address[MAX_ADDR_LEN];
155};
156
157union tpacket_uhdr {
158 struct tpacket_hdr *h1;
159 struct tpacket2_hdr *h2;
160 struct tpacket3_hdr *h3;
161 void *raw;
162};
163
164static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
165 int closing, int tx_ring);
166
167#define V3_ALIGNMENT (8)
168
169#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
170
171#define BLK_PLUS_PRIV(sz_of_priv) \
172 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
173
174#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
175#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
176#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
177#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
178#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
179#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
180#define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
181
182struct packet_sock;
183static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
184 struct packet_type *pt, struct net_device *orig_dev);
185
186static void *packet_previous_frame(struct packet_sock *po,
187 struct packet_ring_buffer *rb,
188 int status);
189static void packet_increment_head(struct packet_ring_buffer *buff);
190static int prb_curr_blk_in_use(struct tpacket_block_desc *);
191static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
192 struct packet_sock *);
193static void prb_retire_current_block(struct tpacket_kbdq_core *,
194 struct packet_sock *, unsigned int status);
195static int prb_queue_frozen(struct tpacket_kbdq_core *);
196static void prb_open_block(struct tpacket_kbdq_core *,
197 struct tpacket_block_desc *);
198static void prb_retire_rx_blk_timer_expired(struct timer_list *);
199static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
200static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
201static void prb_clear_rxhash(struct tpacket_kbdq_core *,
202 struct tpacket3_hdr *);
203static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
204 struct tpacket3_hdr *);
205static void packet_flush_mclist(struct sock *sk);
206static u16 packet_pick_tx_queue(struct sk_buff *skb);
207
208struct packet_skb_cb {
209 union {
210 struct sockaddr_pkt pkt;
211 union {
212 /* Trick: alias skb original length with
213 * ll.sll_family and ll.protocol in order
214 * to save room.
215 */
216 unsigned int origlen;
217 struct sockaddr_ll ll;
218 };
219 } sa;
220};
221
222#define vio_le() virtio_legacy_is_little_endian()
223
224#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
225
226#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
227#define GET_PBLOCK_DESC(x, bid) \
228 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
229#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
230 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
231#define GET_NEXT_PRB_BLK_NUM(x) \
232 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
233 ((x)->kactive_blk_num+1) : 0)
234
235static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
236static void __fanout_link(struct sock *sk, struct packet_sock *po);
237
238static int packet_direct_xmit(struct sk_buff *skb)
239{
240 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
241}
242
243static struct net_device *packet_cached_dev_get(struct packet_sock *po)
244{
245 struct net_device *dev;
246
247 rcu_read_lock();
248 dev = rcu_dereference(po->cached_dev);
249 if (likely(dev))
250 dev_hold(dev);
251 rcu_read_unlock();
252
253 return dev;
254}
255
256static void packet_cached_dev_assign(struct packet_sock *po,
257 struct net_device *dev)
258{
259 rcu_assign_pointer(po->cached_dev, dev);
260}
261
262static void packet_cached_dev_reset(struct packet_sock *po)
263{
264 RCU_INIT_POINTER(po->cached_dev, NULL);
265}
266
267static bool packet_use_direct_xmit(const struct packet_sock *po)
268{
269 return po->xmit == packet_direct_xmit;
270}
271
272static u16 packet_pick_tx_queue(struct sk_buff *skb)
273{
274 struct net_device *dev = skb->dev;
275 const struct net_device_ops *ops = dev->netdev_ops;
276 int cpu = raw_smp_processor_id();
277 u16 queue_index;
278
279#ifdef CONFIG_XPS
280 skb->sender_cpu = cpu + 1;
281#endif
282 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
283 if (ops->ndo_select_queue) {
284 queue_index = ops->ndo_select_queue(dev, skb, NULL);
285 queue_index = netdev_cap_txqueue(dev, queue_index);
286 } else {
287 queue_index = netdev_pick_tx(dev, skb, NULL);
288 }
289
290 return queue_index;
291}
292
293/* __register_prot_hook must be invoked through register_prot_hook
294 * or from a context in which asynchronous accesses to the packet
295 * socket is not possible (packet_create()).
296 */
297static void __register_prot_hook(struct sock *sk)
298{
299 struct packet_sock *po = pkt_sk(sk);
300
301 if (!po->running) {
302 if (po->fanout)
303 __fanout_link(sk, po);
304 else
305 dev_add_pack(&po->prot_hook);
306
307 sock_hold(sk);
308 po->running = 1;
309 }
310}
311
312static void register_prot_hook(struct sock *sk)
313{
314 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
315 __register_prot_hook(sk);
316}
317
318/* If the sync parameter is true, we will temporarily drop
319 * the po->bind_lock and do a synchronize_net to make sure no
320 * asynchronous packet processing paths still refer to the elements
321 * of po->prot_hook. If the sync parameter is false, it is the
322 * callers responsibility to take care of this.
323 */
324static void __unregister_prot_hook(struct sock *sk, bool sync)
325{
326 struct packet_sock *po = pkt_sk(sk);
327
328 lockdep_assert_held_once(&po->bind_lock);
329
330 po->running = 0;
331
332 if (po->fanout)
333 __fanout_unlink(sk, po);
334 else
335 __dev_remove_pack(&po->prot_hook);
336
337 __sock_put(sk);
338
339 if (sync) {
340 spin_unlock(&po->bind_lock);
341 synchronize_net();
342 spin_lock(&po->bind_lock);
343 }
344}
345
346static void unregister_prot_hook(struct sock *sk, bool sync)
347{
348 struct packet_sock *po = pkt_sk(sk);
349
350 if (po->running)
351 __unregister_prot_hook(sk, sync);
352}
353
354static inline struct page * __pure pgv_to_page(void *addr)
355{
356 if (is_vmalloc_addr(addr))
357 return vmalloc_to_page(addr);
358 return virt_to_page(addr);
359}
360
361static void __packet_set_status(struct packet_sock *po, void *frame, int status)
362{
363 union tpacket_uhdr h;
364
365 h.raw = frame;
366 switch (po->tp_version) {
367 case TPACKET_V1:
368 h.h1->tp_status = status;
369 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
370 break;
371 case TPACKET_V2:
372 h.h2->tp_status = status;
373 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
374 break;
375 case TPACKET_V3:
376 h.h3->tp_status = status;
377 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
378 break;
379 default:
380 WARN(1, "TPACKET version not supported.\n");
381 BUG();
382 }
383
384 smp_wmb();
385}
386
387static int __packet_get_status(const struct packet_sock *po, void *frame)
388{
389 union tpacket_uhdr h;
390
391 smp_rmb();
392
393 h.raw = frame;
394 switch (po->tp_version) {
395 case TPACKET_V1:
396 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
397 return h.h1->tp_status;
398 case TPACKET_V2:
399 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
400 return h.h2->tp_status;
401 case TPACKET_V3:
402 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
403 return h.h3->tp_status;
404 default:
405 WARN(1, "TPACKET version not supported.\n");
406 BUG();
407 return 0;
408 }
409}
410
411static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
412 unsigned int flags)
413{
414 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
415
416 if (shhwtstamps &&
417 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
418 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
419 return TP_STATUS_TS_RAW_HARDWARE;
420
421 if (ktime_to_timespec64_cond(skb->tstamp, ts))
422 return TP_STATUS_TS_SOFTWARE;
423
424 return 0;
425}
426
427static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
428 struct sk_buff *skb)
429{
430 union tpacket_uhdr h;
431 struct timespec64 ts;
432 __u32 ts_status;
433
434 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
435 return 0;
436
437 h.raw = frame;
438 /*
439 * versions 1 through 3 overflow the timestamps in y2106, since they
440 * all store the seconds in a 32-bit unsigned integer.
441 * If we create a version 4, that should have a 64-bit timestamp,
442 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
443 * nanoseconds.
444 */
445 switch (po->tp_version) {
446 case TPACKET_V1:
447 h.h1->tp_sec = ts.tv_sec;
448 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
449 break;
450 case TPACKET_V2:
451 h.h2->tp_sec = ts.tv_sec;
452 h.h2->tp_nsec = ts.tv_nsec;
453 break;
454 case TPACKET_V3:
455 h.h3->tp_sec = ts.tv_sec;
456 h.h3->tp_nsec = ts.tv_nsec;
457 break;
458 default:
459 WARN(1, "TPACKET version not supported.\n");
460 BUG();
461 }
462
463 /* one flush is safe, as both fields always lie on the same cacheline */
464 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
465 smp_wmb();
466
467 return ts_status;
468}
469
470static void *packet_lookup_frame(const struct packet_sock *po,
471 const struct packet_ring_buffer *rb,
472 unsigned int position,
473 int status)
474{
475 unsigned int pg_vec_pos, frame_offset;
476 union tpacket_uhdr h;
477
478 pg_vec_pos = position / rb->frames_per_block;
479 frame_offset = position % rb->frames_per_block;
480
481 h.raw = rb->pg_vec[pg_vec_pos].buffer +
482 (frame_offset * rb->frame_size);
483
484 if (status != __packet_get_status(po, h.raw))
485 return NULL;
486
487 return h.raw;
488}
489
490static void *packet_current_frame(struct packet_sock *po,
491 struct packet_ring_buffer *rb,
492 int status)
493{
494 return packet_lookup_frame(po, rb, rb->head, status);
495}
496
497static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
498{
499 del_timer_sync(&pkc->retire_blk_timer);
500}
501
502static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
503 struct sk_buff_head *rb_queue)
504{
505 struct tpacket_kbdq_core *pkc;
506
507 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
508
509 spin_lock_bh(&rb_queue->lock);
510 pkc->delete_blk_timer = 1;
511 spin_unlock_bh(&rb_queue->lock);
512
513 prb_del_retire_blk_timer(pkc);
514}
515
516static void prb_setup_retire_blk_timer(struct packet_sock *po)
517{
518 struct tpacket_kbdq_core *pkc;
519
520 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
521 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
522 0);
523 pkc->retire_blk_timer.expires = jiffies;
524}
525
526static int prb_calc_retire_blk_tmo(struct packet_sock *po,
527 int blk_size_in_bytes)
528{
529 struct net_device *dev;
530 unsigned int mbits, div;
531 struct ethtool_link_ksettings ecmd;
532 int err;
533
534 rtnl_lock();
535 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
536 if (unlikely(!dev)) {
537 rtnl_unlock();
538 return DEFAULT_PRB_RETIRE_TOV;
539 }
540 err = __ethtool_get_link_ksettings(dev, &ecmd);
541 rtnl_unlock();
542 if (err)
543 return DEFAULT_PRB_RETIRE_TOV;
544
545 /* If the link speed is so slow you don't really
546 * need to worry about perf anyways
547 */
548 if (ecmd.base.speed < SPEED_1000 ||
549 ecmd.base.speed == SPEED_UNKNOWN)
550 return DEFAULT_PRB_RETIRE_TOV;
551
552 div = ecmd.base.speed / 1000;
553 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
554
555 if (div)
556 mbits /= div;
557
558 if (div)
559 return mbits + 1;
560 return mbits;
561}
562
563static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
564 union tpacket_req_u *req_u)
565{
566 p1->feature_req_word = req_u->req3.tp_feature_req_word;
567}
568
569static void init_prb_bdqc(struct packet_sock *po,
570 struct packet_ring_buffer *rb,
571 struct pgv *pg_vec,
572 union tpacket_req_u *req_u)
573{
574 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
575 struct tpacket_block_desc *pbd;
576
577 memset(p1, 0x0, sizeof(*p1));
578
579 p1->knxt_seq_num = 1;
580 p1->pkbdq = pg_vec;
581 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
582 p1->pkblk_start = pg_vec[0].buffer;
583 p1->kblk_size = req_u->req3.tp_block_size;
584 p1->knum_blocks = req_u->req3.tp_block_nr;
585 p1->hdrlen = po->tp_hdrlen;
586 p1->version = po->tp_version;
587 p1->last_kactive_blk_num = 0;
588 po->stats.stats3.tp_freeze_q_cnt = 0;
589 if (req_u->req3.tp_retire_blk_tov)
590 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
591 else
592 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
593 req_u->req3.tp_block_size);
594 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
595 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
596 rwlock_init(&p1->blk_fill_in_prog_lock);
597
598 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
599 prb_init_ft_ops(p1, req_u);
600 prb_setup_retire_blk_timer(po);
601 prb_open_block(p1, pbd);
602}
603
604/* Do NOT update the last_blk_num first.
605 * Assumes sk_buff_head lock is held.
606 */
607static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
608{
609 mod_timer(&pkc->retire_blk_timer,
610 jiffies + pkc->tov_in_jiffies);
611 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
612}
613
614/*
615 * Timer logic:
616 * 1) We refresh the timer only when we open a block.
617 * By doing this we don't waste cycles refreshing the timer
618 * on packet-by-packet basis.
619 *
620 * With a 1MB block-size, on a 1Gbps line, it will take
621 * i) ~8 ms to fill a block + ii) memcpy etc.
622 * In this cut we are not accounting for the memcpy time.
623 *
624 * So, if the user sets the 'tmo' to 10ms then the timer
625 * will never fire while the block is still getting filled
626 * (which is what we want). However, the user could choose
627 * to close a block early and that's fine.
628 *
629 * But when the timer does fire, we check whether or not to refresh it.
630 * Since the tmo granularity is in msecs, it is not too expensive
631 * to refresh the timer, lets say every '8' msecs.
632 * Either the user can set the 'tmo' or we can derive it based on
633 * a) line-speed and b) block-size.
634 * prb_calc_retire_blk_tmo() calculates the tmo.
635 *
636 */
637static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
638{
639 struct packet_sock *po =
640 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
641 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
642 unsigned int frozen;
643 struct tpacket_block_desc *pbd;
644
645 spin_lock(&po->sk.sk_receive_queue.lock);
646
647 frozen = prb_queue_frozen(pkc);
648 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
649
650 if (unlikely(pkc->delete_blk_timer))
651 goto out;
652
653 /* We only need to plug the race when the block is partially filled.
654 * tpacket_rcv:
655 * lock(); increment BLOCK_NUM_PKTS; unlock()
656 * copy_bits() is in progress ...
657 * timer fires on other cpu:
658 * we can't retire the current block because copy_bits
659 * is in progress.
660 *
661 */
662 if (BLOCK_NUM_PKTS(pbd)) {
663 /* Waiting for skb_copy_bits to finish... */
664 write_lock(&pkc->blk_fill_in_prog_lock);
665 write_unlock(&pkc->blk_fill_in_prog_lock);
666 }
667
668 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
669 if (!frozen) {
670 if (!BLOCK_NUM_PKTS(pbd)) {
671 /* An empty block. Just refresh the timer. */
672 goto refresh_timer;
673 }
674 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
675 if (!prb_dispatch_next_block(pkc, po))
676 goto refresh_timer;
677 else
678 goto out;
679 } else {
680 /* Case 1. Queue was frozen because user-space was
681 * lagging behind.
682 */
683 if (prb_curr_blk_in_use(pbd)) {
684 /*
685 * Ok, user-space is still behind.
686 * So just refresh the timer.
687 */
688 goto refresh_timer;
689 } else {
690 /* Case 2. queue was frozen,user-space caught up,
691 * now the link went idle && the timer fired.
692 * We don't have a block to close.So we open this
693 * block and restart the timer.
694 * opening a block thaws the queue,restarts timer
695 * Thawing/timer-refresh is a side effect.
696 */
697 prb_open_block(pkc, pbd);
698 goto out;
699 }
700 }
701 }
702
703refresh_timer:
704 _prb_refresh_rx_retire_blk_timer(pkc);
705
706out:
707 spin_unlock(&po->sk.sk_receive_queue.lock);
708}
709
710static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
711 struct tpacket_block_desc *pbd1, __u32 status)
712{
713 /* Flush everything minus the block header */
714
715#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
716 u8 *start, *end;
717
718 start = (u8 *)pbd1;
719
720 /* Skip the block header(we know header WILL fit in 4K) */
721 start += PAGE_SIZE;
722
723 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
724 for (; start < end; start += PAGE_SIZE)
725 flush_dcache_page(pgv_to_page(start));
726
727 smp_wmb();
728#endif
729
730 /* Now update the block status. */
731
732 BLOCK_STATUS(pbd1) = status;
733
734 /* Flush the block header */
735
736#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
737 start = (u8 *)pbd1;
738 flush_dcache_page(pgv_to_page(start));
739
740 smp_wmb();
741#endif
742}
743
744/*
745 * Side effect:
746 *
747 * 1) flush the block
748 * 2) Increment active_blk_num
749 *
750 * Note:We DONT refresh the timer on purpose.
751 * Because almost always the next block will be opened.
752 */
753static void prb_close_block(struct tpacket_kbdq_core *pkc1,
754 struct tpacket_block_desc *pbd1,
755 struct packet_sock *po, unsigned int stat)
756{
757 __u32 status = TP_STATUS_USER | stat;
758
759 struct tpacket3_hdr *last_pkt;
760 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
761 struct sock *sk = &po->sk;
762
763 if (atomic_read(&po->tp_drops))
764 status |= TP_STATUS_LOSING;
765
766 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
767 last_pkt->tp_next_offset = 0;
768
769 /* Get the ts of the last pkt */
770 if (BLOCK_NUM_PKTS(pbd1)) {
771 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
772 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
773 } else {
774 /* Ok, we tmo'd - so get the current time.
775 *
776 * It shouldn't really happen as we don't close empty
777 * blocks. See prb_retire_rx_blk_timer_expired().
778 */
779 struct timespec64 ts;
780 ktime_get_real_ts64(&ts);
781 h1->ts_last_pkt.ts_sec = ts.tv_sec;
782 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
783 }
784
785 smp_wmb();
786
787 /* Flush the block */
788 prb_flush_block(pkc1, pbd1, status);
789
790 sk->sk_data_ready(sk);
791
792 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
793}
794
795static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
796{
797 pkc->reset_pending_on_curr_blk = 0;
798}
799
800/*
801 * Side effect of opening a block:
802 *
803 * 1) prb_queue is thawed.
804 * 2) retire_blk_timer is refreshed.
805 *
806 */
807static void prb_open_block(struct tpacket_kbdq_core *pkc1,
808 struct tpacket_block_desc *pbd1)
809{
810 struct timespec64 ts;
811 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
812
813 smp_rmb();
814
815 /* We could have just memset this but we will lose the
816 * flexibility of making the priv area sticky
817 */
818
819 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
820 BLOCK_NUM_PKTS(pbd1) = 0;
821 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
822
823 ktime_get_real_ts64(&ts);
824
825 h1->ts_first_pkt.ts_sec = ts.tv_sec;
826 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
827
828 pkc1->pkblk_start = (char *)pbd1;
829 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830
831 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
832 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
833
834 pbd1->version = pkc1->version;
835 pkc1->prev = pkc1->nxt_offset;
836 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
837
838 prb_thaw_queue(pkc1);
839 _prb_refresh_rx_retire_blk_timer(pkc1);
840
841 smp_wmb();
842}
843
844/*
845 * Queue freeze logic:
846 * 1) Assume tp_block_nr = 8 blocks.
847 * 2) At time 't0', user opens Rx ring.
848 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
849 * 4) user-space is either sleeping or processing block '0'.
850 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
851 * it will close block-7,loop around and try to fill block '0'.
852 * call-flow:
853 * __packet_lookup_frame_in_block
854 * prb_retire_current_block()
855 * prb_dispatch_next_block()
856 * |->(BLOCK_STATUS == USER) evaluates to true
857 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
858 * 6) Now there are two cases:
859 * 6.1) Link goes idle right after the queue is frozen.
860 * But remember, the last open_block() refreshed the timer.
861 * When this timer expires,it will refresh itself so that we can
862 * re-open block-0 in near future.
863 * 6.2) Link is busy and keeps on receiving packets. This is a simple
864 * case and __packet_lookup_frame_in_block will check if block-0
865 * is free and can now be re-used.
866 */
867static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
868 struct packet_sock *po)
869{
870 pkc->reset_pending_on_curr_blk = 1;
871 po->stats.stats3.tp_freeze_q_cnt++;
872}
873
874#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
875
876/*
877 * If the next block is free then we will dispatch it
878 * and return a good offset.
879 * Else, we will freeze the queue.
880 * So, caller must check the return value.
881 */
882static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
883 struct packet_sock *po)
884{
885 struct tpacket_block_desc *pbd;
886
887 smp_rmb();
888
889 /* 1. Get current block num */
890 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
891
892 /* 2. If this block is currently in_use then freeze the queue */
893 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
894 prb_freeze_queue(pkc, po);
895 return NULL;
896 }
897
898 /*
899 * 3.
900 * open this block and return the offset where the first packet
901 * needs to get stored.
902 */
903 prb_open_block(pkc, pbd);
904 return (void *)pkc->nxt_offset;
905}
906
907static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
908 struct packet_sock *po, unsigned int status)
909{
910 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
911
912 /* retire/close the current block */
913 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
914 /*
915 * Plug the case where copy_bits() is in progress on
916 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
917 * have space to copy the pkt in the current block and
918 * called prb_retire_current_block()
919 *
920 * We don't need to worry about the TMO case because
921 * the timer-handler already handled this case.
922 */
923 if (!(status & TP_STATUS_BLK_TMO)) {
924 /* Waiting for skb_copy_bits to finish... */
925 write_lock(&pkc->blk_fill_in_prog_lock);
926 write_unlock(&pkc->blk_fill_in_prog_lock);
927 }
928 prb_close_block(pkc, pbd, po, status);
929 return;
930 }
931}
932
933static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
934{
935 return TP_STATUS_USER & BLOCK_STATUS(pbd);
936}
937
938static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
939{
940 return pkc->reset_pending_on_curr_blk;
941}
942
943static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
944 __releases(&pkc->blk_fill_in_prog_lock)
945{
946 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
947
948 read_unlock(&pkc->blk_fill_in_prog_lock);
949}
950
951static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
952 struct tpacket3_hdr *ppd)
953{
954 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
955}
956
957static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
958 struct tpacket3_hdr *ppd)
959{
960 ppd->hv1.tp_rxhash = 0;
961}
962
963static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
964 struct tpacket3_hdr *ppd)
965{
966 if (skb_vlan_tag_present(pkc->skb)) {
967 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
968 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
969 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
970 } else {
971 ppd->hv1.tp_vlan_tci = 0;
972 ppd->hv1.tp_vlan_tpid = 0;
973 ppd->tp_status = TP_STATUS_AVAILABLE;
974 }
975}
976
977static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
978 struct tpacket3_hdr *ppd)
979{
980 ppd->hv1.tp_padding = 0;
981 prb_fill_vlan_info(pkc, ppd);
982
983 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
984 prb_fill_rxhash(pkc, ppd);
985 else
986 prb_clear_rxhash(pkc, ppd);
987}
988
989static void prb_fill_curr_block(char *curr,
990 struct tpacket_kbdq_core *pkc,
991 struct tpacket_block_desc *pbd,
992 unsigned int len)
993 __acquires(&pkc->blk_fill_in_prog_lock)
994{
995 struct tpacket3_hdr *ppd;
996
997 ppd = (struct tpacket3_hdr *)curr;
998 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
999 pkc->prev = curr;
1000 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1001 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1002 BLOCK_NUM_PKTS(pbd) += 1;
1003 read_lock(&pkc->blk_fill_in_prog_lock);
1004 prb_run_all_ft_ops(pkc, ppd);
1005}
1006
1007/* Assumes caller has the sk->rx_queue.lock */
1008static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1009 struct sk_buff *skb,
1010 unsigned int len
1011 )
1012{
1013 struct tpacket_kbdq_core *pkc;
1014 struct tpacket_block_desc *pbd;
1015 char *curr, *end;
1016
1017 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1018 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1019
1020 /* Queue is frozen when user space is lagging behind */
1021 if (prb_queue_frozen(pkc)) {
1022 /*
1023 * Check if that last block which caused the queue to freeze,
1024 * is still in_use by user-space.
1025 */
1026 if (prb_curr_blk_in_use(pbd)) {
1027 /* Can't record this packet */
1028 return NULL;
1029 } else {
1030 /*
1031 * Ok, the block was released by user-space.
1032 * Now let's open that block.
1033 * opening a block also thaws the queue.
1034 * Thawing is a side effect.
1035 */
1036 prb_open_block(pkc, pbd);
1037 }
1038 }
1039
1040 smp_mb();
1041 curr = pkc->nxt_offset;
1042 pkc->skb = skb;
1043 end = (char *)pbd + pkc->kblk_size;
1044
1045 /* first try the current block */
1046 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1047 prb_fill_curr_block(curr, pkc, pbd, len);
1048 return (void *)curr;
1049 }
1050
1051 /* Ok, close the current block */
1052 prb_retire_current_block(pkc, po, 0);
1053
1054 /* Now, try to dispatch the next block */
1055 curr = (char *)prb_dispatch_next_block(pkc, po);
1056 if (curr) {
1057 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1058 prb_fill_curr_block(curr, pkc, pbd, len);
1059 return (void *)curr;
1060 }
1061
1062 /*
1063 * No free blocks are available.user_space hasn't caught up yet.
1064 * Queue was just frozen and now this packet will get dropped.
1065 */
1066 return NULL;
1067}
1068
1069static void *packet_current_rx_frame(struct packet_sock *po,
1070 struct sk_buff *skb,
1071 int status, unsigned int len)
1072{
1073 char *curr = NULL;
1074 switch (po->tp_version) {
1075 case TPACKET_V1:
1076 case TPACKET_V2:
1077 curr = packet_lookup_frame(po, &po->rx_ring,
1078 po->rx_ring.head, status);
1079 return curr;
1080 case TPACKET_V3:
1081 return __packet_lookup_frame_in_block(po, skb, len);
1082 default:
1083 WARN(1, "TPACKET version not supported\n");
1084 BUG();
1085 return NULL;
1086 }
1087}
1088
1089static void *prb_lookup_block(const struct packet_sock *po,
1090 const struct packet_ring_buffer *rb,
1091 unsigned int idx,
1092 int status)
1093{
1094 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1095 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1096
1097 if (status != BLOCK_STATUS(pbd))
1098 return NULL;
1099 return pbd;
1100}
1101
1102static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1103{
1104 unsigned int prev;
1105 if (rb->prb_bdqc.kactive_blk_num)
1106 prev = rb->prb_bdqc.kactive_blk_num-1;
1107 else
1108 prev = rb->prb_bdqc.knum_blocks-1;
1109 return prev;
1110}
1111
1112/* Assumes caller has held the rx_queue.lock */
1113static void *__prb_previous_block(struct packet_sock *po,
1114 struct packet_ring_buffer *rb,
1115 int status)
1116{
1117 unsigned int previous = prb_previous_blk_num(rb);
1118 return prb_lookup_block(po, rb, previous, status);
1119}
1120
1121static void *packet_previous_rx_frame(struct packet_sock *po,
1122 struct packet_ring_buffer *rb,
1123 int status)
1124{
1125 if (po->tp_version <= TPACKET_V2)
1126 return packet_previous_frame(po, rb, status);
1127
1128 return __prb_previous_block(po, rb, status);
1129}
1130
1131static void packet_increment_rx_head(struct packet_sock *po,
1132 struct packet_ring_buffer *rb)
1133{
1134 switch (po->tp_version) {
1135 case TPACKET_V1:
1136 case TPACKET_V2:
1137 return packet_increment_head(rb);
1138 case TPACKET_V3:
1139 default:
1140 WARN(1, "TPACKET version not supported.\n");
1141 BUG();
1142 return;
1143 }
1144}
1145
1146static void *packet_previous_frame(struct packet_sock *po,
1147 struct packet_ring_buffer *rb,
1148 int status)
1149{
1150 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1151 return packet_lookup_frame(po, rb, previous, status);
1152}
1153
1154static void packet_increment_head(struct packet_ring_buffer *buff)
1155{
1156 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1157}
1158
1159static void packet_inc_pending(struct packet_ring_buffer *rb)
1160{
1161 this_cpu_inc(*rb->pending_refcnt);
1162}
1163
1164static void packet_dec_pending(struct packet_ring_buffer *rb)
1165{
1166 this_cpu_dec(*rb->pending_refcnt);
1167}
1168
1169static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1170{
1171 unsigned int refcnt = 0;
1172 int cpu;
1173
1174 /* We don't use pending refcount in rx_ring. */
1175 if (rb->pending_refcnt == NULL)
1176 return 0;
1177
1178 for_each_possible_cpu(cpu)
1179 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1180
1181 return refcnt;
1182}
1183
1184static int packet_alloc_pending(struct packet_sock *po)
1185{
1186 po->rx_ring.pending_refcnt = NULL;
1187
1188 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1189 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1190 return -ENOBUFS;
1191
1192 return 0;
1193}
1194
1195static void packet_free_pending(struct packet_sock *po)
1196{
1197 free_percpu(po->tx_ring.pending_refcnt);
1198}
1199
1200#define ROOM_POW_OFF 2
1201#define ROOM_NONE 0x0
1202#define ROOM_LOW 0x1
1203#define ROOM_NORMAL 0x2
1204
1205static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1206{
1207 int idx, len;
1208
1209 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1210 idx = READ_ONCE(po->rx_ring.head);
1211 if (pow_off)
1212 idx += len >> pow_off;
1213 if (idx >= len)
1214 idx -= len;
1215 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1216}
1217
1218static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1219{
1220 int idx, len;
1221
1222 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1223 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1224 if (pow_off)
1225 idx += len >> pow_off;
1226 if (idx >= len)
1227 idx -= len;
1228 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1229}
1230
1231static int __packet_rcv_has_room(const struct packet_sock *po,
1232 const struct sk_buff *skb)
1233{
1234 const struct sock *sk = &po->sk;
1235 int ret = ROOM_NONE;
1236
1237 if (po->prot_hook.func != tpacket_rcv) {
1238 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1239 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1240 - (skb ? skb->truesize : 0);
1241
1242 if (avail > (rcvbuf >> ROOM_POW_OFF))
1243 return ROOM_NORMAL;
1244 else if (avail > 0)
1245 return ROOM_LOW;
1246 else
1247 return ROOM_NONE;
1248 }
1249
1250 if (po->tp_version == TPACKET_V3) {
1251 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1252 ret = ROOM_NORMAL;
1253 else if (__tpacket_v3_has_room(po, 0))
1254 ret = ROOM_LOW;
1255 } else {
1256 if (__tpacket_has_room(po, ROOM_POW_OFF))
1257 ret = ROOM_NORMAL;
1258 else if (__tpacket_has_room(po, 0))
1259 ret = ROOM_LOW;
1260 }
1261
1262 return ret;
1263}
1264
1265static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266{
1267 int pressure, ret;
1268
1269 ret = __packet_rcv_has_room(po, skb);
1270 pressure = ret != ROOM_NORMAL;
1271
1272 if (READ_ONCE(po->pressure) != pressure)
1273 WRITE_ONCE(po->pressure, pressure);
1274
1275 return ret;
1276}
1277
1278static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1279{
1280 if (READ_ONCE(po->pressure) &&
1281 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1282 WRITE_ONCE(po->pressure, 0);
1283}
1284
1285static void packet_sock_destruct(struct sock *sk)
1286{
1287 skb_queue_purge(&sk->sk_error_queue);
1288
1289 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1290 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1291
1292 if (!sock_flag(sk, SOCK_DEAD)) {
1293 pr_err("Attempt to release alive packet socket: %p\n", sk);
1294 return;
1295 }
1296
1297 sk_refcnt_debug_dec(sk);
1298}
1299
1300static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1301{
1302 u32 *history = po->rollover->history;
1303 u32 victim, rxhash;
1304 int i, count = 0;
1305
1306 rxhash = skb_get_hash(skb);
1307 for (i = 0; i < ROLLOVER_HLEN; i++)
1308 if (READ_ONCE(history[i]) == rxhash)
1309 count++;
1310
1311 victim = prandom_u32() % ROLLOVER_HLEN;
1312
1313 /* Avoid dirtying the cache line if possible */
1314 if (READ_ONCE(history[victim]) != rxhash)
1315 WRITE_ONCE(history[victim], rxhash);
1316
1317 return count > (ROLLOVER_HLEN >> 1);
1318}
1319
1320static unsigned int fanout_demux_hash(struct packet_fanout *f,
1321 struct sk_buff *skb,
1322 unsigned int num)
1323{
1324 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1325}
1326
1327static unsigned int fanout_demux_lb(struct packet_fanout *f,
1328 struct sk_buff *skb,
1329 unsigned int num)
1330{
1331 unsigned int val = atomic_inc_return(&f->rr_cur);
1332
1333 return val % num;
1334}
1335
1336static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1337 struct sk_buff *skb,
1338 unsigned int num)
1339{
1340 return smp_processor_id() % num;
1341}
1342
1343static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1344 struct sk_buff *skb,
1345 unsigned int num)
1346{
1347 return prandom_u32_max(num);
1348}
1349
1350static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1351 struct sk_buff *skb,
1352 unsigned int idx, bool try_self,
1353 unsigned int num)
1354{
1355 struct packet_sock *po, *po_next, *po_skip = NULL;
1356 unsigned int i, j, room = ROOM_NONE;
1357
1358 po = pkt_sk(f->arr[idx]);
1359
1360 if (try_self) {
1361 room = packet_rcv_has_room(po, skb);
1362 if (room == ROOM_NORMAL ||
1363 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1364 return idx;
1365 po_skip = po;
1366 }
1367
1368 i = j = min_t(int, po->rollover->sock, num - 1);
1369 do {
1370 po_next = pkt_sk(f->arr[i]);
1371 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1372 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1373 if (i != j)
1374 po->rollover->sock = i;
1375 atomic_long_inc(&po->rollover->num);
1376 if (room == ROOM_LOW)
1377 atomic_long_inc(&po->rollover->num_huge);
1378 return i;
1379 }
1380
1381 if (++i == num)
1382 i = 0;
1383 } while (i != j);
1384
1385 atomic_long_inc(&po->rollover->num_failed);
1386 return idx;
1387}
1388
1389static unsigned int fanout_demux_qm(struct packet_fanout *f,
1390 struct sk_buff *skb,
1391 unsigned int num)
1392{
1393 return skb_get_queue_mapping(skb) % num;
1394}
1395
1396static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1397 struct sk_buff *skb,
1398 unsigned int num)
1399{
1400 struct bpf_prog *prog;
1401 unsigned int ret = 0;
1402
1403 rcu_read_lock();
1404 prog = rcu_dereference(f->bpf_prog);
1405 if (prog)
1406 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1407 rcu_read_unlock();
1408
1409 return ret;
1410}
1411
1412static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1413{
1414 return f->flags & (flag >> 8);
1415}
1416
1417static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1418 struct packet_type *pt, struct net_device *orig_dev)
1419{
1420 struct packet_fanout *f = pt->af_packet_priv;
1421 unsigned int num = READ_ONCE(f->num_members);
1422 struct net *net = read_pnet(&f->net);
1423 struct packet_sock *po;
1424 unsigned int idx;
1425
1426 if (!net_eq(dev_net(dev), net) || !num) {
1427 kfree_skb(skb);
1428 return 0;
1429 }
1430
1431 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1432 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1433 if (!skb)
1434 return 0;
1435 }
1436 switch (f->type) {
1437 case PACKET_FANOUT_HASH:
1438 default:
1439 idx = fanout_demux_hash(f, skb, num);
1440 break;
1441 case PACKET_FANOUT_LB:
1442 idx = fanout_demux_lb(f, skb, num);
1443 break;
1444 case PACKET_FANOUT_CPU:
1445 idx = fanout_demux_cpu(f, skb, num);
1446 break;
1447 case PACKET_FANOUT_RND:
1448 idx = fanout_demux_rnd(f, skb, num);
1449 break;
1450 case PACKET_FANOUT_QM:
1451 idx = fanout_demux_qm(f, skb, num);
1452 break;
1453 case PACKET_FANOUT_ROLLOVER:
1454 idx = fanout_demux_rollover(f, skb, 0, false, num);
1455 break;
1456 case PACKET_FANOUT_CBPF:
1457 case PACKET_FANOUT_EBPF:
1458 idx = fanout_demux_bpf(f, skb, num);
1459 break;
1460 }
1461
1462 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1463 idx = fanout_demux_rollover(f, skb, idx, true, num);
1464
1465 po = pkt_sk(f->arr[idx]);
1466 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1467}
1468
1469DEFINE_MUTEX(fanout_mutex);
1470EXPORT_SYMBOL_GPL(fanout_mutex);
1471static LIST_HEAD(fanout_list);
1472static u16 fanout_next_id;
1473
1474static void __fanout_link(struct sock *sk, struct packet_sock *po)
1475{
1476 struct packet_fanout *f = po->fanout;
1477
1478 spin_lock(&f->lock);
1479 f->arr[f->num_members] = sk;
1480 smp_wmb();
1481 f->num_members++;
1482 if (f->num_members == 1)
1483 dev_add_pack(&f->prot_hook);
1484 spin_unlock(&f->lock);
1485}
1486
1487static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1488{
1489 struct packet_fanout *f = po->fanout;
1490 int i;
1491
1492 spin_lock(&f->lock);
1493 for (i = 0; i < f->num_members; i++) {
1494 if (f->arr[i] == sk)
1495 break;
1496 }
1497 BUG_ON(i >= f->num_members);
1498 f->arr[i] = f->arr[f->num_members - 1];
1499 f->num_members--;
1500 if (f->num_members == 0)
1501 __dev_remove_pack(&f->prot_hook);
1502 spin_unlock(&f->lock);
1503}
1504
1505static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1506{
1507 if (sk->sk_family != PF_PACKET)
1508 return false;
1509
1510 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1511}
1512
1513static void fanout_init_data(struct packet_fanout *f)
1514{
1515 switch (f->type) {
1516 case PACKET_FANOUT_LB:
1517 atomic_set(&f->rr_cur, 0);
1518 break;
1519 case PACKET_FANOUT_CBPF:
1520 case PACKET_FANOUT_EBPF:
1521 RCU_INIT_POINTER(f->bpf_prog, NULL);
1522 break;
1523 }
1524}
1525
1526static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1527{
1528 struct bpf_prog *old;
1529
1530 spin_lock(&f->lock);
1531 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1532 rcu_assign_pointer(f->bpf_prog, new);
1533 spin_unlock(&f->lock);
1534
1535 if (old) {
1536 synchronize_net();
1537 bpf_prog_destroy(old);
1538 }
1539}
1540
1541static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1542 unsigned int len)
1543{
1544 struct bpf_prog *new;
1545 struct sock_fprog fprog;
1546 int ret;
1547
1548 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1549 return -EPERM;
1550
1551 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1552 if (ret)
1553 return ret;
1554
1555 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1556 if (ret)
1557 return ret;
1558
1559 __fanout_set_data_bpf(po->fanout, new);
1560 return 0;
1561}
1562
1563static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1564 unsigned int len)
1565{
1566 struct bpf_prog *new;
1567 u32 fd;
1568
1569 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1570 return -EPERM;
1571 if (len != sizeof(fd))
1572 return -EINVAL;
1573 if (copy_from_sockptr(&fd, data, len))
1574 return -EFAULT;
1575
1576 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1577 if (IS_ERR(new))
1578 return PTR_ERR(new);
1579
1580 __fanout_set_data_bpf(po->fanout, new);
1581 return 0;
1582}
1583
1584static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1585 unsigned int len)
1586{
1587 switch (po->fanout->type) {
1588 case PACKET_FANOUT_CBPF:
1589 return fanout_set_data_cbpf(po, data, len);
1590 case PACKET_FANOUT_EBPF:
1591 return fanout_set_data_ebpf(po, data, len);
1592 default:
1593 return -EINVAL;
1594 }
1595}
1596
1597static void fanout_release_data(struct packet_fanout *f)
1598{
1599 switch (f->type) {
1600 case PACKET_FANOUT_CBPF:
1601 case PACKET_FANOUT_EBPF:
1602 __fanout_set_data_bpf(f, NULL);
1603 }
1604}
1605
1606static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1607{
1608 struct packet_fanout *f;
1609
1610 list_for_each_entry(f, &fanout_list, list) {
1611 if (f->id == candidate_id &&
1612 read_pnet(&f->net) == sock_net(sk)) {
1613 return false;
1614 }
1615 }
1616 return true;
1617}
1618
1619static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1620{
1621 u16 id = fanout_next_id;
1622
1623 do {
1624 if (__fanout_id_is_free(sk, id)) {
1625 *new_id = id;
1626 fanout_next_id = id + 1;
1627 return true;
1628 }
1629
1630 id++;
1631 } while (id != fanout_next_id);
1632
1633 return false;
1634}
1635
1636static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1637{
1638 struct packet_rollover *rollover = NULL;
1639 struct packet_sock *po = pkt_sk(sk);
1640 struct packet_fanout *f, *match;
1641 u8 type = type_flags & 0xff;
1642 u8 flags = type_flags >> 8;
1643 int err;
1644
1645 switch (type) {
1646 case PACKET_FANOUT_ROLLOVER:
1647 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1648 return -EINVAL;
1649 case PACKET_FANOUT_HASH:
1650 case PACKET_FANOUT_LB:
1651 case PACKET_FANOUT_CPU:
1652 case PACKET_FANOUT_RND:
1653 case PACKET_FANOUT_QM:
1654 case PACKET_FANOUT_CBPF:
1655 case PACKET_FANOUT_EBPF:
1656 break;
1657 default:
1658 return -EINVAL;
1659 }
1660
1661 mutex_lock(&fanout_mutex);
1662
1663 err = -EALREADY;
1664 if (po->fanout)
1665 goto out;
1666
1667 if (type == PACKET_FANOUT_ROLLOVER ||
1668 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1669 err = -ENOMEM;
1670 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1671 if (!rollover)
1672 goto out;
1673 atomic_long_set(&rollover->num, 0);
1674 atomic_long_set(&rollover->num_huge, 0);
1675 atomic_long_set(&rollover->num_failed, 0);
1676 }
1677
1678 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1679 if (id != 0) {
1680 err = -EINVAL;
1681 goto out;
1682 }
1683 if (!fanout_find_new_id(sk, &id)) {
1684 err = -ENOMEM;
1685 goto out;
1686 }
1687 /* ephemeral flag for the first socket in the group: drop it */
1688 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1689 }
1690
1691 match = NULL;
1692 list_for_each_entry(f, &fanout_list, list) {
1693 if (f->id == id &&
1694 read_pnet(&f->net) == sock_net(sk)) {
1695 match = f;
1696 break;
1697 }
1698 }
1699 err = -EINVAL;
1700 if (match && match->flags != flags)
1701 goto out;
1702 if (!match) {
1703 err = -ENOMEM;
1704 match = kzalloc(sizeof(*match), GFP_KERNEL);
1705 if (!match)
1706 goto out;
1707 write_pnet(&match->net, sock_net(sk));
1708 match->id = id;
1709 match->type = type;
1710 match->flags = flags;
1711 INIT_LIST_HEAD(&match->list);
1712 spin_lock_init(&match->lock);
1713 refcount_set(&match->sk_ref, 0);
1714 fanout_init_data(match);
1715 match->prot_hook.type = po->prot_hook.type;
1716 match->prot_hook.dev = po->prot_hook.dev;
1717 match->prot_hook.func = packet_rcv_fanout;
1718 match->prot_hook.af_packet_priv = match;
1719 match->prot_hook.id_match = match_fanout_group;
1720 list_add(&match->list, &fanout_list);
1721 }
1722 err = -EINVAL;
1723
1724 spin_lock(&po->bind_lock);
1725 if (po->running &&
1726 match->type == type &&
1727 match->prot_hook.type == po->prot_hook.type &&
1728 match->prot_hook.dev == po->prot_hook.dev) {
1729 err = -ENOSPC;
1730 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1731 __dev_remove_pack(&po->prot_hook);
1732 po->fanout = match;
1733 po->rollover = rollover;
1734 rollover = NULL;
1735 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1736 __fanout_link(sk, po);
1737 err = 0;
1738 }
1739 }
1740 spin_unlock(&po->bind_lock);
1741
1742 if (err && !refcount_read(&match->sk_ref)) {
1743 list_del(&match->list);
1744 kfree(match);
1745 }
1746
1747out:
1748 kfree(rollover);
1749 mutex_unlock(&fanout_mutex);
1750 return err;
1751}
1752
1753/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1754 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1755 * It is the responsibility of the caller to call fanout_release_data() and
1756 * free the returned packet_fanout (after synchronize_net())
1757 */
1758static struct packet_fanout *fanout_release(struct sock *sk)
1759{
1760 struct packet_sock *po = pkt_sk(sk);
1761 struct packet_fanout *f;
1762
1763 mutex_lock(&fanout_mutex);
1764 f = po->fanout;
1765 if (f) {
1766 po->fanout = NULL;
1767
1768 if (refcount_dec_and_test(&f->sk_ref))
1769 list_del(&f->list);
1770 else
1771 f = NULL;
1772 }
1773 mutex_unlock(&fanout_mutex);
1774
1775 return f;
1776}
1777
1778static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1779 struct sk_buff *skb)
1780{
1781 /* Earlier code assumed this would be a VLAN pkt, double-check
1782 * this now that we have the actual packet in hand. We can only
1783 * do this check on Ethernet devices.
1784 */
1785 if (unlikely(dev->type != ARPHRD_ETHER))
1786 return false;
1787
1788 skb_reset_mac_header(skb);
1789 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1790}
1791
1792static const struct proto_ops packet_ops;
1793
1794static const struct proto_ops packet_ops_spkt;
1795
1796static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1797 struct packet_type *pt, struct net_device *orig_dev)
1798{
1799 struct sock *sk;
1800 struct sockaddr_pkt *spkt;
1801
1802 /*
1803 * When we registered the protocol we saved the socket in the data
1804 * field for just this event.
1805 */
1806
1807 sk = pt->af_packet_priv;
1808
1809 /*
1810 * Yank back the headers [hope the device set this
1811 * right or kerboom...]
1812 *
1813 * Incoming packets have ll header pulled,
1814 * push it back.
1815 *
1816 * For outgoing ones skb->data == skb_mac_header(skb)
1817 * so that this procedure is noop.
1818 */
1819
1820 if (skb->pkt_type == PACKET_LOOPBACK)
1821 goto out;
1822
1823 if (!net_eq(dev_net(dev), sock_net(sk)))
1824 goto out;
1825
1826 skb = skb_share_check(skb, GFP_ATOMIC);
1827 if (skb == NULL)
1828 goto oom;
1829
1830 /* drop any routing info */
1831 skb_dst_drop(skb);
1832
1833 /* drop conntrack reference */
1834 nf_reset_ct(skb);
1835
1836 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1837
1838 skb_push(skb, skb->data - skb_mac_header(skb));
1839
1840 /*
1841 * The SOCK_PACKET socket receives _all_ frames.
1842 */
1843
1844 spkt->spkt_family = dev->type;
1845 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1846 spkt->spkt_protocol = skb->protocol;
1847
1848 /*
1849 * Charge the memory to the socket. This is done specifically
1850 * to prevent sockets using all the memory up.
1851 */
1852
1853 if (sock_queue_rcv_skb(sk, skb) == 0)
1854 return 0;
1855
1856out:
1857 kfree_skb(skb);
1858oom:
1859 return 0;
1860}
1861
1862static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1863{
1864 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1865 sock->type == SOCK_RAW) {
1866 skb_reset_mac_header(skb);
1867 skb->protocol = dev_parse_header_protocol(skb);
1868 }
1869
1870 skb_probe_transport_header(skb);
1871}
1872
1873/*
1874 * Output a raw packet to a device layer. This bypasses all the other
1875 * protocol layers and you must therefore supply it with a complete frame
1876 */
1877
1878static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1879 size_t len)
1880{
1881 struct sock *sk = sock->sk;
1882 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1883 struct sk_buff *skb = NULL;
1884 struct net_device *dev;
1885 struct sockcm_cookie sockc;
1886 __be16 proto = 0;
1887 int err;
1888 int extra_len = 0;
1889
1890 /*
1891 * Get and verify the address.
1892 */
1893
1894 if (saddr) {
1895 if (msg->msg_namelen < sizeof(struct sockaddr))
1896 return -EINVAL;
1897 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1898 proto = saddr->spkt_protocol;
1899 } else
1900 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1901
1902 /*
1903 * Find the device first to size check it
1904 */
1905
1906 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1907retry:
1908 rcu_read_lock();
1909 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1910 err = -ENODEV;
1911 if (dev == NULL)
1912 goto out_unlock;
1913
1914 err = -ENETDOWN;
1915 if (!(dev->flags & IFF_UP))
1916 goto out_unlock;
1917
1918 /*
1919 * You may not queue a frame bigger than the mtu. This is the lowest level
1920 * raw protocol and you must do your own fragmentation at this level.
1921 */
1922
1923 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1924 if (!netif_supports_nofcs(dev)) {
1925 err = -EPROTONOSUPPORT;
1926 goto out_unlock;
1927 }
1928 extra_len = 4; /* We're doing our own CRC */
1929 }
1930
1931 err = -EMSGSIZE;
1932 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1933 goto out_unlock;
1934
1935 if (!skb) {
1936 size_t reserved = LL_RESERVED_SPACE(dev);
1937 int tlen = dev->needed_tailroom;
1938 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1939
1940 rcu_read_unlock();
1941 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1942 if (skb == NULL)
1943 return -ENOBUFS;
1944 /* FIXME: Save some space for broken drivers that write a hard
1945 * header at transmission time by themselves. PPP is the notable
1946 * one here. This should really be fixed at the driver level.
1947 */
1948 skb_reserve(skb, reserved);
1949 skb_reset_network_header(skb);
1950
1951 /* Try to align data part correctly */
1952 if (hhlen) {
1953 skb->data -= hhlen;
1954 skb->tail -= hhlen;
1955 if (len < hhlen)
1956 skb_reset_network_header(skb);
1957 }
1958 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1959 if (err)
1960 goto out_free;
1961 goto retry;
1962 }
1963
1964 if (!dev_validate_header(dev, skb->data, len)) {
1965 err = -EINVAL;
1966 goto out_unlock;
1967 }
1968 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1969 !packet_extra_vlan_len_allowed(dev, skb)) {
1970 err = -EMSGSIZE;
1971 goto out_unlock;
1972 }
1973
1974 sockcm_init(&sockc, sk);
1975 if (msg->msg_controllen) {
1976 err = sock_cmsg_send(sk, msg, &sockc);
1977 if (unlikely(err))
1978 goto out_unlock;
1979 }
1980
1981 skb->protocol = proto;
1982 skb->dev = dev;
1983 skb->priority = sk->sk_priority;
1984 skb->mark = sk->sk_mark;
1985 skb->tstamp = sockc.transmit_time;
1986
1987 skb_setup_tx_timestamp(skb, sockc.tsflags);
1988
1989 if (unlikely(extra_len == 4))
1990 skb->no_fcs = 1;
1991
1992 packet_parse_headers(skb, sock);
1993
1994 dev_queue_xmit(skb);
1995 rcu_read_unlock();
1996 return len;
1997
1998out_unlock:
1999 rcu_read_unlock();
2000out_free:
2001 kfree_skb(skb);
2002 return err;
2003}
2004
2005static unsigned int run_filter(struct sk_buff *skb,
2006 const struct sock *sk,
2007 unsigned int res)
2008{
2009 struct sk_filter *filter;
2010
2011 rcu_read_lock();
2012 filter = rcu_dereference(sk->sk_filter);
2013 if (filter != NULL)
2014 res = bpf_prog_run_clear_cb(filter->prog, skb);
2015 rcu_read_unlock();
2016
2017 return res;
2018}
2019
2020static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2021 size_t *len)
2022{
2023 struct virtio_net_hdr vnet_hdr;
2024
2025 if (*len < sizeof(vnet_hdr))
2026 return -EINVAL;
2027 *len -= sizeof(vnet_hdr);
2028
2029 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2030 return -EINVAL;
2031
2032 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2033}
2034
2035/*
2036 * This function makes lazy skb cloning in hope that most of packets
2037 * are discarded by BPF.
2038 *
2039 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2040 * and skb->cb are mangled. It works because (and until) packets
2041 * falling here are owned by current CPU. Output packets are cloned
2042 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2043 * sequencially, so that if we return skb to original state on exit,
2044 * we will not harm anyone.
2045 */
2046
2047static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2048 struct packet_type *pt, struct net_device *orig_dev)
2049{
2050 struct sock *sk;
2051 struct sockaddr_ll *sll;
2052 struct packet_sock *po;
2053 u8 *skb_head = skb->data;
2054 int skb_len = skb->len;
2055 unsigned int snaplen, res;
2056 bool is_drop_n_account = false;
2057
2058 if (skb->pkt_type == PACKET_LOOPBACK)
2059 goto drop;
2060
2061 sk = pt->af_packet_priv;
2062 po = pkt_sk(sk);
2063
2064 if (!net_eq(dev_net(dev), sock_net(sk)))
2065 goto drop;
2066
2067 skb->dev = dev;
2068
2069 if (dev->header_ops) {
2070 /* The device has an explicit notion of ll header,
2071 * exported to higher levels.
2072 *
2073 * Otherwise, the device hides details of its frame
2074 * structure, so that corresponding packet head is
2075 * never delivered to user.
2076 */
2077 if (sk->sk_type != SOCK_DGRAM)
2078 skb_push(skb, skb->data - skb_mac_header(skb));
2079 else if (skb->pkt_type == PACKET_OUTGOING) {
2080 /* Special case: outgoing packets have ll header at head */
2081 skb_pull(skb, skb_network_offset(skb));
2082 }
2083 }
2084
2085 snaplen = skb->len;
2086
2087 res = run_filter(skb, sk, snaplen);
2088 if (!res)
2089 goto drop_n_restore;
2090 if (snaplen > res)
2091 snaplen = res;
2092
2093 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2094 goto drop_n_acct;
2095
2096 if (skb_shared(skb)) {
2097 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2098 if (nskb == NULL)
2099 goto drop_n_acct;
2100
2101 if (skb_head != skb->data) {
2102 skb->data = skb_head;
2103 skb->len = skb_len;
2104 }
2105 consume_skb(skb);
2106 skb = nskb;
2107 }
2108
2109 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2110
2111 sll = &PACKET_SKB_CB(skb)->sa.ll;
2112 sll->sll_hatype = dev->type;
2113 sll->sll_pkttype = skb->pkt_type;
2114 if (unlikely(po->origdev))
2115 sll->sll_ifindex = orig_dev->ifindex;
2116 else
2117 sll->sll_ifindex = dev->ifindex;
2118
2119 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2120
2121 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2122 * Use their space for storing the original skb length.
2123 */
2124 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2125
2126 if (pskb_trim(skb, snaplen))
2127 goto drop_n_acct;
2128
2129 skb_set_owner_r(skb, sk);
2130 skb->dev = NULL;
2131 skb_dst_drop(skb);
2132
2133 /* drop conntrack reference */
2134 nf_reset_ct(skb);
2135
2136 spin_lock(&sk->sk_receive_queue.lock);
2137 po->stats.stats1.tp_packets++;
2138 sock_skb_set_dropcount(sk, skb);
2139 __skb_queue_tail(&sk->sk_receive_queue, skb);
2140 spin_unlock(&sk->sk_receive_queue.lock);
2141 sk->sk_data_ready(sk);
2142 return 0;
2143
2144drop_n_acct:
2145 is_drop_n_account = true;
2146 atomic_inc(&po->tp_drops);
2147 atomic_inc(&sk->sk_drops);
2148
2149drop_n_restore:
2150 if (skb_head != skb->data && skb_shared(skb)) {
2151 skb->data = skb_head;
2152 skb->len = skb_len;
2153 }
2154drop:
2155 if (!is_drop_n_account)
2156 consume_skb(skb);
2157 else
2158 kfree_skb(skb);
2159 return 0;
2160}
2161
2162static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2163 struct packet_type *pt, struct net_device *orig_dev)
2164{
2165 struct sock *sk;
2166 struct packet_sock *po;
2167 struct sockaddr_ll *sll;
2168 union tpacket_uhdr h;
2169 u8 *skb_head = skb->data;
2170 int skb_len = skb->len;
2171 unsigned int snaplen, res;
2172 unsigned long status = TP_STATUS_USER;
2173 unsigned short macoff, hdrlen;
2174 unsigned int netoff;
2175 struct sk_buff *copy_skb = NULL;
2176 struct timespec64 ts;
2177 __u32 ts_status;
2178 bool is_drop_n_account = false;
2179 unsigned int slot_id = 0;
2180 bool do_vnet = false;
2181
2182 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2183 * We may add members to them until current aligned size without forcing
2184 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2185 */
2186 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2187 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2188
2189 if (skb->pkt_type == PACKET_LOOPBACK)
2190 goto drop;
2191
2192 sk = pt->af_packet_priv;
2193 po = pkt_sk(sk);
2194
2195 if (!net_eq(dev_net(dev), sock_net(sk)))
2196 goto drop;
2197
2198 if (dev->header_ops) {
2199 if (sk->sk_type != SOCK_DGRAM)
2200 skb_push(skb, skb->data - skb_mac_header(skb));
2201 else if (skb->pkt_type == PACKET_OUTGOING) {
2202 /* Special case: outgoing packets have ll header at head */
2203 skb_pull(skb, skb_network_offset(skb));
2204 }
2205 }
2206
2207 snaplen = skb->len;
2208
2209 res = run_filter(skb, sk, snaplen);
2210 if (!res)
2211 goto drop_n_restore;
2212
2213 /* If we are flooded, just give up */
2214 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2215 atomic_inc(&po->tp_drops);
2216 goto drop_n_restore;
2217 }
2218
2219 if (skb->ip_summed == CHECKSUM_PARTIAL)
2220 status |= TP_STATUS_CSUMNOTREADY;
2221 else if (skb->pkt_type != PACKET_OUTGOING &&
2222 (skb->ip_summed == CHECKSUM_COMPLETE ||
2223 skb_csum_unnecessary(skb)))
2224 status |= TP_STATUS_CSUM_VALID;
2225
2226 if (snaplen > res)
2227 snaplen = res;
2228
2229 if (sk->sk_type == SOCK_DGRAM) {
2230 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2231 po->tp_reserve;
2232 } else {
2233 unsigned int maclen = skb_network_offset(skb);
2234 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2235 (maclen < 16 ? 16 : maclen)) +
2236 po->tp_reserve;
2237 if (po->has_vnet_hdr) {
2238 netoff += sizeof(struct virtio_net_hdr);
2239 do_vnet = true;
2240 }
2241 macoff = netoff - maclen;
2242 }
2243 if (netoff > USHRT_MAX) {
2244 atomic_inc(&po->tp_drops);
2245 goto drop_n_restore;
2246 }
2247 if (po->tp_version <= TPACKET_V2) {
2248 if (macoff + snaplen > po->rx_ring.frame_size) {
2249 if (po->copy_thresh &&
2250 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2251 if (skb_shared(skb)) {
2252 copy_skb = skb_clone(skb, GFP_ATOMIC);
2253 } else {
2254 copy_skb = skb_get(skb);
2255 skb_head = skb->data;
2256 }
2257 if (copy_skb)
2258 skb_set_owner_r(copy_skb, sk);
2259 }
2260 snaplen = po->rx_ring.frame_size - macoff;
2261 if ((int)snaplen < 0) {
2262 snaplen = 0;
2263 do_vnet = false;
2264 }
2265 }
2266 } else if (unlikely(macoff + snaplen >
2267 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2268 u32 nval;
2269
2270 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2271 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2272 snaplen, nval, macoff);
2273 snaplen = nval;
2274 if (unlikely((int)snaplen < 0)) {
2275 snaplen = 0;
2276 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2277 do_vnet = false;
2278 }
2279 }
2280 spin_lock(&sk->sk_receive_queue.lock);
2281 h.raw = packet_current_rx_frame(po, skb,
2282 TP_STATUS_KERNEL, (macoff+snaplen));
2283 if (!h.raw)
2284 goto drop_n_account;
2285
2286 if (po->tp_version <= TPACKET_V2) {
2287 slot_id = po->rx_ring.head;
2288 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2289 goto drop_n_account;
2290 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2291 }
2292
2293 if (do_vnet &&
2294 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2295 sizeof(struct virtio_net_hdr),
2296 vio_le(), true, 0)) {
2297 if (po->tp_version == TPACKET_V3)
2298 prb_clear_blk_fill_status(&po->rx_ring);
2299 goto drop_n_account;
2300 }
2301
2302 if (po->tp_version <= TPACKET_V2) {
2303 packet_increment_rx_head(po, &po->rx_ring);
2304 /*
2305 * LOSING will be reported till you read the stats,
2306 * because it's COR - Clear On Read.
2307 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2308 * at packet level.
2309 */
2310 if (atomic_read(&po->tp_drops))
2311 status |= TP_STATUS_LOSING;
2312 }
2313
2314 po->stats.stats1.tp_packets++;
2315 if (copy_skb) {
2316 status |= TP_STATUS_COPY;
2317 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2318 }
2319 spin_unlock(&sk->sk_receive_queue.lock);
2320
2321 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2322
2323 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2324 ktime_get_real_ts64(&ts);
2325
2326 status |= ts_status;
2327
2328 switch (po->tp_version) {
2329 case TPACKET_V1:
2330 h.h1->tp_len = skb->len;
2331 h.h1->tp_snaplen = snaplen;
2332 h.h1->tp_mac = macoff;
2333 h.h1->tp_net = netoff;
2334 h.h1->tp_sec = ts.tv_sec;
2335 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2336 hdrlen = sizeof(*h.h1);
2337 break;
2338 case TPACKET_V2:
2339 h.h2->tp_len = skb->len;
2340 h.h2->tp_snaplen = snaplen;
2341 h.h2->tp_mac = macoff;
2342 h.h2->tp_net = netoff;
2343 h.h2->tp_sec = ts.tv_sec;
2344 h.h2->tp_nsec = ts.tv_nsec;
2345 if (skb_vlan_tag_present(skb)) {
2346 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2347 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2348 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2349 } else {
2350 h.h2->tp_vlan_tci = 0;
2351 h.h2->tp_vlan_tpid = 0;
2352 }
2353 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2354 hdrlen = sizeof(*h.h2);
2355 break;
2356 case TPACKET_V3:
2357 /* tp_nxt_offset,vlan are already populated above.
2358 * So DONT clear those fields here
2359 */
2360 h.h3->tp_status |= status;
2361 h.h3->tp_len = skb->len;
2362 h.h3->tp_snaplen = snaplen;
2363 h.h3->tp_mac = macoff;
2364 h.h3->tp_net = netoff;
2365 h.h3->tp_sec = ts.tv_sec;
2366 h.h3->tp_nsec = ts.tv_nsec;
2367 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2368 hdrlen = sizeof(*h.h3);
2369 break;
2370 default:
2371 BUG();
2372 }
2373
2374 sll = h.raw + TPACKET_ALIGN(hdrlen);
2375 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2376 sll->sll_family = AF_PACKET;
2377 sll->sll_hatype = dev->type;
2378 sll->sll_protocol = skb->protocol;
2379 sll->sll_pkttype = skb->pkt_type;
2380 if (unlikely(po->origdev))
2381 sll->sll_ifindex = orig_dev->ifindex;
2382 else
2383 sll->sll_ifindex = dev->ifindex;
2384
2385 smp_mb();
2386
2387#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2388 if (po->tp_version <= TPACKET_V2) {
2389 u8 *start, *end;
2390
2391 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2392 macoff + snaplen);
2393
2394 for (start = h.raw; start < end; start += PAGE_SIZE)
2395 flush_dcache_page(pgv_to_page(start));
2396 }
2397 smp_wmb();
2398#endif
2399
2400 if (po->tp_version <= TPACKET_V2) {
2401 spin_lock(&sk->sk_receive_queue.lock);
2402 __packet_set_status(po, h.raw, status);
2403 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2404 spin_unlock(&sk->sk_receive_queue.lock);
2405 sk->sk_data_ready(sk);
2406 } else if (po->tp_version == TPACKET_V3) {
2407 prb_clear_blk_fill_status(&po->rx_ring);
2408 }
2409
2410drop_n_restore:
2411 if (skb_head != skb->data && skb_shared(skb)) {
2412 skb->data = skb_head;
2413 skb->len = skb_len;
2414 }
2415drop:
2416 if (!is_drop_n_account)
2417 consume_skb(skb);
2418 else
2419 kfree_skb(skb);
2420 return 0;
2421
2422drop_n_account:
2423 spin_unlock(&sk->sk_receive_queue.lock);
2424 atomic_inc(&po->tp_drops);
2425 is_drop_n_account = true;
2426
2427 sk->sk_data_ready(sk);
2428 kfree_skb(copy_skb);
2429 goto drop_n_restore;
2430}
2431
2432static void tpacket_destruct_skb(struct sk_buff *skb)
2433{
2434 struct packet_sock *po = pkt_sk(skb->sk);
2435
2436 if (likely(po->tx_ring.pg_vec)) {
2437 void *ph;
2438 __u32 ts;
2439
2440 ph = skb_zcopy_get_nouarg(skb);
2441 packet_dec_pending(&po->tx_ring);
2442
2443 ts = __packet_set_timestamp(po, ph, skb);
2444 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2445
2446 if (!packet_read_pending(&po->tx_ring))
2447 complete(&po->skb_completion);
2448 }
2449
2450 sock_wfree(skb);
2451}
2452
2453static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2454{
2455 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2456 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2457 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2458 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2459 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2460 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2461 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2462
2463 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2464 return -EINVAL;
2465
2466 return 0;
2467}
2468
2469static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2470 struct virtio_net_hdr *vnet_hdr)
2471{
2472 if (*len < sizeof(*vnet_hdr))
2473 return -EINVAL;
2474 *len -= sizeof(*vnet_hdr);
2475
2476 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2477 return -EFAULT;
2478
2479 return __packet_snd_vnet_parse(vnet_hdr, *len);
2480}
2481
2482static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2483 void *frame, struct net_device *dev, void *data, int tp_len,
2484 __be16 proto, unsigned char *addr, int hlen, int copylen,
2485 const struct sockcm_cookie *sockc)
2486{
2487 union tpacket_uhdr ph;
2488 int to_write, offset, len, nr_frags, len_max;
2489 struct socket *sock = po->sk.sk_socket;
2490 struct page *page;
2491 int err;
2492
2493 ph.raw = frame;
2494
2495 skb->protocol = proto;
2496 skb->dev = dev;
2497 skb->priority = po->sk.sk_priority;
2498 skb->mark = po->sk.sk_mark;
2499 skb->tstamp = sockc->transmit_time;
2500 skb_setup_tx_timestamp(skb, sockc->tsflags);
2501 skb_zcopy_set_nouarg(skb, ph.raw);
2502
2503 skb_reserve(skb, hlen);
2504 skb_reset_network_header(skb);
2505
2506 to_write = tp_len;
2507
2508 if (sock->type == SOCK_DGRAM) {
2509 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2510 NULL, tp_len);
2511 if (unlikely(err < 0))
2512 return -EINVAL;
2513 } else if (copylen) {
2514 int hdrlen = min_t(int, copylen, tp_len);
2515
2516 skb_push(skb, dev->hard_header_len);
2517 skb_put(skb, copylen - dev->hard_header_len);
2518 err = skb_store_bits(skb, 0, data, hdrlen);
2519 if (unlikely(err))
2520 return err;
2521 if (!dev_validate_header(dev, skb->data, hdrlen))
2522 return -EINVAL;
2523
2524 data += hdrlen;
2525 to_write -= hdrlen;
2526 }
2527
2528 offset = offset_in_page(data);
2529 len_max = PAGE_SIZE - offset;
2530 len = ((to_write > len_max) ? len_max : to_write);
2531
2532 skb->data_len = to_write;
2533 skb->len += to_write;
2534 skb->truesize += to_write;
2535 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2536
2537 while (likely(to_write)) {
2538 nr_frags = skb_shinfo(skb)->nr_frags;
2539
2540 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2541 pr_err("Packet exceed the number of skb frags(%lu)\n",
2542 MAX_SKB_FRAGS);
2543 return -EFAULT;
2544 }
2545
2546 page = pgv_to_page(data);
2547 data += len;
2548 flush_dcache_page(page);
2549 get_page(page);
2550 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2551 to_write -= len;
2552 offset = 0;
2553 len_max = PAGE_SIZE;
2554 len = ((to_write > len_max) ? len_max : to_write);
2555 }
2556
2557 packet_parse_headers(skb, sock);
2558
2559 return tp_len;
2560}
2561
2562static int tpacket_parse_header(struct packet_sock *po, void *frame,
2563 int size_max, void **data)
2564{
2565 union tpacket_uhdr ph;
2566 int tp_len, off;
2567
2568 ph.raw = frame;
2569
2570 switch (po->tp_version) {
2571 case TPACKET_V3:
2572 if (ph.h3->tp_next_offset != 0) {
2573 pr_warn_once("variable sized slot not supported");
2574 return -EINVAL;
2575 }
2576 tp_len = ph.h3->tp_len;
2577 break;
2578 case TPACKET_V2:
2579 tp_len = ph.h2->tp_len;
2580 break;
2581 default:
2582 tp_len = ph.h1->tp_len;
2583 break;
2584 }
2585 if (unlikely(tp_len > size_max)) {
2586 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2587 return -EMSGSIZE;
2588 }
2589
2590 if (unlikely(po->tp_tx_has_off)) {
2591 int off_min, off_max;
2592
2593 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2594 off_max = po->tx_ring.frame_size - tp_len;
2595 if (po->sk.sk_type == SOCK_DGRAM) {
2596 switch (po->tp_version) {
2597 case TPACKET_V3:
2598 off = ph.h3->tp_net;
2599 break;
2600 case TPACKET_V2:
2601 off = ph.h2->tp_net;
2602 break;
2603 default:
2604 off = ph.h1->tp_net;
2605 break;
2606 }
2607 } else {
2608 switch (po->tp_version) {
2609 case TPACKET_V3:
2610 off = ph.h3->tp_mac;
2611 break;
2612 case TPACKET_V2:
2613 off = ph.h2->tp_mac;
2614 break;
2615 default:
2616 off = ph.h1->tp_mac;
2617 break;
2618 }
2619 }
2620 if (unlikely((off < off_min) || (off_max < off)))
2621 return -EINVAL;
2622 } else {
2623 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2624 }
2625
2626 *data = frame + off;
2627 return tp_len;
2628}
2629
2630static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2631{
2632 struct sk_buff *skb = NULL;
2633 struct net_device *dev;
2634 struct virtio_net_hdr *vnet_hdr = NULL;
2635 struct sockcm_cookie sockc;
2636 __be16 proto;
2637 int err, reserve = 0;
2638 void *ph;
2639 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2640 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2641 unsigned char *addr = NULL;
2642 int tp_len, size_max;
2643 void *data;
2644 int len_sum = 0;
2645 int status = TP_STATUS_AVAILABLE;
2646 int hlen, tlen, copylen = 0;
2647 long timeo = 0;
2648
2649 mutex_lock(&po->pg_vec_lock);
2650
2651 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2652 * we need to confirm it under protection of pg_vec_lock.
2653 */
2654 if (unlikely(!po->tx_ring.pg_vec)) {
2655 err = -EBUSY;
2656 goto out;
2657 }
2658 if (likely(saddr == NULL)) {
2659 dev = packet_cached_dev_get(po);
2660 proto = po->num;
2661 } else {
2662 err = -EINVAL;
2663 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2664 goto out;
2665 if (msg->msg_namelen < (saddr->sll_halen
2666 + offsetof(struct sockaddr_ll,
2667 sll_addr)))
2668 goto out;
2669 proto = saddr->sll_protocol;
2670 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2671 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2672 if (dev && msg->msg_namelen < dev->addr_len +
2673 offsetof(struct sockaddr_ll, sll_addr))
2674 goto out_put;
2675 addr = saddr->sll_addr;
2676 }
2677 }
2678
2679 err = -ENXIO;
2680 if (unlikely(dev == NULL))
2681 goto out;
2682 err = -ENETDOWN;
2683 if (unlikely(!(dev->flags & IFF_UP)))
2684 goto out_put;
2685
2686 sockcm_init(&sockc, &po->sk);
2687 if (msg->msg_controllen) {
2688 err = sock_cmsg_send(&po->sk, msg, &sockc);
2689 if (unlikely(err))
2690 goto out_put;
2691 }
2692
2693 if (po->sk.sk_socket->type == SOCK_RAW)
2694 reserve = dev->hard_header_len;
2695 size_max = po->tx_ring.frame_size
2696 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2697
2698 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2699 size_max = dev->mtu + reserve + VLAN_HLEN;
2700
2701 reinit_completion(&po->skb_completion);
2702
2703 do {
2704 ph = packet_current_frame(po, &po->tx_ring,
2705 TP_STATUS_SEND_REQUEST);
2706 if (unlikely(ph == NULL)) {
2707 if (need_wait && skb) {
2708 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2709 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2710 if (timeo <= 0) {
2711 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2712 goto out_put;
2713 }
2714 }
2715 /* check for additional frames */
2716 continue;
2717 }
2718
2719 skb = NULL;
2720 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2721 if (tp_len < 0)
2722 goto tpacket_error;
2723
2724 status = TP_STATUS_SEND_REQUEST;
2725 hlen = LL_RESERVED_SPACE(dev);
2726 tlen = dev->needed_tailroom;
2727 if (po->has_vnet_hdr) {
2728 vnet_hdr = data;
2729 data += sizeof(*vnet_hdr);
2730 tp_len -= sizeof(*vnet_hdr);
2731 if (tp_len < 0 ||
2732 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2733 tp_len = -EINVAL;
2734 goto tpacket_error;
2735 }
2736 copylen = __virtio16_to_cpu(vio_le(),
2737 vnet_hdr->hdr_len);
2738 }
2739 copylen = max_t(int, copylen, dev->hard_header_len);
2740 skb = sock_alloc_send_skb(&po->sk,
2741 hlen + tlen + sizeof(struct sockaddr_ll) +
2742 (copylen - dev->hard_header_len),
2743 !need_wait, &err);
2744
2745 if (unlikely(skb == NULL)) {
2746 /* we assume the socket was initially writeable ... */
2747 if (likely(len_sum > 0))
2748 err = len_sum;
2749 goto out_status;
2750 }
2751 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2752 addr, hlen, copylen, &sockc);
2753 if (likely(tp_len >= 0) &&
2754 tp_len > dev->mtu + reserve &&
2755 !po->has_vnet_hdr &&
2756 !packet_extra_vlan_len_allowed(dev, skb))
2757 tp_len = -EMSGSIZE;
2758
2759 if (unlikely(tp_len < 0)) {
2760tpacket_error:
2761 if (po->tp_loss) {
2762 __packet_set_status(po, ph,
2763 TP_STATUS_AVAILABLE);
2764 packet_increment_head(&po->tx_ring);
2765 kfree_skb(skb);
2766 continue;
2767 } else {
2768 status = TP_STATUS_WRONG_FORMAT;
2769 err = tp_len;
2770 goto out_status;
2771 }
2772 }
2773
2774 if (po->has_vnet_hdr) {
2775 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2776 tp_len = -EINVAL;
2777 goto tpacket_error;
2778 }
2779 virtio_net_hdr_set_proto(skb, vnet_hdr);
2780 }
2781
2782 skb->destructor = tpacket_destruct_skb;
2783 __packet_set_status(po, ph, TP_STATUS_SENDING);
2784 packet_inc_pending(&po->tx_ring);
2785
2786 status = TP_STATUS_SEND_REQUEST;
2787 err = po->xmit(skb);
2788 if (unlikely(err > 0)) {
2789 err = net_xmit_errno(err);
2790 if (err && __packet_get_status(po, ph) ==
2791 TP_STATUS_AVAILABLE) {
2792 /* skb was destructed already */
2793 skb = NULL;
2794 goto out_status;
2795 }
2796 /*
2797 * skb was dropped but not destructed yet;
2798 * let's treat it like congestion or err < 0
2799 */
2800 err = 0;
2801 }
2802 packet_increment_head(&po->tx_ring);
2803 len_sum += tp_len;
2804 } while (likely((ph != NULL) ||
2805 /* Note: packet_read_pending() might be slow if we have
2806 * to call it as it's per_cpu variable, but in fast-path
2807 * we already short-circuit the loop with the first
2808 * condition, and luckily don't have to go that path
2809 * anyway.
2810 */
2811 (need_wait && packet_read_pending(&po->tx_ring))));
2812
2813 err = len_sum;
2814 goto out_put;
2815
2816out_status:
2817 __packet_set_status(po, ph, status);
2818 kfree_skb(skb);
2819out_put:
2820 dev_put(dev);
2821out:
2822 mutex_unlock(&po->pg_vec_lock);
2823 return err;
2824}
2825
2826static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2827 size_t reserve, size_t len,
2828 size_t linear, int noblock,
2829 int *err)
2830{
2831 struct sk_buff *skb;
2832
2833 /* Under a page? Don't bother with paged skb. */
2834 if (prepad + len < PAGE_SIZE || !linear)
2835 linear = len;
2836
2837 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2838 err, 0);
2839 if (!skb)
2840 return NULL;
2841
2842 skb_reserve(skb, reserve);
2843 skb_put(skb, linear);
2844 skb->data_len = len - linear;
2845 skb->len += len - linear;
2846
2847 return skb;
2848}
2849
2850static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2851{
2852 struct sock *sk = sock->sk;
2853 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2854 struct sk_buff *skb;
2855 struct net_device *dev;
2856 __be16 proto;
2857 unsigned char *addr = NULL;
2858 int err, reserve = 0;
2859 struct sockcm_cookie sockc;
2860 struct virtio_net_hdr vnet_hdr = { 0 };
2861 int offset = 0;
2862 struct packet_sock *po = pkt_sk(sk);
2863 bool has_vnet_hdr = false;
2864 int hlen, tlen, linear;
2865 int extra_len = 0;
2866
2867 /*
2868 * Get and verify the address.
2869 */
2870
2871 if (likely(saddr == NULL)) {
2872 dev = packet_cached_dev_get(po);
2873 proto = po->num;
2874 } else {
2875 err = -EINVAL;
2876 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2877 goto out;
2878 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2879 goto out;
2880 proto = saddr->sll_protocol;
2881 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2882 if (sock->type == SOCK_DGRAM) {
2883 if (dev && msg->msg_namelen < dev->addr_len +
2884 offsetof(struct sockaddr_ll, sll_addr))
2885 goto out_unlock;
2886 addr = saddr->sll_addr;
2887 }
2888 }
2889
2890 err = -ENXIO;
2891 if (unlikely(dev == NULL))
2892 goto out_unlock;
2893 err = -ENETDOWN;
2894 if (unlikely(!(dev->flags & IFF_UP)))
2895 goto out_unlock;
2896
2897 sockcm_init(&sockc, sk);
2898 sockc.mark = sk->sk_mark;
2899 if (msg->msg_controllen) {
2900 err = sock_cmsg_send(sk, msg, &sockc);
2901 if (unlikely(err))
2902 goto out_unlock;
2903 }
2904
2905 if (sock->type == SOCK_RAW)
2906 reserve = dev->hard_header_len;
2907 if (po->has_vnet_hdr) {
2908 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2909 if (err)
2910 goto out_unlock;
2911 has_vnet_hdr = true;
2912 }
2913
2914 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2915 if (!netif_supports_nofcs(dev)) {
2916 err = -EPROTONOSUPPORT;
2917 goto out_unlock;
2918 }
2919 extra_len = 4; /* We're doing our own CRC */
2920 }
2921
2922 err = -EMSGSIZE;
2923 if (!vnet_hdr.gso_type &&
2924 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2925 goto out_unlock;
2926
2927 err = -ENOBUFS;
2928 hlen = LL_RESERVED_SPACE(dev);
2929 tlen = dev->needed_tailroom;
2930 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2931 linear = max(linear, min_t(int, len, dev->hard_header_len));
2932 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2933 msg->msg_flags & MSG_DONTWAIT, &err);
2934 if (skb == NULL)
2935 goto out_unlock;
2936
2937 skb_reset_network_header(skb);
2938
2939 err = -EINVAL;
2940 if (sock->type == SOCK_DGRAM) {
2941 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2942 if (unlikely(offset < 0))
2943 goto out_free;
2944 } else if (reserve) {
2945 skb_reserve(skb, -reserve);
2946 if (len < reserve + sizeof(struct ipv6hdr) &&
2947 dev->min_header_len != dev->hard_header_len)
2948 skb_reset_network_header(skb);
2949 }
2950
2951 /* Returns -EFAULT on error */
2952 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2953 if (err)
2954 goto out_free;
2955
2956 if (sock->type == SOCK_RAW &&
2957 !dev_validate_header(dev, skb->data, len)) {
2958 err = -EINVAL;
2959 goto out_free;
2960 }
2961
2962 skb_setup_tx_timestamp(skb, sockc.tsflags);
2963
2964 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2965 !packet_extra_vlan_len_allowed(dev, skb)) {
2966 err = -EMSGSIZE;
2967 goto out_free;
2968 }
2969
2970 skb->protocol = proto;
2971 skb->dev = dev;
2972 skb->priority = sk->sk_priority;
2973 skb->mark = sockc.mark;
2974 skb->tstamp = sockc.transmit_time;
2975
2976 if (has_vnet_hdr) {
2977 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2978 if (err)
2979 goto out_free;
2980 len += sizeof(vnet_hdr);
2981 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2982 }
2983
2984 packet_parse_headers(skb, sock);
2985
2986 if (unlikely(extra_len == 4))
2987 skb->no_fcs = 1;
2988
2989 err = po->xmit(skb);
2990 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2991 goto out_unlock;
2992
2993 dev_put(dev);
2994
2995 return len;
2996
2997out_free:
2998 kfree_skb(skb);
2999out_unlock:
3000 if (dev)
3001 dev_put(dev);
3002out:
3003 return err;
3004}
3005
3006static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3007{
3008 struct sock *sk = sock->sk;
3009 struct packet_sock *po = pkt_sk(sk);
3010
3011 if (po->tx_ring.pg_vec)
3012 return tpacket_snd(po, msg);
3013 else
3014 return packet_snd(sock, msg, len);
3015}
3016
3017/*
3018 * Close a PACKET socket. This is fairly simple. We immediately go
3019 * to 'closed' state and remove our protocol entry in the device list.
3020 */
3021
3022static int packet_release(struct socket *sock)
3023{
3024 struct sock *sk = sock->sk;
3025 struct packet_sock *po;
3026 struct packet_fanout *f;
3027 struct net *net;
3028 union tpacket_req_u req_u;
3029
3030 if (!sk)
3031 return 0;
3032
3033 net = sock_net(sk);
3034 po = pkt_sk(sk);
3035
3036 mutex_lock(&net->packet.sklist_lock);
3037 sk_del_node_init_rcu(sk);
3038 mutex_unlock(&net->packet.sklist_lock);
3039
3040 preempt_disable();
3041 sock_prot_inuse_add(net, sk->sk_prot, -1);
3042 preempt_enable();
3043
3044 spin_lock(&po->bind_lock);
3045 unregister_prot_hook(sk, false);
3046 packet_cached_dev_reset(po);
3047
3048 if (po->prot_hook.dev) {
3049 dev_put(po->prot_hook.dev);
3050 po->prot_hook.dev = NULL;
3051 }
3052 spin_unlock(&po->bind_lock);
3053
3054 packet_flush_mclist(sk);
3055
3056 lock_sock(sk);
3057 if (po->rx_ring.pg_vec) {
3058 memset(&req_u, 0, sizeof(req_u));
3059 packet_set_ring(sk, &req_u, 1, 0);
3060 }
3061
3062 if (po->tx_ring.pg_vec) {
3063 memset(&req_u, 0, sizeof(req_u));
3064 packet_set_ring(sk, &req_u, 1, 1);
3065 }
3066 release_sock(sk);
3067
3068 f = fanout_release(sk);
3069
3070 synchronize_net();
3071
3072 kfree(po->rollover);
3073 if (f) {
3074 fanout_release_data(f);
3075 kfree(f);
3076 }
3077 /*
3078 * Now the socket is dead. No more input will appear.
3079 */
3080 sock_orphan(sk);
3081 sock->sk = NULL;
3082
3083 /* Purge queues */
3084
3085 skb_queue_purge(&sk->sk_receive_queue);
3086 packet_free_pending(po);
3087 sk_refcnt_debug_release(sk);
3088
3089 sock_put(sk);
3090 return 0;
3091}
3092
3093/*
3094 * Attach a packet hook.
3095 */
3096
3097static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3098 __be16 proto)
3099{
3100 struct packet_sock *po = pkt_sk(sk);
3101 struct net_device *dev_curr;
3102 __be16 proto_curr;
3103 bool need_rehook;
3104 struct net_device *dev = NULL;
3105 int ret = 0;
3106 bool unlisted = false;
3107
3108 lock_sock(sk);
3109 spin_lock(&po->bind_lock);
3110 rcu_read_lock();
3111
3112 if (po->fanout) {
3113 ret = -EINVAL;
3114 goto out_unlock;
3115 }
3116
3117 if (name) {
3118 dev = dev_get_by_name_rcu(sock_net(sk), name);
3119 if (!dev) {
3120 ret = -ENODEV;
3121 goto out_unlock;
3122 }
3123 } else if (ifindex) {
3124 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3125 if (!dev) {
3126 ret = -ENODEV;
3127 goto out_unlock;
3128 }
3129 }
3130
3131 if (dev)
3132 dev_hold(dev);
3133
3134 proto_curr = po->prot_hook.type;
3135 dev_curr = po->prot_hook.dev;
3136
3137 need_rehook = proto_curr != proto || dev_curr != dev;
3138
3139 if (need_rehook) {
3140 if (po->running) {
3141 rcu_read_unlock();
3142 /* prevents packet_notifier() from calling
3143 * register_prot_hook()
3144 */
3145 po->num = 0;
3146 __unregister_prot_hook(sk, true);
3147 rcu_read_lock();
3148 dev_curr = po->prot_hook.dev;
3149 if (dev)
3150 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3151 dev->ifindex);
3152 }
3153
3154 BUG_ON(po->running);
3155 po->num = proto;
3156 po->prot_hook.type = proto;
3157
3158 if (unlikely(unlisted)) {
3159 dev_put(dev);
3160 po->prot_hook.dev = NULL;
3161 po->ifindex = -1;
3162 packet_cached_dev_reset(po);
3163 } else {
3164 po->prot_hook.dev = dev;
3165 po->ifindex = dev ? dev->ifindex : 0;
3166 packet_cached_dev_assign(po, dev);
3167 }
3168 }
3169 if (dev_curr)
3170 dev_put(dev_curr);
3171
3172 if (proto == 0 || !need_rehook)
3173 goto out_unlock;
3174
3175 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3176 register_prot_hook(sk);
3177 } else {
3178 sk->sk_err = ENETDOWN;
3179 if (!sock_flag(sk, SOCK_DEAD))
3180 sk->sk_error_report(sk);
3181 }
3182
3183out_unlock:
3184 rcu_read_unlock();
3185 spin_unlock(&po->bind_lock);
3186 release_sock(sk);
3187 return ret;
3188}
3189
3190/*
3191 * Bind a packet socket to a device
3192 */
3193
3194static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3195 int addr_len)
3196{
3197 struct sock *sk = sock->sk;
3198 char name[sizeof(uaddr->sa_data) + 1];
3199
3200 /*
3201 * Check legality
3202 */
3203
3204 if (addr_len != sizeof(struct sockaddr))
3205 return -EINVAL;
3206 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3207 * zero-terminated.
3208 */
3209 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3210 name[sizeof(uaddr->sa_data)] = 0;
3211
3212 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3213}
3214
3215static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3216{
3217 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3218 struct sock *sk = sock->sk;
3219
3220 /*
3221 * Check legality
3222 */
3223
3224 if (addr_len < sizeof(struct sockaddr_ll))
3225 return -EINVAL;
3226 if (sll->sll_family != AF_PACKET)
3227 return -EINVAL;
3228
3229 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3230 sll->sll_protocol ? : pkt_sk(sk)->num);
3231}
3232
3233static struct proto packet_proto = {
3234 .name = "PACKET",
3235 .owner = THIS_MODULE,
3236 .obj_size = sizeof(struct packet_sock),
3237};
3238
3239/*
3240 * Create a packet of type SOCK_PACKET.
3241 */
3242
3243static int packet_create(struct net *net, struct socket *sock, int protocol,
3244 int kern)
3245{
3246 struct sock *sk;
3247 struct packet_sock *po;
3248 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3249 int err;
3250
3251 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3252 return -EPERM;
3253 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3254 sock->type != SOCK_PACKET)
3255 return -ESOCKTNOSUPPORT;
3256
3257 sock->state = SS_UNCONNECTED;
3258
3259 err = -ENOBUFS;
3260 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3261 if (sk == NULL)
3262 goto out;
3263
3264 sock->ops = &packet_ops;
3265 if (sock->type == SOCK_PACKET)
3266 sock->ops = &packet_ops_spkt;
3267
3268 sock_init_data(sock, sk);
3269
3270 po = pkt_sk(sk);
3271 init_completion(&po->skb_completion);
3272 sk->sk_family = PF_PACKET;
3273 po->num = proto;
3274 po->xmit = dev_queue_xmit;
3275
3276 err = packet_alloc_pending(po);
3277 if (err)
3278 goto out2;
3279
3280 packet_cached_dev_reset(po);
3281
3282 sk->sk_destruct = packet_sock_destruct;
3283 sk_refcnt_debug_inc(sk);
3284
3285 /*
3286 * Attach a protocol block
3287 */
3288
3289 spin_lock_init(&po->bind_lock);
3290 mutex_init(&po->pg_vec_lock);
3291 po->rollover = NULL;
3292 po->prot_hook.func = packet_rcv;
3293
3294 if (sock->type == SOCK_PACKET)
3295 po->prot_hook.func = packet_rcv_spkt;
3296
3297 po->prot_hook.af_packet_priv = sk;
3298
3299 if (proto) {
3300 po->prot_hook.type = proto;
3301 __register_prot_hook(sk);
3302 }
3303
3304 mutex_lock(&net->packet.sklist_lock);
3305 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3306 mutex_unlock(&net->packet.sklist_lock);
3307
3308 preempt_disable();
3309 sock_prot_inuse_add(net, &packet_proto, 1);
3310 preempt_enable();
3311
3312 return 0;
3313out2:
3314 sk_free(sk);
3315out:
3316 return err;
3317}
3318
3319/*
3320 * Pull a packet from our receive queue and hand it to the user.
3321 * If necessary we block.
3322 */
3323
3324static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3325 int flags)
3326{
3327 struct sock *sk = sock->sk;
3328 struct sk_buff *skb;
3329 int copied, err;
3330 int vnet_hdr_len = 0;
3331 unsigned int origlen = 0;
3332
3333 err = -EINVAL;
3334 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3335 goto out;
3336
3337#if 0
3338 /* What error should we return now? EUNATTACH? */
3339 if (pkt_sk(sk)->ifindex < 0)
3340 return -ENODEV;
3341#endif
3342
3343 if (flags & MSG_ERRQUEUE) {
3344 err = sock_recv_errqueue(sk, msg, len,
3345 SOL_PACKET, PACKET_TX_TIMESTAMP);
3346 goto out;
3347 }
3348
3349 /*
3350 * Call the generic datagram receiver. This handles all sorts
3351 * of horrible races and re-entrancy so we can forget about it
3352 * in the protocol layers.
3353 *
3354 * Now it will return ENETDOWN, if device have just gone down,
3355 * but then it will block.
3356 */
3357
3358 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3359
3360 /*
3361 * An error occurred so return it. Because skb_recv_datagram()
3362 * handles the blocking we don't see and worry about blocking
3363 * retries.
3364 */
3365
3366 if (skb == NULL)
3367 goto out;
3368
3369 packet_rcv_try_clear_pressure(pkt_sk(sk));
3370
3371 if (pkt_sk(sk)->has_vnet_hdr) {
3372 err = packet_rcv_vnet(msg, skb, &len);
3373 if (err)
3374 goto out_free;
3375 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3376 }
3377
3378 /* You lose any data beyond the buffer you gave. If it worries
3379 * a user program they can ask the device for its MTU
3380 * anyway.
3381 */
3382 copied = skb->len;
3383 if (copied > len) {
3384 copied = len;
3385 msg->msg_flags |= MSG_TRUNC;
3386 }
3387
3388 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3389 if (err)
3390 goto out_free;
3391
3392 if (sock->type != SOCK_PACKET) {
3393 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3394
3395 /* Original length was stored in sockaddr_ll fields */
3396 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3397 sll->sll_family = AF_PACKET;
3398 sll->sll_protocol = skb->protocol;
3399 }
3400
3401 sock_recv_ts_and_drops(msg, sk, skb);
3402
3403 if (msg->msg_name) {
3404 int copy_len;
3405
3406 /* If the address length field is there to be filled
3407 * in, we fill it in now.
3408 */
3409 if (sock->type == SOCK_PACKET) {
3410 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3411 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3412 copy_len = msg->msg_namelen;
3413 } else {
3414 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3415
3416 msg->msg_namelen = sll->sll_halen +
3417 offsetof(struct sockaddr_ll, sll_addr);
3418 copy_len = msg->msg_namelen;
3419 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3420 memset(msg->msg_name +
3421 offsetof(struct sockaddr_ll, sll_addr),
3422 0, sizeof(sll->sll_addr));
3423 msg->msg_namelen = sizeof(struct sockaddr_ll);
3424 }
3425 }
3426 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3427 }
3428
3429 if (pkt_sk(sk)->auxdata) {
3430 struct tpacket_auxdata aux;
3431
3432 aux.tp_status = TP_STATUS_USER;
3433 if (skb->ip_summed == CHECKSUM_PARTIAL)
3434 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3435 else if (skb->pkt_type != PACKET_OUTGOING &&
3436 (skb->ip_summed == CHECKSUM_COMPLETE ||
3437 skb_csum_unnecessary(skb)))
3438 aux.tp_status |= TP_STATUS_CSUM_VALID;
3439
3440 aux.tp_len = origlen;
3441 aux.tp_snaplen = skb->len;
3442 aux.tp_mac = 0;
3443 aux.tp_net = skb_network_offset(skb);
3444 if (skb_vlan_tag_present(skb)) {
3445 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3446 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3447 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3448 } else {
3449 aux.tp_vlan_tci = 0;
3450 aux.tp_vlan_tpid = 0;
3451 }
3452 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3453 }
3454
3455 /*
3456 * Free or return the buffer as appropriate. Again this
3457 * hides all the races and re-entrancy issues from us.
3458 */
3459 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3460
3461out_free:
3462 skb_free_datagram(sk, skb);
3463out:
3464 return err;
3465}
3466
3467static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3468 int peer)
3469{
3470 struct net_device *dev;
3471 struct sock *sk = sock->sk;
3472
3473 if (peer)
3474 return -EOPNOTSUPP;
3475
3476 uaddr->sa_family = AF_PACKET;
3477 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3478 rcu_read_lock();
3479 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3480 if (dev)
3481 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3482 rcu_read_unlock();
3483
3484 return sizeof(*uaddr);
3485}
3486
3487static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3488 int peer)
3489{
3490 struct net_device *dev;
3491 struct sock *sk = sock->sk;
3492 struct packet_sock *po = pkt_sk(sk);
3493 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3494
3495 if (peer)
3496 return -EOPNOTSUPP;
3497
3498 sll->sll_family = AF_PACKET;
3499 sll->sll_ifindex = po->ifindex;
3500 sll->sll_protocol = po->num;
3501 sll->sll_pkttype = 0;
3502 rcu_read_lock();
3503 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3504 if (dev) {
3505 sll->sll_hatype = dev->type;
3506 sll->sll_halen = dev->addr_len;
3507 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3508 } else {
3509 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3510 sll->sll_halen = 0;
3511 }
3512 rcu_read_unlock();
3513
3514 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3515}
3516
3517static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3518 int what)
3519{
3520 switch (i->type) {
3521 case PACKET_MR_MULTICAST:
3522 if (i->alen != dev->addr_len)
3523 return -EINVAL;
3524 if (what > 0)
3525 return dev_mc_add(dev, i->addr);
3526 else
3527 return dev_mc_del(dev, i->addr);
3528 break;
3529 case PACKET_MR_PROMISC:
3530 return dev_set_promiscuity(dev, what);
3531 case PACKET_MR_ALLMULTI:
3532 return dev_set_allmulti(dev, what);
3533 case PACKET_MR_UNICAST:
3534 if (i->alen != dev->addr_len)
3535 return -EINVAL;
3536 if (what > 0)
3537 return dev_uc_add(dev, i->addr);
3538 else
3539 return dev_uc_del(dev, i->addr);
3540 break;
3541 default:
3542 break;
3543 }
3544 return 0;
3545}
3546
3547static void packet_dev_mclist_delete(struct net_device *dev,
3548 struct packet_mclist **mlp)
3549{
3550 struct packet_mclist *ml;
3551
3552 while ((ml = *mlp) != NULL) {
3553 if (ml->ifindex == dev->ifindex) {
3554 packet_dev_mc(dev, ml, -1);
3555 *mlp = ml->next;
3556 kfree(ml);
3557 } else
3558 mlp = &ml->next;
3559 }
3560}
3561
3562static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3563{
3564 struct packet_sock *po = pkt_sk(sk);
3565 struct packet_mclist *ml, *i;
3566 struct net_device *dev;
3567 int err;
3568
3569 rtnl_lock();
3570
3571 err = -ENODEV;
3572 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3573 if (!dev)
3574 goto done;
3575
3576 err = -EINVAL;
3577 if (mreq->mr_alen > dev->addr_len)
3578 goto done;
3579
3580 err = -ENOBUFS;
3581 i = kmalloc(sizeof(*i), GFP_KERNEL);
3582 if (i == NULL)
3583 goto done;
3584
3585 err = 0;
3586 for (ml = po->mclist; ml; ml = ml->next) {
3587 if (ml->ifindex == mreq->mr_ifindex &&
3588 ml->type == mreq->mr_type &&
3589 ml->alen == mreq->mr_alen &&
3590 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3591 ml->count++;
3592 /* Free the new element ... */
3593 kfree(i);
3594 goto done;
3595 }
3596 }
3597
3598 i->type = mreq->mr_type;
3599 i->ifindex = mreq->mr_ifindex;
3600 i->alen = mreq->mr_alen;
3601 memcpy(i->addr, mreq->mr_address, i->alen);
3602 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3603 i->count = 1;
3604 i->next = po->mclist;
3605 po->mclist = i;
3606 err = packet_dev_mc(dev, i, 1);
3607 if (err) {
3608 po->mclist = i->next;
3609 kfree(i);
3610 }
3611
3612done:
3613 rtnl_unlock();
3614 return err;
3615}
3616
3617static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3618{
3619 struct packet_mclist *ml, **mlp;
3620
3621 rtnl_lock();
3622
3623 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3624 if (ml->ifindex == mreq->mr_ifindex &&
3625 ml->type == mreq->mr_type &&
3626 ml->alen == mreq->mr_alen &&
3627 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3628 if (--ml->count == 0) {
3629 struct net_device *dev;
3630 *mlp = ml->next;
3631 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3632 if (dev)
3633 packet_dev_mc(dev, ml, -1);
3634 kfree(ml);
3635 }
3636 break;
3637 }
3638 }
3639 rtnl_unlock();
3640 return 0;
3641}
3642
3643static void packet_flush_mclist(struct sock *sk)
3644{
3645 struct packet_sock *po = pkt_sk(sk);
3646 struct packet_mclist *ml;
3647
3648 if (!po->mclist)
3649 return;
3650
3651 rtnl_lock();
3652 while ((ml = po->mclist) != NULL) {
3653 struct net_device *dev;
3654
3655 po->mclist = ml->next;
3656 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3657 if (dev != NULL)
3658 packet_dev_mc(dev, ml, -1);
3659 kfree(ml);
3660 }
3661 rtnl_unlock();
3662}
3663
3664static int
3665packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3666 unsigned int optlen)
3667{
3668 struct sock *sk = sock->sk;
3669 struct packet_sock *po = pkt_sk(sk);
3670 int ret;
3671
3672 if (level != SOL_PACKET)
3673 return -ENOPROTOOPT;
3674
3675 switch (optname) {
3676 case PACKET_ADD_MEMBERSHIP:
3677 case PACKET_DROP_MEMBERSHIP:
3678 {
3679 struct packet_mreq_max mreq;
3680 int len = optlen;
3681 memset(&mreq, 0, sizeof(mreq));
3682 if (len < sizeof(struct packet_mreq))
3683 return -EINVAL;
3684 if (len > sizeof(mreq))
3685 len = sizeof(mreq);
3686 if (copy_from_sockptr(&mreq, optval, len))
3687 return -EFAULT;
3688 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3689 return -EINVAL;
3690 if (optname == PACKET_ADD_MEMBERSHIP)
3691 ret = packet_mc_add(sk, &mreq);
3692 else
3693 ret = packet_mc_drop(sk, &mreq);
3694 return ret;
3695 }
3696
3697 case PACKET_RX_RING:
3698 case PACKET_TX_RING:
3699 {
3700 union tpacket_req_u req_u;
3701 int len;
3702
3703 lock_sock(sk);
3704 switch (po->tp_version) {
3705 case TPACKET_V1:
3706 case TPACKET_V2:
3707 len = sizeof(req_u.req);
3708 break;
3709 case TPACKET_V3:
3710 default:
3711 len = sizeof(req_u.req3);
3712 break;
3713 }
3714 if (optlen < len) {
3715 ret = -EINVAL;
3716 } else {
3717 if (copy_from_sockptr(&req_u.req, optval, len))
3718 ret = -EFAULT;
3719 else
3720 ret = packet_set_ring(sk, &req_u, 0,
3721 optname == PACKET_TX_RING);
3722 }
3723 release_sock(sk);
3724 return ret;
3725 }
3726 case PACKET_COPY_THRESH:
3727 {
3728 int val;
3729
3730 if (optlen != sizeof(val))
3731 return -EINVAL;
3732 if (copy_from_sockptr(&val, optval, sizeof(val)))
3733 return -EFAULT;
3734
3735 pkt_sk(sk)->copy_thresh = val;
3736 return 0;
3737 }
3738 case PACKET_VERSION:
3739 {
3740 int val;
3741
3742 if (optlen != sizeof(val))
3743 return -EINVAL;
3744 if (copy_from_sockptr(&val, optval, sizeof(val)))
3745 return -EFAULT;
3746 switch (val) {
3747 case TPACKET_V1:
3748 case TPACKET_V2:
3749 case TPACKET_V3:
3750 break;
3751 default:
3752 return -EINVAL;
3753 }
3754 lock_sock(sk);
3755 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3756 ret = -EBUSY;
3757 } else {
3758 po->tp_version = val;
3759 ret = 0;
3760 }
3761 release_sock(sk);
3762 return ret;
3763 }
3764 case PACKET_RESERVE:
3765 {
3766 unsigned int val;
3767
3768 if (optlen != sizeof(val))
3769 return -EINVAL;
3770 if (copy_from_sockptr(&val, optval, sizeof(val)))
3771 return -EFAULT;
3772 if (val > INT_MAX)
3773 return -EINVAL;
3774 lock_sock(sk);
3775 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3776 ret = -EBUSY;
3777 } else {
3778 po->tp_reserve = val;
3779 ret = 0;
3780 }
3781 release_sock(sk);
3782 return ret;
3783 }
3784 case PACKET_LOSS:
3785 {
3786 unsigned int val;
3787
3788 if (optlen != sizeof(val))
3789 return -EINVAL;
3790 if (copy_from_sockptr(&val, optval, sizeof(val)))
3791 return -EFAULT;
3792
3793 lock_sock(sk);
3794 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3795 ret = -EBUSY;
3796 } else {
3797 po->tp_loss = !!val;
3798 ret = 0;
3799 }
3800 release_sock(sk);
3801 return ret;
3802 }
3803 case PACKET_AUXDATA:
3804 {
3805 int val;
3806
3807 if (optlen < sizeof(val))
3808 return -EINVAL;
3809 if (copy_from_sockptr(&val, optval, sizeof(val)))
3810 return -EFAULT;
3811
3812 lock_sock(sk);
3813 po->auxdata = !!val;
3814 release_sock(sk);
3815 return 0;
3816 }
3817 case PACKET_ORIGDEV:
3818 {
3819 int val;
3820
3821 if (optlen < sizeof(val))
3822 return -EINVAL;
3823 if (copy_from_sockptr(&val, optval, sizeof(val)))
3824 return -EFAULT;
3825
3826 lock_sock(sk);
3827 po->origdev = !!val;
3828 release_sock(sk);
3829 return 0;
3830 }
3831 case PACKET_VNET_HDR:
3832 {
3833 int val;
3834
3835 if (sock->type != SOCK_RAW)
3836 return -EINVAL;
3837 if (optlen < sizeof(val))
3838 return -EINVAL;
3839 if (copy_from_sockptr(&val, optval, sizeof(val)))
3840 return -EFAULT;
3841
3842 lock_sock(sk);
3843 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3844 ret = -EBUSY;
3845 } else {
3846 po->has_vnet_hdr = !!val;
3847 ret = 0;
3848 }
3849 release_sock(sk);
3850 return ret;
3851 }
3852 case PACKET_TIMESTAMP:
3853 {
3854 int val;
3855
3856 if (optlen != sizeof(val))
3857 return -EINVAL;
3858 if (copy_from_sockptr(&val, optval, sizeof(val)))
3859 return -EFAULT;
3860
3861 po->tp_tstamp = val;
3862 return 0;
3863 }
3864 case PACKET_FANOUT:
3865 {
3866 int val;
3867
3868 if (optlen != sizeof(val))
3869 return -EINVAL;
3870 if (copy_from_sockptr(&val, optval, sizeof(val)))
3871 return -EFAULT;
3872
3873 return fanout_add(sk, val & 0xffff, val >> 16);
3874 }
3875 case PACKET_FANOUT_DATA:
3876 {
3877 if (!po->fanout)
3878 return -EINVAL;
3879
3880 return fanout_set_data(po, optval, optlen);
3881 }
3882 case PACKET_IGNORE_OUTGOING:
3883 {
3884 int val;
3885
3886 if (optlen != sizeof(val))
3887 return -EINVAL;
3888 if (copy_from_sockptr(&val, optval, sizeof(val)))
3889 return -EFAULT;
3890 if (val < 0 || val > 1)
3891 return -EINVAL;
3892
3893 po->prot_hook.ignore_outgoing = !!val;
3894 return 0;
3895 }
3896 case PACKET_TX_HAS_OFF:
3897 {
3898 unsigned int val;
3899
3900 if (optlen != sizeof(val))
3901 return -EINVAL;
3902 if (copy_from_sockptr(&val, optval, sizeof(val)))
3903 return -EFAULT;
3904
3905 lock_sock(sk);
3906 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3907 ret = -EBUSY;
3908 } else {
3909 po->tp_tx_has_off = !!val;
3910 ret = 0;
3911 }
3912 release_sock(sk);
3913 return 0;
3914 }
3915 case PACKET_QDISC_BYPASS:
3916 {
3917 int val;
3918
3919 if (optlen != sizeof(val))
3920 return -EINVAL;
3921 if (copy_from_sockptr(&val, optval, sizeof(val)))
3922 return -EFAULT;
3923
3924 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3925 return 0;
3926 }
3927 default:
3928 return -ENOPROTOOPT;
3929 }
3930}
3931
3932static int packet_getsockopt(struct socket *sock, int level, int optname,
3933 char __user *optval, int __user *optlen)
3934{
3935 int len;
3936 int val, lv = sizeof(val);
3937 struct sock *sk = sock->sk;
3938 struct packet_sock *po = pkt_sk(sk);
3939 void *data = &val;
3940 union tpacket_stats_u st;
3941 struct tpacket_rollover_stats rstats;
3942 int drops;
3943
3944 if (level != SOL_PACKET)
3945 return -ENOPROTOOPT;
3946
3947 if (get_user(len, optlen))
3948 return -EFAULT;
3949
3950 if (len < 0)
3951 return -EINVAL;
3952
3953 switch (optname) {
3954 case PACKET_STATISTICS:
3955 spin_lock_bh(&sk->sk_receive_queue.lock);
3956 memcpy(&st, &po->stats, sizeof(st));
3957 memset(&po->stats, 0, sizeof(po->stats));
3958 spin_unlock_bh(&sk->sk_receive_queue.lock);
3959 drops = atomic_xchg(&po->tp_drops, 0);
3960
3961 if (po->tp_version == TPACKET_V3) {
3962 lv = sizeof(struct tpacket_stats_v3);
3963 st.stats3.tp_drops = drops;
3964 st.stats3.tp_packets += drops;
3965 data = &st.stats3;
3966 } else {
3967 lv = sizeof(struct tpacket_stats);
3968 st.stats1.tp_drops = drops;
3969 st.stats1.tp_packets += drops;
3970 data = &st.stats1;
3971 }
3972
3973 break;
3974 case PACKET_AUXDATA:
3975 val = po->auxdata;
3976 break;
3977 case PACKET_ORIGDEV:
3978 val = po->origdev;
3979 break;
3980 case PACKET_VNET_HDR:
3981 val = po->has_vnet_hdr;
3982 break;
3983 case PACKET_VERSION:
3984 val = po->tp_version;
3985 break;
3986 case PACKET_HDRLEN:
3987 if (len > sizeof(int))
3988 len = sizeof(int);
3989 if (len < sizeof(int))
3990 return -EINVAL;
3991 if (copy_from_user(&val, optval, len))
3992 return -EFAULT;
3993 switch (val) {
3994 case TPACKET_V1:
3995 val = sizeof(struct tpacket_hdr);
3996 break;
3997 case TPACKET_V2:
3998 val = sizeof(struct tpacket2_hdr);
3999 break;
4000 case TPACKET_V3:
4001 val = sizeof(struct tpacket3_hdr);
4002 break;
4003 default:
4004 return -EINVAL;
4005 }
4006 break;
4007 case PACKET_RESERVE:
4008 val = po->tp_reserve;
4009 break;
4010 case PACKET_LOSS:
4011 val = po->tp_loss;
4012 break;
4013 case PACKET_TIMESTAMP:
4014 val = po->tp_tstamp;
4015 break;
4016 case PACKET_FANOUT:
4017 val = (po->fanout ?
4018 ((u32)po->fanout->id |
4019 ((u32)po->fanout->type << 16) |
4020 ((u32)po->fanout->flags << 24)) :
4021 0);
4022 break;
4023 case PACKET_IGNORE_OUTGOING:
4024 val = po->prot_hook.ignore_outgoing;
4025 break;
4026 case PACKET_ROLLOVER_STATS:
4027 if (!po->rollover)
4028 return -EINVAL;
4029 rstats.tp_all = atomic_long_read(&po->rollover->num);
4030 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4031 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4032 data = &rstats;
4033 lv = sizeof(rstats);
4034 break;
4035 case PACKET_TX_HAS_OFF:
4036 val = po->tp_tx_has_off;
4037 break;
4038 case PACKET_QDISC_BYPASS:
4039 val = packet_use_direct_xmit(po);
4040 break;
4041 default:
4042 return -ENOPROTOOPT;
4043 }
4044
4045 if (len > lv)
4046 len = lv;
4047 if (put_user(len, optlen))
4048 return -EFAULT;
4049 if (copy_to_user(optval, data, len))
4050 return -EFAULT;
4051 return 0;
4052}
4053
4054static int packet_notifier(struct notifier_block *this,
4055 unsigned long msg, void *ptr)
4056{
4057 struct sock *sk;
4058 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4059 struct net *net = dev_net(dev);
4060
4061 rcu_read_lock();
4062 sk_for_each_rcu(sk, &net->packet.sklist) {
4063 struct packet_sock *po = pkt_sk(sk);
4064
4065 switch (msg) {
4066 case NETDEV_UNREGISTER:
4067 if (po->mclist)
4068 packet_dev_mclist_delete(dev, &po->mclist);
4069 fallthrough;
4070
4071 case NETDEV_DOWN:
4072 if (dev->ifindex == po->ifindex) {
4073 spin_lock(&po->bind_lock);
4074 if (po->running) {
4075 __unregister_prot_hook(sk, false);
4076 sk->sk_err = ENETDOWN;
4077 if (!sock_flag(sk, SOCK_DEAD))
4078 sk->sk_error_report(sk);
4079 }
4080 if (msg == NETDEV_UNREGISTER) {
4081 packet_cached_dev_reset(po);
4082 po->ifindex = -1;
4083 if (po->prot_hook.dev)
4084 dev_put(po->prot_hook.dev);
4085 po->prot_hook.dev = NULL;
4086 }
4087 spin_unlock(&po->bind_lock);
4088 }
4089 break;
4090 case NETDEV_UP:
4091 if (dev->ifindex == po->ifindex) {
4092 spin_lock(&po->bind_lock);
4093 if (po->num)
4094 register_prot_hook(sk);
4095 spin_unlock(&po->bind_lock);
4096 }
4097 break;
4098 }
4099 }
4100 rcu_read_unlock();
4101 return NOTIFY_DONE;
4102}
4103
4104
4105static int packet_ioctl(struct socket *sock, unsigned int cmd,
4106 unsigned long arg)
4107{
4108 struct sock *sk = sock->sk;
4109
4110 switch (cmd) {
4111 case SIOCOUTQ:
4112 {
4113 int amount = sk_wmem_alloc_get(sk);
4114
4115 return put_user(amount, (int __user *)arg);
4116 }
4117 case SIOCINQ:
4118 {
4119 struct sk_buff *skb;
4120 int amount = 0;
4121
4122 spin_lock_bh(&sk->sk_receive_queue.lock);
4123 skb = skb_peek(&sk->sk_receive_queue);
4124 if (skb)
4125 amount = skb->len;
4126 spin_unlock_bh(&sk->sk_receive_queue.lock);
4127 return put_user(amount, (int __user *)arg);
4128 }
4129#ifdef CONFIG_INET
4130 case SIOCADDRT:
4131 case SIOCDELRT:
4132 case SIOCDARP:
4133 case SIOCGARP:
4134 case SIOCSARP:
4135 case SIOCGIFADDR:
4136 case SIOCSIFADDR:
4137 case SIOCGIFBRDADDR:
4138 case SIOCSIFBRDADDR:
4139 case SIOCGIFNETMASK:
4140 case SIOCSIFNETMASK:
4141 case SIOCGIFDSTADDR:
4142 case SIOCSIFDSTADDR:
4143 case SIOCSIFFLAGS:
4144 return inet_dgram_ops.ioctl(sock, cmd, arg);
4145#endif
4146
4147 default:
4148 return -ENOIOCTLCMD;
4149 }
4150 return 0;
4151}
4152
4153static __poll_t packet_poll(struct file *file, struct socket *sock,
4154 poll_table *wait)
4155{
4156 struct sock *sk = sock->sk;
4157 struct packet_sock *po = pkt_sk(sk);
4158 __poll_t mask = datagram_poll(file, sock, wait);
4159
4160 spin_lock_bh(&sk->sk_receive_queue.lock);
4161 if (po->rx_ring.pg_vec) {
4162 if (!packet_previous_rx_frame(po, &po->rx_ring,
4163 TP_STATUS_KERNEL))
4164 mask |= EPOLLIN | EPOLLRDNORM;
4165 }
4166 packet_rcv_try_clear_pressure(po);
4167 spin_unlock_bh(&sk->sk_receive_queue.lock);
4168 spin_lock_bh(&sk->sk_write_queue.lock);
4169 if (po->tx_ring.pg_vec) {
4170 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4171 mask |= EPOLLOUT | EPOLLWRNORM;
4172 }
4173 spin_unlock_bh(&sk->sk_write_queue.lock);
4174 return mask;
4175}
4176
4177
4178/* Dirty? Well, I still did not learn better way to account
4179 * for user mmaps.
4180 */
4181
4182static void packet_mm_open(struct vm_area_struct *vma)
4183{
4184 struct file *file = vma->vm_file;
4185 struct socket *sock = file->private_data;
4186 struct sock *sk = sock->sk;
4187
4188 if (sk)
4189 atomic_inc(&pkt_sk(sk)->mapped);
4190}
4191
4192static void packet_mm_close(struct vm_area_struct *vma)
4193{
4194 struct file *file = vma->vm_file;
4195 struct socket *sock = file->private_data;
4196 struct sock *sk = sock->sk;
4197
4198 if (sk)
4199 atomic_dec(&pkt_sk(sk)->mapped);
4200}
4201
4202static const struct vm_operations_struct packet_mmap_ops = {
4203 .open = packet_mm_open,
4204 .close = packet_mm_close,
4205};
4206
4207static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4208 unsigned int len)
4209{
4210 int i;
4211
4212 for (i = 0; i < len; i++) {
4213 if (likely(pg_vec[i].buffer)) {
4214 if (is_vmalloc_addr(pg_vec[i].buffer))
4215 vfree(pg_vec[i].buffer);
4216 else
4217 free_pages((unsigned long)pg_vec[i].buffer,
4218 order);
4219 pg_vec[i].buffer = NULL;
4220 }
4221 }
4222 kfree(pg_vec);
4223}
4224
4225static char *alloc_one_pg_vec_page(unsigned long order)
4226{
4227 char *buffer;
4228 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4229 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4230
4231 buffer = (char *) __get_free_pages(gfp_flags, order);
4232 if (buffer)
4233 return buffer;
4234
4235 /* __get_free_pages failed, fall back to vmalloc */
4236 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4237 if (buffer)
4238 return buffer;
4239
4240 /* vmalloc failed, lets dig into swap here */
4241 gfp_flags &= ~__GFP_NORETRY;
4242 buffer = (char *) __get_free_pages(gfp_flags, order);
4243 if (buffer)
4244 return buffer;
4245
4246 /* complete and utter failure */
4247 return NULL;
4248}
4249
4250static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4251{
4252 unsigned int block_nr = req->tp_block_nr;
4253 struct pgv *pg_vec;
4254 int i;
4255
4256 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4257 if (unlikely(!pg_vec))
4258 goto out;
4259
4260 for (i = 0; i < block_nr; i++) {
4261 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4262 if (unlikely(!pg_vec[i].buffer))
4263 goto out_free_pgvec;
4264 }
4265
4266out:
4267 return pg_vec;
4268
4269out_free_pgvec:
4270 free_pg_vec(pg_vec, order, block_nr);
4271 pg_vec = NULL;
4272 goto out;
4273}
4274
4275static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4276 int closing, int tx_ring)
4277{
4278 struct pgv *pg_vec = NULL;
4279 struct packet_sock *po = pkt_sk(sk);
4280 unsigned long *rx_owner_map = NULL;
4281 int was_running, order = 0;
4282 struct packet_ring_buffer *rb;
4283 struct sk_buff_head *rb_queue;
4284 __be16 num;
4285 int err;
4286 /* Added to avoid minimal code churn */
4287 struct tpacket_req *req = &req_u->req;
4288
4289 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4290 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4291
4292 err = -EBUSY;
4293 if (!closing) {
4294 if (atomic_read(&po->mapped))
4295 goto out;
4296 if (packet_read_pending(rb))
4297 goto out;
4298 }
4299
4300 if (req->tp_block_nr) {
4301 unsigned int min_frame_size;
4302
4303 /* Sanity tests and some calculations */
4304 err = -EBUSY;
4305 if (unlikely(rb->pg_vec))
4306 goto out;
4307
4308 switch (po->tp_version) {
4309 case TPACKET_V1:
4310 po->tp_hdrlen = TPACKET_HDRLEN;
4311 break;
4312 case TPACKET_V2:
4313 po->tp_hdrlen = TPACKET2_HDRLEN;
4314 break;
4315 case TPACKET_V3:
4316 po->tp_hdrlen = TPACKET3_HDRLEN;
4317 break;
4318 }
4319
4320 err = -EINVAL;
4321 if (unlikely((int)req->tp_block_size <= 0))
4322 goto out;
4323 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4324 goto out;
4325 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4326 if (po->tp_version >= TPACKET_V3 &&
4327 req->tp_block_size <
4328 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4329 goto out;
4330 if (unlikely(req->tp_frame_size < min_frame_size))
4331 goto out;
4332 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4333 goto out;
4334
4335 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4336 if (unlikely(rb->frames_per_block == 0))
4337 goto out;
4338 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4339 goto out;
4340 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4341 req->tp_frame_nr))
4342 goto out;
4343
4344 err = -ENOMEM;
4345 order = get_order(req->tp_block_size);
4346 pg_vec = alloc_pg_vec(req, order);
4347 if (unlikely(!pg_vec))
4348 goto out;
4349 switch (po->tp_version) {
4350 case TPACKET_V3:
4351 /* Block transmit is not supported yet */
4352 if (!tx_ring) {
4353 init_prb_bdqc(po, rb, pg_vec, req_u);
4354 } else {
4355 struct tpacket_req3 *req3 = &req_u->req3;
4356
4357 if (req3->tp_retire_blk_tov ||
4358 req3->tp_sizeof_priv ||
4359 req3->tp_feature_req_word) {
4360 err = -EINVAL;
4361 goto out_free_pg_vec;
4362 }
4363 }
4364 break;
4365 default:
4366 if (!tx_ring) {
4367 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4368 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4369 if (!rx_owner_map)
4370 goto out_free_pg_vec;
4371 }
4372 break;
4373 }
4374 }
4375 /* Done */
4376 else {
4377 err = -EINVAL;
4378 if (unlikely(req->tp_frame_nr))
4379 goto out;
4380 }
4381
4382
4383 /* Detach socket from network */
4384 spin_lock(&po->bind_lock);
4385 was_running = po->running;
4386 num = po->num;
4387 if (was_running) {
4388 po->num = 0;
4389 __unregister_prot_hook(sk, false);
4390 }
4391 spin_unlock(&po->bind_lock);
4392
4393 synchronize_net();
4394
4395 err = -EBUSY;
4396 mutex_lock(&po->pg_vec_lock);
4397 if (closing || atomic_read(&po->mapped) == 0) {
4398 err = 0;
4399 spin_lock_bh(&rb_queue->lock);
4400 swap(rb->pg_vec, pg_vec);
4401 if (po->tp_version <= TPACKET_V2)
4402 swap(rb->rx_owner_map, rx_owner_map);
4403 rb->frame_max = (req->tp_frame_nr - 1);
4404 rb->head = 0;
4405 rb->frame_size = req->tp_frame_size;
4406 spin_unlock_bh(&rb_queue->lock);
4407
4408 swap(rb->pg_vec_order, order);
4409 swap(rb->pg_vec_len, req->tp_block_nr);
4410
4411 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4412 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4413 tpacket_rcv : packet_rcv;
4414 skb_queue_purge(rb_queue);
4415 if (atomic_read(&po->mapped))
4416 pr_err("packet_mmap: vma is busy: %d\n",
4417 atomic_read(&po->mapped));
4418 }
4419 mutex_unlock(&po->pg_vec_lock);
4420
4421 spin_lock(&po->bind_lock);
4422 if (was_running) {
4423 po->num = num;
4424 register_prot_hook(sk);
4425 }
4426 spin_unlock(&po->bind_lock);
4427 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4428 /* Because we don't support block-based V3 on tx-ring */
4429 if (!tx_ring)
4430 prb_shutdown_retire_blk_timer(po, rb_queue);
4431 }
4432
4433out_free_pg_vec:
4434 bitmap_free(rx_owner_map);
4435 if (pg_vec)
4436 free_pg_vec(pg_vec, order, req->tp_block_nr);
4437out:
4438 return err;
4439}
4440
4441static int packet_mmap(struct file *file, struct socket *sock,
4442 struct vm_area_struct *vma)
4443{
4444 struct sock *sk = sock->sk;
4445 struct packet_sock *po = pkt_sk(sk);
4446 unsigned long size, expected_size;
4447 struct packet_ring_buffer *rb;
4448 unsigned long start;
4449 int err = -EINVAL;
4450 int i;
4451
4452 if (vma->vm_pgoff)
4453 return -EINVAL;
4454
4455 mutex_lock(&po->pg_vec_lock);
4456
4457 expected_size = 0;
4458 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4459 if (rb->pg_vec) {
4460 expected_size += rb->pg_vec_len
4461 * rb->pg_vec_pages
4462 * PAGE_SIZE;
4463 }
4464 }
4465
4466 if (expected_size == 0)
4467 goto out;
4468
4469 size = vma->vm_end - vma->vm_start;
4470 if (size != expected_size)
4471 goto out;
4472
4473 start = vma->vm_start;
4474 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4475 if (rb->pg_vec == NULL)
4476 continue;
4477
4478 for (i = 0; i < rb->pg_vec_len; i++) {
4479 struct page *page;
4480 void *kaddr = rb->pg_vec[i].buffer;
4481 int pg_num;
4482
4483 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4484 page = pgv_to_page(kaddr);
4485 err = vm_insert_page(vma, start, page);
4486 if (unlikely(err))
4487 goto out;
4488 start += PAGE_SIZE;
4489 kaddr += PAGE_SIZE;
4490 }
4491 }
4492 }
4493
4494 atomic_inc(&po->mapped);
4495 vma->vm_ops = &packet_mmap_ops;
4496 err = 0;
4497
4498out:
4499 mutex_unlock(&po->pg_vec_lock);
4500 return err;
4501}
4502
4503static const struct proto_ops packet_ops_spkt = {
4504 .family = PF_PACKET,
4505 .owner = THIS_MODULE,
4506 .release = packet_release,
4507 .bind = packet_bind_spkt,
4508 .connect = sock_no_connect,
4509 .socketpair = sock_no_socketpair,
4510 .accept = sock_no_accept,
4511 .getname = packet_getname_spkt,
4512 .poll = datagram_poll,
4513 .ioctl = packet_ioctl,
4514 .gettstamp = sock_gettstamp,
4515 .listen = sock_no_listen,
4516 .shutdown = sock_no_shutdown,
4517 .sendmsg = packet_sendmsg_spkt,
4518 .recvmsg = packet_recvmsg,
4519 .mmap = sock_no_mmap,
4520 .sendpage = sock_no_sendpage,
4521};
4522
4523static const struct proto_ops packet_ops = {
4524 .family = PF_PACKET,
4525 .owner = THIS_MODULE,
4526 .release = packet_release,
4527 .bind = packet_bind,
4528 .connect = sock_no_connect,
4529 .socketpair = sock_no_socketpair,
4530 .accept = sock_no_accept,
4531 .getname = packet_getname,
4532 .poll = packet_poll,
4533 .ioctl = packet_ioctl,
4534 .gettstamp = sock_gettstamp,
4535 .listen = sock_no_listen,
4536 .shutdown = sock_no_shutdown,
4537 .setsockopt = packet_setsockopt,
4538 .getsockopt = packet_getsockopt,
4539 .sendmsg = packet_sendmsg,
4540 .recvmsg = packet_recvmsg,
4541 .mmap = packet_mmap,
4542 .sendpage = sock_no_sendpage,
4543};
4544
4545static const struct net_proto_family packet_family_ops = {
4546 .family = PF_PACKET,
4547 .create = packet_create,
4548 .owner = THIS_MODULE,
4549};
4550
4551static struct notifier_block packet_netdev_notifier = {
4552 .notifier_call = packet_notifier,
4553};
4554
4555#ifdef CONFIG_PROC_FS
4556
4557static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4558 __acquires(RCU)
4559{
4560 struct net *net = seq_file_net(seq);
4561
4562 rcu_read_lock();
4563 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4564}
4565
4566static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4567{
4568 struct net *net = seq_file_net(seq);
4569 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4570}
4571
4572static void packet_seq_stop(struct seq_file *seq, void *v)
4573 __releases(RCU)
4574{
4575 rcu_read_unlock();
4576}
4577
4578static int packet_seq_show(struct seq_file *seq, void *v)
4579{
4580 if (v == SEQ_START_TOKEN)
4581 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4582 else {
4583 struct sock *s = sk_entry(v);
4584 const struct packet_sock *po = pkt_sk(s);
4585
4586 seq_printf(seq,
4587 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4588 s,
4589 refcount_read(&s->sk_refcnt),
4590 s->sk_type,
4591 ntohs(po->num),
4592 po->ifindex,
4593 po->running,
4594 atomic_read(&s->sk_rmem_alloc),
4595 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4596 sock_i_ino(s));
4597 }
4598
4599 return 0;
4600}
4601
4602static const struct seq_operations packet_seq_ops = {
4603 .start = packet_seq_start,
4604 .next = packet_seq_next,
4605 .stop = packet_seq_stop,
4606 .show = packet_seq_show,
4607};
4608#endif
4609
4610static int __net_init packet_net_init(struct net *net)
4611{
4612 mutex_init(&net->packet.sklist_lock);
4613 INIT_HLIST_HEAD(&net->packet.sklist);
4614
4615 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4616 sizeof(struct seq_net_private)))
4617 return -ENOMEM;
4618
4619 return 0;
4620}
4621
4622static void __net_exit packet_net_exit(struct net *net)
4623{
4624 remove_proc_entry("packet", net->proc_net);
4625 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4626}
4627
4628static struct pernet_operations packet_net_ops = {
4629 .init = packet_net_init,
4630 .exit = packet_net_exit,
4631};
4632
4633
4634static void __exit packet_exit(void)
4635{
4636 unregister_netdevice_notifier(&packet_netdev_notifier);
4637 unregister_pernet_subsys(&packet_net_ops);
4638 sock_unregister(PF_PACKET);
4639 proto_unregister(&packet_proto);
4640}
4641
4642static int __init packet_init(void)
4643{
4644 int rc;
4645
4646 rc = proto_register(&packet_proto, 0);
4647 if (rc)
4648 goto out;
4649 rc = sock_register(&packet_family_ops);
4650 if (rc)
4651 goto out_proto;
4652 rc = register_pernet_subsys(&packet_net_ops);
4653 if (rc)
4654 goto out_sock;
4655 rc = register_netdevice_notifier(&packet_netdev_notifier);
4656 if (rc)
4657 goto out_pernet;
4658
4659 return 0;
4660
4661out_pernet:
4662 unregister_pernet_subsys(&packet_net_ops);
4663out_sock:
4664 sock_unregister(PF_PACKET);
4665out_proto:
4666 proto_unregister(&packet_proto);
4667out:
4668 return rc;
4669}
4670
4671module_init(packet_init);
4672module_exit(packet_exit);
4673MODULE_LICENSE("GPL");
4674MODULE_ALIAS_NETPROTO(PF_PACKET);