?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
staging: erofs: integrate decompression inplace
[linux/fpc-iii.git] / net / packet / af_packet.c
bloba29d66da7394bb48c4012b29ff2ad3d256cf7bff
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
105 On receive:
106 -----------
107
108 Incoming, dev->hard_header!=NULL
109 mac_header -> ll header
110 data -> data
111
112 Outgoing, dev->hard_header!=NULL
113 mac_header -> ll header
114 data -> ll header
115
116 Incoming, 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
122 Outgoing, dev->hard_header==NULL
123 mac_header -> data. ll header is still not built!
124 data -> data
125
126 Resume
127 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
128
129
130 On transmit:
131 ------------
132
133 dev->hard_header != NULL
134 mac_header -> ll header
135 data -> ll header
136
137 dev->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 */
150 struct 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
157 union tpacket_uhdr {
158 struct tpacket_hdr *h1;
159 struct tpacket2_hdr *h2;
160 struct tpacket3_hdr *h3;
161 void *raw;
162 };
163
164 static 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
182 struct packet_sock;
183 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
184 struct packet_type *pt, struct net_device *orig_dev);
185
186 static void *packet_previous_frame(struct packet_sock *po,
187 struct packet_ring_buffer *rb,
188 int status);
189 static void packet_increment_head(struct packet_ring_buffer *buff);
190 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
191 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
192 struct packet_sock *);
193 static void prb_retire_current_block(struct tpacket_kbdq_core *,
194 struct packet_sock *, unsigned int status);
195 static int prb_queue_frozen(struct tpacket_kbdq_core *);
196 static void prb_open_block(struct tpacket_kbdq_core *,
197 struct tpacket_block_desc *);
198 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
199 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
200 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
201 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
202 struct tpacket3_hdr *);
203 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
204 struct tpacket3_hdr *);
205 static void packet_flush_mclist(struct sock *sk);
206 static u16 packet_pick_tx_queue(struct sk_buff *skb);
207
208 struct 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
235 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
236 static void __fanout_link(struct sock *sk, struct packet_sock *po);
237
238 static int packet_direct_xmit(struct sk_buff *skb)
239 {
240 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
241 }
242
243 static 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
256 static 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
262 static void packet_cached_dev_reset(struct packet_sock *po)
263 {
264 RCU_INIT_POINTER(po->cached_dev, NULL);
265 }
266
267 static bool packet_use_direct_xmit(const struct packet_sock *po)
268 {
269 return po->xmit == packet_direct_xmit;
270 }
271
272 static 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 */
297 static 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
312 static 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 */
324 static 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
346 static 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
354 static 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
361 static 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
387 static int __packet_get_status(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
411 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *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_timespec_cond(shhwtstamps->hwtstamp, ts))
419 return TP_STATUS_TS_RAW_HARDWARE;
420
421 if (ktime_to_timespec_cond(skb->tstamp, ts))
422 return TP_STATUS_TS_SOFTWARE;
423
424 return 0;
425 }
426
427 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
428 struct sk_buff *skb)
429 {
430 union tpacket_uhdr h;
431 struct timespec 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 switch (po->tp_version) {
439 case TPACKET_V1:
440 h.h1->tp_sec = ts.tv_sec;
441 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
442 break;
443 case TPACKET_V2:
444 h.h2->tp_sec = ts.tv_sec;
445 h.h2->tp_nsec = ts.tv_nsec;
446 break;
447 case TPACKET_V3:
448 h.h3->tp_sec = ts.tv_sec;
449 h.h3->tp_nsec = ts.tv_nsec;
450 break;
451 default:
452 WARN(1, "TPACKET version not supported.\n");
453 BUG();
454 }
455
456 /* one flush is safe, as both fields always lie on the same cacheline */
457 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
458 smp_wmb();
459
460 return ts_status;
461 }
462
463 static void *packet_lookup_frame(struct packet_sock *po,
464 struct packet_ring_buffer *rb,
465 unsigned int position,
466 int status)
467 {
468 unsigned int pg_vec_pos, frame_offset;
469 union tpacket_uhdr h;
470
471 pg_vec_pos = position / rb->frames_per_block;
472 frame_offset = position % rb->frames_per_block;
473
474 h.raw = rb->pg_vec[pg_vec_pos].buffer +
475 (frame_offset * rb->frame_size);
476
477 if (status != __packet_get_status(po, h.raw))
478 return NULL;
479
480 return h.raw;
481 }
482
483 static void *packet_current_frame(struct packet_sock *po,
484 struct packet_ring_buffer *rb,
485 int status)
486 {
487 return packet_lookup_frame(po, rb, rb->head, status);
488 }
489
490 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
491 {
492 del_timer_sync(&pkc->retire_blk_timer);
493 }
494
495 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
496 struct sk_buff_head *rb_queue)
497 {
498 struct tpacket_kbdq_core *pkc;
499
500 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
501
502 spin_lock_bh(&rb_queue->lock);
503 pkc->delete_blk_timer = 1;
504 spin_unlock_bh(&rb_queue->lock);
505
506 prb_del_retire_blk_timer(pkc);
507 }
508
509 static void prb_setup_retire_blk_timer(struct packet_sock *po)
510 {
511 struct tpacket_kbdq_core *pkc;
512
513 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
514 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
515 0);
516 pkc->retire_blk_timer.expires = jiffies;
517 }
518
519 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
520 int blk_size_in_bytes)
521 {
522 struct net_device *dev;
523 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
524 struct ethtool_link_ksettings ecmd;
525 int err;
526
527 rtnl_lock();
528 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
529 if (unlikely(!dev)) {
530 rtnl_unlock();
531 return DEFAULT_PRB_RETIRE_TOV;
532 }
533 err = __ethtool_get_link_ksettings(dev, &ecmd);
534 rtnl_unlock();
535 if (!err) {
536 /*
537 * If the link speed is so slow you don't really
538 * need to worry about perf anyways
539 */
540 if (ecmd.base.speed < SPEED_1000 ||
541 ecmd.base.speed == SPEED_UNKNOWN) {
542 return DEFAULT_PRB_RETIRE_TOV;
543 } else {
544 msec = 1;
545 div = ecmd.base.speed / 1000;
546 }
547 }
548
549 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
550
551 if (div)
552 mbits /= div;
553
554 tmo = mbits * msec;
555
556 if (div)
557 return tmo+1;
558 return tmo;
559 }
560
561 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
562 union tpacket_req_u *req_u)
563 {
564 p1->feature_req_word = req_u->req3.tp_feature_req_word;
565 }
566
567 static void init_prb_bdqc(struct packet_sock *po,
568 struct packet_ring_buffer *rb,
569 struct pgv *pg_vec,
570 union tpacket_req_u *req_u)
571 {
572 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
573 struct tpacket_block_desc *pbd;
574
575 memset(p1, 0x0, sizeof(*p1));
576
577 p1->knxt_seq_num = 1;
578 p1->pkbdq = pg_vec;
579 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
580 p1->pkblk_start = pg_vec[0].buffer;
581 p1->kblk_size = req_u->req3.tp_block_size;
582 p1->knum_blocks = req_u->req3.tp_block_nr;
583 p1->hdrlen = po->tp_hdrlen;
584 p1->version = po->tp_version;
585 p1->last_kactive_blk_num = 0;
586 po->stats.stats3.tp_freeze_q_cnt = 0;
587 if (req_u->req3.tp_retire_blk_tov)
588 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
589 else
590 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
591 req_u->req3.tp_block_size);
592 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
593 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
594
595 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
596 prb_init_ft_ops(p1, req_u);
597 prb_setup_retire_blk_timer(po);
598 prb_open_block(p1, pbd);
599 }
600
601 /* Do NOT update the last_blk_num first.
602 * Assumes sk_buff_head lock is held.
603 */
604 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
605 {
606 mod_timer(&pkc->retire_blk_timer,
607 jiffies + pkc->tov_in_jiffies);
608 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
609 }
610
611 /*
612 * Timer logic:
613 * 1) We refresh the timer only when we open a block.
614 * By doing this we don't waste cycles refreshing the timer
615 * on packet-by-packet basis.
616 *
617 * With a 1MB block-size, on a 1Gbps line, it will take
618 * i) ~8 ms to fill a block + ii) memcpy etc.
619 * In this cut we are not accounting for the memcpy time.
620 *
621 * So, if the user sets the 'tmo' to 10ms then the timer
622 * will never fire while the block is still getting filled
623 * (which is what we want). However, the user could choose
624 * to close a block early and that's fine.
625 *
626 * But when the timer does fire, we check whether or not to refresh it.
627 * Since the tmo granularity is in msecs, it is not too expensive
628 * to refresh the timer, lets say every '8' msecs.
629 * Either the user can set the 'tmo' or we can derive it based on
630 * a) line-speed and b) block-size.
631 * prb_calc_retire_blk_tmo() calculates the tmo.
632 *
633 */
634 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
635 {
636 struct packet_sock *po =
637 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
638 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
639 unsigned int frozen;
640 struct tpacket_block_desc *pbd;
641
642 spin_lock(&po->sk.sk_receive_queue.lock);
643
644 frozen = prb_queue_frozen(pkc);
645 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
646
647 if (unlikely(pkc->delete_blk_timer))
648 goto out;
649
650 /* We only need to plug the race when the block is partially filled.
651 * tpacket_rcv:
652 * lock(); increment BLOCK_NUM_PKTS; unlock()
653 * copy_bits() is in progress ...
654 * timer fires on other cpu:
655 * we can't retire the current block because copy_bits
656 * is in progress.
657 *
658 */
659 if (BLOCK_NUM_PKTS(pbd)) {
660 while (atomic_read(&pkc->blk_fill_in_prog)) {
661 /* Waiting for skb_copy_bits to finish... */
662 cpu_relax();
663 }
664 }
665
666 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
667 if (!frozen) {
668 if (!BLOCK_NUM_PKTS(pbd)) {
669 /* An empty block. Just refresh the timer. */
670 goto refresh_timer;
671 }
672 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
673 if (!prb_dispatch_next_block(pkc, po))
674 goto refresh_timer;
675 else
676 goto out;
677 } else {
678 /* Case 1. Queue was frozen because user-space was
679 * lagging behind.
680 */
681 if (prb_curr_blk_in_use(pbd)) {
682 /*
683 * Ok, user-space is still behind.
684 * So just refresh the timer.
685 */
686 goto refresh_timer;
687 } else {
688 /* Case 2. queue was frozen,user-space caught up,
689 * now the link went idle && the timer fired.
690 * We don't have a block to close.So we open this
691 * block and restart the timer.
692 * opening a block thaws the queue,restarts timer
693 * Thawing/timer-refresh is a side effect.
694 */
695 prb_open_block(pkc, pbd);
696 goto out;
697 }
698 }
699 }
700
701 refresh_timer:
702 _prb_refresh_rx_retire_blk_timer(pkc);
703
704 out:
705 spin_unlock(&po->sk.sk_receive_queue.lock);
706 }
707
708 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
709 struct tpacket_block_desc *pbd1, __u32 status)
710 {
711 /* Flush everything minus the block header */
712
713 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
714 u8 *start, *end;
715
716 start = (u8 *)pbd1;
717
718 /* Skip the block header(we know header WILL fit in 4K) */
719 start += PAGE_SIZE;
720
721 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
722 for (; start < end; start += PAGE_SIZE)
723 flush_dcache_page(pgv_to_page(start));
724
725 smp_wmb();
726 #endif
727
728 /* Now update the block status. */
729
730 BLOCK_STATUS(pbd1) = status;
731
732 /* Flush the block header */
733
734 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
735 start = (u8 *)pbd1;
736 flush_dcache_page(pgv_to_page(start));
737
738 smp_wmb();
739 #endif
740 }
741
742 /*
743 * Side effect:
744 *
745 * 1) flush the block
746 * 2) Increment active_blk_num
747 *
748 * Note:We DONT refresh the timer on purpose.
749 * Because almost always the next block will be opened.
750 */
751 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
752 struct tpacket_block_desc *pbd1,
753 struct packet_sock *po, unsigned int stat)
754 {
755 __u32 status = TP_STATUS_USER | stat;
756
757 struct tpacket3_hdr *last_pkt;
758 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
759 struct sock *sk = &po->sk;
760
761 if (po->stats.stats3.tp_drops)
762 status |= TP_STATUS_LOSING;
763
764 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
765 last_pkt->tp_next_offset = 0;
766
767 /* Get the ts of the last pkt */
768 if (BLOCK_NUM_PKTS(pbd1)) {
769 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
770 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
771 } else {
772 /* Ok, we tmo'd - so get the current time.
773 *
774 * It shouldn't really happen as we don't close empty
775 * blocks. See prb_retire_rx_blk_timer_expired().
776 */
777 struct timespec ts;
778 getnstimeofday(&ts);
779 h1->ts_last_pkt.ts_sec = ts.tv_sec;
780 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
781 }
782
783 smp_wmb();
784
785 /* Flush the block */
786 prb_flush_block(pkc1, pbd1, status);
787
788 sk->sk_data_ready(sk);
789
790 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
791 }
792
793 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
794 {
795 pkc->reset_pending_on_curr_blk = 0;
796 }
797
798 /*
799 * Side effect of opening a block:
800 *
801 * 1) prb_queue is thawed.
802 * 2) retire_blk_timer is refreshed.
803 *
804 */
805 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
806 struct tpacket_block_desc *pbd1)
807 {
808 struct timespec ts;
809 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
810
811 smp_rmb();
812
813 /* We could have just memset this but we will lose the
814 * flexibility of making the priv area sticky
815 */
816
817 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
818 BLOCK_NUM_PKTS(pbd1) = 0;
819 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
820
821 getnstimeofday(&ts);
822
823 h1->ts_first_pkt.ts_sec = ts.tv_sec;
824 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
825
826 pkc1->pkblk_start = (char *)pbd1;
827 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
828
829 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
831
832 pbd1->version = pkc1->version;
833 pkc1->prev = pkc1->nxt_offset;
834 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
835
836 prb_thaw_queue(pkc1);
837 _prb_refresh_rx_retire_blk_timer(pkc1);
838
839 smp_wmb();
840 }
841
842 /*
843 * Queue freeze logic:
844 * 1) Assume tp_block_nr = 8 blocks.
845 * 2) At time 't0', user opens Rx ring.
846 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
847 * 4) user-space is either sleeping or processing block '0'.
848 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
849 * it will close block-7,loop around and try to fill block '0'.
850 * call-flow:
851 * __packet_lookup_frame_in_block
852 * prb_retire_current_block()
853 * prb_dispatch_next_block()
854 * |->(BLOCK_STATUS == USER) evaluates to true
855 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
856 * 6) Now there are two cases:
857 * 6.1) Link goes idle right after the queue is frozen.
858 * But remember, the last open_block() refreshed the timer.
859 * When this timer expires,it will refresh itself so that we can
860 * re-open block-0 in near future.
861 * 6.2) Link is busy and keeps on receiving packets. This is a simple
862 * case and __packet_lookup_frame_in_block will check if block-0
863 * is free and can now be re-used.
864 */
865 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
866 struct packet_sock *po)
867 {
868 pkc->reset_pending_on_curr_blk = 1;
869 po->stats.stats3.tp_freeze_q_cnt++;
870 }
871
872 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
873
874 /*
875 * If the next block is free then we will dispatch it
876 * and return a good offset.
877 * Else, we will freeze the queue.
878 * So, caller must check the return value.
879 */
880 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
881 struct packet_sock *po)
882 {
883 struct tpacket_block_desc *pbd;
884
885 smp_rmb();
886
887 /* 1. Get current block num */
888 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
889
890 /* 2. If this block is currently in_use then freeze the queue */
891 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
892 prb_freeze_queue(pkc, po);
893 return NULL;
894 }
895
896 /*
897 * 3.
898 * open this block and return the offset where the first packet
899 * needs to get stored.
900 */
901 prb_open_block(pkc, pbd);
902 return (void *)pkc->nxt_offset;
903 }
904
905 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
906 struct packet_sock *po, unsigned int status)
907 {
908 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
909
910 /* retire/close the current block */
911 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
912 /*
913 * Plug the case where copy_bits() is in progress on
914 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
915 * have space to copy the pkt in the current block and
916 * called prb_retire_current_block()
917 *
918 * We don't need to worry about the TMO case because
919 * the timer-handler already handled this case.
920 */
921 if (!(status & TP_STATUS_BLK_TMO)) {
922 while (atomic_read(&pkc->blk_fill_in_prog)) {
923 /* Waiting for skb_copy_bits to finish... */
924 cpu_relax();
925 }
926 }
927 prb_close_block(pkc, pbd, po, status);
928 return;
929 }
930 }
931
932 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
933 {
934 return TP_STATUS_USER & BLOCK_STATUS(pbd);
935 }
936
937 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
938 {
939 return pkc->reset_pending_on_curr_blk;
940 }
941
942 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
943 {
944 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
945 atomic_dec(&pkc->blk_fill_in_prog);
946 }
947
948 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
949 struct tpacket3_hdr *ppd)
950 {
951 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
952 }
953
954 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
955 struct tpacket3_hdr *ppd)
956 {
957 ppd->hv1.tp_rxhash = 0;
958 }
959
960 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
961 struct tpacket3_hdr *ppd)
962 {
963 if (skb_vlan_tag_present(pkc->skb)) {
964 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
965 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
966 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
967 } else {
968 ppd->hv1.tp_vlan_tci = 0;
969 ppd->hv1.tp_vlan_tpid = 0;
970 ppd->tp_status = TP_STATUS_AVAILABLE;
971 }
972 }
973
974 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
975 struct tpacket3_hdr *ppd)
976 {
977 ppd->hv1.tp_padding = 0;
978 prb_fill_vlan_info(pkc, ppd);
979
980 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
981 prb_fill_rxhash(pkc, ppd);
982 else
983 prb_clear_rxhash(pkc, ppd);
984 }
985
986 static void prb_fill_curr_block(char *curr,
987 struct tpacket_kbdq_core *pkc,
988 struct tpacket_block_desc *pbd,
989 unsigned int len)
990 {
991 struct tpacket3_hdr *ppd;
992
993 ppd = (struct tpacket3_hdr *)curr;
994 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
995 pkc->prev = curr;
996 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
997 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
998 BLOCK_NUM_PKTS(pbd) += 1;
999 atomic_inc(&pkc->blk_fill_in_prog);
1000 prb_run_all_ft_ops(pkc, ppd);
1001 }
1002
1003 /* Assumes caller has the sk->rx_queue.lock */
1004 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1005 struct sk_buff *skb,
1006 int status,
1007 unsigned int len
1008 )
1009 {
1010 struct tpacket_kbdq_core *pkc;
1011 struct tpacket_block_desc *pbd;
1012 char *curr, *end;
1013
1014 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1015 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1016
1017 /* Queue is frozen when user space is lagging behind */
1018 if (prb_queue_frozen(pkc)) {
1019 /*
1020 * Check if that last block which caused the queue to freeze,
1021 * is still in_use by user-space.
1022 */
1023 if (prb_curr_blk_in_use(pbd)) {
1024 /* Can't record this packet */
1025 return NULL;
1026 } else {
1027 /*
1028 * Ok, the block was released by user-space.
1029 * Now let's open that block.
1030 * opening a block also thaws the queue.
1031 * Thawing is a side effect.
1032 */
1033 prb_open_block(pkc, pbd);
1034 }
1035 }
1036
1037 smp_mb();
1038 curr = pkc->nxt_offset;
1039 pkc->skb = skb;
1040 end = (char *)pbd + pkc->kblk_size;
1041
1042 /* first try the current block */
1043 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1044 prb_fill_curr_block(curr, pkc, pbd, len);
1045 return (void *)curr;
1046 }
1047
1048 /* Ok, close the current block */
1049 prb_retire_current_block(pkc, po, 0);
1050
1051 /* Now, try to dispatch the next block */
1052 curr = (char *)prb_dispatch_next_block(pkc, po);
1053 if (curr) {
1054 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1055 prb_fill_curr_block(curr, pkc, pbd, len);
1056 return (void *)curr;
1057 }
1058
1059 /*
1060 * No free blocks are available.user_space hasn't caught up yet.
1061 * Queue was just frozen and now this packet will get dropped.
1062 */
1063 return NULL;
1064 }
1065
1066 static void *packet_current_rx_frame(struct packet_sock *po,
1067 struct sk_buff *skb,
1068 int status, unsigned int len)
1069 {
1070 char *curr = NULL;
1071 switch (po->tp_version) {
1072 case TPACKET_V1:
1073 case TPACKET_V2:
1074 curr = packet_lookup_frame(po, &po->rx_ring,
1075 po->rx_ring.head, status);
1076 return curr;
1077 case TPACKET_V3:
1078 return __packet_lookup_frame_in_block(po, skb, status, len);
1079 default:
1080 WARN(1, "TPACKET version not supported\n");
1081 BUG();
1082 return NULL;
1083 }
1084 }
1085
1086 static void *prb_lookup_block(struct packet_sock *po,
1087 struct packet_ring_buffer *rb,
1088 unsigned int idx,
1089 int status)
1090 {
1091 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1092 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1093
1094 if (status != BLOCK_STATUS(pbd))
1095 return NULL;
1096 return pbd;
1097 }
1098
1099 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1100 {
1101 unsigned int prev;
1102 if (rb->prb_bdqc.kactive_blk_num)
1103 prev = rb->prb_bdqc.kactive_blk_num-1;
1104 else
1105 prev = rb->prb_bdqc.knum_blocks-1;
1106 return prev;
1107 }
1108
1109 /* Assumes caller has held the rx_queue.lock */
1110 static void *__prb_previous_block(struct packet_sock *po,
1111 struct packet_ring_buffer *rb,
1112 int status)
1113 {
1114 unsigned int previous = prb_previous_blk_num(rb);
1115 return prb_lookup_block(po, rb, previous, status);
1116 }
1117
1118 static void *packet_previous_rx_frame(struct packet_sock *po,
1119 struct packet_ring_buffer *rb,
1120 int status)
1121 {
1122 if (po->tp_version <= TPACKET_V2)
1123 return packet_previous_frame(po, rb, status);
1124
1125 return __prb_previous_block(po, rb, status);
1126 }
1127
1128 static void packet_increment_rx_head(struct packet_sock *po,
1129 struct packet_ring_buffer *rb)
1130 {
1131 switch (po->tp_version) {
1132 case TPACKET_V1:
1133 case TPACKET_V2:
1134 return packet_increment_head(rb);
1135 case TPACKET_V3:
1136 default:
1137 WARN(1, "TPACKET version not supported.\n");
1138 BUG();
1139 return;
1140 }
1141 }
1142
1143 static void *packet_previous_frame(struct packet_sock *po,
1144 struct packet_ring_buffer *rb,
1145 int status)
1146 {
1147 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1148 return packet_lookup_frame(po, rb, previous, status);
1149 }
1150
1151 static void packet_increment_head(struct packet_ring_buffer *buff)
1152 {
1153 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1154 }
1155
1156 static void packet_inc_pending(struct packet_ring_buffer *rb)
1157 {
1158 this_cpu_inc(*rb->pending_refcnt);
1159 }
1160
1161 static void packet_dec_pending(struct packet_ring_buffer *rb)
1162 {
1163 this_cpu_dec(*rb->pending_refcnt);
1164 }
1165
1166 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1167 {
1168 unsigned int refcnt = 0;
1169 int cpu;
1170
1171 /* We don't use pending refcount in rx_ring. */
1172 if (rb->pending_refcnt == NULL)
1173 return 0;
1174
1175 for_each_possible_cpu(cpu)
1176 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1177
1178 return refcnt;
1179 }
1180
1181 static int packet_alloc_pending(struct packet_sock *po)
1182 {
1183 po->rx_ring.pending_refcnt = NULL;
1184
1185 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1186 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1187 return -ENOBUFS;
1188
1189 return 0;
1190 }
1191
1192 static void packet_free_pending(struct packet_sock *po)
1193 {
1194 free_percpu(po->tx_ring.pending_refcnt);
1195 }
1196
1197 #define ROOM_POW_OFF 2
1198 #define ROOM_NONE 0x0
1199 #define ROOM_LOW 0x1
1200 #define ROOM_NORMAL 0x2
1201
1202 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1203 {
1204 int idx, len;
1205
1206 len = po->rx_ring.frame_max + 1;
1207 idx = po->rx_ring.head;
1208 if (pow_off)
1209 idx += len >> pow_off;
1210 if (idx >= len)
1211 idx -= len;
1212 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1213 }
1214
1215 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1216 {
1217 int idx, len;
1218
1219 len = po->rx_ring.prb_bdqc.knum_blocks;
1220 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1221 if (pow_off)
1222 idx += len >> pow_off;
1223 if (idx >= len)
1224 idx -= len;
1225 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1226 }
1227
1228 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1229 {
1230 struct sock *sk = &po->sk;
1231 int ret = ROOM_NONE;
1232
1233 if (po->prot_hook.func != tpacket_rcv) {
1234 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1235 - (skb ? skb->truesize : 0);
1236 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1237 return ROOM_NORMAL;
1238 else if (avail > 0)
1239 return ROOM_LOW;
1240 else
1241 return ROOM_NONE;
1242 }
1243
1244 if (po->tp_version == TPACKET_V3) {
1245 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1246 ret = ROOM_NORMAL;
1247 else if (__tpacket_v3_has_room(po, 0))
1248 ret = ROOM_LOW;
1249 } else {
1250 if (__tpacket_has_room(po, ROOM_POW_OFF))
1251 ret = ROOM_NORMAL;
1252 else if (__tpacket_has_room(po, 0))
1253 ret = ROOM_LOW;
1254 }
1255
1256 return ret;
1257 }
1258
1259 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1260 {
1261 int ret;
1262 bool has_room;
1263
1264 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1265 ret = __packet_rcv_has_room(po, skb);
1266 has_room = ret == ROOM_NORMAL;
1267 if (po->pressure == has_room)
1268 po->pressure = !has_room;
1269 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1270
1271 return ret;
1272 }
1273
1274 static void packet_sock_destruct(struct sock *sk)
1275 {
1276 skb_queue_purge(&sk->sk_error_queue);
1277
1278 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1279 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1280
1281 if (!sock_flag(sk, SOCK_DEAD)) {
1282 pr_err("Attempt to release alive packet socket: %p\n", sk);
1283 return;
1284 }
1285
1286 sk_refcnt_debug_dec(sk);
1287 }
1288
1289 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1290 {
1291 u32 rxhash;
1292 int i, count = 0;
1293
1294 rxhash = skb_get_hash(skb);
1295 for (i = 0; i < ROLLOVER_HLEN; i++)
1296 if (po->rollover->history[i] == rxhash)
1297 count++;
1298
1299 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1300 return count > (ROLLOVER_HLEN >> 1);
1301 }
1302
1303 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1304 struct sk_buff *skb,
1305 unsigned int num)
1306 {
1307 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1308 }
1309
1310 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1311 struct sk_buff *skb,
1312 unsigned int num)
1313 {
1314 unsigned int val = atomic_inc_return(&f->rr_cur);
1315
1316 return val % num;
1317 }
1318
1319 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1320 struct sk_buff *skb,
1321 unsigned int num)
1322 {
1323 return smp_processor_id() % num;
1324 }
1325
1326 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1327 struct sk_buff *skb,
1328 unsigned int num)
1329 {
1330 return prandom_u32_max(num);
1331 }
1332
1333 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1334 struct sk_buff *skb,
1335 unsigned int idx, bool try_self,
1336 unsigned int num)
1337 {
1338 struct packet_sock *po, *po_next, *po_skip = NULL;
1339 unsigned int i, j, room = ROOM_NONE;
1340
1341 po = pkt_sk(f->arr[idx]);
1342
1343 if (try_self) {
1344 room = packet_rcv_has_room(po, skb);
1345 if (room == ROOM_NORMAL ||
1346 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1347 return idx;
1348 po_skip = po;
1349 }
1350
1351 i = j = min_t(int, po->rollover->sock, num - 1);
1352 do {
1353 po_next = pkt_sk(f->arr[i]);
1354 if (po_next != po_skip && !po_next->pressure &&
1355 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1356 if (i != j)
1357 po->rollover->sock = i;
1358 atomic_long_inc(&po->rollover->num);
1359 if (room == ROOM_LOW)
1360 atomic_long_inc(&po->rollover->num_huge);
1361 return i;
1362 }
1363
1364 if (++i == num)
1365 i = 0;
1366 } while (i != j);
1367
1368 atomic_long_inc(&po->rollover->num_failed);
1369 return idx;
1370 }
1371
1372 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1373 struct sk_buff *skb,
1374 unsigned int num)
1375 {
1376 return skb_get_queue_mapping(skb) % num;
1377 }
1378
1379 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1380 struct sk_buff *skb,
1381 unsigned int num)
1382 {
1383 struct bpf_prog *prog;
1384 unsigned int ret = 0;
1385
1386 rcu_read_lock();
1387 prog = rcu_dereference(f->bpf_prog);
1388 if (prog)
1389 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1390 rcu_read_unlock();
1391
1392 return ret;
1393 }
1394
1395 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1396 {
1397 return f->flags & (flag >> 8);
1398 }
1399
1400 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1401 struct packet_type *pt, struct net_device *orig_dev)
1402 {
1403 struct packet_fanout *f = pt->af_packet_priv;
1404 unsigned int num = READ_ONCE(f->num_members);
1405 struct net *net = read_pnet(&f->net);
1406 struct packet_sock *po;
1407 unsigned int idx;
1408
1409 if (!net_eq(dev_net(dev), net) || !num) {
1410 kfree_skb(skb);
1411 return 0;
1412 }
1413
1414 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1415 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1416 if (!skb)
1417 return 0;
1418 }
1419 switch (f->type) {
1420 case PACKET_FANOUT_HASH:
1421 default:
1422 idx = fanout_demux_hash(f, skb, num);
1423 break;
1424 case PACKET_FANOUT_LB:
1425 idx = fanout_demux_lb(f, skb, num);
1426 break;
1427 case PACKET_FANOUT_CPU:
1428 idx = fanout_demux_cpu(f, skb, num);
1429 break;
1430 case PACKET_FANOUT_RND:
1431 idx = fanout_demux_rnd(f, skb, num);
1432 break;
1433 case PACKET_FANOUT_QM:
1434 idx = fanout_demux_qm(f, skb, num);
1435 break;
1436 case PACKET_FANOUT_ROLLOVER:
1437 idx = fanout_demux_rollover(f, skb, 0, false, num);
1438 break;
1439 case PACKET_FANOUT_CBPF:
1440 case PACKET_FANOUT_EBPF:
1441 idx = fanout_demux_bpf(f, skb, num);
1442 break;
1443 }
1444
1445 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1446 idx = fanout_demux_rollover(f, skb, idx, true, num);
1447
1448 po = pkt_sk(f->arr[idx]);
1449 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1450 }
1451
1452 DEFINE_MUTEX(fanout_mutex);
1453 EXPORT_SYMBOL_GPL(fanout_mutex);
1454 static LIST_HEAD(fanout_list);
1455 static u16 fanout_next_id;
1456
1457 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1458 {
1459 struct packet_fanout *f = po->fanout;
1460
1461 spin_lock(&f->lock);
1462 f->arr[f->num_members] = sk;
1463 smp_wmb();
1464 f->num_members++;
1465 if (f->num_members == 1)
1466 dev_add_pack(&f->prot_hook);
1467 spin_unlock(&f->lock);
1468 }
1469
1470 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1471 {
1472 struct packet_fanout *f = po->fanout;
1473 int i;
1474
1475 spin_lock(&f->lock);
1476 for (i = 0; i < f->num_members; i++) {
1477 if (f->arr[i] == sk)
1478 break;
1479 }
1480 BUG_ON(i >= f->num_members);
1481 f->arr[i] = f->arr[f->num_members - 1];
1482 f->num_members--;
1483 if (f->num_members == 0)
1484 __dev_remove_pack(&f->prot_hook);
1485 spin_unlock(&f->lock);
1486 }
1487
1488 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1489 {
1490 if (sk->sk_family != PF_PACKET)
1491 return false;
1492
1493 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1494 }
1495
1496 static void fanout_init_data(struct packet_fanout *f)
1497 {
1498 switch (f->type) {
1499 case PACKET_FANOUT_LB:
1500 atomic_set(&f->rr_cur, 0);
1501 break;
1502 case PACKET_FANOUT_CBPF:
1503 case PACKET_FANOUT_EBPF:
1504 RCU_INIT_POINTER(f->bpf_prog, NULL);
1505 break;
1506 }
1507 }
1508
1509 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1510 {
1511 struct bpf_prog *old;
1512
1513 spin_lock(&f->lock);
1514 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1515 rcu_assign_pointer(f->bpf_prog, new);
1516 spin_unlock(&f->lock);
1517
1518 if (old) {
1519 synchronize_net();
1520 bpf_prog_destroy(old);
1521 }
1522 }
1523
1524 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1525 unsigned int len)
1526 {
1527 struct bpf_prog *new;
1528 struct sock_fprog fprog;
1529 int ret;
1530
1531 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1532 return -EPERM;
1533 if (len != sizeof(fprog))
1534 return -EINVAL;
1535 if (copy_from_user(&fprog, data, len))
1536 return -EFAULT;
1537
1538 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1539 if (ret)
1540 return ret;
1541
1542 __fanout_set_data_bpf(po->fanout, new);
1543 return 0;
1544 }
1545
1546 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1547 unsigned int len)
1548 {
1549 struct bpf_prog *new;
1550 u32 fd;
1551
1552 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1553 return -EPERM;
1554 if (len != sizeof(fd))
1555 return -EINVAL;
1556 if (copy_from_user(&fd, data, len))
1557 return -EFAULT;
1558
1559 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1560 if (IS_ERR(new))
1561 return PTR_ERR(new);
1562
1563 __fanout_set_data_bpf(po->fanout, new);
1564 return 0;
1565 }
1566
1567 static int fanout_set_data(struct packet_sock *po, char __user *data,
1568 unsigned int len)
1569 {
1570 switch (po->fanout->type) {
1571 case PACKET_FANOUT_CBPF:
1572 return fanout_set_data_cbpf(po, data, len);
1573 case PACKET_FANOUT_EBPF:
1574 return fanout_set_data_ebpf(po, data, len);
1575 default:
1576 return -EINVAL;
1577 }
1578 }
1579
1580 static void fanout_release_data(struct packet_fanout *f)
1581 {
1582 switch (f->type) {
1583 case PACKET_FANOUT_CBPF:
1584 case PACKET_FANOUT_EBPF:
1585 __fanout_set_data_bpf(f, NULL);
1586 }
1587 }
1588
1589 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1590 {
1591 struct packet_fanout *f;
1592
1593 list_for_each_entry(f, &fanout_list, list) {
1594 if (f->id == candidate_id &&
1595 read_pnet(&f->net) == sock_net(sk)) {
1596 return false;
1597 }
1598 }
1599 return true;
1600 }
1601
1602 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1603 {
1604 u16 id = fanout_next_id;
1605
1606 do {
1607 if (__fanout_id_is_free(sk, id)) {
1608 *new_id = id;
1609 fanout_next_id = id + 1;
1610 return true;
1611 }
1612
1613 id++;
1614 } while (id != fanout_next_id);
1615
1616 return false;
1617 }
1618
1619 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1620 {
1621 struct packet_rollover *rollover = NULL;
1622 struct packet_sock *po = pkt_sk(sk);
1623 struct packet_fanout *f, *match;
1624 u8 type = type_flags & 0xff;
1625 u8 flags = type_flags >> 8;
1626 int err;
1627
1628 switch (type) {
1629 case PACKET_FANOUT_ROLLOVER:
1630 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1631 return -EINVAL;
1632 case PACKET_FANOUT_HASH:
1633 case PACKET_FANOUT_LB:
1634 case PACKET_FANOUT_CPU:
1635 case PACKET_FANOUT_RND:
1636 case PACKET_FANOUT_QM:
1637 case PACKET_FANOUT_CBPF:
1638 case PACKET_FANOUT_EBPF:
1639 break;
1640 default:
1641 return -EINVAL;
1642 }
1643
1644 mutex_lock(&fanout_mutex);
1645
1646 err = -EALREADY;
1647 if (po->fanout)
1648 goto out;
1649
1650 if (type == PACKET_FANOUT_ROLLOVER ||
1651 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1652 err = -ENOMEM;
1653 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1654 if (!rollover)
1655 goto out;
1656 atomic_long_set(&rollover->num, 0);
1657 atomic_long_set(&rollover->num_huge, 0);
1658 atomic_long_set(&rollover->num_failed, 0);
1659 }
1660
1661 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1662 if (id != 0) {
1663 err = -EINVAL;
1664 goto out;
1665 }
1666 if (!fanout_find_new_id(sk, &id)) {
1667 err = -ENOMEM;
1668 goto out;
1669 }
1670 /* ephemeral flag for the first socket in the group: drop it */
1671 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1672 }
1673
1674 match = NULL;
1675 list_for_each_entry(f, &fanout_list, list) {
1676 if (f->id == id &&
1677 read_pnet(&f->net) == sock_net(sk)) {
1678 match = f;
1679 break;
1680 }
1681 }
1682 err = -EINVAL;
1683 if (match && match->flags != flags)
1684 goto out;
1685 if (!match) {
1686 err = -ENOMEM;
1687 match = kzalloc(sizeof(*match), GFP_KERNEL);
1688 if (!match)
1689 goto out;
1690 write_pnet(&match->net, sock_net(sk));
1691 match->id = id;
1692 match->type = type;
1693 match->flags = flags;
1694 INIT_LIST_HEAD(&match->list);
1695 spin_lock_init(&match->lock);
1696 refcount_set(&match->sk_ref, 0);
1697 fanout_init_data(match);
1698 match->prot_hook.type = po->prot_hook.type;
1699 match->prot_hook.dev = po->prot_hook.dev;
1700 match->prot_hook.func = packet_rcv_fanout;
1701 match->prot_hook.af_packet_priv = match;
1702 match->prot_hook.id_match = match_fanout_group;
1703 list_add(&match->list, &fanout_list);
1704 }
1705 err = -EINVAL;
1706
1707 spin_lock(&po->bind_lock);
1708 if (po->running &&
1709 match->type == type &&
1710 match->prot_hook.type == po->prot_hook.type &&
1711 match->prot_hook.dev == po->prot_hook.dev) {
1712 err = -ENOSPC;
1713 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1714 __dev_remove_pack(&po->prot_hook);
1715 po->fanout = match;
1716 po->rollover = rollover;
1717 rollover = NULL;
1718 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1719 __fanout_link(sk, po);
1720 err = 0;
1721 }
1722 }
1723 spin_unlock(&po->bind_lock);
1724
1725 if (err && !refcount_read(&match->sk_ref)) {
1726 list_del(&match->list);
1727 kfree(match);
1728 }
1729
1730 out:
1731 kfree(rollover);
1732 mutex_unlock(&fanout_mutex);
1733 return err;
1734 }
1735
1736 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1737 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1738 * It is the responsibility of the caller to call fanout_release_data() and
1739 * free the returned packet_fanout (after synchronize_net())
1740 */
1741 static struct packet_fanout *fanout_release(struct sock *sk)
1742 {
1743 struct packet_sock *po = pkt_sk(sk);
1744 struct packet_fanout *f;
1745
1746 mutex_lock(&fanout_mutex);
1747 f = po->fanout;
1748 if (f) {
1749 po->fanout = NULL;
1750
1751 if (refcount_dec_and_test(&f->sk_ref))
1752 list_del(&f->list);
1753 else
1754 f = NULL;
1755 }
1756 mutex_unlock(&fanout_mutex);
1757
1758 return f;
1759 }
1760
1761 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1762 struct sk_buff *skb)
1763 {
1764 /* Earlier code assumed this would be a VLAN pkt, double-check
1765 * this now that we have the actual packet in hand. We can only
1766 * do this check on Ethernet devices.
1767 */
1768 if (unlikely(dev->type != ARPHRD_ETHER))
1769 return false;
1770
1771 skb_reset_mac_header(skb);
1772 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1773 }
1774
1775 static const struct proto_ops packet_ops;
1776
1777 static const struct proto_ops packet_ops_spkt;
1778
1779 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1780 struct packet_type *pt, struct net_device *orig_dev)
1781 {
1782 struct sock *sk;
1783 struct sockaddr_pkt *spkt;
1784
1785 /*
1786 * When we registered the protocol we saved the socket in the data
1787 * field for just this event.
1788 */
1789
1790 sk = pt->af_packet_priv;
1791
1792 /*
1793 * Yank back the headers [hope the device set this
1794 * right or kerboom...]
1795 *
1796 * Incoming packets have ll header pulled,
1797 * push it back.
1798 *
1799 * For outgoing ones skb->data == skb_mac_header(skb)
1800 * so that this procedure is noop.
1801 */
1802
1803 if (skb->pkt_type == PACKET_LOOPBACK)
1804 goto out;
1805
1806 if (!net_eq(dev_net(dev), sock_net(sk)))
1807 goto out;
1808
1809 skb = skb_share_check(skb, GFP_ATOMIC);
1810 if (skb == NULL)
1811 goto oom;
1812
1813 /* drop any routing info */
1814 skb_dst_drop(skb);
1815
1816 /* drop conntrack reference */
1817 nf_reset(skb);
1818
1819 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1820
1821 skb_push(skb, skb->data - skb_mac_header(skb));
1822
1823 /*
1824 * The SOCK_PACKET socket receives _all_ frames.
1825 */
1826
1827 spkt->spkt_family = dev->type;
1828 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1829 spkt->spkt_protocol = skb->protocol;
1830
1831 /*
1832 * Charge the memory to the socket. This is done specifically
1833 * to prevent sockets using all the memory up.
1834 */
1835
1836 if (sock_queue_rcv_skb(sk, skb) == 0)
1837 return 0;
1838
1839 out:
1840 kfree_skb(skb);
1841 oom:
1842 return 0;
1843 }
1844
1845 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1846 {
1847 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1848 sock->type == SOCK_RAW) {
1849 skb_reset_mac_header(skb);
1850 skb->protocol = dev_parse_header_protocol(skb);
1851 }
1852
1853 skb_probe_transport_header(skb);
1854 }
1855
1856 /*
1857 * Output a raw packet to a device layer. This bypasses all the other
1858 * protocol layers and you must therefore supply it with a complete frame
1859 */
1860
1861 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1862 size_t len)
1863 {
1864 struct sock *sk = sock->sk;
1865 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1866 struct sk_buff *skb = NULL;
1867 struct net_device *dev;
1868 struct sockcm_cookie sockc;
1869 __be16 proto = 0;
1870 int err;
1871 int extra_len = 0;
1872
1873 /*
1874 * Get and verify the address.
1875 */
1876
1877 if (saddr) {
1878 if (msg->msg_namelen < sizeof(struct sockaddr))
1879 return -EINVAL;
1880 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1881 proto = saddr->spkt_protocol;
1882 } else
1883 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1884
1885 /*
1886 * Find the device first to size check it
1887 */
1888
1889 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1890 retry:
1891 rcu_read_lock();
1892 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1893 err = -ENODEV;
1894 if (dev == NULL)
1895 goto out_unlock;
1896
1897 err = -ENETDOWN;
1898 if (!(dev->flags & IFF_UP))
1899 goto out_unlock;
1900
1901 /*
1902 * You may not queue a frame bigger than the mtu. This is the lowest level
1903 * raw protocol and you must do your own fragmentation at this level.
1904 */
1905
1906 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1907 if (!netif_supports_nofcs(dev)) {
1908 err = -EPROTONOSUPPORT;
1909 goto out_unlock;
1910 }
1911 extra_len = 4; /* We're doing our own CRC */
1912 }
1913
1914 err = -EMSGSIZE;
1915 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1916 goto out_unlock;
1917
1918 if (!skb) {
1919 size_t reserved = LL_RESERVED_SPACE(dev);
1920 int tlen = dev->needed_tailroom;
1921 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1922
1923 rcu_read_unlock();
1924 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1925 if (skb == NULL)
1926 return -ENOBUFS;
1927 /* FIXME: Save some space for broken drivers that write a hard
1928 * header at transmission time by themselves. PPP is the notable
1929 * one here. This should really be fixed at the driver level.
1930 */
1931 skb_reserve(skb, reserved);
1932 skb_reset_network_header(skb);
1933
1934 /* Try to align data part correctly */
1935 if (hhlen) {
1936 skb->data -= hhlen;
1937 skb->tail -= hhlen;
1938 if (len < hhlen)
1939 skb_reset_network_header(skb);
1940 }
1941 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1942 if (err)
1943 goto out_free;
1944 goto retry;
1945 }
1946
1947 if (!dev_validate_header(dev, skb->data, len)) {
1948 err = -EINVAL;
1949 goto out_unlock;
1950 }
1951 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1952 !packet_extra_vlan_len_allowed(dev, skb)) {
1953 err = -EMSGSIZE;
1954 goto out_unlock;
1955 }
1956
1957 sockcm_init(&sockc, sk);
1958 if (msg->msg_controllen) {
1959 err = sock_cmsg_send(sk, msg, &sockc);
1960 if (unlikely(err))
1961 goto out_unlock;
1962 }
1963
1964 skb->protocol = proto;
1965 skb->dev = dev;
1966 skb->priority = sk->sk_priority;
1967 skb->mark = sk->sk_mark;
1968 skb->tstamp = sockc.transmit_time;
1969
1970 skb_setup_tx_timestamp(skb, sockc.tsflags);
1971
1972 if (unlikely(extra_len == 4))
1973 skb->no_fcs = 1;
1974
1975 packet_parse_headers(skb, sock);
1976
1977 dev_queue_xmit(skb);
1978 rcu_read_unlock();
1979 return len;
1980
1981 out_unlock:
1982 rcu_read_unlock();
1983 out_free:
1984 kfree_skb(skb);
1985 return err;
1986 }
1987
1988 static unsigned int run_filter(struct sk_buff *skb,
1989 const struct sock *sk,
1990 unsigned int res)
1991 {
1992 struct sk_filter *filter;
1993
1994 rcu_read_lock();
1995 filter = rcu_dereference(sk->sk_filter);
1996 if (filter != NULL)
1997 res = bpf_prog_run_clear_cb(filter->prog, skb);
1998 rcu_read_unlock();
1999
2000 return res;
2001 }
2002
2003 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2004 size_t *len)
2005 {
2006 struct virtio_net_hdr vnet_hdr;
2007
2008 if (*len < sizeof(vnet_hdr))
2009 return -EINVAL;
2010 *len -= sizeof(vnet_hdr);
2011
2012 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2013 return -EINVAL;
2014
2015 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2016 }
2017
2018 /*
2019 * This function makes lazy skb cloning in hope that most of packets
2020 * are discarded by BPF.
2021 *
2022 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2023 * and skb->cb are mangled. It works because (and until) packets
2024 * falling here are owned by current CPU. Output packets are cloned
2025 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2026 * sequencially, so that if we return skb to original state on exit,
2027 * we will not harm anyone.
2028 */
2029
2030 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2031 struct packet_type *pt, struct net_device *orig_dev)
2032 {
2033 struct sock *sk;
2034 struct sockaddr_ll *sll;
2035 struct packet_sock *po;
2036 u8 *skb_head = skb->data;
2037 int skb_len = skb->len;
2038 unsigned int snaplen, res;
2039 bool is_drop_n_account = false;
2040
2041 if (skb->pkt_type == PACKET_LOOPBACK)
2042 goto drop;
2043
2044 sk = pt->af_packet_priv;
2045 po = pkt_sk(sk);
2046
2047 if (!net_eq(dev_net(dev), sock_net(sk)))
2048 goto drop;
2049
2050 skb->dev = dev;
2051
2052 if (dev->header_ops) {
2053 /* The device has an explicit notion of ll header,
2054 * exported to higher levels.
2055 *
2056 * Otherwise, the device hides details of its frame
2057 * structure, so that corresponding packet head is
2058 * never delivered to user.
2059 */
2060 if (sk->sk_type != SOCK_DGRAM)
2061 skb_push(skb, skb->data - skb_mac_header(skb));
2062 else if (skb->pkt_type == PACKET_OUTGOING) {
2063 /* Special case: outgoing packets have ll header at head */
2064 skb_pull(skb, skb_network_offset(skb));
2065 }
2066 }
2067
2068 snaplen = skb->len;
2069
2070 res = run_filter(skb, sk, snaplen);
2071 if (!res)
2072 goto drop_n_restore;
2073 if (snaplen > res)
2074 snaplen = res;
2075
2076 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2077 goto drop_n_acct;
2078
2079 if (skb_shared(skb)) {
2080 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2081 if (nskb == NULL)
2082 goto drop_n_acct;
2083
2084 if (skb_head != skb->data) {
2085 skb->data = skb_head;
2086 skb->len = skb_len;
2087 }
2088 consume_skb(skb);
2089 skb = nskb;
2090 }
2091
2092 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2093
2094 sll = &PACKET_SKB_CB(skb)->sa.ll;
2095 sll->sll_hatype = dev->type;
2096 sll->sll_pkttype = skb->pkt_type;
2097 if (unlikely(po->origdev))
2098 sll->sll_ifindex = orig_dev->ifindex;
2099 else
2100 sll->sll_ifindex = dev->ifindex;
2101
2102 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2103
2104 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2105 * Use their space for storing the original skb length.
2106 */
2107 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2108
2109 if (pskb_trim(skb, snaplen))
2110 goto drop_n_acct;
2111
2112 skb_set_owner_r(skb, sk);
2113 skb->dev = NULL;
2114 skb_dst_drop(skb);
2115
2116 /* drop conntrack reference */
2117 nf_reset(skb);
2118
2119 spin_lock(&sk->sk_receive_queue.lock);
2120 po->stats.stats1.tp_packets++;
2121 sock_skb_set_dropcount(sk, skb);
2122 __skb_queue_tail(&sk->sk_receive_queue, skb);
2123 spin_unlock(&sk->sk_receive_queue.lock);
2124 sk->sk_data_ready(sk);
2125 return 0;
2126
2127 drop_n_acct:
2128 is_drop_n_account = true;
2129 spin_lock(&sk->sk_receive_queue.lock);
2130 po->stats.stats1.tp_drops++;
2131 atomic_inc(&sk->sk_drops);
2132 spin_unlock(&sk->sk_receive_queue.lock);
2133
2134 drop_n_restore:
2135 if (skb_head != skb->data && skb_shared(skb)) {
2136 skb->data = skb_head;
2137 skb->len = skb_len;
2138 }
2139 drop:
2140 if (!is_drop_n_account)
2141 consume_skb(skb);
2142 else
2143 kfree_skb(skb);
2144 return 0;
2145 }
2146
2147 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2148 struct packet_type *pt, struct net_device *orig_dev)
2149 {
2150 struct sock *sk;
2151 struct packet_sock *po;
2152 struct sockaddr_ll *sll;
2153 union tpacket_uhdr h;
2154 u8 *skb_head = skb->data;
2155 int skb_len = skb->len;
2156 unsigned int snaplen, res;
2157 unsigned long status = TP_STATUS_USER;
2158 unsigned short macoff, netoff, hdrlen;
2159 struct sk_buff *copy_skb = NULL;
2160 struct timespec ts;
2161 __u32 ts_status;
2162 bool is_drop_n_account = false;
2163 bool do_vnet = false;
2164
2165 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2166 * We may add members to them until current aligned size without forcing
2167 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2168 */
2169 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2170 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2171
2172 if (skb->pkt_type == PACKET_LOOPBACK)
2173 goto drop;
2174
2175 sk = pt->af_packet_priv;
2176 po = pkt_sk(sk);
2177
2178 if (!net_eq(dev_net(dev), sock_net(sk)))
2179 goto drop;
2180
2181 if (dev->header_ops) {
2182 if (sk->sk_type != SOCK_DGRAM)
2183 skb_push(skb, skb->data - skb_mac_header(skb));
2184 else if (skb->pkt_type == PACKET_OUTGOING) {
2185 /* Special case: outgoing packets have ll header at head */
2186 skb_pull(skb, skb_network_offset(skb));
2187 }
2188 }
2189
2190 snaplen = skb->len;
2191
2192 res = run_filter(skb, sk, snaplen);
2193 if (!res)
2194 goto drop_n_restore;
2195
2196 if (skb->ip_summed == CHECKSUM_PARTIAL)
2197 status |= TP_STATUS_CSUMNOTREADY;
2198 else if (skb->pkt_type != PACKET_OUTGOING &&
2199 (skb->ip_summed == CHECKSUM_COMPLETE ||
2200 skb_csum_unnecessary(skb)))
2201 status |= TP_STATUS_CSUM_VALID;
2202
2203 if (snaplen > res)
2204 snaplen = res;
2205
2206 if (sk->sk_type == SOCK_DGRAM) {
2207 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2208 po->tp_reserve;
2209 } else {
2210 unsigned int maclen = skb_network_offset(skb);
2211 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2212 (maclen < 16 ? 16 : maclen)) +
2213 po->tp_reserve;
2214 if (po->has_vnet_hdr) {
2215 netoff += sizeof(struct virtio_net_hdr);
2216 do_vnet = true;
2217 }
2218 macoff = netoff - maclen;
2219 }
2220 if (po->tp_version <= TPACKET_V2) {
2221 if (macoff + snaplen > po->rx_ring.frame_size) {
2222 if (po->copy_thresh &&
2223 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2224 if (skb_shared(skb)) {
2225 copy_skb = skb_clone(skb, GFP_ATOMIC);
2226 } else {
2227 copy_skb = skb_get(skb);
2228 skb_head = skb->data;
2229 }
2230 if (copy_skb)
2231 skb_set_owner_r(copy_skb, sk);
2232 }
2233 snaplen = po->rx_ring.frame_size - macoff;
2234 if ((int)snaplen < 0) {
2235 snaplen = 0;
2236 do_vnet = false;
2237 }
2238 }
2239 } else if (unlikely(macoff + snaplen >
2240 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2241 u32 nval;
2242
2243 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2244 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2245 snaplen, nval, macoff);
2246 snaplen = nval;
2247 if (unlikely((int)snaplen < 0)) {
2248 snaplen = 0;
2249 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2250 do_vnet = false;
2251 }
2252 }
2253 spin_lock(&sk->sk_receive_queue.lock);
2254 h.raw = packet_current_rx_frame(po, skb,
2255 TP_STATUS_KERNEL, (macoff+snaplen));
2256 if (!h.raw)
2257 goto drop_n_account;
2258 if (po->tp_version <= TPACKET_V2) {
2259 packet_increment_rx_head(po, &po->rx_ring);
2260 /*
2261 * LOSING will be reported till you read the stats,
2262 * because it's COR - Clear On Read.
2263 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2264 * at packet level.
2265 */
2266 if (po->stats.stats1.tp_drops)
2267 status |= TP_STATUS_LOSING;
2268 }
2269
2270 if (do_vnet &&
2271 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2272 sizeof(struct virtio_net_hdr),
2273 vio_le(), true, 0))
2274 goto drop_n_account;
2275
2276 po->stats.stats1.tp_packets++;
2277 if (copy_skb) {
2278 status |= TP_STATUS_COPY;
2279 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2280 }
2281 spin_unlock(&sk->sk_receive_queue.lock);
2282
2283 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2284
2285 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2286 getnstimeofday(&ts);
2287
2288 status |= ts_status;
2289
2290 switch (po->tp_version) {
2291 case TPACKET_V1:
2292 h.h1->tp_len = skb->len;
2293 h.h1->tp_snaplen = snaplen;
2294 h.h1->tp_mac = macoff;
2295 h.h1->tp_net = netoff;
2296 h.h1->tp_sec = ts.tv_sec;
2297 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2298 hdrlen = sizeof(*h.h1);
2299 break;
2300 case TPACKET_V2:
2301 h.h2->tp_len = skb->len;
2302 h.h2->tp_snaplen = snaplen;
2303 h.h2->tp_mac = macoff;
2304 h.h2->tp_net = netoff;
2305 h.h2->tp_sec = ts.tv_sec;
2306 h.h2->tp_nsec = ts.tv_nsec;
2307 if (skb_vlan_tag_present(skb)) {
2308 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2309 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2310 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2311 } else {
2312 h.h2->tp_vlan_tci = 0;
2313 h.h2->tp_vlan_tpid = 0;
2314 }
2315 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2316 hdrlen = sizeof(*h.h2);
2317 break;
2318 case TPACKET_V3:
2319 /* tp_nxt_offset,vlan are already populated above.
2320 * So DONT clear those fields here
2321 */
2322 h.h3->tp_status |= status;
2323 h.h3->tp_len = skb->len;
2324 h.h3->tp_snaplen = snaplen;
2325 h.h3->tp_mac = macoff;
2326 h.h3->tp_net = netoff;
2327 h.h3->tp_sec = ts.tv_sec;
2328 h.h3->tp_nsec = ts.tv_nsec;
2329 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2330 hdrlen = sizeof(*h.h3);
2331 break;
2332 default:
2333 BUG();
2334 }
2335
2336 sll = h.raw + TPACKET_ALIGN(hdrlen);
2337 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2338 sll->sll_family = AF_PACKET;
2339 sll->sll_hatype = dev->type;
2340 sll->sll_protocol = skb->protocol;
2341 sll->sll_pkttype = skb->pkt_type;
2342 if (unlikely(po->origdev))
2343 sll->sll_ifindex = orig_dev->ifindex;
2344 else
2345 sll->sll_ifindex = dev->ifindex;
2346
2347 smp_mb();
2348
2349 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2350 if (po->tp_version <= TPACKET_V2) {
2351 u8 *start, *end;
2352
2353 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2354 macoff + snaplen);
2355
2356 for (start = h.raw; start < end; start += PAGE_SIZE)
2357 flush_dcache_page(pgv_to_page(start));
2358 }
2359 smp_wmb();
2360 #endif
2361
2362 if (po->tp_version <= TPACKET_V2) {
2363 __packet_set_status(po, h.raw, status);
2364 sk->sk_data_ready(sk);
2365 } else {
2366 prb_clear_blk_fill_status(&po->rx_ring);
2367 }
2368
2369 drop_n_restore:
2370 if (skb_head != skb->data && skb_shared(skb)) {
2371 skb->data = skb_head;
2372 skb->len = skb_len;
2373 }
2374 drop:
2375 if (!is_drop_n_account)
2376 consume_skb(skb);
2377 else
2378 kfree_skb(skb);
2379 return 0;
2380
2381 drop_n_account:
2382 is_drop_n_account = true;
2383 po->stats.stats1.tp_drops++;
2384 spin_unlock(&sk->sk_receive_queue.lock);
2385
2386 sk->sk_data_ready(sk);
2387 kfree_skb(copy_skb);
2388 goto drop_n_restore;
2389 }
2390
2391 static void tpacket_destruct_skb(struct sk_buff *skb)
2392 {
2393 struct packet_sock *po = pkt_sk(skb->sk);
2394
2395 if (likely(po->tx_ring.pg_vec)) {
2396 void *ph;
2397 __u32 ts;
2398
2399 ph = skb_zcopy_get_nouarg(skb);
2400 packet_dec_pending(&po->tx_ring);
2401
2402 ts = __packet_set_timestamp(po, ph, skb);
2403 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2404 }
2405
2406 sock_wfree(skb);
2407 }
2408
2409 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2410 {
2411 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2412 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2413 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2414 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2415 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2416 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2417 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2418
2419 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2420 return -EINVAL;
2421
2422 return 0;
2423 }
2424
2425 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2426 struct virtio_net_hdr *vnet_hdr)
2427 {
2428 if (*len < sizeof(*vnet_hdr))
2429 return -EINVAL;
2430 *len -= sizeof(*vnet_hdr);
2431
2432 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2433 return -EFAULT;
2434
2435 return __packet_snd_vnet_parse(vnet_hdr, *len);
2436 }
2437
2438 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2439 void *frame, struct net_device *dev, void *data, int tp_len,
2440 __be16 proto, unsigned char *addr, int hlen, int copylen,
2441 const struct sockcm_cookie *sockc)
2442 {
2443 union tpacket_uhdr ph;
2444 int to_write, offset, len, nr_frags, len_max;
2445 struct socket *sock = po->sk.sk_socket;
2446 struct page *page;
2447 int err;
2448
2449 ph.raw = frame;
2450
2451 skb->protocol = proto;
2452 skb->dev = dev;
2453 skb->priority = po->sk.sk_priority;
2454 skb->mark = po->sk.sk_mark;
2455 skb->tstamp = sockc->transmit_time;
2456 skb_setup_tx_timestamp(skb, sockc->tsflags);
2457 skb_zcopy_set_nouarg(skb, ph.raw);
2458
2459 skb_reserve(skb, hlen);
2460 skb_reset_network_header(skb);
2461
2462 to_write = tp_len;
2463
2464 if (sock->type == SOCK_DGRAM) {
2465 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2466 NULL, tp_len);
2467 if (unlikely(err < 0))
2468 return -EINVAL;
2469 } else if (copylen) {
2470 int hdrlen = min_t(int, copylen, tp_len);
2471
2472 skb_push(skb, dev->hard_header_len);
2473 skb_put(skb, copylen - dev->hard_header_len);
2474 err = skb_store_bits(skb, 0, data, hdrlen);
2475 if (unlikely(err))
2476 return err;
2477 if (!dev_validate_header(dev, skb->data, hdrlen))
2478 return -EINVAL;
2479
2480 data += hdrlen;
2481 to_write -= hdrlen;
2482 }
2483
2484 offset = offset_in_page(data);
2485 len_max = PAGE_SIZE - offset;
2486 len = ((to_write > len_max) ? len_max : to_write);
2487
2488 skb->data_len = to_write;
2489 skb->len += to_write;
2490 skb->truesize += to_write;
2491 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2492
2493 while (likely(to_write)) {
2494 nr_frags = skb_shinfo(skb)->nr_frags;
2495
2496 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2497 pr_err("Packet exceed the number of skb frags(%lu)\n",
2498 MAX_SKB_FRAGS);
2499 return -EFAULT;
2500 }
2501
2502 page = pgv_to_page(data);
2503 data += len;
2504 flush_dcache_page(page);
2505 get_page(page);
2506 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2507 to_write -= len;
2508 offset = 0;
2509 len_max = PAGE_SIZE;
2510 len = ((to_write > len_max) ? len_max : to_write);
2511 }
2512
2513 packet_parse_headers(skb, sock);
2514
2515 return tp_len;
2516 }
2517
2518 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2519 int size_max, void **data)
2520 {
2521 union tpacket_uhdr ph;
2522 int tp_len, off;
2523
2524 ph.raw = frame;
2525
2526 switch (po->tp_version) {
2527 case TPACKET_V3:
2528 if (ph.h3->tp_next_offset != 0) {
2529 pr_warn_once("variable sized slot not supported");
2530 return -EINVAL;
2531 }
2532 tp_len = ph.h3->tp_len;
2533 break;
2534 case TPACKET_V2:
2535 tp_len = ph.h2->tp_len;
2536 break;
2537 default:
2538 tp_len = ph.h1->tp_len;
2539 break;
2540 }
2541 if (unlikely(tp_len > size_max)) {
2542 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2543 return -EMSGSIZE;
2544 }
2545
2546 if (unlikely(po->tp_tx_has_off)) {
2547 int off_min, off_max;
2548
2549 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2550 off_max = po->tx_ring.frame_size - tp_len;
2551 if (po->sk.sk_type == SOCK_DGRAM) {
2552 switch (po->tp_version) {
2553 case TPACKET_V3:
2554 off = ph.h3->tp_net;
2555 break;
2556 case TPACKET_V2:
2557 off = ph.h2->tp_net;
2558 break;
2559 default:
2560 off = ph.h1->tp_net;
2561 break;
2562 }
2563 } else {
2564 switch (po->tp_version) {
2565 case TPACKET_V3:
2566 off = ph.h3->tp_mac;
2567 break;
2568 case TPACKET_V2:
2569 off = ph.h2->tp_mac;
2570 break;
2571 default:
2572 off = ph.h1->tp_mac;
2573 break;
2574 }
2575 }
2576 if (unlikely((off < off_min) || (off_max < off)))
2577 return -EINVAL;
2578 } else {
2579 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2580 }
2581
2582 *data = frame + off;
2583 return tp_len;
2584 }
2585
2586 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2587 {
2588 struct sk_buff *skb;
2589 struct net_device *dev;
2590 struct virtio_net_hdr *vnet_hdr = NULL;
2591 struct sockcm_cookie sockc;
2592 __be16 proto;
2593 int err, reserve = 0;
2594 void *ph;
2595 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2596 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2597 unsigned char *addr = NULL;
2598 int tp_len, size_max;
2599 void *data;
2600 int len_sum = 0;
2601 int status = TP_STATUS_AVAILABLE;
2602 int hlen, tlen, copylen = 0;
2603
2604 mutex_lock(&po->pg_vec_lock);
2605
2606 if (likely(saddr == NULL)) {
2607 dev = packet_cached_dev_get(po);
2608 proto = po->num;
2609 } else {
2610 err = -EINVAL;
2611 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2612 goto out;
2613 if (msg->msg_namelen < (saddr->sll_halen
2614 + offsetof(struct sockaddr_ll,
2615 sll_addr)))
2616 goto out;
2617 proto = saddr->sll_protocol;
2618 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2619 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2620 if (dev && msg->msg_namelen < dev->addr_len +
2621 offsetof(struct sockaddr_ll, sll_addr))
2622 goto out_put;
2623 addr = saddr->sll_addr;
2624 }
2625 }
2626
2627 err = -ENXIO;
2628 if (unlikely(dev == NULL))
2629 goto out;
2630 err = -ENETDOWN;
2631 if (unlikely(!(dev->flags & IFF_UP)))
2632 goto out_put;
2633
2634 sockcm_init(&sockc, &po->sk);
2635 if (msg->msg_controllen) {
2636 err = sock_cmsg_send(&po->sk, msg, &sockc);
2637 if (unlikely(err))
2638 goto out_put;
2639 }
2640
2641 if (po->sk.sk_socket->type == SOCK_RAW)
2642 reserve = dev->hard_header_len;
2643 size_max = po->tx_ring.frame_size
2644 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2645
2646 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2647 size_max = dev->mtu + reserve + VLAN_HLEN;
2648
2649 do {
2650 ph = packet_current_frame(po, &po->tx_ring,
2651 TP_STATUS_SEND_REQUEST);
2652 if (unlikely(ph == NULL)) {
2653 if (need_wait && need_resched())
2654 schedule();
2655 continue;
2656 }
2657
2658 skb = NULL;
2659 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2660 if (tp_len < 0)
2661 goto tpacket_error;
2662
2663 status = TP_STATUS_SEND_REQUEST;
2664 hlen = LL_RESERVED_SPACE(dev);
2665 tlen = dev->needed_tailroom;
2666 if (po->has_vnet_hdr) {
2667 vnet_hdr = data;
2668 data += sizeof(*vnet_hdr);
2669 tp_len -= sizeof(*vnet_hdr);
2670 if (tp_len < 0 ||
2671 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2672 tp_len = -EINVAL;
2673 goto tpacket_error;
2674 }
2675 copylen = __virtio16_to_cpu(vio_le(),
2676 vnet_hdr->hdr_len);
2677 }
2678 copylen = max_t(int, copylen, dev->hard_header_len);
2679 skb = sock_alloc_send_skb(&po->sk,
2680 hlen + tlen + sizeof(struct sockaddr_ll) +
2681 (copylen - dev->hard_header_len),
2682 !need_wait, &err);
2683
2684 if (unlikely(skb == NULL)) {
2685 /* we assume the socket was initially writeable ... */
2686 if (likely(len_sum > 0))
2687 err = len_sum;
2688 goto out_status;
2689 }
2690 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2691 addr, hlen, copylen, &sockc);
2692 if (likely(tp_len >= 0) &&
2693 tp_len > dev->mtu + reserve &&
2694 !po->has_vnet_hdr &&
2695 !packet_extra_vlan_len_allowed(dev, skb))
2696 tp_len = -EMSGSIZE;
2697
2698 if (unlikely(tp_len < 0)) {
2699 tpacket_error:
2700 if (po->tp_loss) {
2701 __packet_set_status(po, ph,
2702 TP_STATUS_AVAILABLE);
2703 packet_increment_head(&po->tx_ring);
2704 kfree_skb(skb);
2705 continue;
2706 } else {
2707 status = TP_STATUS_WRONG_FORMAT;
2708 err = tp_len;
2709 goto out_status;
2710 }
2711 }
2712
2713 if (po->has_vnet_hdr) {
2714 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2715 tp_len = -EINVAL;
2716 goto tpacket_error;
2717 }
2718 virtio_net_hdr_set_proto(skb, vnet_hdr);
2719 }
2720
2721 skb->destructor = tpacket_destruct_skb;
2722 __packet_set_status(po, ph, TP_STATUS_SENDING);
2723 packet_inc_pending(&po->tx_ring);
2724
2725 status = TP_STATUS_SEND_REQUEST;
2726 err = po->xmit(skb);
2727 if (unlikely(err > 0)) {
2728 err = net_xmit_errno(err);
2729 if (err && __packet_get_status(po, ph) ==
2730 TP_STATUS_AVAILABLE) {
2731 /* skb was destructed already */
2732 skb = NULL;
2733 goto out_status;
2734 }
2735 /*
2736 * skb was dropped but not destructed yet;
2737 * let's treat it like congestion or err < 0
2738 */
2739 err = 0;
2740 }
2741 packet_increment_head(&po->tx_ring);
2742 len_sum += tp_len;
2743 } while (likely((ph != NULL) ||
2744 /* Note: packet_read_pending() might be slow if we have
2745 * to call it as it's per_cpu variable, but in fast-path
2746 * we already short-circuit the loop with the first
2747 * condition, and luckily don't have to go that path
2748 * anyway.
2749 */
2750 (need_wait && packet_read_pending(&po->tx_ring))));
2751
2752 err = len_sum;
2753 goto out_put;
2754
2755 out_status:
2756 __packet_set_status(po, ph, status);
2757 kfree_skb(skb);
2758 out_put:
2759 dev_put(dev);
2760 out:
2761 mutex_unlock(&po->pg_vec_lock);
2762 return err;
2763 }
2764
2765 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2766 size_t reserve, size_t len,
2767 size_t linear, int noblock,
2768 int *err)
2769 {
2770 struct sk_buff *skb;
2771
2772 /* Under a page? Don't bother with paged skb. */
2773 if (prepad + len < PAGE_SIZE || !linear)
2774 linear = len;
2775
2776 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2777 err, 0);
2778 if (!skb)
2779 return NULL;
2780
2781 skb_reserve(skb, reserve);
2782 skb_put(skb, linear);
2783 skb->data_len = len - linear;
2784 skb->len += len - linear;
2785
2786 return skb;
2787 }
2788
2789 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2790 {
2791 struct sock *sk = sock->sk;
2792 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2793 struct sk_buff *skb;
2794 struct net_device *dev;
2795 __be16 proto;
2796 unsigned char *addr = NULL;
2797 int err, reserve = 0;
2798 struct sockcm_cookie sockc;
2799 struct virtio_net_hdr vnet_hdr = { 0 };
2800 int offset = 0;
2801 struct packet_sock *po = pkt_sk(sk);
2802 bool has_vnet_hdr = false;
2803 int hlen, tlen, linear;
2804 int extra_len = 0;
2805
2806 /*
2807 * Get and verify the address.
2808 */
2809
2810 if (likely(saddr == NULL)) {
2811 dev = packet_cached_dev_get(po);
2812 proto = po->num;
2813 } else {
2814 err = -EINVAL;
2815 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2816 goto out;
2817 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2818 goto out;
2819 proto = saddr->sll_protocol;
2820 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2821 if (sock->type == SOCK_DGRAM) {
2822 if (dev && msg->msg_namelen < dev->addr_len +
2823 offsetof(struct sockaddr_ll, sll_addr))
2824 goto out_unlock;
2825 addr = saddr->sll_addr;
2826 }
2827 }
2828
2829 err = -ENXIO;
2830 if (unlikely(dev == NULL))
2831 goto out_unlock;
2832 err = -ENETDOWN;
2833 if (unlikely(!(dev->flags & IFF_UP)))
2834 goto out_unlock;
2835
2836 sockcm_init(&sockc, sk);
2837 sockc.mark = sk->sk_mark;
2838 if (msg->msg_controllen) {
2839 err = sock_cmsg_send(sk, msg, &sockc);
2840 if (unlikely(err))
2841 goto out_unlock;
2842 }
2843
2844 if (sock->type == SOCK_RAW)
2845 reserve = dev->hard_header_len;
2846 if (po->has_vnet_hdr) {
2847 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2848 if (err)
2849 goto out_unlock;
2850 has_vnet_hdr = true;
2851 }
2852
2853 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2854 if (!netif_supports_nofcs(dev)) {
2855 err = -EPROTONOSUPPORT;
2856 goto out_unlock;
2857 }
2858 extra_len = 4; /* We're doing our own CRC */
2859 }
2860
2861 err = -EMSGSIZE;
2862 if (!vnet_hdr.gso_type &&
2863 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2864 goto out_unlock;
2865
2866 err = -ENOBUFS;
2867 hlen = LL_RESERVED_SPACE(dev);
2868 tlen = dev->needed_tailroom;
2869 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2870 linear = max(linear, min_t(int, len, dev->hard_header_len));
2871 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2872 msg->msg_flags & MSG_DONTWAIT, &err);
2873 if (skb == NULL)
2874 goto out_unlock;
2875
2876 skb_reset_network_header(skb);
2877
2878 err = -EINVAL;
2879 if (sock->type == SOCK_DGRAM) {
2880 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2881 if (unlikely(offset < 0))
2882 goto out_free;
2883 } else if (reserve) {
2884 skb_reserve(skb, -reserve);
2885 if (len < reserve + sizeof(struct ipv6hdr) &&
2886 dev->min_header_len != dev->hard_header_len)
2887 skb_reset_network_header(skb);
2888 }
2889
2890 /* Returns -EFAULT on error */
2891 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2892 if (err)
2893 goto out_free;
2894
2895 if (sock->type == SOCK_RAW &&
2896 !dev_validate_header(dev, skb->data, len)) {
2897 err = -EINVAL;
2898 goto out_free;
2899 }
2900
2901 skb_setup_tx_timestamp(skb, sockc.tsflags);
2902
2903 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2904 !packet_extra_vlan_len_allowed(dev, skb)) {
2905 err = -EMSGSIZE;
2906 goto out_free;
2907 }
2908
2909 skb->protocol = proto;
2910 skb->dev = dev;
2911 skb->priority = sk->sk_priority;
2912 skb->mark = sockc.mark;
2913 skb->tstamp = sockc.transmit_time;
2914
2915 if (has_vnet_hdr) {
2916 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2917 if (err)
2918 goto out_free;
2919 len += sizeof(vnet_hdr);
2920 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2921 }
2922
2923 packet_parse_headers(skb, sock);
2924
2925 if (unlikely(extra_len == 4))
2926 skb->no_fcs = 1;
2927
2928 err = po->xmit(skb);
2929 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2930 goto out_unlock;
2931
2932 dev_put(dev);
2933
2934 return len;
2935
2936 out_free:
2937 kfree_skb(skb);
2938 out_unlock:
2939 if (dev)
2940 dev_put(dev);
2941 out:
2942 return err;
2943 }
2944
2945 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2946 {
2947 struct sock *sk = sock->sk;
2948 struct packet_sock *po = pkt_sk(sk);
2949
2950 if (po->tx_ring.pg_vec)
2951 return tpacket_snd(po, msg);
2952 else
2953 return packet_snd(sock, msg, len);
2954 }
2955
2956 /*
2957 * Close a PACKET socket. This is fairly simple. We immediately go
2958 * to 'closed' state and remove our protocol entry in the device list.
2959 */
2960
2961 static int packet_release(struct socket *sock)
2962 {
2963 struct sock *sk = sock->sk;
2964 struct packet_sock *po;
2965 struct packet_fanout *f;
2966 struct net *net;
2967 union tpacket_req_u req_u;
2968
2969 if (!sk)
2970 return 0;
2971
2972 net = sock_net(sk);
2973 po = pkt_sk(sk);
2974
2975 mutex_lock(&net->packet.sklist_lock);
2976 sk_del_node_init_rcu(sk);
2977 mutex_unlock(&net->packet.sklist_lock);
2978
2979 preempt_disable();
2980 sock_prot_inuse_add(net, sk->sk_prot, -1);
2981 preempt_enable();
2982
2983 spin_lock(&po->bind_lock);
2984 unregister_prot_hook(sk, false);
2985 packet_cached_dev_reset(po);
2986
2987 if (po->prot_hook.dev) {
2988 dev_put(po->prot_hook.dev);
2989 po->prot_hook.dev = NULL;
2990 }
2991 spin_unlock(&po->bind_lock);
2992
2993 packet_flush_mclist(sk);
2994
2995 lock_sock(sk);
2996 if (po->rx_ring.pg_vec) {
2997 memset(&req_u, 0, sizeof(req_u));
2998 packet_set_ring(sk, &req_u, 1, 0);
2999 }
3000
3001 if (po->tx_ring.pg_vec) {
3002 memset(&req_u, 0, sizeof(req_u));
3003 packet_set_ring(sk, &req_u, 1, 1);
3004 }
3005 release_sock(sk);
3006
3007 f = fanout_release(sk);
3008
3009 synchronize_net();
3010
3011 kfree(po->rollover);
3012 if (f) {
3013 fanout_release_data(f);
3014 kfree(f);
3015 }
3016 /*
3017 * Now the socket is dead. No more input will appear.
3018 */
3019 sock_orphan(sk);
3020 sock->sk = NULL;
3021
3022 /* Purge queues */
3023
3024 skb_queue_purge(&sk->sk_receive_queue);
3025 packet_free_pending(po);
3026 sk_refcnt_debug_release(sk);
3027
3028 sock_put(sk);
3029 return 0;
3030 }
3031
3032 /*
3033 * Attach a packet hook.
3034 */
3035
3036 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3037 __be16 proto)
3038 {
3039 struct packet_sock *po = pkt_sk(sk);
3040 struct net_device *dev_curr;
3041 __be16 proto_curr;
3042 bool need_rehook;
3043 struct net_device *dev = NULL;
3044 int ret = 0;
3045 bool unlisted = false;
3046
3047 lock_sock(sk);
3048 spin_lock(&po->bind_lock);
3049 rcu_read_lock();
3050
3051 if (po->fanout) {
3052 ret = -EINVAL;
3053 goto out_unlock;
3054 }
3055
3056 if (name) {
3057 dev = dev_get_by_name_rcu(sock_net(sk), name);
3058 if (!dev) {
3059 ret = -ENODEV;
3060 goto out_unlock;
3061 }
3062 } else if (ifindex) {
3063 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3064 if (!dev) {
3065 ret = -ENODEV;
3066 goto out_unlock;
3067 }
3068 }
3069
3070 if (dev)
3071 dev_hold(dev);
3072
3073 proto_curr = po->prot_hook.type;
3074 dev_curr = po->prot_hook.dev;
3075
3076 need_rehook = proto_curr != proto || dev_curr != dev;
3077
3078 if (need_rehook) {
3079 if (po->running) {
3080 rcu_read_unlock();
3081 /* prevents packet_notifier() from calling
3082 * register_prot_hook()
3083 */
3084 po->num = 0;
3085 __unregister_prot_hook(sk, true);
3086 rcu_read_lock();
3087 dev_curr = po->prot_hook.dev;
3088 if (dev)
3089 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3090 dev->ifindex);
3091 }
3092
3093 BUG_ON(po->running);
3094 po->num = proto;
3095 po->prot_hook.type = proto;
3096
3097 if (unlikely(unlisted)) {
3098 dev_put(dev);
3099 po->prot_hook.dev = NULL;
3100 po->ifindex = -1;
3101 packet_cached_dev_reset(po);
3102 } else {
3103 po->prot_hook.dev = dev;
3104 po->ifindex = dev ? dev->ifindex : 0;
3105 packet_cached_dev_assign(po, dev);
3106 }
3107 }
3108 if (dev_curr)
3109 dev_put(dev_curr);
3110
3111 if (proto == 0 || !need_rehook)
3112 goto out_unlock;
3113
3114 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3115 register_prot_hook(sk);
3116 } else {
3117 sk->sk_err = ENETDOWN;
3118 if (!sock_flag(sk, SOCK_DEAD))
3119 sk->sk_error_report(sk);
3120 }
3121
3122 out_unlock:
3123 rcu_read_unlock();
3124 spin_unlock(&po->bind_lock);
3125 release_sock(sk);
3126 return ret;
3127 }
3128
3129 /*
3130 * Bind a packet socket to a device
3131 */
3132
3133 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3134 int addr_len)
3135 {
3136 struct sock *sk = sock->sk;
3137 char name[sizeof(uaddr->sa_data) + 1];
3138
3139 /*
3140 * Check legality
3141 */
3142
3143 if (addr_len != sizeof(struct sockaddr))
3144 return -EINVAL;
3145 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3146 * zero-terminated.
3147 */
3148 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3149 name[sizeof(uaddr->sa_data)] = 0;
3150
3151 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3152 }
3153
3154 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3155 {
3156 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3157 struct sock *sk = sock->sk;
3158
3159 /*
3160 * Check legality
3161 */
3162
3163 if (addr_len < sizeof(struct sockaddr_ll))
3164 return -EINVAL;
3165 if (sll->sll_family != AF_PACKET)
3166 return -EINVAL;
3167
3168 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3169 sll->sll_protocol ? : pkt_sk(sk)->num);
3170 }
3171
3172 static struct proto packet_proto = {
3173 .name = "PACKET",
3174 .owner = THIS_MODULE,
3175 .obj_size = sizeof(struct packet_sock),
3176 };
3177
3178 /*
3179 * Create a packet of type SOCK_PACKET.
3180 */
3181
3182 static int packet_create(struct net *net, struct socket *sock, int protocol,
3183 int kern)
3184 {
3185 struct sock *sk;
3186 struct packet_sock *po;
3187 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3188 int err;
3189
3190 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3191 return -EPERM;
3192 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3193 sock->type != SOCK_PACKET)
3194 return -ESOCKTNOSUPPORT;
3195
3196 sock->state = SS_UNCONNECTED;
3197
3198 err = -ENOBUFS;
3199 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3200 if (sk == NULL)
3201 goto out;
3202
3203 sock->ops = &packet_ops;
3204 if (sock->type == SOCK_PACKET)
3205 sock->ops = &packet_ops_spkt;
3206
3207 sock_init_data(sock, sk);
3208
3209 po = pkt_sk(sk);
3210 sk->sk_family = PF_PACKET;
3211 po->num = proto;
3212 po->xmit = dev_queue_xmit;
3213
3214 err = packet_alloc_pending(po);
3215 if (err)
3216 goto out2;
3217
3218 packet_cached_dev_reset(po);
3219
3220 sk->sk_destruct = packet_sock_destruct;
3221 sk_refcnt_debug_inc(sk);
3222
3223 /*
3224 * Attach a protocol block
3225 */
3226
3227 spin_lock_init(&po->bind_lock);
3228 mutex_init(&po->pg_vec_lock);
3229 po->rollover = NULL;
3230 po->prot_hook.func = packet_rcv;
3231
3232 if (sock->type == SOCK_PACKET)
3233 po->prot_hook.func = packet_rcv_spkt;
3234
3235 po->prot_hook.af_packet_priv = sk;
3236
3237 if (proto) {
3238 po->prot_hook.type = proto;
3239 __register_prot_hook(sk);
3240 }
3241
3242 mutex_lock(&net->packet.sklist_lock);
3243 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3244 mutex_unlock(&net->packet.sklist_lock);
3245
3246 preempt_disable();
3247 sock_prot_inuse_add(net, &packet_proto, 1);
3248 preempt_enable();
3249
3250 return 0;
3251 out2:
3252 sk_free(sk);
3253 out:
3254 return err;
3255 }
3256
3257 /*
3258 * Pull a packet from our receive queue and hand it to the user.
3259 * If necessary we block.
3260 */
3261
3262 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3263 int flags)
3264 {
3265 struct sock *sk = sock->sk;
3266 struct sk_buff *skb;
3267 int copied, err;
3268 int vnet_hdr_len = 0;
3269 unsigned int origlen = 0;
3270
3271 err = -EINVAL;
3272 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3273 goto out;
3274
3275 #if 0
3276 /* What error should we return now? EUNATTACH? */
3277 if (pkt_sk(sk)->ifindex < 0)
3278 return -ENODEV;
3279 #endif
3280
3281 if (flags & MSG_ERRQUEUE) {
3282 err = sock_recv_errqueue(sk, msg, len,
3283 SOL_PACKET, PACKET_TX_TIMESTAMP);
3284 goto out;
3285 }
3286
3287 /*
3288 * Call the generic datagram receiver. This handles all sorts
3289 * of horrible races and re-entrancy so we can forget about it
3290 * in the protocol layers.
3291 *
3292 * Now it will return ENETDOWN, if device have just gone down,
3293 * but then it will block.
3294 */
3295
3296 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3297
3298 /*
3299 * An error occurred so return it. Because skb_recv_datagram()
3300 * handles the blocking we don't see and worry about blocking
3301 * retries.
3302 */
3303
3304 if (skb == NULL)
3305 goto out;
3306
3307 if (pkt_sk(sk)->pressure)
3308 packet_rcv_has_room(pkt_sk(sk), NULL);
3309
3310 if (pkt_sk(sk)->has_vnet_hdr) {
3311 err = packet_rcv_vnet(msg, skb, &len);
3312 if (err)
3313 goto out_free;
3314 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3315 }
3316
3317 /* You lose any data beyond the buffer you gave. If it worries
3318 * a user program they can ask the device for its MTU
3319 * anyway.
3320 */
3321 copied = skb->len;
3322 if (copied > len) {
3323 copied = len;
3324 msg->msg_flags |= MSG_TRUNC;
3325 }
3326
3327 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3328 if (err)
3329 goto out_free;
3330
3331 if (sock->type != SOCK_PACKET) {
3332 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3333
3334 /* Original length was stored in sockaddr_ll fields */
3335 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3336 sll->sll_family = AF_PACKET;
3337 sll->sll_protocol = skb->protocol;
3338 }
3339
3340 sock_recv_ts_and_drops(msg, sk, skb);
3341
3342 if (msg->msg_name) {
3343 int copy_len;
3344
3345 /* If the address length field is there to be filled
3346 * in, we fill it in now.
3347 */
3348 if (sock->type == SOCK_PACKET) {
3349 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3350 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3351 copy_len = msg->msg_namelen;
3352 } else {
3353 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3354
3355 msg->msg_namelen = sll->sll_halen +
3356 offsetof(struct sockaddr_ll, sll_addr);
3357 copy_len = msg->msg_namelen;
3358 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3359 memset(msg->msg_name +
3360 offsetof(struct sockaddr_ll, sll_addr),
3361 0, sizeof(sll->sll_addr));
3362 msg->msg_namelen = sizeof(struct sockaddr_ll);
3363 }
3364 }
3365 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3366 }
3367
3368 if (pkt_sk(sk)->auxdata) {
3369 struct tpacket_auxdata aux;
3370
3371 aux.tp_status = TP_STATUS_USER;
3372 if (skb->ip_summed == CHECKSUM_PARTIAL)
3373 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3374 else if (skb->pkt_type != PACKET_OUTGOING &&
3375 (skb->ip_summed == CHECKSUM_COMPLETE ||
3376 skb_csum_unnecessary(skb)))
3377 aux.tp_status |= TP_STATUS_CSUM_VALID;
3378
3379 aux.tp_len = origlen;
3380 aux.tp_snaplen = skb->len;
3381 aux.tp_mac = 0;
3382 aux.tp_net = skb_network_offset(skb);
3383 if (skb_vlan_tag_present(skb)) {
3384 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3385 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3386 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3387 } else {
3388 aux.tp_vlan_tci = 0;
3389 aux.tp_vlan_tpid = 0;
3390 }
3391 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3392 }
3393
3394 /*
3395 * Free or return the buffer as appropriate. Again this
3396 * hides all the races and re-entrancy issues from us.
3397 */
3398 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3399
3400 out_free:
3401 skb_free_datagram(sk, skb);
3402 out:
3403 return err;
3404 }
3405
3406 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3407 int peer)
3408 {
3409 struct net_device *dev;
3410 struct sock *sk = sock->sk;
3411
3412 if (peer)
3413 return -EOPNOTSUPP;
3414
3415 uaddr->sa_family = AF_PACKET;
3416 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3417 rcu_read_lock();
3418 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3419 if (dev)
3420 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3421 rcu_read_unlock();
3422
3423 return sizeof(*uaddr);
3424 }
3425
3426 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3427 int peer)
3428 {
3429 struct net_device *dev;
3430 struct sock *sk = sock->sk;
3431 struct packet_sock *po = pkt_sk(sk);
3432 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3433
3434 if (peer)
3435 return -EOPNOTSUPP;
3436
3437 sll->sll_family = AF_PACKET;
3438 sll->sll_ifindex = po->ifindex;
3439 sll->sll_protocol = po->num;
3440 sll->sll_pkttype = 0;
3441 rcu_read_lock();
3442 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3443 if (dev) {
3444 sll->sll_hatype = dev->type;
3445 sll->sll_halen = dev->addr_len;
3446 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3447 } else {
3448 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3449 sll->sll_halen = 0;
3450 }
3451 rcu_read_unlock();
3452
3453 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3454 }
3455
3456 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3457 int what)
3458 {
3459 switch (i->type) {
3460 case PACKET_MR_MULTICAST:
3461 if (i->alen != dev->addr_len)
3462 return -EINVAL;
3463 if (what > 0)
3464 return dev_mc_add(dev, i->addr);
3465 else
3466 return dev_mc_del(dev, i->addr);
3467 break;
3468 case PACKET_MR_PROMISC:
3469 return dev_set_promiscuity(dev, what);
3470 case PACKET_MR_ALLMULTI:
3471 return dev_set_allmulti(dev, what);
3472 case PACKET_MR_UNICAST:
3473 if (i->alen != dev->addr_len)
3474 return -EINVAL;
3475 if (what > 0)
3476 return dev_uc_add(dev, i->addr);
3477 else
3478 return dev_uc_del(dev, i->addr);
3479 break;
3480 default:
3481 break;
3482 }
3483 return 0;
3484 }
3485
3486 static void packet_dev_mclist_delete(struct net_device *dev,
3487 struct packet_mclist **mlp)
3488 {
3489 struct packet_mclist *ml;
3490
3491 while ((ml = *mlp) != NULL) {
3492 if (ml->ifindex == dev->ifindex) {
3493 packet_dev_mc(dev, ml, -1);
3494 *mlp = ml->next;
3495 kfree(ml);
3496 } else
3497 mlp = &ml->next;
3498 }
3499 }
3500
3501 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3502 {
3503 struct packet_sock *po = pkt_sk(sk);
3504 struct packet_mclist *ml, *i;
3505 struct net_device *dev;
3506 int err;
3507
3508 rtnl_lock();
3509
3510 err = -ENODEV;
3511 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3512 if (!dev)
3513 goto done;
3514
3515 err = -EINVAL;
3516 if (mreq->mr_alen > dev->addr_len)
3517 goto done;
3518
3519 err = -ENOBUFS;
3520 i = kmalloc(sizeof(*i), GFP_KERNEL);
3521 if (i == NULL)
3522 goto done;
3523
3524 err = 0;
3525 for (ml = po->mclist; ml; ml = ml->next) {
3526 if (ml->ifindex == mreq->mr_ifindex &&
3527 ml->type == mreq->mr_type &&
3528 ml->alen == mreq->mr_alen &&
3529 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3530 ml->count++;
3531 /* Free the new element ... */
3532 kfree(i);
3533 goto done;
3534 }
3535 }
3536
3537 i->type = mreq->mr_type;
3538 i->ifindex = mreq->mr_ifindex;
3539 i->alen = mreq->mr_alen;
3540 memcpy(i->addr, mreq->mr_address, i->alen);
3541 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3542 i->count = 1;
3543 i->next = po->mclist;
3544 po->mclist = i;
3545 err = packet_dev_mc(dev, i, 1);
3546 if (err) {
3547 po->mclist = i->next;
3548 kfree(i);
3549 }
3550
3551 done:
3552 rtnl_unlock();
3553 return err;
3554 }
3555
3556 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3557 {
3558 struct packet_mclist *ml, **mlp;
3559
3560 rtnl_lock();
3561
3562 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3563 if (ml->ifindex == mreq->mr_ifindex &&
3564 ml->type == mreq->mr_type &&
3565 ml->alen == mreq->mr_alen &&
3566 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3567 if (--ml->count == 0) {
3568 struct net_device *dev;
3569 *mlp = ml->next;
3570 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3571 if (dev)
3572 packet_dev_mc(dev, ml, -1);
3573 kfree(ml);
3574 }
3575 break;
3576 }
3577 }
3578 rtnl_unlock();
3579 return 0;
3580 }
3581
3582 static void packet_flush_mclist(struct sock *sk)
3583 {
3584 struct packet_sock *po = pkt_sk(sk);
3585 struct packet_mclist *ml;
3586
3587 if (!po->mclist)
3588 return;
3589
3590 rtnl_lock();
3591 while ((ml = po->mclist) != NULL) {
3592 struct net_device *dev;
3593
3594 po->mclist = ml->next;
3595 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3596 if (dev != NULL)
3597 packet_dev_mc(dev, ml, -1);
3598 kfree(ml);
3599 }
3600 rtnl_unlock();
3601 }
3602
3603 static int
3604 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3605 {
3606 struct sock *sk = sock->sk;
3607 struct packet_sock *po = pkt_sk(sk);
3608 int ret;
3609
3610 if (level != SOL_PACKET)
3611 return -ENOPROTOOPT;
3612
3613 switch (optname) {
3614 case PACKET_ADD_MEMBERSHIP:
3615 case PACKET_DROP_MEMBERSHIP:
3616 {
3617 struct packet_mreq_max mreq;
3618 int len = optlen;
3619 memset(&mreq, 0, sizeof(mreq));
3620 if (len < sizeof(struct packet_mreq))
3621 return -EINVAL;
3622 if (len > sizeof(mreq))
3623 len = sizeof(mreq);
3624 if (copy_from_user(&mreq, optval, len))
3625 return -EFAULT;
3626 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3627 return -EINVAL;
3628 if (optname == PACKET_ADD_MEMBERSHIP)
3629 ret = packet_mc_add(sk, &mreq);
3630 else
3631 ret = packet_mc_drop(sk, &mreq);
3632 return ret;
3633 }
3634
3635 case PACKET_RX_RING:
3636 case PACKET_TX_RING:
3637 {
3638 union tpacket_req_u req_u;
3639 int len;
3640
3641 lock_sock(sk);
3642 switch (po->tp_version) {
3643 case TPACKET_V1:
3644 case TPACKET_V2:
3645 len = sizeof(req_u.req);
3646 break;
3647 case TPACKET_V3:
3648 default:
3649 len = sizeof(req_u.req3);
3650 break;
3651 }
3652 if (optlen < len) {
3653 ret = -EINVAL;
3654 } else {
3655 if (copy_from_user(&req_u.req, optval, len))
3656 ret = -EFAULT;
3657 else
3658 ret = packet_set_ring(sk, &req_u, 0,
3659 optname == PACKET_TX_RING);
3660 }
3661 release_sock(sk);
3662 return ret;
3663 }
3664 case PACKET_COPY_THRESH:
3665 {
3666 int val;
3667
3668 if (optlen != sizeof(val))
3669 return -EINVAL;
3670 if (copy_from_user(&val, optval, sizeof(val)))
3671 return -EFAULT;
3672
3673 pkt_sk(sk)->copy_thresh = val;
3674 return 0;
3675 }
3676 case PACKET_VERSION:
3677 {
3678 int val;
3679
3680 if (optlen != sizeof(val))
3681 return -EINVAL;
3682 if (copy_from_user(&val, optval, sizeof(val)))
3683 return -EFAULT;
3684 switch (val) {
3685 case TPACKET_V1:
3686 case TPACKET_V2:
3687 case TPACKET_V3:
3688 break;
3689 default:
3690 return -EINVAL;
3691 }
3692 lock_sock(sk);
3693 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3694 ret = -EBUSY;
3695 } else {
3696 po->tp_version = val;
3697 ret = 0;
3698 }
3699 release_sock(sk);
3700 return ret;
3701 }
3702 case PACKET_RESERVE:
3703 {
3704 unsigned int val;
3705
3706 if (optlen != sizeof(val))
3707 return -EINVAL;
3708 if (copy_from_user(&val, optval, sizeof(val)))
3709 return -EFAULT;
3710 if (val > INT_MAX)
3711 return -EINVAL;
3712 lock_sock(sk);
3713 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3714 ret = -EBUSY;
3715 } else {
3716 po->tp_reserve = val;
3717 ret = 0;
3718 }
3719 release_sock(sk);
3720 return ret;
3721 }
3722 case PACKET_LOSS:
3723 {
3724 unsigned int val;
3725
3726 if (optlen != sizeof(val))
3727 return -EINVAL;
3728 if (copy_from_user(&val, optval, sizeof(val)))
3729 return -EFAULT;
3730
3731 lock_sock(sk);
3732 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3733 ret = -EBUSY;
3734 } else {
3735 po->tp_loss = !!val;
3736 ret = 0;
3737 }
3738 release_sock(sk);
3739 return ret;
3740 }
3741 case PACKET_AUXDATA:
3742 {
3743 int val;
3744
3745 if (optlen < sizeof(val))
3746 return -EINVAL;
3747 if (copy_from_user(&val, optval, sizeof(val)))
3748 return -EFAULT;
3749
3750 lock_sock(sk);
3751 po->auxdata = !!val;
3752 release_sock(sk);
3753 return 0;
3754 }
3755 case PACKET_ORIGDEV:
3756 {
3757 int val;
3758
3759 if (optlen < sizeof(val))
3760 return -EINVAL;
3761 if (copy_from_user(&val, optval, sizeof(val)))
3762 return -EFAULT;
3763
3764 lock_sock(sk);
3765 po->origdev = !!val;
3766 release_sock(sk);
3767 return 0;
3768 }
3769 case PACKET_VNET_HDR:
3770 {
3771 int val;
3772
3773 if (sock->type != SOCK_RAW)
3774 return -EINVAL;
3775 if (optlen < sizeof(val))
3776 return -EINVAL;
3777 if (copy_from_user(&val, optval, sizeof(val)))
3778 return -EFAULT;
3779
3780 lock_sock(sk);
3781 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3782 ret = -EBUSY;
3783 } else {
3784 po->has_vnet_hdr = !!val;
3785 ret = 0;
3786 }
3787 release_sock(sk);
3788 return ret;
3789 }
3790 case PACKET_TIMESTAMP:
3791 {
3792 int val;
3793
3794 if (optlen != sizeof(val))
3795 return -EINVAL;
3796 if (copy_from_user(&val, optval, sizeof(val)))
3797 return -EFAULT;
3798
3799 po->tp_tstamp = val;
3800 return 0;
3801 }
3802 case PACKET_FANOUT:
3803 {
3804 int val;
3805
3806 if (optlen != sizeof(val))
3807 return -EINVAL;
3808 if (copy_from_user(&val, optval, sizeof(val)))
3809 return -EFAULT;
3810
3811 return fanout_add(sk, val & 0xffff, val >> 16);
3812 }
3813 case PACKET_FANOUT_DATA:
3814 {
3815 if (!po->fanout)
3816 return -EINVAL;
3817
3818 return fanout_set_data(po, optval, optlen);
3819 }
3820 case PACKET_IGNORE_OUTGOING:
3821 {
3822 int val;
3823
3824 if (optlen != sizeof(val))
3825 return -EINVAL;
3826 if (copy_from_user(&val, optval, sizeof(val)))
3827 return -EFAULT;
3828 if (val < 0 || val > 1)
3829 return -EINVAL;
3830
3831 po->prot_hook.ignore_outgoing = !!val;
3832 return 0;
3833 }
3834 case PACKET_TX_HAS_OFF:
3835 {
3836 unsigned int val;
3837
3838 if (optlen != sizeof(val))
3839 return -EINVAL;
3840 if (copy_from_user(&val, optval, sizeof(val)))
3841 return -EFAULT;
3842
3843 lock_sock(sk);
3844 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3845 ret = -EBUSY;
3846 } else {
3847 po->tp_tx_has_off = !!val;
3848 ret = 0;
3849 }
3850 release_sock(sk);
3851 return 0;
3852 }
3853 case PACKET_QDISC_BYPASS:
3854 {
3855 int val;
3856
3857 if (optlen != sizeof(val))
3858 return -EINVAL;
3859 if (copy_from_user(&val, optval, sizeof(val)))
3860 return -EFAULT;
3861
3862 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3863 return 0;
3864 }
3865 default:
3866 return -ENOPROTOOPT;
3867 }
3868 }
3869
3870 static int packet_getsockopt(struct socket *sock, int level, int optname,
3871 char __user *optval, int __user *optlen)
3872 {
3873 int len;
3874 int val, lv = sizeof(val);
3875 struct sock *sk = sock->sk;
3876 struct packet_sock *po = pkt_sk(sk);
3877 void *data = &val;
3878 union tpacket_stats_u st;
3879 struct tpacket_rollover_stats rstats;
3880
3881 if (level != SOL_PACKET)
3882 return -ENOPROTOOPT;
3883
3884 if (get_user(len, optlen))
3885 return -EFAULT;
3886
3887 if (len < 0)
3888 return -EINVAL;
3889
3890 switch (optname) {
3891 case PACKET_STATISTICS:
3892 spin_lock_bh(&sk->sk_receive_queue.lock);
3893 memcpy(&st, &po->stats, sizeof(st));
3894 memset(&po->stats, 0, sizeof(po->stats));
3895 spin_unlock_bh(&sk->sk_receive_queue.lock);
3896
3897 if (po->tp_version == TPACKET_V3) {
3898 lv = sizeof(struct tpacket_stats_v3);
3899 st.stats3.tp_packets += st.stats3.tp_drops;
3900 data = &st.stats3;
3901 } else {
3902 lv = sizeof(struct tpacket_stats);
3903 st.stats1.tp_packets += st.stats1.tp_drops;
3904 data = &st.stats1;
3905 }
3906
3907 break;
3908 case PACKET_AUXDATA:
3909 val = po->auxdata;
3910 break;
3911 case PACKET_ORIGDEV:
3912 val = po->origdev;
3913 break;
3914 case PACKET_VNET_HDR:
3915 val = po->has_vnet_hdr;
3916 break;
3917 case PACKET_VERSION:
3918 val = po->tp_version;
3919 break;
3920 case PACKET_HDRLEN:
3921 if (len > sizeof(int))
3922 len = sizeof(int);
3923 if (len < sizeof(int))
3924 return -EINVAL;
3925 if (copy_from_user(&val, optval, len))
3926 return -EFAULT;
3927 switch (val) {
3928 case TPACKET_V1:
3929 val = sizeof(struct tpacket_hdr);
3930 break;
3931 case TPACKET_V2:
3932 val = sizeof(struct tpacket2_hdr);
3933 break;
3934 case TPACKET_V3:
3935 val = sizeof(struct tpacket3_hdr);
3936 break;
3937 default:
3938 return -EINVAL;
3939 }
3940 break;
3941 case PACKET_RESERVE:
3942 val = po->tp_reserve;
3943 break;
3944 case PACKET_LOSS:
3945 val = po->tp_loss;
3946 break;
3947 case PACKET_TIMESTAMP:
3948 val = po->tp_tstamp;
3949 break;
3950 case PACKET_FANOUT:
3951 val = (po->fanout ?
3952 ((u32)po->fanout->id |
3953 ((u32)po->fanout->type << 16) |
3954 ((u32)po->fanout->flags << 24)) :
3955 0);
3956 break;
3957 case PACKET_IGNORE_OUTGOING:
3958 val = po->prot_hook.ignore_outgoing;
3959 break;
3960 case PACKET_ROLLOVER_STATS:
3961 if (!po->rollover)
3962 return -EINVAL;
3963 rstats.tp_all = atomic_long_read(&po->rollover->num);
3964 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3965 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3966 data = &rstats;
3967 lv = sizeof(rstats);
3968 break;
3969 case PACKET_TX_HAS_OFF:
3970 val = po->tp_tx_has_off;
3971 break;
3972 case PACKET_QDISC_BYPASS:
3973 val = packet_use_direct_xmit(po);
3974 break;
3975 default:
3976 return -ENOPROTOOPT;
3977 }
3978
3979 if (len > lv)
3980 len = lv;
3981 if (put_user(len, optlen))
3982 return -EFAULT;
3983 if (copy_to_user(optval, data, len))
3984 return -EFAULT;
3985 return 0;
3986 }
3987
3988
3989 #ifdef CONFIG_COMPAT
3990 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3991 char __user *optval, unsigned int optlen)
3992 {
3993 struct packet_sock *po = pkt_sk(sock->sk);
3994
3995 if (level != SOL_PACKET)
3996 return -ENOPROTOOPT;
3997
3998 if (optname == PACKET_FANOUT_DATA &&
3999 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4000 optval = (char __user *)get_compat_bpf_fprog(optval);
4001 if (!optval)
4002 return -EFAULT;
4003 optlen = sizeof(struct sock_fprog);
4004 }
4005
4006 return packet_setsockopt(sock, level, optname, optval, optlen);
4007 }
4008 #endif
4009
4010 static int packet_notifier(struct notifier_block *this,
4011 unsigned long msg, void *ptr)
4012 {
4013 struct sock *sk;
4014 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4015 struct net *net = dev_net(dev);
4016
4017 rcu_read_lock();
4018 sk_for_each_rcu(sk, &net->packet.sklist) {
4019 struct packet_sock *po = pkt_sk(sk);
4020
4021 switch (msg) {
4022 case NETDEV_UNREGISTER:
4023 if (po->mclist)
4024 packet_dev_mclist_delete(dev, &po->mclist);
4025 /* fallthrough */
4026
4027 case NETDEV_DOWN:
4028 if (dev->ifindex == po->ifindex) {
4029 spin_lock(&po->bind_lock);
4030 if (po->running) {
4031 __unregister_prot_hook(sk, false);
4032 sk->sk_err = ENETDOWN;
4033 if (!sock_flag(sk, SOCK_DEAD))
4034 sk->sk_error_report(sk);
4035 }
4036 if (msg == NETDEV_UNREGISTER) {
4037 packet_cached_dev_reset(po);
4038 po->ifindex = -1;
4039 if (po->prot_hook.dev)
4040 dev_put(po->prot_hook.dev);
4041 po->prot_hook.dev = NULL;
4042 }
4043 spin_unlock(&po->bind_lock);
4044 }
4045 break;
4046 case NETDEV_UP:
4047 if (dev->ifindex == po->ifindex) {
4048 spin_lock(&po->bind_lock);
4049 if (po->num)
4050 register_prot_hook(sk);
4051 spin_unlock(&po->bind_lock);
4052 }
4053 break;
4054 }
4055 }
4056 rcu_read_unlock();
4057 return NOTIFY_DONE;
4058 }
4059
4060
4061 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4062 unsigned long arg)
4063 {
4064 struct sock *sk = sock->sk;
4065
4066 switch (cmd) {
4067 case SIOCOUTQ:
4068 {
4069 int amount = sk_wmem_alloc_get(sk);
4070
4071 return put_user(amount, (int __user *)arg);
4072 }
4073 case SIOCINQ:
4074 {
4075 struct sk_buff *skb;
4076 int amount = 0;
4077
4078 spin_lock_bh(&sk->sk_receive_queue.lock);
4079 skb = skb_peek(&sk->sk_receive_queue);
4080 if (skb)
4081 amount = skb->len;
4082 spin_unlock_bh(&sk->sk_receive_queue.lock);
4083 return put_user(amount, (int __user *)arg);
4084 }
4085 #ifdef CONFIG_INET
4086 case SIOCADDRT:
4087 case SIOCDELRT:
4088 case SIOCDARP:
4089 case SIOCGARP:
4090 case SIOCSARP:
4091 case SIOCGIFADDR:
4092 case SIOCSIFADDR:
4093 case SIOCGIFBRDADDR:
4094 case SIOCSIFBRDADDR:
4095 case SIOCGIFNETMASK:
4096 case SIOCSIFNETMASK:
4097 case SIOCGIFDSTADDR:
4098 case SIOCSIFDSTADDR:
4099 case SIOCSIFFLAGS:
4100 return inet_dgram_ops.ioctl(sock, cmd, arg);
4101 #endif
4102
4103 default:
4104 return -ENOIOCTLCMD;
4105 }
4106 return 0;
4107 }
4108
4109 static __poll_t packet_poll(struct file *file, struct socket *sock,
4110 poll_table *wait)
4111 {
4112 struct sock *sk = sock->sk;
4113 struct packet_sock *po = pkt_sk(sk);
4114 __poll_t mask = datagram_poll(file, sock, wait);
4115
4116 spin_lock_bh(&sk->sk_receive_queue.lock);
4117 if (po->rx_ring.pg_vec) {
4118 if (!packet_previous_rx_frame(po, &po->rx_ring,
4119 TP_STATUS_KERNEL))
4120 mask |= EPOLLIN | EPOLLRDNORM;
4121 }
4122 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4123 po->pressure = 0;
4124 spin_unlock_bh(&sk->sk_receive_queue.lock);
4125 spin_lock_bh(&sk->sk_write_queue.lock);
4126 if (po->tx_ring.pg_vec) {
4127 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4128 mask |= EPOLLOUT | EPOLLWRNORM;
4129 }
4130 spin_unlock_bh(&sk->sk_write_queue.lock);
4131 return mask;
4132 }
4133
4134
4135 /* Dirty? Well, I still did not learn better way to account
4136 * for user mmaps.
4137 */
4138
4139 static void packet_mm_open(struct vm_area_struct *vma)
4140 {
4141 struct file *file = vma->vm_file;
4142 struct socket *sock = file->private_data;
4143 struct sock *sk = sock->sk;
4144
4145 if (sk)
4146 atomic_inc(&pkt_sk(sk)->mapped);
4147 }
4148
4149 static void packet_mm_close(struct vm_area_struct *vma)
4150 {
4151 struct file *file = vma->vm_file;
4152 struct socket *sock = file->private_data;
4153 struct sock *sk = sock->sk;
4154
4155 if (sk)
4156 atomic_dec(&pkt_sk(sk)->mapped);
4157 }
4158
4159 static const struct vm_operations_struct packet_mmap_ops = {
4160 .open = packet_mm_open,
4161 .close = packet_mm_close,
4162 };
4163
4164 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4165 unsigned int len)
4166 {
4167 int i;
4168
4169 for (i = 0; i < len; i++) {
4170 if (likely(pg_vec[i].buffer)) {
4171 if (is_vmalloc_addr(pg_vec[i].buffer))
4172 vfree(pg_vec[i].buffer);
4173 else
4174 free_pages((unsigned long)pg_vec[i].buffer,
4175 order);
4176 pg_vec[i].buffer = NULL;
4177 }
4178 }
4179 kfree(pg_vec);
4180 }
4181
4182 static char *alloc_one_pg_vec_page(unsigned long order)
4183 {
4184 char *buffer;
4185 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4186 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4187
4188 buffer = (char *) __get_free_pages(gfp_flags, order);
4189 if (buffer)
4190 return buffer;
4191
4192 /* __get_free_pages failed, fall back to vmalloc */
4193 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4194 if (buffer)
4195 return buffer;
4196
4197 /* vmalloc failed, lets dig into swap here */
4198 gfp_flags &= ~__GFP_NORETRY;
4199 buffer = (char *) __get_free_pages(gfp_flags, order);
4200 if (buffer)
4201 return buffer;
4202
4203 /* complete and utter failure */
4204 return NULL;
4205 }
4206
4207 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4208 {
4209 unsigned int block_nr = req->tp_block_nr;
4210 struct pgv *pg_vec;
4211 int i;
4212
4213 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4214 if (unlikely(!pg_vec))
4215 goto out;
4216
4217 for (i = 0; i < block_nr; i++) {
4218 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4219 if (unlikely(!pg_vec[i].buffer))
4220 goto out_free_pgvec;
4221 }
4222
4223 out:
4224 return pg_vec;
4225
4226 out_free_pgvec:
4227 free_pg_vec(pg_vec, order, block_nr);
4228 pg_vec = NULL;
4229 goto out;
4230 }
4231
4232 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4233 int closing, int tx_ring)
4234 {
4235 struct pgv *pg_vec = NULL;
4236 struct packet_sock *po = pkt_sk(sk);
4237 int was_running, order = 0;
4238 struct packet_ring_buffer *rb;
4239 struct sk_buff_head *rb_queue;
4240 __be16 num;
4241 int err = -EINVAL;
4242 /* Added to avoid minimal code churn */
4243 struct tpacket_req *req = &req_u->req;
4244
4245 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4246 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4247
4248 err = -EBUSY;
4249 if (!closing) {
4250 if (atomic_read(&po->mapped))
4251 goto out;
4252 if (packet_read_pending(rb))
4253 goto out;
4254 }
4255
4256 if (req->tp_block_nr) {
4257 unsigned int min_frame_size;
4258
4259 /* Sanity tests and some calculations */
4260 err = -EBUSY;
4261 if (unlikely(rb->pg_vec))
4262 goto out;
4263
4264 switch (po->tp_version) {
4265 case TPACKET_V1:
4266 po->tp_hdrlen = TPACKET_HDRLEN;
4267 break;
4268 case TPACKET_V2:
4269 po->tp_hdrlen = TPACKET2_HDRLEN;
4270 break;
4271 case TPACKET_V3:
4272 po->tp_hdrlen = TPACKET3_HDRLEN;
4273 break;
4274 }
4275
4276 err = -EINVAL;
4277 if (unlikely((int)req->tp_block_size <= 0))
4278 goto out;
4279 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4280 goto out;
4281 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4282 if (po->tp_version >= TPACKET_V3 &&
4283 req->tp_block_size <
4284 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4285 goto out;
4286 if (unlikely(req->tp_frame_size < min_frame_size))
4287 goto out;
4288 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4289 goto out;
4290
4291 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4292 if (unlikely(rb->frames_per_block == 0))
4293 goto out;
4294 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4295 goto out;
4296 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4297 req->tp_frame_nr))
4298 goto out;
4299
4300 err = -ENOMEM;
4301 order = get_order(req->tp_block_size);
4302 pg_vec = alloc_pg_vec(req, order);
4303 if (unlikely(!pg_vec))
4304 goto out;
4305 switch (po->tp_version) {
4306 case TPACKET_V3:
4307 /* Block transmit is not supported yet */
4308 if (!tx_ring) {
4309 init_prb_bdqc(po, rb, pg_vec, req_u);
4310 } else {
4311 struct tpacket_req3 *req3 = &req_u->req3;
4312
4313 if (req3->tp_retire_blk_tov ||
4314 req3->tp_sizeof_priv ||
4315 req3->tp_feature_req_word) {
4316 err = -EINVAL;
4317 goto out;
4318 }
4319 }
4320 break;
4321 default:
4322 break;
4323 }
4324 }
4325 /* Done */
4326 else {
4327 err = -EINVAL;
4328 if (unlikely(req->tp_frame_nr))
4329 goto out;
4330 }
4331
4332
4333 /* Detach socket from network */
4334 spin_lock(&po->bind_lock);
4335 was_running = po->running;
4336 num = po->num;
4337 if (was_running) {
4338 po->num = 0;
4339 __unregister_prot_hook(sk, false);
4340 }
4341 spin_unlock(&po->bind_lock);
4342
4343 synchronize_net();
4344
4345 err = -EBUSY;
4346 mutex_lock(&po->pg_vec_lock);
4347 if (closing || atomic_read(&po->mapped) == 0) {
4348 err = 0;
4349 spin_lock_bh(&rb_queue->lock);
4350 swap(rb->pg_vec, pg_vec);
4351 rb->frame_max = (req->tp_frame_nr - 1);
4352 rb->head = 0;
4353 rb->frame_size = req->tp_frame_size;
4354 spin_unlock_bh(&rb_queue->lock);
4355
4356 swap(rb->pg_vec_order, order);
4357 swap(rb->pg_vec_len, req->tp_block_nr);
4358
4359 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4360 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4361 tpacket_rcv : packet_rcv;
4362 skb_queue_purge(rb_queue);
4363 if (atomic_read(&po->mapped))
4364 pr_err("packet_mmap: vma is busy: %d\n",
4365 atomic_read(&po->mapped));
4366 }
4367 mutex_unlock(&po->pg_vec_lock);
4368
4369 spin_lock(&po->bind_lock);
4370 if (was_running) {
4371 po->num = num;
4372 register_prot_hook(sk);
4373 }
4374 spin_unlock(&po->bind_lock);
4375 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4376 /* Because we don't support block-based V3 on tx-ring */
4377 if (!tx_ring)
4378 prb_shutdown_retire_blk_timer(po, rb_queue);
4379 }
4380
4381 if (pg_vec)
4382 free_pg_vec(pg_vec, order, req->tp_block_nr);
4383 out:
4384 return err;
4385 }
4386
4387 static int packet_mmap(struct file *file, struct socket *sock,
4388 struct vm_area_struct *vma)
4389 {
4390 struct sock *sk = sock->sk;
4391 struct packet_sock *po = pkt_sk(sk);
4392 unsigned long size, expected_size;
4393 struct packet_ring_buffer *rb;
4394 unsigned long start;
4395 int err = -EINVAL;
4396 int i;
4397
4398 if (vma->vm_pgoff)
4399 return -EINVAL;
4400
4401 mutex_lock(&po->pg_vec_lock);
4402
4403 expected_size = 0;
4404 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4405 if (rb->pg_vec) {
4406 expected_size += rb->pg_vec_len
4407 * rb->pg_vec_pages
4408 * PAGE_SIZE;
4409 }
4410 }
4411
4412 if (expected_size == 0)
4413 goto out;
4414
4415 size = vma->vm_end - vma->vm_start;
4416 if (size != expected_size)
4417 goto out;
4418
4419 start = vma->vm_start;
4420 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4421 if (rb->pg_vec == NULL)
4422 continue;
4423
4424 for (i = 0; i < rb->pg_vec_len; i++) {
4425 struct page *page;
4426 void *kaddr = rb->pg_vec[i].buffer;
4427 int pg_num;
4428
4429 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4430 page = pgv_to_page(kaddr);
4431 err = vm_insert_page(vma, start, page);
4432 if (unlikely(err))
4433 goto out;
4434 start += PAGE_SIZE;
4435 kaddr += PAGE_SIZE;
4436 }
4437 }
4438 }
4439
4440 atomic_inc(&po->mapped);
4441 vma->vm_ops = &packet_mmap_ops;
4442 err = 0;
4443
4444 out:
4445 mutex_unlock(&po->pg_vec_lock);
4446 return err;
4447 }
4448
4449 static const struct proto_ops packet_ops_spkt = {
4450 .family = PF_PACKET,
4451 .owner = THIS_MODULE,
4452 .release = packet_release,
4453 .bind = packet_bind_spkt,
4454 .connect = sock_no_connect,
4455 .socketpair = sock_no_socketpair,
4456 .accept = sock_no_accept,
4457 .getname = packet_getname_spkt,
4458 .poll = datagram_poll,
4459 .ioctl = packet_ioctl,
4460 .gettstamp = sock_gettstamp,
4461 .listen = sock_no_listen,
4462 .shutdown = sock_no_shutdown,
4463 .setsockopt = sock_no_setsockopt,
4464 .getsockopt = sock_no_getsockopt,
4465 .sendmsg = packet_sendmsg_spkt,
4466 .recvmsg = packet_recvmsg,
4467 .mmap = sock_no_mmap,
4468 .sendpage = sock_no_sendpage,
4469 };
4470
4471 static const struct proto_ops packet_ops = {
4472 .family = PF_PACKET,
4473 .owner = THIS_MODULE,
4474 .release = packet_release,
4475 .bind = packet_bind,
4476 .connect = sock_no_connect,
4477 .socketpair = sock_no_socketpair,
4478 .accept = sock_no_accept,
4479 .getname = packet_getname,
4480 .poll = packet_poll,
4481 .ioctl = packet_ioctl,
4482 .gettstamp = sock_gettstamp,
4483 .listen = sock_no_listen,
4484 .shutdown = sock_no_shutdown,
4485 .setsockopt = packet_setsockopt,
4486 .getsockopt = packet_getsockopt,
4487 #ifdef CONFIG_COMPAT
4488 .compat_setsockopt = compat_packet_setsockopt,
4489 #endif
4490 .sendmsg = packet_sendmsg,
4491 .recvmsg = packet_recvmsg,
4492 .mmap = packet_mmap,
4493 .sendpage = sock_no_sendpage,
4494 };
4495
4496 static const struct net_proto_family packet_family_ops = {
4497 .family = PF_PACKET,
4498 .create = packet_create,
4499 .owner = THIS_MODULE,
4500 };
4501
4502 static struct notifier_block packet_netdev_notifier = {
4503 .notifier_call = packet_notifier,
4504 };
4505
4506 #ifdef CONFIG_PROC_FS
4507
4508 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4509 __acquires(RCU)
4510 {
4511 struct net *net = seq_file_net(seq);
4512
4513 rcu_read_lock();
4514 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4515 }
4516
4517 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4518 {
4519 struct net *net = seq_file_net(seq);
4520 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4521 }
4522
4523 static void packet_seq_stop(struct seq_file *seq, void *v)
4524 __releases(RCU)
4525 {
4526 rcu_read_unlock();
4527 }
4528
4529 static int packet_seq_show(struct seq_file *seq, void *v)
4530 {
4531 if (v == SEQ_START_TOKEN)
4532 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4533 else {
4534 struct sock *s = sk_entry(v);
4535 const struct packet_sock *po = pkt_sk(s);
4536
4537 seq_printf(seq,
4538 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4539 s,
4540 refcount_read(&s->sk_refcnt),
4541 s->sk_type,
4542 ntohs(po->num),
4543 po->ifindex,
4544 po->running,
4545 atomic_read(&s->sk_rmem_alloc),
4546 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4547 sock_i_ino(s));
4548 }
4549
4550 return 0;
4551 }
4552
4553 static const struct seq_operations packet_seq_ops = {
4554 .start = packet_seq_start,
4555 .next = packet_seq_next,
4556 .stop = packet_seq_stop,
4557 .show = packet_seq_show,
4558 };
4559 #endif
4560
4561 static int __net_init packet_net_init(struct net *net)
4562 {
4563 mutex_init(&net->packet.sklist_lock);
4564 INIT_HLIST_HEAD(&net->packet.sklist);
4565
4566 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4567 sizeof(struct seq_net_private)))
4568 return -ENOMEM;
4569
4570 return 0;
4571 }
4572
4573 static void __net_exit packet_net_exit(struct net *net)
4574 {
4575 remove_proc_entry("packet", net->proc_net);
4576 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4577 }
4578
4579 static struct pernet_operations packet_net_ops = {
4580 .init = packet_net_init,
4581 .exit = packet_net_exit,
4582 };
4583
4584
4585 static void __exit packet_exit(void)
4586 {
4587 unregister_netdevice_notifier(&packet_netdev_notifier);
4588 unregister_pernet_subsys(&packet_net_ops);
4589 sock_unregister(PF_PACKET);
4590 proto_unregister(&packet_proto);
4591 }
4592
4593 static int __init packet_init(void)
4594 {
4595 int rc;
4596
4597 rc = proto_register(&packet_proto, 0);
4598 if (rc)
4599 goto out;
4600 rc = sock_register(&packet_family_ops);
4601 if (rc)
4602 goto out_proto;
4603 rc = register_pernet_subsys(&packet_net_ops);
4604 if (rc)
4605 goto out_sock;
4606 rc = register_netdevice_notifier(&packet_netdev_notifier);
4607 if (rc)
4608 goto out_pernet;
4609
4610 return 0;
4611
4612 out_pernet:
4613 unregister_pernet_subsys(&packet_net_ops);
4614 out_sock:
4615 sock_unregister(PF_PACKET);
4616 out_proto:
4617 proto_unregister(&packet_proto);
4618 out:
4619 return rc;
4620 }
4621
4622 module_init(packet_init);
4623 module_exit(packet_exit);
4624 MODULE_LICENSE("GPL");
4625 MODULE_ALIAS_NETPROTO(PF_PACKET);
Linux with added FPC-III support