]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - net/packet/af_packet.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
af_packet: Prefixed tpacket_v3 structs to avoid name space collision
[mirror_ubuntu-jammy-kernel.git] / net / packet / af_packet.c
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PACKET - implements raw packet sockets.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 *
12 * Fixes:
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
41 * and packet_mreq.
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
44 * layer.
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
46 *
47 *
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
52 *
53 */
54
55 #include <linux/types.h>
56 #include <linux/mm.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
60 #include <linux/in.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
70 #include <net/ip.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/system.h>
77 #include <asm/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <asm/page.h>
80 #include <asm/cacheflush.h>
81 #include <asm/io.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84 #include <linux/poll.h>
85 #include <linux/module.h>
86 #include <linux/init.h>
87 #include <linux/mutex.h>
88 #include <linux/if_vlan.h>
89 #include <linux/virtio_net.h>
90 #include <linux/errqueue.h>
91 #include <linux/net_tstamp.h>
92
93 #ifdef CONFIG_INET
94 #include <net/inet_common.h>
95 #endif
96
97 /*
98 Assumptions:
99 - if device has no dev->hard_header routine, it adds and removes ll header
100 inside itself. In this case ll header is invisible outside of device,
101 but higher levels still should reserve dev->hard_header_len.
102 Some devices are enough clever to reallocate skb, when header
103 will not fit to reserved space (tunnel), another ones are silly
104 (PPP).
105 - packet socket receives packets with pulled ll header,
106 so that SOCK_RAW should push it back.
107
108 On receive:
109 -----------
110
111 Incoming, dev->hard_header!=NULL
112 mac_header -> ll header
113 data -> data
114
115 Outgoing, dev->hard_header!=NULL
116 mac_header -> ll header
117 data -> ll header
118
119 Incoming, dev->hard_header==NULL
120 mac_header -> UNKNOWN position. It is very likely, that it points to ll
121 header. PPP makes it, that is wrong, because introduce
122 assymetry between rx and tx paths.
123 data -> data
124
125 Outgoing, dev->hard_header==NULL
126 mac_header -> data. ll header is still not built!
127 data -> data
128
129 Resume
130 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
131
132
133 On transmit:
134 ------------
135
136 dev->hard_header != NULL
137 mac_header -> ll header
138 data -> ll header
139
140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
141 mac_header -> data
142 data -> data
143
144 We should set nh.raw on output to correct posistion,
145 packet classifier depends on it.
146 */
147
148 /* Private packet socket structures. */
149
150 struct packet_mclist {
151 struct packet_mclist *next;
152 int ifindex;
153 int count;
154 unsigned short type;
155 unsigned short alen;
156 unsigned char addr[MAX_ADDR_LEN];
157 };
158 /* identical to struct packet_mreq except it has
159 * a longer address field.
160 */
161 struct packet_mreq_max {
162 int mr_ifindex;
163 unsigned short mr_type;
164 unsigned short mr_alen;
165 unsigned char mr_address[MAX_ADDR_LEN];
166 };
167
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
170
171
172 #define V3_ALIGNMENT (8)
173
174 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178
179 /* kbdq - kernel block descriptor queue */
180 struct tpacket_kbdq_core {
181 struct pgv *pkbdq;
182 unsigned int feature_req_word;
183 unsigned int hdrlen;
184 unsigned char reset_pending_on_curr_blk;
185 unsigned char delete_blk_timer;
186 unsigned short kactive_blk_num;
187 unsigned short blk_sizeof_priv;
188
189 /* last_kactive_blk_num:
190 * trick to see if user-space has caught up
191 * in order to avoid refreshing timer when every single pkt arrives.
192 */
193 unsigned short last_kactive_blk_num;
194
195 char *pkblk_start;
196 char *pkblk_end;
197 int kblk_size;
198 unsigned int knum_blocks;
199 uint64_t knxt_seq_num;
200 char *prev;
201 char *nxt_offset;
202 struct sk_buff *skb;
203
204 atomic_t blk_fill_in_prog;
205
206 /* Default is set to 8ms */
207 #define DEFAULT_PRB_RETIRE_TOV (8)
208
209 unsigned short retire_blk_tov;
210 unsigned short version;
211 unsigned long tov_in_jiffies;
212
213 /* timer to retire an outstanding block */
214 struct timer_list retire_blk_timer;
215 };
216
217 #define PGV_FROM_VMALLOC 1
218 struct pgv {
219 char *buffer;
220 };
221
222 struct packet_ring_buffer {
223 struct pgv *pg_vec;
224 unsigned int head;
225 unsigned int frames_per_block;
226 unsigned int frame_size;
227 unsigned int frame_max;
228
229 unsigned int pg_vec_order;
230 unsigned int pg_vec_pages;
231 unsigned int pg_vec_len;
232
233 struct tpacket_kbdq_core prb_bdqc;
234 atomic_t pending;
235 };
236
237 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
238 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
239 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
240 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
241 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
242 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
243 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
244
245 struct packet_sock;
246 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
247
248 static void *packet_previous_frame(struct packet_sock *po,
249 struct packet_ring_buffer *rb,
250 int status);
251 static void packet_increment_head(struct packet_ring_buffer *buff);
252 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
253 struct tpacket_block_desc *);
254 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
255 struct packet_sock *);
256 static void prb_retire_current_block(struct tpacket_kbdq_core *,
257 struct packet_sock *, unsigned int status);
258 static int prb_queue_frozen(struct tpacket_kbdq_core *);
259 static void prb_open_block(struct tpacket_kbdq_core *,
260 struct tpacket_block_desc *);
261 static void prb_retire_rx_blk_timer_expired(unsigned long);
262 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
263 static void prb_init_blk_timer(struct packet_sock *,
264 struct tpacket_kbdq_core *,
265 void (*func) (unsigned long));
266 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
267 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
268 struct tpacket3_hdr *);
269 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
270 struct tpacket3_hdr *);
271 static void packet_flush_mclist(struct sock *sk);
272
273 struct packet_fanout;
274 struct packet_sock {
275 /* struct sock has to be the first member of packet_sock */
276 struct sock sk;
277 struct packet_fanout *fanout;
278 struct tpacket_stats stats;
279 union tpacket_stats_u stats_u;
280 struct packet_ring_buffer rx_ring;
281 struct packet_ring_buffer tx_ring;
282 int copy_thresh;
283 spinlock_t bind_lock;
284 struct mutex pg_vec_lock;
285 unsigned int running:1, /* prot_hook is attached*/
286 auxdata:1,
287 origdev:1,
288 has_vnet_hdr:1;
289 int ifindex; /* bound device */
290 __be16 num;
291 struct packet_mclist *mclist;
292 atomic_t mapped;
293 enum tpacket_versions tp_version;
294 unsigned int tp_hdrlen;
295 unsigned int tp_reserve;
296 unsigned int tp_loss:1;
297 unsigned int tp_tstamp;
298 struct packet_type prot_hook ____cacheline_aligned_in_smp;
299 };
300
301 #define PACKET_FANOUT_MAX 256
302
303 struct packet_fanout {
304 #ifdef CONFIG_NET_NS
305 struct net *net;
306 #endif
307 unsigned int num_members;
308 u16 id;
309 u8 type;
310 u8 defrag;
311 atomic_t rr_cur;
312 struct list_head list;
313 struct sock *arr[PACKET_FANOUT_MAX];
314 spinlock_t lock;
315 atomic_t sk_ref;
316 struct packet_type prot_hook ____cacheline_aligned_in_smp;
317 };
318
319 struct packet_skb_cb {
320 unsigned int origlen;
321 union {
322 struct sockaddr_pkt pkt;
323 struct sockaddr_ll ll;
324 } sa;
325 };
326
327 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
328
329 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
330 #define GET_PBLOCK_DESC(x, bid) \
331 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
332 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
333 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
334 #define GET_NEXT_PRB_BLK_NUM(x) \
335 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
336 ((x)->kactive_blk_num+1) : 0)
337
338 static inline struct packet_sock *pkt_sk(struct sock *sk)
339 {
340 return (struct packet_sock *)sk;
341 }
342
343 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
344 static void __fanout_link(struct sock *sk, struct packet_sock *po);
345
346 /* register_prot_hook must be invoked with the po->bind_lock held,
347 * or from a context in which asynchronous accesses to the packet
348 * socket is not possible (packet_create()).
349 */
350 static void register_prot_hook(struct sock *sk)
351 {
352 struct packet_sock *po = pkt_sk(sk);
353 if (!po->running) {
354 if (po->fanout)
355 __fanout_link(sk, po);
356 else
357 dev_add_pack(&po->prot_hook);
358 sock_hold(sk);
359 po->running = 1;
360 }
361 }
362
363 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
364 * held. If the sync parameter is true, we will temporarily drop
365 * the po->bind_lock and do a synchronize_net to make sure no
366 * asynchronous packet processing paths still refer to the elements
367 * of po->prot_hook. If the sync parameter is false, it is the
368 * callers responsibility to take care of this.
369 */
370 static void __unregister_prot_hook(struct sock *sk, bool sync)
371 {
372 struct packet_sock *po = pkt_sk(sk);
373
374 po->running = 0;
375 if (po->fanout)
376 __fanout_unlink(sk, po);
377 else
378 __dev_remove_pack(&po->prot_hook);
379 __sock_put(sk);
380
381 if (sync) {
382 spin_unlock(&po->bind_lock);
383 synchronize_net();
384 spin_lock(&po->bind_lock);
385 }
386 }
387
388 static void unregister_prot_hook(struct sock *sk, bool sync)
389 {
390 struct packet_sock *po = pkt_sk(sk);
391
392 if (po->running)
393 __unregister_prot_hook(sk, sync);
394 }
395
396 static inline __pure struct page *pgv_to_page(void *addr)
397 {
398 if (is_vmalloc_addr(addr))
399 return vmalloc_to_page(addr);
400 return virt_to_page(addr);
401 }
402
403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
404 {
405 union {
406 struct tpacket_hdr *h1;
407 struct tpacket2_hdr *h2;
408 void *raw;
409 } h;
410
411 h.raw = frame;
412 switch (po->tp_version) {
413 case TPACKET_V1:
414 h.h1->tp_status = status;
415 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
416 break;
417 case TPACKET_V2:
418 h.h2->tp_status = status;
419 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
420 break;
421 case TPACKET_V3:
422 default:
423 WARN(1, "TPACKET version not supported.\n");
424 BUG();
425 }
426
427 smp_wmb();
428 }
429
430 static int __packet_get_status(struct packet_sock *po, void *frame)
431 {
432 union {
433 struct tpacket_hdr *h1;
434 struct tpacket2_hdr *h2;
435 void *raw;
436 } h;
437
438 smp_rmb();
439
440 h.raw = frame;
441 switch (po->tp_version) {
442 case TPACKET_V1:
443 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
444 return h.h1->tp_status;
445 case TPACKET_V2:
446 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
447 return h.h2->tp_status;
448 case TPACKET_V3:
449 default:
450 WARN(1, "TPACKET version not supported.\n");
451 BUG();
452 return 0;
453 }
454 }
455
456 static void *packet_lookup_frame(struct packet_sock *po,
457 struct packet_ring_buffer *rb,
458 unsigned int position,
459 int status)
460 {
461 unsigned int pg_vec_pos, frame_offset;
462 union {
463 struct tpacket_hdr *h1;
464 struct tpacket2_hdr *h2;
465 void *raw;
466 } h;
467
468 pg_vec_pos = position / rb->frames_per_block;
469 frame_offset = position % rb->frames_per_block;
470
471 h.raw = rb->pg_vec[pg_vec_pos].buffer +
472 (frame_offset * rb->frame_size);
473
474 if (status != __packet_get_status(po, h.raw))
475 return NULL;
476
477 return h.raw;
478 }
479
480 static inline void *packet_current_frame(struct packet_sock *po,
481 struct packet_ring_buffer *rb,
482 int status)
483 {
484 return packet_lookup_frame(po, rb, rb->head, status);
485 }
486
487 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
488 {
489 del_timer_sync(&pkc->retire_blk_timer);
490 }
491
492 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
493 int tx_ring,
494 struct sk_buff_head *rb_queue)
495 {
496 struct tpacket_kbdq_core *pkc;
497
498 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
499
500 spin_lock(&rb_queue->lock);
501 pkc->delete_blk_timer = 1;
502 spin_unlock(&rb_queue->lock);
503
504 prb_del_retire_blk_timer(pkc);
505 }
506
507 static void prb_init_blk_timer(struct packet_sock *po,
508 struct tpacket_kbdq_core *pkc,
509 void (*func) (unsigned long))
510 {
511 init_timer(&pkc->retire_blk_timer);
512 pkc->retire_blk_timer.data = (long)po;
513 pkc->retire_blk_timer.function = func;
514 pkc->retire_blk_timer.expires = jiffies;
515 }
516
517 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
518 {
519 struct tpacket_kbdq_core *pkc;
520
521 if (tx_ring)
522 BUG();
523
524 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
525 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
526 }
527
528 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
529 int blk_size_in_bytes)
530 {
531 struct net_device *dev;
532 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
533
534 dev = dev_get_by_index(sock_net(&po->sk), po->ifindex);
535 if (unlikely(dev == NULL))
536 return DEFAULT_PRB_RETIRE_TOV;
537
538 if (dev->ethtool_ops && dev->ethtool_ops->get_settings) {
539 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET, };
540
541 if (!dev->ethtool_ops->get_settings(dev, &ecmd)) {
542 switch (ecmd.speed) {
543 case SPEED_10000:
544 msec = 1;
545 div = 10000/1000;
546 break;
547 case SPEED_1000:
548 msec = 1;
549 div = 1000/1000;
550 break;
551 /*
552 * If the link speed is so slow you don't really
553 * need to worry about perf anyways
554 */
555 case SPEED_100:
556 case SPEED_10:
557 default:
558 return DEFAULT_PRB_RETIRE_TOV;
559 }
560 }
561 }
562
563 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
564
565 if (div)
566 mbits /= div;
567
568 tmo = mbits * msec;
569
570 if (div)
571 return tmo+1;
572 return tmo;
573 }
574
575 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
576 union tpacket_req_u *req_u)
577 {
578 p1->feature_req_word = req_u->req3.tp_feature_req_word;
579 }
580
581 static void init_prb_bdqc(struct packet_sock *po,
582 struct packet_ring_buffer *rb,
583 struct pgv *pg_vec,
584 union tpacket_req_u *req_u, int tx_ring)
585 {
586 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
587 struct tpacket_block_desc *pbd;
588
589 memset(p1, 0x0, sizeof(*p1));
590
591 p1->knxt_seq_num = 1;
592 p1->pkbdq = pg_vec;
593 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
594 p1->pkblk_start = (char *)pg_vec[0].buffer;
595 p1->kblk_size = req_u->req3.tp_block_size;
596 p1->knum_blocks = req_u->req3.tp_block_nr;
597 p1->hdrlen = po->tp_hdrlen;
598 p1->version = po->tp_version;
599 p1->last_kactive_blk_num = 0;
600 po->stats_u.stats3.tp_freeze_q_cnt = 0;
601 if (req_u->req3.tp_retire_blk_tov)
602 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
603 else
604 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
605 req_u->req3.tp_block_size);
606 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
607 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
608
609 prb_init_ft_ops(p1, req_u);
610 prb_setup_retire_blk_timer(po, tx_ring);
611 prb_open_block(p1, pbd);
612 }
613
614 /* Do NOT update the last_blk_num first.
615 * Assumes sk_buff_head lock is held.
616 */
617 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
618 {
619 mod_timer(&pkc->retire_blk_timer,
620 jiffies + pkc->tov_in_jiffies);
621 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
622 }
623
624 /*
625 * Timer logic:
626 * 1) We refresh the timer only when we open a block.
627 * By doing this we don't waste cycles refreshing the timer
628 * on packet-by-packet basis.
629 *
630 * With a 1MB block-size, on a 1Gbps line, it will take
631 * i) ~8 ms to fill a block + ii) memcpy etc.
632 * In this cut we are not accounting for the memcpy time.
633 *
634 * So, if the user sets the 'tmo' to 10ms then the timer
635 * will never fire while the block is still getting filled
636 * (which is what we want). However, the user could choose
637 * to close a block early and that's fine.
638 *
639 * But when the timer does fire, we check whether or not to refresh it.
640 * Since the tmo granularity is in msecs, it is not too expensive
641 * to refresh the timer, lets say every '8' msecs.
642 * Either the user can set the 'tmo' or we can derive it based on
643 * a) line-speed and b) block-size.
644 * prb_calc_retire_blk_tmo() calculates the tmo.
645 *
646 */
647 static void prb_retire_rx_blk_timer_expired(unsigned long data)
648 {
649 struct packet_sock *po = (struct packet_sock *)data;
650 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
651 unsigned int frozen;
652 struct tpacket_block_desc *pbd;
653
654 spin_lock(&po->sk.sk_receive_queue.lock);
655
656 frozen = prb_queue_frozen(pkc);
657 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
658
659 if (unlikely(pkc->delete_blk_timer))
660 goto out;
661
662 /* We only need to plug the race when the block is partially filled.
663 * tpacket_rcv:
664 * lock(); increment BLOCK_NUM_PKTS; unlock()
665 * copy_bits() is in progress ...
666 * timer fires on other cpu:
667 * we can't retire the current block because copy_bits
668 * is in progress.
669 *
670 */
671 if (BLOCK_NUM_PKTS(pbd)) {
672 while (atomic_read(&pkc->blk_fill_in_prog)) {
673 /* Waiting for skb_copy_bits to finish... */
674 cpu_relax();
675 }
676 }
677
678 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
679 if (!frozen) {
680 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
681 if (!prb_dispatch_next_block(pkc, po))
682 goto refresh_timer;
683 else
684 goto out;
685 } else {
686 /* Case 1. Queue was frozen because user-space was
687 * lagging behind.
688 */
689 if (prb_curr_blk_in_use(pkc, pbd)) {
690 /*
691 * Ok, user-space is still behind.
692 * So just refresh the timer.
693 */
694 goto refresh_timer;
695 } else {
696 /* Case 2. queue was frozen,user-space caught up,
697 * now the link went idle && the timer fired.
698 * We don't have a block to close.So we open this
699 * block and restart the timer.
700 * opening a block thaws the queue,restarts timer
701 * Thawing/timer-refresh is a side effect.
702 */
703 prb_open_block(pkc, pbd);
704 goto out;
705 }
706 }
707 }
708
709 refresh_timer:
710 _prb_refresh_rx_retire_blk_timer(pkc);
711
712 out:
713 spin_unlock(&po->sk.sk_receive_queue.lock);
714 }
715
716 static inline void prb_flush_block(struct tpacket_kbdq_core *pkc1,
717 struct tpacket_block_desc *pbd1, __u32 status)
718 {
719 /* Flush everything minus the block header */
720
721 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
722 u8 *start, *end;
723
724 start = (u8 *)pbd1;
725
726 /* Skip the block header(we know header WILL fit in 4K) */
727 start += PAGE_SIZE;
728
729 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
730 for (; start < end; start += PAGE_SIZE)
731 flush_dcache_page(pgv_to_page(start));
732
733 smp_wmb();
734 #endif
735
736 /* Now update the block status. */
737
738 BLOCK_STATUS(pbd1) = status;
739
740 /* Flush the block header */
741
742 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
743 start = (u8 *)pbd1;
744 flush_dcache_page(pgv_to_page(start));
745
746 smp_wmb();
747 #endif
748 }
749
750 /*
751 * Side effect:
752 *
753 * 1) flush the block
754 * 2) Increment active_blk_num
755 *
756 * Note:We DONT refresh the timer on purpose.
757 * Because almost always the next block will be opened.
758 */
759 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
760 struct tpacket_block_desc *pbd1,
761 struct packet_sock *po, unsigned int stat)
762 {
763 __u32 status = TP_STATUS_USER | stat;
764
765 struct tpacket3_hdr *last_pkt;
766 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
767
768 if (po->stats.tp_drops)
769 status |= TP_STATUS_LOSING;
770
771 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
772 last_pkt->tp_next_offset = 0;
773
774 /* Get the ts of the last pkt */
775 if (BLOCK_NUM_PKTS(pbd1)) {
776 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
777 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
778 } else {
779 /* Ok, we tmo'd - so get the current time */
780 struct timespec ts;
781 getnstimeofday(&ts);
782 h1->ts_last_pkt.ts_sec = ts.tv_sec;
783 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
784 }
785
786 smp_wmb();
787
788 /* Flush the block */
789 prb_flush_block(pkc1, pbd1, status);
790
791 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
792 }
793
794 static inline void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
795 {
796 pkc->reset_pending_on_curr_blk = 0;
797 }
798
799 /*
800 * Side effect of opening a block:
801 *
802 * 1) prb_queue is thawed.
803 * 2) retire_blk_timer is refreshed.
804 *
805 */
806 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
807 struct tpacket_block_desc *pbd1)
808 {
809 struct timespec ts;
810 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
811
812 smp_rmb();
813
814 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
815
816 /* We could have just memset this but we will lose the
817 * flexibility of making the priv area sticky
818 */
819 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
820 BLOCK_NUM_PKTS(pbd1) = 0;
821 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
822 getnstimeofday(&ts);
823 h1->ts_first_pkt.ts_sec = ts.tv_sec;
824 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
825 pkc1->pkblk_start = (char *)pbd1;
826 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
827 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
828 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
829 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
830 pbd1->version = pkc1->version;
831 pkc1->prev = pkc1->nxt_offset;
832 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
833 prb_thaw_queue(pkc1);
834 _prb_refresh_rx_retire_blk_timer(pkc1);
835
836 smp_wmb();
837
838 return;
839 }
840
841 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
842 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
843 dump_stack();
844 BUG();
845 }
846
847 /*
848 * Queue freeze logic:
849 * 1) Assume tp_block_nr = 8 blocks.
850 * 2) At time 't0', user opens Rx ring.
851 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
852 * 4) user-space is either sleeping or processing block '0'.
853 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
854 * it will close block-7,loop around and try to fill block '0'.
855 * call-flow:
856 * __packet_lookup_frame_in_block
857 * prb_retire_current_block()
858 * prb_dispatch_next_block()
859 * |->(BLOCK_STATUS == USER) evaluates to true
860 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
861 * 6) Now there are two cases:
862 * 6.1) Link goes idle right after the queue is frozen.
863 * But remember, the last open_block() refreshed the timer.
864 * When this timer expires,it will refresh itself so that we can
865 * re-open block-0 in near future.
866 * 6.2) Link is busy and keeps on receiving packets. This is a simple
867 * case and __packet_lookup_frame_in_block will check if block-0
868 * is free and can now be re-used.
869 */
870 static inline void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
871 struct packet_sock *po)
872 {
873 pkc->reset_pending_on_curr_blk = 1;
874 po->stats_u.stats3.tp_freeze_q_cnt++;
875 }
876
877 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
878
879 /*
880 * If the next block is free then we will dispatch it
881 * and return a good offset.
882 * Else, we will freeze the queue.
883 * So, caller must check the return value.
884 */
885 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
886 struct packet_sock *po)
887 {
888 struct tpacket_block_desc *pbd;
889
890 smp_rmb();
891
892 /* 1. Get current block num */
893 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
894
895 /* 2. If this block is currently in_use then freeze the queue */
896 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
897 prb_freeze_queue(pkc, po);
898 return NULL;
899 }
900
901 /*
902 * 3.
903 * open this block and return the offset where the first packet
904 * needs to get stored.
905 */
906 prb_open_block(pkc, pbd);
907 return (void *)pkc->nxt_offset;
908 }
909
910 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
911 struct packet_sock *po, unsigned int status)
912 {
913 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
914
915 /* retire/close the current block */
916 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
917 /*
918 * Plug the case where copy_bits() is in progress on
919 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
920 * have space to copy the pkt in the current block and
921 * called prb_retire_current_block()
922 *
923 * We don't need to worry about the TMO case because
924 * the timer-handler already handled this case.
925 */
926 if (!(status & TP_STATUS_BLK_TMO)) {
927 while (atomic_read(&pkc->blk_fill_in_prog)) {
928 /* Waiting for skb_copy_bits to finish... */
929 cpu_relax();
930 }
931 }
932 prb_close_block(pkc, pbd, po, status);
933 return;
934 }
935
936 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
937 dump_stack();
938 BUG();
939 }
940
941 static inline int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
942 struct tpacket_block_desc *pbd)
943 {
944 return TP_STATUS_USER & BLOCK_STATUS(pbd);
945 }
946
947 static inline int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
948 {
949 return pkc->reset_pending_on_curr_blk;
950 }
951
952 static inline void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
953 {
954 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
955 atomic_dec(&pkc->blk_fill_in_prog);
956 }
957
958 static inline void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
959 struct tpacket3_hdr *ppd)
960 {
961 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
962 }
963
964 static inline void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
965 struct tpacket3_hdr *ppd)
966 {
967 ppd->hv1.tp_rxhash = 0;
968 }
969
970 static inline void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
971 struct tpacket3_hdr *ppd)
972 {
973 if (vlan_tx_tag_present(pkc->skb)) {
974 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
975 ppd->tp_status = TP_STATUS_VLAN_VALID;
976 } else {
977 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
978 }
979 }
980
981 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
982 struct tpacket3_hdr *ppd)
983 {
984 prb_fill_vlan_info(pkc, ppd);
985
986 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
987 prb_fill_rxhash(pkc, ppd);
988 else
989 prb_clear_rxhash(pkc, ppd);
990 }
991
992 static inline void prb_fill_curr_block(char *curr,
993 struct tpacket_kbdq_core *pkc,
994 struct tpacket_block_desc *pbd,
995 unsigned int len)
996 {
997 struct tpacket3_hdr *ppd;
998
999 ppd = (struct tpacket3_hdr *)curr;
1000 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1001 pkc->prev = curr;
1002 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1003 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1004 BLOCK_NUM_PKTS(pbd) += 1;
1005 atomic_inc(&pkc->blk_fill_in_prog);
1006 prb_run_all_ft_ops(pkc, ppd);
1007 }
1008
1009 /* Assumes caller has the sk->rx_queue.lock */
1010 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1011 struct sk_buff *skb,
1012 int status,
1013 unsigned int len
1014 )
1015 {
1016 struct tpacket_kbdq_core *pkc;
1017 struct tpacket_block_desc *pbd;
1018 char *curr, *end;
1019
1020 pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1021 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1022
1023 /* Queue is frozen when user space is lagging behind */
1024 if (prb_queue_frozen(pkc)) {
1025 /*
1026 * Check if that last block which caused the queue to freeze,
1027 * is still in_use by user-space.
1028 */
1029 if (prb_curr_blk_in_use(pkc, pbd)) {
1030 /* Can't record this packet */
1031 return NULL;
1032 } else {
1033 /*
1034 * Ok, the block was released by user-space.
1035 * Now let's open that block.
1036 * opening a block also thaws the queue.
1037 * Thawing is a side effect.
1038 */
1039 prb_open_block(pkc, pbd);
1040 }
1041 }
1042
1043 smp_mb();
1044 curr = pkc->nxt_offset;
1045 pkc->skb = skb;
1046 end = (char *) ((char *)pbd + pkc->kblk_size);
1047
1048 /* first try the current block */
1049 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1050 prb_fill_curr_block(curr, pkc, pbd, len);
1051 return (void *)curr;
1052 }
1053
1054 /* Ok, close the current block */
1055 prb_retire_current_block(pkc, po, 0);
1056
1057 /* Now, try to dispatch the next block */
1058 curr = (char *)prb_dispatch_next_block(pkc, po);
1059 if (curr) {
1060 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1061 prb_fill_curr_block(curr, pkc, pbd, len);
1062 return (void *)curr;
1063 }
1064
1065 /*
1066 * No free blocks are available.user_space hasn't caught up yet.
1067 * Queue was just frozen and now this packet will get dropped.
1068 */
1069 return NULL;
1070 }
1071
1072 static inline void *packet_current_rx_frame(struct packet_sock *po,
1073 struct sk_buff *skb,
1074 int status, unsigned int len)
1075 {
1076 char *curr = NULL;
1077 switch (po->tp_version) {
1078 case TPACKET_V1:
1079 case TPACKET_V2:
1080 curr = packet_lookup_frame(po, &po->rx_ring,
1081 po->rx_ring.head, status);
1082 return curr;
1083 case TPACKET_V3:
1084 return __packet_lookup_frame_in_block(po, skb, status, len);
1085 default:
1086 WARN(1, "TPACKET version not supported\n");
1087 BUG();
1088 return 0;
1089 }
1090 }
1091
1092 static inline void *prb_lookup_block(struct packet_sock *po,
1093 struct packet_ring_buffer *rb,
1094 unsigned int previous,
1095 int status)
1096 {
1097 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1098 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1099
1100 if (status != BLOCK_STATUS(pbd))
1101 return NULL;
1102 return pbd;
1103 }
1104
1105 static inline int prb_previous_blk_num(struct packet_ring_buffer *rb)
1106 {
1107 unsigned int prev;
1108 if (rb->prb_bdqc.kactive_blk_num)
1109 prev = rb->prb_bdqc.kactive_blk_num-1;
1110 else
1111 prev = rb->prb_bdqc.knum_blocks-1;
1112 return prev;
1113 }
1114
1115 /* Assumes caller has held the rx_queue.lock */
1116 static inline void *__prb_previous_block(struct packet_sock *po,
1117 struct packet_ring_buffer *rb,
1118 int status)
1119 {
1120 unsigned int previous = prb_previous_blk_num(rb);
1121 return prb_lookup_block(po, rb, previous, status);
1122 }
1123
1124 static inline void *packet_previous_rx_frame(struct packet_sock *po,
1125 struct packet_ring_buffer *rb,
1126 int status)
1127 {
1128 if (po->tp_version <= TPACKET_V2)
1129 return packet_previous_frame(po, rb, status);
1130
1131 return __prb_previous_block(po, rb, status);
1132 }
1133
1134 static inline void packet_increment_rx_head(struct packet_sock *po,
1135 struct packet_ring_buffer *rb)
1136 {
1137 switch (po->tp_version) {
1138 case TPACKET_V1:
1139 case TPACKET_V2:
1140 return packet_increment_head(rb);
1141 case TPACKET_V3:
1142 default:
1143 WARN(1, "TPACKET version not supported.\n");
1144 BUG();
1145 return;
1146 }
1147 }
1148
1149 static inline void *packet_previous_frame(struct packet_sock *po,
1150 struct packet_ring_buffer *rb,
1151 int status)
1152 {
1153 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1154 return packet_lookup_frame(po, rb, previous, status);
1155 }
1156
1157 static inline void packet_increment_head(struct packet_ring_buffer *buff)
1158 {
1159 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1160 }
1161
1162 static void packet_sock_destruct(struct sock *sk)
1163 {
1164 skb_queue_purge(&sk->sk_error_queue);
1165
1166 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1167 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1168
1169 if (!sock_flag(sk, SOCK_DEAD)) {
1170 pr_err("Attempt to release alive packet socket: %p\n", sk);
1171 return;
1172 }
1173
1174 sk_refcnt_debug_dec(sk);
1175 }
1176
1177 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1178 {
1179 int x = atomic_read(&f->rr_cur) + 1;
1180
1181 if (x >= num)
1182 x = 0;
1183
1184 return x;
1185 }
1186
1187 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1188 {
1189 u32 idx, hash = skb->rxhash;
1190
1191 idx = ((u64)hash * num) >> 32;
1192
1193 return f->arr[idx];
1194 }
1195
1196 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1197 {
1198 int cur, old;
1199
1200 cur = atomic_read(&f->rr_cur);
1201 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1202 fanout_rr_next(f, num))) != cur)
1203 cur = old;
1204 return f->arr[cur];
1205 }
1206
1207 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1208 {
1209 unsigned int cpu = smp_processor_id();
1210
1211 return f->arr[cpu % num];
1212 }
1213
1214 static struct sk_buff *fanout_check_defrag(struct sk_buff *skb)
1215 {
1216 #ifdef CONFIG_INET
1217 const struct iphdr *iph;
1218 u32 len;
1219
1220 if (skb->protocol != htons(ETH_P_IP))
1221 return skb;
1222
1223 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1224 return skb;
1225
1226 iph = ip_hdr(skb);
1227 if (iph->ihl < 5 || iph->version != 4)
1228 return skb;
1229 if (!pskb_may_pull(skb, iph->ihl*4))
1230 return skb;
1231 iph = ip_hdr(skb);
1232 len = ntohs(iph->tot_len);
1233 if (skb->len < len || len < (iph->ihl * 4))
1234 return skb;
1235
1236 if (ip_is_fragment(ip_hdr(skb))) {
1237 skb = skb_share_check(skb, GFP_ATOMIC);
1238 if (skb) {
1239 if (pskb_trim_rcsum(skb, len))
1240 return skb;
1241 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
1242 if (ip_defrag(skb, IP_DEFRAG_AF_PACKET))
1243 return NULL;
1244 skb->rxhash = 0;
1245 }
1246 }
1247 #endif
1248 return skb;
1249 }
1250
1251 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1252 struct packet_type *pt, struct net_device *orig_dev)
1253 {
1254 struct packet_fanout *f = pt->af_packet_priv;
1255 unsigned int num = f->num_members;
1256 struct packet_sock *po;
1257 struct sock *sk;
1258
1259 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1260 !num) {
1261 kfree_skb(skb);
1262 return 0;
1263 }
1264
1265 switch (f->type) {
1266 case PACKET_FANOUT_HASH:
1267 default:
1268 if (f->defrag) {
1269 skb = fanout_check_defrag(skb);
1270 if (!skb)
1271 return 0;
1272 }
1273 skb_get_rxhash(skb);
1274 sk = fanout_demux_hash(f, skb, num);
1275 break;
1276 case PACKET_FANOUT_LB:
1277 sk = fanout_demux_lb(f, skb, num);
1278 break;
1279 case PACKET_FANOUT_CPU:
1280 sk = fanout_demux_cpu(f, skb, num);
1281 break;
1282 }
1283
1284 po = pkt_sk(sk);
1285
1286 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1287 }
1288
1289 static DEFINE_MUTEX(fanout_mutex);
1290 static LIST_HEAD(fanout_list);
1291
1292 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1293 {
1294 struct packet_fanout *f = po->fanout;
1295
1296 spin_lock(&f->lock);
1297 f->arr[f->num_members] = sk;
1298 smp_wmb();
1299 f->num_members++;
1300 spin_unlock(&f->lock);
1301 }
1302
1303 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1304 {
1305 struct packet_fanout *f = po->fanout;
1306 int i;
1307
1308 spin_lock(&f->lock);
1309 for (i = 0; i < f->num_members; i++) {
1310 if (f->arr[i] == sk)
1311 break;
1312 }
1313 BUG_ON(i >= f->num_members);
1314 f->arr[i] = f->arr[f->num_members - 1];
1315 f->num_members--;
1316 spin_unlock(&f->lock);
1317 }
1318
1319 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1320 {
1321 struct packet_sock *po = pkt_sk(sk);
1322 struct packet_fanout *f, *match;
1323 u8 type = type_flags & 0xff;
1324 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1325 int err;
1326
1327 switch (type) {
1328 case PACKET_FANOUT_HASH:
1329 case PACKET_FANOUT_LB:
1330 case PACKET_FANOUT_CPU:
1331 break;
1332 default:
1333 return -EINVAL;
1334 }
1335
1336 if (!po->running)
1337 return -EINVAL;
1338
1339 if (po->fanout)
1340 return -EALREADY;
1341
1342 mutex_lock(&fanout_mutex);
1343 match = NULL;
1344 list_for_each_entry(f, &fanout_list, list) {
1345 if (f->id == id &&
1346 read_pnet(&f->net) == sock_net(sk)) {
1347 match = f;
1348 break;
1349 }
1350 }
1351 err = -EINVAL;
1352 if (match && match->defrag != defrag)
1353 goto out;
1354 if (!match) {
1355 err = -ENOMEM;
1356 match = kzalloc(sizeof(*match), GFP_KERNEL);
1357 if (!match)
1358 goto out;
1359 write_pnet(&match->net, sock_net(sk));
1360 match->id = id;
1361 match->type = type;
1362 match->defrag = defrag;
1363 atomic_set(&match->rr_cur, 0);
1364 INIT_LIST_HEAD(&match->list);
1365 spin_lock_init(&match->lock);
1366 atomic_set(&match->sk_ref, 0);
1367 match->prot_hook.type = po->prot_hook.type;
1368 match->prot_hook.dev = po->prot_hook.dev;
1369 match->prot_hook.func = packet_rcv_fanout;
1370 match->prot_hook.af_packet_priv = match;
1371 dev_add_pack(&match->prot_hook);
1372 list_add(&match->list, &fanout_list);
1373 }
1374 err = -EINVAL;
1375 if (match->type == type &&
1376 match->prot_hook.type == po->prot_hook.type &&
1377 match->prot_hook.dev == po->prot_hook.dev) {
1378 err = -ENOSPC;
1379 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1380 __dev_remove_pack(&po->prot_hook);
1381 po->fanout = match;
1382 atomic_inc(&match->sk_ref);
1383 __fanout_link(sk, po);
1384 err = 0;
1385 }
1386 }
1387 out:
1388 mutex_unlock(&fanout_mutex);
1389 return err;
1390 }
1391
1392 static void fanout_release(struct sock *sk)
1393 {
1394 struct packet_sock *po = pkt_sk(sk);
1395 struct packet_fanout *f;
1396
1397 f = po->fanout;
1398 if (!f)
1399 return;
1400
1401 po->fanout = NULL;
1402
1403 mutex_lock(&fanout_mutex);
1404 if (atomic_dec_and_test(&f->sk_ref)) {
1405 list_del(&f->list);
1406 dev_remove_pack(&f->prot_hook);
1407 kfree(f);
1408 }
1409 mutex_unlock(&fanout_mutex);
1410 }
1411
1412 static const struct proto_ops packet_ops;
1413
1414 static const struct proto_ops packet_ops_spkt;
1415
1416 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1417 struct packet_type *pt, struct net_device *orig_dev)
1418 {
1419 struct sock *sk;
1420 struct sockaddr_pkt *spkt;
1421
1422 /*
1423 * When we registered the protocol we saved the socket in the data
1424 * field for just this event.
1425 */
1426
1427 sk = pt->af_packet_priv;
1428
1429 /*
1430 * Yank back the headers [hope the device set this
1431 * right or kerboom...]
1432 *
1433 * Incoming packets have ll header pulled,
1434 * push it back.
1435 *
1436 * For outgoing ones skb->data == skb_mac_header(skb)
1437 * so that this procedure is noop.
1438 */
1439
1440 if (skb->pkt_type == PACKET_LOOPBACK)
1441 goto out;
1442
1443 if (!net_eq(dev_net(dev), sock_net(sk)))
1444 goto out;
1445
1446 skb = skb_share_check(skb, GFP_ATOMIC);
1447 if (skb == NULL)
1448 goto oom;
1449
1450 /* drop any routing info */
1451 skb_dst_drop(skb);
1452
1453 /* drop conntrack reference */
1454 nf_reset(skb);
1455
1456 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1457
1458 skb_push(skb, skb->data - skb_mac_header(skb));
1459
1460 /*
1461 * The SOCK_PACKET socket receives _all_ frames.
1462 */
1463
1464 spkt->spkt_family = dev->type;
1465 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1466 spkt->spkt_protocol = skb->protocol;
1467
1468 /*
1469 * Charge the memory to the socket. This is done specifically
1470 * to prevent sockets using all the memory up.
1471 */
1472
1473 if (sock_queue_rcv_skb(sk, skb) == 0)
1474 return 0;
1475
1476 out:
1477 kfree_skb(skb);
1478 oom:
1479 return 0;
1480 }
1481
1482
1483 /*
1484 * Output a raw packet to a device layer. This bypasses all the other
1485 * protocol layers and you must therefore supply it with a complete frame
1486 */
1487
1488 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1489 struct msghdr *msg, size_t len)
1490 {
1491 struct sock *sk = sock->sk;
1492 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1493 struct sk_buff *skb = NULL;
1494 struct net_device *dev;
1495 __be16 proto = 0;
1496 int err;
1497
1498 /*
1499 * Get and verify the address.
1500 */
1501
1502 if (saddr) {
1503 if (msg->msg_namelen < sizeof(struct sockaddr))
1504 return -EINVAL;
1505 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1506 proto = saddr->spkt_protocol;
1507 } else
1508 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1509
1510 /*
1511 * Find the device first to size check it
1512 */
1513
1514 saddr->spkt_device[13] = 0;
1515 retry:
1516 rcu_read_lock();
1517 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1518 err = -ENODEV;
1519 if (dev == NULL)
1520 goto out_unlock;
1521
1522 err = -ENETDOWN;
1523 if (!(dev->flags & IFF_UP))
1524 goto out_unlock;
1525
1526 /*
1527 * You may not queue a frame bigger than the mtu. This is the lowest level
1528 * raw protocol and you must do your own fragmentation at this level.
1529 */
1530
1531 err = -EMSGSIZE;
1532 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1533 goto out_unlock;
1534
1535 if (!skb) {
1536 size_t reserved = LL_RESERVED_SPACE(dev);
1537 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1538
1539 rcu_read_unlock();
1540 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1541 if (skb == NULL)
1542 return -ENOBUFS;
1543 /* FIXME: Save some space for broken drivers that write a hard
1544 * header at transmission time by themselves. PPP is the notable
1545 * one here. This should really be fixed at the driver level.
1546 */
1547 skb_reserve(skb, reserved);
1548 skb_reset_network_header(skb);
1549
1550 /* Try to align data part correctly */
1551 if (hhlen) {
1552 skb->data -= hhlen;
1553 skb->tail -= hhlen;
1554 if (len < hhlen)
1555 skb_reset_network_header(skb);
1556 }
1557 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1558 if (err)
1559 goto out_free;
1560 goto retry;
1561 }
1562
1563 if (len > (dev->mtu + dev->hard_header_len)) {
1564 /* Earlier code assumed this would be a VLAN pkt,
1565 * double-check this now that we have the actual
1566 * packet in hand.
1567 */
1568 struct ethhdr *ehdr;
1569 skb_reset_mac_header(skb);
1570 ehdr = eth_hdr(skb);
1571 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1572 err = -EMSGSIZE;
1573 goto out_unlock;
1574 }
1575 }
1576
1577 skb->protocol = proto;
1578 skb->dev = dev;
1579 skb->priority = sk->sk_priority;
1580 skb->mark = sk->sk_mark;
1581 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1582 if (err < 0)
1583 goto out_unlock;
1584
1585 dev_queue_xmit(skb);
1586 rcu_read_unlock();
1587 return len;
1588
1589 out_unlock:
1590 rcu_read_unlock();
1591 out_free:
1592 kfree_skb(skb);
1593 return err;
1594 }
1595
1596 static inline unsigned int run_filter(const struct sk_buff *skb,
1597 const struct sock *sk,
1598 unsigned int res)
1599 {
1600 struct sk_filter *filter;
1601
1602 rcu_read_lock();
1603 filter = rcu_dereference(sk->sk_filter);
1604 if (filter != NULL)
1605 res = SK_RUN_FILTER(filter, skb);
1606 rcu_read_unlock();
1607
1608 return res;
1609 }
1610
1611 /*
1612 * This function makes lazy skb cloning in hope that most of packets
1613 * are discarded by BPF.
1614 *
1615 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1616 * and skb->cb are mangled. It works because (and until) packets
1617 * falling here are owned by current CPU. Output packets are cloned
1618 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1619 * sequencially, so that if we return skb to original state on exit,
1620 * we will not harm anyone.
1621 */
1622
1623 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1624 struct packet_type *pt, struct net_device *orig_dev)
1625 {
1626 struct sock *sk;
1627 struct sockaddr_ll *sll;
1628 struct packet_sock *po;
1629 u8 *skb_head = skb->data;
1630 int skb_len = skb->len;
1631 unsigned int snaplen, res;
1632
1633 if (skb->pkt_type == PACKET_LOOPBACK)
1634 goto drop;
1635
1636 sk = pt->af_packet_priv;
1637 po = pkt_sk(sk);
1638
1639 if (!net_eq(dev_net(dev), sock_net(sk)))
1640 goto drop;
1641
1642 skb->dev = dev;
1643
1644 if (dev->header_ops) {
1645 /* The device has an explicit notion of ll header,
1646 * exported to higher levels.
1647 *
1648 * Otherwise, the device hides details of its frame
1649 * structure, so that corresponding packet head is
1650 * never delivered to user.
1651 */
1652 if (sk->sk_type != SOCK_DGRAM)
1653 skb_push(skb, skb->data - skb_mac_header(skb));
1654 else if (skb->pkt_type == PACKET_OUTGOING) {
1655 /* Special case: outgoing packets have ll header at head */
1656 skb_pull(skb, skb_network_offset(skb));
1657 }
1658 }
1659
1660 snaplen = skb->len;
1661
1662 res = run_filter(skb, sk, snaplen);
1663 if (!res)
1664 goto drop_n_restore;
1665 if (snaplen > res)
1666 snaplen = res;
1667
1668 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1669 (unsigned)sk->sk_rcvbuf)
1670 goto drop_n_acct;
1671
1672 if (skb_shared(skb)) {
1673 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1674 if (nskb == NULL)
1675 goto drop_n_acct;
1676
1677 if (skb_head != skb->data) {
1678 skb->data = skb_head;
1679 skb->len = skb_len;
1680 }
1681 kfree_skb(skb);
1682 skb = nskb;
1683 }
1684
1685 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1686 sizeof(skb->cb));
1687
1688 sll = &PACKET_SKB_CB(skb)->sa.ll;
1689 sll->sll_family = AF_PACKET;
1690 sll->sll_hatype = dev->type;
1691 sll->sll_protocol = skb->protocol;
1692 sll->sll_pkttype = skb->pkt_type;
1693 if (unlikely(po->origdev))
1694 sll->sll_ifindex = orig_dev->ifindex;
1695 else
1696 sll->sll_ifindex = dev->ifindex;
1697
1698 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1699
1700 PACKET_SKB_CB(skb)->origlen = skb->len;
1701
1702 if (pskb_trim(skb, snaplen))
1703 goto drop_n_acct;
1704
1705 skb_set_owner_r(skb, sk);
1706 skb->dev = NULL;
1707 skb_dst_drop(skb);
1708
1709 /* drop conntrack reference */
1710 nf_reset(skb);
1711
1712 spin_lock(&sk->sk_receive_queue.lock);
1713 po->stats.tp_packets++;
1714 skb->dropcount = atomic_read(&sk->sk_drops);
1715 __skb_queue_tail(&sk->sk_receive_queue, skb);
1716 spin_unlock(&sk->sk_receive_queue.lock);
1717 sk->sk_data_ready(sk, skb->len);
1718 return 0;
1719
1720 drop_n_acct:
1721 po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
1722
1723 drop_n_restore:
1724 if (skb_head != skb->data && skb_shared(skb)) {
1725 skb->data = skb_head;
1726 skb->len = skb_len;
1727 }
1728 drop:
1729 consume_skb(skb);
1730 return 0;
1731 }
1732
1733 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1734 struct packet_type *pt, struct net_device *orig_dev)
1735 {
1736 struct sock *sk;
1737 struct packet_sock *po;
1738 struct sockaddr_ll *sll;
1739 union {
1740 struct tpacket_hdr *h1;
1741 struct tpacket2_hdr *h2;
1742 struct tpacket3_hdr *h3;
1743 void *raw;
1744 } h;
1745 u8 *skb_head = skb->data;
1746 int skb_len = skb->len;
1747 unsigned int snaplen, res;
1748 unsigned long status = TP_STATUS_USER;
1749 unsigned short macoff, netoff, hdrlen;
1750 struct sk_buff *copy_skb = NULL;
1751 struct timeval tv;
1752 struct timespec ts;
1753 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1754
1755 if (skb->pkt_type == PACKET_LOOPBACK)
1756 goto drop;
1757
1758 sk = pt->af_packet_priv;
1759 po = pkt_sk(sk);
1760
1761 if (!net_eq(dev_net(dev), sock_net(sk)))
1762 goto drop;
1763
1764 if (dev->header_ops) {
1765 if (sk->sk_type != SOCK_DGRAM)
1766 skb_push(skb, skb->data - skb_mac_header(skb));
1767 else if (skb->pkt_type == PACKET_OUTGOING) {
1768 /* Special case: outgoing packets have ll header at head */
1769 skb_pull(skb, skb_network_offset(skb));
1770 }
1771 }
1772
1773 if (skb->ip_summed == CHECKSUM_PARTIAL)
1774 status |= TP_STATUS_CSUMNOTREADY;
1775
1776 snaplen = skb->len;
1777
1778 res = run_filter(skb, sk, snaplen);
1779 if (!res)
1780 goto drop_n_restore;
1781 if (snaplen > res)
1782 snaplen = res;
1783
1784 if (sk->sk_type == SOCK_DGRAM) {
1785 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1786 po->tp_reserve;
1787 } else {
1788 unsigned maclen = skb_network_offset(skb);
1789 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1790 (maclen < 16 ? 16 : maclen)) +
1791 po->tp_reserve;
1792 macoff = netoff - maclen;
1793 }
1794 if (po->tp_version <= TPACKET_V2) {
1795 if (macoff + snaplen > po->rx_ring.frame_size) {
1796 if (po->copy_thresh &&
1797 atomic_read(&sk->sk_rmem_alloc) + skb->truesize
1798 < (unsigned)sk->sk_rcvbuf) {
1799 if (skb_shared(skb)) {
1800 copy_skb = skb_clone(skb, GFP_ATOMIC);
1801 } else {
1802 copy_skb = skb_get(skb);
1803 skb_head = skb->data;
1804 }
1805 if (copy_skb)
1806 skb_set_owner_r(copy_skb, sk);
1807 }
1808 snaplen = po->rx_ring.frame_size - macoff;
1809 if ((int)snaplen < 0)
1810 snaplen = 0;
1811 }
1812 }
1813 spin_lock(&sk->sk_receive_queue.lock);
1814 h.raw = packet_current_rx_frame(po, skb,
1815 TP_STATUS_KERNEL, (macoff+snaplen));
1816 if (!h.raw)
1817 goto ring_is_full;
1818 if (po->tp_version <= TPACKET_V2) {
1819 packet_increment_rx_head(po, &po->rx_ring);
1820 /*
1821 * LOSING will be reported till you read the stats,
1822 * because it's COR - Clear On Read.
1823 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1824 * at packet level.
1825 */
1826 if (po->stats.tp_drops)
1827 status |= TP_STATUS_LOSING;
1828 }
1829 po->stats.tp_packets++;
1830 if (copy_skb) {
1831 status |= TP_STATUS_COPY;
1832 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1833 }
1834 spin_unlock(&sk->sk_receive_queue.lock);
1835
1836 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1837
1838 switch (po->tp_version) {
1839 case TPACKET_V1:
1840 h.h1->tp_len = skb->len;
1841 h.h1->tp_snaplen = snaplen;
1842 h.h1->tp_mac = macoff;
1843 h.h1->tp_net = netoff;
1844 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1845 && shhwtstamps->syststamp.tv64)
1846 tv = ktime_to_timeval(shhwtstamps->syststamp);
1847 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1848 && shhwtstamps->hwtstamp.tv64)
1849 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1850 else if (skb->tstamp.tv64)
1851 tv = ktime_to_timeval(skb->tstamp);
1852 else
1853 do_gettimeofday(&tv);
1854 h.h1->tp_sec = tv.tv_sec;
1855 h.h1->tp_usec = tv.tv_usec;
1856 hdrlen = sizeof(*h.h1);
1857 break;
1858 case TPACKET_V2:
1859 h.h2->tp_len = skb->len;
1860 h.h2->tp_snaplen = snaplen;
1861 h.h2->tp_mac = macoff;
1862 h.h2->tp_net = netoff;
1863 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1864 && shhwtstamps->syststamp.tv64)
1865 ts = ktime_to_timespec(shhwtstamps->syststamp);
1866 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1867 && shhwtstamps->hwtstamp.tv64)
1868 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1869 else if (skb->tstamp.tv64)
1870 ts = ktime_to_timespec(skb->tstamp);
1871 else
1872 getnstimeofday(&ts);
1873 h.h2->tp_sec = ts.tv_sec;
1874 h.h2->tp_nsec = ts.tv_nsec;
1875 if (vlan_tx_tag_present(skb)) {
1876 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1877 status |= TP_STATUS_VLAN_VALID;
1878 } else {
1879 h.h2->tp_vlan_tci = 0;
1880 }
1881 h.h2->tp_padding = 0;
1882 hdrlen = sizeof(*h.h2);
1883 break;
1884 case TPACKET_V3:
1885 /* tp_nxt_offset,vlan are already populated above.
1886 * So DONT clear those fields here
1887 */
1888 h.h3->tp_status |= status;
1889 h.h3->tp_len = skb->len;
1890 h.h3->tp_snaplen = snaplen;
1891 h.h3->tp_mac = macoff;
1892 h.h3->tp_net = netoff;
1893 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1894 && shhwtstamps->syststamp.tv64)
1895 ts = ktime_to_timespec(shhwtstamps->syststamp);
1896 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1897 && shhwtstamps->hwtstamp.tv64)
1898 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1899 else if (skb->tstamp.tv64)
1900 ts = ktime_to_timespec(skb->tstamp);
1901 else
1902 getnstimeofday(&ts);
1903 h.h3->tp_sec = ts.tv_sec;
1904 h.h3->tp_nsec = ts.tv_nsec;
1905 hdrlen = sizeof(*h.h3);
1906 break;
1907 default:
1908 BUG();
1909 }
1910
1911 sll = h.raw + TPACKET_ALIGN(hdrlen);
1912 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1913 sll->sll_family = AF_PACKET;
1914 sll->sll_hatype = dev->type;
1915 sll->sll_protocol = skb->protocol;
1916 sll->sll_pkttype = skb->pkt_type;
1917 if (unlikely(po->origdev))
1918 sll->sll_ifindex = orig_dev->ifindex;
1919 else
1920 sll->sll_ifindex = dev->ifindex;
1921
1922 smp_mb();
1923 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1924 {
1925 u8 *start, *end;
1926
1927 if (po->tp_version <= TPACKET_V2) {
1928 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1929 + macoff + snaplen);
1930 for (start = h.raw; start < end; start += PAGE_SIZE)
1931 flush_dcache_page(pgv_to_page(start));
1932 }
1933 smp_wmb();
1934 }
1935 #endif
1936 if (po->tp_version <= TPACKET_V2)
1937 __packet_set_status(po, h.raw, status);
1938 else
1939 prb_clear_blk_fill_status(&po->rx_ring);
1940
1941 sk->sk_data_ready(sk, 0);
1942
1943 drop_n_restore:
1944 if (skb_head != skb->data && skb_shared(skb)) {
1945 skb->data = skb_head;
1946 skb->len = skb_len;
1947 }
1948 drop:
1949 kfree_skb(skb);
1950 return 0;
1951
1952 ring_is_full:
1953 po->stats.tp_drops++;
1954 spin_unlock(&sk->sk_receive_queue.lock);
1955
1956 sk->sk_data_ready(sk, 0);
1957 kfree_skb(copy_skb);
1958 goto drop_n_restore;
1959 }
1960
1961 static void tpacket_destruct_skb(struct sk_buff *skb)
1962 {
1963 struct packet_sock *po = pkt_sk(skb->sk);
1964 void *ph;
1965
1966 BUG_ON(skb == NULL);
1967
1968 if (likely(po->tx_ring.pg_vec)) {
1969 ph = skb_shinfo(skb)->destructor_arg;
1970 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1971 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1972 atomic_dec(&po->tx_ring.pending);
1973 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1974 }
1975
1976 sock_wfree(skb);
1977 }
1978
1979 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1980 void *frame, struct net_device *dev, int size_max,
1981 __be16 proto, unsigned char *addr)
1982 {
1983 union {
1984 struct tpacket_hdr *h1;
1985 struct tpacket2_hdr *h2;
1986 void *raw;
1987 } ph;
1988 int to_write, offset, len, tp_len, nr_frags, len_max;
1989 struct socket *sock = po->sk.sk_socket;
1990 struct page *page;
1991 void *data;
1992 int err;
1993
1994 ph.raw = frame;
1995
1996 skb->protocol = proto;
1997 skb->dev = dev;
1998 skb->priority = po->sk.sk_priority;
1999 skb->mark = po->sk.sk_mark;
2000 skb_shinfo(skb)->destructor_arg = ph.raw;
2001
2002 switch (po->tp_version) {
2003 case TPACKET_V2:
2004 tp_len = ph.h2->tp_len;
2005 break;
2006 default:
2007 tp_len = ph.h1->tp_len;
2008 break;
2009 }
2010 if (unlikely(tp_len > size_max)) {
2011 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2012 return -EMSGSIZE;
2013 }
2014
2015 skb_reserve(skb, LL_RESERVED_SPACE(dev));
2016 skb_reset_network_header(skb);
2017
2018 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2019 to_write = tp_len;
2020
2021 if (sock->type == SOCK_DGRAM) {
2022 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2023 NULL, tp_len);
2024 if (unlikely(err < 0))
2025 return -EINVAL;
2026 } else if (dev->hard_header_len) {
2027 /* net device doesn't like empty head */
2028 if (unlikely(tp_len <= dev->hard_header_len)) {
2029 pr_err("packet size is too short (%d < %d)\n",
2030 tp_len, dev->hard_header_len);
2031 return -EINVAL;
2032 }
2033
2034 skb_push(skb, dev->hard_header_len);
2035 err = skb_store_bits(skb, 0, data,
2036 dev->hard_header_len);
2037 if (unlikely(err))
2038 return err;
2039
2040 data += dev->hard_header_len;
2041 to_write -= dev->hard_header_len;
2042 }
2043
2044 err = -EFAULT;
2045 offset = offset_in_page(data);
2046 len_max = PAGE_SIZE - offset;
2047 len = ((to_write > len_max) ? len_max : to_write);
2048
2049 skb->data_len = to_write;
2050 skb->len += to_write;
2051 skb->truesize += to_write;
2052 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2053
2054 while (likely(to_write)) {
2055 nr_frags = skb_shinfo(skb)->nr_frags;
2056
2057 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2058 pr_err("Packet exceed the number of skb frags(%lu)\n",
2059 MAX_SKB_FRAGS);
2060 return -EFAULT;
2061 }
2062
2063 page = pgv_to_page(data);
2064 data += len;
2065 flush_dcache_page(page);
2066 get_page(page);
2067 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2068 to_write -= len;
2069 offset = 0;
2070 len_max = PAGE_SIZE;
2071 len = ((to_write > len_max) ? len_max : to_write);
2072 }
2073
2074 return tp_len;
2075 }
2076
2077 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2078 {
2079 struct sk_buff *skb;
2080 struct net_device *dev;
2081 __be16 proto;
2082 bool need_rls_dev = false;
2083 int err, reserve = 0;
2084 void *ph;
2085 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2086 int tp_len, size_max;
2087 unsigned char *addr;
2088 int len_sum = 0;
2089 int status = 0;
2090
2091 mutex_lock(&po->pg_vec_lock);
2092
2093 err = -EBUSY;
2094 if (saddr == NULL) {
2095 dev = po->prot_hook.dev;
2096 proto = po->num;
2097 addr = NULL;
2098 } else {
2099 err = -EINVAL;
2100 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2101 goto out;
2102 if (msg->msg_namelen < (saddr->sll_halen
2103 + offsetof(struct sockaddr_ll,
2104 sll_addr)))
2105 goto out;
2106 proto = saddr->sll_protocol;
2107 addr = saddr->sll_addr;
2108 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2109 need_rls_dev = true;
2110 }
2111
2112 err = -ENXIO;
2113 if (unlikely(dev == NULL))
2114 goto out;
2115
2116 reserve = dev->hard_header_len;
2117
2118 err = -ENETDOWN;
2119 if (unlikely(!(dev->flags & IFF_UP)))
2120 goto out_put;
2121
2122 size_max = po->tx_ring.frame_size
2123 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2124
2125 if (size_max > dev->mtu + reserve)
2126 size_max = dev->mtu + reserve;
2127
2128 do {
2129 ph = packet_current_frame(po, &po->tx_ring,
2130 TP_STATUS_SEND_REQUEST);
2131
2132 if (unlikely(ph == NULL)) {
2133 schedule();
2134 continue;
2135 }
2136
2137 status = TP_STATUS_SEND_REQUEST;
2138 skb = sock_alloc_send_skb(&po->sk,
2139 LL_ALLOCATED_SPACE(dev)
2140 + sizeof(struct sockaddr_ll),
2141 0, &err);
2142
2143 if (unlikely(skb == NULL))
2144 goto out_status;
2145
2146 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2147 addr);
2148
2149 if (unlikely(tp_len < 0)) {
2150 if (po->tp_loss) {
2151 __packet_set_status(po, ph,
2152 TP_STATUS_AVAILABLE);
2153 packet_increment_head(&po->tx_ring);
2154 kfree_skb(skb);
2155 continue;
2156 } else {
2157 status = TP_STATUS_WRONG_FORMAT;
2158 err = tp_len;
2159 goto out_status;
2160 }
2161 }
2162
2163 skb->destructor = tpacket_destruct_skb;
2164 __packet_set_status(po, ph, TP_STATUS_SENDING);
2165 atomic_inc(&po->tx_ring.pending);
2166
2167 status = TP_STATUS_SEND_REQUEST;
2168 err = dev_queue_xmit(skb);
2169 if (unlikely(err > 0)) {
2170 err = net_xmit_errno(err);
2171 if (err && __packet_get_status(po, ph) ==
2172 TP_STATUS_AVAILABLE) {
2173 /* skb was destructed already */
2174 skb = NULL;
2175 goto out_status;
2176 }
2177 /*
2178 * skb was dropped but not destructed yet;
2179 * let's treat it like congestion or err < 0
2180 */
2181 err = 0;
2182 }
2183 packet_increment_head(&po->tx_ring);
2184 len_sum += tp_len;
2185 } while (likely((ph != NULL) ||
2186 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2187 (atomic_read(&po->tx_ring.pending))))
2188 );
2189
2190 err = len_sum;
2191 goto out_put;
2192
2193 out_status:
2194 __packet_set_status(po, ph, status);
2195 kfree_skb(skb);
2196 out_put:
2197 if (need_rls_dev)
2198 dev_put(dev);
2199 out:
2200 mutex_unlock(&po->pg_vec_lock);
2201 return err;
2202 }
2203
2204 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2205 size_t reserve, size_t len,
2206 size_t linear, int noblock,
2207 int *err)
2208 {
2209 struct sk_buff *skb;
2210
2211 /* Under a page? Don't bother with paged skb. */
2212 if (prepad + len < PAGE_SIZE || !linear)
2213 linear = len;
2214
2215 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2216 err);
2217 if (!skb)
2218 return NULL;
2219
2220 skb_reserve(skb, reserve);
2221 skb_put(skb, linear);
2222 skb->data_len = len - linear;
2223 skb->len += len - linear;
2224
2225 return skb;
2226 }
2227
2228 static int packet_snd(struct socket *sock,
2229 struct msghdr *msg, size_t len)
2230 {
2231 struct sock *sk = sock->sk;
2232 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2233 struct sk_buff *skb;
2234 struct net_device *dev;
2235 __be16 proto;
2236 bool need_rls_dev = false;
2237 unsigned char *addr;
2238 int err, reserve = 0;
2239 struct virtio_net_hdr vnet_hdr = { 0 };
2240 int offset = 0;
2241 int vnet_hdr_len;
2242 struct packet_sock *po = pkt_sk(sk);
2243 unsigned short gso_type = 0;
2244
2245 /*
2246 * Get and verify the address.
2247 */
2248
2249 if (saddr == NULL) {
2250 dev = po->prot_hook.dev;
2251 proto = po->num;
2252 addr = NULL;
2253 } else {
2254 err = -EINVAL;
2255 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2256 goto out;
2257 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2258 goto out;
2259 proto = saddr->sll_protocol;
2260 addr = saddr->sll_addr;
2261 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2262 need_rls_dev = true;
2263 }
2264
2265 err = -ENXIO;
2266 if (dev == NULL)
2267 goto out_unlock;
2268 if (sock->type == SOCK_RAW)
2269 reserve = dev->hard_header_len;
2270
2271 err = -ENETDOWN;
2272 if (!(dev->flags & IFF_UP))
2273 goto out_unlock;
2274
2275 if (po->has_vnet_hdr) {
2276 vnet_hdr_len = sizeof(vnet_hdr);
2277
2278 err = -EINVAL;
2279 if (len < vnet_hdr_len)
2280 goto out_unlock;
2281
2282 len -= vnet_hdr_len;
2283
2284 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2285 vnet_hdr_len);
2286 if (err < 0)
2287 goto out_unlock;
2288
2289 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2290 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2291 vnet_hdr.hdr_len))
2292 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2293 vnet_hdr.csum_offset + 2;
2294
2295 err = -EINVAL;
2296 if (vnet_hdr.hdr_len > len)
2297 goto out_unlock;
2298
2299 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2300 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2301 case VIRTIO_NET_HDR_GSO_TCPV4:
2302 gso_type = SKB_GSO_TCPV4;
2303 break;
2304 case VIRTIO_NET_HDR_GSO_TCPV6:
2305 gso_type = SKB_GSO_TCPV6;
2306 break;
2307 case VIRTIO_NET_HDR_GSO_UDP:
2308 gso_type = SKB_GSO_UDP;
2309 break;
2310 default:
2311 goto out_unlock;
2312 }
2313
2314 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2315 gso_type |= SKB_GSO_TCP_ECN;
2316
2317 if (vnet_hdr.gso_size == 0)
2318 goto out_unlock;
2319
2320 }
2321 }
2322
2323 err = -EMSGSIZE;
2324 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2325 goto out_unlock;
2326
2327 err = -ENOBUFS;
2328 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2329 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2330 msg->msg_flags & MSG_DONTWAIT, &err);
2331 if (skb == NULL)
2332 goto out_unlock;
2333
2334 skb_set_network_header(skb, reserve);
2335
2336 err = -EINVAL;
2337 if (sock->type == SOCK_DGRAM &&
2338 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2339 goto out_free;
2340
2341 /* Returns -EFAULT on error */
2342 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2343 if (err)
2344 goto out_free;
2345 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2346 if (err < 0)
2347 goto out_free;
2348
2349 if (!gso_type && (len > dev->mtu + reserve)) {
2350 /* Earlier code assumed this would be a VLAN pkt,
2351 * double-check this now that we have the actual
2352 * packet in hand.
2353 */
2354 struct ethhdr *ehdr;
2355 skb_reset_mac_header(skb);
2356 ehdr = eth_hdr(skb);
2357 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2358 err = -EMSGSIZE;
2359 goto out_free;
2360 }
2361 }
2362
2363 skb->protocol = proto;
2364 skb->dev = dev;
2365 skb->priority = sk->sk_priority;
2366 skb->mark = sk->sk_mark;
2367
2368 if (po->has_vnet_hdr) {
2369 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2370 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2371 vnet_hdr.csum_offset)) {
2372 err = -EINVAL;
2373 goto out_free;
2374 }
2375 }
2376
2377 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2378 skb_shinfo(skb)->gso_type = gso_type;
2379
2380 /* Header must be checked, and gso_segs computed. */
2381 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2382 skb_shinfo(skb)->gso_segs = 0;
2383
2384 len += vnet_hdr_len;
2385 }
2386
2387 /*
2388 * Now send it
2389 */
2390
2391 err = dev_queue_xmit(skb);
2392 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2393 goto out_unlock;
2394
2395 if (need_rls_dev)
2396 dev_put(dev);
2397
2398 return len;
2399
2400 out_free:
2401 kfree_skb(skb);
2402 out_unlock:
2403 if (dev && need_rls_dev)
2404 dev_put(dev);
2405 out:
2406 return err;
2407 }
2408
2409 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2410 struct msghdr *msg, size_t len)
2411 {
2412 struct sock *sk = sock->sk;
2413 struct packet_sock *po = pkt_sk(sk);
2414 if (po->tx_ring.pg_vec)
2415 return tpacket_snd(po, msg);
2416 else
2417 return packet_snd(sock, msg, len);
2418 }
2419
2420 /*
2421 * Close a PACKET socket. This is fairly simple. We immediately go
2422 * to 'closed' state and remove our protocol entry in the device list.
2423 */
2424
2425 static int packet_release(struct socket *sock)
2426 {
2427 struct sock *sk = sock->sk;
2428 struct packet_sock *po;
2429 struct net *net;
2430 union tpacket_req_u req_u;
2431
2432 if (!sk)
2433 return 0;
2434
2435 net = sock_net(sk);
2436 po = pkt_sk(sk);
2437
2438 spin_lock_bh(&net->packet.sklist_lock);
2439 sk_del_node_init_rcu(sk);
2440 sock_prot_inuse_add(net, sk->sk_prot, -1);
2441 spin_unlock_bh(&net->packet.sklist_lock);
2442
2443 spin_lock(&po->bind_lock);
2444 unregister_prot_hook(sk, false);
2445 if (po->prot_hook.dev) {
2446 dev_put(po->prot_hook.dev);
2447 po->prot_hook.dev = NULL;
2448 }
2449 spin_unlock(&po->bind_lock);
2450
2451 packet_flush_mclist(sk);
2452
2453 memset(&req_u, 0, sizeof(req_u));
2454
2455 if (po->rx_ring.pg_vec)
2456 packet_set_ring(sk, &req_u, 1, 0);
2457
2458 if (po->tx_ring.pg_vec)
2459 packet_set_ring(sk, &req_u, 1, 1);
2460
2461 fanout_release(sk);
2462
2463 synchronize_net();
2464 /*
2465 * Now the socket is dead. No more input will appear.
2466 */
2467 sock_orphan(sk);
2468 sock->sk = NULL;
2469
2470 /* Purge queues */
2471
2472 skb_queue_purge(&sk->sk_receive_queue);
2473 sk_refcnt_debug_release(sk);
2474
2475 sock_put(sk);
2476 return 0;
2477 }
2478
2479 /*
2480 * Attach a packet hook.
2481 */
2482
2483 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2484 {
2485 struct packet_sock *po = pkt_sk(sk);
2486
2487 if (po->fanout)
2488 return -EINVAL;
2489
2490 lock_sock(sk);
2491
2492 spin_lock(&po->bind_lock);
2493 unregister_prot_hook(sk, true);
2494 po->num = protocol;
2495 po->prot_hook.type = protocol;
2496 if (po->prot_hook.dev)
2497 dev_put(po->prot_hook.dev);
2498 po->prot_hook.dev = dev;
2499
2500 po->ifindex = dev ? dev->ifindex : 0;
2501
2502 if (protocol == 0)
2503 goto out_unlock;
2504
2505 if (!dev || (dev->flags & IFF_UP)) {
2506 register_prot_hook(sk);
2507 } else {
2508 sk->sk_err = ENETDOWN;
2509 if (!sock_flag(sk, SOCK_DEAD))
2510 sk->sk_error_report(sk);
2511 }
2512
2513 out_unlock:
2514 spin_unlock(&po->bind_lock);
2515 release_sock(sk);
2516 return 0;
2517 }
2518
2519 /*
2520 * Bind a packet socket to a device
2521 */
2522
2523 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2524 int addr_len)
2525 {
2526 struct sock *sk = sock->sk;
2527 char name[15];
2528 struct net_device *dev;
2529 int err = -ENODEV;
2530
2531 /*
2532 * Check legality
2533 */
2534
2535 if (addr_len != sizeof(struct sockaddr))
2536 return -EINVAL;
2537 strlcpy(name, uaddr->sa_data, sizeof(name));
2538
2539 dev = dev_get_by_name(sock_net(sk), name);
2540 if (dev)
2541 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2542 return err;
2543 }
2544
2545 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2546 {
2547 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2548 struct sock *sk = sock->sk;
2549 struct net_device *dev = NULL;
2550 int err;
2551
2552
2553 /*
2554 * Check legality
2555 */
2556
2557 if (addr_len < sizeof(struct sockaddr_ll))
2558 return -EINVAL;
2559 if (sll->sll_family != AF_PACKET)
2560 return -EINVAL;
2561
2562 if (sll->sll_ifindex) {
2563 err = -ENODEV;
2564 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2565 if (dev == NULL)
2566 goto out;
2567 }
2568 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2569
2570 out:
2571 return err;
2572 }
2573
2574 static struct proto packet_proto = {
2575 .name = "PACKET",
2576 .owner = THIS_MODULE,
2577 .obj_size = sizeof(struct packet_sock),
2578 };
2579
2580 /*
2581 * Create a packet of type SOCK_PACKET.
2582 */
2583
2584 static int packet_create(struct net *net, struct socket *sock, int protocol,
2585 int kern)
2586 {
2587 struct sock *sk;
2588 struct packet_sock *po;
2589 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2590 int err;
2591
2592 if (!capable(CAP_NET_RAW))
2593 return -EPERM;
2594 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2595 sock->type != SOCK_PACKET)
2596 return -ESOCKTNOSUPPORT;
2597
2598 sock->state = SS_UNCONNECTED;
2599
2600 err = -ENOBUFS;
2601 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2602 if (sk == NULL)
2603 goto out;
2604
2605 sock->ops = &packet_ops;
2606 if (sock->type == SOCK_PACKET)
2607 sock->ops = &packet_ops_spkt;
2608
2609 sock_init_data(sock, sk);
2610
2611 po = pkt_sk(sk);
2612 sk->sk_family = PF_PACKET;
2613 po->num = proto;
2614
2615 sk->sk_destruct = packet_sock_destruct;
2616 sk_refcnt_debug_inc(sk);
2617
2618 /*
2619 * Attach a protocol block
2620 */
2621
2622 spin_lock_init(&po->bind_lock);
2623 mutex_init(&po->pg_vec_lock);
2624 po->prot_hook.func = packet_rcv;
2625
2626 if (sock->type == SOCK_PACKET)
2627 po->prot_hook.func = packet_rcv_spkt;
2628
2629 po->prot_hook.af_packet_priv = sk;
2630
2631 if (proto) {
2632 po->prot_hook.type = proto;
2633 register_prot_hook(sk);
2634 }
2635
2636 spin_lock_bh(&net->packet.sklist_lock);
2637 sk_add_node_rcu(sk, &net->packet.sklist);
2638 sock_prot_inuse_add(net, &packet_proto, 1);
2639 spin_unlock_bh(&net->packet.sklist_lock);
2640
2641 return 0;
2642 out:
2643 return err;
2644 }
2645
2646 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2647 {
2648 struct sock_exterr_skb *serr;
2649 struct sk_buff *skb, *skb2;
2650 int copied, err;
2651
2652 err = -EAGAIN;
2653 skb = skb_dequeue(&sk->sk_error_queue);
2654 if (skb == NULL)
2655 goto out;
2656
2657 copied = skb->len;
2658 if (copied > len) {
2659 msg->msg_flags |= MSG_TRUNC;
2660 copied = len;
2661 }
2662 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2663 if (err)
2664 goto out_free_skb;
2665
2666 sock_recv_timestamp(msg, sk, skb);
2667
2668 serr = SKB_EXT_ERR(skb);
2669 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2670 sizeof(serr->ee), &serr->ee);
2671
2672 msg->msg_flags |= MSG_ERRQUEUE;
2673 err = copied;
2674
2675 /* Reset and regenerate socket error */
2676 spin_lock_bh(&sk->sk_error_queue.lock);
2677 sk->sk_err = 0;
2678 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2679 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2680 spin_unlock_bh(&sk->sk_error_queue.lock);
2681 sk->sk_error_report(sk);
2682 } else
2683 spin_unlock_bh(&sk->sk_error_queue.lock);
2684
2685 out_free_skb:
2686 kfree_skb(skb);
2687 out:
2688 return err;
2689 }
2690
2691 /*
2692 * Pull a packet from our receive queue and hand it to the user.
2693 * If necessary we block.
2694 */
2695
2696 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2697 struct msghdr *msg, size_t len, int flags)
2698 {
2699 struct sock *sk = sock->sk;
2700 struct sk_buff *skb;
2701 int copied, err;
2702 struct sockaddr_ll *sll;
2703 int vnet_hdr_len = 0;
2704
2705 err = -EINVAL;
2706 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2707 goto out;
2708
2709 #if 0
2710 /* What error should we return now? EUNATTACH? */
2711 if (pkt_sk(sk)->ifindex < 0)
2712 return -ENODEV;
2713 #endif
2714
2715 if (flags & MSG_ERRQUEUE) {
2716 err = packet_recv_error(sk, msg, len);
2717 goto out;
2718 }
2719
2720 /*
2721 * Call the generic datagram receiver. This handles all sorts
2722 * of horrible races and re-entrancy so we can forget about it
2723 * in the protocol layers.
2724 *
2725 * Now it will return ENETDOWN, if device have just gone down,
2726 * but then it will block.
2727 */
2728
2729 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2730
2731 /*
2732 * An error occurred so return it. Because skb_recv_datagram()
2733 * handles the blocking we don't see and worry about blocking
2734 * retries.
2735 */
2736
2737 if (skb == NULL)
2738 goto out;
2739
2740 if (pkt_sk(sk)->has_vnet_hdr) {
2741 struct virtio_net_hdr vnet_hdr = { 0 };
2742
2743 err = -EINVAL;
2744 vnet_hdr_len = sizeof(vnet_hdr);
2745 if (len < vnet_hdr_len)
2746 goto out_free;
2747
2748 len -= vnet_hdr_len;
2749
2750 if (skb_is_gso(skb)) {
2751 struct skb_shared_info *sinfo = skb_shinfo(skb);
2752
2753 /* This is a hint as to how much should be linear. */
2754 vnet_hdr.hdr_len = skb_headlen(skb);
2755 vnet_hdr.gso_size = sinfo->gso_size;
2756 if (sinfo->gso_type & SKB_GSO_TCPV4)
2757 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2758 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2759 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2760 else if (sinfo->gso_type & SKB_GSO_UDP)
2761 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2762 else if (sinfo->gso_type & SKB_GSO_FCOE)
2763 goto out_free;
2764 else
2765 BUG();
2766 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2767 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2768 } else
2769 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2770
2771 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2772 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2773 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2774 vnet_hdr.csum_offset = skb->csum_offset;
2775 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2776 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2777 } /* else everything is zero */
2778
2779 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2780 vnet_hdr_len);
2781 if (err < 0)
2782 goto out_free;
2783 }
2784
2785 /*
2786 * If the address length field is there to be filled in, we fill
2787 * it in now.
2788 */
2789
2790 sll = &PACKET_SKB_CB(skb)->sa.ll;
2791 if (sock->type == SOCK_PACKET)
2792 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2793 else
2794 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2795
2796 /*
2797 * You lose any data beyond the buffer you gave. If it worries a
2798 * user program they can ask the device for its MTU anyway.
2799 */
2800
2801 copied = skb->len;
2802 if (copied > len) {
2803 copied = len;
2804 msg->msg_flags |= MSG_TRUNC;
2805 }
2806
2807 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2808 if (err)
2809 goto out_free;
2810
2811 sock_recv_ts_and_drops(msg, sk, skb);
2812
2813 if (msg->msg_name)
2814 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2815 msg->msg_namelen);
2816
2817 if (pkt_sk(sk)->auxdata) {
2818 struct tpacket_auxdata aux;
2819
2820 aux.tp_status = TP_STATUS_USER;
2821 if (skb->ip_summed == CHECKSUM_PARTIAL)
2822 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2823 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2824 aux.tp_snaplen = skb->len;
2825 aux.tp_mac = 0;
2826 aux.tp_net = skb_network_offset(skb);
2827 if (vlan_tx_tag_present(skb)) {
2828 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2829 aux.tp_status |= TP_STATUS_VLAN_VALID;
2830 } else {
2831 aux.tp_vlan_tci = 0;
2832 }
2833 aux.tp_padding = 0;
2834 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2835 }
2836
2837 /*
2838 * Free or return the buffer as appropriate. Again this
2839 * hides all the races and re-entrancy issues from us.
2840 */
2841 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2842
2843 out_free:
2844 skb_free_datagram(sk, skb);
2845 out:
2846 return err;
2847 }
2848
2849 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2850 int *uaddr_len, int peer)
2851 {
2852 struct net_device *dev;
2853 struct sock *sk = sock->sk;
2854
2855 if (peer)
2856 return -EOPNOTSUPP;
2857
2858 uaddr->sa_family = AF_PACKET;
2859 rcu_read_lock();
2860 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2861 if (dev)
2862 strncpy(uaddr->sa_data, dev->name, 14);
2863 else
2864 memset(uaddr->sa_data, 0, 14);
2865 rcu_read_unlock();
2866 *uaddr_len = sizeof(*uaddr);
2867
2868 return 0;
2869 }
2870
2871 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2872 int *uaddr_len, int peer)
2873 {
2874 struct net_device *dev;
2875 struct sock *sk = sock->sk;
2876 struct packet_sock *po = pkt_sk(sk);
2877 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2878
2879 if (peer)
2880 return -EOPNOTSUPP;
2881
2882 sll->sll_family = AF_PACKET;
2883 sll->sll_ifindex = po->ifindex;
2884 sll->sll_protocol = po->num;
2885 sll->sll_pkttype = 0;
2886 rcu_read_lock();
2887 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2888 if (dev) {
2889 sll->sll_hatype = dev->type;
2890 sll->sll_halen = dev->addr_len;
2891 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2892 } else {
2893 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2894 sll->sll_halen = 0;
2895 }
2896 rcu_read_unlock();
2897 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2898
2899 return 0;
2900 }
2901
2902 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2903 int what)
2904 {
2905 switch (i->type) {
2906 case PACKET_MR_MULTICAST:
2907 if (i->alen != dev->addr_len)
2908 return -EINVAL;
2909 if (what > 0)
2910 return dev_mc_add(dev, i->addr);
2911 else
2912 return dev_mc_del(dev, i->addr);
2913 break;
2914 case PACKET_MR_PROMISC:
2915 return dev_set_promiscuity(dev, what);
2916 break;
2917 case PACKET_MR_ALLMULTI:
2918 return dev_set_allmulti(dev, what);
2919 break;
2920 case PACKET_MR_UNICAST:
2921 if (i->alen != dev->addr_len)
2922 return -EINVAL;
2923 if (what > 0)
2924 return dev_uc_add(dev, i->addr);
2925 else
2926 return dev_uc_del(dev, i->addr);
2927 break;
2928 default:
2929 break;
2930 }
2931 return 0;
2932 }
2933
2934 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2935 {
2936 for ( ; i; i = i->next) {
2937 if (i->ifindex == dev->ifindex)
2938 packet_dev_mc(dev, i, what);
2939 }
2940 }
2941
2942 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2943 {
2944 struct packet_sock *po = pkt_sk(sk);
2945 struct packet_mclist *ml, *i;
2946 struct net_device *dev;
2947 int err;
2948
2949 rtnl_lock();
2950
2951 err = -ENODEV;
2952 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2953 if (!dev)
2954 goto done;
2955
2956 err = -EINVAL;
2957 if (mreq->mr_alen > dev->addr_len)
2958 goto done;
2959
2960 err = -ENOBUFS;
2961 i = kmalloc(sizeof(*i), GFP_KERNEL);
2962 if (i == NULL)
2963 goto done;
2964
2965 err = 0;
2966 for (ml = po->mclist; ml; ml = ml->next) {
2967 if (ml->ifindex == mreq->mr_ifindex &&
2968 ml->type == mreq->mr_type &&
2969 ml->alen == mreq->mr_alen &&
2970 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2971 ml->count++;
2972 /* Free the new element ... */
2973 kfree(i);
2974 goto done;
2975 }
2976 }
2977
2978 i->type = mreq->mr_type;
2979 i->ifindex = mreq->mr_ifindex;
2980 i->alen = mreq->mr_alen;
2981 memcpy(i->addr, mreq->mr_address, i->alen);
2982 i->count = 1;
2983 i->next = po->mclist;
2984 po->mclist = i;
2985 err = packet_dev_mc(dev, i, 1);
2986 if (err) {
2987 po->mclist = i->next;
2988 kfree(i);
2989 }
2990
2991 done:
2992 rtnl_unlock();
2993 return err;
2994 }
2995
2996 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2997 {
2998 struct packet_mclist *ml, **mlp;
2999
3000 rtnl_lock();
3001
3002 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3003 if (ml->ifindex == mreq->mr_ifindex &&
3004 ml->type == mreq->mr_type &&
3005 ml->alen == mreq->mr_alen &&
3006 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3007 if (--ml->count == 0) {
3008 struct net_device *dev;
3009 *mlp = ml->next;
3010 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3011 if (dev)
3012 packet_dev_mc(dev, ml, -1);
3013 kfree(ml);
3014 }
3015 rtnl_unlock();
3016 return 0;
3017 }
3018 }
3019 rtnl_unlock();
3020 return -EADDRNOTAVAIL;
3021 }
3022
3023 static void packet_flush_mclist(struct sock *sk)
3024 {
3025 struct packet_sock *po = pkt_sk(sk);
3026 struct packet_mclist *ml;
3027
3028 if (!po->mclist)
3029 return;
3030
3031 rtnl_lock();
3032 while ((ml = po->mclist) != NULL) {
3033 struct net_device *dev;
3034
3035 po->mclist = ml->next;
3036 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3037 if (dev != NULL)
3038 packet_dev_mc(dev, ml, -1);
3039 kfree(ml);
3040 }
3041 rtnl_unlock();
3042 }
3043
3044 static int
3045 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3046 {
3047 struct sock *sk = sock->sk;
3048 struct packet_sock *po = pkt_sk(sk);
3049 int ret;
3050
3051 if (level != SOL_PACKET)
3052 return -ENOPROTOOPT;
3053
3054 switch (optname) {
3055 case PACKET_ADD_MEMBERSHIP:
3056 case PACKET_DROP_MEMBERSHIP:
3057 {
3058 struct packet_mreq_max mreq;
3059 int len = optlen;
3060 memset(&mreq, 0, sizeof(mreq));
3061 if (len < sizeof(struct packet_mreq))
3062 return -EINVAL;
3063 if (len > sizeof(mreq))
3064 len = sizeof(mreq);
3065 if (copy_from_user(&mreq, optval, len))
3066 return -EFAULT;
3067 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3068 return -EINVAL;
3069 if (optname == PACKET_ADD_MEMBERSHIP)
3070 ret = packet_mc_add(sk, &mreq);
3071 else
3072 ret = packet_mc_drop(sk, &mreq);
3073 return ret;
3074 }
3075
3076 case PACKET_RX_RING:
3077 case PACKET_TX_RING:
3078 {
3079 union tpacket_req_u req_u;
3080 int len;
3081
3082 switch (po->tp_version) {
3083 case TPACKET_V1:
3084 case TPACKET_V2:
3085 len = sizeof(req_u.req);
3086 break;
3087 case TPACKET_V3:
3088 default:
3089 len = sizeof(req_u.req3);
3090 break;
3091 }
3092 if (optlen < len)
3093 return -EINVAL;
3094 if (pkt_sk(sk)->has_vnet_hdr)
3095 return -EINVAL;
3096 if (copy_from_user(&req_u.req, optval, len))
3097 return -EFAULT;
3098 return packet_set_ring(sk, &req_u, 0,
3099 optname == PACKET_TX_RING);
3100 }
3101 case PACKET_COPY_THRESH:
3102 {
3103 int val;
3104
3105 if (optlen != sizeof(val))
3106 return -EINVAL;
3107 if (copy_from_user(&val, optval, sizeof(val)))
3108 return -EFAULT;
3109
3110 pkt_sk(sk)->copy_thresh = val;
3111 return 0;
3112 }
3113 case PACKET_VERSION:
3114 {
3115 int val;
3116
3117 if (optlen != sizeof(val))
3118 return -EINVAL;
3119 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3120 return -EBUSY;
3121 if (copy_from_user(&val, optval, sizeof(val)))
3122 return -EFAULT;
3123 switch (val) {
3124 case TPACKET_V1:
3125 case TPACKET_V2:
3126 case TPACKET_V3:
3127 po->tp_version = val;
3128 return 0;
3129 default:
3130 return -EINVAL;
3131 }
3132 }
3133 case PACKET_RESERVE:
3134 {
3135 unsigned int val;
3136
3137 if (optlen != sizeof(val))
3138 return -EINVAL;
3139 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3140 return -EBUSY;
3141 if (copy_from_user(&val, optval, sizeof(val)))
3142 return -EFAULT;
3143 po->tp_reserve = val;
3144 return 0;
3145 }
3146 case PACKET_LOSS:
3147 {
3148 unsigned int val;
3149
3150 if (optlen != sizeof(val))
3151 return -EINVAL;
3152 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3153 return -EBUSY;
3154 if (copy_from_user(&val, optval, sizeof(val)))
3155 return -EFAULT;
3156 po->tp_loss = !!val;
3157 return 0;
3158 }
3159 case PACKET_AUXDATA:
3160 {
3161 int val;
3162
3163 if (optlen < sizeof(val))
3164 return -EINVAL;
3165 if (copy_from_user(&val, optval, sizeof(val)))
3166 return -EFAULT;
3167
3168 po->auxdata = !!val;
3169 return 0;
3170 }
3171 case PACKET_ORIGDEV:
3172 {
3173 int val;
3174
3175 if (optlen < sizeof(val))
3176 return -EINVAL;
3177 if (copy_from_user(&val, optval, sizeof(val)))
3178 return -EFAULT;
3179
3180 po->origdev = !!val;
3181 return 0;
3182 }
3183 case PACKET_VNET_HDR:
3184 {
3185 int val;
3186
3187 if (sock->type != SOCK_RAW)
3188 return -EINVAL;
3189 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3190 return -EBUSY;
3191 if (optlen < sizeof(val))
3192 return -EINVAL;
3193 if (copy_from_user(&val, optval, sizeof(val)))
3194 return -EFAULT;
3195
3196 po->has_vnet_hdr = !!val;
3197 return 0;
3198 }
3199 case PACKET_TIMESTAMP:
3200 {
3201 int val;
3202
3203 if (optlen != sizeof(val))
3204 return -EINVAL;
3205 if (copy_from_user(&val, optval, sizeof(val)))
3206 return -EFAULT;
3207
3208 po->tp_tstamp = val;
3209 return 0;
3210 }
3211 case PACKET_FANOUT:
3212 {
3213 int val;
3214
3215 if (optlen != sizeof(val))
3216 return -EINVAL;
3217 if (copy_from_user(&val, optval, sizeof(val)))
3218 return -EFAULT;
3219
3220 return fanout_add(sk, val & 0xffff, val >> 16);
3221 }
3222 default:
3223 return -ENOPROTOOPT;
3224 }
3225 }
3226
3227 static int packet_getsockopt(struct socket *sock, int level, int optname,
3228 char __user *optval, int __user *optlen)
3229 {
3230 int len;
3231 int val;
3232 struct sock *sk = sock->sk;
3233 struct packet_sock *po = pkt_sk(sk);
3234 void *data;
3235 struct tpacket_stats st;
3236 union tpacket_stats_u st_u;
3237
3238 if (level != SOL_PACKET)
3239 return -ENOPROTOOPT;
3240
3241 if (get_user(len, optlen))
3242 return -EFAULT;
3243
3244 if (len < 0)
3245 return -EINVAL;
3246
3247 switch (optname) {
3248 case PACKET_STATISTICS:
3249 if (po->tp_version == TPACKET_V3) {
3250 len = sizeof(struct tpacket_stats_v3);
3251 } else {
3252 if (len > sizeof(struct tpacket_stats))
3253 len = sizeof(struct tpacket_stats);
3254 }
3255 spin_lock_bh(&sk->sk_receive_queue.lock);
3256 if (po->tp_version == TPACKET_V3) {
3257 memcpy(&st_u.stats3, &po->stats,
3258 sizeof(struct tpacket_stats));
3259 st_u.stats3.tp_freeze_q_cnt =
3260 po->stats_u.stats3.tp_freeze_q_cnt;
3261 st_u.stats3.tp_packets += po->stats.tp_drops;
3262 data = &st_u.stats3;
3263 } else {
3264 st = po->stats;
3265 st.tp_packets += st.tp_drops;
3266 data = &st;
3267 }
3268 memset(&po->stats, 0, sizeof(st));
3269 spin_unlock_bh(&sk->sk_receive_queue.lock);
3270 break;
3271 case PACKET_AUXDATA:
3272 if (len > sizeof(int))
3273 len = sizeof(int);
3274 val = po->auxdata;
3275
3276 data = &val;
3277 break;
3278 case PACKET_ORIGDEV:
3279 if (len > sizeof(int))
3280 len = sizeof(int);
3281 val = po->origdev;
3282
3283 data = &val;
3284 break;
3285 case PACKET_VNET_HDR:
3286 if (len > sizeof(int))
3287 len = sizeof(int);
3288 val = po->has_vnet_hdr;
3289
3290 data = &val;
3291 break;
3292 case PACKET_VERSION:
3293 if (len > sizeof(int))
3294 len = sizeof(int);
3295 val = po->tp_version;
3296 data = &val;
3297 break;
3298 case PACKET_HDRLEN:
3299 if (len > sizeof(int))
3300 len = sizeof(int);
3301 if (copy_from_user(&val, optval, len))
3302 return -EFAULT;
3303 switch (val) {
3304 case TPACKET_V1:
3305 val = sizeof(struct tpacket_hdr);
3306 break;
3307 case TPACKET_V2:
3308 val = sizeof(struct tpacket2_hdr);
3309 break;
3310 case TPACKET_V3:
3311 val = sizeof(struct tpacket3_hdr);
3312 break;
3313 default:
3314 return -EINVAL;
3315 }
3316 data = &val;
3317 break;
3318 case PACKET_RESERVE:
3319 if (len > sizeof(unsigned int))
3320 len = sizeof(unsigned int);
3321 val = po->tp_reserve;
3322 data = &val;
3323 break;
3324 case PACKET_LOSS:
3325 if (len > sizeof(unsigned int))
3326 len = sizeof(unsigned int);
3327 val = po->tp_loss;
3328 data = &val;
3329 break;
3330 case PACKET_TIMESTAMP:
3331 if (len > sizeof(int))
3332 len = sizeof(int);
3333 val = po->tp_tstamp;
3334 data = &val;
3335 break;
3336 case PACKET_FANOUT:
3337 if (len > sizeof(int))
3338 len = sizeof(int);
3339 val = (po->fanout ?
3340 ((u32)po->fanout->id |
3341 ((u32)po->fanout->type << 16)) :
3342 0);
3343 data = &val;
3344 break;
3345 default:
3346 return -ENOPROTOOPT;
3347 }
3348
3349 if (put_user(len, optlen))
3350 return -EFAULT;
3351 if (copy_to_user(optval, data, len))
3352 return -EFAULT;
3353 return 0;
3354 }
3355
3356
3357 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3358 {
3359 struct sock *sk;
3360 struct hlist_node *node;
3361 struct net_device *dev = data;
3362 struct net *net = dev_net(dev);
3363
3364 rcu_read_lock();
3365 sk_for_each_rcu(sk, node, &net->packet.sklist) {
3366 struct packet_sock *po = pkt_sk(sk);
3367
3368 switch (msg) {
3369 case NETDEV_UNREGISTER:
3370 if (po->mclist)
3371 packet_dev_mclist(dev, po->mclist, -1);
3372 /* fallthrough */
3373
3374 case NETDEV_DOWN:
3375 if (dev->ifindex == po->ifindex) {
3376 spin_lock(&po->bind_lock);
3377 if (po->running) {
3378 __unregister_prot_hook(sk, false);
3379 sk->sk_err = ENETDOWN;
3380 if (!sock_flag(sk, SOCK_DEAD))
3381 sk->sk_error_report(sk);
3382 }
3383 if (msg == NETDEV_UNREGISTER) {
3384 po->ifindex = -1;
3385 if (po->prot_hook.dev)
3386 dev_put(po->prot_hook.dev);
3387 po->prot_hook.dev = NULL;
3388 }
3389 spin_unlock(&po->bind_lock);
3390 }
3391 break;
3392 case NETDEV_UP:
3393 if (dev->ifindex == po->ifindex) {
3394 spin_lock(&po->bind_lock);
3395 if (po->num)
3396 register_prot_hook(sk);
3397 spin_unlock(&po->bind_lock);
3398 }
3399 break;
3400 }
3401 }
3402 rcu_read_unlock();
3403 return NOTIFY_DONE;
3404 }
3405
3406
3407 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3408 unsigned long arg)
3409 {
3410 struct sock *sk = sock->sk;
3411
3412 switch (cmd) {
3413 case SIOCOUTQ:
3414 {
3415 int amount = sk_wmem_alloc_get(sk);
3416
3417 return put_user(amount, (int __user *)arg);
3418 }
3419 case SIOCINQ:
3420 {
3421 struct sk_buff *skb;
3422 int amount = 0;
3423
3424 spin_lock_bh(&sk->sk_receive_queue.lock);
3425 skb = skb_peek(&sk->sk_receive_queue);
3426 if (skb)
3427 amount = skb->len;
3428 spin_unlock_bh(&sk->sk_receive_queue.lock);
3429 return put_user(amount, (int __user *)arg);
3430 }
3431 case SIOCGSTAMP:
3432 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3433 case SIOCGSTAMPNS:
3434 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3435
3436 #ifdef CONFIG_INET
3437 case SIOCADDRT:
3438 case SIOCDELRT:
3439 case SIOCDARP:
3440 case SIOCGARP:
3441 case SIOCSARP:
3442 case SIOCGIFADDR:
3443 case SIOCSIFADDR:
3444 case SIOCGIFBRDADDR:
3445 case SIOCSIFBRDADDR:
3446 case SIOCGIFNETMASK:
3447 case SIOCSIFNETMASK:
3448 case SIOCGIFDSTADDR:
3449 case SIOCSIFDSTADDR:
3450 case SIOCSIFFLAGS:
3451 return inet_dgram_ops.ioctl(sock, cmd, arg);
3452 #endif
3453
3454 default:
3455 return -ENOIOCTLCMD;
3456 }
3457 return 0;
3458 }
3459
3460 static unsigned int packet_poll(struct file *file, struct socket *sock,
3461 poll_table *wait)
3462 {
3463 struct sock *sk = sock->sk;
3464 struct packet_sock *po = pkt_sk(sk);
3465 unsigned int mask = datagram_poll(file, sock, wait);
3466
3467 spin_lock_bh(&sk->sk_receive_queue.lock);
3468 if (po->rx_ring.pg_vec) {
3469 if (!packet_previous_rx_frame(po, &po->rx_ring,
3470 TP_STATUS_KERNEL))
3471 mask |= POLLIN | POLLRDNORM;
3472 }
3473 spin_unlock_bh(&sk->sk_receive_queue.lock);
3474 spin_lock_bh(&sk->sk_write_queue.lock);
3475 if (po->tx_ring.pg_vec) {
3476 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3477 mask |= POLLOUT | POLLWRNORM;
3478 }
3479 spin_unlock_bh(&sk->sk_write_queue.lock);
3480 return mask;
3481 }
3482
3483
3484 /* Dirty? Well, I still did not learn better way to account
3485 * for user mmaps.
3486 */
3487
3488 static void packet_mm_open(struct vm_area_struct *vma)
3489 {
3490 struct file *file = vma->vm_file;
3491 struct socket *sock = file->private_data;
3492 struct sock *sk = sock->sk;
3493
3494 if (sk)
3495 atomic_inc(&pkt_sk(sk)->mapped);
3496 }
3497
3498 static void packet_mm_close(struct vm_area_struct *vma)
3499 {
3500 struct file *file = vma->vm_file;
3501 struct socket *sock = file->private_data;
3502 struct sock *sk = sock->sk;
3503
3504 if (sk)
3505 atomic_dec(&pkt_sk(sk)->mapped);
3506 }
3507
3508 static const struct vm_operations_struct packet_mmap_ops = {
3509 .open = packet_mm_open,
3510 .close = packet_mm_close,
3511 };
3512
3513 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3514 unsigned int len)
3515 {
3516 int i;
3517
3518 for (i = 0; i < len; i++) {
3519 if (likely(pg_vec[i].buffer)) {
3520 if (is_vmalloc_addr(pg_vec[i].buffer))
3521 vfree(pg_vec[i].buffer);
3522 else
3523 free_pages((unsigned long)pg_vec[i].buffer,
3524 order);
3525 pg_vec[i].buffer = NULL;
3526 }
3527 }
3528 kfree(pg_vec);
3529 }
3530
3531 static inline char *alloc_one_pg_vec_page(unsigned long order)
3532 {
3533 char *buffer = NULL;
3534 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3535 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3536
3537 buffer = (char *) __get_free_pages(gfp_flags, order);
3538
3539 if (buffer)
3540 return buffer;
3541
3542 /*
3543 * __get_free_pages failed, fall back to vmalloc
3544 */
3545 buffer = vzalloc((1 << order) * PAGE_SIZE);
3546
3547 if (buffer)
3548 return buffer;
3549
3550 /*
3551 * vmalloc failed, lets dig into swap here
3552 */
3553 gfp_flags &= ~__GFP_NORETRY;
3554 buffer = (char *)__get_free_pages(gfp_flags, order);
3555 if (buffer)
3556 return buffer;
3557
3558 /*
3559 * complete and utter failure
3560 */
3561 return NULL;
3562 }
3563
3564 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3565 {
3566 unsigned int block_nr = req->tp_block_nr;
3567 struct pgv *pg_vec;
3568 int i;
3569
3570 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3571 if (unlikely(!pg_vec))
3572 goto out;
3573
3574 for (i = 0; i < block_nr; i++) {
3575 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3576 if (unlikely(!pg_vec[i].buffer))
3577 goto out_free_pgvec;
3578 }
3579
3580 out:
3581 return pg_vec;
3582
3583 out_free_pgvec:
3584 free_pg_vec(pg_vec, order, block_nr);
3585 pg_vec = NULL;
3586 goto out;
3587 }
3588
3589 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3590 int closing, int tx_ring)
3591 {
3592 struct pgv *pg_vec = NULL;
3593 struct packet_sock *po = pkt_sk(sk);
3594 int was_running, order = 0;
3595 struct packet_ring_buffer *rb;
3596 struct sk_buff_head *rb_queue;
3597 __be16 num;
3598 int err = -EINVAL;
3599 /* Added to avoid minimal code churn */
3600 struct tpacket_req *req = &req_u->req;
3601
3602 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3603 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3604 WARN(1, "Tx-ring is not supported.\n");
3605 goto out;
3606 }
3607
3608 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3609 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3610
3611 err = -EBUSY;
3612 if (!closing) {
3613 if (atomic_read(&po->mapped))
3614 goto out;
3615 if (atomic_read(&rb->pending))
3616 goto out;
3617 }
3618
3619 if (req->tp_block_nr) {
3620 /* Sanity tests and some calculations */
3621 err = -EBUSY;
3622 if (unlikely(rb->pg_vec))
3623 goto out;
3624
3625 switch (po->tp_version) {
3626 case TPACKET_V1:
3627 po->tp_hdrlen = TPACKET_HDRLEN;
3628 break;
3629 case TPACKET_V2:
3630 po->tp_hdrlen = TPACKET2_HDRLEN;
3631 break;
3632 case TPACKET_V3:
3633 po->tp_hdrlen = TPACKET3_HDRLEN;
3634 break;
3635 }
3636
3637 err = -EINVAL;
3638 if (unlikely((int)req->tp_block_size <= 0))
3639 goto out;
3640 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3641 goto out;
3642 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3643 po->tp_reserve))
3644 goto out;
3645 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3646 goto out;
3647
3648 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3649 if (unlikely(rb->frames_per_block <= 0))
3650 goto out;
3651 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3652 req->tp_frame_nr))
3653 goto out;
3654
3655 err = -ENOMEM;
3656 order = get_order(req->tp_block_size);
3657 pg_vec = alloc_pg_vec(req, order);
3658 if (unlikely(!pg_vec))
3659 goto out;
3660 switch (po->tp_version) {
3661 case TPACKET_V3:
3662 /* Transmit path is not supported. We checked
3663 * it above but just being paranoid
3664 */
3665 if (!tx_ring)
3666 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3667 break;
3668 default:
3669 break;
3670 }
3671 }
3672 /* Done */
3673 else {
3674 err = -EINVAL;
3675 if (unlikely(req->tp_frame_nr))
3676 goto out;
3677 }
3678
3679 lock_sock(sk);
3680
3681 /* Detach socket from network */
3682 spin_lock(&po->bind_lock);
3683 was_running = po->running;
3684 num = po->num;
3685 if (was_running) {
3686 po->num = 0;
3687 __unregister_prot_hook(sk, false);
3688 }
3689 spin_unlock(&po->bind_lock);
3690
3691 synchronize_net();
3692
3693 err = -EBUSY;
3694 mutex_lock(&po->pg_vec_lock);
3695 if (closing || atomic_read(&po->mapped) == 0) {
3696 err = 0;
3697 spin_lock_bh(&rb_queue->lock);
3698 swap(rb->pg_vec, pg_vec);
3699 rb->frame_max = (req->tp_frame_nr - 1);
3700 rb->head = 0;
3701 rb->frame_size = req->tp_frame_size;
3702 spin_unlock_bh(&rb_queue->lock);
3703
3704 swap(rb->pg_vec_order, order);
3705 swap(rb->pg_vec_len, req->tp_block_nr);
3706
3707 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3708 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3709 tpacket_rcv : packet_rcv;
3710 skb_queue_purge(rb_queue);
3711 if (atomic_read(&po->mapped))
3712 pr_err("packet_mmap: vma is busy: %d\n",
3713 atomic_read(&po->mapped));
3714 }
3715 mutex_unlock(&po->pg_vec_lock);
3716
3717 spin_lock(&po->bind_lock);
3718 if (was_running) {
3719 po->num = num;
3720 register_prot_hook(sk);
3721 }
3722 spin_unlock(&po->bind_lock);
3723 if (closing && (po->tp_version > TPACKET_V2)) {
3724 /* Because we don't support block-based V3 on tx-ring */
3725 if (!tx_ring)
3726 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3727 }
3728 release_sock(sk);
3729
3730 if (pg_vec)
3731 free_pg_vec(pg_vec, order, req->tp_block_nr);
3732 out:
3733 return err;
3734 }
3735
3736 static int packet_mmap(struct file *file, struct socket *sock,
3737 struct vm_area_struct *vma)
3738 {
3739 struct sock *sk = sock->sk;
3740 struct packet_sock *po = pkt_sk(sk);
3741 unsigned long size, expected_size;
3742 struct packet_ring_buffer *rb;
3743 unsigned long start;
3744 int err = -EINVAL;
3745 int i;
3746
3747 if (vma->vm_pgoff)
3748 return -EINVAL;
3749
3750 mutex_lock(&po->pg_vec_lock);
3751
3752 expected_size = 0;
3753 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3754 if (rb->pg_vec) {
3755 expected_size += rb->pg_vec_len
3756 * rb->pg_vec_pages
3757 * PAGE_SIZE;
3758 }
3759 }
3760
3761 if (expected_size == 0)
3762 goto out;
3763
3764 size = vma->vm_end - vma->vm_start;
3765 if (size != expected_size)
3766 goto out;
3767
3768 start = vma->vm_start;
3769 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3770 if (rb->pg_vec == NULL)
3771 continue;
3772
3773 for (i = 0; i < rb->pg_vec_len; i++) {
3774 struct page *page;
3775 void *kaddr = rb->pg_vec[i].buffer;
3776 int pg_num;
3777
3778 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3779 page = pgv_to_page(kaddr);
3780 err = vm_insert_page(vma, start, page);
3781 if (unlikely(err))
3782 goto out;
3783 start += PAGE_SIZE;
3784 kaddr += PAGE_SIZE;
3785 }
3786 }
3787 }
3788
3789 atomic_inc(&po->mapped);
3790 vma->vm_ops = &packet_mmap_ops;
3791 err = 0;
3792
3793 out:
3794 mutex_unlock(&po->pg_vec_lock);
3795 return err;
3796 }
3797
3798 static const struct proto_ops packet_ops_spkt = {
3799 .family = PF_PACKET,
3800 .owner = THIS_MODULE,
3801 .release = packet_release,
3802 .bind = packet_bind_spkt,
3803 .connect = sock_no_connect,
3804 .socketpair = sock_no_socketpair,
3805 .accept = sock_no_accept,
3806 .getname = packet_getname_spkt,
3807 .poll = datagram_poll,
3808 .ioctl = packet_ioctl,
3809 .listen = sock_no_listen,
3810 .shutdown = sock_no_shutdown,
3811 .setsockopt = sock_no_setsockopt,
3812 .getsockopt = sock_no_getsockopt,
3813 .sendmsg = packet_sendmsg_spkt,
3814 .recvmsg = packet_recvmsg,
3815 .mmap = sock_no_mmap,
3816 .sendpage = sock_no_sendpage,
3817 };
3818
3819 static const struct proto_ops packet_ops = {
3820 .family = PF_PACKET,
3821 .owner = THIS_MODULE,
3822 .release = packet_release,
3823 .bind = packet_bind,
3824 .connect = sock_no_connect,
3825 .socketpair = sock_no_socketpair,
3826 .accept = sock_no_accept,
3827 .getname = packet_getname,
3828 .poll = packet_poll,
3829 .ioctl = packet_ioctl,
3830 .listen = sock_no_listen,
3831 .shutdown = sock_no_shutdown,
3832 .setsockopt = packet_setsockopt,
3833 .getsockopt = packet_getsockopt,
3834 .sendmsg = packet_sendmsg,
3835 .recvmsg = packet_recvmsg,
3836 .mmap = packet_mmap,
3837 .sendpage = sock_no_sendpage,
3838 };
3839
3840 static const struct net_proto_family packet_family_ops = {
3841 .family = PF_PACKET,
3842 .create = packet_create,
3843 .owner = THIS_MODULE,
3844 };
3845
3846 static struct notifier_block packet_netdev_notifier = {
3847 .notifier_call = packet_notifier,
3848 };
3849
3850 #ifdef CONFIG_PROC_FS
3851
3852 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3853 __acquires(RCU)
3854 {
3855 struct net *net = seq_file_net(seq);
3856
3857 rcu_read_lock();
3858 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3859 }
3860
3861 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3862 {
3863 struct net *net = seq_file_net(seq);
3864 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3865 }
3866
3867 static void packet_seq_stop(struct seq_file *seq, void *v)
3868 __releases(RCU)
3869 {
3870 rcu_read_unlock();
3871 }
3872
3873 static int packet_seq_show(struct seq_file *seq, void *v)
3874 {
3875 if (v == SEQ_START_TOKEN)
3876 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3877 else {
3878 struct sock *s = sk_entry(v);
3879 const struct packet_sock *po = pkt_sk(s);
3880
3881 seq_printf(seq,
3882 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
3883 s,
3884 atomic_read(&s->sk_refcnt),
3885 s->sk_type,
3886 ntohs(po->num),
3887 po->ifindex,
3888 po->running,
3889 atomic_read(&s->sk_rmem_alloc),
3890 sock_i_uid(s),
3891 sock_i_ino(s));
3892 }
3893
3894 return 0;
3895 }
3896
3897 static const struct seq_operations packet_seq_ops = {
3898 .start = packet_seq_start,
3899 .next = packet_seq_next,
3900 .stop = packet_seq_stop,
3901 .show = packet_seq_show,
3902 };
3903
3904 static int packet_seq_open(struct inode *inode, struct file *file)
3905 {
3906 return seq_open_net(inode, file, &packet_seq_ops,
3907 sizeof(struct seq_net_private));
3908 }
3909
3910 static const struct file_operations packet_seq_fops = {
3911 .owner = THIS_MODULE,
3912 .open = packet_seq_open,
3913 .read = seq_read,
3914 .llseek = seq_lseek,
3915 .release = seq_release_net,
3916 };
3917
3918 #endif
3919
3920 static int __net_init packet_net_init(struct net *net)
3921 {
3922 spin_lock_init(&net->packet.sklist_lock);
3923 INIT_HLIST_HEAD(&net->packet.sklist);
3924
3925 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3926 return -ENOMEM;
3927
3928 return 0;
3929 }
3930
3931 static void __net_exit packet_net_exit(struct net *net)
3932 {
3933 proc_net_remove(net, "packet");
3934 }
3935
3936 static struct pernet_operations packet_net_ops = {
3937 .init = packet_net_init,
3938 .exit = packet_net_exit,
3939 };
3940
3941
3942 static void __exit packet_exit(void)
3943 {
3944 unregister_netdevice_notifier(&packet_netdev_notifier);
3945 unregister_pernet_subsys(&packet_net_ops);
3946 sock_unregister(PF_PACKET);
3947 proto_unregister(&packet_proto);
3948 }
3949
3950 static int __init packet_init(void)
3951 {
3952 int rc = proto_register(&packet_proto, 0);
3953
3954 if (rc != 0)
3955 goto out;
3956
3957 sock_register(&packet_family_ops);
3958 register_pernet_subsys(&packet_net_ops);
3959 register_netdevice_notifier(&packet_netdev_notifier);
3960 out:
3961 return rc;
3962 }
3963
3964 module_init(packet_init);
3965 module_exit(packet_exit);
3966 MODULE_LICENSE("GPL");
3967 MODULE_ALIAS_NETPROTO(PF_PACKET);
Ubuntu Jammy Linux kernel mirror