Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * | |
02c30a84 | 8 | * Authors: Ross Biro |
1da177e4 LT |
9 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
10 | * Alan Cox, <gw4pts@gw4pts.ampr.org> | |
11 | * | |
1ce4f28b | 12 | * Fixes: |
1da177e4 LT |
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. | |
1ce4f28b | 35 | * Ulises Alonso : Frame number limit removal and |
1da177e4 | 36 | * packet_set_ring memory leak. |
0fb375fb EB |
37 | * Eric Biederman : Allow for > 8 byte hardware addresses. |
38 | * The convention is that longer addresses | |
39 | * will simply extend the hardware address | |
1ce4f28b | 40 | * byte arrays at the end of sockaddr_ll |
0fb375fb | 41 | * and packet_mreq. |
69e3c75f | 42 | * Johann Baudy : Added TX RING. |
f6fb8f10 | 43 | * Chetan Loke : Implemented TPACKET_V3 block abstraction |
44 | * layer. | |
45 | * Copyright (C) 2011, <lokec@ccs.neu.edu> | |
46 | * | |
1da177e4 LT |
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 | */ | |
1ce4f28b | 54 | |
1da177e4 | 55 | #include <linux/types.h> |
1da177e4 | 56 | #include <linux/mm.h> |
4fc268d2 | 57 | #include <linux/capability.h> |
1da177e4 LT |
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> | |
ffbc6111 | 65 | #include <linux/kernel.h> |
1da177e4 | 66 | #include <linux/kmod.h> |
5a0e3ad6 | 67 | #include <linux/slab.h> |
0e3125c7 | 68 | #include <linux/vmalloc.h> |
457c4cbc | 69 | #include <net/net_namespace.h> |
1da177e4 LT |
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> | |
1da177e4 LT |
76 | #include <asm/uaccess.h> |
77 | #include <asm/ioctls.h> | |
78 | #include <asm/page.h> | |
a1f8e7f7 | 79 | #include <asm/cacheflush.h> |
1da177e4 LT |
80 | #include <asm/io.h> |
81 | #include <linux/proc_fs.h> | |
82 | #include <linux/seq_file.h> | |
83 | #include <linux/poll.h> | |
84 | #include <linux/module.h> | |
85 | #include <linux/init.h> | |
905db440 | 86 | #include <linux/mutex.h> |
05423b24 | 87 | #include <linux/if_vlan.h> |
bfd5f4a3 | 88 | #include <linux/virtio_net.h> |
ed85b565 | 89 | #include <linux/errqueue.h> |
614f60fa | 90 | #include <linux/net_tstamp.h> |
1da177e4 LT |
91 | |
92 | #ifdef CONFIG_INET | |
93 | #include <net/inet_common.h> | |
94 | #endif | |
95 | ||
2787b04b PE |
96 | #include "internal.h" |
97 | ||
1da177e4 LT |
98 | /* |
99 | Assumptions: | |
100 | - if device has no dev->hard_header routine, it adds and removes ll header | |
101 | inside itself. In this case ll header is invisible outside of device, | |
102 | but higher levels still should reserve dev->hard_header_len. | |
103 | Some devices are enough clever to reallocate skb, when header | |
104 | will not fit to reserved space (tunnel), another ones are silly | |
105 | (PPP). | |
106 | - packet socket receives packets with pulled ll header, | |
107 | so that SOCK_RAW should push it back. | |
108 | ||
109 | On receive: | |
110 | ----------- | |
111 | ||
112 | Incoming, dev->hard_header!=NULL | |
b0e380b1 ACM |
113 | mac_header -> ll header |
114 | data -> data | |
1da177e4 LT |
115 | |
116 | Outgoing, dev->hard_header!=NULL | |
b0e380b1 ACM |
117 | mac_header -> ll header |
118 | data -> ll header | |
1da177e4 LT |
119 | |
120 | Incoming, dev->hard_header==NULL | |
b0e380b1 ACM |
121 | mac_header -> UNKNOWN position. It is very likely, that it points to ll |
122 | header. PPP makes it, that is wrong, because introduce | |
db0c58f9 | 123 | assymetry between rx and tx paths. |
b0e380b1 | 124 | data -> data |
1da177e4 LT |
125 | |
126 | Outgoing, dev->hard_header==NULL | |
b0e380b1 ACM |
127 | mac_header -> data. ll header is still not built! |
128 | data -> data | |
1da177e4 LT |
129 | |
130 | Resume | |
131 | If dev->hard_header==NULL we are unlikely to restore sensible ll header. | |
132 | ||
133 | ||
134 | On transmit: | |
135 | ------------ | |
136 | ||
137 | dev->hard_header != NULL | |
b0e380b1 ACM |
138 | mac_header -> ll header |
139 | data -> ll header | |
1da177e4 LT |
140 | |
141 | dev->hard_header == NULL (ll header is added by device, we cannot control it) | |
b0e380b1 ACM |
142 | mac_header -> data |
143 | data -> data | |
1da177e4 LT |
144 | |
145 | We should set nh.raw on output to correct posistion, | |
146 | packet classifier depends on it. | |
147 | */ | |
148 | ||
1da177e4 LT |
149 | /* Private packet socket structures. */ |
150 | ||
0fb375fb EB |
151 | /* identical to struct packet_mreq except it has |
152 | * a longer address field. | |
153 | */ | |
40d4e3df | 154 | struct packet_mreq_max { |
0fb375fb EB |
155 | int mr_ifindex; |
156 | unsigned short mr_type; | |
157 | unsigned short mr_alen; | |
158 | unsigned char mr_address[MAX_ADDR_LEN]; | |
1da177e4 | 159 | }; |
a2efcfa0 | 160 | |
184f489e DB |
161 | union tpacket_uhdr { |
162 | struct tpacket_hdr *h1; | |
163 | struct tpacket2_hdr *h2; | |
164 | struct tpacket3_hdr *h3; | |
165 | void *raw; | |
166 | }; | |
167 | ||
f6fb8f10 | 168 | static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, |
69e3c75f JB |
169 | int closing, int tx_ring); |
170 | ||
f6fb8f10 | 171 | #define V3_ALIGNMENT (8) |
172 | ||
bc59ba39 | 173 | #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) |
f6fb8f10 | 174 | |
175 | #define BLK_PLUS_PRIV(sz_of_priv) \ | |
176 | (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) | |
177 | ||
f6fb8f10 | 178 | #define PGV_FROM_VMALLOC 1 |
69e3c75f | 179 | |
f6fb8f10 | 180 | #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) |
181 | #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) | |
182 | #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) | |
183 | #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) | |
184 | #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) | |
185 | #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) | |
186 | #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x))) | |
187 | ||
69e3c75f JB |
188 | struct packet_sock; |
189 | static int tpacket_snd(struct packet_sock *po, struct msghdr *msg); | |
77f65ebd WB |
190 | static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, |
191 | struct packet_type *pt, struct net_device *orig_dev); | |
1da177e4 | 192 | |
f6fb8f10 | 193 | static void *packet_previous_frame(struct packet_sock *po, |
194 | struct packet_ring_buffer *rb, | |
195 | int status); | |
196 | static void packet_increment_head(struct packet_ring_buffer *buff); | |
bc59ba39 | 197 | static int prb_curr_blk_in_use(struct tpacket_kbdq_core *, |
198 | struct tpacket_block_desc *); | |
199 | static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, | |
f6fb8f10 | 200 | struct packet_sock *); |
bc59ba39 | 201 | static void prb_retire_current_block(struct tpacket_kbdq_core *, |
f6fb8f10 | 202 | struct packet_sock *, unsigned int status); |
bc59ba39 | 203 | static int prb_queue_frozen(struct tpacket_kbdq_core *); |
204 | static void prb_open_block(struct tpacket_kbdq_core *, | |
205 | struct tpacket_block_desc *); | |
f6fb8f10 | 206 | static void prb_retire_rx_blk_timer_expired(unsigned long); |
bc59ba39 | 207 | static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); |
208 | static void prb_init_blk_timer(struct packet_sock *, | |
209 | struct tpacket_kbdq_core *, | |
210 | void (*func) (unsigned long)); | |
211 | static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); | |
212 | static void prb_clear_rxhash(struct tpacket_kbdq_core *, | |
213 | struct tpacket3_hdr *); | |
214 | static void prb_fill_vlan_info(struct tpacket_kbdq_core *, | |
215 | struct tpacket3_hdr *); | |
1da177e4 LT |
216 | static void packet_flush_mclist(struct sock *sk); |
217 | ||
ffbc6111 HX |
218 | struct packet_skb_cb { |
219 | unsigned int origlen; | |
220 | union { | |
221 | struct sockaddr_pkt pkt; | |
222 | struct sockaddr_ll ll; | |
223 | } sa; | |
224 | }; | |
225 | ||
226 | #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) | |
8dc41944 | 227 | |
bc59ba39 | 228 | #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) |
f6fb8f10 | 229 | #define GET_PBLOCK_DESC(x, bid) \ |
bc59ba39 | 230 | ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) |
f6fb8f10 | 231 | #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ |
bc59ba39 | 232 | ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) |
f6fb8f10 | 233 | #define GET_NEXT_PRB_BLK_NUM(x) \ |
234 | (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ | |
235 | ((x)->kactive_blk_num+1) : 0) | |
236 | ||
dc99f600 DM |
237 | static void __fanout_unlink(struct sock *sk, struct packet_sock *po); |
238 | static void __fanout_link(struct sock *sk, struct packet_sock *po); | |
239 | ||
ce06b03e DM |
240 | /* register_prot_hook must be invoked with the po->bind_lock held, |
241 | * or from a context in which asynchronous accesses to the packet | |
242 | * socket is not possible (packet_create()). | |
243 | */ | |
244 | static void register_prot_hook(struct sock *sk) | |
245 | { | |
246 | struct packet_sock *po = pkt_sk(sk); | |
247 | if (!po->running) { | |
dc99f600 DM |
248 | if (po->fanout) |
249 | __fanout_link(sk, po); | |
250 | else | |
251 | dev_add_pack(&po->prot_hook); | |
ce06b03e DM |
252 | sock_hold(sk); |
253 | po->running = 1; | |
254 | } | |
255 | } | |
256 | ||
257 | /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock | |
258 | * held. If the sync parameter is true, we will temporarily drop | |
259 | * the po->bind_lock and do a synchronize_net to make sure no | |
260 | * asynchronous packet processing paths still refer to the elements | |
261 | * of po->prot_hook. If the sync parameter is false, it is the | |
262 | * callers responsibility to take care of this. | |
263 | */ | |
264 | static void __unregister_prot_hook(struct sock *sk, bool sync) | |
265 | { | |
266 | struct packet_sock *po = pkt_sk(sk); | |
267 | ||
268 | po->running = 0; | |
dc99f600 DM |
269 | if (po->fanout) |
270 | __fanout_unlink(sk, po); | |
271 | else | |
272 | __dev_remove_pack(&po->prot_hook); | |
ce06b03e DM |
273 | __sock_put(sk); |
274 | ||
275 | if (sync) { | |
276 | spin_unlock(&po->bind_lock); | |
277 | synchronize_net(); | |
278 | spin_lock(&po->bind_lock); | |
279 | } | |
280 | } | |
281 | ||
282 | static void unregister_prot_hook(struct sock *sk, bool sync) | |
283 | { | |
284 | struct packet_sock *po = pkt_sk(sk); | |
285 | ||
286 | if (po->running) | |
287 | __unregister_prot_hook(sk, sync); | |
288 | } | |
289 | ||
f6dafa95 | 290 | static inline __pure struct page *pgv_to_page(void *addr) |
0af55bb5 CG |
291 | { |
292 | if (is_vmalloc_addr(addr)) | |
293 | return vmalloc_to_page(addr); | |
294 | return virt_to_page(addr); | |
295 | } | |
296 | ||
69e3c75f | 297 | static void __packet_set_status(struct packet_sock *po, void *frame, int status) |
1da177e4 | 298 | { |
184f489e | 299 | union tpacket_uhdr h; |
1da177e4 | 300 | |
69e3c75f | 301 | h.raw = frame; |
bbd6ef87 PM |
302 | switch (po->tp_version) { |
303 | case TPACKET_V1: | |
69e3c75f | 304 | h.h1->tp_status = status; |
0af55bb5 | 305 | flush_dcache_page(pgv_to_page(&h.h1->tp_status)); |
bbd6ef87 PM |
306 | break; |
307 | case TPACKET_V2: | |
69e3c75f | 308 | h.h2->tp_status = status; |
0af55bb5 | 309 | flush_dcache_page(pgv_to_page(&h.h2->tp_status)); |
bbd6ef87 | 310 | break; |
f6fb8f10 | 311 | case TPACKET_V3: |
69e3c75f | 312 | default: |
f6fb8f10 | 313 | WARN(1, "TPACKET version not supported.\n"); |
69e3c75f | 314 | BUG(); |
bbd6ef87 | 315 | } |
69e3c75f JB |
316 | |
317 | smp_wmb(); | |
bbd6ef87 PM |
318 | } |
319 | ||
69e3c75f | 320 | static int __packet_get_status(struct packet_sock *po, void *frame) |
bbd6ef87 | 321 | { |
184f489e | 322 | union tpacket_uhdr h; |
bbd6ef87 | 323 | |
69e3c75f JB |
324 | smp_rmb(); |
325 | ||
bbd6ef87 PM |
326 | h.raw = frame; |
327 | switch (po->tp_version) { | |
328 | case TPACKET_V1: | |
0af55bb5 | 329 | flush_dcache_page(pgv_to_page(&h.h1->tp_status)); |
69e3c75f | 330 | return h.h1->tp_status; |
bbd6ef87 | 331 | case TPACKET_V2: |
0af55bb5 | 332 | flush_dcache_page(pgv_to_page(&h.h2->tp_status)); |
69e3c75f | 333 | return h.h2->tp_status; |
f6fb8f10 | 334 | case TPACKET_V3: |
69e3c75f | 335 | default: |
f6fb8f10 | 336 | WARN(1, "TPACKET version not supported.\n"); |
69e3c75f JB |
337 | BUG(); |
338 | return 0; | |
bbd6ef87 | 339 | } |
1da177e4 | 340 | } |
69e3c75f JB |
341 | |
342 | static void *packet_lookup_frame(struct packet_sock *po, | |
343 | struct packet_ring_buffer *rb, | |
344 | unsigned int position, | |
345 | int status) | |
346 | { | |
347 | unsigned int pg_vec_pos, frame_offset; | |
184f489e | 348 | union tpacket_uhdr h; |
69e3c75f JB |
349 | |
350 | pg_vec_pos = position / rb->frames_per_block; | |
351 | frame_offset = position % rb->frames_per_block; | |
352 | ||
0e3125c7 NH |
353 | h.raw = rb->pg_vec[pg_vec_pos].buffer + |
354 | (frame_offset * rb->frame_size); | |
69e3c75f JB |
355 | |
356 | if (status != __packet_get_status(po, h.raw)) | |
357 | return NULL; | |
358 | ||
359 | return h.raw; | |
360 | } | |
361 | ||
eea49cc9 | 362 | static void *packet_current_frame(struct packet_sock *po, |
69e3c75f JB |
363 | struct packet_ring_buffer *rb, |
364 | int status) | |
365 | { | |
366 | return packet_lookup_frame(po, rb, rb->head, status); | |
367 | } | |
368 | ||
bc59ba39 | 369 | static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 370 | { |
371 | del_timer_sync(&pkc->retire_blk_timer); | |
372 | } | |
373 | ||
374 | static void prb_shutdown_retire_blk_timer(struct packet_sock *po, | |
375 | int tx_ring, | |
376 | struct sk_buff_head *rb_queue) | |
377 | { | |
bc59ba39 | 378 | struct tpacket_kbdq_core *pkc; |
f6fb8f10 | 379 | |
380 | pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc; | |
381 | ||
382 | spin_lock(&rb_queue->lock); | |
383 | pkc->delete_blk_timer = 1; | |
384 | spin_unlock(&rb_queue->lock); | |
385 | ||
386 | prb_del_retire_blk_timer(pkc); | |
387 | } | |
388 | ||
389 | static void prb_init_blk_timer(struct packet_sock *po, | |
bc59ba39 | 390 | struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 391 | void (*func) (unsigned long)) |
392 | { | |
393 | init_timer(&pkc->retire_blk_timer); | |
394 | pkc->retire_blk_timer.data = (long)po; | |
395 | pkc->retire_blk_timer.function = func; | |
396 | pkc->retire_blk_timer.expires = jiffies; | |
397 | } | |
398 | ||
399 | static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring) | |
400 | { | |
bc59ba39 | 401 | struct tpacket_kbdq_core *pkc; |
f6fb8f10 | 402 | |
403 | if (tx_ring) | |
404 | BUG(); | |
405 | ||
406 | pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc; | |
407 | prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired); | |
408 | } | |
409 | ||
410 | static int prb_calc_retire_blk_tmo(struct packet_sock *po, | |
411 | int blk_size_in_bytes) | |
412 | { | |
413 | struct net_device *dev; | |
414 | unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; | |
4bc71cb9 JP |
415 | struct ethtool_cmd ecmd; |
416 | int err; | |
e440cf2c | 417 | u32 speed; |
f6fb8f10 | 418 | |
4bc71cb9 JP |
419 | rtnl_lock(); |
420 | dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); | |
421 | if (unlikely(!dev)) { | |
422 | rtnl_unlock(); | |
f6fb8f10 | 423 | return DEFAULT_PRB_RETIRE_TOV; |
4bc71cb9 JP |
424 | } |
425 | err = __ethtool_get_settings(dev, &ecmd); | |
e440cf2c | 426 | speed = ethtool_cmd_speed(&ecmd); |
4bc71cb9 JP |
427 | rtnl_unlock(); |
428 | if (!err) { | |
4bc71cb9 JP |
429 | /* |
430 | * If the link speed is so slow you don't really | |
431 | * need to worry about perf anyways | |
432 | */ | |
e440cf2c | 433 | if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) { |
4bc71cb9 | 434 | return DEFAULT_PRB_RETIRE_TOV; |
e440cf2c | 435 | } else { |
436 | msec = 1; | |
437 | div = speed / 1000; | |
f6fb8f10 | 438 | } |
439 | } | |
440 | ||
441 | mbits = (blk_size_in_bytes * 8) / (1024 * 1024); | |
442 | ||
443 | if (div) | |
444 | mbits /= div; | |
445 | ||
446 | tmo = mbits * msec; | |
447 | ||
448 | if (div) | |
449 | return tmo+1; | |
450 | return tmo; | |
451 | } | |
452 | ||
bc59ba39 | 453 | static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, |
f6fb8f10 | 454 | union tpacket_req_u *req_u) |
455 | { | |
456 | p1->feature_req_word = req_u->req3.tp_feature_req_word; | |
457 | } | |
458 | ||
459 | static void init_prb_bdqc(struct packet_sock *po, | |
460 | struct packet_ring_buffer *rb, | |
461 | struct pgv *pg_vec, | |
462 | union tpacket_req_u *req_u, int tx_ring) | |
463 | { | |
bc59ba39 | 464 | struct tpacket_kbdq_core *p1 = &rb->prb_bdqc; |
465 | struct tpacket_block_desc *pbd; | |
f6fb8f10 | 466 | |
467 | memset(p1, 0x0, sizeof(*p1)); | |
468 | ||
469 | p1->knxt_seq_num = 1; | |
470 | p1->pkbdq = pg_vec; | |
bc59ba39 | 471 | pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; |
e3192690 | 472 | p1->pkblk_start = pg_vec[0].buffer; |
f6fb8f10 | 473 | p1->kblk_size = req_u->req3.tp_block_size; |
474 | p1->knum_blocks = req_u->req3.tp_block_nr; | |
475 | p1->hdrlen = po->tp_hdrlen; | |
476 | p1->version = po->tp_version; | |
477 | p1->last_kactive_blk_num = 0; | |
478 | po->stats_u.stats3.tp_freeze_q_cnt = 0; | |
479 | if (req_u->req3.tp_retire_blk_tov) | |
480 | p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; | |
481 | else | |
482 | p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, | |
483 | req_u->req3.tp_block_size); | |
484 | p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); | |
485 | p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; | |
486 | ||
487 | prb_init_ft_ops(p1, req_u); | |
488 | prb_setup_retire_blk_timer(po, tx_ring); | |
489 | prb_open_block(p1, pbd); | |
490 | } | |
491 | ||
492 | /* Do NOT update the last_blk_num first. | |
493 | * Assumes sk_buff_head lock is held. | |
494 | */ | |
bc59ba39 | 495 | static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 496 | { |
497 | mod_timer(&pkc->retire_blk_timer, | |
498 | jiffies + pkc->tov_in_jiffies); | |
499 | pkc->last_kactive_blk_num = pkc->kactive_blk_num; | |
500 | } | |
501 | ||
502 | /* | |
503 | * Timer logic: | |
504 | * 1) We refresh the timer only when we open a block. | |
505 | * By doing this we don't waste cycles refreshing the timer | |
506 | * on packet-by-packet basis. | |
507 | * | |
508 | * With a 1MB block-size, on a 1Gbps line, it will take | |
509 | * i) ~8 ms to fill a block + ii) memcpy etc. | |
510 | * In this cut we are not accounting for the memcpy time. | |
511 | * | |
512 | * So, if the user sets the 'tmo' to 10ms then the timer | |
513 | * will never fire while the block is still getting filled | |
514 | * (which is what we want). However, the user could choose | |
515 | * to close a block early and that's fine. | |
516 | * | |
517 | * But when the timer does fire, we check whether or not to refresh it. | |
518 | * Since the tmo granularity is in msecs, it is not too expensive | |
519 | * to refresh the timer, lets say every '8' msecs. | |
520 | * Either the user can set the 'tmo' or we can derive it based on | |
521 | * a) line-speed and b) block-size. | |
522 | * prb_calc_retire_blk_tmo() calculates the tmo. | |
523 | * | |
524 | */ | |
525 | static void prb_retire_rx_blk_timer_expired(unsigned long data) | |
526 | { | |
527 | struct packet_sock *po = (struct packet_sock *)data; | |
bc59ba39 | 528 | struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc; |
f6fb8f10 | 529 | unsigned int frozen; |
bc59ba39 | 530 | struct tpacket_block_desc *pbd; |
f6fb8f10 | 531 | |
532 | spin_lock(&po->sk.sk_receive_queue.lock); | |
533 | ||
534 | frozen = prb_queue_frozen(pkc); | |
535 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); | |
536 | ||
537 | if (unlikely(pkc->delete_blk_timer)) | |
538 | goto out; | |
539 | ||
540 | /* We only need to plug the race when the block is partially filled. | |
541 | * tpacket_rcv: | |
542 | * lock(); increment BLOCK_NUM_PKTS; unlock() | |
543 | * copy_bits() is in progress ... | |
544 | * timer fires on other cpu: | |
545 | * we can't retire the current block because copy_bits | |
546 | * is in progress. | |
547 | * | |
548 | */ | |
549 | if (BLOCK_NUM_PKTS(pbd)) { | |
550 | while (atomic_read(&pkc->blk_fill_in_prog)) { | |
551 | /* Waiting for skb_copy_bits to finish... */ | |
552 | cpu_relax(); | |
553 | } | |
554 | } | |
555 | ||
556 | if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { | |
557 | if (!frozen) { | |
558 | prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); | |
559 | if (!prb_dispatch_next_block(pkc, po)) | |
560 | goto refresh_timer; | |
561 | else | |
562 | goto out; | |
563 | } else { | |
564 | /* Case 1. Queue was frozen because user-space was | |
565 | * lagging behind. | |
566 | */ | |
567 | if (prb_curr_blk_in_use(pkc, pbd)) { | |
568 | /* | |
569 | * Ok, user-space is still behind. | |
570 | * So just refresh the timer. | |
571 | */ | |
572 | goto refresh_timer; | |
573 | } else { | |
574 | /* Case 2. queue was frozen,user-space caught up, | |
575 | * now the link went idle && the timer fired. | |
576 | * We don't have a block to close.So we open this | |
577 | * block and restart the timer. | |
578 | * opening a block thaws the queue,restarts timer | |
579 | * Thawing/timer-refresh is a side effect. | |
580 | */ | |
581 | prb_open_block(pkc, pbd); | |
582 | goto out; | |
583 | } | |
584 | } | |
585 | } | |
586 | ||
587 | refresh_timer: | |
588 | _prb_refresh_rx_retire_blk_timer(pkc); | |
589 | ||
590 | out: | |
591 | spin_unlock(&po->sk.sk_receive_queue.lock); | |
592 | } | |
593 | ||
eea49cc9 | 594 | static void prb_flush_block(struct tpacket_kbdq_core *pkc1, |
bc59ba39 | 595 | struct tpacket_block_desc *pbd1, __u32 status) |
f6fb8f10 | 596 | { |
597 | /* Flush everything minus the block header */ | |
598 | ||
599 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 | |
600 | u8 *start, *end; | |
601 | ||
602 | start = (u8 *)pbd1; | |
603 | ||
604 | /* Skip the block header(we know header WILL fit in 4K) */ | |
605 | start += PAGE_SIZE; | |
606 | ||
607 | end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); | |
608 | for (; start < end; start += PAGE_SIZE) | |
609 | flush_dcache_page(pgv_to_page(start)); | |
610 | ||
611 | smp_wmb(); | |
612 | #endif | |
613 | ||
614 | /* Now update the block status. */ | |
615 | ||
616 | BLOCK_STATUS(pbd1) = status; | |
617 | ||
618 | /* Flush the block header */ | |
619 | ||
620 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 | |
621 | start = (u8 *)pbd1; | |
622 | flush_dcache_page(pgv_to_page(start)); | |
623 | ||
624 | smp_wmb(); | |
625 | #endif | |
626 | } | |
627 | ||
628 | /* | |
629 | * Side effect: | |
630 | * | |
631 | * 1) flush the block | |
632 | * 2) Increment active_blk_num | |
633 | * | |
634 | * Note:We DONT refresh the timer on purpose. | |
635 | * Because almost always the next block will be opened. | |
636 | */ | |
bc59ba39 | 637 | static void prb_close_block(struct tpacket_kbdq_core *pkc1, |
638 | struct tpacket_block_desc *pbd1, | |
f6fb8f10 | 639 | struct packet_sock *po, unsigned int stat) |
640 | { | |
641 | __u32 status = TP_STATUS_USER | stat; | |
642 | ||
643 | struct tpacket3_hdr *last_pkt; | |
bc59ba39 | 644 | struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; |
f6fb8f10 | 645 | |
646 | if (po->stats.tp_drops) | |
647 | status |= TP_STATUS_LOSING; | |
648 | ||
649 | last_pkt = (struct tpacket3_hdr *)pkc1->prev; | |
650 | last_pkt->tp_next_offset = 0; | |
651 | ||
652 | /* Get the ts of the last pkt */ | |
653 | if (BLOCK_NUM_PKTS(pbd1)) { | |
654 | h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; | |
655 | h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; | |
656 | } else { | |
657 | /* Ok, we tmo'd - so get the current time */ | |
658 | struct timespec ts; | |
659 | getnstimeofday(&ts); | |
660 | h1->ts_last_pkt.ts_sec = ts.tv_sec; | |
661 | h1->ts_last_pkt.ts_nsec = ts.tv_nsec; | |
662 | } | |
663 | ||
664 | smp_wmb(); | |
665 | ||
666 | /* Flush the block */ | |
667 | prb_flush_block(pkc1, pbd1, status); | |
668 | ||
669 | pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); | |
670 | } | |
671 | ||
eea49cc9 | 672 | static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 673 | { |
674 | pkc->reset_pending_on_curr_blk = 0; | |
675 | } | |
676 | ||
677 | /* | |
678 | * Side effect of opening a block: | |
679 | * | |
680 | * 1) prb_queue is thawed. | |
681 | * 2) retire_blk_timer is refreshed. | |
682 | * | |
683 | */ | |
bc59ba39 | 684 | static void prb_open_block(struct tpacket_kbdq_core *pkc1, |
685 | struct tpacket_block_desc *pbd1) | |
f6fb8f10 | 686 | { |
687 | struct timespec ts; | |
bc59ba39 | 688 | struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; |
f6fb8f10 | 689 | |
690 | smp_rmb(); | |
691 | ||
692 | if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) { | |
693 | ||
694 | /* We could have just memset this but we will lose the | |
695 | * flexibility of making the priv area sticky | |
696 | */ | |
697 | BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; | |
698 | BLOCK_NUM_PKTS(pbd1) = 0; | |
699 | BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); | |
700 | getnstimeofday(&ts); | |
701 | h1->ts_first_pkt.ts_sec = ts.tv_sec; | |
702 | h1->ts_first_pkt.ts_nsec = ts.tv_nsec; | |
703 | pkc1->pkblk_start = (char *)pbd1; | |
e3192690 | 704 | pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); |
f6fb8f10 | 705 | BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); |
706 | BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; | |
707 | pbd1->version = pkc1->version; | |
708 | pkc1->prev = pkc1->nxt_offset; | |
709 | pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; | |
710 | prb_thaw_queue(pkc1); | |
711 | _prb_refresh_rx_retire_blk_timer(pkc1); | |
712 | ||
713 | smp_wmb(); | |
714 | ||
715 | return; | |
716 | } | |
717 | ||
718 | WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n", | |
719 | pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num); | |
720 | dump_stack(); | |
721 | BUG(); | |
722 | } | |
723 | ||
724 | /* | |
725 | * Queue freeze logic: | |
726 | * 1) Assume tp_block_nr = 8 blocks. | |
727 | * 2) At time 't0', user opens Rx ring. | |
728 | * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 | |
729 | * 4) user-space is either sleeping or processing block '0'. | |
730 | * 5) tpacket_rcv is currently filling block '7', since there is no space left, | |
731 | * it will close block-7,loop around and try to fill block '0'. | |
732 | * call-flow: | |
733 | * __packet_lookup_frame_in_block | |
734 | * prb_retire_current_block() | |
735 | * prb_dispatch_next_block() | |
736 | * |->(BLOCK_STATUS == USER) evaluates to true | |
737 | * 5.1) Since block-0 is currently in-use, we just freeze the queue. | |
738 | * 6) Now there are two cases: | |
739 | * 6.1) Link goes idle right after the queue is frozen. | |
740 | * But remember, the last open_block() refreshed the timer. | |
741 | * When this timer expires,it will refresh itself so that we can | |
742 | * re-open block-0 in near future. | |
743 | * 6.2) Link is busy and keeps on receiving packets. This is a simple | |
744 | * case and __packet_lookup_frame_in_block will check if block-0 | |
745 | * is free and can now be re-used. | |
746 | */ | |
eea49cc9 | 747 | static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 748 | struct packet_sock *po) |
749 | { | |
750 | pkc->reset_pending_on_curr_blk = 1; | |
751 | po->stats_u.stats3.tp_freeze_q_cnt++; | |
752 | } | |
753 | ||
754 | #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) | |
755 | ||
756 | /* | |
757 | * If the next block is free then we will dispatch it | |
758 | * and return a good offset. | |
759 | * Else, we will freeze the queue. | |
760 | * So, caller must check the return value. | |
761 | */ | |
bc59ba39 | 762 | static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 763 | struct packet_sock *po) |
764 | { | |
bc59ba39 | 765 | struct tpacket_block_desc *pbd; |
f6fb8f10 | 766 | |
767 | smp_rmb(); | |
768 | ||
769 | /* 1. Get current block num */ | |
770 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); | |
771 | ||
772 | /* 2. If this block is currently in_use then freeze the queue */ | |
773 | if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { | |
774 | prb_freeze_queue(pkc, po); | |
775 | return NULL; | |
776 | } | |
777 | ||
778 | /* | |
779 | * 3. | |
780 | * open this block and return the offset where the first packet | |
781 | * needs to get stored. | |
782 | */ | |
783 | prb_open_block(pkc, pbd); | |
784 | return (void *)pkc->nxt_offset; | |
785 | } | |
786 | ||
bc59ba39 | 787 | static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 788 | struct packet_sock *po, unsigned int status) |
789 | { | |
bc59ba39 | 790 | struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); |
f6fb8f10 | 791 | |
792 | /* retire/close the current block */ | |
793 | if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { | |
794 | /* | |
795 | * Plug the case where copy_bits() is in progress on | |
796 | * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't | |
797 | * have space to copy the pkt in the current block and | |
798 | * called prb_retire_current_block() | |
799 | * | |
800 | * We don't need to worry about the TMO case because | |
801 | * the timer-handler already handled this case. | |
802 | */ | |
803 | if (!(status & TP_STATUS_BLK_TMO)) { | |
804 | while (atomic_read(&pkc->blk_fill_in_prog)) { | |
805 | /* Waiting for skb_copy_bits to finish... */ | |
806 | cpu_relax(); | |
807 | } | |
808 | } | |
809 | prb_close_block(pkc, pbd, po, status); | |
810 | return; | |
811 | } | |
812 | ||
813 | WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd); | |
814 | dump_stack(); | |
815 | BUG(); | |
816 | } | |
817 | ||
eea49cc9 | 818 | static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc, |
bc59ba39 | 819 | struct tpacket_block_desc *pbd) |
f6fb8f10 | 820 | { |
821 | return TP_STATUS_USER & BLOCK_STATUS(pbd); | |
822 | } | |
823 | ||
eea49cc9 | 824 | static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 825 | { |
826 | return pkc->reset_pending_on_curr_blk; | |
827 | } | |
828 | ||
eea49cc9 | 829 | static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) |
f6fb8f10 | 830 | { |
bc59ba39 | 831 | struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); |
f6fb8f10 | 832 | atomic_dec(&pkc->blk_fill_in_prog); |
833 | } | |
834 | ||
eea49cc9 | 835 | static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 836 | struct tpacket3_hdr *ppd) |
837 | { | |
838 | ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb); | |
839 | } | |
840 | ||
eea49cc9 | 841 | static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 842 | struct tpacket3_hdr *ppd) |
843 | { | |
844 | ppd->hv1.tp_rxhash = 0; | |
845 | } | |
846 | ||
eea49cc9 | 847 | static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 848 | struct tpacket3_hdr *ppd) |
849 | { | |
850 | if (vlan_tx_tag_present(pkc->skb)) { | |
851 | ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb); | |
852 | ppd->tp_status = TP_STATUS_VLAN_VALID; | |
853 | } else { | |
9e67030a | 854 | ppd->hv1.tp_vlan_tci = 0; |
855 | ppd->tp_status = TP_STATUS_AVAILABLE; | |
f6fb8f10 | 856 | } |
857 | } | |
858 | ||
bc59ba39 | 859 | static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 860 | struct tpacket3_hdr *ppd) |
861 | { | |
862 | prb_fill_vlan_info(pkc, ppd); | |
863 | ||
864 | if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) | |
865 | prb_fill_rxhash(pkc, ppd); | |
866 | else | |
867 | prb_clear_rxhash(pkc, ppd); | |
868 | } | |
869 | ||
eea49cc9 | 870 | static void prb_fill_curr_block(char *curr, |
bc59ba39 | 871 | struct tpacket_kbdq_core *pkc, |
872 | struct tpacket_block_desc *pbd, | |
f6fb8f10 | 873 | unsigned int len) |
874 | { | |
875 | struct tpacket3_hdr *ppd; | |
876 | ||
877 | ppd = (struct tpacket3_hdr *)curr; | |
878 | ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); | |
879 | pkc->prev = curr; | |
880 | pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); | |
881 | BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); | |
882 | BLOCK_NUM_PKTS(pbd) += 1; | |
883 | atomic_inc(&pkc->blk_fill_in_prog); | |
884 | prb_run_all_ft_ops(pkc, ppd); | |
885 | } | |
886 | ||
887 | /* Assumes caller has the sk->rx_queue.lock */ | |
888 | static void *__packet_lookup_frame_in_block(struct packet_sock *po, | |
889 | struct sk_buff *skb, | |
890 | int status, | |
891 | unsigned int len | |
892 | ) | |
893 | { | |
bc59ba39 | 894 | struct tpacket_kbdq_core *pkc; |
895 | struct tpacket_block_desc *pbd; | |
f6fb8f10 | 896 | char *curr, *end; |
897 | ||
e3192690 | 898 | pkc = GET_PBDQC_FROM_RB(&po->rx_ring); |
f6fb8f10 | 899 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); |
900 | ||
901 | /* Queue is frozen when user space is lagging behind */ | |
902 | if (prb_queue_frozen(pkc)) { | |
903 | /* | |
904 | * Check if that last block which caused the queue to freeze, | |
905 | * is still in_use by user-space. | |
906 | */ | |
907 | if (prb_curr_blk_in_use(pkc, pbd)) { | |
908 | /* Can't record this packet */ | |
909 | return NULL; | |
910 | } else { | |
911 | /* | |
912 | * Ok, the block was released by user-space. | |
913 | * Now let's open that block. | |
914 | * opening a block also thaws the queue. | |
915 | * Thawing is a side effect. | |
916 | */ | |
917 | prb_open_block(pkc, pbd); | |
918 | } | |
919 | } | |
920 | ||
921 | smp_mb(); | |
922 | curr = pkc->nxt_offset; | |
923 | pkc->skb = skb; | |
e3192690 | 924 | end = (char *)pbd + pkc->kblk_size; |
f6fb8f10 | 925 | |
926 | /* first try the current block */ | |
927 | if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { | |
928 | prb_fill_curr_block(curr, pkc, pbd, len); | |
929 | return (void *)curr; | |
930 | } | |
931 | ||
932 | /* Ok, close the current block */ | |
933 | prb_retire_current_block(pkc, po, 0); | |
934 | ||
935 | /* Now, try to dispatch the next block */ | |
936 | curr = (char *)prb_dispatch_next_block(pkc, po); | |
937 | if (curr) { | |
938 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); | |
939 | prb_fill_curr_block(curr, pkc, pbd, len); | |
940 | return (void *)curr; | |
941 | } | |
942 | ||
943 | /* | |
944 | * No free blocks are available.user_space hasn't caught up yet. | |
945 | * Queue was just frozen and now this packet will get dropped. | |
946 | */ | |
947 | return NULL; | |
948 | } | |
949 | ||
eea49cc9 | 950 | static void *packet_current_rx_frame(struct packet_sock *po, |
f6fb8f10 | 951 | struct sk_buff *skb, |
952 | int status, unsigned int len) | |
953 | { | |
954 | char *curr = NULL; | |
955 | switch (po->tp_version) { | |
956 | case TPACKET_V1: | |
957 | case TPACKET_V2: | |
958 | curr = packet_lookup_frame(po, &po->rx_ring, | |
959 | po->rx_ring.head, status); | |
960 | return curr; | |
961 | case TPACKET_V3: | |
962 | return __packet_lookup_frame_in_block(po, skb, status, len); | |
963 | default: | |
964 | WARN(1, "TPACKET version not supported\n"); | |
965 | BUG(); | |
99aa3473 | 966 | return NULL; |
f6fb8f10 | 967 | } |
968 | } | |
969 | ||
eea49cc9 | 970 | static void *prb_lookup_block(struct packet_sock *po, |
f6fb8f10 | 971 | struct packet_ring_buffer *rb, |
77f65ebd | 972 | unsigned int idx, |
f6fb8f10 | 973 | int status) |
974 | { | |
bc59ba39 | 975 | struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); |
77f65ebd | 976 | struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); |
f6fb8f10 | 977 | |
978 | if (status != BLOCK_STATUS(pbd)) | |
979 | return NULL; | |
980 | return pbd; | |
981 | } | |
982 | ||
eea49cc9 | 983 | static int prb_previous_blk_num(struct packet_ring_buffer *rb) |
f6fb8f10 | 984 | { |
985 | unsigned int prev; | |
986 | if (rb->prb_bdqc.kactive_blk_num) | |
987 | prev = rb->prb_bdqc.kactive_blk_num-1; | |
988 | else | |
989 | prev = rb->prb_bdqc.knum_blocks-1; | |
990 | return prev; | |
991 | } | |
992 | ||
993 | /* Assumes caller has held the rx_queue.lock */ | |
eea49cc9 | 994 | static void *__prb_previous_block(struct packet_sock *po, |
f6fb8f10 | 995 | struct packet_ring_buffer *rb, |
996 | int status) | |
997 | { | |
998 | unsigned int previous = prb_previous_blk_num(rb); | |
999 | return prb_lookup_block(po, rb, previous, status); | |
1000 | } | |
1001 | ||
eea49cc9 | 1002 | static void *packet_previous_rx_frame(struct packet_sock *po, |
f6fb8f10 | 1003 | struct packet_ring_buffer *rb, |
1004 | int status) | |
1005 | { | |
1006 | if (po->tp_version <= TPACKET_V2) | |
1007 | return packet_previous_frame(po, rb, status); | |
1008 | ||
1009 | return __prb_previous_block(po, rb, status); | |
1010 | } | |
1011 | ||
eea49cc9 | 1012 | static void packet_increment_rx_head(struct packet_sock *po, |
f6fb8f10 | 1013 | struct packet_ring_buffer *rb) |
1014 | { | |
1015 | switch (po->tp_version) { | |
1016 | case TPACKET_V1: | |
1017 | case TPACKET_V2: | |
1018 | return packet_increment_head(rb); | |
1019 | case TPACKET_V3: | |
1020 | default: | |
1021 | WARN(1, "TPACKET version not supported.\n"); | |
1022 | BUG(); | |
1023 | return; | |
1024 | } | |
1025 | } | |
1026 | ||
eea49cc9 | 1027 | static void *packet_previous_frame(struct packet_sock *po, |
69e3c75f JB |
1028 | struct packet_ring_buffer *rb, |
1029 | int status) | |
1030 | { | |
1031 | unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; | |
1032 | return packet_lookup_frame(po, rb, previous, status); | |
1033 | } | |
1034 | ||
eea49cc9 | 1035 | static void packet_increment_head(struct packet_ring_buffer *buff) |
69e3c75f JB |
1036 | { |
1037 | buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; | |
1038 | } | |
1039 | ||
77f65ebd WB |
1040 | static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) |
1041 | { | |
1042 | struct sock *sk = &po->sk; | |
1043 | bool has_room; | |
1044 | ||
1045 | if (po->prot_hook.func != tpacket_rcv) | |
1046 | return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize) | |
1047 | <= sk->sk_rcvbuf; | |
1048 | ||
1049 | spin_lock(&sk->sk_receive_queue.lock); | |
1050 | if (po->tp_version == TPACKET_V3) | |
1051 | has_room = prb_lookup_block(po, &po->rx_ring, | |
1052 | po->rx_ring.prb_bdqc.kactive_blk_num, | |
1053 | TP_STATUS_KERNEL); | |
1054 | else | |
1055 | has_room = packet_lookup_frame(po, &po->rx_ring, | |
1056 | po->rx_ring.head, | |
1057 | TP_STATUS_KERNEL); | |
1058 | spin_unlock(&sk->sk_receive_queue.lock); | |
1059 | ||
1060 | return has_room; | |
1061 | } | |
1062 | ||
1da177e4 LT |
1063 | static void packet_sock_destruct(struct sock *sk) |
1064 | { | |
ed85b565 RC |
1065 | skb_queue_purge(&sk->sk_error_queue); |
1066 | ||
547b792c IJ |
1067 | WARN_ON(atomic_read(&sk->sk_rmem_alloc)); |
1068 | WARN_ON(atomic_read(&sk->sk_wmem_alloc)); | |
1da177e4 LT |
1069 | |
1070 | if (!sock_flag(sk, SOCK_DEAD)) { | |
40d4e3df | 1071 | pr_err("Attempt to release alive packet socket: %p\n", sk); |
1da177e4 LT |
1072 | return; |
1073 | } | |
1074 | ||
17ab56a2 | 1075 | sk_refcnt_debug_dec(sk); |
1da177e4 LT |
1076 | } |
1077 | ||
dc99f600 DM |
1078 | static int fanout_rr_next(struct packet_fanout *f, unsigned int num) |
1079 | { | |
1080 | int x = atomic_read(&f->rr_cur) + 1; | |
1081 | ||
1082 | if (x >= num) | |
1083 | x = 0; | |
1084 | ||
1085 | return x; | |
1086 | } | |
1087 | ||
77f65ebd WB |
1088 | static unsigned int fanout_demux_hash(struct packet_fanout *f, |
1089 | struct sk_buff *skb, | |
1090 | unsigned int num) | |
dc99f600 | 1091 | { |
77f65ebd | 1092 | return (((u64)skb->rxhash) * num) >> 32; |
dc99f600 DM |
1093 | } |
1094 | ||
77f65ebd WB |
1095 | static unsigned int fanout_demux_lb(struct packet_fanout *f, |
1096 | struct sk_buff *skb, | |
1097 | unsigned int num) | |
dc99f600 DM |
1098 | { |
1099 | int cur, old; | |
1100 | ||
1101 | cur = atomic_read(&f->rr_cur); | |
1102 | while ((old = atomic_cmpxchg(&f->rr_cur, cur, | |
1103 | fanout_rr_next(f, num))) != cur) | |
1104 | cur = old; | |
77f65ebd WB |
1105 | return cur; |
1106 | } | |
1107 | ||
1108 | static unsigned int fanout_demux_cpu(struct packet_fanout *f, | |
1109 | struct sk_buff *skb, | |
1110 | unsigned int num) | |
1111 | { | |
1112 | return smp_processor_id() % num; | |
dc99f600 DM |
1113 | } |
1114 | ||
77f65ebd WB |
1115 | static unsigned int fanout_demux_rollover(struct packet_fanout *f, |
1116 | struct sk_buff *skb, | |
1117 | unsigned int idx, unsigned int skip, | |
1118 | unsigned int num) | |
95ec3eb4 | 1119 | { |
77f65ebd | 1120 | unsigned int i, j; |
95ec3eb4 | 1121 | |
77f65ebd WB |
1122 | i = j = min_t(int, f->next[idx], num - 1); |
1123 | do { | |
1124 | if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) { | |
1125 | if (i != j) | |
1126 | f->next[idx] = i; | |
1127 | return i; | |
1128 | } | |
1129 | if (++i == num) | |
1130 | i = 0; | |
1131 | } while (i != j); | |
1132 | ||
1133 | return idx; | |
1134 | } | |
1135 | ||
1136 | static bool fanout_has_flag(struct packet_fanout *f, u16 flag) | |
1137 | { | |
1138 | return f->flags & (flag >> 8); | |
95ec3eb4 DM |
1139 | } |
1140 | ||
95ec3eb4 DM |
1141 | static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, |
1142 | struct packet_type *pt, struct net_device *orig_dev) | |
dc99f600 DM |
1143 | { |
1144 | struct packet_fanout *f = pt->af_packet_priv; | |
1145 | unsigned int num = f->num_members; | |
1146 | struct packet_sock *po; | |
77f65ebd | 1147 | unsigned int idx; |
dc99f600 DM |
1148 | |
1149 | if (!net_eq(dev_net(dev), read_pnet(&f->net)) || | |
1150 | !num) { | |
1151 | kfree_skb(skb); | |
1152 | return 0; | |
1153 | } | |
1154 | ||
95ec3eb4 DM |
1155 | switch (f->type) { |
1156 | case PACKET_FANOUT_HASH: | |
1157 | default: | |
77f65ebd | 1158 | if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { |
bc416d97 | 1159 | skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET); |
95ec3eb4 DM |
1160 | if (!skb) |
1161 | return 0; | |
1162 | } | |
1163 | skb_get_rxhash(skb); | |
77f65ebd | 1164 | idx = fanout_demux_hash(f, skb, num); |
95ec3eb4 DM |
1165 | break; |
1166 | case PACKET_FANOUT_LB: | |
77f65ebd | 1167 | idx = fanout_demux_lb(f, skb, num); |
95ec3eb4 DM |
1168 | break; |
1169 | case PACKET_FANOUT_CPU: | |
77f65ebd WB |
1170 | idx = fanout_demux_cpu(f, skb, num); |
1171 | break; | |
1172 | case PACKET_FANOUT_ROLLOVER: | |
1173 | idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num); | |
95ec3eb4 | 1174 | break; |
dc99f600 DM |
1175 | } |
1176 | ||
77f65ebd WB |
1177 | po = pkt_sk(f->arr[idx]); |
1178 | if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) && | |
1179 | unlikely(!packet_rcv_has_room(po, skb))) { | |
1180 | idx = fanout_demux_rollover(f, skb, idx, idx, num); | |
1181 | po = pkt_sk(f->arr[idx]); | |
1182 | } | |
dc99f600 DM |
1183 | |
1184 | return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); | |
1185 | } | |
1186 | ||
fff3321d PE |
1187 | DEFINE_MUTEX(fanout_mutex); |
1188 | EXPORT_SYMBOL_GPL(fanout_mutex); | |
dc99f600 DM |
1189 | static LIST_HEAD(fanout_list); |
1190 | ||
1191 | static void __fanout_link(struct sock *sk, struct packet_sock *po) | |
1192 | { | |
1193 | struct packet_fanout *f = po->fanout; | |
1194 | ||
1195 | spin_lock(&f->lock); | |
1196 | f->arr[f->num_members] = sk; | |
1197 | smp_wmb(); | |
1198 | f->num_members++; | |
1199 | spin_unlock(&f->lock); | |
1200 | } | |
1201 | ||
1202 | static void __fanout_unlink(struct sock *sk, struct packet_sock *po) | |
1203 | { | |
1204 | struct packet_fanout *f = po->fanout; | |
1205 | int i; | |
1206 | ||
1207 | spin_lock(&f->lock); | |
1208 | for (i = 0; i < f->num_members; i++) { | |
1209 | if (f->arr[i] == sk) | |
1210 | break; | |
1211 | } | |
1212 | BUG_ON(i >= f->num_members); | |
1213 | f->arr[i] = f->arr[f->num_members - 1]; | |
1214 | f->num_members--; | |
1215 | spin_unlock(&f->lock); | |
1216 | } | |
1217 | ||
a0dfb263 | 1218 | static bool match_fanout_group(struct packet_type *ptype, struct sock * sk) |
c0de08d0 EL |
1219 | { |
1220 | if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout) | |
1221 | return true; | |
1222 | ||
1223 | return false; | |
1224 | } | |
1225 | ||
7736d33f | 1226 | static int fanout_add(struct sock *sk, u16 id, u16 type_flags) |
dc99f600 DM |
1227 | { |
1228 | struct packet_sock *po = pkt_sk(sk); | |
1229 | struct packet_fanout *f, *match; | |
7736d33f | 1230 | u8 type = type_flags & 0xff; |
77f65ebd | 1231 | u8 flags = type_flags >> 8; |
dc99f600 DM |
1232 | int err; |
1233 | ||
1234 | switch (type) { | |
77f65ebd WB |
1235 | case PACKET_FANOUT_ROLLOVER: |
1236 | if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) | |
1237 | return -EINVAL; | |
dc99f600 DM |
1238 | case PACKET_FANOUT_HASH: |
1239 | case PACKET_FANOUT_LB: | |
95ec3eb4 | 1240 | case PACKET_FANOUT_CPU: |
dc99f600 DM |
1241 | break; |
1242 | default: | |
1243 | return -EINVAL; | |
1244 | } | |
1245 | ||
1246 | if (!po->running) | |
1247 | return -EINVAL; | |
1248 | ||
1249 | if (po->fanout) | |
1250 | return -EALREADY; | |
1251 | ||
1252 | mutex_lock(&fanout_mutex); | |
1253 | match = NULL; | |
1254 | list_for_each_entry(f, &fanout_list, list) { | |
1255 | if (f->id == id && | |
1256 | read_pnet(&f->net) == sock_net(sk)) { | |
1257 | match = f; | |
1258 | break; | |
1259 | } | |
1260 | } | |
afe62c68 | 1261 | err = -EINVAL; |
77f65ebd | 1262 | if (match && match->flags != flags) |
afe62c68 | 1263 | goto out; |
dc99f600 | 1264 | if (!match) { |
afe62c68 | 1265 | err = -ENOMEM; |
dc99f600 | 1266 | match = kzalloc(sizeof(*match), GFP_KERNEL); |
afe62c68 ED |
1267 | if (!match) |
1268 | goto out; | |
1269 | write_pnet(&match->net, sock_net(sk)); | |
1270 | match->id = id; | |
1271 | match->type = type; | |
77f65ebd | 1272 | match->flags = flags; |
afe62c68 ED |
1273 | atomic_set(&match->rr_cur, 0); |
1274 | INIT_LIST_HEAD(&match->list); | |
1275 | spin_lock_init(&match->lock); | |
1276 | atomic_set(&match->sk_ref, 0); | |
1277 | match->prot_hook.type = po->prot_hook.type; | |
1278 | match->prot_hook.dev = po->prot_hook.dev; | |
1279 | match->prot_hook.func = packet_rcv_fanout; | |
1280 | match->prot_hook.af_packet_priv = match; | |
c0de08d0 | 1281 | match->prot_hook.id_match = match_fanout_group; |
afe62c68 ED |
1282 | dev_add_pack(&match->prot_hook); |
1283 | list_add(&match->list, &fanout_list); | |
dc99f600 | 1284 | } |
afe62c68 ED |
1285 | err = -EINVAL; |
1286 | if (match->type == type && | |
1287 | match->prot_hook.type == po->prot_hook.type && | |
1288 | match->prot_hook.dev == po->prot_hook.dev) { | |
1289 | err = -ENOSPC; | |
1290 | if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) { | |
1291 | __dev_remove_pack(&po->prot_hook); | |
1292 | po->fanout = match; | |
1293 | atomic_inc(&match->sk_ref); | |
1294 | __fanout_link(sk, po); | |
1295 | err = 0; | |
dc99f600 DM |
1296 | } |
1297 | } | |
afe62c68 | 1298 | out: |
dc99f600 DM |
1299 | mutex_unlock(&fanout_mutex); |
1300 | return err; | |
1301 | } | |
1302 | ||
1303 | static void fanout_release(struct sock *sk) | |
1304 | { | |
1305 | struct packet_sock *po = pkt_sk(sk); | |
1306 | struct packet_fanout *f; | |
1307 | ||
1308 | f = po->fanout; | |
1309 | if (!f) | |
1310 | return; | |
1311 | ||
fff3321d | 1312 | mutex_lock(&fanout_mutex); |
dc99f600 DM |
1313 | po->fanout = NULL; |
1314 | ||
dc99f600 DM |
1315 | if (atomic_dec_and_test(&f->sk_ref)) { |
1316 | list_del(&f->list); | |
1317 | dev_remove_pack(&f->prot_hook); | |
1318 | kfree(f); | |
1319 | } | |
1320 | mutex_unlock(&fanout_mutex); | |
1321 | } | |
1da177e4 | 1322 | |
90ddc4f0 | 1323 | static const struct proto_ops packet_ops; |
1da177e4 | 1324 | |
90ddc4f0 | 1325 | static const struct proto_ops packet_ops_spkt; |
1da177e4 | 1326 | |
40d4e3df ED |
1327 | static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, |
1328 | struct packet_type *pt, struct net_device *orig_dev) | |
1da177e4 LT |
1329 | { |
1330 | struct sock *sk; | |
1331 | struct sockaddr_pkt *spkt; | |
1332 | ||
1333 | /* | |
1334 | * When we registered the protocol we saved the socket in the data | |
1335 | * field for just this event. | |
1336 | */ | |
1337 | ||
1338 | sk = pt->af_packet_priv; | |
1ce4f28b | 1339 | |
1da177e4 LT |
1340 | /* |
1341 | * Yank back the headers [hope the device set this | |
1342 | * right or kerboom...] | |
1343 | * | |
1344 | * Incoming packets have ll header pulled, | |
1345 | * push it back. | |
1346 | * | |
98e399f8 | 1347 | * For outgoing ones skb->data == skb_mac_header(skb) |
1da177e4 LT |
1348 | * so that this procedure is noop. |
1349 | */ | |
1350 | ||
1351 | if (skb->pkt_type == PACKET_LOOPBACK) | |
1352 | goto out; | |
1353 | ||
09ad9bc7 | 1354 | if (!net_eq(dev_net(dev), sock_net(sk))) |
d12d01d6 DL |
1355 | goto out; |
1356 | ||
40d4e3df ED |
1357 | skb = skb_share_check(skb, GFP_ATOMIC); |
1358 | if (skb == NULL) | |
1da177e4 LT |
1359 | goto oom; |
1360 | ||
1361 | /* drop any routing info */ | |
adf30907 | 1362 | skb_dst_drop(skb); |
1da177e4 | 1363 | |
84531c24 PO |
1364 | /* drop conntrack reference */ |
1365 | nf_reset(skb); | |
1366 | ||
ffbc6111 | 1367 | spkt = &PACKET_SKB_CB(skb)->sa.pkt; |
1da177e4 | 1368 | |
98e399f8 | 1369 | skb_push(skb, skb->data - skb_mac_header(skb)); |
1da177e4 LT |
1370 | |
1371 | /* | |
1372 | * The SOCK_PACKET socket receives _all_ frames. | |
1373 | */ | |
1374 | ||
1375 | spkt->spkt_family = dev->type; | |
1376 | strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); | |
1377 | spkt->spkt_protocol = skb->protocol; | |
1378 | ||
1379 | /* | |
1380 | * Charge the memory to the socket. This is done specifically | |
1381 | * to prevent sockets using all the memory up. | |
1382 | */ | |
1383 | ||
40d4e3df | 1384 | if (sock_queue_rcv_skb(sk, skb) == 0) |
1da177e4 LT |
1385 | return 0; |
1386 | ||
1387 | out: | |
1388 | kfree_skb(skb); | |
1389 | oom: | |
1390 | return 0; | |
1391 | } | |
1392 | ||
1393 | ||
1394 | /* | |
1395 | * Output a raw packet to a device layer. This bypasses all the other | |
1396 | * protocol layers and you must therefore supply it with a complete frame | |
1397 | */ | |
1ce4f28b | 1398 | |
1da177e4 LT |
1399 | static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, |
1400 | struct msghdr *msg, size_t len) | |
1401 | { | |
1402 | struct sock *sk = sock->sk; | |
40d4e3df | 1403 | struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name; |
1a35ca80 | 1404 | struct sk_buff *skb = NULL; |
1da177e4 | 1405 | struct net_device *dev; |
40d4e3df | 1406 | __be16 proto = 0; |
1da177e4 | 1407 | int err; |
3bdc0eba | 1408 | int extra_len = 0; |
1ce4f28b | 1409 | |
1da177e4 | 1410 | /* |
1ce4f28b | 1411 | * Get and verify the address. |
1da177e4 LT |
1412 | */ |
1413 | ||
40d4e3df | 1414 | if (saddr) { |
1da177e4 | 1415 | if (msg->msg_namelen < sizeof(struct sockaddr)) |
40d4e3df ED |
1416 | return -EINVAL; |
1417 | if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) | |
1418 | proto = saddr->spkt_protocol; | |
1419 | } else | |
1420 | return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ | |
1da177e4 LT |
1421 | |
1422 | /* | |
1ce4f28b | 1423 | * Find the device first to size check it |
1da177e4 LT |
1424 | */ |
1425 | ||
de74e92a | 1426 | saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; |
1a35ca80 | 1427 | retry: |
654d1f8a ED |
1428 | rcu_read_lock(); |
1429 | dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); | |
1da177e4 LT |
1430 | err = -ENODEV; |
1431 | if (dev == NULL) | |
1432 | goto out_unlock; | |
1ce4f28b | 1433 | |
d5e76b0a DM |
1434 | err = -ENETDOWN; |
1435 | if (!(dev->flags & IFF_UP)) | |
1436 | goto out_unlock; | |
1437 | ||
1da177e4 | 1438 | /* |
40d4e3df ED |
1439 | * You may not queue a frame bigger than the mtu. This is the lowest level |
1440 | * raw protocol and you must do your own fragmentation at this level. | |
1da177e4 | 1441 | */ |
1ce4f28b | 1442 | |
3bdc0eba BG |
1443 | if (unlikely(sock_flag(sk, SOCK_NOFCS))) { |
1444 | if (!netif_supports_nofcs(dev)) { | |
1445 | err = -EPROTONOSUPPORT; | |
1446 | goto out_unlock; | |
1447 | } | |
1448 | extra_len = 4; /* We're doing our own CRC */ | |
1449 | } | |
1450 | ||
1da177e4 | 1451 | err = -EMSGSIZE; |
3bdc0eba | 1452 | if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) |
1da177e4 LT |
1453 | goto out_unlock; |
1454 | ||
1a35ca80 ED |
1455 | if (!skb) { |
1456 | size_t reserved = LL_RESERVED_SPACE(dev); | |
4ce40912 | 1457 | int tlen = dev->needed_tailroom; |
1a35ca80 ED |
1458 | unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; |
1459 | ||
1460 | rcu_read_unlock(); | |
4ce40912 | 1461 | skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); |
1a35ca80 ED |
1462 | if (skb == NULL) |
1463 | return -ENOBUFS; | |
1464 | /* FIXME: Save some space for broken drivers that write a hard | |
1465 | * header at transmission time by themselves. PPP is the notable | |
1466 | * one here. This should really be fixed at the driver level. | |
1467 | */ | |
1468 | skb_reserve(skb, reserved); | |
1469 | skb_reset_network_header(skb); | |
1470 | ||
1471 | /* Try to align data part correctly */ | |
1472 | if (hhlen) { | |
1473 | skb->data -= hhlen; | |
1474 | skb->tail -= hhlen; | |
1475 | if (len < hhlen) | |
1476 | skb_reset_network_header(skb); | |
1477 | } | |
1478 | err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len); | |
1479 | if (err) | |
1480 | goto out_free; | |
1481 | goto retry; | |
1da177e4 LT |
1482 | } |
1483 | ||
3bdc0eba | 1484 | if (len > (dev->mtu + dev->hard_header_len + extra_len)) { |
57f89bfa BG |
1485 | /* Earlier code assumed this would be a VLAN pkt, |
1486 | * double-check this now that we have the actual | |
1487 | * packet in hand. | |
1488 | */ | |
1489 | struct ethhdr *ehdr; | |
1490 | skb_reset_mac_header(skb); | |
1491 | ehdr = eth_hdr(skb); | |
1492 | if (ehdr->h_proto != htons(ETH_P_8021Q)) { | |
1493 | err = -EMSGSIZE; | |
1494 | goto out_unlock; | |
1495 | } | |
1496 | } | |
1a35ca80 | 1497 | |
1da177e4 LT |
1498 | skb->protocol = proto; |
1499 | skb->dev = dev; | |
1500 | skb->priority = sk->sk_priority; | |
2d37a186 | 1501 | skb->mark = sk->sk_mark; |
bf84a010 DB |
1502 | |
1503 | sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags); | |
1da177e4 | 1504 | |
3bdc0eba BG |
1505 | if (unlikely(extra_len == 4)) |
1506 | skb->no_fcs = 1; | |
1507 | ||
40893fd0 | 1508 | skb_probe_transport_header(skb, 0); |
c1aad275 | 1509 | |
1da177e4 | 1510 | dev_queue_xmit(skb); |
654d1f8a | 1511 | rcu_read_unlock(); |
40d4e3df | 1512 | return len; |
1da177e4 | 1513 | |
1da177e4 | 1514 | out_unlock: |
654d1f8a | 1515 | rcu_read_unlock(); |
1a35ca80 ED |
1516 | out_free: |
1517 | kfree_skb(skb); | |
1da177e4 LT |
1518 | return err; |
1519 | } | |
1da177e4 | 1520 | |
eea49cc9 | 1521 | static unsigned int run_filter(const struct sk_buff *skb, |
62ab0812 | 1522 | const struct sock *sk, |
dbcb5855 | 1523 | unsigned int res) |
1da177e4 LT |
1524 | { |
1525 | struct sk_filter *filter; | |
fda9ef5d | 1526 | |
80f8f102 ED |
1527 | rcu_read_lock(); |
1528 | filter = rcu_dereference(sk->sk_filter); | |
dbcb5855 | 1529 | if (filter != NULL) |
0a14842f | 1530 | res = SK_RUN_FILTER(filter, skb); |
80f8f102 | 1531 | rcu_read_unlock(); |
1da177e4 | 1532 | |
dbcb5855 | 1533 | return res; |
1da177e4 LT |
1534 | } |
1535 | ||
1536 | /* | |
62ab0812 ED |
1537 | * This function makes lazy skb cloning in hope that most of packets |
1538 | * are discarded by BPF. | |
1539 | * | |
1540 | * Note tricky part: we DO mangle shared skb! skb->data, skb->len | |
1541 | * and skb->cb are mangled. It works because (and until) packets | |
1542 | * falling here are owned by current CPU. Output packets are cloned | |
1543 | * by dev_queue_xmit_nit(), input packets are processed by net_bh | |
1544 | * sequencially, so that if we return skb to original state on exit, | |
1545 | * we will not harm anyone. | |
1da177e4 LT |
1546 | */ |
1547 | ||
40d4e3df ED |
1548 | static int packet_rcv(struct sk_buff *skb, struct net_device *dev, |
1549 | struct packet_type *pt, struct net_device *orig_dev) | |
1da177e4 LT |
1550 | { |
1551 | struct sock *sk; | |
1552 | struct sockaddr_ll *sll; | |
1553 | struct packet_sock *po; | |
40d4e3df | 1554 | u8 *skb_head = skb->data; |
1da177e4 | 1555 | int skb_len = skb->len; |
dbcb5855 | 1556 | unsigned int snaplen, res; |
1da177e4 LT |
1557 | |
1558 | if (skb->pkt_type == PACKET_LOOPBACK) | |
1559 | goto drop; | |
1560 | ||
1561 | sk = pt->af_packet_priv; | |
1562 | po = pkt_sk(sk); | |
1563 | ||
09ad9bc7 | 1564 | if (!net_eq(dev_net(dev), sock_net(sk))) |
d12d01d6 DL |
1565 | goto drop; |
1566 | ||
1da177e4 LT |
1567 | skb->dev = dev; |
1568 | ||
3b04ddde | 1569 | if (dev->header_ops) { |
1da177e4 | 1570 | /* The device has an explicit notion of ll header, |
62ab0812 ED |
1571 | * exported to higher levels. |
1572 | * | |
1573 | * Otherwise, the device hides details of its frame | |
1574 | * structure, so that corresponding packet head is | |
1575 | * never delivered to user. | |
1da177e4 LT |
1576 | */ |
1577 | if (sk->sk_type != SOCK_DGRAM) | |
98e399f8 | 1578 | skb_push(skb, skb->data - skb_mac_header(skb)); |
1da177e4 LT |
1579 | else if (skb->pkt_type == PACKET_OUTGOING) { |
1580 | /* Special case: outgoing packets have ll header at head */ | |
bbe735e4 | 1581 | skb_pull(skb, skb_network_offset(skb)); |
1da177e4 LT |
1582 | } |
1583 | } | |
1584 | ||
1585 | snaplen = skb->len; | |
1586 | ||
dbcb5855 DM |
1587 | res = run_filter(skb, sk, snaplen); |
1588 | if (!res) | |
fda9ef5d | 1589 | goto drop_n_restore; |
dbcb5855 DM |
1590 | if (snaplen > res) |
1591 | snaplen = res; | |
1da177e4 | 1592 | |
0fd7bac6 | 1593 | if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) |
1da177e4 LT |
1594 | goto drop_n_acct; |
1595 | ||
1596 | if (skb_shared(skb)) { | |
1597 | struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); | |
1598 | if (nskb == NULL) | |
1599 | goto drop_n_acct; | |
1600 | ||
1601 | if (skb_head != skb->data) { | |
1602 | skb->data = skb_head; | |
1603 | skb->len = skb_len; | |
1604 | } | |
abc4e4fa | 1605 | consume_skb(skb); |
1da177e4 LT |
1606 | skb = nskb; |
1607 | } | |
1608 | ||
ffbc6111 HX |
1609 | BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 > |
1610 | sizeof(skb->cb)); | |
1611 | ||
1612 | sll = &PACKET_SKB_CB(skb)->sa.ll; | |
1da177e4 LT |
1613 | sll->sll_family = AF_PACKET; |
1614 | sll->sll_hatype = dev->type; | |
1615 | sll->sll_protocol = skb->protocol; | |
1616 | sll->sll_pkttype = skb->pkt_type; | |
8032b464 | 1617 | if (unlikely(po->origdev)) |
80feaacb PWJ |
1618 | sll->sll_ifindex = orig_dev->ifindex; |
1619 | else | |
1620 | sll->sll_ifindex = dev->ifindex; | |
1da177e4 | 1621 | |
b95cce35 | 1622 | sll->sll_halen = dev_parse_header(skb, sll->sll_addr); |
1da177e4 | 1623 | |
ffbc6111 | 1624 | PACKET_SKB_CB(skb)->origlen = skb->len; |
8dc41944 | 1625 | |
1da177e4 LT |
1626 | if (pskb_trim(skb, snaplen)) |
1627 | goto drop_n_acct; | |
1628 | ||
1629 | skb_set_owner_r(skb, sk); | |
1630 | skb->dev = NULL; | |
adf30907 | 1631 | skb_dst_drop(skb); |
1da177e4 | 1632 | |
84531c24 PO |
1633 | /* drop conntrack reference */ |
1634 | nf_reset(skb); | |
1635 | ||
1da177e4 LT |
1636 | spin_lock(&sk->sk_receive_queue.lock); |
1637 | po->stats.tp_packets++; | |
3b885787 | 1638 | skb->dropcount = atomic_read(&sk->sk_drops); |
1da177e4 LT |
1639 | __skb_queue_tail(&sk->sk_receive_queue, skb); |
1640 | spin_unlock(&sk->sk_receive_queue.lock); | |
1641 | sk->sk_data_ready(sk, skb->len); | |
1642 | return 0; | |
1643 | ||
1644 | drop_n_acct: | |
7091fbd8 WB |
1645 | spin_lock(&sk->sk_receive_queue.lock); |
1646 | po->stats.tp_drops++; | |
1647 | atomic_inc(&sk->sk_drops); | |
1648 | spin_unlock(&sk->sk_receive_queue.lock); | |
1da177e4 LT |
1649 | |
1650 | drop_n_restore: | |
1651 | if (skb_head != skb->data && skb_shared(skb)) { | |
1652 | skb->data = skb_head; | |
1653 | skb->len = skb_len; | |
1654 | } | |
1655 | drop: | |
ead2ceb0 | 1656 | consume_skb(skb); |
1da177e4 LT |
1657 | return 0; |
1658 | } | |
1659 | ||
4b457bdf DB |
1660 | static void tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, |
1661 | unsigned int flags) | |
1662 | { | |
1663 | struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); | |
1664 | ||
1665 | if (shhwtstamps) { | |
1666 | if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) && | |
1667 | ktime_to_timespec_cond(shhwtstamps->syststamp, ts)) | |
1668 | return; | |
1669 | if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) && | |
1670 | ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) | |
1671 | return; | |
1672 | } | |
1673 | ||
1674 | if (ktime_to_timespec_cond(skb->tstamp, ts)) | |
1675 | return; | |
1676 | ||
1677 | getnstimeofday(ts); | |
1678 | } | |
1679 | ||
40d4e3df ED |
1680 | static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, |
1681 | struct packet_type *pt, struct net_device *orig_dev) | |
1da177e4 LT |
1682 | { |
1683 | struct sock *sk; | |
1684 | struct packet_sock *po; | |
1685 | struct sockaddr_ll *sll; | |
184f489e | 1686 | union tpacket_uhdr h; |
40d4e3df | 1687 | u8 *skb_head = skb->data; |
1da177e4 | 1688 | int skb_len = skb->len; |
dbcb5855 | 1689 | unsigned int snaplen, res; |
f6fb8f10 | 1690 | unsigned long status = TP_STATUS_USER; |
bbd6ef87 | 1691 | unsigned short macoff, netoff, hdrlen; |
1da177e4 | 1692 | struct sk_buff *copy_skb = NULL; |
bbd6ef87 | 1693 | struct timespec ts; |
1da177e4 LT |
1694 | |
1695 | if (skb->pkt_type == PACKET_LOOPBACK) | |
1696 | goto drop; | |
1697 | ||
1698 | sk = pt->af_packet_priv; | |
1699 | po = pkt_sk(sk); | |
1700 | ||
09ad9bc7 | 1701 | if (!net_eq(dev_net(dev), sock_net(sk))) |
d12d01d6 DL |
1702 | goto drop; |
1703 | ||
3b04ddde | 1704 | if (dev->header_ops) { |
1da177e4 | 1705 | if (sk->sk_type != SOCK_DGRAM) |
98e399f8 | 1706 | skb_push(skb, skb->data - skb_mac_header(skb)); |
1da177e4 LT |
1707 | else if (skb->pkt_type == PACKET_OUTGOING) { |
1708 | /* Special case: outgoing packets have ll header at head */ | |
bbe735e4 | 1709 | skb_pull(skb, skb_network_offset(skb)); |
1da177e4 LT |
1710 | } |
1711 | } | |
1712 | ||
8dc41944 HX |
1713 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
1714 | status |= TP_STATUS_CSUMNOTREADY; | |
1715 | ||
1da177e4 LT |
1716 | snaplen = skb->len; |
1717 | ||
dbcb5855 DM |
1718 | res = run_filter(skb, sk, snaplen); |
1719 | if (!res) | |
fda9ef5d | 1720 | goto drop_n_restore; |
dbcb5855 DM |
1721 | if (snaplen > res) |
1722 | snaplen = res; | |
1da177e4 LT |
1723 | |
1724 | if (sk->sk_type == SOCK_DGRAM) { | |
8913336a PM |
1725 | macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + |
1726 | po->tp_reserve; | |
1da177e4 | 1727 | } else { |
95c96174 | 1728 | unsigned int maclen = skb_network_offset(skb); |
bbd6ef87 | 1729 | netoff = TPACKET_ALIGN(po->tp_hdrlen + |
8913336a PM |
1730 | (maclen < 16 ? 16 : maclen)) + |
1731 | po->tp_reserve; | |
1da177e4 LT |
1732 | macoff = netoff - maclen; |
1733 | } | |
f6fb8f10 | 1734 | if (po->tp_version <= TPACKET_V2) { |
1735 | if (macoff + snaplen > po->rx_ring.frame_size) { | |
1736 | if (po->copy_thresh && | |
0fd7bac6 | 1737 | atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { |
f6fb8f10 | 1738 | if (skb_shared(skb)) { |
1739 | copy_skb = skb_clone(skb, GFP_ATOMIC); | |
1740 | } else { | |
1741 | copy_skb = skb_get(skb); | |
1742 | skb_head = skb->data; | |
1743 | } | |
1744 | if (copy_skb) | |
1745 | skb_set_owner_r(copy_skb, sk); | |
1da177e4 | 1746 | } |
f6fb8f10 | 1747 | snaplen = po->rx_ring.frame_size - macoff; |
1748 | if ((int)snaplen < 0) | |
1749 | snaplen = 0; | |
1da177e4 | 1750 | } |
1da177e4 | 1751 | } |
1da177e4 | 1752 | spin_lock(&sk->sk_receive_queue.lock); |
f6fb8f10 | 1753 | h.raw = packet_current_rx_frame(po, skb, |
1754 | TP_STATUS_KERNEL, (macoff+snaplen)); | |
bbd6ef87 | 1755 | if (!h.raw) |
1da177e4 | 1756 | goto ring_is_full; |
f6fb8f10 | 1757 | if (po->tp_version <= TPACKET_V2) { |
1758 | packet_increment_rx_head(po, &po->rx_ring); | |
1759 | /* | |
1760 | * LOSING will be reported till you read the stats, | |
1761 | * because it's COR - Clear On Read. | |
1762 | * Anyways, moving it for V1/V2 only as V3 doesn't need this | |
1763 | * at packet level. | |
1764 | */ | |
1765 | if (po->stats.tp_drops) | |
1766 | status |= TP_STATUS_LOSING; | |
1767 | } | |
1da177e4 LT |
1768 | po->stats.tp_packets++; |
1769 | if (copy_skb) { | |
1770 | status |= TP_STATUS_COPY; | |
1771 | __skb_queue_tail(&sk->sk_receive_queue, copy_skb); | |
1772 | } | |
1da177e4 LT |
1773 | spin_unlock(&sk->sk_receive_queue.lock); |
1774 | ||
bbd6ef87 | 1775 | skb_copy_bits(skb, 0, h.raw + macoff, snaplen); |
4b457bdf | 1776 | tpacket_get_timestamp(skb, &ts, po->tp_tstamp); |
1da177e4 | 1777 | |
bbd6ef87 PM |
1778 | switch (po->tp_version) { |
1779 | case TPACKET_V1: | |
1780 | h.h1->tp_len = skb->len; | |
1781 | h.h1->tp_snaplen = snaplen; | |
1782 | h.h1->tp_mac = macoff; | |
1783 | h.h1->tp_net = netoff; | |
4b457bdf DB |
1784 | h.h1->tp_sec = ts.tv_sec; |
1785 | h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; | |
bbd6ef87 PM |
1786 | hdrlen = sizeof(*h.h1); |
1787 | break; | |
1788 | case TPACKET_V2: | |
1789 | h.h2->tp_len = skb->len; | |
1790 | h.h2->tp_snaplen = snaplen; | |
1791 | h.h2->tp_mac = macoff; | |
1792 | h.h2->tp_net = netoff; | |
bbd6ef87 PM |
1793 | h.h2->tp_sec = ts.tv_sec; |
1794 | h.h2->tp_nsec = ts.tv_nsec; | |
a3bcc23e BG |
1795 | if (vlan_tx_tag_present(skb)) { |
1796 | h.h2->tp_vlan_tci = vlan_tx_tag_get(skb); | |
1797 | status |= TP_STATUS_VLAN_VALID; | |
1798 | } else { | |
1799 | h.h2->tp_vlan_tci = 0; | |
1800 | } | |
13fcb7bd | 1801 | h.h2->tp_padding = 0; |
bbd6ef87 PM |
1802 | hdrlen = sizeof(*h.h2); |
1803 | break; | |
f6fb8f10 | 1804 | case TPACKET_V3: |
1805 | /* tp_nxt_offset,vlan are already populated above. | |
1806 | * So DONT clear those fields here | |
1807 | */ | |
1808 | h.h3->tp_status |= status; | |
1809 | h.h3->tp_len = skb->len; | |
1810 | h.h3->tp_snaplen = snaplen; | |
1811 | h.h3->tp_mac = macoff; | |
1812 | h.h3->tp_net = netoff; | |
f6fb8f10 | 1813 | h.h3->tp_sec = ts.tv_sec; |
1814 | h.h3->tp_nsec = ts.tv_nsec; | |
1815 | hdrlen = sizeof(*h.h3); | |
1816 | break; | |
bbd6ef87 PM |
1817 | default: |
1818 | BUG(); | |
1819 | } | |
1da177e4 | 1820 | |
bbd6ef87 | 1821 | sll = h.raw + TPACKET_ALIGN(hdrlen); |
b95cce35 | 1822 | sll->sll_halen = dev_parse_header(skb, sll->sll_addr); |
1da177e4 LT |
1823 | sll->sll_family = AF_PACKET; |
1824 | sll->sll_hatype = dev->type; | |
1825 | sll->sll_protocol = skb->protocol; | |
1826 | sll->sll_pkttype = skb->pkt_type; | |
8032b464 | 1827 | if (unlikely(po->origdev)) |
80feaacb PWJ |
1828 | sll->sll_ifindex = orig_dev->ifindex; |
1829 | else | |
1830 | sll->sll_ifindex = dev->ifindex; | |
1da177e4 | 1831 | |
e16aa207 | 1832 | smp_mb(); |
f6dafa95 | 1833 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 |
1da177e4 | 1834 | { |
0af55bb5 CG |
1835 | u8 *start, *end; |
1836 | ||
f6fb8f10 | 1837 | if (po->tp_version <= TPACKET_V2) { |
1838 | end = (u8 *)PAGE_ALIGN((unsigned long)h.raw | |
1839 | + macoff + snaplen); | |
1840 | for (start = h.raw; start < end; start += PAGE_SIZE) | |
1841 | flush_dcache_page(pgv_to_page(start)); | |
1842 | } | |
cc9f01b2 | 1843 | smp_wmb(); |
1da177e4 | 1844 | } |
f6dafa95 | 1845 | #endif |
f6fb8f10 | 1846 | if (po->tp_version <= TPACKET_V2) |
1847 | __packet_set_status(po, h.raw, status); | |
1848 | else | |
1849 | prb_clear_blk_fill_status(&po->rx_ring); | |
1da177e4 LT |
1850 | |
1851 | sk->sk_data_ready(sk, 0); | |
1852 | ||
1853 | drop_n_restore: | |
1854 | if (skb_head != skb->data && skb_shared(skb)) { | |
1855 | skb->data = skb_head; | |
1856 | skb->len = skb_len; | |
1857 | } | |
1858 | drop: | |
1ce4f28b | 1859 | kfree_skb(skb); |
1da177e4 LT |
1860 | return 0; |
1861 | ||
1862 | ring_is_full: | |
1863 | po->stats.tp_drops++; | |
1864 | spin_unlock(&sk->sk_receive_queue.lock); | |
1865 | ||
1866 | sk->sk_data_ready(sk, 0); | |
acb5d75b | 1867 | kfree_skb(copy_skb); |
1da177e4 LT |
1868 | goto drop_n_restore; |
1869 | } | |
1870 | ||
69e3c75f JB |
1871 | static void tpacket_destruct_skb(struct sk_buff *skb) |
1872 | { | |
1873 | struct packet_sock *po = pkt_sk(skb->sk); | |
40d4e3df | 1874 | void *ph; |
1da177e4 | 1875 | |
69e3c75f JB |
1876 | if (likely(po->tx_ring.pg_vec)) { |
1877 | ph = skb_shinfo(skb)->destructor_arg; | |
69e3c75f JB |
1878 | BUG_ON(atomic_read(&po->tx_ring.pending) == 0); |
1879 | atomic_dec(&po->tx_ring.pending); | |
1880 | __packet_set_status(po, ph, TP_STATUS_AVAILABLE); | |
1881 | } | |
1882 | ||
1883 | sock_wfree(skb); | |
1884 | } | |
1885 | ||
40d4e3df ED |
1886 | static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, |
1887 | void *frame, struct net_device *dev, int size_max, | |
ae641949 | 1888 | __be16 proto, unsigned char *addr, int hlen) |
69e3c75f | 1889 | { |
184f489e | 1890 | union tpacket_uhdr ph; |
69e3c75f JB |
1891 | int to_write, offset, len, tp_len, nr_frags, len_max; |
1892 | struct socket *sock = po->sk.sk_socket; | |
1893 | struct page *page; | |
1894 | void *data; | |
1895 | int err; | |
1896 | ||
1897 | ph.raw = frame; | |
1898 | ||
1899 | skb->protocol = proto; | |
1900 | skb->dev = dev; | |
1901 | skb->priority = po->sk.sk_priority; | |
2d37a186 | 1902 | skb->mark = po->sk.sk_mark; |
69e3c75f JB |
1903 | skb_shinfo(skb)->destructor_arg = ph.raw; |
1904 | ||
1905 | switch (po->tp_version) { | |
1906 | case TPACKET_V2: | |
1907 | tp_len = ph.h2->tp_len; | |
1908 | break; | |
1909 | default: | |
1910 | tp_len = ph.h1->tp_len; | |
1911 | break; | |
1912 | } | |
1913 | if (unlikely(tp_len > size_max)) { | |
40d4e3df | 1914 | pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); |
69e3c75f JB |
1915 | return -EMSGSIZE; |
1916 | } | |
1917 | ||
ae641949 | 1918 | skb_reserve(skb, hlen); |
69e3c75f | 1919 | skb_reset_network_header(skb); |
40893fd0 | 1920 | skb_probe_transport_header(skb, 0); |
c1aad275 | 1921 | |
5920cd3a PC |
1922 | if (po->tp_tx_has_off) { |
1923 | int off_min, off_max, off; | |
1924 | off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); | |
1925 | off_max = po->tx_ring.frame_size - tp_len; | |
1926 | if (sock->type == SOCK_DGRAM) { | |
1927 | switch (po->tp_version) { | |
1928 | case TPACKET_V2: | |
1929 | off = ph.h2->tp_net; | |
1930 | break; | |
1931 | default: | |
1932 | off = ph.h1->tp_net; | |
1933 | break; | |
1934 | } | |
1935 | } else { | |
1936 | switch (po->tp_version) { | |
1937 | case TPACKET_V2: | |
1938 | off = ph.h2->tp_mac; | |
1939 | break; | |
1940 | default: | |
1941 | off = ph.h1->tp_mac; | |
1942 | break; | |
1943 | } | |
1944 | } | |
1945 | if (unlikely((off < off_min) || (off_max < off))) | |
1946 | return -EINVAL; | |
1947 | data = ph.raw + off; | |
1948 | } else { | |
1949 | data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll); | |
1950 | } | |
69e3c75f JB |
1951 | to_write = tp_len; |
1952 | ||
1953 | if (sock->type == SOCK_DGRAM) { | |
1954 | err = dev_hard_header(skb, dev, ntohs(proto), addr, | |
1955 | NULL, tp_len); | |
1956 | if (unlikely(err < 0)) | |
1957 | return -EINVAL; | |
40d4e3df | 1958 | } else if (dev->hard_header_len) { |
69e3c75f JB |
1959 | /* net device doesn't like empty head */ |
1960 | if (unlikely(tp_len <= dev->hard_header_len)) { | |
40d4e3df ED |
1961 | pr_err("packet size is too short (%d < %d)\n", |
1962 | tp_len, dev->hard_header_len); | |
69e3c75f JB |
1963 | return -EINVAL; |
1964 | } | |
1965 | ||
1966 | skb_push(skb, dev->hard_header_len); | |
1967 | err = skb_store_bits(skb, 0, data, | |
1968 | dev->hard_header_len); | |
1969 | if (unlikely(err)) | |
1970 | return err; | |
1971 | ||
1972 | data += dev->hard_header_len; | |
1973 | to_write -= dev->hard_header_len; | |
1974 | } | |
1975 | ||
69e3c75f JB |
1976 | offset = offset_in_page(data); |
1977 | len_max = PAGE_SIZE - offset; | |
1978 | len = ((to_write > len_max) ? len_max : to_write); | |
1979 | ||
1980 | skb->data_len = to_write; | |
1981 | skb->len += to_write; | |
1982 | skb->truesize += to_write; | |
1983 | atomic_add(to_write, &po->sk.sk_wmem_alloc); | |
1984 | ||
1985 | while (likely(to_write)) { | |
1986 | nr_frags = skb_shinfo(skb)->nr_frags; | |
1987 | ||
1988 | if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { | |
40d4e3df ED |
1989 | pr_err("Packet exceed the number of skb frags(%lu)\n", |
1990 | MAX_SKB_FRAGS); | |
69e3c75f JB |
1991 | return -EFAULT; |
1992 | } | |
1993 | ||
0af55bb5 CG |
1994 | page = pgv_to_page(data); |
1995 | data += len; | |
69e3c75f JB |
1996 | flush_dcache_page(page); |
1997 | get_page(page); | |
0af55bb5 | 1998 | skb_fill_page_desc(skb, nr_frags, page, offset, len); |
69e3c75f JB |
1999 | to_write -= len; |
2000 | offset = 0; | |
2001 | len_max = PAGE_SIZE; | |
2002 | len = ((to_write > len_max) ? len_max : to_write); | |
2003 | } | |
2004 | ||
2005 | return tp_len; | |
2006 | } | |
2007 | ||
2008 | static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) | |
2009 | { | |
69e3c75f JB |
2010 | struct sk_buff *skb; |
2011 | struct net_device *dev; | |
2012 | __be16 proto; | |
827d9780 BG |
2013 | bool need_rls_dev = false; |
2014 | int err, reserve = 0; | |
40d4e3df ED |
2015 | void *ph; |
2016 | struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name; | |
69e3c75f JB |
2017 | int tp_len, size_max; |
2018 | unsigned char *addr; | |
2019 | int len_sum = 0; | |
9e67030a | 2020 | int status = TP_STATUS_AVAILABLE; |
ae641949 | 2021 | int hlen, tlen; |
69e3c75f | 2022 | |
69e3c75f JB |
2023 | mutex_lock(&po->pg_vec_lock); |
2024 | ||
69e3c75f | 2025 | if (saddr == NULL) { |
827d9780 | 2026 | dev = po->prot_hook.dev; |
69e3c75f JB |
2027 | proto = po->num; |
2028 | addr = NULL; | |
2029 | } else { | |
2030 | err = -EINVAL; | |
2031 | if (msg->msg_namelen < sizeof(struct sockaddr_ll)) | |
2032 | goto out; | |
2033 | if (msg->msg_namelen < (saddr->sll_halen | |
2034 | + offsetof(struct sockaddr_ll, | |
2035 | sll_addr))) | |
2036 | goto out; | |
69e3c75f JB |
2037 | proto = saddr->sll_protocol; |
2038 | addr = saddr->sll_addr; | |
827d9780 BG |
2039 | dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); |
2040 | need_rls_dev = true; | |
69e3c75f JB |
2041 | } |
2042 | ||
69e3c75f JB |
2043 | err = -ENXIO; |
2044 | if (unlikely(dev == NULL)) | |
2045 | goto out; | |
2046 | ||
2047 | reserve = dev->hard_header_len; | |
2048 | ||
2049 | err = -ENETDOWN; | |
2050 | if (unlikely(!(dev->flags & IFF_UP))) | |
2051 | goto out_put; | |
2052 | ||
2053 | size_max = po->tx_ring.frame_size | |
b5dd884e | 2054 | - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); |
69e3c75f JB |
2055 | |
2056 | if (size_max > dev->mtu + reserve) | |
2057 | size_max = dev->mtu + reserve; | |
2058 | ||
2059 | do { | |
2060 | ph = packet_current_frame(po, &po->tx_ring, | |
2061 | TP_STATUS_SEND_REQUEST); | |
2062 | ||
2063 | if (unlikely(ph == NULL)) { | |
2064 | schedule(); | |
2065 | continue; | |
2066 | } | |
2067 | ||
2068 | status = TP_STATUS_SEND_REQUEST; | |
ae641949 HX |
2069 | hlen = LL_RESERVED_SPACE(dev); |
2070 | tlen = dev->needed_tailroom; | |
69e3c75f | 2071 | skb = sock_alloc_send_skb(&po->sk, |
ae641949 | 2072 | hlen + tlen + sizeof(struct sockaddr_ll), |
69e3c75f JB |
2073 | 0, &err); |
2074 | ||
2075 | if (unlikely(skb == NULL)) | |
2076 | goto out_status; | |
2077 | ||
2078 | tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto, | |
ae641949 | 2079 | addr, hlen); |
69e3c75f JB |
2080 | |
2081 | if (unlikely(tp_len < 0)) { | |
2082 | if (po->tp_loss) { | |
2083 | __packet_set_status(po, ph, | |
2084 | TP_STATUS_AVAILABLE); | |
2085 | packet_increment_head(&po->tx_ring); | |
2086 | kfree_skb(skb); | |
2087 | continue; | |
2088 | } else { | |
2089 | status = TP_STATUS_WRONG_FORMAT; | |
2090 | err = tp_len; | |
2091 | goto out_status; | |
2092 | } | |
2093 | } | |
2094 | ||
2095 | skb->destructor = tpacket_destruct_skb; | |
2096 | __packet_set_status(po, ph, TP_STATUS_SENDING); | |
2097 | atomic_inc(&po->tx_ring.pending); | |
2098 | ||
2099 | status = TP_STATUS_SEND_REQUEST; | |
2100 | err = dev_queue_xmit(skb); | |
eb70df13 JP |
2101 | if (unlikely(err > 0)) { |
2102 | err = net_xmit_errno(err); | |
2103 | if (err && __packet_get_status(po, ph) == | |
2104 | TP_STATUS_AVAILABLE) { | |
2105 | /* skb was destructed already */ | |
2106 | skb = NULL; | |
2107 | goto out_status; | |
2108 | } | |
2109 | /* | |
2110 | * skb was dropped but not destructed yet; | |
2111 | * let's treat it like congestion or err < 0 | |
2112 | */ | |
2113 | err = 0; | |
2114 | } | |
69e3c75f JB |
2115 | packet_increment_head(&po->tx_ring); |
2116 | len_sum += tp_len; | |
f64f9e71 JP |
2117 | } while (likely((ph != NULL) || |
2118 | ((!(msg->msg_flags & MSG_DONTWAIT)) && | |
2119 | (atomic_read(&po->tx_ring.pending)))) | |
2120 | ); | |
69e3c75f JB |
2121 | |
2122 | err = len_sum; | |
2123 | goto out_put; | |
2124 | ||
69e3c75f JB |
2125 | out_status: |
2126 | __packet_set_status(po, ph, status); | |
2127 | kfree_skb(skb); | |
2128 | out_put: | |
827d9780 BG |
2129 | if (need_rls_dev) |
2130 | dev_put(dev); | |
69e3c75f JB |
2131 | out: |
2132 | mutex_unlock(&po->pg_vec_lock); | |
2133 | return err; | |
2134 | } | |
69e3c75f | 2135 | |
eea49cc9 OJ |
2136 | static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, |
2137 | size_t reserve, size_t len, | |
2138 | size_t linear, int noblock, | |
2139 | int *err) | |
bfd5f4a3 SS |
2140 | { |
2141 | struct sk_buff *skb; | |
2142 | ||
2143 | /* Under a page? Don't bother with paged skb. */ | |
2144 | if (prepad + len < PAGE_SIZE || !linear) | |
2145 | linear = len; | |
2146 | ||
2147 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
2148 | err); | |
2149 | if (!skb) | |
2150 | return NULL; | |
2151 | ||
2152 | skb_reserve(skb, reserve); | |
2153 | skb_put(skb, linear); | |
2154 | skb->data_len = len - linear; | |
2155 | skb->len += len - linear; | |
2156 | ||
2157 | return skb; | |
2158 | } | |
2159 | ||
69e3c75f | 2160 | static int packet_snd(struct socket *sock, |
1da177e4 LT |
2161 | struct msghdr *msg, size_t len) |
2162 | { | |
2163 | struct sock *sk = sock->sk; | |
40d4e3df | 2164 | struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name; |
1da177e4 LT |
2165 | struct sk_buff *skb; |
2166 | struct net_device *dev; | |
0e11c91e | 2167 | __be16 proto; |
827d9780 | 2168 | bool need_rls_dev = false; |
1da177e4 | 2169 | unsigned char *addr; |
827d9780 | 2170 | int err, reserve = 0; |
bfd5f4a3 SS |
2171 | struct virtio_net_hdr vnet_hdr = { 0 }; |
2172 | int offset = 0; | |
2173 | int vnet_hdr_len; | |
2174 | struct packet_sock *po = pkt_sk(sk); | |
2175 | unsigned short gso_type = 0; | |
ae641949 | 2176 | int hlen, tlen; |
3bdc0eba | 2177 | int extra_len = 0; |
1da177e4 LT |
2178 | |
2179 | /* | |
1ce4f28b | 2180 | * Get and verify the address. |
1da177e4 | 2181 | */ |
1ce4f28b | 2182 | |
1da177e4 | 2183 | if (saddr == NULL) { |
827d9780 | 2184 | dev = po->prot_hook.dev; |
1da177e4 LT |
2185 | proto = po->num; |
2186 | addr = NULL; | |
2187 | } else { | |
2188 | err = -EINVAL; | |
2189 | if (msg->msg_namelen < sizeof(struct sockaddr_ll)) | |
2190 | goto out; | |
0fb375fb EB |
2191 | if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) |
2192 | goto out; | |
1da177e4 LT |
2193 | proto = saddr->sll_protocol; |
2194 | addr = saddr->sll_addr; | |
827d9780 BG |
2195 | dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); |
2196 | need_rls_dev = true; | |
1da177e4 LT |
2197 | } |
2198 | ||
1da177e4 LT |
2199 | err = -ENXIO; |
2200 | if (dev == NULL) | |
2201 | goto out_unlock; | |
2202 | if (sock->type == SOCK_RAW) | |
2203 | reserve = dev->hard_header_len; | |
2204 | ||
d5e76b0a DM |
2205 | err = -ENETDOWN; |
2206 | if (!(dev->flags & IFF_UP)) | |
2207 | goto out_unlock; | |
2208 | ||
bfd5f4a3 SS |
2209 | if (po->has_vnet_hdr) { |
2210 | vnet_hdr_len = sizeof(vnet_hdr); | |
2211 | ||
2212 | err = -EINVAL; | |
2213 | if (len < vnet_hdr_len) | |
2214 | goto out_unlock; | |
2215 | ||
2216 | len -= vnet_hdr_len; | |
2217 | ||
2218 | err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov, | |
2219 | vnet_hdr_len); | |
2220 | if (err < 0) | |
2221 | goto out_unlock; | |
2222 | ||
2223 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
2224 | (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
2225 | vnet_hdr.hdr_len)) | |
2226 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
2227 | vnet_hdr.csum_offset + 2; | |
2228 | ||
2229 | err = -EINVAL; | |
2230 | if (vnet_hdr.hdr_len > len) | |
2231 | goto out_unlock; | |
2232 | ||
2233 | if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
2234 | switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
2235 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
2236 | gso_type = SKB_GSO_TCPV4; | |
2237 | break; | |
2238 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
2239 | gso_type = SKB_GSO_TCPV6; | |
2240 | break; | |
2241 | case VIRTIO_NET_HDR_GSO_UDP: | |
2242 | gso_type = SKB_GSO_UDP; | |
2243 | break; | |
2244 | default: | |
2245 | goto out_unlock; | |
2246 | } | |
2247 | ||
2248 | if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
2249 | gso_type |= SKB_GSO_TCP_ECN; | |
2250 | ||
2251 | if (vnet_hdr.gso_size == 0) | |
2252 | goto out_unlock; | |
2253 | ||
2254 | } | |
2255 | } | |
2256 | ||
3bdc0eba BG |
2257 | if (unlikely(sock_flag(sk, SOCK_NOFCS))) { |
2258 | if (!netif_supports_nofcs(dev)) { | |
2259 | err = -EPROTONOSUPPORT; | |
2260 | goto out_unlock; | |
2261 | } | |
2262 | extra_len = 4; /* We're doing our own CRC */ | |
2263 | } | |
2264 | ||
1da177e4 | 2265 | err = -EMSGSIZE; |
3bdc0eba | 2266 | if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) |
1da177e4 LT |
2267 | goto out_unlock; |
2268 | ||
bfd5f4a3 | 2269 | err = -ENOBUFS; |
ae641949 HX |
2270 | hlen = LL_RESERVED_SPACE(dev); |
2271 | tlen = dev->needed_tailroom; | |
2272 | skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len, | |
bfd5f4a3 | 2273 | msg->msg_flags & MSG_DONTWAIT, &err); |
40d4e3df | 2274 | if (skb == NULL) |
1da177e4 LT |
2275 | goto out_unlock; |
2276 | ||
bfd5f4a3 | 2277 | skb_set_network_header(skb, reserve); |
1da177e4 | 2278 | |
0c4e8581 SH |
2279 | err = -EINVAL; |
2280 | if (sock->type == SOCK_DGRAM && | |
bfd5f4a3 | 2281 | (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0) |
0c4e8581 | 2282 | goto out_free; |
1da177e4 LT |
2283 | |
2284 | /* Returns -EFAULT on error */ | |
bfd5f4a3 | 2285 | err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len); |
1da177e4 LT |
2286 | if (err) |
2287 | goto out_free; | |
bf84a010 DB |
2288 | |
2289 | sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags); | |
1da177e4 | 2290 | |
3bdc0eba | 2291 | if (!gso_type && (len > dev->mtu + reserve + extra_len)) { |
57f89bfa BG |
2292 | /* Earlier code assumed this would be a VLAN pkt, |
2293 | * double-check this now that we have the actual | |
2294 | * packet in hand. | |
2295 | */ | |
2296 | struct ethhdr *ehdr; | |
2297 | skb_reset_mac_header(skb); | |
2298 | ehdr = eth_hdr(skb); | |
2299 | if (ehdr->h_proto != htons(ETH_P_8021Q)) { | |
2300 | err = -EMSGSIZE; | |
2301 | goto out_free; | |
2302 | } | |
2303 | } | |
2304 | ||
1da177e4 LT |
2305 | skb->protocol = proto; |
2306 | skb->dev = dev; | |
2307 | skb->priority = sk->sk_priority; | |
2d37a186 | 2308 | skb->mark = sk->sk_mark; |
1da177e4 | 2309 | |
bfd5f4a3 SS |
2310 | if (po->has_vnet_hdr) { |
2311 | if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
2312 | if (!skb_partial_csum_set(skb, vnet_hdr.csum_start, | |
2313 | vnet_hdr.csum_offset)) { | |
2314 | err = -EINVAL; | |
2315 | goto out_free; | |
2316 | } | |
2317 | } | |
2318 | ||
2319 | skb_shinfo(skb)->gso_size = vnet_hdr.gso_size; | |
2320 | skb_shinfo(skb)->gso_type = gso_type; | |
2321 | ||
2322 | /* Header must be checked, and gso_segs computed. */ | |
2323 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
2324 | skb_shinfo(skb)->gso_segs = 0; | |
2325 | ||
2326 | len += vnet_hdr_len; | |
2327 | } | |
2328 | ||
40893fd0 | 2329 | skb_probe_transport_header(skb, reserve); |
c1aad275 | 2330 | |
3bdc0eba BG |
2331 | if (unlikely(extra_len == 4)) |
2332 | skb->no_fcs = 1; | |
2333 | ||
1da177e4 LT |
2334 | /* |
2335 | * Now send it | |
2336 | */ | |
2337 | ||
2338 | err = dev_queue_xmit(skb); | |
2339 | if (err > 0 && (err = net_xmit_errno(err)) != 0) | |
2340 | goto out_unlock; | |
2341 | ||
827d9780 BG |
2342 | if (need_rls_dev) |
2343 | dev_put(dev); | |
1da177e4 | 2344 | |
40d4e3df | 2345 | return len; |
1da177e4 LT |
2346 | |
2347 | out_free: | |
2348 | kfree_skb(skb); | |
2349 | out_unlock: | |
827d9780 | 2350 | if (dev && need_rls_dev) |
1da177e4 LT |
2351 | dev_put(dev); |
2352 | out: | |
2353 | return err; | |
2354 | } | |
2355 | ||
69e3c75f JB |
2356 | static int packet_sendmsg(struct kiocb *iocb, struct socket *sock, |
2357 | struct msghdr *msg, size_t len) | |
2358 | { | |
69e3c75f JB |
2359 | struct sock *sk = sock->sk; |
2360 | struct packet_sock *po = pkt_sk(sk); | |
2361 | if (po->tx_ring.pg_vec) | |
2362 | return tpacket_snd(po, msg); | |
2363 | else | |
69e3c75f JB |
2364 | return packet_snd(sock, msg, len); |
2365 | } | |
2366 | ||
1da177e4 LT |
2367 | /* |
2368 | * Close a PACKET socket. This is fairly simple. We immediately go | |
2369 | * to 'closed' state and remove our protocol entry in the device list. | |
2370 | */ | |
2371 | ||
2372 | static int packet_release(struct socket *sock) | |
2373 | { | |
2374 | struct sock *sk = sock->sk; | |
2375 | struct packet_sock *po; | |
d12d01d6 | 2376 | struct net *net; |
f6fb8f10 | 2377 | union tpacket_req_u req_u; |
1da177e4 LT |
2378 | |
2379 | if (!sk) | |
2380 | return 0; | |
2381 | ||
3b1e0a65 | 2382 | net = sock_net(sk); |
1da177e4 LT |
2383 | po = pkt_sk(sk); |
2384 | ||
0fa7fa98 | 2385 | mutex_lock(&net->packet.sklist_lock); |
808f5114 | 2386 | sk_del_node_init_rcu(sk); |
0fa7fa98 PE |
2387 | mutex_unlock(&net->packet.sklist_lock); |
2388 | ||
2389 | preempt_disable(); | |
920de804 | 2390 | sock_prot_inuse_add(net, sk->sk_prot, -1); |
0fa7fa98 | 2391 | preempt_enable(); |
1da177e4 | 2392 | |
808f5114 | 2393 | spin_lock(&po->bind_lock); |
ce06b03e | 2394 | unregister_prot_hook(sk, false); |
160ff18a BG |
2395 | if (po->prot_hook.dev) { |
2396 | dev_put(po->prot_hook.dev); | |
2397 | po->prot_hook.dev = NULL; | |
2398 | } | |
808f5114 | 2399 | spin_unlock(&po->bind_lock); |
1da177e4 | 2400 | |
1da177e4 | 2401 | packet_flush_mclist(sk); |
1da177e4 | 2402 | |
9665d5d6 PS |
2403 | if (po->rx_ring.pg_vec) { |
2404 | memset(&req_u, 0, sizeof(req_u)); | |
f6fb8f10 | 2405 | packet_set_ring(sk, &req_u, 1, 0); |
9665d5d6 | 2406 | } |
69e3c75f | 2407 | |
9665d5d6 PS |
2408 | if (po->tx_ring.pg_vec) { |
2409 | memset(&req_u, 0, sizeof(req_u)); | |
f6fb8f10 | 2410 | packet_set_ring(sk, &req_u, 1, 1); |
9665d5d6 | 2411 | } |
1da177e4 | 2412 | |
dc99f600 DM |
2413 | fanout_release(sk); |
2414 | ||
808f5114 | 2415 | synchronize_net(); |
1da177e4 LT |
2416 | /* |
2417 | * Now the socket is dead. No more input will appear. | |
2418 | */ | |
1da177e4 LT |
2419 | sock_orphan(sk); |
2420 | sock->sk = NULL; | |
2421 | ||
2422 | /* Purge queues */ | |
2423 | ||
2424 | skb_queue_purge(&sk->sk_receive_queue); | |
17ab56a2 | 2425 | sk_refcnt_debug_release(sk); |
1da177e4 LT |
2426 | |
2427 | sock_put(sk); | |
2428 | return 0; | |
2429 | } | |
2430 | ||
2431 | /* | |
2432 | * Attach a packet hook. | |
2433 | */ | |
2434 | ||
0e11c91e | 2435 | static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol) |
1da177e4 LT |
2436 | { |
2437 | struct packet_sock *po = pkt_sk(sk); | |
dc99f600 | 2438 | |
aef950b4 WY |
2439 | if (po->fanout) { |
2440 | if (dev) | |
2441 | dev_put(dev); | |
2442 | ||
dc99f600 | 2443 | return -EINVAL; |
aef950b4 | 2444 | } |
1da177e4 LT |
2445 | |
2446 | lock_sock(sk); | |
2447 | ||
2448 | spin_lock(&po->bind_lock); | |
ce06b03e | 2449 | unregister_prot_hook(sk, true); |
1da177e4 LT |
2450 | po->num = protocol; |
2451 | po->prot_hook.type = protocol; | |
160ff18a BG |
2452 | if (po->prot_hook.dev) |
2453 | dev_put(po->prot_hook.dev); | |
1da177e4 LT |
2454 | po->prot_hook.dev = dev; |
2455 | ||
2456 | po->ifindex = dev ? dev->ifindex : 0; | |
2457 | ||
2458 | if (protocol == 0) | |
2459 | goto out_unlock; | |
2460 | ||
be85d4ad | 2461 | if (!dev || (dev->flags & IFF_UP)) { |
ce06b03e | 2462 | register_prot_hook(sk); |
be85d4ad UT |
2463 | } else { |
2464 | sk->sk_err = ENETDOWN; | |
2465 | if (!sock_flag(sk, SOCK_DEAD)) | |
2466 | sk->sk_error_report(sk); | |
1da177e4 LT |
2467 | } |
2468 | ||
2469 | out_unlock: | |
2470 | spin_unlock(&po->bind_lock); | |
2471 | release_sock(sk); | |
2472 | return 0; | |
2473 | } | |
2474 | ||
2475 | /* | |
2476 | * Bind a packet socket to a device | |
2477 | */ | |
2478 | ||
40d4e3df ED |
2479 | static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, |
2480 | int addr_len) | |
1da177e4 | 2481 | { |
40d4e3df | 2482 | struct sock *sk = sock->sk; |
1da177e4 LT |
2483 | char name[15]; |
2484 | struct net_device *dev; | |
2485 | int err = -ENODEV; | |
1ce4f28b | 2486 | |
1da177e4 LT |
2487 | /* |
2488 | * Check legality | |
2489 | */ | |
1ce4f28b | 2490 | |
8ae55f04 | 2491 | if (addr_len != sizeof(struct sockaddr)) |
1da177e4 | 2492 | return -EINVAL; |
40d4e3df | 2493 | strlcpy(name, uaddr->sa_data, sizeof(name)); |
1da177e4 | 2494 | |
3b1e0a65 | 2495 | dev = dev_get_by_name(sock_net(sk), name); |
160ff18a | 2496 | if (dev) |
1da177e4 | 2497 | err = packet_do_bind(sk, dev, pkt_sk(sk)->num); |
1da177e4 LT |
2498 | return err; |
2499 | } | |
1da177e4 LT |
2500 | |
2501 | static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) | |
2502 | { | |
40d4e3df ED |
2503 | struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; |
2504 | struct sock *sk = sock->sk; | |
1da177e4 LT |
2505 | struct net_device *dev = NULL; |
2506 | int err; | |
2507 | ||
2508 | ||
2509 | /* | |
2510 | * Check legality | |
2511 | */ | |
1ce4f28b | 2512 | |
1da177e4 LT |
2513 | if (addr_len < sizeof(struct sockaddr_ll)) |
2514 | return -EINVAL; | |
2515 | if (sll->sll_family != AF_PACKET) | |
2516 | return -EINVAL; | |
2517 | ||
2518 | if (sll->sll_ifindex) { | |
2519 | err = -ENODEV; | |
3b1e0a65 | 2520 | dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex); |
1da177e4 LT |
2521 | if (dev == NULL) |
2522 | goto out; | |
2523 | } | |
2524 | err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num); | |
1da177e4 LT |
2525 | |
2526 | out: | |
2527 | return err; | |
2528 | } | |
2529 | ||
2530 | static struct proto packet_proto = { | |
2531 | .name = "PACKET", | |
2532 | .owner = THIS_MODULE, | |
2533 | .obj_size = sizeof(struct packet_sock), | |
2534 | }; | |
2535 | ||
2536 | /* | |
1ce4f28b | 2537 | * Create a packet of type SOCK_PACKET. |
1da177e4 LT |
2538 | */ |
2539 | ||
3f378b68 EP |
2540 | static int packet_create(struct net *net, struct socket *sock, int protocol, |
2541 | int kern) | |
1da177e4 LT |
2542 | { |
2543 | struct sock *sk; | |
2544 | struct packet_sock *po; | |
0e11c91e | 2545 | __be16 proto = (__force __be16)protocol; /* weird, but documented */ |
1da177e4 LT |
2546 | int err; |
2547 | ||
df008c91 | 2548 | if (!ns_capable(net->user_ns, CAP_NET_RAW)) |
1da177e4 | 2549 | return -EPERM; |
be02097c DM |
2550 | if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && |
2551 | sock->type != SOCK_PACKET) | |
1da177e4 LT |
2552 | return -ESOCKTNOSUPPORT; |
2553 | ||
2554 | sock->state = SS_UNCONNECTED; | |
2555 | ||
2556 | err = -ENOBUFS; | |
6257ff21 | 2557 | sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto); |
1da177e4 LT |
2558 | if (sk == NULL) |
2559 | goto out; | |
2560 | ||
2561 | sock->ops = &packet_ops; | |
1da177e4 LT |
2562 | if (sock->type == SOCK_PACKET) |
2563 | sock->ops = &packet_ops_spkt; | |
be02097c | 2564 | |
1da177e4 LT |
2565 | sock_init_data(sock, sk); |
2566 | ||
2567 | po = pkt_sk(sk); | |
2568 | sk->sk_family = PF_PACKET; | |
0e11c91e | 2569 | po->num = proto; |
1da177e4 LT |
2570 | |
2571 | sk->sk_destruct = packet_sock_destruct; | |
17ab56a2 | 2572 | sk_refcnt_debug_inc(sk); |
1da177e4 LT |
2573 | |
2574 | /* | |
2575 | * Attach a protocol block | |
2576 | */ | |
2577 | ||
2578 | spin_lock_init(&po->bind_lock); | |
905db440 | 2579 | mutex_init(&po->pg_vec_lock); |
1da177e4 | 2580 | po->prot_hook.func = packet_rcv; |
be02097c | 2581 | |
1da177e4 LT |
2582 | if (sock->type == SOCK_PACKET) |
2583 | po->prot_hook.func = packet_rcv_spkt; | |
be02097c | 2584 | |
1da177e4 LT |
2585 | po->prot_hook.af_packet_priv = sk; |
2586 | ||
0e11c91e AV |
2587 | if (proto) { |
2588 | po->prot_hook.type = proto; | |
ce06b03e | 2589 | register_prot_hook(sk); |
1da177e4 LT |
2590 | } |
2591 | ||
0fa7fa98 | 2592 | mutex_lock(&net->packet.sklist_lock); |
808f5114 | 2593 | sk_add_node_rcu(sk, &net->packet.sklist); |
0fa7fa98 PE |
2594 | mutex_unlock(&net->packet.sklist_lock); |
2595 | ||
2596 | preempt_disable(); | |
3680453c | 2597 | sock_prot_inuse_add(net, &packet_proto, 1); |
0fa7fa98 | 2598 | preempt_enable(); |
808f5114 | 2599 | |
40d4e3df | 2600 | return 0; |
1da177e4 LT |
2601 | out: |
2602 | return err; | |
2603 | } | |
2604 | ||
ed85b565 RC |
2605 | static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len) |
2606 | { | |
2607 | struct sock_exterr_skb *serr; | |
2608 | struct sk_buff *skb, *skb2; | |
2609 | int copied, err; | |
2610 | ||
2611 | err = -EAGAIN; | |
2612 | skb = skb_dequeue(&sk->sk_error_queue); | |
2613 | if (skb == NULL) | |
2614 | goto out; | |
2615 | ||
2616 | copied = skb->len; | |
2617 | if (copied > len) { | |
2618 | msg->msg_flags |= MSG_TRUNC; | |
2619 | copied = len; | |
2620 | } | |
2621 | err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); | |
2622 | if (err) | |
2623 | goto out_free_skb; | |
2624 | ||
2625 | sock_recv_timestamp(msg, sk, skb); | |
2626 | ||
2627 | serr = SKB_EXT_ERR(skb); | |
2628 | put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP, | |
2629 | sizeof(serr->ee), &serr->ee); | |
2630 | ||
2631 | msg->msg_flags |= MSG_ERRQUEUE; | |
2632 | err = copied; | |
2633 | ||
2634 | /* Reset and regenerate socket error */ | |
2635 | spin_lock_bh(&sk->sk_error_queue.lock); | |
2636 | sk->sk_err = 0; | |
2637 | if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) { | |
2638 | sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno; | |
2639 | spin_unlock_bh(&sk->sk_error_queue.lock); | |
2640 | sk->sk_error_report(sk); | |
2641 | } else | |
2642 | spin_unlock_bh(&sk->sk_error_queue.lock); | |
2643 | ||
2644 | out_free_skb: | |
2645 | kfree_skb(skb); | |
2646 | out: | |
2647 | return err; | |
2648 | } | |
2649 | ||
1da177e4 LT |
2650 | /* |
2651 | * Pull a packet from our receive queue and hand it to the user. | |
2652 | * If necessary we block. | |
2653 | */ | |
2654 | ||
2655 | static int packet_recvmsg(struct kiocb *iocb, struct socket *sock, | |
2656 | struct msghdr *msg, size_t len, int flags) | |
2657 | { | |
2658 | struct sock *sk = sock->sk; | |
2659 | struct sk_buff *skb; | |
2660 | int copied, err; | |
0fb375fb | 2661 | struct sockaddr_ll *sll; |
bfd5f4a3 | 2662 | int vnet_hdr_len = 0; |
1da177e4 LT |
2663 | |
2664 | err = -EINVAL; | |
ed85b565 | 2665 | if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) |
1da177e4 LT |
2666 | goto out; |
2667 | ||
2668 | #if 0 | |
2669 | /* What error should we return now? EUNATTACH? */ | |
2670 | if (pkt_sk(sk)->ifindex < 0) | |
2671 | return -ENODEV; | |
2672 | #endif | |
2673 | ||
ed85b565 RC |
2674 | if (flags & MSG_ERRQUEUE) { |
2675 | err = packet_recv_error(sk, msg, len); | |
2676 | goto out; | |
2677 | } | |
2678 | ||
1da177e4 LT |
2679 | /* |
2680 | * Call the generic datagram receiver. This handles all sorts | |
2681 | * of horrible races and re-entrancy so we can forget about it | |
2682 | * in the protocol layers. | |
2683 | * | |
2684 | * Now it will return ENETDOWN, if device have just gone down, | |
2685 | * but then it will block. | |
2686 | */ | |
2687 | ||
40d4e3df | 2688 | skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); |
1da177e4 LT |
2689 | |
2690 | /* | |
1ce4f28b | 2691 | * An error occurred so return it. Because skb_recv_datagram() |
1da177e4 LT |
2692 | * handles the blocking we don't see and worry about blocking |
2693 | * retries. | |
2694 | */ | |
2695 | ||
8ae55f04 | 2696 | if (skb == NULL) |
1da177e4 LT |
2697 | goto out; |
2698 | ||
bfd5f4a3 SS |
2699 | if (pkt_sk(sk)->has_vnet_hdr) { |
2700 | struct virtio_net_hdr vnet_hdr = { 0 }; | |
2701 | ||
2702 | err = -EINVAL; | |
2703 | vnet_hdr_len = sizeof(vnet_hdr); | |
1f18b717 | 2704 | if (len < vnet_hdr_len) |
bfd5f4a3 SS |
2705 | goto out_free; |
2706 | ||
1f18b717 MK |
2707 | len -= vnet_hdr_len; |
2708 | ||
bfd5f4a3 SS |
2709 | if (skb_is_gso(skb)) { |
2710 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
2711 | ||
2712 | /* This is a hint as to how much should be linear. */ | |
2713 | vnet_hdr.hdr_len = skb_headlen(skb); | |
2714 | vnet_hdr.gso_size = sinfo->gso_size; | |
2715 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
2716 | vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
2717 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
2718 | vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
2719 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
2720 | vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
2721 | else if (sinfo->gso_type & SKB_GSO_FCOE) | |
2722 | goto out_free; | |
2723 | else | |
2724 | BUG(); | |
2725 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
2726 | vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
2727 | } else | |
2728 | vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
2729 | ||
2730 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2731 | vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 2732 | vnet_hdr.csum_start = skb_checksum_start_offset(skb); |
bfd5f4a3 | 2733 | vnet_hdr.csum_offset = skb->csum_offset; |
10a8d94a JW |
2734 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
2735 | vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
bfd5f4a3 SS |
2736 | } /* else everything is zero */ |
2737 | ||
2738 | err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr, | |
2739 | vnet_hdr_len); | |
2740 | if (err < 0) | |
2741 | goto out_free; | |
2742 | } | |
2743 | ||
0fb375fb EB |
2744 | /* |
2745 | * If the address length field is there to be filled in, we fill | |
2746 | * it in now. | |
2747 | */ | |
2748 | ||
ffbc6111 | 2749 | sll = &PACKET_SKB_CB(skb)->sa.ll; |
0fb375fb EB |
2750 | if (sock->type == SOCK_PACKET) |
2751 | msg->msg_namelen = sizeof(struct sockaddr_pkt); | |
2752 | else | |
2753 | msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr); | |
2754 | ||
1da177e4 LT |
2755 | /* |
2756 | * You lose any data beyond the buffer you gave. If it worries a | |
2757 | * user program they can ask the device for its MTU anyway. | |
2758 | */ | |
2759 | ||
2760 | copied = skb->len; | |
40d4e3df ED |
2761 | if (copied > len) { |
2762 | copied = len; | |
2763 | msg->msg_flags |= MSG_TRUNC; | |
1da177e4 LT |
2764 | } |
2765 | ||
2766 | err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); | |
2767 | if (err) | |
2768 | goto out_free; | |
2769 | ||
3b885787 | 2770 | sock_recv_ts_and_drops(msg, sk, skb); |
1da177e4 LT |
2771 | |
2772 | if (msg->msg_name) | |
ffbc6111 HX |
2773 | memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, |
2774 | msg->msg_namelen); | |
1da177e4 | 2775 | |
8dc41944 | 2776 | if (pkt_sk(sk)->auxdata) { |
ffbc6111 HX |
2777 | struct tpacket_auxdata aux; |
2778 | ||
2779 | aux.tp_status = TP_STATUS_USER; | |
2780 | if (skb->ip_summed == CHECKSUM_PARTIAL) | |
2781 | aux.tp_status |= TP_STATUS_CSUMNOTREADY; | |
2782 | aux.tp_len = PACKET_SKB_CB(skb)->origlen; | |
2783 | aux.tp_snaplen = skb->len; | |
2784 | aux.tp_mac = 0; | |
bbe735e4 | 2785 | aux.tp_net = skb_network_offset(skb); |
a3bcc23e BG |
2786 | if (vlan_tx_tag_present(skb)) { |
2787 | aux.tp_vlan_tci = vlan_tx_tag_get(skb); | |
2788 | aux.tp_status |= TP_STATUS_VLAN_VALID; | |
2789 | } else { | |
2790 | aux.tp_vlan_tci = 0; | |
2791 | } | |
13fcb7bd | 2792 | aux.tp_padding = 0; |
ffbc6111 | 2793 | put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); |
8dc41944 HX |
2794 | } |
2795 | ||
1da177e4 LT |
2796 | /* |
2797 | * Free or return the buffer as appropriate. Again this | |
2798 | * hides all the races and re-entrancy issues from us. | |
2799 | */ | |
bfd5f4a3 | 2800 | err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); |
1da177e4 LT |
2801 | |
2802 | out_free: | |
2803 | skb_free_datagram(sk, skb); | |
2804 | out: | |
2805 | return err; | |
2806 | } | |
2807 | ||
1da177e4 LT |
2808 | static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, |
2809 | int *uaddr_len, int peer) | |
2810 | { | |
2811 | struct net_device *dev; | |
2812 | struct sock *sk = sock->sk; | |
2813 | ||
2814 | if (peer) | |
2815 | return -EOPNOTSUPP; | |
2816 | ||
2817 | uaddr->sa_family = AF_PACKET; | |
654d1f8a ED |
2818 | rcu_read_lock(); |
2819 | dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); | |
2820 | if (dev) | |
67286640 | 2821 | strncpy(uaddr->sa_data, dev->name, 14); |
654d1f8a | 2822 | else |
1da177e4 | 2823 | memset(uaddr->sa_data, 0, 14); |
654d1f8a | 2824 | rcu_read_unlock(); |
1da177e4 LT |
2825 | *uaddr_len = sizeof(*uaddr); |
2826 | ||
2827 | return 0; | |
2828 | } | |
1da177e4 LT |
2829 | |
2830 | static int packet_getname(struct socket *sock, struct sockaddr *uaddr, | |
2831 | int *uaddr_len, int peer) | |
2832 | { | |
2833 | struct net_device *dev; | |
2834 | struct sock *sk = sock->sk; | |
2835 | struct packet_sock *po = pkt_sk(sk); | |
13cfa97b | 2836 | DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); |
1da177e4 LT |
2837 | |
2838 | if (peer) | |
2839 | return -EOPNOTSUPP; | |
2840 | ||
2841 | sll->sll_family = AF_PACKET; | |
2842 | sll->sll_ifindex = po->ifindex; | |
2843 | sll->sll_protocol = po->num; | |
67286640 | 2844 | sll->sll_pkttype = 0; |
654d1f8a ED |
2845 | rcu_read_lock(); |
2846 | dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); | |
1da177e4 LT |
2847 | if (dev) { |
2848 | sll->sll_hatype = dev->type; | |
2849 | sll->sll_halen = dev->addr_len; | |
2850 | memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); | |
1da177e4 LT |
2851 | } else { |
2852 | sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ | |
2853 | sll->sll_halen = 0; | |
2854 | } | |
654d1f8a | 2855 | rcu_read_unlock(); |
0fb375fb | 2856 | *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; |
1da177e4 LT |
2857 | |
2858 | return 0; | |
2859 | } | |
2860 | ||
2aeb0b88 WC |
2861 | static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, |
2862 | int what) | |
1da177e4 LT |
2863 | { |
2864 | switch (i->type) { | |
2865 | case PACKET_MR_MULTICAST: | |
1162563f JP |
2866 | if (i->alen != dev->addr_len) |
2867 | return -EINVAL; | |
1da177e4 | 2868 | if (what > 0) |
22bedad3 | 2869 | return dev_mc_add(dev, i->addr); |
1da177e4 | 2870 | else |
22bedad3 | 2871 | return dev_mc_del(dev, i->addr); |
1da177e4 LT |
2872 | break; |
2873 | case PACKET_MR_PROMISC: | |
2aeb0b88 | 2874 | return dev_set_promiscuity(dev, what); |
1da177e4 LT |
2875 | break; |
2876 | case PACKET_MR_ALLMULTI: | |
2aeb0b88 | 2877 | return dev_set_allmulti(dev, what); |
1da177e4 | 2878 | break; |
d95ed927 | 2879 | case PACKET_MR_UNICAST: |
1162563f JP |
2880 | if (i->alen != dev->addr_len) |
2881 | return -EINVAL; | |
d95ed927 | 2882 | if (what > 0) |
a748ee24 | 2883 | return dev_uc_add(dev, i->addr); |
d95ed927 | 2884 | else |
a748ee24 | 2885 | return dev_uc_del(dev, i->addr); |
d95ed927 | 2886 | break; |
40d4e3df ED |
2887 | default: |
2888 | break; | |
1da177e4 | 2889 | } |
2aeb0b88 | 2890 | return 0; |
1da177e4 LT |
2891 | } |
2892 | ||
2893 | static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what) | |
2894 | { | |
40d4e3df | 2895 | for ( ; i; i = i->next) { |
1da177e4 LT |
2896 | if (i->ifindex == dev->ifindex) |
2897 | packet_dev_mc(dev, i, what); | |
2898 | } | |
2899 | } | |
2900 | ||
0fb375fb | 2901 | static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) |
1da177e4 LT |
2902 | { |
2903 | struct packet_sock *po = pkt_sk(sk); | |
2904 | struct packet_mclist *ml, *i; | |
2905 | struct net_device *dev; | |
2906 | int err; | |
2907 | ||
2908 | rtnl_lock(); | |
2909 | ||
2910 | err = -ENODEV; | |
3b1e0a65 | 2911 | dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); |
1da177e4 LT |
2912 | if (!dev) |
2913 | goto done; | |
2914 | ||
2915 | err = -EINVAL; | |
1162563f | 2916 | if (mreq->mr_alen > dev->addr_len) |
1da177e4 LT |
2917 | goto done; |
2918 | ||
2919 | err = -ENOBUFS; | |
8b3a7005 | 2920 | i = kmalloc(sizeof(*i), GFP_KERNEL); |
1da177e4 LT |
2921 | if (i == NULL) |
2922 | goto done; | |
2923 | ||
2924 | err = 0; | |
2925 | for (ml = po->mclist; ml; ml = ml->next) { | |
2926 | if (ml->ifindex == mreq->mr_ifindex && | |
2927 | ml->type == mreq->mr_type && | |
2928 | ml->alen == mreq->mr_alen && | |
2929 | memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { | |
2930 | ml->count++; | |
2931 | /* Free the new element ... */ | |
2932 | kfree(i); | |
2933 | goto done; | |
2934 | } | |
2935 | } | |
2936 | ||
2937 | i->type = mreq->mr_type; | |
2938 | i->ifindex = mreq->mr_ifindex; | |
2939 | i->alen = mreq->mr_alen; | |
2940 | memcpy(i->addr, mreq->mr_address, i->alen); | |
2941 | i->count = 1; | |
2942 | i->next = po->mclist; | |
2943 | po->mclist = i; | |
2aeb0b88 WC |
2944 | err = packet_dev_mc(dev, i, 1); |
2945 | if (err) { | |
2946 | po->mclist = i->next; | |
2947 | kfree(i); | |
2948 | } | |
1da177e4 LT |
2949 | |
2950 | done: | |
2951 | rtnl_unlock(); | |
2952 | return err; | |
2953 | } | |
2954 | ||
0fb375fb | 2955 | static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) |
1da177e4 LT |
2956 | { |
2957 | struct packet_mclist *ml, **mlp; | |
2958 | ||
2959 | rtnl_lock(); | |
2960 | ||
2961 | for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { | |
2962 | if (ml->ifindex == mreq->mr_ifindex && | |
2963 | ml->type == mreq->mr_type && | |
2964 | ml->alen == mreq->mr_alen && | |
2965 | memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { | |
2966 | if (--ml->count == 0) { | |
2967 | struct net_device *dev; | |
2968 | *mlp = ml->next; | |
ad959e76 ED |
2969 | dev = __dev_get_by_index(sock_net(sk), ml->ifindex); |
2970 | if (dev) | |
1da177e4 | 2971 | packet_dev_mc(dev, ml, -1); |
1da177e4 LT |
2972 | kfree(ml); |
2973 | } | |
2974 | rtnl_unlock(); | |
2975 | return 0; | |
2976 | } | |
2977 | } | |
2978 | rtnl_unlock(); | |
2979 | return -EADDRNOTAVAIL; | |
2980 | } | |
2981 | ||
2982 | static void packet_flush_mclist(struct sock *sk) | |
2983 | { | |
2984 | struct packet_sock *po = pkt_sk(sk); | |
2985 | struct packet_mclist *ml; | |
2986 | ||
2987 | if (!po->mclist) | |
2988 | return; | |
2989 | ||
2990 | rtnl_lock(); | |
2991 | while ((ml = po->mclist) != NULL) { | |
2992 | struct net_device *dev; | |
2993 | ||
2994 | po->mclist = ml->next; | |
ad959e76 ED |
2995 | dev = __dev_get_by_index(sock_net(sk), ml->ifindex); |
2996 | if (dev != NULL) | |
1da177e4 | 2997 | packet_dev_mc(dev, ml, -1); |
1da177e4 LT |
2998 | kfree(ml); |
2999 | } | |
3000 | rtnl_unlock(); | |
3001 | } | |
1da177e4 LT |
3002 | |
3003 | static int | |
b7058842 | 3004 | packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) |
1da177e4 LT |
3005 | { |
3006 | struct sock *sk = sock->sk; | |
8dc41944 | 3007 | struct packet_sock *po = pkt_sk(sk); |
1da177e4 LT |
3008 | int ret; |
3009 | ||
3010 | if (level != SOL_PACKET) | |
3011 | return -ENOPROTOOPT; | |
3012 | ||
69e3c75f | 3013 | switch (optname) { |
1ce4f28b | 3014 | case PACKET_ADD_MEMBERSHIP: |
1da177e4 LT |
3015 | case PACKET_DROP_MEMBERSHIP: |
3016 | { | |
0fb375fb EB |
3017 | struct packet_mreq_max mreq; |
3018 | int len = optlen; | |
3019 | memset(&mreq, 0, sizeof(mreq)); | |
3020 | if (len < sizeof(struct packet_mreq)) | |
1da177e4 | 3021 | return -EINVAL; |
0fb375fb EB |
3022 | if (len > sizeof(mreq)) |
3023 | len = sizeof(mreq); | |
40d4e3df | 3024 | if (copy_from_user(&mreq, optval, len)) |
1da177e4 | 3025 | return -EFAULT; |
0fb375fb EB |
3026 | if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) |
3027 | return -EINVAL; | |
1da177e4 LT |
3028 | if (optname == PACKET_ADD_MEMBERSHIP) |
3029 | ret = packet_mc_add(sk, &mreq); | |
3030 | else | |
3031 | ret = packet_mc_drop(sk, &mreq); | |
3032 | return ret; | |
3033 | } | |
a2efcfa0 | 3034 | |
1da177e4 | 3035 | case PACKET_RX_RING: |
69e3c75f | 3036 | case PACKET_TX_RING: |
1da177e4 | 3037 | { |
f6fb8f10 | 3038 | union tpacket_req_u req_u; |
3039 | int len; | |
1da177e4 | 3040 | |
f6fb8f10 | 3041 | switch (po->tp_version) { |
3042 | case TPACKET_V1: | |
3043 | case TPACKET_V2: | |
3044 | len = sizeof(req_u.req); | |
3045 | break; | |
3046 | case TPACKET_V3: | |
3047 | default: | |
3048 | len = sizeof(req_u.req3); | |
3049 | break; | |
3050 | } | |
3051 | if (optlen < len) | |
1da177e4 | 3052 | return -EINVAL; |
bfd5f4a3 SS |
3053 | if (pkt_sk(sk)->has_vnet_hdr) |
3054 | return -EINVAL; | |
f6fb8f10 | 3055 | if (copy_from_user(&req_u.req, optval, len)) |
1da177e4 | 3056 | return -EFAULT; |
f6fb8f10 | 3057 | return packet_set_ring(sk, &req_u, 0, |
3058 | optname == PACKET_TX_RING); | |
1da177e4 LT |
3059 | } |
3060 | case PACKET_COPY_THRESH: | |
3061 | { | |
3062 | int val; | |
3063 | ||
40d4e3df | 3064 | if (optlen != sizeof(val)) |
1da177e4 | 3065 | return -EINVAL; |
40d4e3df | 3066 | if (copy_from_user(&val, optval, sizeof(val))) |
1da177e4 LT |
3067 | return -EFAULT; |
3068 | ||
3069 | pkt_sk(sk)->copy_thresh = val; | |
3070 | return 0; | |
3071 | } | |
bbd6ef87 PM |
3072 | case PACKET_VERSION: |
3073 | { | |
3074 | int val; | |
3075 | ||
3076 | if (optlen != sizeof(val)) | |
3077 | return -EINVAL; | |
69e3c75f | 3078 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) |
bbd6ef87 PM |
3079 | return -EBUSY; |
3080 | if (copy_from_user(&val, optval, sizeof(val))) | |
3081 | return -EFAULT; | |
3082 | switch (val) { | |
3083 | case TPACKET_V1: | |
3084 | case TPACKET_V2: | |
f6fb8f10 | 3085 | case TPACKET_V3: |
bbd6ef87 PM |
3086 | po->tp_version = val; |
3087 | return 0; | |
3088 | default: | |
3089 | return -EINVAL; | |
3090 | } | |
3091 | } | |
8913336a PM |
3092 | case PACKET_RESERVE: |
3093 | { | |
3094 | unsigned int val; | |
3095 | ||
3096 | if (optlen != sizeof(val)) | |
3097 | return -EINVAL; | |
69e3c75f | 3098 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) |
8913336a PM |
3099 | return -EBUSY; |
3100 | if (copy_from_user(&val, optval, sizeof(val))) | |
3101 | return -EFAULT; | |
3102 | po->tp_reserve = val; | |
3103 | return 0; | |
3104 | } | |
69e3c75f JB |
3105 | case PACKET_LOSS: |
3106 | { | |
3107 | unsigned int val; | |
3108 | ||
3109 | if (optlen != sizeof(val)) | |
3110 | return -EINVAL; | |
3111 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) | |
3112 | return -EBUSY; | |
3113 | if (copy_from_user(&val, optval, sizeof(val))) | |
3114 | return -EFAULT; | |
3115 | po->tp_loss = !!val; | |
3116 | return 0; | |
3117 | } | |
8dc41944 HX |
3118 | case PACKET_AUXDATA: |
3119 | { | |
3120 | int val; | |
3121 | ||
3122 | if (optlen < sizeof(val)) | |
3123 | return -EINVAL; | |
3124 | if (copy_from_user(&val, optval, sizeof(val))) | |
3125 | return -EFAULT; | |
3126 | ||
3127 | po->auxdata = !!val; | |
3128 | return 0; | |
3129 | } | |
80feaacb PWJ |
3130 | case PACKET_ORIGDEV: |
3131 | { | |
3132 | int val; | |
3133 | ||
3134 | if (optlen < sizeof(val)) | |
3135 | return -EINVAL; | |
3136 | if (copy_from_user(&val, optval, sizeof(val))) | |
3137 | return -EFAULT; | |
3138 | ||
3139 | po->origdev = !!val; | |
3140 | return 0; | |
3141 | } | |
bfd5f4a3 SS |
3142 | case PACKET_VNET_HDR: |
3143 | { | |
3144 | int val; | |
3145 | ||
3146 | if (sock->type != SOCK_RAW) | |
3147 | return -EINVAL; | |
3148 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) | |
3149 | return -EBUSY; | |
3150 | if (optlen < sizeof(val)) | |
3151 | return -EINVAL; | |
3152 | if (copy_from_user(&val, optval, sizeof(val))) | |
3153 | return -EFAULT; | |
3154 | ||
3155 | po->has_vnet_hdr = !!val; | |
3156 | return 0; | |
3157 | } | |
614f60fa SM |
3158 | case PACKET_TIMESTAMP: |
3159 | { | |
3160 | int val; | |
3161 | ||
3162 | if (optlen != sizeof(val)) | |
3163 | return -EINVAL; | |
3164 | if (copy_from_user(&val, optval, sizeof(val))) | |
3165 | return -EFAULT; | |
3166 | ||
3167 | po->tp_tstamp = val; | |
3168 | return 0; | |
3169 | } | |
dc99f600 DM |
3170 | case PACKET_FANOUT: |
3171 | { | |
3172 | int val; | |
3173 | ||
3174 | if (optlen != sizeof(val)) | |
3175 | return -EINVAL; | |
3176 | if (copy_from_user(&val, optval, sizeof(val))) | |
3177 | return -EFAULT; | |
3178 | ||
3179 | return fanout_add(sk, val & 0xffff, val >> 16); | |
3180 | } | |
5920cd3a PC |
3181 | case PACKET_TX_HAS_OFF: |
3182 | { | |
3183 | unsigned int val; | |
3184 | ||
3185 | if (optlen != sizeof(val)) | |
3186 | return -EINVAL; | |
3187 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) | |
3188 | return -EBUSY; | |
3189 | if (copy_from_user(&val, optval, sizeof(val))) | |
3190 | return -EFAULT; | |
3191 | po->tp_tx_has_off = !!val; | |
3192 | return 0; | |
3193 | } | |
1da177e4 LT |
3194 | default: |
3195 | return -ENOPROTOOPT; | |
3196 | } | |
3197 | } | |
3198 | ||
3199 | static int packet_getsockopt(struct socket *sock, int level, int optname, | |
3200 | char __user *optval, int __user *optlen) | |
3201 | { | |
3202 | int len; | |
c06fff6e | 3203 | int val, lv = sizeof(val); |
1da177e4 LT |
3204 | struct sock *sk = sock->sk; |
3205 | struct packet_sock *po = pkt_sk(sk); | |
c06fff6e | 3206 | void *data = &val; |
8dc41944 | 3207 | struct tpacket_stats st; |
f6fb8f10 | 3208 | union tpacket_stats_u st_u; |
1da177e4 LT |
3209 | |
3210 | if (level != SOL_PACKET) | |
3211 | return -ENOPROTOOPT; | |
3212 | ||
8ae55f04 KK |
3213 | if (get_user(len, optlen)) |
3214 | return -EFAULT; | |
1da177e4 LT |
3215 | |
3216 | if (len < 0) | |
3217 | return -EINVAL; | |
1ce4f28b | 3218 | |
69e3c75f | 3219 | switch (optname) { |
1da177e4 | 3220 | case PACKET_STATISTICS: |
1da177e4 | 3221 | spin_lock_bh(&sk->sk_receive_queue.lock); |
f6fb8f10 | 3222 | if (po->tp_version == TPACKET_V3) { |
c06fff6e | 3223 | lv = sizeof(struct tpacket_stats_v3); |
f6fb8f10 | 3224 | memcpy(&st_u.stats3, &po->stats, |
c06fff6e | 3225 | sizeof(struct tpacket_stats)); |
f6fb8f10 | 3226 | st_u.stats3.tp_freeze_q_cnt = |
c06fff6e | 3227 | po->stats_u.stats3.tp_freeze_q_cnt; |
f6fb8f10 | 3228 | st_u.stats3.tp_packets += po->stats.tp_drops; |
3229 | data = &st_u.stats3; | |
3230 | } else { | |
c06fff6e | 3231 | lv = sizeof(struct tpacket_stats); |
f6fb8f10 | 3232 | st = po->stats; |
3233 | st.tp_packets += st.tp_drops; | |
3234 | data = &st; | |
3235 | } | |
1da177e4 LT |
3236 | memset(&po->stats, 0, sizeof(st)); |
3237 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
8dc41944 HX |
3238 | break; |
3239 | case PACKET_AUXDATA: | |
8dc41944 | 3240 | val = po->auxdata; |
80feaacb PWJ |
3241 | break; |
3242 | case PACKET_ORIGDEV: | |
80feaacb | 3243 | val = po->origdev; |
bfd5f4a3 SS |
3244 | break; |
3245 | case PACKET_VNET_HDR: | |
bfd5f4a3 | 3246 | val = po->has_vnet_hdr; |
1da177e4 | 3247 | break; |
bbd6ef87 | 3248 | case PACKET_VERSION: |
bbd6ef87 | 3249 | val = po->tp_version; |
bbd6ef87 PM |
3250 | break; |
3251 | case PACKET_HDRLEN: | |
3252 | if (len > sizeof(int)) | |
3253 | len = sizeof(int); | |
3254 | if (copy_from_user(&val, optval, len)) | |
3255 | return -EFAULT; | |
3256 | switch (val) { | |
3257 | case TPACKET_V1: | |
3258 | val = sizeof(struct tpacket_hdr); | |
3259 | break; | |
3260 | case TPACKET_V2: | |
3261 | val = sizeof(struct tpacket2_hdr); | |
3262 | break; | |
f6fb8f10 | 3263 | case TPACKET_V3: |
3264 | val = sizeof(struct tpacket3_hdr); | |
3265 | break; | |
bbd6ef87 PM |
3266 | default: |
3267 | return -EINVAL; | |
3268 | } | |
bbd6ef87 | 3269 | break; |
8913336a | 3270 | case PACKET_RESERVE: |
8913336a | 3271 | val = po->tp_reserve; |
8913336a | 3272 | break; |
69e3c75f | 3273 | case PACKET_LOSS: |
69e3c75f | 3274 | val = po->tp_loss; |
69e3c75f | 3275 | break; |
614f60fa | 3276 | case PACKET_TIMESTAMP: |
614f60fa | 3277 | val = po->tp_tstamp; |
614f60fa | 3278 | break; |
dc99f600 | 3279 | case PACKET_FANOUT: |
dc99f600 DM |
3280 | val = (po->fanout ? |
3281 | ((u32)po->fanout->id | | |
77f65ebd WB |
3282 | ((u32)po->fanout->type << 16) | |
3283 | ((u32)po->fanout->flags << 24)) : | |
dc99f600 | 3284 | 0); |
dc99f600 | 3285 | break; |
5920cd3a PC |
3286 | case PACKET_TX_HAS_OFF: |
3287 | val = po->tp_tx_has_off; | |
3288 | break; | |
1da177e4 LT |
3289 | default: |
3290 | return -ENOPROTOOPT; | |
3291 | } | |
3292 | ||
c06fff6e ED |
3293 | if (len > lv) |
3294 | len = lv; | |
8ae55f04 KK |
3295 | if (put_user(len, optlen)) |
3296 | return -EFAULT; | |
8dc41944 HX |
3297 | if (copy_to_user(optval, data, len)) |
3298 | return -EFAULT; | |
8ae55f04 | 3299 | return 0; |
1da177e4 LT |
3300 | } |
3301 | ||
3302 | ||
3303 | static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data) | |
3304 | { | |
3305 | struct sock *sk; | |
ad930650 | 3306 | struct net_device *dev = data; |
c346dca1 | 3307 | struct net *net = dev_net(dev); |
1da177e4 | 3308 | |
808f5114 | 3309 | rcu_read_lock(); |
b67bfe0d | 3310 | sk_for_each_rcu(sk, &net->packet.sklist) { |
1da177e4 LT |
3311 | struct packet_sock *po = pkt_sk(sk); |
3312 | ||
3313 | switch (msg) { | |
3314 | case NETDEV_UNREGISTER: | |
1da177e4 LT |
3315 | if (po->mclist) |
3316 | packet_dev_mclist(dev, po->mclist, -1); | |
a2efcfa0 DM |
3317 | /* fallthrough */ |
3318 | ||
1da177e4 LT |
3319 | case NETDEV_DOWN: |
3320 | if (dev->ifindex == po->ifindex) { | |
3321 | spin_lock(&po->bind_lock); | |
3322 | if (po->running) { | |
ce06b03e | 3323 | __unregister_prot_hook(sk, false); |
1da177e4 LT |
3324 | sk->sk_err = ENETDOWN; |
3325 | if (!sock_flag(sk, SOCK_DEAD)) | |
3326 | sk->sk_error_report(sk); | |
3327 | } | |
3328 | if (msg == NETDEV_UNREGISTER) { | |
3329 | po->ifindex = -1; | |
160ff18a BG |
3330 | if (po->prot_hook.dev) |
3331 | dev_put(po->prot_hook.dev); | |
1da177e4 LT |
3332 | po->prot_hook.dev = NULL; |
3333 | } | |
3334 | spin_unlock(&po->bind_lock); | |
3335 | } | |
3336 | break; | |
3337 | case NETDEV_UP: | |
808f5114 | 3338 | if (dev->ifindex == po->ifindex) { |
3339 | spin_lock(&po->bind_lock); | |
ce06b03e DM |
3340 | if (po->num) |
3341 | register_prot_hook(sk); | |
808f5114 | 3342 | spin_unlock(&po->bind_lock); |
1da177e4 | 3343 | } |
1da177e4 LT |
3344 | break; |
3345 | } | |
3346 | } | |
808f5114 | 3347 | rcu_read_unlock(); |
1da177e4 LT |
3348 | return NOTIFY_DONE; |
3349 | } | |
3350 | ||
3351 | ||
3352 | static int packet_ioctl(struct socket *sock, unsigned int cmd, | |
3353 | unsigned long arg) | |
3354 | { | |
3355 | struct sock *sk = sock->sk; | |
3356 | ||
69e3c75f | 3357 | switch (cmd) { |
40d4e3df ED |
3358 | case SIOCOUTQ: |
3359 | { | |
3360 | int amount = sk_wmem_alloc_get(sk); | |
31e6d363 | 3361 | |
40d4e3df ED |
3362 | return put_user(amount, (int __user *)arg); |
3363 | } | |
3364 | case SIOCINQ: | |
3365 | { | |
3366 | struct sk_buff *skb; | |
3367 | int amount = 0; | |
3368 | ||
3369 | spin_lock_bh(&sk->sk_receive_queue.lock); | |
3370 | skb = skb_peek(&sk->sk_receive_queue); | |
3371 | if (skb) | |
3372 | amount = skb->len; | |
3373 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
3374 | return put_user(amount, (int __user *)arg); | |
3375 | } | |
3376 | case SIOCGSTAMP: | |
3377 | return sock_get_timestamp(sk, (struct timeval __user *)arg); | |
3378 | case SIOCGSTAMPNS: | |
3379 | return sock_get_timestampns(sk, (struct timespec __user *)arg); | |
1ce4f28b | 3380 | |
1da177e4 | 3381 | #ifdef CONFIG_INET |
40d4e3df ED |
3382 | case SIOCADDRT: |
3383 | case SIOCDELRT: | |
3384 | case SIOCDARP: | |
3385 | case SIOCGARP: | |
3386 | case SIOCSARP: | |
3387 | case SIOCGIFADDR: | |
3388 | case SIOCSIFADDR: | |
3389 | case SIOCGIFBRDADDR: | |
3390 | case SIOCSIFBRDADDR: | |
3391 | case SIOCGIFNETMASK: | |
3392 | case SIOCSIFNETMASK: | |
3393 | case SIOCGIFDSTADDR: | |
3394 | case SIOCSIFDSTADDR: | |
3395 | case SIOCSIFFLAGS: | |
40d4e3df | 3396 | return inet_dgram_ops.ioctl(sock, cmd, arg); |
1da177e4 LT |
3397 | #endif |
3398 | ||
40d4e3df ED |
3399 | default: |
3400 | return -ENOIOCTLCMD; | |
1da177e4 LT |
3401 | } |
3402 | return 0; | |
3403 | } | |
3404 | ||
40d4e3df | 3405 | static unsigned int packet_poll(struct file *file, struct socket *sock, |
1da177e4 LT |
3406 | poll_table *wait) |
3407 | { | |
3408 | struct sock *sk = sock->sk; | |
3409 | struct packet_sock *po = pkt_sk(sk); | |
3410 | unsigned int mask = datagram_poll(file, sock, wait); | |
3411 | ||
3412 | spin_lock_bh(&sk->sk_receive_queue.lock); | |
69e3c75f | 3413 | if (po->rx_ring.pg_vec) { |
f6fb8f10 | 3414 | if (!packet_previous_rx_frame(po, &po->rx_ring, |
3415 | TP_STATUS_KERNEL)) | |
1da177e4 LT |
3416 | mask |= POLLIN | POLLRDNORM; |
3417 | } | |
3418 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
69e3c75f JB |
3419 | spin_lock_bh(&sk->sk_write_queue.lock); |
3420 | if (po->tx_ring.pg_vec) { | |
3421 | if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) | |
3422 | mask |= POLLOUT | POLLWRNORM; | |
3423 | } | |
3424 | spin_unlock_bh(&sk->sk_write_queue.lock); | |
1da177e4 LT |
3425 | return mask; |
3426 | } | |
3427 | ||
3428 | ||
3429 | /* Dirty? Well, I still did not learn better way to account | |
3430 | * for user mmaps. | |
3431 | */ | |
3432 | ||
3433 | static void packet_mm_open(struct vm_area_struct *vma) | |
3434 | { | |
3435 | struct file *file = vma->vm_file; | |
40d4e3df | 3436 | struct socket *sock = file->private_data; |
1da177e4 | 3437 | struct sock *sk = sock->sk; |
1ce4f28b | 3438 | |
1da177e4 LT |
3439 | if (sk) |
3440 | atomic_inc(&pkt_sk(sk)->mapped); | |
3441 | } | |
3442 | ||
3443 | static void packet_mm_close(struct vm_area_struct *vma) | |
3444 | { | |
3445 | struct file *file = vma->vm_file; | |
40d4e3df | 3446 | struct socket *sock = file->private_data; |
1da177e4 | 3447 | struct sock *sk = sock->sk; |
1ce4f28b | 3448 | |
1da177e4 LT |
3449 | if (sk) |
3450 | atomic_dec(&pkt_sk(sk)->mapped); | |
3451 | } | |
3452 | ||
f0f37e2f | 3453 | static const struct vm_operations_struct packet_mmap_ops = { |
40d4e3df ED |
3454 | .open = packet_mm_open, |
3455 | .close = packet_mm_close, | |
1da177e4 LT |
3456 | }; |
3457 | ||
0e3125c7 NH |
3458 | static void free_pg_vec(struct pgv *pg_vec, unsigned int order, |
3459 | unsigned int len) | |
1da177e4 LT |
3460 | { |
3461 | int i; | |
3462 | ||
4ebf0ae2 | 3463 | for (i = 0; i < len; i++) { |
0e3125c7 | 3464 | if (likely(pg_vec[i].buffer)) { |
c56b4d90 | 3465 | if (is_vmalloc_addr(pg_vec[i].buffer)) |
0e3125c7 NH |
3466 | vfree(pg_vec[i].buffer); |
3467 | else | |
3468 | free_pages((unsigned long)pg_vec[i].buffer, | |
3469 | order); | |
3470 | pg_vec[i].buffer = NULL; | |
3471 | } | |
1da177e4 LT |
3472 | } |
3473 | kfree(pg_vec); | |
3474 | } | |
3475 | ||
eea49cc9 | 3476 | static char *alloc_one_pg_vec_page(unsigned long order) |
4ebf0ae2 | 3477 | { |
0e3125c7 NH |
3478 | char *buffer = NULL; |
3479 | gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | | |
3480 | __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; | |
3481 | ||
3482 | buffer = (char *) __get_free_pages(gfp_flags, order); | |
3483 | ||
3484 | if (buffer) | |
3485 | return buffer; | |
3486 | ||
3487 | /* | |
3488 | * __get_free_pages failed, fall back to vmalloc | |
3489 | */ | |
bbce5a59 | 3490 | buffer = vzalloc((1 << order) * PAGE_SIZE); |
719bfeaa | 3491 | |
0e3125c7 NH |
3492 | if (buffer) |
3493 | return buffer; | |
3494 | ||
3495 | /* | |
3496 | * vmalloc failed, lets dig into swap here | |
3497 | */ | |
0e3125c7 NH |
3498 | gfp_flags &= ~__GFP_NORETRY; |
3499 | buffer = (char *)__get_free_pages(gfp_flags, order); | |
3500 | if (buffer) | |
3501 | return buffer; | |
3502 | ||
3503 | /* | |
3504 | * complete and utter failure | |
3505 | */ | |
3506 | return NULL; | |
4ebf0ae2 DM |
3507 | } |
3508 | ||
0e3125c7 | 3509 | static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) |
4ebf0ae2 DM |
3510 | { |
3511 | unsigned int block_nr = req->tp_block_nr; | |
0e3125c7 | 3512 | struct pgv *pg_vec; |
4ebf0ae2 DM |
3513 | int i; |
3514 | ||
0e3125c7 | 3515 | pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL); |
4ebf0ae2 DM |
3516 | if (unlikely(!pg_vec)) |
3517 | goto out; | |
3518 | ||
3519 | for (i = 0; i < block_nr; i++) { | |
c56b4d90 | 3520 | pg_vec[i].buffer = alloc_one_pg_vec_page(order); |
0e3125c7 | 3521 | if (unlikely(!pg_vec[i].buffer)) |
4ebf0ae2 DM |
3522 | goto out_free_pgvec; |
3523 | } | |
3524 | ||
3525 | out: | |
3526 | return pg_vec; | |
3527 | ||
3528 | out_free_pgvec: | |
3529 | free_pg_vec(pg_vec, order, block_nr); | |
3530 | pg_vec = NULL; | |
3531 | goto out; | |
3532 | } | |
1da177e4 | 3533 | |
f6fb8f10 | 3534 | static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, |
69e3c75f | 3535 | int closing, int tx_ring) |
1da177e4 | 3536 | { |
0e3125c7 | 3537 | struct pgv *pg_vec = NULL; |
1da177e4 | 3538 | struct packet_sock *po = pkt_sk(sk); |
0e11c91e | 3539 | int was_running, order = 0; |
69e3c75f JB |
3540 | struct packet_ring_buffer *rb; |
3541 | struct sk_buff_head *rb_queue; | |
0e11c91e | 3542 | __be16 num; |
f6fb8f10 | 3543 | int err = -EINVAL; |
3544 | /* Added to avoid minimal code churn */ | |
3545 | struct tpacket_req *req = &req_u->req; | |
3546 | ||
3547 | /* Opening a Tx-ring is NOT supported in TPACKET_V3 */ | |
3548 | if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) { | |
3549 | WARN(1, "Tx-ring is not supported.\n"); | |
3550 | goto out; | |
3551 | } | |
1ce4f28b | 3552 | |
69e3c75f JB |
3553 | rb = tx_ring ? &po->tx_ring : &po->rx_ring; |
3554 | rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; | |
1da177e4 | 3555 | |
69e3c75f JB |
3556 | err = -EBUSY; |
3557 | if (!closing) { | |
3558 | if (atomic_read(&po->mapped)) | |
3559 | goto out; | |
3560 | if (atomic_read(&rb->pending)) | |
3561 | goto out; | |
3562 | } | |
1da177e4 | 3563 | |
69e3c75f JB |
3564 | if (req->tp_block_nr) { |
3565 | /* Sanity tests and some calculations */ | |
3566 | err = -EBUSY; | |
3567 | if (unlikely(rb->pg_vec)) | |
3568 | goto out; | |
1da177e4 | 3569 | |
bbd6ef87 PM |
3570 | switch (po->tp_version) { |
3571 | case TPACKET_V1: | |
3572 | po->tp_hdrlen = TPACKET_HDRLEN; | |
3573 | break; | |
3574 | case TPACKET_V2: | |
3575 | po->tp_hdrlen = TPACKET2_HDRLEN; | |
3576 | break; | |
f6fb8f10 | 3577 | case TPACKET_V3: |
3578 | po->tp_hdrlen = TPACKET3_HDRLEN; | |
3579 | break; | |
bbd6ef87 PM |
3580 | } |
3581 | ||
69e3c75f | 3582 | err = -EINVAL; |
4ebf0ae2 | 3583 | if (unlikely((int)req->tp_block_size <= 0)) |
69e3c75f | 3584 | goto out; |
4ebf0ae2 | 3585 | if (unlikely(req->tp_block_size & (PAGE_SIZE - 1))) |
69e3c75f | 3586 | goto out; |
8913336a | 3587 | if (unlikely(req->tp_frame_size < po->tp_hdrlen + |
69e3c75f JB |
3588 | po->tp_reserve)) |
3589 | goto out; | |
4ebf0ae2 | 3590 | if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) |
69e3c75f | 3591 | goto out; |
1da177e4 | 3592 | |
69e3c75f JB |
3593 | rb->frames_per_block = req->tp_block_size/req->tp_frame_size; |
3594 | if (unlikely(rb->frames_per_block <= 0)) | |
3595 | goto out; | |
3596 | if (unlikely((rb->frames_per_block * req->tp_block_nr) != | |
3597 | req->tp_frame_nr)) | |
3598 | goto out; | |
1da177e4 LT |
3599 | |
3600 | err = -ENOMEM; | |
4ebf0ae2 DM |
3601 | order = get_order(req->tp_block_size); |
3602 | pg_vec = alloc_pg_vec(req, order); | |
3603 | if (unlikely(!pg_vec)) | |
1da177e4 | 3604 | goto out; |
f6fb8f10 | 3605 | switch (po->tp_version) { |
3606 | case TPACKET_V3: | |
3607 | /* Transmit path is not supported. We checked | |
3608 | * it above but just being paranoid | |
3609 | */ | |
3610 | if (!tx_ring) | |
3611 | init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring); | |
3612 | break; | |
3613 | default: | |
3614 | break; | |
3615 | } | |
69e3c75f JB |
3616 | } |
3617 | /* Done */ | |
3618 | else { | |
3619 | err = -EINVAL; | |
4ebf0ae2 | 3620 | if (unlikely(req->tp_frame_nr)) |
69e3c75f | 3621 | goto out; |
1da177e4 LT |
3622 | } |
3623 | ||
3624 | lock_sock(sk); | |
3625 | ||
3626 | /* Detach socket from network */ | |
3627 | spin_lock(&po->bind_lock); | |
3628 | was_running = po->running; | |
3629 | num = po->num; | |
3630 | if (was_running) { | |
1da177e4 | 3631 | po->num = 0; |
ce06b03e | 3632 | __unregister_prot_hook(sk, false); |
1da177e4 LT |
3633 | } |
3634 | spin_unlock(&po->bind_lock); | |
1ce4f28b | 3635 | |
1da177e4 LT |
3636 | synchronize_net(); |
3637 | ||
3638 | err = -EBUSY; | |
905db440 | 3639 | mutex_lock(&po->pg_vec_lock); |
1da177e4 LT |
3640 | if (closing || atomic_read(&po->mapped) == 0) { |
3641 | err = 0; | |
69e3c75f | 3642 | spin_lock_bh(&rb_queue->lock); |
c053fd96 | 3643 | swap(rb->pg_vec, pg_vec); |
69e3c75f JB |
3644 | rb->frame_max = (req->tp_frame_nr - 1); |
3645 | rb->head = 0; | |
3646 | rb->frame_size = req->tp_frame_size; | |
3647 | spin_unlock_bh(&rb_queue->lock); | |
3648 | ||
c053fd96 CG |
3649 | swap(rb->pg_vec_order, order); |
3650 | swap(rb->pg_vec_len, req->tp_block_nr); | |
69e3c75f JB |
3651 | |
3652 | rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; | |
3653 | po->prot_hook.func = (po->rx_ring.pg_vec) ? | |
3654 | tpacket_rcv : packet_rcv; | |
3655 | skb_queue_purge(rb_queue); | |
1da177e4 | 3656 | if (atomic_read(&po->mapped)) |
40d4e3df ED |
3657 | pr_err("packet_mmap: vma is busy: %d\n", |
3658 | atomic_read(&po->mapped)); | |
1da177e4 | 3659 | } |
905db440 | 3660 | mutex_unlock(&po->pg_vec_lock); |
1da177e4 LT |
3661 | |
3662 | spin_lock(&po->bind_lock); | |
ce06b03e | 3663 | if (was_running) { |
1da177e4 | 3664 | po->num = num; |
ce06b03e | 3665 | register_prot_hook(sk); |
1da177e4 LT |
3666 | } |
3667 | spin_unlock(&po->bind_lock); | |
f6fb8f10 | 3668 | if (closing && (po->tp_version > TPACKET_V2)) { |
3669 | /* Because we don't support block-based V3 on tx-ring */ | |
3670 | if (!tx_ring) | |
3671 | prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue); | |
3672 | } | |
1da177e4 LT |
3673 | release_sock(sk); |
3674 | ||
1da177e4 LT |
3675 | if (pg_vec) |
3676 | free_pg_vec(pg_vec, order, req->tp_block_nr); | |
3677 | out: | |
3678 | return err; | |
3679 | } | |
3680 | ||
69e3c75f JB |
3681 | static int packet_mmap(struct file *file, struct socket *sock, |
3682 | struct vm_area_struct *vma) | |
1da177e4 LT |
3683 | { |
3684 | struct sock *sk = sock->sk; | |
3685 | struct packet_sock *po = pkt_sk(sk); | |
69e3c75f JB |
3686 | unsigned long size, expected_size; |
3687 | struct packet_ring_buffer *rb; | |
1da177e4 LT |
3688 | unsigned long start; |
3689 | int err = -EINVAL; | |
3690 | int i; | |
3691 | ||
3692 | if (vma->vm_pgoff) | |
3693 | return -EINVAL; | |
3694 | ||
905db440 | 3695 | mutex_lock(&po->pg_vec_lock); |
69e3c75f JB |
3696 | |
3697 | expected_size = 0; | |
3698 | for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { | |
3699 | if (rb->pg_vec) { | |
3700 | expected_size += rb->pg_vec_len | |
3701 | * rb->pg_vec_pages | |
3702 | * PAGE_SIZE; | |
3703 | } | |
3704 | } | |
3705 | ||
3706 | if (expected_size == 0) | |
1da177e4 | 3707 | goto out; |
69e3c75f JB |
3708 | |
3709 | size = vma->vm_end - vma->vm_start; | |
3710 | if (size != expected_size) | |
1da177e4 LT |
3711 | goto out; |
3712 | ||
1da177e4 | 3713 | start = vma->vm_start; |
69e3c75f JB |
3714 | for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { |
3715 | if (rb->pg_vec == NULL) | |
3716 | continue; | |
3717 | ||
3718 | for (i = 0; i < rb->pg_vec_len; i++) { | |
0e3125c7 NH |
3719 | struct page *page; |
3720 | void *kaddr = rb->pg_vec[i].buffer; | |
69e3c75f JB |
3721 | int pg_num; |
3722 | ||
c56b4d90 CG |
3723 | for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { |
3724 | page = pgv_to_page(kaddr); | |
69e3c75f JB |
3725 | err = vm_insert_page(vma, start, page); |
3726 | if (unlikely(err)) | |
3727 | goto out; | |
3728 | start += PAGE_SIZE; | |
0e3125c7 | 3729 | kaddr += PAGE_SIZE; |
69e3c75f | 3730 | } |
4ebf0ae2 | 3731 | } |
1da177e4 | 3732 | } |
69e3c75f | 3733 | |
4ebf0ae2 | 3734 | atomic_inc(&po->mapped); |
1da177e4 LT |
3735 | vma->vm_ops = &packet_mmap_ops; |
3736 | err = 0; | |
3737 | ||
3738 | out: | |
905db440 | 3739 | mutex_unlock(&po->pg_vec_lock); |
1da177e4 LT |
3740 | return err; |
3741 | } | |
1da177e4 | 3742 | |
90ddc4f0 | 3743 | static const struct proto_ops packet_ops_spkt = { |
1da177e4 LT |
3744 | .family = PF_PACKET, |
3745 | .owner = THIS_MODULE, | |
3746 | .release = packet_release, | |
3747 | .bind = packet_bind_spkt, | |
3748 | .connect = sock_no_connect, | |
3749 | .socketpair = sock_no_socketpair, | |
3750 | .accept = sock_no_accept, | |
3751 | .getname = packet_getname_spkt, | |
3752 | .poll = datagram_poll, | |
3753 | .ioctl = packet_ioctl, | |
3754 | .listen = sock_no_listen, | |
3755 | .shutdown = sock_no_shutdown, | |
3756 | .setsockopt = sock_no_setsockopt, | |
3757 | .getsockopt = sock_no_getsockopt, | |
3758 | .sendmsg = packet_sendmsg_spkt, | |
3759 | .recvmsg = packet_recvmsg, | |
3760 | .mmap = sock_no_mmap, | |
3761 | .sendpage = sock_no_sendpage, | |
3762 | }; | |
1da177e4 | 3763 | |
90ddc4f0 | 3764 | static const struct proto_ops packet_ops = { |
1da177e4 LT |
3765 | .family = PF_PACKET, |
3766 | .owner = THIS_MODULE, | |
3767 | .release = packet_release, | |
3768 | .bind = packet_bind, | |
3769 | .connect = sock_no_connect, | |
3770 | .socketpair = sock_no_socketpair, | |
3771 | .accept = sock_no_accept, | |
1ce4f28b | 3772 | .getname = packet_getname, |
1da177e4 LT |
3773 | .poll = packet_poll, |
3774 | .ioctl = packet_ioctl, | |
3775 | .listen = sock_no_listen, | |
3776 | .shutdown = sock_no_shutdown, | |
3777 | .setsockopt = packet_setsockopt, | |
3778 | .getsockopt = packet_getsockopt, | |
3779 | .sendmsg = packet_sendmsg, | |
3780 | .recvmsg = packet_recvmsg, | |
3781 | .mmap = packet_mmap, | |
3782 | .sendpage = sock_no_sendpage, | |
3783 | }; | |
3784 | ||
ec1b4cf7 | 3785 | static const struct net_proto_family packet_family_ops = { |
1da177e4 LT |
3786 | .family = PF_PACKET, |
3787 | .create = packet_create, | |
3788 | .owner = THIS_MODULE, | |
3789 | }; | |
3790 | ||
3791 | static struct notifier_block packet_netdev_notifier = { | |
40d4e3df | 3792 | .notifier_call = packet_notifier, |
1da177e4 LT |
3793 | }; |
3794 | ||
3795 | #ifdef CONFIG_PROC_FS | |
1da177e4 LT |
3796 | |
3797 | static void *packet_seq_start(struct seq_file *seq, loff_t *pos) | |
808f5114 | 3798 | __acquires(RCU) |
1da177e4 | 3799 | { |
e372c414 | 3800 | struct net *net = seq_file_net(seq); |
808f5114 | 3801 | |
3802 | rcu_read_lock(); | |
3803 | return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); | |
1da177e4 LT |
3804 | } |
3805 | ||
3806 | static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3807 | { | |
1bf40954 | 3808 | struct net *net = seq_file_net(seq); |
808f5114 | 3809 | return seq_hlist_next_rcu(v, &net->packet.sklist, pos); |
1da177e4 LT |
3810 | } |
3811 | ||
3812 | static void packet_seq_stop(struct seq_file *seq, void *v) | |
808f5114 | 3813 | __releases(RCU) |
1da177e4 | 3814 | { |
808f5114 | 3815 | rcu_read_unlock(); |
1da177e4 LT |
3816 | } |
3817 | ||
1ce4f28b | 3818 | static int packet_seq_show(struct seq_file *seq, void *v) |
1da177e4 LT |
3819 | { |
3820 | if (v == SEQ_START_TOKEN) | |
3821 | seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); | |
3822 | else { | |
b7ceabd9 | 3823 | struct sock *s = sk_entry(v); |
1da177e4 LT |
3824 | const struct packet_sock *po = pkt_sk(s); |
3825 | ||
3826 | seq_printf(seq, | |
71338aa7 | 3827 | "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", |
1da177e4 LT |
3828 | s, |
3829 | atomic_read(&s->sk_refcnt), | |
3830 | s->sk_type, | |
3831 | ntohs(po->num), | |
3832 | po->ifindex, | |
3833 | po->running, | |
3834 | atomic_read(&s->sk_rmem_alloc), | |
a7cb5a49 | 3835 | from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), |
40d4e3df | 3836 | sock_i_ino(s)); |
1da177e4 LT |
3837 | } |
3838 | ||
3839 | return 0; | |
3840 | } | |
3841 | ||
56b3d975 | 3842 | static const struct seq_operations packet_seq_ops = { |
1da177e4 LT |
3843 | .start = packet_seq_start, |
3844 | .next = packet_seq_next, | |
3845 | .stop = packet_seq_stop, | |
3846 | .show = packet_seq_show, | |
3847 | }; | |
3848 | ||
3849 | static int packet_seq_open(struct inode *inode, struct file *file) | |
3850 | { | |
e372c414 DL |
3851 | return seq_open_net(inode, file, &packet_seq_ops, |
3852 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
3853 | } |
3854 | ||
da7071d7 | 3855 | static const struct file_operations packet_seq_fops = { |
1da177e4 LT |
3856 | .owner = THIS_MODULE, |
3857 | .open = packet_seq_open, | |
3858 | .read = seq_read, | |
3859 | .llseek = seq_lseek, | |
e372c414 | 3860 | .release = seq_release_net, |
1da177e4 LT |
3861 | }; |
3862 | ||
3863 | #endif | |
3864 | ||
2c8c1e72 | 3865 | static int __net_init packet_net_init(struct net *net) |
d12d01d6 | 3866 | { |
0fa7fa98 | 3867 | mutex_init(&net->packet.sklist_lock); |
2aaef4e4 | 3868 | INIT_HLIST_HEAD(&net->packet.sklist); |
d12d01d6 | 3869 | |
d4beaa66 | 3870 | if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops)) |
d12d01d6 DL |
3871 | return -ENOMEM; |
3872 | ||
3873 | return 0; | |
3874 | } | |
3875 | ||
2c8c1e72 | 3876 | static void __net_exit packet_net_exit(struct net *net) |
d12d01d6 | 3877 | { |
ece31ffd | 3878 | remove_proc_entry("packet", net->proc_net); |
d12d01d6 DL |
3879 | } |
3880 | ||
3881 | static struct pernet_operations packet_net_ops = { | |
3882 | .init = packet_net_init, | |
3883 | .exit = packet_net_exit, | |
3884 | }; | |
3885 | ||
3886 | ||
1da177e4 LT |
3887 | static void __exit packet_exit(void) |
3888 | { | |
1da177e4 | 3889 | unregister_netdevice_notifier(&packet_netdev_notifier); |
d12d01d6 | 3890 | unregister_pernet_subsys(&packet_net_ops); |
1da177e4 LT |
3891 | sock_unregister(PF_PACKET); |
3892 | proto_unregister(&packet_proto); | |
3893 | } | |
3894 | ||
3895 | static int __init packet_init(void) | |
3896 | { | |
3897 | int rc = proto_register(&packet_proto, 0); | |
3898 | ||
3899 | if (rc != 0) | |
3900 | goto out; | |
3901 | ||
3902 | sock_register(&packet_family_ops); | |
d12d01d6 | 3903 | register_pernet_subsys(&packet_net_ops); |
1da177e4 | 3904 | register_netdevice_notifier(&packet_netdev_notifier); |
1da177e4 LT |
3905 | out: |
3906 | return rc; | |
3907 | } | |
3908 | ||
3909 | module_init(packet_init); | |
3910 | module_exit(packet_exit); | |
3911 | MODULE_LICENSE("GPL"); | |
3912 | MODULE_ALIAS_NETPROTO(PF_PACKET); |