]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
4 | * operating system. INET is implemented using the BSD Socket | |
5 | * interface as the means of communication with the user level. | |
6 | * | |
7 | * PACKET - implements raw packet sockets. | |
8 | * | |
02c30a84 | 9 | * Authors: Ross Biro |
1da177e4 LT |
10 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
11 | * Alan Cox, <gw4pts@gw4pts.ampr.org> | |
12 | * | |
1ce4f28b | 13 | * Fixes: |
1da177e4 LT |
14 | * Alan Cox : verify_area() now used correctly |
15 | * Alan Cox : new skbuff lists, look ma no backlogs! | |
16 | * Alan Cox : tidied skbuff lists. | |
17 | * Alan Cox : Now uses generic datagram routines I | |
18 | * added. Also fixed the peek/read crash | |
19 | * from all old Linux datagram code. | |
20 | * Alan Cox : Uses the improved datagram code. | |
21 | * Alan Cox : Added NULL's for socket options. | |
22 | * Alan Cox : Re-commented the code. | |
23 | * Alan Cox : Use new kernel side addressing | |
24 | * Rob Janssen : Correct MTU usage. | |
25 | * Dave Platt : Counter leaks caused by incorrect | |
26 | * interrupt locking and some slightly | |
27 | * dubious gcc output. Can you read | |
28 | * compiler: it said _VOLATILE_ | |
29 | * Richard Kooijman : Timestamp fixes. | |
30 | * Alan Cox : New buffers. Use sk->mac.raw. | |
31 | * Alan Cox : sendmsg/recvmsg support. | |
32 | * Alan Cox : Protocol setting support | |
33 | * Alexey Kuznetsov : Untied from IPv4 stack. | |
34 | * Cyrus Durgin : Fixed kerneld for kmod. | |
35 | * Michal Ostrowski : Module initialization cleanup. | |
1ce4f28b | 36 | * Ulises Alonso : Frame number limit removal and |
1da177e4 | 37 | * packet_set_ring memory leak. |
0fb375fb EB |
38 | * Eric Biederman : Allow for > 8 byte hardware addresses. |
39 | * The convention is that longer addresses | |
40 | * will simply extend the hardware address | |
1ce4f28b | 41 | * byte arrays at the end of sockaddr_ll |
0fb375fb | 42 | * and packet_mreq. |
69e3c75f | 43 | * Johann Baudy : Added TX RING. |
f6fb8f10 | 44 | * Chetan Loke : Implemented TPACKET_V3 block abstraction |
45 | * layer. | |
46 | * Copyright (C) 2011, <lokec@ccs.neu.edu> | |
1da177e4 | 47 | */ |
1ce4f28b | 48 | |
dc41c4a9 BS |
49 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
50 | ||
cc69837f | 51 | #include <linux/ethtool.h> |
1da177e4 | 52 | #include <linux/types.h> |
1da177e4 | 53 | #include <linux/mm.h> |
4fc268d2 | 54 | #include <linux/capability.h> |
1da177e4 LT |
55 | #include <linux/fcntl.h> |
56 | #include <linux/socket.h> | |
57 | #include <linux/in.h> | |
58 | #include <linux/inet.h> | |
59 | #include <linux/netdevice.h> | |
60 | #include <linux/if_packet.h> | |
61 | #include <linux/wireless.h> | |
ffbc6111 | 62 | #include <linux/kernel.h> |
1da177e4 | 63 | #include <linux/kmod.h> |
5a0e3ad6 | 64 | #include <linux/slab.h> |
0e3125c7 | 65 | #include <linux/vmalloc.h> |
457c4cbc | 66 | #include <net/net_namespace.h> |
1da177e4 LT |
67 | #include <net/ip.h> |
68 | #include <net/protocol.h> | |
69 | #include <linux/skbuff.h> | |
70 | #include <net/sock.h> | |
71 | #include <linux/errno.h> | |
72 | #include <linux/timer.h> | |
7c0f6ba6 | 73 | #include <linux/uaccess.h> |
1da177e4 LT |
74 | #include <asm/ioctls.h> |
75 | #include <asm/page.h> | |
a1f8e7f7 | 76 | #include <asm/cacheflush.h> |
1da177e4 LT |
77 | #include <asm/io.h> |
78 | #include <linux/proc_fs.h> | |
79 | #include <linux/seq_file.h> | |
80 | #include <linux/poll.h> | |
81 | #include <linux/module.h> | |
82 | #include <linux/init.h> | |
905db440 | 83 | #include <linux/mutex.h> |
05423b24 | 84 | #include <linux/if_vlan.h> |
bfd5f4a3 | 85 | #include <linux/virtio_net.h> |
ed85b565 | 86 | #include <linux/errqueue.h> |
614f60fa | 87 | #include <linux/net_tstamp.h> |
b0138408 | 88 | #include <linux/percpu.h> |
1da177e4 LT |
89 | #ifdef CONFIG_INET |
90 | #include <net/inet_common.h> | |
91 | #endif | |
47dceb8e | 92 | #include <linux/bpf.h> |
719c44d3 | 93 | #include <net/compat.h> |
1da177e4 | 94 | |
2787b04b PE |
95 | #include "internal.h" |
96 | ||
1da177e4 LT |
97 | /* |
98 | Assumptions: | |
d5496990 EB |
99 | - If the device has no dev->header_ops->create, there is no LL header |
100 | visible above the device. In this case, its hard_header_len should be 0. | |
b4c58814 XH |
101 | The device may prepend its own header internally. In this case, its |
102 | needed_headroom should be set to the space needed for it to add its | |
103 | internal header. | |
104 | For example, a WiFi driver pretending to be an Ethernet driver should | |
105 | set its hard_header_len to be the Ethernet header length, and set its | |
106 | needed_headroom to be (the real WiFi header length - the fake Ethernet | |
107 | header length). | |
1da177e4 LT |
108 | - packet socket receives packets with pulled ll header, |
109 | so that SOCK_RAW should push it back. | |
110 | ||
111 | On receive: | |
112 | ----------- | |
113 | ||
d5496990 | 114 | Incoming, dev_has_header(dev) == true |
b0e380b1 ACM |
115 | mac_header -> ll header |
116 | data -> data | |
1da177e4 | 117 | |
d5496990 | 118 | Outgoing, dev_has_header(dev) == true |
b0e380b1 ACM |
119 | mac_header -> ll header |
120 | data -> ll header | |
1da177e4 | 121 | |
d5496990 | 122 | Incoming, dev_has_header(dev) == false |
b79a80bd XH |
123 | mac_header -> data |
124 | However drivers often make it point to the ll header. | |
125 | This is incorrect because the ll header should be invisible to us. | |
b0e380b1 | 126 | data -> data |
1da177e4 | 127 | |
d5496990 | 128 | Outgoing, dev_has_header(dev) == false |
b79a80bd | 129 | mac_header -> data. ll header is invisible to us. |
b0e380b1 | 130 | data -> data |
1da177e4 LT |
131 | |
132 | Resume | |
d5496990 | 133 | If dev_has_header(dev) == false we are unable to restore the ll header, |
b79a80bd | 134 | because it is invisible to us. |
1da177e4 LT |
135 | |
136 | ||
137 | On transmit: | |
138 | ------------ | |
139 | ||
21c85974 | 140 | dev_has_header(dev) == true |
b0e380b1 ACM |
141 | mac_header -> ll header |
142 | data -> ll header | |
1da177e4 | 143 | |
21c85974 | 144 | dev_has_header(dev) == false (ll header is invisible to us) |
b0e380b1 ACM |
145 | mac_header -> data |
146 | data -> data | |
1da177e4 | 147 | |
09599729 | 148 | We should set network_header on output to the correct position, |
1da177e4 LT |
149 | packet classifier depends on it. |
150 | */ | |
151 | ||
1da177e4 LT |
152 | /* Private packet socket structures. */ |
153 | ||
0fb375fb EB |
154 | /* identical to struct packet_mreq except it has |
155 | * a longer address field. | |
156 | */ | |
40d4e3df | 157 | struct packet_mreq_max { |
0fb375fb EB |
158 | int mr_ifindex; |
159 | unsigned short mr_type; | |
160 | unsigned short mr_alen; | |
161 | unsigned char mr_address[MAX_ADDR_LEN]; | |
1da177e4 | 162 | }; |
a2efcfa0 | 163 | |
184f489e DB |
164 | union tpacket_uhdr { |
165 | struct tpacket_hdr *h1; | |
166 | struct tpacket2_hdr *h2; | |
167 | struct tpacket3_hdr *h3; | |
168 | void *raw; | |
169 | }; | |
170 | ||
f6fb8f10 | 171 | static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, |
69e3c75f JB |
172 | int closing, int tx_ring); |
173 | ||
f6fb8f10 | 174 | #define V3_ALIGNMENT (8) |
175 | ||
bc59ba39 | 176 | #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) |
f6fb8f10 | 177 | |
178 | #define BLK_PLUS_PRIV(sz_of_priv) \ | |
179 | (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) | |
180 | ||
f6fb8f10 | 181 | #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) |
182 | #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) | |
183 | #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) | |
184 | #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) | |
185 | #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) | |
186 | #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) | |
f6fb8f10 | 187 | |
69e3c75f | 188 | struct packet_sock; |
77f65ebd WB |
189 | static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, |
190 | struct packet_type *pt, struct net_device *orig_dev); | |
1da177e4 | 191 | |
f6fb8f10 | 192 | static void *packet_previous_frame(struct packet_sock *po, |
193 | struct packet_ring_buffer *rb, | |
194 | int status); | |
195 | static void packet_increment_head(struct packet_ring_buffer *buff); | |
878cd3ba | 196 | static int prb_curr_blk_in_use(struct tpacket_block_desc *); |
bc59ba39 | 197 | static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, |
f6fb8f10 | 198 | struct packet_sock *); |
bc59ba39 | 199 | static void prb_retire_current_block(struct tpacket_kbdq_core *, |
f6fb8f10 | 200 | struct packet_sock *, unsigned int status); |
bc59ba39 | 201 | static int prb_queue_frozen(struct tpacket_kbdq_core *); |
202 | static void prb_open_block(struct tpacket_kbdq_core *, | |
203 | struct tpacket_block_desc *); | |
17bfd8c8 | 204 | static void prb_retire_rx_blk_timer_expired(struct timer_list *); |
bc59ba39 | 205 | static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); |
bc59ba39 | 206 | static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); |
207 | static void prb_clear_rxhash(struct tpacket_kbdq_core *, | |
208 | struct tpacket3_hdr *); | |
209 | static void prb_fill_vlan_info(struct tpacket_kbdq_core *, | |
210 | struct tpacket3_hdr *); | |
1da177e4 | 211 | static void packet_flush_mclist(struct sock *sk); |
865b03f2 | 212 | static u16 packet_pick_tx_queue(struct sk_buff *skb); |
1da177e4 | 213 | |
ffbc6111 | 214 | struct packet_skb_cb { |
ffbc6111 HX |
215 | union { |
216 | struct sockaddr_pkt pkt; | |
2472d761 EB |
217 | union { |
218 | /* Trick: alias skb original length with | |
219 | * ll.sll_family and ll.protocol in order | |
220 | * to save room. | |
221 | */ | |
222 | unsigned int origlen; | |
223 | struct sockaddr_ll ll; | |
224 | }; | |
ffbc6111 HX |
225 | } sa; |
226 | }; | |
227 | ||
d3869efe DW |
228 | #define vio_le() virtio_legacy_is_little_endian() |
229 | ||
ffbc6111 | 230 | #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) |
8dc41944 | 231 | |
bc59ba39 | 232 | #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) |
f6fb8f10 | 233 | #define GET_PBLOCK_DESC(x, bid) \ |
bc59ba39 | 234 | ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) |
f6fb8f10 | 235 | #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ |
bc59ba39 | 236 | ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) |
f6fb8f10 | 237 | #define GET_NEXT_PRB_BLK_NUM(x) \ |
238 | (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ | |
239 | ((x)->kactive_blk_num+1) : 0) | |
240 | ||
dc99f600 DM |
241 | static void __fanout_unlink(struct sock *sk, struct packet_sock *po); |
242 | static void __fanout_link(struct sock *sk, struct packet_sock *po); | |
243 | ||
d346a3fa DB |
244 | static int packet_direct_xmit(struct sk_buff *skb) |
245 | { | |
865b03f2 | 246 | return dev_direct_xmit(skb, packet_pick_tx_queue(skb)); |
d346a3fa DB |
247 | } |
248 | ||
66e56cd4 DB |
249 | static struct net_device *packet_cached_dev_get(struct packet_sock *po) |
250 | { | |
251 | struct net_device *dev; | |
252 | ||
253 | rcu_read_lock(); | |
254 | dev = rcu_dereference(po->cached_dev); | |
1160dfa1 | 255 | dev_hold(dev); |
66e56cd4 DB |
256 | rcu_read_unlock(); |
257 | ||
258 | return dev; | |
259 | } | |
260 | ||
261 | static void packet_cached_dev_assign(struct packet_sock *po, | |
262 | struct net_device *dev) | |
263 | { | |
264 | rcu_assign_pointer(po->cached_dev, dev); | |
265 | } | |
266 | ||
267 | static void packet_cached_dev_reset(struct packet_sock *po) | |
268 | { | |
269 | RCU_INIT_POINTER(po->cached_dev, NULL); | |
270 | } | |
271 | ||
d346a3fa DB |
272 | static bool packet_use_direct_xmit(const struct packet_sock *po) |
273 | { | |
274 | return po->xmit == packet_direct_xmit; | |
275 | } | |
276 | ||
865b03f2 | 277 | static u16 packet_pick_tx_queue(struct sk_buff *skb) |
0fd5d57b | 278 | { |
865b03f2 | 279 | struct net_device *dev = skb->dev; |
0fd5d57b | 280 | const struct net_device_ops *ops = dev->netdev_ops; |
b71b5837 | 281 | int cpu = raw_smp_processor_id(); |
0fd5d57b DB |
282 | u16 queue_index; |
283 | ||
b71b5837 PA |
284 | #ifdef CONFIG_XPS |
285 | skb->sender_cpu = cpu + 1; | |
286 | #endif | |
287 | skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues); | |
0fd5d57b | 288 | if (ops->ndo_select_queue) { |
a350ecce | 289 | queue_index = ops->ndo_select_queue(dev, skb, NULL); |
0fd5d57b DB |
290 | queue_index = netdev_cap_txqueue(dev, queue_index); |
291 | } else { | |
b71b5837 | 292 | queue_index = netdev_pick_tx(dev, skb, NULL); |
0fd5d57b DB |
293 | } |
294 | ||
865b03f2 | 295 | return queue_index; |
0fd5d57b DB |
296 | } |
297 | ||
a6361f0c | 298 | /* __register_prot_hook must be invoked through register_prot_hook |
ce06b03e DM |
299 | * or from a context in which asynchronous accesses to the packet |
300 | * socket is not possible (packet_create()). | |
301 | */ | |
a6361f0c | 302 | static void __register_prot_hook(struct sock *sk) |
ce06b03e DM |
303 | { |
304 | struct packet_sock *po = pkt_sk(sk); | |
e40526cb | 305 | |
ce06b03e | 306 | if (!po->running) { |
66e56cd4 | 307 | if (po->fanout) |
dc99f600 | 308 | __fanout_link(sk, po); |
66e56cd4 | 309 | else |
dc99f600 | 310 | dev_add_pack(&po->prot_hook); |
e40526cb | 311 | |
ce06b03e DM |
312 | sock_hold(sk); |
313 | po->running = 1; | |
314 | } | |
315 | } | |
316 | ||
a6361f0c WB |
317 | static void register_prot_hook(struct sock *sk) |
318 | { | |
319 | lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); | |
320 | __register_prot_hook(sk); | |
321 | } | |
322 | ||
323 | /* If the sync parameter is true, we will temporarily drop | |
ce06b03e DM |
324 | * the po->bind_lock and do a synchronize_net to make sure no |
325 | * asynchronous packet processing paths still refer to the elements | |
326 | * of po->prot_hook. If the sync parameter is false, it is the | |
327 | * callers responsibility to take care of this. | |
328 | */ | |
329 | static void __unregister_prot_hook(struct sock *sk, bool sync) | |
330 | { | |
331 | struct packet_sock *po = pkt_sk(sk); | |
332 | ||
a6361f0c WB |
333 | lockdep_assert_held_once(&po->bind_lock); |
334 | ||
ce06b03e | 335 | po->running = 0; |
66e56cd4 DB |
336 | |
337 | if (po->fanout) | |
dc99f600 | 338 | __fanout_unlink(sk, po); |
66e56cd4 | 339 | else |
dc99f600 | 340 | __dev_remove_pack(&po->prot_hook); |
e40526cb | 341 | |
ce06b03e DM |
342 | __sock_put(sk); |
343 | ||
344 | if (sync) { | |
345 | spin_unlock(&po->bind_lock); | |
346 | synchronize_net(); | |
347 | spin_lock(&po->bind_lock); | |
348 | } | |
349 | } | |
350 | ||
351 | static void unregister_prot_hook(struct sock *sk, bool sync) | |
352 | { | |
353 | struct packet_sock *po = pkt_sk(sk); | |
354 | ||
355 | if (po->running) | |
356 | __unregister_prot_hook(sk, sync); | |
357 | } | |
358 | ||
6e58040b | 359 | static inline struct page * __pure pgv_to_page(void *addr) |
0af55bb5 CG |
360 | { |
361 | if (is_vmalloc_addr(addr)) | |
362 | return vmalloc_to_page(addr); | |
363 | return virt_to_page(addr); | |
364 | } | |
365 | ||
69e3c75f | 366 | static void __packet_set_status(struct packet_sock *po, void *frame, int status) |
1da177e4 | 367 | { |
184f489e | 368 | union tpacket_uhdr h; |
1da177e4 | 369 | |
69e3c75f | 370 | h.raw = frame; |
bbd6ef87 PM |
371 | switch (po->tp_version) { |
372 | case TPACKET_V1: | |
69e3c75f | 373 | h.h1->tp_status = status; |
0af55bb5 | 374 | flush_dcache_page(pgv_to_page(&h.h1->tp_status)); |
bbd6ef87 PM |
375 | break; |
376 | case TPACKET_V2: | |
69e3c75f | 377 | h.h2->tp_status = status; |
0af55bb5 | 378 | flush_dcache_page(pgv_to_page(&h.h2->tp_status)); |
bbd6ef87 | 379 | break; |
f6fb8f10 | 380 | case TPACKET_V3: |
7f953ab2 SV |
381 | h.h3->tp_status = status; |
382 | flush_dcache_page(pgv_to_page(&h.h3->tp_status)); | |
383 | break; | |
69e3c75f | 384 | default: |
f6fb8f10 | 385 | WARN(1, "TPACKET version not supported.\n"); |
69e3c75f | 386 | BUG(); |
bbd6ef87 | 387 | } |
69e3c75f JB |
388 | |
389 | smp_wmb(); | |
bbd6ef87 PM |
390 | } |
391 | ||
96f657e6 | 392 | static int __packet_get_status(const struct packet_sock *po, void *frame) |
bbd6ef87 | 393 | { |
184f489e | 394 | union tpacket_uhdr h; |
bbd6ef87 | 395 | |
69e3c75f JB |
396 | smp_rmb(); |
397 | ||
bbd6ef87 PM |
398 | h.raw = frame; |
399 | switch (po->tp_version) { | |
400 | case TPACKET_V1: | |
0af55bb5 | 401 | flush_dcache_page(pgv_to_page(&h.h1->tp_status)); |
69e3c75f | 402 | return h.h1->tp_status; |
bbd6ef87 | 403 | case TPACKET_V2: |
0af55bb5 | 404 | flush_dcache_page(pgv_to_page(&h.h2->tp_status)); |
69e3c75f | 405 | return h.h2->tp_status; |
f6fb8f10 | 406 | case TPACKET_V3: |
7f953ab2 SV |
407 | flush_dcache_page(pgv_to_page(&h.h3->tp_status)); |
408 | return h.h3->tp_status; | |
69e3c75f | 409 | default: |
f6fb8f10 | 410 | WARN(1, "TPACKET version not supported.\n"); |
69e3c75f JB |
411 | BUG(); |
412 | return 0; | |
bbd6ef87 | 413 | } |
1da177e4 | 414 | } |
69e3c75f | 415 | |
d413fcb4 | 416 | static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts, |
b9c32fb2 | 417 | unsigned int flags) |
7a51384c DB |
418 | { |
419 | struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); | |
420 | ||
68a360e8 WB |
421 | if (shhwtstamps && |
422 | (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && | |
d413fcb4 | 423 | ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts)) |
68a360e8 | 424 | return TP_STATUS_TS_RAW_HARDWARE; |
7a51384c | 425 | |
171c3b15 RS |
426 | if ((flags & SOF_TIMESTAMPING_SOFTWARE) && |
427 | ktime_to_timespec64_cond(skb->tstamp, ts)) | |
b9c32fb2 | 428 | return TP_STATUS_TS_SOFTWARE; |
7a51384c | 429 | |
b9c32fb2 | 430 | return 0; |
7a51384c DB |
431 | } |
432 | ||
b9c32fb2 DB |
433 | static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, |
434 | struct sk_buff *skb) | |
2e31396f WB |
435 | { |
436 | union tpacket_uhdr h; | |
d413fcb4 | 437 | struct timespec64 ts; |
b9c32fb2 | 438 | __u32 ts_status; |
2e31396f | 439 | |
b9c32fb2 DB |
440 | if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) |
441 | return 0; | |
2e31396f WB |
442 | |
443 | h.raw = frame; | |
d413fcb4 AB |
444 | /* |
445 | * versions 1 through 3 overflow the timestamps in y2106, since they | |
446 | * all store the seconds in a 32-bit unsigned integer. | |
447 | * If we create a version 4, that should have a 64-bit timestamp, | |
448 | * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit | |
449 | * nanoseconds. | |
450 | */ | |
2e31396f WB |
451 | switch (po->tp_version) { |
452 | case TPACKET_V1: | |
453 | h.h1->tp_sec = ts.tv_sec; | |
454 | h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; | |
455 | break; | |
456 | case TPACKET_V2: | |
457 | h.h2->tp_sec = ts.tv_sec; | |
458 | h.h2->tp_nsec = ts.tv_nsec; | |
459 | break; | |
460 | case TPACKET_V3: | |
57ea884b DB |
461 | h.h3->tp_sec = ts.tv_sec; |
462 | h.h3->tp_nsec = ts.tv_nsec; | |
463 | break; | |
2e31396f WB |
464 | default: |
465 | WARN(1, "TPACKET version not supported.\n"); | |
466 | BUG(); | |
467 | } | |
468 | ||
469 | /* one flush is safe, as both fields always lie on the same cacheline */ | |
470 | flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); | |
471 | smp_wmb(); | |
b9c32fb2 DB |
472 | |
473 | return ts_status; | |
2e31396f WB |
474 | } |
475 | ||
d4b5bd98 ED |
476 | static void *packet_lookup_frame(const struct packet_sock *po, |
477 | const struct packet_ring_buffer *rb, | |
478 | unsigned int position, | |
479 | int status) | |
69e3c75f JB |
480 | { |
481 | unsigned int pg_vec_pos, frame_offset; | |
184f489e | 482 | union tpacket_uhdr h; |
69e3c75f JB |
483 | |
484 | pg_vec_pos = position / rb->frames_per_block; | |
485 | frame_offset = position % rb->frames_per_block; | |
486 | ||
0e3125c7 NH |
487 | h.raw = rb->pg_vec[pg_vec_pos].buffer + |
488 | (frame_offset * rb->frame_size); | |
69e3c75f JB |
489 | |
490 | if (status != __packet_get_status(po, h.raw)) | |
491 | return NULL; | |
492 | ||
493 | return h.raw; | |
494 | } | |
495 | ||
eea49cc9 | 496 | static void *packet_current_frame(struct packet_sock *po, |
69e3c75f JB |
497 | struct packet_ring_buffer *rb, |
498 | int status) | |
499 | { | |
500 | return packet_lookup_frame(po, rb, rb->head, status); | |
501 | } | |
502 | ||
bc59ba39 | 503 | static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 504 | { |
505 | del_timer_sync(&pkc->retire_blk_timer); | |
506 | } | |
507 | ||
508 | static void prb_shutdown_retire_blk_timer(struct packet_sock *po, | |
f6fb8f10 | 509 | struct sk_buff_head *rb_queue) |
510 | { | |
bc59ba39 | 511 | struct tpacket_kbdq_core *pkc; |
f6fb8f10 | 512 | |
73d0fcf2 | 513 | pkc = GET_PBDQC_FROM_RB(&po->rx_ring); |
f6fb8f10 | 514 | |
ec6f809f | 515 | spin_lock_bh(&rb_queue->lock); |
f6fb8f10 | 516 | pkc->delete_blk_timer = 1; |
ec6f809f | 517 | spin_unlock_bh(&rb_queue->lock); |
f6fb8f10 | 518 | |
519 | prb_del_retire_blk_timer(pkc); | |
520 | } | |
521 | ||
e8e85cc5 | 522 | static void prb_setup_retire_blk_timer(struct packet_sock *po) |
f6fb8f10 | 523 | { |
bc59ba39 | 524 | struct tpacket_kbdq_core *pkc; |
f6fb8f10 | 525 | |
e8e85cc5 | 526 | pkc = GET_PBDQC_FROM_RB(&po->rx_ring); |
17bfd8c8 KC |
527 | timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, |
528 | 0); | |
529 | pkc->retire_blk_timer.expires = jiffies; | |
f6fb8f10 | 530 | } |
531 | ||
532 | static int prb_calc_retire_blk_tmo(struct packet_sock *po, | |
533 | int blk_size_in_bytes) | |
534 | { | |
535 | struct net_device *dev; | |
0914d2bb | 536 | unsigned int mbits, div; |
7cad1bac | 537 | struct ethtool_link_ksettings ecmd; |
4bc71cb9 | 538 | int err; |
f6fb8f10 | 539 | |
4bc71cb9 JP |
540 | rtnl_lock(); |
541 | dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); | |
542 | if (unlikely(!dev)) { | |
543 | rtnl_unlock(); | |
f6fb8f10 | 544 | return DEFAULT_PRB_RETIRE_TOV; |
4bc71cb9 | 545 | } |
7cad1bac | 546 | err = __ethtool_get_link_ksettings(dev, &ecmd); |
4bc71cb9 | 547 | rtnl_unlock(); |
0914d2bb | 548 | if (err) |
b43d1f9f | 549 | return DEFAULT_PRB_RETIRE_TOV; |
f6fb8f10 | 550 | |
0914d2bb MW |
551 | /* If the link speed is so slow you don't really |
552 | * need to worry about perf anyways | |
553 | */ | |
554 | if (ecmd.base.speed < SPEED_1000 || | |
555 | ecmd.base.speed == SPEED_UNKNOWN) | |
556 | return DEFAULT_PRB_RETIRE_TOV; | |
557 | ||
558 | div = ecmd.base.speed / 1000; | |
f6fb8f10 | 559 | mbits = (blk_size_in_bytes * 8) / (1024 * 1024); |
560 | ||
561 | if (div) | |
562 | mbits /= div; | |
563 | ||
f6fb8f10 | 564 | if (div) |
0914d2bb MW |
565 | return mbits + 1; |
566 | return mbits; | |
f6fb8f10 | 567 | } |
568 | ||
bc59ba39 | 569 | static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, |
f6fb8f10 | 570 | union tpacket_req_u *req_u) |
571 | { | |
572 | p1->feature_req_word = req_u->req3.tp_feature_req_word; | |
573 | } | |
574 | ||
575 | static void init_prb_bdqc(struct packet_sock *po, | |
576 | struct packet_ring_buffer *rb, | |
577 | struct pgv *pg_vec, | |
e8e85cc5 | 578 | union tpacket_req_u *req_u) |
f6fb8f10 | 579 | { |
22781a5b | 580 | struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); |
bc59ba39 | 581 | struct tpacket_block_desc *pbd; |
f6fb8f10 | 582 | |
583 | memset(p1, 0x0, sizeof(*p1)); | |
584 | ||
585 | p1->knxt_seq_num = 1; | |
586 | p1->pkbdq = pg_vec; | |
bc59ba39 | 587 | pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; |
e3192690 | 588 | p1->pkblk_start = pg_vec[0].buffer; |
f6fb8f10 | 589 | p1->kblk_size = req_u->req3.tp_block_size; |
590 | p1->knum_blocks = req_u->req3.tp_block_nr; | |
591 | p1->hdrlen = po->tp_hdrlen; | |
592 | p1->version = po->tp_version; | |
593 | p1->last_kactive_blk_num = 0; | |
ee80fbf3 | 594 | po->stats.stats3.tp_freeze_q_cnt = 0; |
f6fb8f10 | 595 | if (req_u->req3.tp_retire_blk_tov) |
596 | p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; | |
597 | else | |
598 | p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, | |
599 | req_u->req3.tp_block_size); | |
600 | p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); | |
601 | p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; | |
632ca50f | 602 | rwlock_init(&p1->blk_fill_in_prog_lock); |
f6fb8f10 | 603 | |
dc808110 | 604 | p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); |
f6fb8f10 | 605 | prb_init_ft_ops(p1, req_u); |
e8e85cc5 | 606 | prb_setup_retire_blk_timer(po); |
f6fb8f10 | 607 | prb_open_block(p1, pbd); |
608 | } | |
609 | ||
610 | /* Do NOT update the last_blk_num first. | |
611 | * Assumes sk_buff_head lock is held. | |
612 | */ | |
bc59ba39 | 613 | static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 614 | { |
615 | mod_timer(&pkc->retire_blk_timer, | |
616 | jiffies + pkc->tov_in_jiffies); | |
617 | pkc->last_kactive_blk_num = pkc->kactive_blk_num; | |
618 | } | |
619 | ||
620 | /* | |
621 | * Timer logic: | |
622 | * 1) We refresh the timer only when we open a block. | |
623 | * By doing this we don't waste cycles refreshing the timer | |
624 | * on packet-by-packet basis. | |
625 | * | |
626 | * With a 1MB block-size, on a 1Gbps line, it will take | |
627 | * i) ~8 ms to fill a block + ii) memcpy etc. | |
628 | * In this cut we are not accounting for the memcpy time. | |
629 | * | |
630 | * So, if the user sets the 'tmo' to 10ms then the timer | |
631 | * will never fire while the block is still getting filled | |
632 | * (which is what we want). However, the user could choose | |
633 | * to close a block early and that's fine. | |
634 | * | |
635 | * But when the timer does fire, we check whether or not to refresh it. | |
636 | * Since the tmo granularity is in msecs, it is not too expensive | |
637 | * to refresh the timer, lets say every '8' msecs. | |
638 | * Either the user can set the 'tmo' or we can derive it based on | |
639 | * a) line-speed and b) block-size. | |
640 | * prb_calc_retire_blk_tmo() calculates the tmo. | |
641 | * | |
642 | */ | |
17bfd8c8 | 643 | static void prb_retire_rx_blk_timer_expired(struct timer_list *t) |
f6fb8f10 | 644 | { |
17bfd8c8 KC |
645 | struct packet_sock *po = |
646 | from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer); | |
22781a5b | 647 | struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); |
f6fb8f10 | 648 | unsigned int frozen; |
bc59ba39 | 649 | struct tpacket_block_desc *pbd; |
f6fb8f10 | 650 | |
651 | spin_lock(&po->sk.sk_receive_queue.lock); | |
652 | ||
653 | frozen = prb_queue_frozen(pkc); | |
654 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); | |
655 | ||
656 | if (unlikely(pkc->delete_blk_timer)) | |
657 | goto out; | |
658 | ||
659 | /* We only need to plug the race when the block is partially filled. | |
660 | * tpacket_rcv: | |
661 | * lock(); increment BLOCK_NUM_PKTS; unlock() | |
662 | * copy_bits() is in progress ... | |
663 | * timer fires on other cpu: | |
664 | * we can't retire the current block because copy_bits | |
665 | * is in progress. | |
666 | * | |
667 | */ | |
668 | if (BLOCK_NUM_PKTS(pbd)) { | |
632ca50f JO |
669 | /* Waiting for skb_copy_bits to finish... */ |
670 | write_lock(&pkc->blk_fill_in_prog_lock); | |
671 | write_unlock(&pkc->blk_fill_in_prog_lock); | |
f6fb8f10 | 672 | } |
673 | ||
674 | if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { | |
675 | if (!frozen) { | |
41a50d62 AD |
676 | if (!BLOCK_NUM_PKTS(pbd)) { |
677 | /* An empty block. Just refresh the timer. */ | |
678 | goto refresh_timer; | |
679 | } | |
f6fb8f10 | 680 | prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); |
681 | if (!prb_dispatch_next_block(pkc, po)) | |
682 | goto refresh_timer; | |
683 | else | |
684 | goto out; | |
685 | } else { | |
686 | /* Case 1. Queue was frozen because user-space was | |
687 | * lagging behind. | |
688 | */ | |
878cd3ba | 689 | if (prb_curr_blk_in_use(pbd)) { |
f6fb8f10 | 690 | /* |
691 | * Ok, user-space is still behind. | |
692 | * So just refresh the timer. | |
693 | */ | |
694 | goto refresh_timer; | |
695 | } else { | |
696 | /* Case 2. queue was frozen,user-space caught up, | |
697 | * now the link went idle && the timer fired. | |
698 | * We don't have a block to close.So we open this | |
699 | * block and restart the timer. | |
700 | * opening a block thaws the queue,restarts timer | |
701 | * Thawing/timer-refresh is a side effect. | |
702 | */ | |
703 | prb_open_block(pkc, pbd); | |
704 | goto out; | |
705 | } | |
706 | } | |
707 | } | |
708 | ||
709 | refresh_timer: | |
710 | _prb_refresh_rx_retire_blk_timer(pkc); | |
711 | ||
712 | out: | |
713 | spin_unlock(&po->sk.sk_receive_queue.lock); | |
714 | } | |
715 | ||
eea49cc9 | 716 | static void prb_flush_block(struct tpacket_kbdq_core *pkc1, |
bc59ba39 | 717 | struct tpacket_block_desc *pbd1, __u32 status) |
f6fb8f10 | 718 | { |
719 | /* Flush everything minus the block header */ | |
720 | ||
721 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 | |
722 | u8 *start, *end; | |
723 | ||
724 | start = (u8 *)pbd1; | |
725 | ||
726 | /* Skip the block header(we know header WILL fit in 4K) */ | |
727 | start += PAGE_SIZE; | |
728 | ||
729 | end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); | |
730 | for (; start < end; start += PAGE_SIZE) | |
731 | flush_dcache_page(pgv_to_page(start)); | |
732 | ||
733 | smp_wmb(); | |
734 | #endif | |
735 | ||
736 | /* Now update the block status. */ | |
737 | ||
738 | BLOCK_STATUS(pbd1) = status; | |
739 | ||
740 | /* Flush the block header */ | |
741 | ||
742 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 | |
743 | start = (u8 *)pbd1; | |
744 | flush_dcache_page(pgv_to_page(start)); | |
745 | ||
746 | smp_wmb(); | |
747 | #endif | |
748 | } | |
749 | ||
750 | /* | |
751 | * Side effect: | |
752 | * | |
753 | * 1) flush the block | |
754 | * 2) Increment active_blk_num | |
755 | * | |
756 | * Note:We DONT refresh the timer on purpose. | |
757 | * Because almost always the next block will be opened. | |
758 | */ | |
bc59ba39 | 759 | static void prb_close_block(struct tpacket_kbdq_core *pkc1, |
760 | struct tpacket_block_desc *pbd1, | |
f6fb8f10 | 761 | struct packet_sock *po, unsigned int stat) |
762 | { | |
763 | __u32 status = TP_STATUS_USER | stat; | |
764 | ||
765 | struct tpacket3_hdr *last_pkt; | |
bc59ba39 | 766 | struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; |
da413eec | 767 | struct sock *sk = &po->sk; |
f6fb8f10 | 768 | |
8e8e2951 | 769 | if (atomic_read(&po->tp_drops)) |
f6fb8f10 | 770 | status |= TP_STATUS_LOSING; |
771 | ||
772 | last_pkt = (struct tpacket3_hdr *)pkc1->prev; | |
773 | last_pkt->tp_next_offset = 0; | |
774 | ||
775 | /* Get the ts of the last pkt */ | |
776 | if (BLOCK_NUM_PKTS(pbd1)) { | |
777 | h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; | |
778 | h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; | |
779 | } else { | |
41a50d62 AD |
780 | /* Ok, we tmo'd - so get the current time. |
781 | * | |
782 | * It shouldn't really happen as we don't close empty | |
783 | * blocks. See prb_retire_rx_blk_timer_expired(). | |
784 | */ | |
d413fcb4 AB |
785 | struct timespec64 ts; |
786 | ktime_get_real_ts64(&ts); | |
f6fb8f10 | 787 | h1->ts_last_pkt.ts_sec = ts.tv_sec; |
788 | h1->ts_last_pkt.ts_nsec = ts.tv_nsec; | |
789 | } | |
790 | ||
791 | smp_wmb(); | |
792 | ||
793 | /* Flush the block */ | |
794 | prb_flush_block(pkc1, pbd1, status); | |
795 | ||
da413eec DC |
796 | sk->sk_data_ready(sk); |
797 | ||
f6fb8f10 | 798 | pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); |
799 | } | |
800 | ||
eea49cc9 | 801 | static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 802 | { |
803 | pkc->reset_pending_on_curr_blk = 0; | |
804 | } | |
805 | ||
806 | /* | |
807 | * Side effect of opening a block: | |
808 | * | |
809 | * 1) prb_queue is thawed. | |
810 | * 2) retire_blk_timer is refreshed. | |
811 | * | |
812 | */ | |
bc59ba39 | 813 | static void prb_open_block(struct tpacket_kbdq_core *pkc1, |
814 | struct tpacket_block_desc *pbd1) | |
f6fb8f10 | 815 | { |
d413fcb4 | 816 | struct timespec64 ts; |
bc59ba39 | 817 | struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; |
f6fb8f10 | 818 | |
819 | smp_rmb(); | |
820 | ||
8da3056c DB |
821 | /* We could have just memset this but we will lose the |
822 | * flexibility of making the priv area sticky | |
823 | */ | |
f6fb8f10 | 824 | |
8da3056c DB |
825 | BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; |
826 | BLOCK_NUM_PKTS(pbd1) = 0; | |
827 | BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); | |
f6fb8f10 | 828 | |
d413fcb4 | 829 | ktime_get_real_ts64(&ts); |
8da3056c DB |
830 | |
831 | h1->ts_first_pkt.ts_sec = ts.tv_sec; | |
832 | h1->ts_first_pkt.ts_nsec = ts.tv_nsec; | |
f6fb8f10 | 833 | |
8da3056c DB |
834 | pkc1->pkblk_start = (char *)pbd1; |
835 | pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); | |
836 | ||
837 | BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); | |
838 | BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; | |
839 | ||
840 | pbd1->version = pkc1->version; | |
841 | pkc1->prev = pkc1->nxt_offset; | |
842 | pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; | |
843 | ||
844 | prb_thaw_queue(pkc1); | |
845 | _prb_refresh_rx_retire_blk_timer(pkc1); | |
846 | ||
847 | smp_wmb(); | |
f6fb8f10 | 848 | } |
849 | ||
850 | /* | |
851 | * Queue freeze logic: | |
852 | * 1) Assume tp_block_nr = 8 blocks. | |
853 | * 2) At time 't0', user opens Rx ring. | |
854 | * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 | |
855 | * 4) user-space is either sleeping or processing block '0'. | |
856 | * 5) tpacket_rcv is currently filling block '7', since there is no space left, | |
857 | * it will close block-7,loop around and try to fill block '0'. | |
858 | * call-flow: | |
859 | * __packet_lookup_frame_in_block | |
860 | * prb_retire_current_block() | |
861 | * prb_dispatch_next_block() | |
862 | * |->(BLOCK_STATUS == USER) evaluates to true | |
863 | * 5.1) Since block-0 is currently in-use, we just freeze the queue. | |
864 | * 6) Now there are two cases: | |
865 | * 6.1) Link goes idle right after the queue is frozen. | |
866 | * But remember, the last open_block() refreshed the timer. | |
867 | * When this timer expires,it will refresh itself so that we can | |
868 | * re-open block-0 in near future. | |
869 | * 6.2) Link is busy and keeps on receiving packets. This is a simple | |
870 | * case and __packet_lookup_frame_in_block will check if block-0 | |
871 | * is free and can now be re-used. | |
872 | */ | |
eea49cc9 | 873 | static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 874 | struct packet_sock *po) |
875 | { | |
876 | pkc->reset_pending_on_curr_blk = 1; | |
ee80fbf3 | 877 | po->stats.stats3.tp_freeze_q_cnt++; |
f6fb8f10 | 878 | } |
879 | ||
880 | #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) | |
881 | ||
882 | /* | |
883 | * If the next block is free then we will dispatch it | |
884 | * and return a good offset. | |
885 | * Else, we will freeze the queue. | |
886 | * So, caller must check the return value. | |
887 | */ | |
bc59ba39 | 888 | static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 889 | struct packet_sock *po) |
890 | { | |
bc59ba39 | 891 | struct tpacket_block_desc *pbd; |
f6fb8f10 | 892 | |
893 | smp_rmb(); | |
894 | ||
895 | /* 1. Get current block num */ | |
896 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); | |
897 | ||
898 | /* 2. If this block is currently in_use then freeze the queue */ | |
899 | if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { | |
900 | prb_freeze_queue(pkc, po); | |
901 | return NULL; | |
902 | } | |
903 | ||
904 | /* | |
905 | * 3. | |
906 | * open this block and return the offset where the first packet | |
907 | * needs to get stored. | |
908 | */ | |
909 | prb_open_block(pkc, pbd); | |
910 | return (void *)pkc->nxt_offset; | |
911 | } | |
912 | ||
bc59ba39 | 913 | static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 914 | struct packet_sock *po, unsigned int status) |
915 | { | |
bc59ba39 | 916 | struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); |
f6fb8f10 | 917 | |
918 | /* retire/close the current block */ | |
919 | if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { | |
920 | /* | |
921 | * Plug the case where copy_bits() is in progress on | |
922 | * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't | |
923 | * have space to copy the pkt in the current block and | |
924 | * called prb_retire_current_block() | |
925 | * | |
926 | * We don't need to worry about the TMO case because | |
927 | * the timer-handler already handled this case. | |
928 | */ | |
929 | if (!(status & TP_STATUS_BLK_TMO)) { | |
632ca50f JO |
930 | /* Waiting for skb_copy_bits to finish... */ |
931 | write_lock(&pkc->blk_fill_in_prog_lock); | |
932 | write_unlock(&pkc->blk_fill_in_prog_lock); | |
f6fb8f10 | 933 | } |
934 | prb_close_block(pkc, pbd, po, status); | |
935 | return; | |
936 | } | |
f6fb8f10 | 937 | } |
938 | ||
878cd3ba | 939 | static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) |
f6fb8f10 | 940 | { |
941 | return TP_STATUS_USER & BLOCK_STATUS(pbd); | |
942 | } | |
943 | ||
eea49cc9 | 944 | static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 945 | { |
946 | return pkc->reset_pending_on_curr_blk; | |
947 | } | |
948 | ||
eea49cc9 | 949 | static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) |
88fd1cb8 | 950 | __releases(&pkc->blk_fill_in_prog_lock) |
f6fb8f10 | 951 | { |
bc59ba39 | 952 | struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); |
632ca50f JO |
953 | |
954 | read_unlock(&pkc->blk_fill_in_prog_lock); | |
f6fb8f10 | 955 | } |
956 | ||
eea49cc9 | 957 | static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 958 | struct tpacket3_hdr *ppd) |
959 | { | |
3958afa1 | 960 | ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); |
f6fb8f10 | 961 | } |
962 | ||
eea49cc9 | 963 | static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 964 | struct tpacket3_hdr *ppd) |
965 | { | |
966 | ppd->hv1.tp_rxhash = 0; | |
967 | } | |
968 | ||
eea49cc9 | 969 | static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 970 | struct tpacket3_hdr *ppd) |
971 | { | |
df8a39de JP |
972 | if (skb_vlan_tag_present(pkc->skb)) { |
973 | ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); | |
a0cdfcf3 AW |
974 | ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); |
975 | ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; | |
f6fb8f10 | 976 | } else { |
9e67030a | 977 | ppd->hv1.tp_vlan_tci = 0; |
a0cdfcf3 | 978 | ppd->hv1.tp_vlan_tpid = 0; |
9e67030a | 979 | ppd->tp_status = TP_STATUS_AVAILABLE; |
f6fb8f10 | 980 | } |
981 | } | |
982 | ||
bc59ba39 | 983 | static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 984 | struct tpacket3_hdr *ppd) |
985 | { | |
a0cdfcf3 | 986 | ppd->hv1.tp_padding = 0; |
f6fb8f10 | 987 | prb_fill_vlan_info(pkc, ppd); |
988 | ||
989 | if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) | |
990 | prb_fill_rxhash(pkc, ppd); | |
991 | else | |
992 | prb_clear_rxhash(pkc, ppd); | |
993 | } | |
994 | ||
eea49cc9 | 995 | static void prb_fill_curr_block(char *curr, |
bc59ba39 | 996 | struct tpacket_kbdq_core *pkc, |
997 | struct tpacket_block_desc *pbd, | |
f6fb8f10 | 998 | unsigned int len) |
88fd1cb8 | 999 | __acquires(&pkc->blk_fill_in_prog_lock) |
f6fb8f10 | 1000 | { |
1001 | struct tpacket3_hdr *ppd; | |
1002 | ||
1003 | ppd = (struct tpacket3_hdr *)curr; | |
1004 | ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); | |
1005 | pkc->prev = curr; | |
1006 | pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); | |
1007 | BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); | |
1008 | BLOCK_NUM_PKTS(pbd) += 1; | |
632ca50f | 1009 | read_lock(&pkc->blk_fill_in_prog_lock); |
f6fb8f10 | 1010 | prb_run_all_ft_ops(pkc, ppd); |
1011 | } | |
1012 | ||
1013 | /* Assumes caller has the sk->rx_queue.lock */ | |
1014 | static void *__packet_lookup_frame_in_block(struct packet_sock *po, | |
1015 | struct sk_buff *skb, | |
f6fb8f10 | 1016 | unsigned int len |
1017 | ) | |
1018 | { | |
bc59ba39 | 1019 | struct tpacket_kbdq_core *pkc; |
1020 | struct tpacket_block_desc *pbd; | |
f6fb8f10 | 1021 | char *curr, *end; |
1022 | ||
e3192690 | 1023 | pkc = GET_PBDQC_FROM_RB(&po->rx_ring); |
f6fb8f10 | 1024 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); |
1025 | ||
1026 | /* Queue is frozen when user space is lagging behind */ | |
1027 | if (prb_queue_frozen(pkc)) { | |
1028 | /* | |
1029 | * Check if that last block which caused the queue to freeze, | |
1030 | * is still in_use by user-space. | |
1031 | */ | |
878cd3ba | 1032 | if (prb_curr_blk_in_use(pbd)) { |
f6fb8f10 | 1033 | /* Can't record this packet */ |
1034 | return NULL; | |
1035 | } else { | |
1036 | /* | |
1037 | * Ok, the block was released by user-space. | |
1038 | * Now let's open that block. | |
1039 | * opening a block also thaws the queue. | |
1040 | * Thawing is a side effect. | |
1041 | */ | |
1042 | prb_open_block(pkc, pbd); | |
1043 | } | |
1044 | } | |
1045 | ||
1046 | smp_mb(); | |
1047 | curr = pkc->nxt_offset; | |
1048 | pkc->skb = skb; | |
e3192690 | 1049 | end = (char *)pbd + pkc->kblk_size; |
f6fb8f10 | 1050 | |
1051 | /* first try the current block */ | |
1052 | if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { | |
1053 | prb_fill_curr_block(curr, pkc, pbd, len); | |
1054 | return (void *)curr; | |
1055 | } | |
1056 | ||
1057 | /* Ok, close the current block */ | |
1058 | prb_retire_current_block(pkc, po, 0); | |
1059 | ||
1060 | /* Now, try to dispatch the next block */ | |
1061 | curr = (char *)prb_dispatch_next_block(pkc, po); | |
1062 | if (curr) { | |
1063 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); | |
1064 | prb_fill_curr_block(curr, pkc, pbd, len); | |
1065 | return (void *)curr; | |
1066 | } | |
1067 | ||
1068 | /* | |
1069 | * No free blocks are available.user_space hasn't caught up yet. | |
1070 | * Queue was just frozen and now this packet will get dropped. | |
1071 | */ | |
1072 | return NULL; | |
1073 | } | |
1074 | ||
eea49cc9 | 1075 | static void *packet_current_rx_frame(struct packet_sock *po, |
f6fb8f10 | 1076 | struct sk_buff *skb, |
1077 | int status, unsigned int len) | |
1078 | { | |
1079 | char *curr = NULL; | |
1080 | switch (po->tp_version) { | |
1081 | case TPACKET_V1: | |
1082 | case TPACKET_V2: | |
1083 | curr = packet_lookup_frame(po, &po->rx_ring, | |
1084 | po->rx_ring.head, status); | |
1085 | return curr; | |
1086 | case TPACKET_V3: | |
46088059 | 1087 | return __packet_lookup_frame_in_block(po, skb, len); |
f6fb8f10 | 1088 | default: |
1089 | WARN(1, "TPACKET version not supported\n"); | |
1090 | BUG(); | |
99aa3473 | 1091 | return NULL; |
f6fb8f10 | 1092 | } |
1093 | } | |
1094 | ||
dcf70cef ED |
1095 | static void *prb_lookup_block(const struct packet_sock *po, |
1096 | const struct packet_ring_buffer *rb, | |
1097 | unsigned int idx, | |
1098 | int status) | |
f6fb8f10 | 1099 | { |
bc59ba39 | 1100 | struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); |
77f65ebd | 1101 | struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); |
f6fb8f10 | 1102 | |
1103 | if (status != BLOCK_STATUS(pbd)) | |
1104 | return NULL; | |
1105 | return pbd; | |
1106 | } | |
1107 | ||
eea49cc9 | 1108 | static int prb_previous_blk_num(struct packet_ring_buffer *rb) |
f6fb8f10 | 1109 | { |
1110 | unsigned int prev; | |
1111 | if (rb->prb_bdqc.kactive_blk_num) | |
1112 | prev = rb->prb_bdqc.kactive_blk_num-1; | |
1113 | else | |
1114 | prev = rb->prb_bdqc.knum_blocks-1; | |
1115 | return prev; | |
1116 | } | |
1117 | ||
1118 | /* Assumes caller has held the rx_queue.lock */ | |
eea49cc9 | 1119 | static void *__prb_previous_block(struct packet_sock *po, |
f6fb8f10 | 1120 | struct packet_ring_buffer *rb, |
1121 | int status) | |
1122 | { | |
1123 | unsigned int previous = prb_previous_blk_num(rb); | |
1124 | return prb_lookup_block(po, rb, previous, status); | |
1125 | } | |
1126 | ||
eea49cc9 | 1127 | static void *packet_previous_rx_frame(struct packet_sock *po, |
f6fb8f10 | 1128 | struct packet_ring_buffer *rb, |
1129 | int status) | |
1130 | { | |
1131 | if (po->tp_version <= TPACKET_V2) | |
1132 | return packet_previous_frame(po, rb, status); | |
1133 | ||
1134 | return __prb_previous_block(po, rb, status); | |
1135 | } | |
1136 | ||
eea49cc9 | 1137 | static void packet_increment_rx_head(struct packet_sock *po, |
f6fb8f10 | 1138 | struct packet_ring_buffer *rb) |
1139 | { | |
1140 | switch (po->tp_version) { | |
1141 | case TPACKET_V1: | |
1142 | case TPACKET_V2: | |
1143 | return packet_increment_head(rb); | |
1144 | case TPACKET_V3: | |
1145 | default: | |
1146 | WARN(1, "TPACKET version not supported.\n"); | |
1147 | BUG(); | |
1148 | return; | |
1149 | } | |
1150 | } | |
1151 | ||
eea49cc9 | 1152 | static void *packet_previous_frame(struct packet_sock *po, |
69e3c75f JB |
1153 | struct packet_ring_buffer *rb, |
1154 | int status) | |
1155 | { | |
1156 | unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; | |
1157 | return packet_lookup_frame(po, rb, previous, status); | |
1158 | } | |
1159 | ||
eea49cc9 | 1160 | static void packet_increment_head(struct packet_ring_buffer *buff) |
69e3c75f JB |
1161 | { |
1162 | buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; | |
1163 | } | |
1164 | ||
b0138408 DB |
1165 | static void packet_inc_pending(struct packet_ring_buffer *rb) |
1166 | { | |
1167 | this_cpu_inc(*rb->pending_refcnt); | |
1168 | } | |
1169 | ||
1170 | static void packet_dec_pending(struct packet_ring_buffer *rb) | |
1171 | { | |
1172 | this_cpu_dec(*rb->pending_refcnt); | |
1173 | } | |
1174 | ||
1175 | static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) | |
1176 | { | |
1177 | unsigned int refcnt = 0; | |
1178 | int cpu; | |
1179 | ||
1180 | /* We don't use pending refcount in rx_ring. */ | |
1181 | if (rb->pending_refcnt == NULL) | |
1182 | return 0; | |
1183 | ||
1184 | for_each_possible_cpu(cpu) | |
1185 | refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); | |
1186 | ||
1187 | return refcnt; | |
1188 | } | |
1189 | ||
1190 | static int packet_alloc_pending(struct packet_sock *po) | |
1191 | { | |
1192 | po->rx_ring.pending_refcnt = NULL; | |
1193 | ||
1194 | po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); | |
1195 | if (unlikely(po->tx_ring.pending_refcnt == NULL)) | |
1196 | return -ENOBUFS; | |
1197 | ||
1198 | return 0; | |
1199 | } | |
1200 | ||
1201 | static void packet_free_pending(struct packet_sock *po) | |
1202 | { | |
1203 | free_percpu(po->tx_ring.pending_refcnt); | |
1204 | } | |
1205 | ||
9954729b WB |
1206 | #define ROOM_POW_OFF 2 |
1207 | #define ROOM_NONE 0x0 | |
1208 | #define ROOM_LOW 0x1 | |
1209 | #define ROOM_NORMAL 0x2 | |
1210 | ||
d4b5bd98 | 1211 | static bool __tpacket_has_room(const struct packet_sock *po, int pow_off) |
77f65ebd | 1212 | { |
9954729b WB |
1213 | int idx, len; |
1214 | ||
d4b5bd98 ED |
1215 | len = READ_ONCE(po->rx_ring.frame_max) + 1; |
1216 | idx = READ_ONCE(po->rx_ring.head); | |
9954729b WB |
1217 | if (pow_off) |
1218 | idx += len >> pow_off; | |
1219 | if (idx >= len) | |
1220 | idx -= len; | |
1221 | return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); | |
1222 | } | |
1223 | ||
dcf70cef | 1224 | static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off) |
9954729b WB |
1225 | { |
1226 | int idx, len; | |
1227 | ||
dcf70cef ED |
1228 | len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks); |
1229 | idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num); | |
9954729b WB |
1230 | if (pow_off) |
1231 | idx += len >> pow_off; | |
1232 | if (idx >= len) | |
1233 | idx -= len; | |
1234 | return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); | |
1235 | } | |
77f65ebd | 1236 | |
0338a145 ED |
1237 | static int __packet_rcv_has_room(const struct packet_sock *po, |
1238 | const struct sk_buff *skb) | |
9954729b | 1239 | { |
0338a145 | 1240 | const struct sock *sk = &po->sk; |
9954729b WB |
1241 | int ret = ROOM_NONE; |
1242 | ||
1243 | if (po->prot_hook.func != tpacket_rcv) { | |
0338a145 ED |
1244 | int rcvbuf = READ_ONCE(sk->sk_rcvbuf); |
1245 | int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc) | |
1246 | - (skb ? skb->truesize : 0); | |
1247 | ||
1248 | if (avail > (rcvbuf >> ROOM_POW_OFF)) | |
9954729b WB |
1249 | return ROOM_NORMAL; |
1250 | else if (avail > 0) | |
1251 | return ROOM_LOW; | |
1252 | else | |
1253 | return ROOM_NONE; | |
1254 | } | |
77f65ebd | 1255 | |
9954729b WB |
1256 | if (po->tp_version == TPACKET_V3) { |
1257 | if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) | |
1258 | ret = ROOM_NORMAL; | |
1259 | else if (__tpacket_v3_has_room(po, 0)) | |
1260 | ret = ROOM_LOW; | |
1261 | } else { | |
1262 | if (__tpacket_has_room(po, ROOM_POW_OFF)) | |
1263 | ret = ROOM_NORMAL; | |
1264 | else if (__tpacket_has_room(po, 0)) | |
1265 | ret = ROOM_LOW; | |
1266 | } | |
2ccdbaa6 WB |
1267 | |
1268 | return ret; | |
1269 | } | |
1270 | ||
1271 | static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) | |
1272 | { | |
3a2bb84e | 1273 | int pressure, ret; |
2ccdbaa6 | 1274 | |
54d7c01d | 1275 | ret = __packet_rcv_has_room(po, skb); |
3a2bb84e ED |
1276 | pressure = ret != ROOM_NORMAL; |
1277 | ||
1278 | if (READ_ONCE(po->pressure) != pressure) | |
1279 | WRITE_ONCE(po->pressure, pressure); | |
77f65ebd | 1280 | |
9954729b | 1281 | return ret; |
77f65ebd WB |
1282 | } |
1283 | ||
9bb6cd65 ED |
1284 | static void packet_rcv_try_clear_pressure(struct packet_sock *po) |
1285 | { | |
1286 | if (READ_ONCE(po->pressure) && | |
1287 | __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) | |
1288 | WRITE_ONCE(po->pressure, 0); | |
1289 | } | |
1290 | ||
1da177e4 LT |
1291 | static void packet_sock_destruct(struct sock *sk) |
1292 | { | |
ed85b565 RC |
1293 | skb_queue_purge(&sk->sk_error_queue); |
1294 | ||
547b792c | 1295 | WARN_ON(atomic_read(&sk->sk_rmem_alloc)); |
14afee4b | 1296 | WARN_ON(refcount_read(&sk->sk_wmem_alloc)); |
1da177e4 LT |
1297 | |
1298 | if (!sock_flag(sk, SOCK_DEAD)) { | |
40d4e3df | 1299 | pr_err("Attempt to release alive packet socket: %p\n", sk); |
1da177e4 LT |
1300 | return; |
1301 | } | |
1302 | ||
17ab56a2 | 1303 | sk_refcnt_debug_dec(sk); |
1da177e4 LT |
1304 | } |
1305 | ||
3b3a5b0a WB |
1306 | static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) |
1307 | { | |
b756ad92 ED |
1308 | u32 *history = po->rollover->history; |
1309 | u32 victim, rxhash; | |
3b3a5b0a WB |
1310 | int i, count = 0; |
1311 | ||
1312 | rxhash = skb_get_hash(skb); | |
1313 | for (i = 0; i < ROLLOVER_HLEN; i++) | |
b756ad92 | 1314 | if (READ_ONCE(history[i]) == rxhash) |
3b3a5b0a WB |
1315 | count++; |
1316 | ||
b756ad92 ED |
1317 | victim = prandom_u32() % ROLLOVER_HLEN; |
1318 | ||
1319 | /* Avoid dirtying the cache line if possible */ | |
1320 | if (READ_ONCE(history[victim]) != rxhash) | |
1321 | WRITE_ONCE(history[victim], rxhash); | |
1322 | ||
3b3a5b0a WB |
1323 | return count > (ROLLOVER_HLEN >> 1); |
1324 | } | |
1325 | ||
77f65ebd WB |
1326 | static unsigned int fanout_demux_hash(struct packet_fanout *f, |
1327 | struct sk_buff *skb, | |
1328 | unsigned int num) | |
dc99f600 | 1329 | { |
eb70db87 | 1330 | return reciprocal_scale(__skb_get_hash_symmetric(skb), num); |
dc99f600 DM |
1331 | } |
1332 | ||
77f65ebd WB |
1333 | static unsigned int fanout_demux_lb(struct packet_fanout *f, |
1334 | struct sk_buff *skb, | |
1335 | unsigned int num) | |
dc99f600 | 1336 | { |
468479e6 | 1337 | unsigned int val = atomic_inc_return(&f->rr_cur); |
dc99f600 | 1338 | |
468479e6 | 1339 | return val % num; |
77f65ebd WB |
1340 | } |
1341 | ||
1342 | static unsigned int fanout_demux_cpu(struct packet_fanout *f, | |
1343 | struct sk_buff *skb, | |
1344 | unsigned int num) | |
1345 | { | |
1346 | return smp_processor_id() % num; | |
dc99f600 DM |
1347 | } |
1348 | ||
5df0ddfb DB |
1349 | static unsigned int fanout_demux_rnd(struct packet_fanout *f, |
1350 | struct sk_buff *skb, | |
1351 | unsigned int num) | |
1352 | { | |
f337db64 | 1353 | return prandom_u32_max(num); |
5df0ddfb DB |
1354 | } |
1355 | ||
77f65ebd WB |
1356 | static unsigned int fanout_demux_rollover(struct packet_fanout *f, |
1357 | struct sk_buff *skb, | |
ad377cab | 1358 | unsigned int idx, bool try_self, |
77f65ebd | 1359 | unsigned int num) |
95ec3eb4 | 1360 | { |
4633c9e0 | 1361 | struct packet_sock *po, *po_next, *po_skip = NULL; |
a9b63918 | 1362 | unsigned int i, j, room = ROOM_NONE; |
95ec3eb4 | 1363 | |
94f633ea | 1364 | po = pkt_sk(rcu_dereference(f->arr[idx])); |
3b3a5b0a WB |
1365 | |
1366 | if (try_self) { | |
1367 | room = packet_rcv_has_room(po, skb); | |
1368 | if (room == ROOM_NORMAL || | |
1369 | (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) | |
1370 | return idx; | |
4633c9e0 | 1371 | po_skip = po; |
3b3a5b0a | 1372 | } |
ad377cab | 1373 | |
0648ab70 | 1374 | i = j = min_t(int, po->rollover->sock, num - 1); |
77f65ebd | 1375 | do { |
94f633ea | 1376 | po_next = pkt_sk(rcu_dereference(f->arr[i])); |
3a2bb84e | 1377 | if (po_next != po_skip && !READ_ONCE(po_next->pressure) && |
2ccdbaa6 | 1378 | packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { |
77f65ebd | 1379 | if (i != j) |
0648ab70 | 1380 | po->rollover->sock = i; |
a9b63918 WB |
1381 | atomic_long_inc(&po->rollover->num); |
1382 | if (room == ROOM_LOW) | |
1383 | atomic_long_inc(&po->rollover->num_huge); | |
77f65ebd WB |
1384 | return i; |
1385 | } | |
ad377cab | 1386 | |
77f65ebd WB |
1387 | if (++i == num) |
1388 | i = 0; | |
1389 | } while (i != j); | |
1390 | ||
a9b63918 | 1391 | atomic_long_inc(&po->rollover->num_failed); |
77f65ebd WB |
1392 | return idx; |
1393 | } | |
1394 | ||
2d36097d NH |
1395 | static unsigned int fanout_demux_qm(struct packet_fanout *f, |
1396 | struct sk_buff *skb, | |
1397 | unsigned int num) | |
1398 | { | |
1399 | return skb_get_queue_mapping(skb) % num; | |
1400 | } | |
1401 | ||
47dceb8e WB |
1402 | static unsigned int fanout_demux_bpf(struct packet_fanout *f, |
1403 | struct sk_buff *skb, | |
1404 | unsigned int num) | |
1405 | { | |
1406 | struct bpf_prog *prog; | |
1407 | unsigned int ret = 0; | |
1408 | ||
1409 | rcu_read_lock(); | |
1410 | prog = rcu_dereference(f->bpf_prog); | |
1411 | if (prog) | |
ff936a04 | 1412 | ret = bpf_prog_run_clear_cb(prog, skb) % num; |
47dceb8e WB |
1413 | rcu_read_unlock(); |
1414 | ||
1415 | return ret; | |
1416 | } | |
1417 | ||
77f65ebd WB |
1418 | static bool fanout_has_flag(struct packet_fanout *f, u16 flag) |
1419 | { | |
1420 | return f->flags & (flag >> 8); | |
95ec3eb4 DM |
1421 | } |
1422 | ||
95ec3eb4 DM |
1423 | static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, |
1424 | struct packet_type *pt, struct net_device *orig_dev) | |
dc99f600 DM |
1425 | { |
1426 | struct packet_fanout *f = pt->af_packet_priv; | |
f98f4514 | 1427 | unsigned int num = READ_ONCE(f->num_members); |
19bcf9f2 | 1428 | struct net *net = read_pnet(&f->net); |
dc99f600 | 1429 | struct packet_sock *po; |
77f65ebd | 1430 | unsigned int idx; |
dc99f600 | 1431 | |
19bcf9f2 | 1432 | if (!net_eq(dev_net(dev), net) || !num) { |
dc99f600 DM |
1433 | kfree_skb(skb); |
1434 | return 0; | |
1435 | } | |
1436 | ||
3f34b24a | 1437 | if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { |
19bcf9f2 | 1438 | skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); |
3f34b24a AD |
1439 | if (!skb) |
1440 | return 0; | |
1441 | } | |
95ec3eb4 DM |
1442 | switch (f->type) { |
1443 | case PACKET_FANOUT_HASH: | |
1444 | default: | |
77f65ebd | 1445 | idx = fanout_demux_hash(f, skb, num); |
95ec3eb4 DM |
1446 | break; |
1447 | case PACKET_FANOUT_LB: | |
77f65ebd | 1448 | idx = fanout_demux_lb(f, skb, num); |
95ec3eb4 DM |
1449 | break; |
1450 | case PACKET_FANOUT_CPU: | |
77f65ebd WB |
1451 | idx = fanout_demux_cpu(f, skb, num); |
1452 | break; | |
5df0ddfb DB |
1453 | case PACKET_FANOUT_RND: |
1454 | idx = fanout_demux_rnd(f, skb, num); | |
1455 | break; | |
2d36097d NH |
1456 | case PACKET_FANOUT_QM: |
1457 | idx = fanout_demux_qm(f, skb, num); | |
1458 | break; | |
77f65ebd | 1459 | case PACKET_FANOUT_ROLLOVER: |
ad377cab | 1460 | idx = fanout_demux_rollover(f, skb, 0, false, num); |
95ec3eb4 | 1461 | break; |
47dceb8e | 1462 | case PACKET_FANOUT_CBPF: |
f2e52095 | 1463 | case PACKET_FANOUT_EBPF: |
47dceb8e WB |
1464 | idx = fanout_demux_bpf(f, skb, num); |
1465 | break; | |
dc99f600 DM |
1466 | } |
1467 | ||
ad377cab WB |
1468 | if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) |
1469 | idx = fanout_demux_rollover(f, skb, idx, true, num); | |
dc99f600 | 1470 | |
94f633ea | 1471 | po = pkt_sk(rcu_dereference(f->arr[idx])); |
dc99f600 DM |
1472 | return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); |
1473 | } | |
1474 | ||
fff3321d PE |
1475 | DEFINE_MUTEX(fanout_mutex); |
1476 | EXPORT_SYMBOL_GPL(fanout_mutex); | |
dc99f600 | 1477 | static LIST_HEAD(fanout_list); |
4a69a864 | 1478 | static u16 fanout_next_id; |
dc99f600 DM |
1479 | |
1480 | static void __fanout_link(struct sock *sk, struct packet_sock *po) | |
1481 | { | |
1482 | struct packet_fanout *f = po->fanout; | |
1483 | ||
1484 | spin_lock(&f->lock); | |
94f633ea | 1485 | rcu_assign_pointer(f->arr[f->num_members], sk); |
dc99f600 DM |
1486 | smp_wmb(); |
1487 | f->num_members++; | |
2bd624b4 AS |
1488 | if (f->num_members == 1) |
1489 | dev_add_pack(&f->prot_hook); | |
dc99f600 DM |
1490 | spin_unlock(&f->lock); |
1491 | } | |
1492 | ||
1493 | static void __fanout_unlink(struct sock *sk, struct packet_sock *po) | |
1494 | { | |
1495 | struct packet_fanout *f = po->fanout; | |
1496 | int i; | |
1497 | ||
1498 | spin_lock(&f->lock); | |
1499 | for (i = 0; i < f->num_members; i++) { | |
94f633ea ED |
1500 | if (rcu_dereference_protected(f->arr[i], |
1501 | lockdep_is_held(&f->lock)) == sk) | |
dc99f600 DM |
1502 | break; |
1503 | } | |
1504 | BUG_ON(i >= f->num_members); | |
94f633ea ED |
1505 | rcu_assign_pointer(f->arr[i], |
1506 | rcu_dereference_protected(f->arr[f->num_members - 1], | |
1507 | lockdep_is_held(&f->lock))); | |
dc99f600 | 1508 | f->num_members--; |
2bd624b4 AS |
1509 | if (f->num_members == 0) |
1510 | __dev_remove_pack(&f->prot_hook); | |
dc99f600 DM |
1511 | spin_unlock(&f->lock); |
1512 | } | |
1513 | ||
d4dd8aee | 1514 | static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) |
c0de08d0 | 1515 | { |
161642e2 ED |
1516 | if (sk->sk_family != PF_PACKET) |
1517 | return false; | |
c0de08d0 | 1518 | |
161642e2 | 1519 | return ptype->af_packet_priv == pkt_sk(sk)->fanout; |
c0de08d0 EL |
1520 | } |
1521 | ||
47dceb8e WB |
1522 | static void fanout_init_data(struct packet_fanout *f) |
1523 | { | |
1524 | switch (f->type) { | |
1525 | case PACKET_FANOUT_LB: | |
1526 | atomic_set(&f->rr_cur, 0); | |
1527 | break; | |
1528 | case PACKET_FANOUT_CBPF: | |
f2e52095 | 1529 | case PACKET_FANOUT_EBPF: |
47dceb8e WB |
1530 | RCU_INIT_POINTER(f->bpf_prog, NULL); |
1531 | break; | |
1532 | } | |
1533 | } | |
1534 | ||
1535 | static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) | |
1536 | { | |
1537 | struct bpf_prog *old; | |
1538 | ||
1539 | spin_lock(&f->lock); | |
1540 | old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); | |
1541 | rcu_assign_pointer(f->bpf_prog, new); | |
1542 | spin_unlock(&f->lock); | |
1543 | ||
1544 | if (old) { | |
1545 | synchronize_net(); | |
1546 | bpf_prog_destroy(old); | |
1547 | } | |
1548 | } | |
1549 | ||
b1ea9ff6 | 1550 | static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data, |
47dceb8e WB |
1551 | unsigned int len) |
1552 | { | |
1553 | struct bpf_prog *new; | |
1554 | struct sock_fprog fprog; | |
1555 | int ret; | |
1556 | ||
1557 | if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) | |
1558 | return -EPERM; | |
4d295e54 CH |
1559 | |
1560 | ret = copy_bpf_fprog_from_user(&fprog, data, len); | |
1561 | if (ret) | |
1562 | return ret; | |
47dceb8e | 1563 | |
bab18991 | 1564 | ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); |
47dceb8e WB |
1565 | if (ret) |
1566 | return ret; | |
1567 | ||
1568 | __fanout_set_data_bpf(po->fanout, new); | |
1569 | return 0; | |
1570 | } | |
1571 | ||
a7b75c5a | 1572 | static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data, |
f2e52095 WB |
1573 | unsigned int len) |
1574 | { | |
1575 | struct bpf_prog *new; | |
1576 | u32 fd; | |
1577 | ||
1578 | if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) | |
1579 | return -EPERM; | |
1580 | if (len != sizeof(fd)) | |
1581 | return -EINVAL; | |
a7b75c5a | 1582 | if (copy_from_sockptr(&fd, data, len)) |
f2e52095 WB |
1583 | return -EFAULT; |
1584 | ||
113214be | 1585 | new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); |
f2e52095 WB |
1586 | if (IS_ERR(new)) |
1587 | return PTR_ERR(new); | |
f2e52095 WB |
1588 | |
1589 | __fanout_set_data_bpf(po->fanout, new); | |
1590 | return 0; | |
1591 | } | |
1592 | ||
a7b75c5a | 1593 | static int fanout_set_data(struct packet_sock *po, sockptr_t data, |
47dceb8e WB |
1594 | unsigned int len) |
1595 | { | |
1596 | switch (po->fanout->type) { | |
1597 | case PACKET_FANOUT_CBPF: | |
a7b75c5a | 1598 | return fanout_set_data_cbpf(po, data, len); |
f2e52095 WB |
1599 | case PACKET_FANOUT_EBPF: |
1600 | return fanout_set_data_ebpf(po, data, len); | |
47dceb8e WB |
1601 | default: |
1602 | return -EINVAL; | |
07d53ae4 | 1603 | } |
47dceb8e WB |
1604 | } |
1605 | ||
1606 | static void fanout_release_data(struct packet_fanout *f) | |
1607 | { | |
1608 | switch (f->type) { | |
1609 | case PACKET_FANOUT_CBPF: | |
f2e52095 | 1610 | case PACKET_FANOUT_EBPF: |
47dceb8e | 1611 | __fanout_set_data_bpf(f, NULL); |
07d53ae4 | 1612 | } |
47dceb8e WB |
1613 | } |
1614 | ||
4a69a864 MM |
1615 | static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) |
1616 | { | |
1617 | struct packet_fanout *f; | |
1618 | ||
1619 | list_for_each_entry(f, &fanout_list, list) { | |
1620 | if (f->id == candidate_id && | |
1621 | read_pnet(&f->net) == sock_net(sk)) { | |
1622 | return false; | |
1623 | } | |
1624 | } | |
1625 | return true; | |
1626 | } | |
1627 | ||
1628 | static bool fanout_find_new_id(struct sock *sk, u16 *new_id) | |
1629 | { | |
1630 | u16 id = fanout_next_id; | |
1631 | ||
1632 | do { | |
1633 | if (__fanout_id_is_free(sk, id)) { | |
1634 | *new_id = id; | |
1635 | fanout_next_id = id + 1; | |
1636 | return true; | |
1637 | } | |
1638 | ||
1639 | id++; | |
1640 | } while (id != fanout_next_id); | |
1641 | ||
1642 | return false; | |
1643 | } | |
1644 | ||
9c661b0b | 1645 | static int fanout_add(struct sock *sk, struct fanout_args *args) |
dc99f600 | 1646 | { |
d199fab6 | 1647 | struct packet_rollover *rollover = NULL; |
dc99f600 | 1648 | struct packet_sock *po = pkt_sk(sk); |
9c661b0b | 1649 | u16 type_flags = args->type_flags; |
dc99f600 | 1650 | struct packet_fanout *f, *match; |
7736d33f | 1651 | u8 type = type_flags & 0xff; |
77f65ebd | 1652 | u8 flags = type_flags >> 8; |
9c661b0b | 1653 | u16 id = args->id; |
dc99f600 DM |
1654 | int err; |
1655 | ||
1656 | switch (type) { | |
77f65ebd WB |
1657 | case PACKET_FANOUT_ROLLOVER: |
1658 | if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) | |
1659 | return -EINVAL; | |
5af5a020 | 1660 | break; |
dc99f600 DM |
1661 | case PACKET_FANOUT_HASH: |
1662 | case PACKET_FANOUT_LB: | |
95ec3eb4 | 1663 | case PACKET_FANOUT_CPU: |
5df0ddfb | 1664 | case PACKET_FANOUT_RND: |
2d36097d | 1665 | case PACKET_FANOUT_QM: |
47dceb8e | 1666 | case PACKET_FANOUT_CBPF: |
f2e52095 | 1667 | case PACKET_FANOUT_EBPF: |
dc99f600 DM |
1668 | break; |
1669 | default: | |
1670 | return -EINVAL; | |
1671 | } | |
1672 | ||
d199fab6 ED |
1673 | mutex_lock(&fanout_mutex); |
1674 | ||
d199fab6 | 1675 | err = -EALREADY; |
dc99f600 | 1676 | if (po->fanout) |
d199fab6 | 1677 | goto out; |
dc99f600 | 1678 | |
4633c9e0 WB |
1679 | if (type == PACKET_FANOUT_ROLLOVER || |
1680 | (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { | |
d199fab6 ED |
1681 | err = -ENOMEM; |
1682 | rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); | |
1683 | if (!rollover) | |
1684 | goto out; | |
1685 | atomic_long_set(&rollover->num, 0); | |
1686 | atomic_long_set(&rollover->num_huge, 0); | |
1687 | atomic_long_set(&rollover->num_failed, 0); | |
0648ab70 WB |
1688 | } |
1689 | ||
4a69a864 MM |
1690 | if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { |
1691 | if (id != 0) { | |
1692 | err = -EINVAL; | |
1693 | goto out; | |
1694 | } | |
1695 | if (!fanout_find_new_id(sk, &id)) { | |
1696 | err = -ENOMEM; | |
1697 | goto out; | |
1698 | } | |
1699 | /* ephemeral flag for the first socket in the group: drop it */ | |
1700 | flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); | |
1701 | } | |
1702 | ||
dc99f600 DM |
1703 | match = NULL; |
1704 | list_for_each_entry(f, &fanout_list, list) { | |
1705 | if (f->id == id && | |
1706 | read_pnet(&f->net) == sock_net(sk)) { | |
1707 | match = f; | |
1708 | break; | |
1709 | } | |
1710 | } | |
afe62c68 | 1711 | err = -EINVAL; |
9c661b0b TL |
1712 | if (match) { |
1713 | if (match->flags != flags) | |
1714 | goto out; | |
1715 | if (args->max_num_members && | |
1716 | args->max_num_members != match->max_num_members) | |
1717 | goto out; | |
1718 | } else { | |
1719 | if (args->max_num_members > PACKET_FANOUT_MAX) | |
1720 | goto out; | |
1721 | if (!args->max_num_members) | |
1722 | /* legacy PACKET_FANOUT_MAX */ | |
1723 | args->max_num_members = 256; | |
afe62c68 | 1724 | err = -ENOMEM; |
9c661b0b TL |
1725 | match = kvzalloc(struct_size(match, arr, args->max_num_members), |
1726 | GFP_KERNEL); | |
afe62c68 ED |
1727 | if (!match) |
1728 | goto out; | |
1729 | write_pnet(&match->net, sock_net(sk)); | |
1730 | match->id = id; | |
1731 | match->type = type; | |
77f65ebd | 1732 | match->flags = flags; |
afe62c68 ED |
1733 | INIT_LIST_HEAD(&match->list); |
1734 | spin_lock_init(&match->lock); | |
fb5c2c17 | 1735 | refcount_set(&match->sk_ref, 0); |
47dceb8e | 1736 | fanout_init_data(match); |
afe62c68 ED |
1737 | match->prot_hook.type = po->prot_hook.type; |
1738 | match->prot_hook.dev = po->prot_hook.dev; | |
1739 | match->prot_hook.func = packet_rcv_fanout; | |
1740 | match->prot_hook.af_packet_priv = match; | |
c0de08d0 | 1741 | match->prot_hook.id_match = match_fanout_group; |
9c661b0b | 1742 | match->max_num_members = args->max_num_members; |
afe62c68 | 1743 | list_add(&match->list, &fanout_list); |
dc99f600 | 1744 | } |
afe62c68 | 1745 | err = -EINVAL; |
008ba2a1 WB |
1746 | |
1747 | spin_lock(&po->bind_lock); | |
1748 | if (po->running && | |
1749 | match->type == type && | |
afe62c68 ED |
1750 | match->prot_hook.type == po->prot_hook.type && |
1751 | match->prot_hook.dev == po->prot_hook.dev) { | |
1752 | err = -ENOSPC; | |
9c661b0b | 1753 | if (refcount_read(&match->sk_ref) < match->max_num_members) { |
afe62c68 ED |
1754 | __dev_remove_pack(&po->prot_hook); |
1755 | po->fanout = match; | |
57f015f5 MM |
1756 | po->rollover = rollover; |
1757 | rollover = NULL; | |
fb5c2c17 | 1758 | refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); |
afe62c68 ED |
1759 | __fanout_link(sk, po); |
1760 | err = 0; | |
dc99f600 DM |
1761 | } |
1762 | } | |
008ba2a1 WB |
1763 | spin_unlock(&po->bind_lock); |
1764 | ||
1765 | if (err && !refcount_read(&match->sk_ref)) { | |
1766 | list_del(&match->list); | |
9c661b0b | 1767 | kvfree(match); |
008ba2a1 WB |
1768 | } |
1769 | ||
afe62c68 | 1770 | out: |
57f015f5 | 1771 | kfree(rollover); |
d199fab6 | 1772 | mutex_unlock(&fanout_mutex); |
dc99f600 DM |
1773 | return err; |
1774 | } | |
1775 | ||
2bd624b4 AS |
1776 | /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes |
1777 | * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. | |
1778 | * It is the responsibility of the caller to call fanout_release_data() and | |
1779 | * free the returned packet_fanout (after synchronize_net()) | |
1780 | */ | |
1781 | static struct packet_fanout *fanout_release(struct sock *sk) | |
dc99f600 DM |
1782 | { |
1783 | struct packet_sock *po = pkt_sk(sk); | |
1784 | struct packet_fanout *f; | |
1785 | ||
fff3321d | 1786 | mutex_lock(&fanout_mutex); |
d199fab6 ED |
1787 | f = po->fanout; |
1788 | if (f) { | |
1789 | po->fanout = NULL; | |
1790 | ||
fb5c2c17 | 1791 | if (refcount_dec_and_test(&f->sk_ref)) |
d199fab6 | 1792 | list_del(&f->list); |
2bd624b4 AS |
1793 | else |
1794 | f = NULL; | |
dc99f600 DM |
1795 | } |
1796 | mutex_unlock(&fanout_mutex); | |
2bd624b4 AS |
1797 | |
1798 | return f; | |
dc99f600 | 1799 | } |
1da177e4 | 1800 | |
3c70c132 DB |
1801 | static bool packet_extra_vlan_len_allowed(const struct net_device *dev, |
1802 | struct sk_buff *skb) | |
1803 | { | |
1804 | /* Earlier code assumed this would be a VLAN pkt, double-check | |
1805 | * this now that we have the actual packet in hand. We can only | |
1806 | * do this check on Ethernet devices. | |
1807 | */ | |
1808 | if (unlikely(dev->type != ARPHRD_ETHER)) | |
1809 | return false; | |
1810 | ||
1811 | skb_reset_mac_header(skb); | |
1812 | return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); | |
1813 | } | |
1814 | ||
90ddc4f0 | 1815 | static const struct proto_ops packet_ops; |
1da177e4 | 1816 | |
90ddc4f0 | 1817 | static const struct proto_ops packet_ops_spkt; |
1da177e4 | 1818 | |
40d4e3df ED |
1819 | static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, |
1820 | struct packet_type *pt, struct net_device *orig_dev) | |
1da177e4 LT |
1821 | { |
1822 | struct sock *sk; | |
1823 | struct sockaddr_pkt *spkt; | |
1824 | ||
1825 | /* | |
1826 | * When we registered the protocol we saved the socket in the data | |
1827 | * field for just this event. | |
1828 | */ | |
1829 | ||
1830 | sk = pt->af_packet_priv; | |
1ce4f28b | 1831 | |
1da177e4 LT |
1832 | /* |
1833 | * Yank back the headers [hope the device set this | |
1834 | * right or kerboom...] | |
1835 | * | |
1836 | * Incoming packets have ll header pulled, | |
1837 | * push it back. | |
1838 | * | |
98e399f8 | 1839 | * For outgoing ones skb->data == skb_mac_header(skb) |
1da177e4 LT |
1840 | * so that this procedure is noop. |
1841 | */ | |
1842 | ||
1843 | if (skb->pkt_type == PACKET_LOOPBACK) | |
1844 | goto out; | |
1845 | ||
09ad9bc7 | 1846 | if (!net_eq(dev_net(dev), sock_net(sk))) |
d12d01d6 DL |
1847 | goto out; |
1848 | ||
40d4e3df ED |
1849 | skb = skb_share_check(skb, GFP_ATOMIC); |
1850 | if (skb == NULL) | |
1da177e4 LT |
1851 | goto oom; |
1852 | ||
1853 | /* drop any routing info */ | |
adf30907 | 1854 | skb_dst_drop(skb); |
1da177e4 | 1855 | |
84531c24 | 1856 | /* drop conntrack reference */ |
895b5c9f | 1857 | nf_reset_ct(skb); |
84531c24 | 1858 | |
ffbc6111 | 1859 | spkt = &PACKET_SKB_CB(skb)->sa.pkt; |
1da177e4 | 1860 | |
98e399f8 | 1861 | skb_push(skb, skb->data - skb_mac_header(skb)); |
1da177e4 LT |
1862 | |
1863 | /* | |
1864 | * The SOCK_PACKET socket receives _all_ frames. | |
1865 | */ | |
1866 | ||
1867 | spkt->spkt_family = dev->type; | |
1868 | strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); | |
1869 | spkt->spkt_protocol = skb->protocol; | |
1870 | ||
1871 | /* | |
1872 | * Charge the memory to the socket. This is done specifically | |
1873 | * to prevent sockets using all the memory up. | |
1874 | */ | |
1875 | ||
40d4e3df | 1876 | if (sock_queue_rcv_skb(sk, skb) == 0) |
1da177e4 LT |
1877 | return 0; |
1878 | ||
1879 | out: | |
1880 | kfree_skb(skb); | |
1881 | oom: | |
1882 | return 0; | |
1883 | } | |
1884 | ||
75c65772 MM |
1885 | static void packet_parse_headers(struct sk_buff *skb, struct socket *sock) |
1886 | { | |
18bed891 YK |
1887 | if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && |
1888 | sock->type == SOCK_RAW) { | |
75c65772 MM |
1889 | skb_reset_mac_header(skb); |
1890 | skb->protocol = dev_parse_header_protocol(skb); | |
1891 | } | |
1892 | ||
1893 | skb_probe_transport_header(skb); | |
1894 | } | |
1da177e4 LT |
1895 | |
1896 | /* | |
1897 | * Output a raw packet to a device layer. This bypasses all the other | |
1898 | * protocol layers and you must therefore supply it with a complete frame | |
1899 | */ | |
1ce4f28b | 1900 | |
1b784140 YX |
1901 | static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, |
1902 | size_t len) | |
1da177e4 LT |
1903 | { |
1904 | struct sock *sk = sock->sk; | |
342dfc30 | 1905 | DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); |
1a35ca80 | 1906 | struct sk_buff *skb = NULL; |
1da177e4 | 1907 | struct net_device *dev; |
c14ac945 | 1908 | struct sockcm_cookie sockc; |
40d4e3df | 1909 | __be16 proto = 0; |
1da177e4 | 1910 | int err; |
3bdc0eba | 1911 | int extra_len = 0; |
1ce4f28b | 1912 | |
1da177e4 | 1913 | /* |
1ce4f28b | 1914 | * Get and verify the address. |
1da177e4 LT |
1915 | */ |
1916 | ||
40d4e3df | 1917 | if (saddr) { |
1da177e4 | 1918 | if (msg->msg_namelen < sizeof(struct sockaddr)) |
40d4e3df ED |
1919 | return -EINVAL; |
1920 | if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) | |
1921 | proto = saddr->spkt_protocol; | |
1922 | } else | |
1923 | return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ | |
1da177e4 LT |
1924 | |
1925 | /* | |
1ce4f28b | 1926 | * Find the device first to size check it |
1da177e4 LT |
1927 | */ |
1928 | ||
de74e92a | 1929 | saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; |
1a35ca80 | 1930 | retry: |
654d1f8a ED |
1931 | rcu_read_lock(); |
1932 | dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); | |
1da177e4 LT |
1933 | err = -ENODEV; |
1934 | if (dev == NULL) | |
1935 | goto out_unlock; | |
1ce4f28b | 1936 | |
d5e76b0a DM |
1937 | err = -ENETDOWN; |
1938 | if (!(dev->flags & IFF_UP)) | |
1939 | goto out_unlock; | |
1940 | ||
1da177e4 | 1941 | /* |
40d4e3df ED |
1942 | * You may not queue a frame bigger than the mtu. This is the lowest level |
1943 | * raw protocol and you must do your own fragmentation at this level. | |
1da177e4 | 1944 | */ |
1ce4f28b | 1945 | |
3bdc0eba BG |
1946 | if (unlikely(sock_flag(sk, SOCK_NOFCS))) { |
1947 | if (!netif_supports_nofcs(dev)) { | |
1948 | err = -EPROTONOSUPPORT; | |
1949 | goto out_unlock; | |
1950 | } | |
1951 | extra_len = 4; /* We're doing our own CRC */ | |
1952 | } | |
1953 | ||
1da177e4 | 1954 | err = -EMSGSIZE; |
3bdc0eba | 1955 | if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) |
1da177e4 LT |
1956 | goto out_unlock; |
1957 | ||
1a35ca80 ED |
1958 | if (!skb) { |
1959 | size_t reserved = LL_RESERVED_SPACE(dev); | |
4ce40912 | 1960 | int tlen = dev->needed_tailroom; |
1a35ca80 ED |
1961 | unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; |
1962 | ||
1963 | rcu_read_unlock(); | |
4ce40912 | 1964 | skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); |
1a35ca80 ED |
1965 | if (skb == NULL) |
1966 | return -ENOBUFS; | |
1967 | /* FIXME: Save some space for broken drivers that write a hard | |
1968 | * header at transmission time by themselves. PPP is the notable | |
1969 | * one here. This should really be fixed at the driver level. | |
1970 | */ | |
1971 | skb_reserve(skb, reserved); | |
1972 | skb_reset_network_header(skb); | |
1973 | ||
1974 | /* Try to align data part correctly */ | |
1975 | if (hhlen) { | |
1976 | skb->data -= hhlen; | |
1977 | skb->tail -= hhlen; | |
1978 | if (len < hhlen) | |
1979 | skb_reset_network_header(skb); | |
1980 | } | |
6ce8e9ce | 1981 | err = memcpy_from_msg(skb_put(skb, len), msg, len); |
1a35ca80 ED |
1982 | if (err) |
1983 | goto out_free; | |
1984 | goto retry; | |
1da177e4 LT |
1985 | } |
1986 | ||
9ed988cd WB |
1987 | if (!dev_validate_header(dev, skb->data, len)) { |
1988 | err = -EINVAL; | |
1989 | goto out_unlock; | |
1990 | } | |
3c70c132 DB |
1991 | if (len > (dev->mtu + dev->hard_header_len + extra_len) && |
1992 | !packet_extra_vlan_len_allowed(dev, skb)) { | |
1993 | err = -EMSGSIZE; | |
1994 | goto out_unlock; | |
57f89bfa | 1995 | } |
1a35ca80 | 1996 | |
657a0667 | 1997 | sockcm_init(&sockc, sk); |
c14ac945 SHY |
1998 | if (msg->msg_controllen) { |
1999 | err = sock_cmsg_send(sk, msg, &sockc); | |
f8e7718c | 2000 | if (unlikely(err)) |
c14ac945 | 2001 | goto out_unlock; |
c14ac945 SHY |
2002 | } |
2003 | ||
1da177e4 LT |
2004 | skb->protocol = proto; |
2005 | skb->dev = dev; | |
2006 | skb->priority = sk->sk_priority; | |
2d37a186 | 2007 | skb->mark = sk->sk_mark; |
3d0ba8c0 | 2008 | skb->tstamp = sockc.transmit_time; |
bf84a010 | 2009 | |
8f932f76 | 2010 | skb_setup_tx_timestamp(skb, sockc.tsflags); |
1da177e4 | 2011 | |
3bdc0eba BG |
2012 | if (unlikely(extra_len == 4)) |
2013 | skb->no_fcs = 1; | |
2014 | ||
75c65772 | 2015 | packet_parse_headers(skb, sock); |
c1aad275 | 2016 | |
1da177e4 | 2017 | dev_queue_xmit(skb); |
654d1f8a | 2018 | rcu_read_unlock(); |
40d4e3df | 2019 | return len; |
1da177e4 | 2020 | |
1da177e4 | 2021 | out_unlock: |
654d1f8a | 2022 | rcu_read_unlock(); |
1a35ca80 ED |
2023 | out_free: |
2024 | kfree_skb(skb); | |
1da177e4 LT |
2025 | return err; |
2026 | } | |
1da177e4 | 2027 | |
ff936a04 AS |
2028 | static unsigned int run_filter(struct sk_buff *skb, |
2029 | const struct sock *sk, | |
2030 | unsigned int res) | |
1da177e4 LT |
2031 | { |
2032 | struct sk_filter *filter; | |
fda9ef5d | 2033 | |
80f8f102 ED |
2034 | rcu_read_lock(); |
2035 | filter = rcu_dereference(sk->sk_filter); | |
dbcb5855 | 2036 | if (filter != NULL) |
ff936a04 | 2037 | res = bpf_prog_run_clear_cb(filter->prog, skb); |
80f8f102 | 2038 | rcu_read_unlock(); |
1da177e4 | 2039 | |
dbcb5855 | 2040 | return res; |
1da177e4 LT |
2041 | } |
2042 | ||
16cc1400 WB |
2043 | static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, |
2044 | size_t *len) | |
2045 | { | |
2046 | struct virtio_net_hdr vnet_hdr; | |
2047 | ||
2048 | if (*len < sizeof(vnet_hdr)) | |
2049 | return -EINVAL; | |
2050 | *len -= sizeof(vnet_hdr); | |
2051 | ||
fd3a8862 | 2052 | if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0)) |
16cc1400 WB |
2053 | return -EINVAL; |
2054 | ||
2055 | return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); | |
2056 | } | |
2057 | ||
1da177e4 | 2058 | /* |
62ab0812 ED |
2059 | * This function makes lazy skb cloning in hope that most of packets |
2060 | * are discarded by BPF. | |
2061 | * | |
2062 | * Note tricky part: we DO mangle shared skb! skb->data, skb->len | |
2063 | * and skb->cb are mangled. It works because (and until) packets | |
2064 | * falling here are owned by current CPU. Output packets are cloned | |
2065 | * by dev_queue_xmit_nit(), input packets are processed by net_bh | |
0e4161d0 | 2066 | * sequentially, so that if we return skb to original state on exit, |
62ab0812 | 2067 | * we will not harm anyone. |
1da177e4 LT |
2068 | */ |
2069 | ||
40d4e3df ED |
2070 | static int packet_rcv(struct sk_buff *skb, struct net_device *dev, |
2071 | struct packet_type *pt, struct net_device *orig_dev) | |
1da177e4 LT |
2072 | { |
2073 | struct sock *sk; | |
2074 | struct sockaddr_ll *sll; | |
2075 | struct packet_sock *po; | |
40d4e3df | 2076 | u8 *skb_head = skb->data; |
1da177e4 | 2077 | int skb_len = skb->len; |
dbcb5855 | 2078 | unsigned int snaplen, res; |
da37845f | 2079 | bool is_drop_n_account = false; |
1da177e4 LT |
2080 | |
2081 | if (skb->pkt_type == PACKET_LOOPBACK) | |
2082 | goto drop; | |
2083 | ||
2084 | sk = pt->af_packet_priv; | |
2085 | po = pkt_sk(sk); | |
2086 | ||
09ad9bc7 | 2087 | if (!net_eq(dev_net(dev), sock_net(sk))) |
d12d01d6 DL |
2088 | goto drop; |
2089 | ||
1da177e4 LT |
2090 | skb->dev = dev; |
2091 | ||
d5496990 | 2092 | if (dev_has_header(dev)) { |
1da177e4 | 2093 | /* The device has an explicit notion of ll header, |
62ab0812 ED |
2094 | * exported to higher levels. |
2095 | * | |
2096 | * Otherwise, the device hides details of its frame | |
2097 | * structure, so that corresponding packet head is | |
2098 | * never delivered to user. | |
1da177e4 LT |
2099 | */ |
2100 | if (sk->sk_type != SOCK_DGRAM) | |
98e399f8 | 2101 | skb_push(skb, skb->data - skb_mac_header(skb)); |
1da177e4 LT |
2102 | else if (skb->pkt_type == PACKET_OUTGOING) { |
2103 | /* Special case: outgoing packets have ll header at head */ | |
bbe735e4 | 2104 | skb_pull(skb, skb_network_offset(skb)); |
1da177e4 LT |
2105 | } |
2106 | } | |
2107 | ||
2108 | snaplen = skb->len; | |
2109 | ||
dbcb5855 DM |
2110 | res = run_filter(skb, sk, snaplen); |
2111 | if (!res) | |
fda9ef5d | 2112 | goto drop_n_restore; |
dbcb5855 DM |
2113 | if (snaplen > res) |
2114 | snaplen = res; | |
1da177e4 | 2115 | |
0fd7bac6 | 2116 | if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) |
1da177e4 LT |
2117 | goto drop_n_acct; |
2118 | ||
2119 | if (skb_shared(skb)) { | |
2120 | struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); | |
2121 | if (nskb == NULL) | |
2122 | goto drop_n_acct; | |
2123 | ||
2124 | if (skb_head != skb->data) { | |
2125 | skb->data = skb_head; | |
2126 | skb->len = skb_len; | |
2127 | } | |
abc4e4fa | 2128 | consume_skb(skb); |
1da177e4 LT |
2129 | skb = nskb; |
2130 | } | |
2131 | ||
b4772ef8 | 2132 | sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); |
ffbc6111 HX |
2133 | |
2134 | sll = &PACKET_SKB_CB(skb)->sa.ll; | |
1da177e4 | 2135 | sll->sll_hatype = dev->type; |
1da177e4 | 2136 | sll->sll_pkttype = skb->pkt_type; |
8032b464 | 2137 | if (unlikely(po->origdev)) |
80feaacb PWJ |
2138 | sll->sll_ifindex = orig_dev->ifindex; |
2139 | else | |
2140 | sll->sll_ifindex = dev->ifindex; | |
1da177e4 | 2141 | |
b95cce35 | 2142 | sll->sll_halen = dev_parse_header(skb, sll->sll_addr); |
1da177e4 | 2143 | |
2472d761 EB |
2144 | /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). |
2145 | * Use their space for storing the original skb length. | |
2146 | */ | |
2147 | PACKET_SKB_CB(skb)->sa.origlen = skb->len; | |
8dc41944 | 2148 | |
1da177e4 LT |
2149 | if (pskb_trim(skb, snaplen)) |
2150 | goto drop_n_acct; | |
2151 | ||
2152 | skb_set_owner_r(skb, sk); | |
2153 | skb->dev = NULL; | |
adf30907 | 2154 | skb_dst_drop(skb); |
1da177e4 | 2155 | |
84531c24 | 2156 | /* drop conntrack reference */ |
895b5c9f | 2157 | nf_reset_ct(skb); |
84531c24 | 2158 | |
1da177e4 | 2159 | spin_lock(&sk->sk_receive_queue.lock); |
ee80fbf3 | 2160 | po->stats.stats1.tp_packets++; |
3bc3b96f | 2161 | sock_skb_set_dropcount(sk, skb); |
1da177e4 LT |
2162 | __skb_queue_tail(&sk->sk_receive_queue, skb); |
2163 | spin_unlock(&sk->sk_receive_queue.lock); | |
676d2369 | 2164 | sk->sk_data_ready(sk); |
1da177e4 LT |
2165 | return 0; |
2166 | ||
2167 | drop_n_acct: | |
da37845f | 2168 | is_drop_n_account = true; |
8e8e2951 | 2169 | atomic_inc(&po->tp_drops); |
7091fbd8 | 2170 | atomic_inc(&sk->sk_drops); |
1da177e4 LT |
2171 | |
2172 | drop_n_restore: | |
2173 | if (skb_head != skb->data && skb_shared(skb)) { | |
2174 | skb->data = skb_head; | |
2175 | skb->len = skb_len; | |
2176 | } | |
2177 | drop: | |
da37845f WJ |
2178 | if (!is_drop_n_account) |
2179 | consume_skb(skb); | |
2180 | else | |
2181 | kfree_skb(skb); | |
1da177e4 LT |
2182 | return 0; |
2183 | } | |
2184 | ||
40d4e3df ED |
2185 | static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, |
2186 | struct packet_type *pt, struct net_device *orig_dev) | |
1da177e4 LT |
2187 | { |
2188 | struct sock *sk; | |
2189 | struct packet_sock *po; | |
2190 | struct sockaddr_ll *sll; | |
184f489e | 2191 | union tpacket_uhdr h; |
40d4e3df | 2192 | u8 *skb_head = skb->data; |
1da177e4 | 2193 | int skb_len = skb->len; |
dbcb5855 | 2194 | unsigned int snaplen, res; |
f6fb8f10 | 2195 | unsigned long status = TP_STATUS_USER; |
acf69c94 OC |
2196 | unsigned short macoff, hdrlen; |
2197 | unsigned int netoff; | |
1da177e4 | 2198 | struct sk_buff *copy_skb = NULL; |
d413fcb4 | 2199 | struct timespec64 ts; |
b9c32fb2 | 2200 | __u32 ts_status; |
da37845f | 2201 | bool is_drop_n_account = false; |
61fad681 | 2202 | unsigned int slot_id = 0; |
edbd58be | 2203 | bool do_vnet = false; |
1da177e4 | 2204 | |
51846355 AW |
2205 | /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. |
2206 | * We may add members to them until current aligned size without forcing | |
2207 | * userspace to call getsockopt(..., PACKET_HDRLEN, ...). | |
2208 | */ | |
2209 | BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); | |
2210 | BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); | |
2211 | ||
1da177e4 LT |
2212 | if (skb->pkt_type == PACKET_LOOPBACK) |
2213 | goto drop; | |
2214 | ||
2215 | sk = pt->af_packet_priv; | |
2216 | po = pkt_sk(sk); | |
2217 | ||
09ad9bc7 | 2218 | if (!net_eq(dev_net(dev), sock_net(sk))) |
d12d01d6 DL |
2219 | goto drop; |
2220 | ||
d5496990 | 2221 | if (dev_has_header(dev)) { |
1da177e4 | 2222 | if (sk->sk_type != SOCK_DGRAM) |
98e399f8 | 2223 | skb_push(skb, skb->data - skb_mac_header(skb)); |
1da177e4 LT |
2224 | else if (skb->pkt_type == PACKET_OUTGOING) { |
2225 | /* Special case: outgoing packets have ll header at head */ | |
bbe735e4 | 2226 | skb_pull(skb, skb_network_offset(skb)); |
1da177e4 LT |
2227 | } |
2228 | } | |
2229 | ||
2230 | snaplen = skb->len; | |
2231 | ||
dbcb5855 DM |
2232 | res = run_filter(skb, sk, snaplen); |
2233 | if (!res) | |
fda9ef5d | 2234 | goto drop_n_restore; |
68c2e5de | 2235 | |
2c51c627 ED |
2236 | /* If we are flooded, just give up */ |
2237 | if (__packet_rcv_has_room(po, skb) == ROOM_NONE) { | |
2238 | atomic_inc(&po->tp_drops); | |
2239 | goto drop_n_restore; | |
2240 | } | |
2241 | ||
68c2e5de AD |
2242 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
2243 | status |= TP_STATUS_CSUMNOTREADY; | |
682f048b AD |
2244 | else if (skb->pkt_type != PACKET_OUTGOING && |
2245 | (skb->ip_summed == CHECKSUM_COMPLETE || | |
2246 | skb_csum_unnecessary(skb))) | |
2247 | status |= TP_STATUS_CSUM_VALID; | |
68c2e5de | 2248 | |
dbcb5855 DM |
2249 | if (snaplen > res) |
2250 | snaplen = res; | |
1da177e4 LT |
2251 | |
2252 | if (sk->sk_type == SOCK_DGRAM) { | |
8913336a PM |
2253 | macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + |
2254 | po->tp_reserve; | |
1da177e4 | 2255 | } else { |
95c96174 | 2256 | unsigned int maclen = skb_network_offset(skb); |
bbd6ef87 | 2257 | netoff = TPACKET_ALIGN(po->tp_hdrlen + |
8913336a | 2258 | (maclen < 16 ? 16 : maclen)) + |
58d19b19 | 2259 | po->tp_reserve; |
edbd58be | 2260 | if (po->has_vnet_hdr) { |
58d19b19 | 2261 | netoff += sizeof(struct virtio_net_hdr); |
edbd58be BP |
2262 | do_vnet = true; |
2263 | } | |
1da177e4 LT |
2264 | macoff = netoff - maclen; |
2265 | } | |
acf69c94 OC |
2266 | if (netoff > USHRT_MAX) { |
2267 | atomic_inc(&po->tp_drops); | |
2268 | goto drop_n_restore; | |
2269 | } | |
f6fb8f10 | 2270 | if (po->tp_version <= TPACKET_V2) { |
2271 | if (macoff + snaplen > po->rx_ring.frame_size) { | |
2272 | if (po->copy_thresh && | |
0fd7bac6 | 2273 | atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { |
f6fb8f10 | 2274 | if (skb_shared(skb)) { |
2275 | copy_skb = skb_clone(skb, GFP_ATOMIC); | |
2276 | } else { | |
2277 | copy_skb = skb_get(skb); | |
2278 | skb_head = skb->data; | |
2279 | } | |
2280 | if (copy_skb) | |
2281 | skb_set_owner_r(copy_skb, sk); | |
1da177e4 | 2282 | } |
f6fb8f10 | 2283 | snaplen = po->rx_ring.frame_size - macoff; |
edbd58be | 2284 | if ((int)snaplen < 0) { |
f6fb8f10 | 2285 | snaplen = 0; |
edbd58be BP |
2286 | do_vnet = false; |
2287 | } | |
1da177e4 | 2288 | } |
dc808110 ED |
2289 | } else if (unlikely(macoff + snaplen > |
2290 | GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { | |
2291 | u32 nval; | |
2292 | ||
2293 | nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; | |
2294 | pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", | |
2295 | snaplen, nval, macoff); | |
2296 | snaplen = nval; | |
2297 | if (unlikely((int)snaplen < 0)) { | |
2298 | snaplen = 0; | |
2299 | macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; | |
edbd58be | 2300 | do_vnet = false; |
dc808110 | 2301 | } |
1da177e4 | 2302 | } |
1da177e4 | 2303 | spin_lock(&sk->sk_receive_queue.lock); |
f6fb8f10 | 2304 | h.raw = packet_current_rx_frame(po, skb, |
2305 | TP_STATUS_KERNEL, (macoff+snaplen)); | |
bbd6ef87 | 2306 | if (!h.raw) |
58d19b19 | 2307 | goto drop_n_account; |
46e4c421 | 2308 | |
61fad681 WB |
2309 | if (po->tp_version <= TPACKET_V2) { |
2310 | slot_id = po->rx_ring.head; | |
2311 | if (test_bit(slot_id, po->rx_ring.rx_owner_map)) | |
2312 | goto drop_n_account; | |
2313 | __set_bit(slot_id, po->rx_ring.rx_owner_map); | |
2314 | } | |
2315 | ||
46e4c421 WB |
2316 | if (do_vnet && |
2317 | virtio_net_hdr_from_skb(skb, h.raw + macoff - | |
2318 | sizeof(struct virtio_net_hdr), | |
88fd1cb8 JO |
2319 | vio_le(), true, 0)) { |
2320 | if (po->tp_version == TPACKET_V3) | |
2321 | prb_clear_blk_fill_status(&po->rx_ring); | |
46e4c421 | 2322 | goto drop_n_account; |
88fd1cb8 | 2323 | } |
46e4c421 | 2324 | |
f6fb8f10 | 2325 | if (po->tp_version <= TPACKET_V2) { |
2326 | packet_increment_rx_head(po, &po->rx_ring); | |
2327 | /* | |
2328 | * LOSING will be reported till you read the stats, | |
2329 | * because it's COR - Clear On Read. | |
2330 | * Anyways, moving it for V1/V2 only as V3 doesn't need this | |
2331 | * at packet level. | |
2332 | */ | |
8e8e2951 | 2333 | if (atomic_read(&po->tp_drops)) |
f6fb8f10 | 2334 | status |= TP_STATUS_LOSING; |
2335 | } | |
945d015e | 2336 | |
ee80fbf3 | 2337 | po->stats.stats1.tp_packets++; |
1da177e4 LT |
2338 | if (copy_skb) { |
2339 | status |= TP_STATUS_COPY; | |
2340 | __skb_queue_tail(&sk->sk_receive_queue, copy_skb); | |
2341 | } | |
1da177e4 LT |
2342 | spin_unlock(&sk->sk_receive_queue.lock); |
2343 | ||
bbd6ef87 | 2344 | skb_copy_bits(skb, 0, h.raw + macoff, snaplen); |
b9c32fb2 | 2345 | |
171c3b15 RS |
2346 | /* Always timestamp; prefer an existing software timestamp taken |
2347 | * closer to the time of capture. | |
2348 | */ | |
2349 | ts_status = tpacket_get_timestamp(skb, &ts, | |
2350 | po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE); | |
2351 | if (!ts_status) | |
d413fcb4 | 2352 | ktime_get_real_ts64(&ts); |
1da177e4 | 2353 | |
b9c32fb2 DB |
2354 | status |= ts_status; |
2355 | ||
bbd6ef87 PM |
2356 | switch (po->tp_version) { |
2357 | case TPACKET_V1: | |
2358 | h.h1->tp_len = skb->len; | |
2359 | h.h1->tp_snaplen = snaplen; | |
2360 | h.h1->tp_mac = macoff; | |
2361 | h.h1->tp_net = netoff; | |
4b457bdf DB |
2362 | h.h1->tp_sec = ts.tv_sec; |
2363 | h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; | |
bbd6ef87 PM |
2364 | hdrlen = sizeof(*h.h1); |
2365 | break; | |
2366 | case TPACKET_V2: | |
2367 | h.h2->tp_len = skb->len; | |
2368 | h.h2->tp_snaplen = snaplen; | |
2369 | h.h2->tp_mac = macoff; | |
2370 | h.h2->tp_net = netoff; | |
bbd6ef87 PM |
2371 | h.h2->tp_sec = ts.tv_sec; |
2372 | h.h2->tp_nsec = ts.tv_nsec; | |
df8a39de JP |
2373 | if (skb_vlan_tag_present(skb)) { |
2374 | h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); | |
a0cdfcf3 AW |
2375 | h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); |
2376 | status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; | |
a3bcc23e BG |
2377 | } else { |
2378 | h.h2->tp_vlan_tci = 0; | |
a0cdfcf3 | 2379 | h.h2->tp_vlan_tpid = 0; |
a3bcc23e | 2380 | } |
e4d26f4b | 2381 | memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); |
bbd6ef87 PM |
2382 | hdrlen = sizeof(*h.h2); |
2383 | break; | |
f6fb8f10 | 2384 | case TPACKET_V3: |
2385 | /* tp_nxt_offset,vlan are already populated above. | |
2386 | * So DONT clear those fields here | |
2387 | */ | |
2388 | h.h3->tp_status |= status; | |
2389 | h.h3->tp_len = skb->len; | |
2390 | h.h3->tp_snaplen = snaplen; | |
2391 | h.h3->tp_mac = macoff; | |
2392 | h.h3->tp_net = netoff; | |
f6fb8f10 | 2393 | h.h3->tp_sec = ts.tv_sec; |
2394 | h.h3->tp_nsec = ts.tv_nsec; | |
e4d26f4b | 2395 | memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); |
f6fb8f10 | 2396 | hdrlen = sizeof(*h.h3); |
2397 | break; | |
bbd6ef87 PM |
2398 | default: |
2399 | BUG(); | |
2400 | } | |
1da177e4 | 2401 | |
bbd6ef87 | 2402 | sll = h.raw + TPACKET_ALIGN(hdrlen); |
b95cce35 | 2403 | sll->sll_halen = dev_parse_header(skb, sll->sll_addr); |
1da177e4 LT |
2404 | sll->sll_family = AF_PACKET; |
2405 | sll->sll_hatype = dev->type; | |
2406 | sll->sll_protocol = skb->protocol; | |
2407 | sll->sll_pkttype = skb->pkt_type; | |
8032b464 | 2408 | if (unlikely(po->origdev)) |
80feaacb PWJ |
2409 | sll->sll_ifindex = orig_dev->ifindex; |
2410 | else | |
2411 | sll->sll_ifindex = dev->ifindex; | |
1da177e4 | 2412 | |
e16aa207 | 2413 | smp_mb(); |
f0d4eb29 | 2414 | |
f6dafa95 | 2415 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 |
f0d4eb29 | 2416 | if (po->tp_version <= TPACKET_V2) { |
0af55bb5 CG |
2417 | u8 *start, *end; |
2418 | ||
f0d4eb29 DB |
2419 | end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + |
2420 | macoff + snaplen); | |
2421 | ||
2422 | for (start = h.raw; start < end; start += PAGE_SIZE) | |
2423 | flush_dcache_page(pgv_to_page(start)); | |
1da177e4 | 2424 | } |
f0d4eb29 | 2425 | smp_wmb(); |
f6dafa95 | 2426 | #endif |
f0d4eb29 | 2427 | |
da413eec | 2428 | if (po->tp_version <= TPACKET_V2) { |
61fad681 | 2429 | spin_lock(&sk->sk_receive_queue.lock); |
f6fb8f10 | 2430 | __packet_set_status(po, h.raw, status); |
61fad681 WB |
2431 | __clear_bit(slot_id, po->rx_ring.rx_owner_map); |
2432 | spin_unlock(&sk->sk_receive_queue.lock); | |
da413eec | 2433 | sk->sk_data_ready(sk); |
88fd1cb8 | 2434 | } else if (po->tp_version == TPACKET_V3) { |
f6fb8f10 | 2435 | prb_clear_blk_fill_status(&po->rx_ring); |
da413eec | 2436 | } |
1da177e4 LT |
2437 | |
2438 | drop_n_restore: | |
2439 | if (skb_head != skb->data && skb_shared(skb)) { | |
2440 | skb->data = skb_head; | |
2441 | skb->len = skb_len; | |
2442 | } | |
2443 | drop: | |
da37845f WJ |
2444 | if (!is_drop_n_account) |
2445 | consume_skb(skb); | |
2446 | else | |
2447 | kfree_skb(skb); | |
1da177e4 LT |
2448 | return 0; |
2449 | ||
58d19b19 | 2450 | drop_n_account: |
1da177e4 | 2451 | spin_unlock(&sk->sk_receive_queue.lock); |
8e8e2951 ED |
2452 | atomic_inc(&po->tp_drops); |
2453 | is_drop_n_account = true; | |
1da177e4 | 2454 | |
676d2369 | 2455 | sk->sk_data_ready(sk); |
acb5d75b | 2456 | kfree_skb(copy_skb); |
1da177e4 LT |
2457 | goto drop_n_restore; |
2458 | } | |
2459 | ||
69e3c75f JB |
2460 | static void tpacket_destruct_skb(struct sk_buff *skb) |
2461 | { | |
2462 | struct packet_sock *po = pkt_sk(skb->sk); | |
1da177e4 | 2463 | |
69e3c75f | 2464 | if (likely(po->tx_ring.pg_vec)) { |
f0d4eb29 | 2465 | void *ph; |
b9c32fb2 DB |
2466 | __u32 ts; |
2467 | ||
5cd8d46e | 2468 | ph = skb_zcopy_get_nouarg(skb); |
b0138408 | 2469 | packet_dec_pending(&po->tx_ring); |
b9c32fb2 DB |
2470 | |
2471 | ts = __packet_set_timestamp(po, ph, skb); | |
2472 | __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); | |
89ed5b51 NH |
2473 | |
2474 | if (!packet_read_pending(&po->tx_ring)) | |
2475 | complete(&po->skb_completion); | |
69e3c75f JB |
2476 | } |
2477 | ||
2478 | sock_wfree(skb); | |
2479 | } | |
2480 | ||
16cc1400 WB |
2481 | static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) |
2482 | { | |
16cc1400 WB |
2483 | if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && |
2484 | (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + | |
2485 | __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > | |
2486 | __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) | |
2487 | vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), | |
2488 | __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + | |
2489 | __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); | |
2490 | ||
2491 | if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) | |
2492 | return -EINVAL; | |
2493 | ||
16cc1400 WB |
2494 | return 0; |
2495 | } | |
2496 | ||
2497 | static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, | |
2498 | struct virtio_net_hdr *vnet_hdr) | |
2499 | { | |
16cc1400 WB |
2500 | if (*len < sizeof(*vnet_hdr)) |
2501 | return -EINVAL; | |
2502 | *len -= sizeof(*vnet_hdr); | |
2503 | ||
cbbd26b8 | 2504 | if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) |
16cc1400 WB |
2505 | return -EFAULT; |
2506 | ||
2507 | return __packet_snd_vnet_parse(vnet_hdr, *len); | |
2508 | } | |
2509 | ||
40d4e3df | 2510 | static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, |
8d39b4a6 | 2511 | void *frame, struct net_device *dev, void *data, int tp_len, |
c14ac945 SHY |
2512 | __be16 proto, unsigned char *addr, int hlen, int copylen, |
2513 | const struct sockcm_cookie *sockc) | |
69e3c75f | 2514 | { |
184f489e | 2515 | union tpacket_uhdr ph; |
8d39b4a6 | 2516 | int to_write, offset, len, nr_frags, len_max; |
69e3c75f JB |
2517 | struct socket *sock = po->sk.sk_socket; |
2518 | struct page *page; | |
69e3c75f JB |
2519 | int err; |
2520 | ||
2521 | ph.raw = frame; | |
2522 | ||
2523 | skb->protocol = proto; | |
2524 | skb->dev = dev; | |
2525 | skb->priority = po->sk.sk_priority; | |
2d37a186 | 2526 | skb->mark = po->sk.sk_mark; |
3d0ba8c0 | 2527 | skb->tstamp = sockc->transmit_time; |
8f932f76 | 2528 | skb_setup_tx_timestamp(skb, sockc->tsflags); |
5cd8d46e | 2529 | skb_zcopy_set_nouarg(skb, ph.raw); |
69e3c75f | 2530 | |
ae641949 | 2531 | skb_reserve(skb, hlen); |
69e3c75f | 2532 | skb_reset_network_header(skb); |
c1aad275 | 2533 | |
69e3c75f JB |
2534 | to_write = tp_len; |
2535 | ||
2536 | if (sock->type == SOCK_DGRAM) { | |
2537 | err = dev_hard_header(skb, dev, ntohs(proto), addr, | |
2538 | NULL, tp_len); | |
2539 | if (unlikely(err < 0)) | |
2540 | return -EINVAL; | |
1d036d25 | 2541 | } else if (copylen) { |
9ed988cd WB |
2542 | int hdrlen = min_t(int, copylen, tp_len); |
2543 | ||
69e3c75f | 2544 | skb_push(skb, dev->hard_header_len); |
1d036d25 | 2545 | skb_put(skb, copylen - dev->hard_header_len); |
9ed988cd | 2546 | err = skb_store_bits(skb, 0, data, hdrlen); |
69e3c75f JB |
2547 | if (unlikely(err)) |
2548 | return err; | |
9ed988cd WB |
2549 | if (!dev_validate_header(dev, skb->data, hdrlen)) |
2550 | return -EINVAL; | |
69e3c75f | 2551 | |
9ed988cd WB |
2552 | data += hdrlen; |
2553 | to_write -= hdrlen; | |
69e3c75f JB |
2554 | } |
2555 | ||
69e3c75f JB |
2556 | offset = offset_in_page(data); |
2557 | len_max = PAGE_SIZE - offset; | |
2558 | len = ((to_write > len_max) ? len_max : to_write); | |
2559 | ||
2560 | skb->data_len = to_write; | |
2561 | skb->len += to_write; | |
2562 | skb->truesize += to_write; | |
14afee4b | 2563 | refcount_add(to_write, &po->sk.sk_wmem_alloc); |
69e3c75f JB |
2564 | |
2565 | while (likely(to_write)) { | |
2566 | nr_frags = skb_shinfo(skb)->nr_frags; | |
2567 | ||
2568 | if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { | |
40d4e3df ED |
2569 | pr_err("Packet exceed the number of skb frags(%lu)\n", |
2570 | MAX_SKB_FRAGS); | |
69e3c75f JB |
2571 | return -EFAULT; |
2572 | } | |
2573 | ||
0af55bb5 CG |
2574 | page = pgv_to_page(data); |
2575 | data += len; | |
69e3c75f JB |
2576 | flush_dcache_page(page); |
2577 | get_page(page); | |
0af55bb5 | 2578 | skb_fill_page_desc(skb, nr_frags, page, offset, len); |
69e3c75f JB |
2579 | to_write -= len; |
2580 | offset = 0; | |
2581 | len_max = PAGE_SIZE; | |
2582 | len = ((to_write > len_max) ? len_max : to_write); | |
2583 | } | |
2584 | ||
75c65772 | 2585 | packet_parse_headers(skb, sock); |
efdfa2f7 | 2586 | |
69e3c75f JB |
2587 | return tp_len; |
2588 | } | |
2589 | ||
8d39b4a6 WB |
2590 | static int tpacket_parse_header(struct packet_sock *po, void *frame, |
2591 | int size_max, void **data) | |
2592 | { | |
2593 | union tpacket_uhdr ph; | |
2594 | int tp_len, off; | |
2595 | ||
2596 | ph.raw = frame; | |
2597 | ||
2598 | switch (po->tp_version) { | |
7f953ab2 SV |
2599 | case TPACKET_V3: |
2600 | if (ph.h3->tp_next_offset != 0) { | |
2601 | pr_warn_once("variable sized slot not supported"); | |
2602 | return -EINVAL; | |
2603 | } | |
2604 | tp_len = ph.h3->tp_len; | |
2605 | break; | |
8d39b4a6 WB |
2606 | case TPACKET_V2: |
2607 | tp_len = ph.h2->tp_len; | |
2608 | break; | |
2609 | default: | |
2610 | tp_len = ph.h1->tp_len; | |
2611 | break; | |
2612 | } | |
2613 | if (unlikely(tp_len > size_max)) { | |
2614 | pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); | |
2615 | return -EMSGSIZE; | |
2616 | } | |
2617 | ||
2618 | if (unlikely(po->tp_tx_has_off)) { | |
2619 | int off_min, off_max; | |
2620 | ||
2621 | off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); | |
2622 | off_max = po->tx_ring.frame_size - tp_len; | |
2623 | if (po->sk.sk_type == SOCK_DGRAM) { | |
2624 | switch (po->tp_version) { | |
7f953ab2 SV |
2625 | case TPACKET_V3: |
2626 | off = ph.h3->tp_net; | |
2627 | break; | |
8d39b4a6 WB |
2628 | case TPACKET_V2: |
2629 | off = ph.h2->tp_net; | |
2630 | break; | |
2631 | default: | |
2632 | off = ph.h1->tp_net; | |
2633 | break; | |
2634 | } | |
2635 | } else { | |
2636 | switch (po->tp_version) { | |
7f953ab2 SV |
2637 | case TPACKET_V3: |
2638 | off = ph.h3->tp_mac; | |
2639 | break; | |
8d39b4a6 WB |
2640 | case TPACKET_V2: |
2641 | off = ph.h2->tp_mac; | |
2642 | break; | |
2643 | default: | |
2644 | off = ph.h1->tp_mac; | |
2645 | break; | |
2646 | } | |
2647 | } | |
2648 | if (unlikely((off < off_min) || (off_max < off))) | |
2649 | return -EINVAL; | |
2650 | } else { | |
2651 | off = po->tp_hdrlen - sizeof(struct sockaddr_ll); | |
2652 | } | |
2653 | ||
2654 | *data = frame + off; | |
2655 | return tp_len; | |
2656 | } | |
2657 | ||
69e3c75f JB |
2658 | static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) |
2659 | { | |
89ed5b51 | 2660 | struct sk_buff *skb = NULL; |
69e3c75f | 2661 | struct net_device *dev; |
1d036d25 | 2662 | struct virtio_net_hdr *vnet_hdr = NULL; |
c14ac945 | 2663 | struct sockcm_cookie sockc; |
69e3c75f | 2664 | __be16 proto; |
09effa67 | 2665 | int err, reserve = 0; |
40d4e3df | 2666 | void *ph; |
342dfc30 | 2667 | DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); |
87a2fd28 | 2668 | bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); |
486efdc8 | 2669 | unsigned char *addr = NULL; |
69e3c75f | 2670 | int tp_len, size_max; |
8d39b4a6 | 2671 | void *data; |
69e3c75f | 2672 | int len_sum = 0; |
9e67030a | 2673 | int status = TP_STATUS_AVAILABLE; |
1d036d25 | 2674 | int hlen, tlen, copylen = 0; |
89ed5b51 | 2675 | long timeo = 0; |
69e3c75f | 2676 | |
69e3c75f JB |
2677 | mutex_lock(&po->pg_vec_lock); |
2678 | ||
32d3182c ED |
2679 | /* packet_sendmsg() check on tx_ring.pg_vec was lockless, |
2680 | * we need to confirm it under protection of pg_vec_lock. | |
2681 | */ | |
2682 | if (unlikely(!po->tx_ring.pg_vec)) { | |
2683 | err = -EBUSY; | |
2684 | goto out; | |
2685 | } | |
66e56cd4 | 2686 | if (likely(saddr == NULL)) { |
e40526cb | 2687 | dev = packet_cached_dev_get(po); |
c7d2ef5d | 2688 | proto = READ_ONCE(po->num); |
69e3c75f JB |
2689 | } else { |
2690 | err = -EINVAL; | |
2691 | if (msg->msg_namelen < sizeof(struct sockaddr_ll)) | |
2692 | goto out; | |
2693 | if (msg->msg_namelen < (saddr->sll_halen | |
2694 | + offsetof(struct sockaddr_ll, | |
2695 | sll_addr))) | |
2696 | goto out; | |
69e3c75f | 2697 | proto = saddr->sll_protocol; |
827d9780 | 2698 | dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); |
486efdc8 WB |
2699 | if (po->sk.sk_socket->type == SOCK_DGRAM) { |
2700 | if (dev && msg->msg_namelen < dev->addr_len + | |
2701 | offsetof(struct sockaddr_ll, sll_addr)) | |
2702 | goto out_put; | |
2703 | addr = saddr->sll_addr; | |
2704 | } | |
69e3c75f JB |
2705 | } |
2706 | ||
69e3c75f JB |
2707 | err = -ENXIO; |
2708 | if (unlikely(dev == NULL)) | |
2709 | goto out; | |
69e3c75f JB |
2710 | err = -ENETDOWN; |
2711 | if (unlikely(!(dev->flags & IFF_UP))) | |
2712 | goto out_put; | |
2713 | ||
657a0667 | 2714 | sockcm_init(&sockc, &po->sk); |
d19b183c DCS |
2715 | if (msg->msg_controllen) { |
2716 | err = sock_cmsg_send(&po->sk, msg, &sockc); | |
2717 | if (unlikely(err)) | |
2718 | goto out_put; | |
2719 | } | |
2720 | ||
5cfb4c8d DB |
2721 | if (po->sk.sk_socket->type == SOCK_RAW) |
2722 | reserve = dev->hard_header_len; | |
69e3c75f | 2723 | size_max = po->tx_ring.frame_size |
b5dd884e | 2724 | - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); |
69e3c75f | 2725 | |
1d036d25 | 2726 | if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) |
5cfb4c8d | 2727 | size_max = dev->mtu + reserve + VLAN_HLEN; |
09effa67 | 2728 | |
89ed5b51 NH |
2729 | reinit_completion(&po->skb_completion); |
2730 | ||
69e3c75f JB |
2731 | do { |
2732 | ph = packet_current_frame(po, &po->tx_ring, | |
87a2fd28 | 2733 | TP_STATUS_SEND_REQUEST); |
69e3c75f | 2734 | if (unlikely(ph == NULL)) { |
89ed5b51 NH |
2735 | if (need_wait && skb) { |
2736 | timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT); | |
2737 | timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo); | |
2738 | if (timeo <= 0) { | |
2739 | err = !timeo ? -ETIMEDOUT : -ERESTARTSYS; | |
2740 | goto out_put; | |
2741 | } | |
2742 | } | |
2743 | /* check for additional frames */ | |
69e3c75f JB |
2744 | continue; |
2745 | } | |
2746 | ||
8d39b4a6 WB |
2747 | skb = NULL; |
2748 | tp_len = tpacket_parse_header(po, ph, size_max, &data); | |
2749 | if (tp_len < 0) | |
2750 | goto tpacket_error; | |
2751 | ||
69e3c75f | 2752 | status = TP_STATUS_SEND_REQUEST; |
ae641949 HX |
2753 | hlen = LL_RESERVED_SPACE(dev); |
2754 | tlen = dev->needed_tailroom; | |
1d036d25 WB |
2755 | if (po->has_vnet_hdr) { |
2756 | vnet_hdr = data; | |
2757 | data += sizeof(*vnet_hdr); | |
2758 | tp_len -= sizeof(*vnet_hdr); | |
2759 | if (tp_len < 0 || | |
2760 | __packet_snd_vnet_parse(vnet_hdr, tp_len)) { | |
2761 | tp_len = -EINVAL; | |
2762 | goto tpacket_error; | |
2763 | } | |
2764 | copylen = __virtio16_to_cpu(vio_le(), | |
2765 | vnet_hdr->hdr_len); | |
2766 | } | |
9ed988cd | 2767 | copylen = max_t(int, copylen, dev->hard_header_len); |
69e3c75f | 2768 | skb = sock_alloc_send_skb(&po->sk, |
1d036d25 WB |
2769 | hlen + tlen + sizeof(struct sockaddr_ll) + |
2770 | (copylen - dev->hard_header_len), | |
fbf33a28 | 2771 | !need_wait, &err); |
69e3c75f | 2772 | |
fbf33a28 KM |
2773 | if (unlikely(skb == NULL)) { |
2774 | /* we assume the socket was initially writeable ... */ | |
2775 | if (likely(len_sum > 0)) | |
2776 | err = len_sum; | |
69e3c75f | 2777 | goto out_status; |
fbf33a28 | 2778 | } |
8d39b4a6 | 2779 | tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, |
c14ac945 | 2780 | addr, hlen, copylen, &sockc); |
dbd46ab4 | 2781 | if (likely(tp_len >= 0) && |
5cfb4c8d | 2782 | tp_len > dev->mtu + reserve && |
1d036d25 | 2783 | !po->has_vnet_hdr && |
3c70c132 DB |
2784 | !packet_extra_vlan_len_allowed(dev, skb)) |
2785 | tp_len = -EMSGSIZE; | |
69e3c75f JB |
2786 | |
2787 | if (unlikely(tp_len < 0)) { | |
8d39b4a6 | 2788 | tpacket_error: |
69e3c75f JB |
2789 | if (po->tp_loss) { |
2790 | __packet_set_status(po, ph, | |
2791 | TP_STATUS_AVAILABLE); | |
2792 | packet_increment_head(&po->tx_ring); | |
2793 | kfree_skb(skb); | |
2794 | continue; | |
2795 | } else { | |
2796 | status = TP_STATUS_WRONG_FORMAT; | |
2797 | err = tp_len; | |
2798 | goto out_status; | |
2799 | } | |
2800 | } | |
2801 | ||
9d2f67e4 JT |
2802 | if (po->has_vnet_hdr) { |
2803 | if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) { | |
2804 | tp_len = -EINVAL; | |
2805 | goto tpacket_error; | |
2806 | } | |
2807 | virtio_net_hdr_set_proto(skb, vnet_hdr); | |
1d036d25 WB |
2808 | } |
2809 | ||
69e3c75f JB |
2810 | skb->destructor = tpacket_destruct_skb; |
2811 | __packet_set_status(po, ph, TP_STATUS_SENDING); | |
b0138408 | 2812 | packet_inc_pending(&po->tx_ring); |
69e3c75f JB |
2813 | |
2814 | status = TP_STATUS_SEND_REQUEST; | |
d346a3fa | 2815 | err = po->xmit(skb); |
eb70df13 JP |
2816 | if (unlikely(err > 0)) { |
2817 | err = net_xmit_errno(err); | |
2818 | if (err && __packet_get_status(po, ph) == | |
2819 | TP_STATUS_AVAILABLE) { | |
2820 | /* skb was destructed already */ | |
2821 | skb = NULL; | |
2822 | goto out_status; | |
2823 | } | |
2824 | /* | |
2825 | * skb was dropped but not destructed yet; | |
2826 | * let's treat it like congestion or err < 0 | |
2827 | */ | |
2828 | err = 0; | |
2829 | } | |
69e3c75f JB |
2830 | packet_increment_head(&po->tx_ring); |
2831 | len_sum += tp_len; | |
b0138408 DB |
2832 | } while (likely((ph != NULL) || |
2833 | /* Note: packet_read_pending() might be slow if we have | |
2834 | * to call it as it's per_cpu variable, but in fast-path | |
2835 | * we already short-circuit the loop with the first | |
2836 | * condition, and luckily don't have to go that path | |
2837 | * anyway. | |
2838 | */ | |
2839 | (need_wait && packet_read_pending(&po->tx_ring)))); | |
69e3c75f JB |
2840 | |
2841 | err = len_sum; | |
2842 | goto out_put; | |
2843 | ||
69e3c75f JB |
2844 | out_status: |
2845 | __packet_set_status(po, ph, status); | |
2846 | kfree_skb(skb); | |
2847 | out_put: | |
e40526cb | 2848 | dev_put(dev); |
69e3c75f JB |
2849 | out: |
2850 | mutex_unlock(&po->pg_vec_lock); | |
2851 | return err; | |
2852 | } | |
69e3c75f | 2853 | |
eea49cc9 OJ |
2854 | static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, |
2855 | size_t reserve, size_t len, | |
2856 | size_t linear, int noblock, | |
2857 | int *err) | |
bfd5f4a3 SS |
2858 | { |
2859 | struct sk_buff *skb; | |
2860 | ||
2861 | /* Under a page? Don't bother with paged skb. */ | |
2862 | if (prepad + len < PAGE_SIZE || !linear) | |
2863 | linear = len; | |
2864 | ||
2865 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
28d64271 | 2866 | err, 0); |
bfd5f4a3 SS |
2867 | if (!skb) |
2868 | return NULL; | |
2869 | ||
2870 | skb_reserve(skb, reserve); | |
2871 | skb_put(skb, linear); | |
2872 | skb->data_len = len - linear; | |
2873 | skb->len += len - linear; | |
2874 | ||
2875 | return skb; | |
2876 | } | |
2877 | ||
d346a3fa | 2878 | static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) |
1da177e4 LT |
2879 | { |
2880 | struct sock *sk = sock->sk; | |
342dfc30 | 2881 | DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); |
1da177e4 LT |
2882 | struct sk_buff *skb; |
2883 | struct net_device *dev; | |
0e11c91e | 2884 | __be16 proto; |
486efdc8 | 2885 | unsigned char *addr = NULL; |
827d9780 | 2886 | int err, reserve = 0; |
c7d39e32 | 2887 | struct sockcm_cookie sockc; |
bfd5f4a3 SS |
2888 | struct virtio_net_hdr vnet_hdr = { 0 }; |
2889 | int offset = 0; | |
bfd5f4a3 | 2890 | struct packet_sock *po = pkt_sk(sk); |
da7c9561 | 2891 | bool has_vnet_hdr = false; |
57031eb7 | 2892 | int hlen, tlen, linear; |
3bdc0eba | 2893 | int extra_len = 0; |
1da177e4 LT |
2894 | |
2895 | /* | |
1ce4f28b | 2896 | * Get and verify the address. |
1da177e4 | 2897 | */ |
1ce4f28b | 2898 | |
66e56cd4 | 2899 | if (likely(saddr == NULL)) { |
e40526cb | 2900 | dev = packet_cached_dev_get(po); |
c7d2ef5d | 2901 | proto = READ_ONCE(po->num); |
1da177e4 LT |
2902 | } else { |
2903 | err = -EINVAL; | |
2904 | if (msg->msg_namelen < sizeof(struct sockaddr_ll)) | |
2905 | goto out; | |
0fb375fb EB |
2906 | if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) |
2907 | goto out; | |
1da177e4 | 2908 | proto = saddr->sll_protocol; |
827d9780 | 2909 | dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); |
486efdc8 WB |
2910 | if (sock->type == SOCK_DGRAM) { |
2911 | if (dev && msg->msg_namelen < dev->addr_len + | |
2912 | offsetof(struct sockaddr_ll, sll_addr)) | |
2913 | goto out_unlock; | |
2914 | addr = saddr->sll_addr; | |
2915 | } | |
1da177e4 LT |
2916 | } |
2917 | ||
1da177e4 | 2918 | err = -ENXIO; |
e40526cb | 2919 | if (unlikely(dev == NULL)) |
1da177e4 | 2920 | goto out_unlock; |
d5e76b0a | 2921 | err = -ENETDOWN; |
e40526cb | 2922 | if (unlikely(!(dev->flags & IFF_UP))) |
d5e76b0a DM |
2923 | goto out_unlock; |
2924 | ||
657a0667 | 2925 | sockcm_init(&sockc, sk); |
c7d39e32 EJ |
2926 | sockc.mark = sk->sk_mark; |
2927 | if (msg->msg_controllen) { | |
2928 | err = sock_cmsg_send(sk, msg, &sockc); | |
2929 | if (unlikely(err)) | |
2930 | goto out_unlock; | |
2931 | } | |
2932 | ||
e40526cb DB |
2933 | if (sock->type == SOCK_RAW) |
2934 | reserve = dev->hard_header_len; | |
bfd5f4a3 | 2935 | if (po->has_vnet_hdr) { |
16cc1400 WB |
2936 | err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); |
2937 | if (err) | |
bfd5f4a3 | 2938 | goto out_unlock; |
da7c9561 | 2939 | has_vnet_hdr = true; |
bfd5f4a3 SS |
2940 | } |
2941 | ||
3bdc0eba BG |
2942 | if (unlikely(sock_flag(sk, SOCK_NOFCS))) { |
2943 | if (!netif_supports_nofcs(dev)) { | |
2944 | err = -EPROTONOSUPPORT; | |
2945 | goto out_unlock; | |
2946 | } | |
2947 | extra_len = 4; /* We're doing our own CRC */ | |
2948 | } | |
2949 | ||
1da177e4 | 2950 | err = -EMSGSIZE; |
16cc1400 WB |
2951 | if (!vnet_hdr.gso_type && |
2952 | (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) | |
1da177e4 LT |
2953 | goto out_unlock; |
2954 | ||
bfd5f4a3 | 2955 | err = -ENOBUFS; |
ae641949 HX |
2956 | hlen = LL_RESERVED_SPACE(dev); |
2957 | tlen = dev->needed_tailroom; | |
57031eb7 WB |
2958 | linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); |
2959 | linear = max(linear, min_t(int, len, dev->hard_header_len)); | |
2960 | skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, | |
bfd5f4a3 | 2961 | msg->msg_flags & MSG_DONTWAIT, &err); |
40d4e3df | 2962 | if (skb == NULL) |
1da177e4 LT |
2963 | goto out_unlock; |
2964 | ||
b84bbaf7 | 2965 | skb_reset_network_header(skb); |
1da177e4 | 2966 | |
0c4e8581 | 2967 | err = -EINVAL; |
9c707762 WB |
2968 | if (sock->type == SOCK_DGRAM) { |
2969 | offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); | |
46d2cfb1 | 2970 | if (unlikely(offset < 0)) |
9c707762 | 2971 | goto out_free; |
b84bbaf7 | 2972 | } else if (reserve) { |
9aad13b0 | 2973 | skb_reserve(skb, -reserve); |
88a8121d ND |
2974 | if (len < reserve + sizeof(struct ipv6hdr) && |
2975 | dev->min_header_len != dev->hard_header_len) | |
993675a3 | 2976 | skb_reset_network_header(skb); |
9c707762 | 2977 | } |
1da177e4 LT |
2978 | |
2979 | /* Returns -EFAULT on error */ | |
c0371da6 | 2980 | err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); |
1da177e4 LT |
2981 | if (err) |
2982 | goto out_free; | |
bf84a010 | 2983 | |
9ed988cd WB |
2984 | if (sock->type == SOCK_RAW && |
2985 | !dev_validate_header(dev, skb->data, len)) { | |
2986 | err = -EINVAL; | |
2987 | goto out_free; | |
2988 | } | |
2989 | ||
8f932f76 | 2990 | skb_setup_tx_timestamp(skb, sockc.tsflags); |
1da177e4 | 2991 | |
16cc1400 | 2992 | if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && |
3c70c132 DB |
2993 | !packet_extra_vlan_len_allowed(dev, skb)) { |
2994 | err = -EMSGSIZE; | |
2995 | goto out_free; | |
57f89bfa BG |
2996 | } |
2997 | ||
09effa67 DM |
2998 | skb->protocol = proto; |
2999 | skb->dev = dev; | |
1da177e4 | 3000 | skb->priority = sk->sk_priority; |
c7d39e32 | 3001 | skb->mark = sockc.mark; |
3d0ba8c0 | 3002 | skb->tstamp = sockc.transmit_time; |
0fd5d57b | 3003 | |
da7c9561 | 3004 | if (has_vnet_hdr) { |
db60eb5f | 3005 | err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); |
16cc1400 WB |
3006 | if (err) |
3007 | goto out_free; | |
3008 | len += sizeof(vnet_hdr); | |
9d2f67e4 | 3009 | virtio_net_hdr_set_proto(skb, &vnet_hdr); |
bfd5f4a3 SS |
3010 | } |
3011 | ||
75c65772 | 3012 | packet_parse_headers(skb, sock); |
8fd6c80d | 3013 | |
3bdc0eba BG |
3014 | if (unlikely(extra_len == 4)) |
3015 | skb->no_fcs = 1; | |
3016 | ||
d346a3fa | 3017 | err = po->xmit(skb); |
1da177e4 LT |
3018 | if (err > 0 && (err = net_xmit_errno(err)) != 0) |
3019 | goto out_unlock; | |
3020 | ||
e40526cb | 3021 | dev_put(dev); |
1da177e4 | 3022 | |
40d4e3df | 3023 | return len; |
1da177e4 LT |
3024 | |
3025 | out_free: | |
3026 | kfree_skb(skb); | |
3027 | out_unlock: | |
1160dfa1 | 3028 | dev_put(dev); |
1da177e4 LT |
3029 | out: |
3030 | return err; | |
3031 | } | |
3032 | ||
1b784140 | 3033 | static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) |
69e3c75f | 3034 | { |
69e3c75f JB |
3035 | struct sock *sk = sock->sk; |
3036 | struct packet_sock *po = pkt_sk(sk); | |
d346a3fa | 3037 | |
d1b5bee4 ED |
3038 | /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy. |
3039 | * tpacket_snd() will redo the check safely. | |
3040 | */ | |
3041 | if (data_race(po->tx_ring.pg_vec)) | |
69e3c75f | 3042 | return tpacket_snd(po, msg); |
d1b5bee4 ED |
3043 | |
3044 | return packet_snd(sock, msg, len); | |
69e3c75f JB |
3045 | } |
3046 | ||
1da177e4 LT |
3047 | /* |
3048 | * Close a PACKET socket. This is fairly simple. We immediately go | |
3049 | * to 'closed' state and remove our protocol entry in the device list. | |
3050 | */ | |
3051 | ||
3052 | static int packet_release(struct socket *sock) | |
3053 | { | |
3054 | struct sock *sk = sock->sk; | |
3055 | struct packet_sock *po; | |
2bd624b4 | 3056 | struct packet_fanout *f; |
d12d01d6 | 3057 | struct net *net; |
f6fb8f10 | 3058 | union tpacket_req_u req_u; |
1da177e4 LT |
3059 | |
3060 | if (!sk) | |
3061 | return 0; | |
3062 | ||
3b1e0a65 | 3063 | net = sock_net(sk); |
1da177e4 LT |
3064 | po = pkt_sk(sk); |
3065 | ||
0fa7fa98 | 3066 | mutex_lock(&net->packet.sklist_lock); |
808f5114 | 3067 | sk_del_node_init_rcu(sk); |
0fa7fa98 PE |
3068 | mutex_unlock(&net->packet.sklist_lock); |
3069 | ||
3070 | preempt_disable(); | |
920de804 | 3071 | sock_prot_inuse_add(net, sk->sk_prot, -1); |
0fa7fa98 | 3072 | preempt_enable(); |
1da177e4 | 3073 | |
808f5114 | 3074 | spin_lock(&po->bind_lock); |
ce06b03e | 3075 | unregister_prot_hook(sk, false); |
66e56cd4 DB |
3076 | packet_cached_dev_reset(po); |
3077 | ||
160ff18a BG |
3078 | if (po->prot_hook.dev) { |
3079 | dev_put(po->prot_hook.dev); | |
3080 | po->prot_hook.dev = NULL; | |
3081 | } | |
808f5114 | 3082 | spin_unlock(&po->bind_lock); |
1da177e4 | 3083 | |
1da177e4 | 3084 | packet_flush_mclist(sk); |
1da177e4 | 3085 | |
5171b37d | 3086 | lock_sock(sk); |
9665d5d6 PS |
3087 | if (po->rx_ring.pg_vec) { |
3088 | memset(&req_u, 0, sizeof(req_u)); | |
f6fb8f10 | 3089 | packet_set_ring(sk, &req_u, 1, 0); |
9665d5d6 | 3090 | } |
69e3c75f | 3091 | |
9665d5d6 PS |
3092 | if (po->tx_ring.pg_vec) { |
3093 | memset(&req_u, 0, sizeof(req_u)); | |
f6fb8f10 | 3094 | packet_set_ring(sk, &req_u, 1, 1); |
9665d5d6 | 3095 | } |
5171b37d | 3096 | release_sock(sk); |
1da177e4 | 3097 | |
2bd624b4 | 3098 | f = fanout_release(sk); |
dc99f600 | 3099 | |
808f5114 | 3100 | synchronize_net(); |
2bd624b4 | 3101 | |
afa0925c | 3102 | kfree(po->rollover); |
2bd624b4 AS |
3103 | if (f) { |
3104 | fanout_release_data(f); | |
9c661b0b | 3105 | kvfree(f); |
2bd624b4 | 3106 | } |
1da177e4 LT |
3107 | /* |
3108 | * Now the socket is dead. No more input will appear. | |
3109 | */ | |
1da177e4 LT |
3110 | sock_orphan(sk); |
3111 | sock->sk = NULL; | |
3112 | ||
3113 | /* Purge queues */ | |
3114 | ||
3115 | skb_queue_purge(&sk->sk_receive_queue); | |
b0138408 | 3116 | packet_free_pending(po); |
17ab56a2 | 3117 | sk_refcnt_debug_release(sk); |
1da177e4 LT |
3118 | |
3119 | sock_put(sk); | |
3120 | return 0; | |
3121 | } | |
3122 | ||
3123 | /* | |
3124 | * Attach a packet hook. | |
3125 | */ | |
3126 | ||
30f7ea1c FR |
3127 | static int packet_do_bind(struct sock *sk, const char *name, int ifindex, |
3128 | __be16 proto) | |
1da177e4 LT |
3129 | { |
3130 | struct packet_sock *po = pkt_sk(sk); | |
158cd4af | 3131 | struct net_device *dev_curr; |
902fefb8 DB |
3132 | __be16 proto_curr; |
3133 | bool need_rehook; | |
30f7ea1c FR |
3134 | struct net_device *dev = NULL; |
3135 | int ret = 0; | |
3136 | bool unlisted = false; | |
dc99f600 | 3137 | |
1da177e4 | 3138 | lock_sock(sk); |
1da177e4 | 3139 | spin_lock(&po->bind_lock); |
30f7ea1c FR |
3140 | rcu_read_lock(); |
3141 | ||
4971613c WB |
3142 | if (po->fanout) { |
3143 | ret = -EINVAL; | |
3144 | goto out_unlock; | |
3145 | } | |
3146 | ||
30f7ea1c FR |
3147 | if (name) { |
3148 | dev = dev_get_by_name_rcu(sock_net(sk), name); | |
3149 | if (!dev) { | |
3150 | ret = -ENODEV; | |
3151 | goto out_unlock; | |
3152 | } | |
3153 | } else if (ifindex) { | |
3154 | dev = dev_get_by_index_rcu(sock_net(sk), ifindex); | |
3155 | if (!dev) { | |
3156 | ret = -ENODEV; | |
3157 | goto out_unlock; | |
3158 | } | |
3159 | } | |
3160 | ||
1160dfa1 | 3161 | dev_hold(dev); |
66e56cd4 | 3162 | |
902fefb8 DB |
3163 | proto_curr = po->prot_hook.type; |
3164 | dev_curr = po->prot_hook.dev; | |
3165 | ||
3166 | need_rehook = proto_curr != proto || dev_curr != dev; | |
3167 | ||
3168 | if (need_rehook) { | |
30f7ea1c FR |
3169 | if (po->running) { |
3170 | rcu_read_unlock(); | |
15fe076e ED |
3171 | /* prevents packet_notifier() from calling |
3172 | * register_prot_hook() | |
3173 | */ | |
c7d2ef5d | 3174 | WRITE_ONCE(po->num, 0); |
30f7ea1c FR |
3175 | __unregister_prot_hook(sk, true); |
3176 | rcu_read_lock(); | |
3177 | dev_curr = po->prot_hook.dev; | |
3178 | if (dev) | |
3179 | unlisted = !dev_get_by_index_rcu(sock_net(sk), | |
3180 | dev->ifindex); | |
3181 | } | |
1da177e4 | 3182 | |
15fe076e | 3183 | BUG_ON(po->running); |
c7d2ef5d | 3184 | WRITE_ONCE(po->num, proto); |
902fefb8 | 3185 | po->prot_hook.type = proto; |
902fefb8 | 3186 | |
30f7ea1c FR |
3187 | if (unlikely(unlisted)) { |
3188 | dev_put(dev); | |
3189 | po->prot_hook.dev = NULL; | |
e032f7c9 | 3190 | WRITE_ONCE(po->ifindex, -1); |
30f7ea1c FR |
3191 | packet_cached_dev_reset(po); |
3192 | } else { | |
3193 | po->prot_hook.dev = dev; | |
e032f7c9 | 3194 | WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0); |
30f7ea1c FR |
3195 | packet_cached_dev_assign(po, dev); |
3196 | } | |
902fefb8 | 3197 | } |
1160dfa1 | 3198 | dev_put(dev_curr); |
66e56cd4 | 3199 | |
902fefb8 | 3200 | if (proto == 0 || !need_rehook) |
1da177e4 LT |
3201 | goto out_unlock; |
3202 | ||
30f7ea1c | 3203 | if (!unlisted && (!dev || (dev->flags & IFF_UP))) { |
ce06b03e | 3204 | register_prot_hook(sk); |
be85d4ad UT |
3205 | } else { |
3206 | sk->sk_err = ENETDOWN; | |
3207 | if (!sock_flag(sk, SOCK_DEAD)) | |
e3ae2365 | 3208 | sk_error_report(sk); |
1da177e4 LT |
3209 | } |
3210 | ||
3211 | out_unlock: | |
30f7ea1c | 3212 | rcu_read_unlock(); |
1da177e4 LT |
3213 | spin_unlock(&po->bind_lock); |
3214 | release_sock(sk); | |
30f7ea1c | 3215 | return ret; |
1da177e4 LT |
3216 | } |
3217 | ||
3218 | /* | |
3219 | * Bind a packet socket to a device | |
3220 | */ | |
3221 | ||
40d4e3df ED |
3222 | static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, |
3223 | int addr_len) | |
1da177e4 | 3224 | { |
40d4e3df | 3225 | struct sock *sk = sock->sk; |
540e2894 | 3226 | char name[sizeof(uaddr->sa_data) + 1]; |
1ce4f28b | 3227 | |
1da177e4 LT |
3228 | /* |
3229 | * Check legality | |
3230 | */ | |
1ce4f28b | 3231 | |
8ae55f04 | 3232 | if (addr_len != sizeof(struct sockaddr)) |
1da177e4 | 3233 | return -EINVAL; |
540e2894 AP |
3234 | /* uaddr->sa_data comes from the userspace, it's not guaranteed to be |
3235 | * zero-terminated. | |
3236 | */ | |
3237 | memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); | |
3238 | name[sizeof(uaddr->sa_data)] = 0; | |
1da177e4 | 3239 | |
30f7ea1c | 3240 | return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); |
1da177e4 | 3241 | } |
1da177e4 LT |
3242 | |
3243 | static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) | |
3244 | { | |
40d4e3df ED |
3245 | struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; |
3246 | struct sock *sk = sock->sk; | |
1da177e4 LT |
3247 | |
3248 | /* | |
3249 | * Check legality | |
3250 | */ | |
1ce4f28b | 3251 | |
1da177e4 LT |
3252 | if (addr_len < sizeof(struct sockaddr_ll)) |
3253 | return -EINVAL; | |
3254 | if (sll->sll_family != AF_PACKET) | |
3255 | return -EINVAL; | |
3256 | ||
30f7ea1c FR |
3257 | return packet_do_bind(sk, NULL, sll->sll_ifindex, |
3258 | sll->sll_protocol ? : pkt_sk(sk)->num); | |
1da177e4 LT |
3259 | } |
3260 | ||
3261 | static struct proto packet_proto = { | |
3262 | .name = "PACKET", | |
3263 | .owner = THIS_MODULE, | |
3264 | .obj_size = sizeof(struct packet_sock), | |
3265 | }; | |
3266 | ||
3267 | /* | |
1ce4f28b | 3268 | * Create a packet of type SOCK_PACKET. |
1da177e4 LT |
3269 | */ |
3270 | ||
3f378b68 EP |
3271 | static int packet_create(struct net *net, struct socket *sock, int protocol, |
3272 | int kern) | |
1da177e4 LT |
3273 | { |
3274 | struct sock *sk; | |
3275 | struct packet_sock *po; | |
0e11c91e | 3276 | __be16 proto = (__force __be16)protocol; /* weird, but documented */ |
1da177e4 LT |
3277 | int err; |
3278 | ||
df008c91 | 3279 | if (!ns_capable(net->user_ns, CAP_NET_RAW)) |
1da177e4 | 3280 | return -EPERM; |
be02097c DM |
3281 | if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && |
3282 | sock->type != SOCK_PACKET) | |
1da177e4 LT |
3283 | return -ESOCKTNOSUPPORT; |
3284 | ||
3285 | sock->state = SS_UNCONNECTED; | |
3286 | ||
3287 | err = -ENOBUFS; | |
11aa9c28 | 3288 | sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); |
1da177e4 LT |
3289 | if (sk == NULL) |
3290 | goto out; | |
3291 | ||
3292 | sock->ops = &packet_ops; | |
1da177e4 LT |
3293 | if (sock->type == SOCK_PACKET) |
3294 | sock->ops = &packet_ops_spkt; | |
be02097c | 3295 | |
1da177e4 LT |
3296 | sock_init_data(sock, sk); |
3297 | ||
3298 | po = pkt_sk(sk); | |
89ed5b51 | 3299 | init_completion(&po->skb_completion); |
1da177e4 | 3300 | sk->sk_family = PF_PACKET; |
0e11c91e | 3301 | po->num = proto; |
d346a3fa | 3302 | po->xmit = dev_queue_xmit; |
66e56cd4 | 3303 | |
b0138408 DB |
3304 | err = packet_alloc_pending(po); |
3305 | if (err) | |
3306 | goto out2; | |
3307 | ||
66e56cd4 | 3308 | packet_cached_dev_reset(po); |
1da177e4 LT |
3309 | |
3310 | sk->sk_destruct = packet_sock_destruct; | |
17ab56a2 | 3311 | sk_refcnt_debug_inc(sk); |
1da177e4 LT |
3312 | |
3313 | /* | |
3314 | * Attach a protocol block | |
3315 | */ | |
3316 | ||
3317 | spin_lock_init(&po->bind_lock); | |
905db440 | 3318 | mutex_init(&po->pg_vec_lock); |
0648ab70 | 3319 | po->rollover = NULL; |
1da177e4 | 3320 | po->prot_hook.func = packet_rcv; |
be02097c | 3321 | |
1da177e4 LT |
3322 | if (sock->type == SOCK_PACKET) |
3323 | po->prot_hook.func = packet_rcv_spkt; | |
be02097c | 3324 | |
1da177e4 LT |
3325 | po->prot_hook.af_packet_priv = sk; |
3326 | ||
0e11c91e AV |
3327 | if (proto) { |
3328 | po->prot_hook.type = proto; | |
a6361f0c | 3329 | __register_prot_hook(sk); |
1da177e4 LT |
3330 | } |
3331 | ||
0fa7fa98 | 3332 | mutex_lock(&net->packet.sklist_lock); |
a4dc6a49 | 3333 | sk_add_node_tail_rcu(sk, &net->packet.sklist); |
0fa7fa98 PE |
3334 | mutex_unlock(&net->packet.sklist_lock); |
3335 | ||
3336 | preempt_disable(); | |
3680453c | 3337 | sock_prot_inuse_add(net, &packet_proto, 1); |
0fa7fa98 | 3338 | preempt_enable(); |
808f5114 | 3339 | |
40d4e3df | 3340 | return 0; |
b0138408 DB |
3341 | out2: |
3342 | sk_free(sk); | |
1da177e4 LT |
3343 | out: |
3344 | return err; | |
3345 | } | |
3346 | ||
3347 | /* | |
3348 | * Pull a packet from our receive queue and hand it to the user. | |
3349 | * If necessary we block. | |
3350 | */ | |
3351 | ||
1b784140 YX |
3352 | static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, |
3353 | int flags) | |
1da177e4 LT |
3354 | { |
3355 | struct sock *sk = sock->sk; | |
3356 | struct sk_buff *skb; | |
3357 | int copied, err; | |
bfd5f4a3 | 3358 | int vnet_hdr_len = 0; |
2472d761 | 3359 | unsigned int origlen = 0; |
1da177e4 LT |
3360 | |
3361 | err = -EINVAL; | |
ed85b565 | 3362 | if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) |
1da177e4 LT |
3363 | goto out; |
3364 | ||
3365 | #if 0 | |
3366 | /* What error should we return now? EUNATTACH? */ | |
3367 | if (pkt_sk(sk)->ifindex < 0) | |
3368 | return -ENODEV; | |
3369 | #endif | |
3370 | ||
ed85b565 | 3371 | if (flags & MSG_ERRQUEUE) { |
cb820f8e RC |
3372 | err = sock_recv_errqueue(sk, msg, len, |
3373 | SOL_PACKET, PACKET_TX_TIMESTAMP); | |
ed85b565 RC |
3374 | goto out; |
3375 | } | |
3376 | ||
1da177e4 LT |
3377 | /* |
3378 | * Call the generic datagram receiver. This handles all sorts | |
3379 | * of horrible races and re-entrancy so we can forget about it | |
3380 | * in the protocol layers. | |
3381 | * | |
3382 | * Now it will return ENETDOWN, if device have just gone down, | |
3383 | * but then it will block. | |
3384 | */ | |
3385 | ||
40d4e3df | 3386 | skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); |
1da177e4 LT |
3387 | |
3388 | /* | |
1ce4f28b | 3389 | * An error occurred so return it. Because skb_recv_datagram() |
1da177e4 LT |
3390 | * handles the blocking we don't see and worry about blocking |
3391 | * retries. | |
3392 | */ | |
3393 | ||
8ae55f04 | 3394 | if (skb == NULL) |
1da177e4 LT |
3395 | goto out; |
3396 | ||
9bb6cd65 | 3397 | packet_rcv_try_clear_pressure(pkt_sk(sk)); |
2ccdbaa6 | 3398 | |
bfd5f4a3 | 3399 | if (pkt_sk(sk)->has_vnet_hdr) { |
16cc1400 WB |
3400 | err = packet_rcv_vnet(msg, skb, &len); |
3401 | if (err) | |
bfd5f4a3 | 3402 | goto out_free; |
16cc1400 | 3403 | vnet_hdr_len = sizeof(struct virtio_net_hdr); |
bfd5f4a3 SS |
3404 | } |
3405 | ||
f3d33426 HFS |
3406 | /* You lose any data beyond the buffer you gave. If it worries |
3407 | * a user program they can ask the device for its MTU | |
3408 | * anyway. | |
1da177e4 | 3409 | */ |
1da177e4 | 3410 | copied = skb->len; |
40d4e3df ED |
3411 | if (copied > len) { |
3412 | copied = len; | |
3413 | msg->msg_flags |= MSG_TRUNC; | |
1da177e4 LT |
3414 | } |
3415 | ||
51f3d02b | 3416 | err = skb_copy_datagram_msg(skb, 0, msg, copied); |
1da177e4 LT |
3417 | if (err) |
3418 | goto out_free; | |
3419 | ||
2472d761 EB |
3420 | if (sock->type != SOCK_PACKET) { |
3421 | struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; | |
3422 | ||
3423 | /* Original length was stored in sockaddr_ll fields */ | |
3424 | origlen = PACKET_SKB_CB(skb)->sa.origlen; | |
3425 | sll->sll_family = AF_PACKET; | |
3426 | sll->sll_protocol = skb->protocol; | |
3427 | } | |
3428 | ||
3b885787 | 3429 | sock_recv_ts_and_drops(msg, sk, skb); |
1da177e4 | 3430 | |
f3d33426 | 3431 | if (msg->msg_name) { |
b2cf86e1 WB |
3432 | int copy_len; |
3433 | ||
f3d33426 HFS |
3434 | /* If the address length field is there to be filled |
3435 | * in, we fill it in now. | |
3436 | */ | |
3437 | if (sock->type == SOCK_PACKET) { | |
342dfc30 | 3438 | __sockaddr_check_size(sizeof(struct sockaddr_pkt)); |
f3d33426 | 3439 | msg->msg_namelen = sizeof(struct sockaddr_pkt); |
b2cf86e1 | 3440 | copy_len = msg->msg_namelen; |
f3d33426 HFS |
3441 | } else { |
3442 | struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; | |
2472d761 | 3443 | |
f3d33426 HFS |
3444 | msg->msg_namelen = sll->sll_halen + |
3445 | offsetof(struct sockaddr_ll, sll_addr); | |
b2cf86e1 WB |
3446 | copy_len = msg->msg_namelen; |
3447 | if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { | |
3448 | memset(msg->msg_name + | |
3449 | offsetof(struct sockaddr_ll, sll_addr), | |
3450 | 0, sizeof(sll->sll_addr)); | |
3451 | msg->msg_namelen = sizeof(struct sockaddr_ll); | |
3452 | } | |
f3d33426 | 3453 | } |
b2cf86e1 | 3454 | memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); |
f3d33426 | 3455 | } |
1da177e4 | 3456 | |
8dc41944 | 3457 | if (pkt_sk(sk)->auxdata) { |
ffbc6111 HX |
3458 | struct tpacket_auxdata aux; |
3459 | ||
3460 | aux.tp_status = TP_STATUS_USER; | |
3461 | if (skb->ip_summed == CHECKSUM_PARTIAL) | |
3462 | aux.tp_status |= TP_STATUS_CSUMNOTREADY; | |
682f048b AD |
3463 | else if (skb->pkt_type != PACKET_OUTGOING && |
3464 | (skb->ip_summed == CHECKSUM_COMPLETE || | |
3465 | skb_csum_unnecessary(skb))) | |
3466 | aux.tp_status |= TP_STATUS_CSUM_VALID; | |
3467 | ||
2472d761 | 3468 | aux.tp_len = origlen; |
ffbc6111 HX |
3469 | aux.tp_snaplen = skb->len; |
3470 | aux.tp_mac = 0; | |
bbe735e4 | 3471 | aux.tp_net = skb_network_offset(skb); |
df8a39de JP |
3472 | if (skb_vlan_tag_present(skb)) { |
3473 | aux.tp_vlan_tci = skb_vlan_tag_get(skb); | |
a0cdfcf3 AW |
3474 | aux.tp_vlan_tpid = ntohs(skb->vlan_proto); |
3475 | aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; | |
a3bcc23e BG |
3476 | } else { |
3477 | aux.tp_vlan_tci = 0; | |
a0cdfcf3 | 3478 | aux.tp_vlan_tpid = 0; |
a3bcc23e | 3479 | } |
ffbc6111 | 3480 | put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); |
8dc41944 HX |
3481 | } |
3482 | ||
1da177e4 LT |
3483 | /* |
3484 | * Free or return the buffer as appropriate. Again this | |
3485 | * hides all the races and re-entrancy issues from us. | |
3486 | */ | |
bfd5f4a3 | 3487 | err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); |
1da177e4 LT |
3488 | |
3489 | out_free: | |
3490 | skb_free_datagram(sk, skb); | |
3491 | out: | |
3492 | return err; | |
3493 | } | |
3494 | ||
1da177e4 | 3495 | static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, |
9b2c45d4 | 3496 | int peer) |
1da177e4 LT |
3497 | { |
3498 | struct net_device *dev; | |
3499 | struct sock *sk = sock->sk; | |
3500 | ||
3501 | if (peer) | |
3502 | return -EOPNOTSUPP; | |
3503 | ||
3504 | uaddr->sa_family = AF_PACKET; | |
2dc85bf3 | 3505 | memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); |
654d1f8a | 3506 | rcu_read_lock(); |
e032f7c9 | 3507 | dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex)); |
654d1f8a | 3508 | if (dev) |
2dc85bf3 | 3509 | strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); |
654d1f8a | 3510 | rcu_read_unlock(); |
1da177e4 | 3511 | |
9b2c45d4 | 3512 | return sizeof(*uaddr); |
1da177e4 | 3513 | } |
1da177e4 LT |
3514 | |
3515 | static int packet_getname(struct socket *sock, struct sockaddr *uaddr, | |
9b2c45d4 | 3516 | int peer) |
1da177e4 LT |
3517 | { |
3518 | struct net_device *dev; | |
3519 | struct sock *sk = sock->sk; | |
3520 | struct packet_sock *po = pkt_sk(sk); | |
13cfa97b | 3521 | DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); |
e032f7c9 | 3522 | int ifindex; |
1da177e4 LT |
3523 | |
3524 | if (peer) | |
3525 | return -EOPNOTSUPP; | |
3526 | ||
e032f7c9 | 3527 | ifindex = READ_ONCE(po->ifindex); |
1da177e4 | 3528 | sll->sll_family = AF_PACKET; |
e032f7c9 | 3529 | sll->sll_ifindex = ifindex; |
c7d2ef5d | 3530 | sll->sll_protocol = READ_ONCE(po->num); |
67286640 | 3531 | sll->sll_pkttype = 0; |
654d1f8a | 3532 | rcu_read_lock(); |
e032f7c9 | 3533 | dev = dev_get_by_index_rcu(sock_net(sk), ifindex); |
1da177e4 LT |
3534 | if (dev) { |
3535 | sll->sll_hatype = dev->type; | |
3536 | sll->sll_halen = dev->addr_len; | |
3537 | memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); | |
1da177e4 LT |
3538 | } else { |
3539 | sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ | |
3540 | sll->sll_halen = 0; | |
3541 | } | |
654d1f8a | 3542 | rcu_read_unlock(); |
1da177e4 | 3543 | |
9b2c45d4 | 3544 | return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; |
1da177e4 LT |
3545 | } |
3546 | ||
2aeb0b88 WC |
3547 | static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, |
3548 | int what) | |
1da177e4 LT |
3549 | { |
3550 | switch (i->type) { | |
3551 | case PACKET_MR_MULTICAST: | |
1162563f JP |
3552 | if (i->alen != dev->addr_len) |
3553 | return -EINVAL; | |
1da177e4 | 3554 | if (what > 0) |
22bedad3 | 3555 | return dev_mc_add(dev, i->addr); |
1da177e4 | 3556 | else |
22bedad3 | 3557 | return dev_mc_del(dev, i->addr); |
1da177e4 LT |
3558 | break; |
3559 | case PACKET_MR_PROMISC: | |
2aeb0b88 | 3560 | return dev_set_promiscuity(dev, what); |
1da177e4 | 3561 | case PACKET_MR_ALLMULTI: |
2aeb0b88 | 3562 | return dev_set_allmulti(dev, what); |
d95ed927 | 3563 | case PACKET_MR_UNICAST: |
1162563f JP |
3564 | if (i->alen != dev->addr_len) |
3565 | return -EINVAL; | |
d95ed927 | 3566 | if (what > 0) |
a748ee24 | 3567 | return dev_uc_add(dev, i->addr); |
d95ed927 | 3568 | else |
a748ee24 | 3569 | return dev_uc_del(dev, i->addr); |
d95ed927 | 3570 | break; |
40d4e3df ED |
3571 | default: |
3572 | break; | |
1da177e4 | 3573 | } |
2aeb0b88 | 3574 | return 0; |
1da177e4 LT |
3575 | } |
3576 | ||
82f17091 FR |
3577 | static void packet_dev_mclist_delete(struct net_device *dev, |
3578 | struct packet_mclist **mlp) | |
1da177e4 | 3579 | { |
82f17091 FR |
3580 | struct packet_mclist *ml; |
3581 | ||
3582 | while ((ml = *mlp) != NULL) { | |
3583 | if (ml->ifindex == dev->ifindex) { | |
3584 | packet_dev_mc(dev, ml, -1); | |
3585 | *mlp = ml->next; | |
3586 | kfree(ml); | |
3587 | } else | |
3588 | mlp = &ml->next; | |
1da177e4 LT |
3589 | } |
3590 | } | |
3591 | ||
0fb375fb | 3592 | static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) |
1da177e4 LT |
3593 | { |
3594 | struct packet_sock *po = pkt_sk(sk); | |
3595 | struct packet_mclist *ml, *i; | |
3596 | struct net_device *dev; | |
3597 | int err; | |
3598 | ||
3599 | rtnl_lock(); | |
3600 | ||
3601 | err = -ENODEV; | |
3b1e0a65 | 3602 | dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); |
1da177e4 LT |
3603 | if (!dev) |
3604 | goto done; | |
3605 | ||
3606 | err = -EINVAL; | |
1162563f | 3607 | if (mreq->mr_alen > dev->addr_len) |
1da177e4 LT |
3608 | goto done; |
3609 | ||
3610 | err = -ENOBUFS; | |
8b3a7005 | 3611 | i = kmalloc(sizeof(*i), GFP_KERNEL); |
1da177e4 LT |
3612 | if (i == NULL) |
3613 | goto done; | |
3614 | ||
3615 | err = 0; | |
3616 | for (ml = po->mclist; ml; ml = ml->next) { | |
3617 | if (ml->ifindex == mreq->mr_ifindex && | |
3618 | ml->type == mreq->mr_type && | |
3619 | ml->alen == mreq->mr_alen && | |
3620 | memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { | |
3621 | ml->count++; | |
3622 | /* Free the new element ... */ | |
3623 | kfree(i); | |
3624 | goto done; | |
3625 | } | |
3626 | } | |
3627 | ||
3628 | i->type = mreq->mr_type; | |
3629 | i->ifindex = mreq->mr_ifindex; | |
3630 | i->alen = mreq->mr_alen; | |
3631 | memcpy(i->addr, mreq->mr_address, i->alen); | |
309cf37f | 3632 | memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); |
1da177e4 LT |
3633 | i->count = 1; |
3634 | i->next = po->mclist; | |
3635 | po->mclist = i; | |
2aeb0b88 WC |
3636 | err = packet_dev_mc(dev, i, 1); |
3637 | if (err) { | |
3638 | po->mclist = i->next; | |
3639 | kfree(i); | |
3640 | } | |
1da177e4 LT |
3641 | |
3642 | done: | |
3643 | rtnl_unlock(); | |
3644 | return err; | |
3645 | } | |
3646 | ||
0fb375fb | 3647 | static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) |
1da177e4 LT |
3648 | { |
3649 | struct packet_mclist *ml, **mlp; | |
3650 | ||
3651 | rtnl_lock(); | |
3652 | ||
3653 | for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { | |
3654 | if (ml->ifindex == mreq->mr_ifindex && | |
3655 | ml->type == mreq->mr_type && | |
3656 | ml->alen == mreq->mr_alen && | |
3657 | memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { | |
3658 | if (--ml->count == 0) { | |
3659 | struct net_device *dev; | |
3660 | *mlp = ml->next; | |
ad959e76 ED |
3661 | dev = __dev_get_by_index(sock_net(sk), ml->ifindex); |
3662 | if (dev) | |
1da177e4 | 3663 | packet_dev_mc(dev, ml, -1); |
1da177e4 LT |
3664 | kfree(ml); |
3665 | } | |
82f17091 | 3666 | break; |
1da177e4 LT |
3667 | } |
3668 | } | |
3669 | rtnl_unlock(); | |
82f17091 | 3670 | return 0; |
1da177e4 LT |
3671 | } |
3672 | ||
3673 | static void packet_flush_mclist(struct sock *sk) | |
3674 | { | |
3675 | struct packet_sock *po = pkt_sk(sk); | |
3676 | struct packet_mclist *ml; | |
3677 | ||
3678 | if (!po->mclist) | |
3679 | return; | |
3680 | ||
3681 | rtnl_lock(); | |
3682 | while ((ml = po->mclist) != NULL) { | |
3683 | struct net_device *dev; | |
3684 | ||
3685 | po->mclist = ml->next; | |
ad959e76 ED |
3686 | dev = __dev_get_by_index(sock_net(sk), ml->ifindex); |
3687 | if (dev != NULL) | |
1da177e4 | 3688 | packet_dev_mc(dev, ml, -1); |
1da177e4 LT |
3689 | kfree(ml); |
3690 | } | |
3691 | rtnl_unlock(); | |
3692 | } | |
1da177e4 LT |
3693 | |
3694 | static int | |
a7b75c5a CH |
3695 | packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval, |
3696 | unsigned int optlen) | |
1da177e4 LT |
3697 | { |
3698 | struct sock *sk = sock->sk; | |
8dc41944 | 3699 | struct packet_sock *po = pkt_sk(sk); |
1da177e4 LT |
3700 | int ret; |
3701 | ||
3702 | if (level != SOL_PACKET) | |
3703 | return -ENOPROTOOPT; | |
3704 | ||
69e3c75f | 3705 | switch (optname) { |
1ce4f28b | 3706 | case PACKET_ADD_MEMBERSHIP: |
1da177e4 LT |
3707 | case PACKET_DROP_MEMBERSHIP: |
3708 | { | |
0fb375fb EB |
3709 | struct packet_mreq_max mreq; |
3710 | int len = optlen; | |
3711 | memset(&mreq, 0, sizeof(mreq)); | |
3712 | if (len < sizeof(struct packet_mreq)) | |
1da177e4 | 3713 | return -EINVAL; |
0fb375fb EB |
3714 | if (len > sizeof(mreq)) |
3715 | len = sizeof(mreq); | |
a7b75c5a | 3716 | if (copy_from_sockptr(&mreq, optval, len)) |
1da177e4 | 3717 | return -EFAULT; |
0fb375fb EB |
3718 | if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) |
3719 | return -EINVAL; | |
1da177e4 LT |
3720 | if (optname == PACKET_ADD_MEMBERSHIP) |
3721 | ret = packet_mc_add(sk, &mreq); | |
3722 | else | |
3723 | ret = packet_mc_drop(sk, &mreq); | |
3724 | return ret; | |
3725 | } | |
a2efcfa0 | 3726 | |
1da177e4 | 3727 | case PACKET_RX_RING: |
69e3c75f | 3728 | case PACKET_TX_RING: |
1da177e4 | 3729 | { |
f6fb8f10 | 3730 | union tpacket_req_u req_u; |
3731 | int len; | |
1da177e4 | 3732 | |
5171b37d | 3733 | lock_sock(sk); |
f6fb8f10 | 3734 | switch (po->tp_version) { |
3735 | case TPACKET_V1: | |
3736 | case TPACKET_V2: | |
3737 | len = sizeof(req_u.req); | |
3738 | break; | |
3739 | case TPACKET_V3: | |
3740 | default: | |
3741 | len = sizeof(req_u.req3); | |
3742 | break; | |
3743 | } | |
5171b37d ED |
3744 | if (optlen < len) { |
3745 | ret = -EINVAL; | |
3746 | } else { | |
a7b75c5a | 3747 | if (copy_from_sockptr(&req_u.req, optval, len)) |
5171b37d ED |
3748 | ret = -EFAULT; |
3749 | else | |
3750 | ret = packet_set_ring(sk, &req_u, 0, | |
3751 | optname == PACKET_TX_RING); | |
3752 | } | |
3753 | release_sock(sk); | |
3754 | return ret; | |
1da177e4 LT |
3755 | } |
3756 | case PACKET_COPY_THRESH: | |
3757 | { | |
3758 | int val; | |
3759 | ||
40d4e3df | 3760 | if (optlen != sizeof(val)) |
1da177e4 | 3761 | return -EINVAL; |
a7b75c5a | 3762 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
1da177e4 LT |
3763 | return -EFAULT; |
3764 | ||
3765 | pkt_sk(sk)->copy_thresh = val; | |
3766 | return 0; | |
3767 | } | |
bbd6ef87 PM |
3768 | case PACKET_VERSION: |
3769 | { | |
3770 | int val; | |
3771 | ||
3772 | if (optlen != sizeof(val)) | |
3773 | return -EINVAL; | |
a7b75c5a | 3774 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
bbd6ef87 PM |
3775 | return -EFAULT; |
3776 | switch (val) { | |
3777 | case TPACKET_V1: | |
3778 | case TPACKET_V2: | |
f6fb8f10 | 3779 | case TPACKET_V3: |
84ac7260 | 3780 | break; |
bbd6ef87 PM |
3781 | default: |
3782 | return -EINVAL; | |
3783 | } | |
84ac7260 PP |
3784 | lock_sock(sk); |
3785 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { | |
3786 | ret = -EBUSY; | |
3787 | } else { | |
3788 | po->tp_version = val; | |
3789 | ret = 0; | |
3790 | } | |
3791 | release_sock(sk); | |
3792 | return ret; | |
bbd6ef87 | 3793 | } |
8913336a PM |
3794 | case PACKET_RESERVE: |
3795 | { | |
3796 | unsigned int val; | |
3797 | ||
3798 | if (optlen != sizeof(val)) | |
3799 | return -EINVAL; | |
a7b75c5a | 3800 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
8913336a | 3801 | return -EFAULT; |
bcc5364b AK |
3802 | if (val > INT_MAX) |
3803 | return -EINVAL; | |
c27927e3 WB |
3804 | lock_sock(sk); |
3805 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { | |
3806 | ret = -EBUSY; | |
3807 | } else { | |
3808 | po->tp_reserve = val; | |
3809 | ret = 0; | |
3810 | } | |
3811 | release_sock(sk); | |
3812 | return ret; | |
8913336a | 3813 | } |
69e3c75f JB |
3814 | case PACKET_LOSS: |
3815 | { | |
3816 | unsigned int val; | |
3817 | ||
3818 | if (optlen != sizeof(val)) | |
3819 | return -EINVAL; | |
a7b75c5a | 3820 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
69e3c75f | 3821 | return -EFAULT; |
a6361f0c WB |
3822 | |
3823 | lock_sock(sk); | |
3824 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { | |
3825 | ret = -EBUSY; | |
3826 | } else { | |
3827 | po->tp_loss = !!val; | |
3828 | ret = 0; | |
3829 | } | |
3830 | release_sock(sk); | |
3831 | return ret; | |
69e3c75f | 3832 | } |
8dc41944 HX |
3833 | case PACKET_AUXDATA: |
3834 | { | |
3835 | int val; | |
3836 | ||
3837 | if (optlen < sizeof(val)) | |
3838 | return -EINVAL; | |
a7b75c5a | 3839 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
8dc41944 HX |
3840 | return -EFAULT; |
3841 | ||
a6361f0c | 3842 | lock_sock(sk); |
8dc41944 | 3843 | po->auxdata = !!val; |
a6361f0c | 3844 | release_sock(sk); |
8dc41944 HX |
3845 | return 0; |
3846 | } | |
80feaacb PWJ |
3847 | case PACKET_ORIGDEV: |
3848 | { | |
3849 | int val; | |
3850 | ||
3851 | if (optlen < sizeof(val)) | |
3852 | return -EINVAL; | |
a7b75c5a | 3853 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
80feaacb PWJ |
3854 | return -EFAULT; |
3855 | ||
a6361f0c | 3856 | lock_sock(sk); |
80feaacb | 3857 | po->origdev = !!val; |
a6361f0c | 3858 | release_sock(sk); |
80feaacb PWJ |
3859 | return 0; |
3860 | } | |
bfd5f4a3 SS |
3861 | case PACKET_VNET_HDR: |
3862 | { | |
3863 | int val; | |
3864 | ||
3865 | if (sock->type != SOCK_RAW) | |
3866 | return -EINVAL; | |
bfd5f4a3 SS |
3867 | if (optlen < sizeof(val)) |
3868 | return -EINVAL; | |
a7b75c5a | 3869 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
bfd5f4a3 SS |
3870 | return -EFAULT; |
3871 | ||
a6361f0c WB |
3872 | lock_sock(sk); |
3873 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { | |
3874 | ret = -EBUSY; | |
3875 | } else { | |
3876 | po->has_vnet_hdr = !!val; | |
3877 | ret = 0; | |
3878 | } | |
3879 | release_sock(sk); | |
3880 | return ret; | |
bfd5f4a3 | 3881 | } |
614f60fa SM |
3882 | case PACKET_TIMESTAMP: |
3883 | { | |
3884 | int val; | |
3885 | ||
3886 | if (optlen != sizeof(val)) | |
3887 | return -EINVAL; | |
a7b75c5a | 3888 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
614f60fa SM |
3889 | return -EFAULT; |
3890 | ||
3891 | po->tp_tstamp = val; | |
3892 | return 0; | |
3893 | } | |
dc99f600 DM |
3894 | case PACKET_FANOUT: |
3895 | { | |
9c661b0b | 3896 | struct fanout_args args = { 0 }; |
dc99f600 | 3897 | |
9c661b0b | 3898 | if (optlen != sizeof(int) && optlen != sizeof(args)) |
dc99f600 | 3899 | return -EINVAL; |
9c661b0b | 3900 | if (copy_from_sockptr(&args, optval, optlen)) |
dc99f600 DM |
3901 | return -EFAULT; |
3902 | ||
9c661b0b | 3903 | return fanout_add(sk, &args); |
dc99f600 | 3904 | } |
47dceb8e WB |
3905 | case PACKET_FANOUT_DATA: |
3906 | { | |
3907 | if (!po->fanout) | |
3908 | return -EINVAL; | |
3909 | ||
3910 | return fanout_set_data(po, optval, optlen); | |
3911 | } | |
fa788d98 VW |
3912 | case PACKET_IGNORE_OUTGOING: |
3913 | { | |
3914 | int val; | |
3915 | ||
3916 | if (optlen != sizeof(val)) | |
3917 | return -EINVAL; | |
a7b75c5a | 3918 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
fa788d98 VW |
3919 | return -EFAULT; |
3920 | if (val < 0 || val > 1) | |
3921 | return -EINVAL; | |
3922 | ||
3923 | po->prot_hook.ignore_outgoing = !!val; | |
3924 | return 0; | |
3925 | } | |
5920cd3a PC |
3926 | case PACKET_TX_HAS_OFF: |
3927 | { | |
3928 | unsigned int val; | |
3929 | ||
3930 | if (optlen != sizeof(val)) | |
3931 | return -EINVAL; | |
a7b75c5a | 3932 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
5920cd3a | 3933 | return -EFAULT; |
a6361f0c WB |
3934 | |
3935 | lock_sock(sk); | |
25c55b38 | 3936 | if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec) |
a6361f0c | 3937 | po->tp_tx_has_off = !!val; |
25c55b38 | 3938 | |
a6361f0c | 3939 | release_sock(sk); |
5920cd3a PC |
3940 | return 0; |
3941 | } | |
d346a3fa DB |
3942 | case PACKET_QDISC_BYPASS: |
3943 | { | |
3944 | int val; | |
3945 | ||
3946 | if (optlen != sizeof(val)) | |
3947 | return -EINVAL; | |
a7b75c5a | 3948 | if (copy_from_sockptr(&val, optval, sizeof(val))) |
d346a3fa DB |
3949 | return -EFAULT; |
3950 | ||
3951 | po->xmit = val ? packet_direct_xmit : dev_queue_xmit; | |
3952 | return 0; | |
3953 | } | |
1da177e4 LT |
3954 | default: |
3955 | return -ENOPROTOOPT; | |
3956 | } | |
3957 | } | |
3958 | ||
3959 | static int packet_getsockopt(struct socket *sock, int level, int optname, | |
3960 | char __user *optval, int __user *optlen) | |
3961 | { | |
3962 | int len; | |
c06fff6e | 3963 | int val, lv = sizeof(val); |
1da177e4 LT |
3964 | struct sock *sk = sock->sk; |
3965 | struct packet_sock *po = pkt_sk(sk); | |
c06fff6e | 3966 | void *data = &val; |
ee80fbf3 | 3967 | union tpacket_stats_u st; |
a9b63918 | 3968 | struct tpacket_rollover_stats rstats; |
8e8e2951 | 3969 | int drops; |
1da177e4 LT |
3970 | |
3971 | if (level != SOL_PACKET) | |
3972 | return -ENOPROTOOPT; | |
3973 | ||
8ae55f04 KK |
3974 | if (get_user(len, optlen)) |
3975 | return -EFAULT; | |
1da177e4 LT |
3976 | |
3977 | if (len < 0) | |
3978 | return -EINVAL; | |
1ce4f28b | 3979 | |
69e3c75f | 3980 | switch (optname) { |
1da177e4 | 3981 | case PACKET_STATISTICS: |
1da177e4 | 3982 | spin_lock_bh(&sk->sk_receive_queue.lock); |
ee80fbf3 DB |
3983 | memcpy(&st, &po->stats, sizeof(st)); |
3984 | memset(&po->stats, 0, sizeof(po->stats)); | |
3985 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
8e8e2951 | 3986 | drops = atomic_xchg(&po->tp_drops, 0); |
ee80fbf3 | 3987 | |
f6fb8f10 | 3988 | if (po->tp_version == TPACKET_V3) { |
c06fff6e | 3989 | lv = sizeof(struct tpacket_stats_v3); |
8e8e2951 ED |
3990 | st.stats3.tp_drops = drops; |
3991 | st.stats3.tp_packets += drops; | |
ee80fbf3 | 3992 | data = &st.stats3; |
f6fb8f10 | 3993 | } else { |
c06fff6e | 3994 | lv = sizeof(struct tpacket_stats); |
8e8e2951 ED |
3995 | st.stats1.tp_drops = drops; |
3996 | st.stats1.tp_packets += drops; | |
ee80fbf3 | 3997 | data = &st.stats1; |
f6fb8f10 | 3998 | } |
ee80fbf3 | 3999 | |
8dc41944 HX |
4000 | break; |
4001 | case PACKET_AUXDATA: | |
8dc41944 | 4002 | val = po->auxdata; |
80feaacb PWJ |
4003 | break; |
4004 | case PACKET_ORIGDEV: | |
80feaacb | 4005 | val = po->origdev; |
bfd5f4a3 SS |
4006 | break; |
4007 | case PACKET_VNET_HDR: | |
bfd5f4a3 | 4008 | val = po->has_vnet_hdr; |
1da177e4 | 4009 | break; |
bbd6ef87 | 4010 | case PACKET_VERSION: |
bbd6ef87 | 4011 | val = po->tp_version; |
bbd6ef87 PM |
4012 | break; |
4013 | case PACKET_HDRLEN: | |
4014 | if (len > sizeof(int)) | |
4015 | len = sizeof(int); | |
fd2c83b3 AP |
4016 | if (len < sizeof(int)) |
4017 | return -EINVAL; | |
bbd6ef87 PM |
4018 | if (copy_from_user(&val, optval, len)) |
4019 | return -EFAULT; | |
4020 | switch (val) { | |
4021 | case TPACKET_V1: | |
4022 | val = sizeof(struct tpacket_hdr); | |
4023 | break; | |
4024 | case TPACKET_V2: | |
4025 | val = sizeof(struct tpacket2_hdr); | |
4026 | break; | |
f6fb8f10 | 4027 | case TPACKET_V3: |
4028 | val = sizeof(struct tpacket3_hdr); | |
4029 | break; | |
bbd6ef87 PM |
4030 | default: |
4031 | return -EINVAL; | |
4032 | } | |
bbd6ef87 | 4033 | break; |
8913336a | 4034 | case PACKET_RESERVE: |
8913336a | 4035 | val = po->tp_reserve; |
8913336a | 4036 | break; |
69e3c75f | 4037 | case PACKET_LOSS: |
69e3c75f | 4038 | val = po->tp_loss; |
69e3c75f | 4039 | break; |
614f60fa | 4040 | case PACKET_TIMESTAMP: |
614f60fa | 4041 | val = po->tp_tstamp; |
614f60fa | 4042 | break; |
dc99f600 | 4043 | case PACKET_FANOUT: |
dc99f600 DM |
4044 | val = (po->fanout ? |
4045 | ((u32)po->fanout->id | | |
77f65ebd WB |
4046 | ((u32)po->fanout->type << 16) | |
4047 | ((u32)po->fanout->flags << 24)) : | |
dc99f600 | 4048 | 0); |
dc99f600 | 4049 | break; |
fa788d98 VW |
4050 | case PACKET_IGNORE_OUTGOING: |
4051 | val = po->prot_hook.ignore_outgoing; | |
4052 | break; | |
a9b63918 | 4053 | case PACKET_ROLLOVER_STATS: |
57f015f5 | 4054 | if (!po->rollover) |
a9b63918 | 4055 | return -EINVAL; |
57f015f5 MM |
4056 | rstats.tp_all = atomic_long_read(&po->rollover->num); |
4057 | rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); | |
4058 | rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); | |
4059 | data = &rstats; | |
4060 | lv = sizeof(rstats); | |
a9b63918 | 4061 | break; |
5920cd3a PC |
4062 | case PACKET_TX_HAS_OFF: |
4063 | val = po->tp_tx_has_off; | |
4064 | break; | |
d346a3fa DB |
4065 | case PACKET_QDISC_BYPASS: |
4066 | val = packet_use_direct_xmit(po); | |
4067 | break; | |
1da177e4 LT |
4068 | default: |
4069 | return -ENOPROTOOPT; | |
4070 | } | |
4071 | ||
c06fff6e ED |
4072 | if (len > lv) |
4073 | len = lv; | |
8ae55f04 KK |
4074 | if (put_user(len, optlen)) |
4075 | return -EFAULT; | |
8dc41944 HX |
4076 | if (copy_to_user(optval, data, len)) |
4077 | return -EFAULT; | |
8ae55f04 | 4078 | return 0; |
1da177e4 LT |
4079 | } |
4080 | ||
351638e7 JP |
4081 | static int packet_notifier(struct notifier_block *this, |
4082 | unsigned long msg, void *ptr) | |
1da177e4 LT |
4083 | { |
4084 | struct sock *sk; | |
351638e7 | 4085 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
c346dca1 | 4086 | struct net *net = dev_net(dev); |
1da177e4 | 4087 | |
808f5114 | 4088 | rcu_read_lock(); |
b67bfe0d | 4089 | sk_for_each_rcu(sk, &net->packet.sklist) { |
1da177e4 LT |
4090 | struct packet_sock *po = pkt_sk(sk); |
4091 | ||
4092 | switch (msg) { | |
4093 | case NETDEV_UNREGISTER: | |
1da177e4 | 4094 | if (po->mclist) |
82f17091 | 4095 | packet_dev_mclist_delete(dev, &po->mclist); |
df561f66 | 4096 | fallthrough; |
a2efcfa0 | 4097 | |
1da177e4 LT |
4098 | case NETDEV_DOWN: |
4099 | if (dev->ifindex == po->ifindex) { | |
4100 | spin_lock(&po->bind_lock); | |
4101 | if (po->running) { | |
ce06b03e | 4102 | __unregister_prot_hook(sk, false); |
1da177e4 LT |
4103 | sk->sk_err = ENETDOWN; |
4104 | if (!sock_flag(sk, SOCK_DEAD)) | |
e3ae2365 | 4105 | sk_error_report(sk); |
1da177e4 LT |
4106 | } |
4107 | if (msg == NETDEV_UNREGISTER) { | |
66e56cd4 | 4108 | packet_cached_dev_reset(po); |
e032f7c9 | 4109 | WRITE_ONCE(po->ifindex, -1); |
1160dfa1 | 4110 | dev_put(po->prot_hook.dev); |
1da177e4 LT |
4111 | po->prot_hook.dev = NULL; |
4112 | } | |
4113 | spin_unlock(&po->bind_lock); | |
4114 | } | |
4115 | break; | |
4116 | case NETDEV_UP: | |
808f5114 | 4117 | if (dev->ifindex == po->ifindex) { |
4118 | spin_lock(&po->bind_lock); | |
ce06b03e DM |
4119 | if (po->num) |
4120 | register_prot_hook(sk); | |
808f5114 | 4121 | spin_unlock(&po->bind_lock); |
1da177e4 | 4122 | } |
1da177e4 LT |
4123 | break; |
4124 | } | |
4125 | } | |
808f5114 | 4126 | rcu_read_unlock(); |
1da177e4 LT |
4127 | return NOTIFY_DONE; |
4128 | } | |
4129 | ||
4130 | ||
4131 | static int packet_ioctl(struct socket *sock, unsigned int cmd, | |
4132 | unsigned long arg) | |
4133 | { | |
4134 | struct sock *sk = sock->sk; | |
4135 | ||
69e3c75f | 4136 | switch (cmd) { |
40d4e3df ED |
4137 | case SIOCOUTQ: |
4138 | { | |
4139 | int amount = sk_wmem_alloc_get(sk); | |
31e6d363 | 4140 | |
40d4e3df ED |
4141 | return put_user(amount, (int __user *)arg); |
4142 | } | |
4143 | case SIOCINQ: | |
4144 | { | |
4145 | struct sk_buff *skb; | |
4146 | int amount = 0; | |
4147 | ||
4148 | spin_lock_bh(&sk->sk_receive_queue.lock); | |
4149 | skb = skb_peek(&sk->sk_receive_queue); | |
4150 | if (skb) | |
4151 | amount = skb->len; | |
4152 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
4153 | return put_user(amount, (int __user *)arg); | |
4154 | } | |
1da177e4 | 4155 | #ifdef CONFIG_INET |
40d4e3df ED |
4156 | case SIOCADDRT: |
4157 | case SIOCDELRT: | |
4158 | case SIOCDARP: | |
4159 | case SIOCGARP: | |
4160 | case SIOCSARP: | |
4161 | case SIOCGIFADDR: | |
4162 | case SIOCSIFADDR: | |
4163 | case SIOCGIFBRDADDR: | |
4164 | case SIOCSIFBRDADDR: | |
4165 | case SIOCGIFNETMASK: | |
4166 | case SIOCSIFNETMASK: | |
4167 | case SIOCGIFDSTADDR: | |
4168 | case SIOCSIFDSTADDR: | |
4169 | case SIOCSIFFLAGS: | |
40d4e3df | 4170 | return inet_dgram_ops.ioctl(sock, cmd, arg); |
1da177e4 LT |
4171 | #endif |
4172 | ||
40d4e3df ED |
4173 | default: |
4174 | return -ENOIOCTLCMD; | |
1da177e4 LT |
4175 | } |
4176 | return 0; | |
4177 | } | |
4178 | ||
a11e1d43 LT |
4179 | static __poll_t packet_poll(struct file *file, struct socket *sock, |
4180 | poll_table *wait) | |
1da177e4 LT |
4181 | { |
4182 | struct sock *sk = sock->sk; | |
4183 | struct packet_sock *po = pkt_sk(sk); | |
a11e1d43 | 4184 | __poll_t mask = datagram_poll(file, sock, wait); |
1da177e4 LT |
4185 | |
4186 | spin_lock_bh(&sk->sk_receive_queue.lock); | |
69e3c75f | 4187 | if (po->rx_ring.pg_vec) { |
f6fb8f10 | 4188 | if (!packet_previous_rx_frame(po, &po->rx_ring, |
4189 | TP_STATUS_KERNEL)) | |
a9a08845 | 4190 | mask |= EPOLLIN | EPOLLRDNORM; |
1da177e4 | 4191 | } |
9bb6cd65 | 4192 | packet_rcv_try_clear_pressure(po); |
1da177e4 | 4193 | spin_unlock_bh(&sk->sk_receive_queue.lock); |
69e3c75f JB |
4194 | spin_lock_bh(&sk->sk_write_queue.lock); |
4195 | if (po->tx_ring.pg_vec) { | |
4196 | if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) | |
a9a08845 | 4197 | mask |= EPOLLOUT | EPOLLWRNORM; |
69e3c75f JB |
4198 | } |
4199 | spin_unlock_bh(&sk->sk_write_queue.lock); | |
1da177e4 LT |
4200 | return mask; |
4201 | } | |
4202 | ||
4203 | ||
4204 | /* Dirty? Well, I still did not learn better way to account | |
4205 | * for user mmaps. | |
4206 | */ | |
4207 | ||
4208 | static void packet_mm_open(struct vm_area_struct *vma) | |
4209 | { | |
4210 | struct file *file = vma->vm_file; | |
40d4e3df | 4211 | struct socket *sock = file->private_data; |
1da177e4 | 4212 | struct sock *sk = sock->sk; |
1ce4f28b | 4213 | |
1da177e4 LT |
4214 | if (sk) |
4215 | atomic_inc(&pkt_sk(sk)->mapped); | |
4216 | } | |
4217 | ||
4218 | static void packet_mm_close(struct vm_area_struct *vma) | |
4219 | { | |
4220 | struct file *file = vma->vm_file; | |
40d4e3df | 4221 | struct socket *sock = file->private_data; |
1da177e4 | 4222 | struct sock *sk = sock->sk; |
1ce4f28b | 4223 | |
1da177e4 LT |
4224 | if (sk) |
4225 | atomic_dec(&pkt_sk(sk)->mapped); | |
4226 | } | |
4227 | ||
f0f37e2f | 4228 | static const struct vm_operations_struct packet_mmap_ops = { |
40d4e3df ED |
4229 | .open = packet_mm_open, |
4230 | .close = packet_mm_close, | |
1da177e4 LT |
4231 | }; |
4232 | ||
3a7ad063 ED |
4233 | static void free_pg_vec(struct pgv *pg_vec, unsigned int order, |
4234 | unsigned int len) | |
1da177e4 LT |
4235 | { |
4236 | int i; | |
4237 | ||
4ebf0ae2 | 4238 | for (i = 0; i < len; i++) { |
0e3125c7 | 4239 | if (likely(pg_vec[i].buffer)) { |
3a7ad063 ED |
4240 | if (is_vmalloc_addr(pg_vec[i].buffer)) |
4241 | vfree(pg_vec[i].buffer); | |
4242 | else | |
4243 | free_pages((unsigned long)pg_vec[i].buffer, | |
4244 | order); | |
0e3125c7 NH |
4245 | pg_vec[i].buffer = NULL; |
4246 | } | |
1da177e4 LT |
4247 | } |
4248 | kfree(pg_vec); | |
4249 | } | |
4250 | ||
3a7ad063 | 4251 | static char *alloc_one_pg_vec_page(unsigned long order) |
4ebf0ae2 | 4252 | { |
f0d4eb29 | 4253 | char *buffer; |
3a7ad063 ED |
4254 | gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | |
4255 | __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; | |
0e3125c7 | 4256 | |
3a7ad063 | 4257 | buffer = (char *) __get_free_pages(gfp_flags, order); |
0e3125c7 NH |
4258 | if (buffer) |
4259 | return buffer; | |
4260 | ||
3a7ad063 ED |
4261 | /* __get_free_pages failed, fall back to vmalloc */ |
4262 | buffer = vzalloc(array_size((1 << order), PAGE_SIZE)); | |
4263 | if (buffer) | |
4264 | return buffer; | |
0e3125c7 | 4265 | |
3a7ad063 ED |
4266 | /* vmalloc failed, lets dig into swap here */ |
4267 | gfp_flags &= ~__GFP_NORETRY; | |
4268 | buffer = (char *) __get_free_pages(gfp_flags, order); | |
4269 | if (buffer) | |
4270 | return buffer; | |
4271 | ||
4272 | /* complete and utter failure */ | |
4273 | return NULL; | |
4ebf0ae2 DM |
4274 | } |
4275 | ||
3a7ad063 | 4276 | static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) |
4ebf0ae2 DM |
4277 | { |
4278 | unsigned int block_nr = req->tp_block_nr; | |
0e3125c7 | 4279 | struct pgv *pg_vec; |
4ebf0ae2 DM |
4280 | int i; |
4281 | ||
398f0132 | 4282 | pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN); |
4ebf0ae2 DM |
4283 | if (unlikely(!pg_vec)) |
4284 | goto out; | |
4285 | ||
4286 | for (i = 0; i < block_nr; i++) { | |
3a7ad063 | 4287 | pg_vec[i].buffer = alloc_one_pg_vec_page(order); |
0e3125c7 | 4288 | if (unlikely(!pg_vec[i].buffer)) |
4ebf0ae2 DM |
4289 | goto out_free_pgvec; |
4290 | } | |
4291 | ||
4292 | out: | |
4293 | return pg_vec; | |
4294 | ||
4295 | out_free_pgvec: | |
3a7ad063 | 4296 | free_pg_vec(pg_vec, order, block_nr); |
4ebf0ae2 DM |
4297 | pg_vec = NULL; |
4298 | goto out; | |
4299 | } | |
1da177e4 | 4300 | |
f6fb8f10 | 4301 | static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, |
69e3c75f | 4302 | int closing, int tx_ring) |
1da177e4 | 4303 | { |
0e3125c7 | 4304 | struct pgv *pg_vec = NULL; |
1da177e4 | 4305 | struct packet_sock *po = pkt_sk(sk); |
61fad681 | 4306 | unsigned long *rx_owner_map = NULL; |
3a7ad063 | 4307 | int was_running, order = 0; |
69e3c75f JB |
4308 | struct packet_ring_buffer *rb; |
4309 | struct sk_buff_head *rb_queue; | |
0e11c91e | 4310 | __be16 num; |
2a6d6c31 | 4311 | int err; |
f6fb8f10 | 4312 | /* Added to avoid minimal code churn */ |
4313 | struct tpacket_req *req = &req_u->req; | |
4314 | ||
69e3c75f JB |
4315 | rb = tx_ring ? &po->tx_ring : &po->rx_ring; |
4316 | rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; | |
1da177e4 | 4317 | |
69e3c75f JB |
4318 | err = -EBUSY; |
4319 | if (!closing) { | |
4320 | if (atomic_read(&po->mapped)) | |
4321 | goto out; | |
b0138408 | 4322 | if (packet_read_pending(rb)) |
69e3c75f JB |
4323 | goto out; |
4324 | } | |
1da177e4 | 4325 | |
69e3c75f | 4326 | if (req->tp_block_nr) { |
4576cd46 WB |
4327 | unsigned int min_frame_size; |
4328 | ||
69e3c75f JB |
4329 | /* Sanity tests and some calculations */ |
4330 | err = -EBUSY; | |
4331 | if (unlikely(rb->pg_vec)) | |
4332 | goto out; | |
1da177e4 | 4333 | |
bbd6ef87 PM |
4334 | switch (po->tp_version) { |
4335 | case TPACKET_V1: | |
4336 | po->tp_hdrlen = TPACKET_HDRLEN; | |
4337 | break; | |
4338 | case TPACKET_V2: | |
4339 | po->tp_hdrlen = TPACKET2_HDRLEN; | |
4340 | break; | |
f6fb8f10 | 4341 | case TPACKET_V3: |
4342 | po->tp_hdrlen = TPACKET3_HDRLEN; | |
4343 | break; | |
bbd6ef87 PM |
4344 | } |
4345 | ||
69e3c75f | 4346 | err = -EINVAL; |
4ebf0ae2 | 4347 | if (unlikely((int)req->tp_block_size <= 0)) |
69e3c75f | 4348 | goto out; |
90836b67 | 4349 | if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) |
69e3c75f | 4350 | goto out; |
4576cd46 | 4351 | min_frame_size = po->tp_hdrlen + po->tp_reserve; |
dc808110 | 4352 | if (po->tp_version >= TPACKET_V3 && |
4576cd46 WB |
4353 | req->tp_block_size < |
4354 | BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) | |
dc808110 | 4355 | goto out; |
4576cd46 | 4356 | if (unlikely(req->tp_frame_size < min_frame_size)) |
69e3c75f | 4357 | goto out; |
4ebf0ae2 | 4358 | if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) |
69e3c75f | 4359 | goto out; |
1da177e4 | 4360 | |
4194b491 TK |
4361 | rb->frames_per_block = req->tp_block_size / req->tp_frame_size; |
4362 | if (unlikely(rb->frames_per_block == 0)) | |
69e3c75f | 4363 | goto out; |
fc62814d | 4364 | if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) |
8f8d28e4 | 4365 | goto out; |
69e3c75f JB |
4366 | if (unlikely((rb->frames_per_block * req->tp_block_nr) != |
4367 | req->tp_frame_nr)) | |
4368 | goto out; | |
1da177e4 LT |
4369 | |
4370 | err = -ENOMEM; | |
3a7ad063 ED |
4371 | order = get_order(req->tp_block_size); |
4372 | pg_vec = alloc_pg_vec(req, order); | |
4ebf0ae2 | 4373 | if (unlikely(!pg_vec)) |
1da177e4 | 4374 | goto out; |
f6fb8f10 | 4375 | switch (po->tp_version) { |
4376 | case TPACKET_V3: | |
7f953ab2 SV |
4377 | /* Block transmit is not supported yet */ |
4378 | if (!tx_ring) { | |
e8e85cc5 | 4379 | init_prb_bdqc(po, rb, pg_vec, req_u); |
7f953ab2 SV |
4380 | } else { |
4381 | struct tpacket_req3 *req3 = &req_u->req3; | |
4382 | ||
4383 | if (req3->tp_retire_blk_tov || | |
4384 | req3->tp_sizeof_priv || | |
4385 | req3->tp_feature_req_word) { | |
4386 | err = -EINVAL; | |
55655e3d | 4387 | goto out_free_pg_vec; |
7f953ab2 SV |
4388 | } |
4389 | } | |
d7cf0c34 | 4390 | break; |
f6fb8f10 | 4391 | default: |
61fad681 WB |
4392 | if (!tx_ring) { |
4393 | rx_owner_map = bitmap_alloc(req->tp_frame_nr, | |
4394 | GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO); | |
4395 | if (!rx_owner_map) | |
4396 | goto out_free_pg_vec; | |
4397 | } | |
f6fb8f10 | 4398 | break; |
4399 | } | |
69e3c75f JB |
4400 | } |
4401 | /* Done */ | |
4402 | else { | |
4403 | err = -EINVAL; | |
4ebf0ae2 | 4404 | if (unlikely(req->tp_frame_nr)) |
69e3c75f | 4405 | goto out; |
1da177e4 LT |
4406 | } |
4407 | ||
1da177e4 LT |
4408 | |
4409 | /* Detach socket from network */ | |
4410 | spin_lock(&po->bind_lock); | |
4411 | was_running = po->running; | |
4412 | num = po->num; | |
4413 | if (was_running) { | |
c7d2ef5d | 4414 | WRITE_ONCE(po->num, 0); |
ce06b03e | 4415 | __unregister_prot_hook(sk, false); |
1da177e4 LT |
4416 | } |
4417 | spin_unlock(&po->bind_lock); | |
1ce4f28b | 4418 | |
1da177e4 LT |
4419 | synchronize_net(); |
4420 | ||
4421 | err = -EBUSY; | |
905db440 | 4422 | mutex_lock(&po->pg_vec_lock); |
1da177e4 LT |
4423 | if (closing || atomic_read(&po->mapped) == 0) { |
4424 | err = 0; | |
69e3c75f | 4425 | spin_lock_bh(&rb_queue->lock); |
c053fd96 | 4426 | swap(rb->pg_vec, pg_vec); |
61fad681 WB |
4427 | if (po->tp_version <= TPACKET_V2) |
4428 | swap(rb->rx_owner_map, rx_owner_map); | |
69e3c75f JB |
4429 | rb->frame_max = (req->tp_frame_nr - 1); |
4430 | rb->head = 0; | |
4431 | rb->frame_size = req->tp_frame_size; | |
4432 | spin_unlock_bh(&rb_queue->lock); | |
4433 | ||
3a7ad063 | 4434 | swap(rb->pg_vec_order, order); |
c053fd96 | 4435 | swap(rb->pg_vec_len, req->tp_block_nr); |
69e3c75f JB |
4436 | |
4437 | rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; | |
4438 | po->prot_hook.func = (po->rx_ring.pg_vec) ? | |
4439 | tpacket_rcv : packet_rcv; | |
4440 | skb_queue_purge(rb_queue); | |
1da177e4 | 4441 | if (atomic_read(&po->mapped)) |
40d4e3df ED |
4442 | pr_err("packet_mmap: vma is busy: %d\n", |
4443 | atomic_read(&po->mapped)); | |
1da177e4 | 4444 | } |
905db440 | 4445 | mutex_unlock(&po->pg_vec_lock); |
1da177e4 LT |
4446 | |
4447 | spin_lock(&po->bind_lock); | |
ce06b03e | 4448 | if (was_running) { |
c7d2ef5d | 4449 | WRITE_ONCE(po->num, num); |
ce06b03e | 4450 | register_prot_hook(sk); |
1da177e4 LT |
4451 | } |
4452 | spin_unlock(&po->bind_lock); | |
c800aaf8 | 4453 | if (pg_vec && (po->tp_version > TPACKET_V2)) { |
f6fb8f10 | 4454 | /* Because we don't support block-based V3 on tx-ring */ |
4455 | if (!tx_ring) | |
73d0fcf2 | 4456 | prb_shutdown_retire_blk_timer(po, rb_queue); |
f6fb8f10 | 4457 | } |
1da177e4 | 4458 | |
55655e3d | 4459 | out_free_pg_vec: |
61fad681 | 4460 | bitmap_free(rx_owner_map); |
1da177e4 | 4461 | if (pg_vec) |
3a7ad063 | 4462 | free_pg_vec(pg_vec, order, req->tp_block_nr); |
1da177e4 LT |
4463 | out: |
4464 | return err; | |
4465 | } | |
4466 | ||
69e3c75f JB |
4467 | static int packet_mmap(struct file *file, struct socket *sock, |
4468 | struct vm_area_struct *vma) | |
1da177e4 LT |
4469 | { |
4470 | struct sock *sk = sock->sk; | |
4471 | struct packet_sock *po = pkt_sk(sk); | |
69e3c75f JB |
4472 | unsigned long size, expected_size; |
4473 | struct packet_ring_buffer *rb; | |
1da177e4 LT |
4474 | unsigned long start; |
4475 | int err = -EINVAL; | |
4476 | int i; | |
4477 | ||
4478 | if (vma->vm_pgoff) | |
4479 | return -EINVAL; | |
4480 | ||
905db440 | 4481 | mutex_lock(&po->pg_vec_lock); |
69e3c75f JB |
4482 | |
4483 | expected_size = 0; | |
4484 | for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { | |
4485 | if (rb->pg_vec) { | |
4486 | expected_size += rb->pg_vec_len | |
4487 | * rb->pg_vec_pages | |
4488 | * PAGE_SIZE; | |
4489 | } | |
4490 | } | |
4491 | ||
4492 | if (expected_size == 0) | |
1da177e4 | 4493 | goto out; |
69e3c75f JB |
4494 | |
4495 | size = vma->vm_end - vma->vm_start; | |
4496 | if (size != expected_size) | |
1da177e4 LT |
4497 | goto out; |
4498 | ||
1da177e4 | 4499 | start = vma->vm_start; |
69e3c75f JB |
4500 | for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { |
4501 | if (rb->pg_vec == NULL) | |
4502 | continue; | |
4503 | ||
4504 | for (i = 0; i < rb->pg_vec_len; i++) { | |
0e3125c7 NH |
4505 | struct page *page; |
4506 | void *kaddr = rb->pg_vec[i].buffer; | |
69e3c75f JB |
4507 | int pg_num; |
4508 | ||
c56b4d90 CG |
4509 | for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { |
4510 | page = pgv_to_page(kaddr); | |
69e3c75f JB |
4511 | err = vm_insert_page(vma, start, page); |
4512 | if (unlikely(err)) | |
4513 | goto out; | |
4514 | start += PAGE_SIZE; | |
0e3125c7 | 4515 | kaddr += PAGE_SIZE; |
69e3c75f | 4516 | } |
4ebf0ae2 | 4517 | } |
1da177e4 | 4518 | } |
69e3c75f | 4519 | |
4ebf0ae2 | 4520 | atomic_inc(&po->mapped); |
1da177e4 LT |
4521 | vma->vm_ops = &packet_mmap_ops; |
4522 | err = 0; | |
4523 | ||
4524 | out: | |
905db440 | 4525 | mutex_unlock(&po->pg_vec_lock); |
1da177e4 LT |
4526 | return err; |
4527 | } | |
1da177e4 | 4528 | |
90ddc4f0 | 4529 | static const struct proto_ops packet_ops_spkt = { |
1da177e4 LT |
4530 | .family = PF_PACKET, |
4531 | .owner = THIS_MODULE, | |
4532 | .release = packet_release, | |
4533 | .bind = packet_bind_spkt, | |
4534 | .connect = sock_no_connect, | |
4535 | .socketpair = sock_no_socketpair, | |
4536 | .accept = sock_no_accept, | |
4537 | .getname = packet_getname_spkt, | |
a11e1d43 | 4538 | .poll = datagram_poll, |
1da177e4 | 4539 | .ioctl = packet_ioctl, |
c7cbdbf2 | 4540 | .gettstamp = sock_gettstamp, |
1da177e4 LT |
4541 | .listen = sock_no_listen, |
4542 | .shutdown = sock_no_shutdown, | |
1da177e4 LT |
4543 | .sendmsg = packet_sendmsg_spkt, |
4544 | .recvmsg = packet_recvmsg, | |
4545 | .mmap = sock_no_mmap, | |
4546 | .sendpage = sock_no_sendpage, | |
4547 | }; | |
1da177e4 | 4548 | |
90ddc4f0 | 4549 | static const struct proto_ops packet_ops = { |
1da177e4 LT |
4550 | .family = PF_PACKET, |
4551 | .owner = THIS_MODULE, | |
4552 | .release = packet_release, | |
4553 | .bind = packet_bind, | |
4554 | .connect = sock_no_connect, | |
4555 | .socketpair = sock_no_socketpair, | |
4556 | .accept = sock_no_accept, | |
1ce4f28b | 4557 | .getname = packet_getname, |
a11e1d43 | 4558 | .poll = packet_poll, |
1da177e4 | 4559 | .ioctl = packet_ioctl, |
c7cbdbf2 | 4560 | .gettstamp = sock_gettstamp, |
1da177e4 LT |
4561 | .listen = sock_no_listen, |
4562 | .shutdown = sock_no_shutdown, | |
4563 | .setsockopt = packet_setsockopt, | |
4564 | .getsockopt = packet_getsockopt, | |
4565 | .sendmsg = packet_sendmsg, | |
4566 | .recvmsg = packet_recvmsg, | |
4567 | .mmap = packet_mmap, | |
4568 | .sendpage = sock_no_sendpage, | |
4569 | }; | |
4570 | ||
ec1b4cf7 | 4571 | static const struct net_proto_family packet_family_ops = { |
1da177e4 LT |
4572 | .family = PF_PACKET, |
4573 | .create = packet_create, | |
4574 | .owner = THIS_MODULE, | |
4575 | }; | |
4576 | ||
4577 | static struct notifier_block packet_netdev_notifier = { | |
40d4e3df | 4578 | .notifier_call = packet_notifier, |
1da177e4 LT |
4579 | }; |
4580 | ||
4581 | #ifdef CONFIG_PROC_FS | |
1da177e4 LT |
4582 | |
4583 | static void *packet_seq_start(struct seq_file *seq, loff_t *pos) | |
808f5114 | 4584 | __acquires(RCU) |
1da177e4 | 4585 | { |
e372c414 | 4586 | struct net *net = seq_file_net(seq); |
808f5114 | 4587 | |
4588 | rcu_read_lock(); | |
4589 | return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); | |
1da177e4 LT |
4590 | } |
4591 | ||
4592 | static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4593 | { | |
1bf40954 | 4594 | struct net *net = seq_file_net(seq); |
808f5114 | 4595 | return seq_hlist_next_rcu(v, &net->packet.sklist, pos); |
1da177e4 LT |
4596 | } |
4597 | ||
4598 | static void packet_seq_stop(struct seq_file *seq, void *v) | |
808f5114 | 4599 | __releases(RCU) |
1da177e4 | 4600 | { |
808f5114 | 4601 | rcu_read_unlock(); |
1da177e4 LT |
4602 | } |
4603 | ||
1ce4f28b | 4604 | static int packet_seq_show(struct seq_file *seq, void *v) |
1da177e4 LT |
4605 | { |
4606 | if (v == SEQ_START_TOKEN) | |
abdcd06c BS |
4607 | seq_printf(seq, |
4608 | "%*sRefCnt Type Proto Iface R Rmem User Inode\n", | |
4609 | IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk"); | |
1da177e4 | 4610 | else { |
b7ceabd9 | 4611 | struct sock *s = sk_entry(v); |
1da177e4 LT |
4612 | const struct packet_sock *po = pkt_sk(s); |
4613 | ||
4614 | seq_printf(seq, | |
71338aa7 | 4615 | "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", |
1da177e4 | 4616 | s, |
41c6d650 | 4617 | refcount_read(&s->sk_refcnt), |
1da177e4 | 4618 | s->sk_type, |
c7d2ef5d | 4619 | ntohs(READ_ONCE(po->num)), |
e032f7c9 | 4620 | READ_ONCE(po->ifindex), |
1da177e4 LT |
4621 | po->running, |
4622 | atomic_read(&s->sk_rmem_alloc), | |
a7cb5a49 | 4623 | from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), |
40d4e3df | 4624 | sock_i_ino(s)); |
1da177e4 LT |
4625 | } |
4626 | ||
4627 | return 0; | |
4628 | } | |
4629 | ||
56b3d975 | 4630 | static const struct seq_operations packet_seq_ops = { |
1da177e4 LT |
4631 | .start = packet_seq_start, |
4632 | .next = packet_seq_next, | |
4633 | .stop = packet_seq_stop, | |
4634 | .show = packet_seq_show, | |
4635 | }; | |
1da177e4 LT |
4636 | #endif |
4637 | ||
2c8c1e72 | 4638 | static int __net_init packet_net_init(struct net *net) |
d12d01d6 | 4639 | { |
0fa7fa98 | 4640 | mutex_init(&net->packet.sklist_lock); |
2aaef4e4 | 4641 | INIT_HLIST_HEAD(&net->packet.sklist); |
d12d01d6 | 4642 | |
a268e0f2 | 4643 | #ifdef CONFIG_PROC_FS |
c3506372 CH |
4644 | if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, |
4645 | sizeof(struct seq_net_private))) | |
d12d01d6 | 4646 | return -ENOMEM; |
a268e0f2 | 4647 | #endif /* CONFIG_PROC_FS */ |
d12d01d6 DL |
4648 | |
4649 | return 0; | |
4650 | } | |
4651 | ||
2c8c1e72 | 4652 | static void __net_exit packet_net_exit(struct net *net) |
d12d01d6 | 4653 | { |
ece31ffd | 4654 | remove_proc_entry("packet", net->proc_net); |
669f8f1a | 4655 | WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); |
d12d01d6 DL |
4656 | } |
4657 | ||
4658 | static struct pernet_operations packet_net_ops = { | |
4659 | .init = packet_net_init, | |
4660 | .exit = packet_net_exit, | |
4661 | }; | |
4662 | ||
4663 | ||
1da177e4 LT |
4664 | static void __exit packet_exit(void) |
4665 | { | |
1da177e4 | 4666 | unregister_netdevice_notifier(&packet_netdev_notifier); |
d12d01d6 | 4667 | unregister_pernet_subsys(&packet_net_ops); |
1da177e4 LT |
4668 | sock_unregister(PF_PACKET); |
4669 | proto_unregister(&packet_proto); | |
4670 | } | |
4671 | ||
4672 | static int __init packet_init(void) | |
4673 | { | |
36096f2f | 4674 | int rc; |
1da177e4 | 4675 | |
36096f2f Y |
4676 | rc = proto_register(&packet_proto, 0); |
4677 | if (rc) | |
1da177e4 | 4678 | goto out; |
36096f2f Y |
4679 | rc = sock_register(&packet_family_ops); |
4680 | if (rc) | |
4681 | goto out_proto; | |
4682 | rc = register_pernet_subsys(&packet_net_ops); | |
4683 | if (rc) | |
4684 | goto out_sock; | |
4685 | rc = register_netdevice_notifier(&packet_netdev_notifier); | |
4686 | if (rc) | |
4687 | goto out_pernet; | |
1da177e4 | 4688 | |
36096f2f Y |
4689 | return 0; |
4690 | ||
4691 | out_pernet: | |
4692 | unregister_pernet_subsys(&packet_net_ops); | |
4693 | out_sock: | |
4694 | sock_unregister(PF_PACKET); | |
4695 | out_proto: | |
4696 | proto_unregister(&packet_proto); | |
1da177e4 LT |
4697 | out: |
4698 | return rc; | |
4699 | } | |
4700 | ||
4701 | module_init(packet_init); | |
4702 | module_exit(packet_exit); | |
4703 | MODULE_LICENSE("GPL"); | |
4704 | MODULE_ALIAS_NETPROTO(PF_PACKET); |