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