]>
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 | |
1da177e4 | 49 | #include <linux/types.h> |
1da177e4 | 50 | #include <linux/mm.h> |
4fc268d2 | 51 | #include <linux/capability.h> |
1da177e4 LT |
52 | #include <linux/fcntl.h> |
53 | #include <linux/socket.h> | |
54 | #include <linux/in.h> | |
55 | #include <linux/inet.h> | |
56 | #include <linux/netdevice.h> | |
57 | #include <linux/if_packet.h> | |
58 | #include <linux/wireless.h> | |
ffbc6111 | 59 | #include <linux/kernel.h> |
1da177e4 | 60 | #include <linux/kmod.h> |
5a0e3ad6 | 61 | #include <linux/slab.h> |
0e3125c7 | 62 | #include <linux/vmalloc.h> |
457c4cbc | 63 | #include <net/net_namespace.h> |
1da177e4 LT |
64 | #include <net/ip.h> |
65 | #include <net/protocol.h> | |
66 | #include <linux/skbuff.h> | |
67 | #include <net/sock.h> | |
68 | #include <linux/errno.h> | |
69 | #include <linux/timer.h> | |
7c0f6ba6 | 70 | #include <linux/uaccess.h> |
1da177e4 LT |
71 | #include <asm/ioctls.h> |
72 | #include <asm/page.h> | |
a1f8e7f7 | 73 | #include <asm/cacheflush.h> |
1da177e4 LT |
74 | #include <asm/io.h> |
75 | #include <linux/proc_fs.h> | |
76 | #include <linux/seq_file.h> | |
77 | #include <linux/poll.h> | |
78 | #include <linux/module.h> | |
79 | #include <linux/init.h> | |
905db440 | 80 | #include <linux/mutex.h> |
05423b24 | 81 | #include <linux/if_vlan.h> |
bfd5f4a3 | 82 | #include <linux/virtio_net.h> |
ed85b565 | 83 | #include <linux/errqueue.h> |
614f60fa | 84 | #include <linux/net_tstamp.h> |
b0138408 | 85 | #include <linux/percpu.h> |
1da177e4 LT |
86 | #ifdef CONFIG_INET |
87 | #include <net/inet_common.h> | |
88 | #endif | |
47dceb8e | 89 | #include <linux/bpf.h> |
719c44d3 | 90 | #include <net/compat.h> |
1da177e4 | 91 | |
2787b04b PE |
92 | #include "internal.h" |
93 | ||
1da177e4 LT |
94 | /* |
95 | Assumptions: | |
96 | - if device has no dev->hard_header routine, it adds and removes ll header | |
97 | inside itself. In this case ll header is invisible outside of device, | |
98 | but higher levels still should reserve dev->hard_header_len. | |
99 | Some devices are enough clever to reallocate skb, when header | |
100 | will not fit to reserved space (tunnel), another ones are silly | |
101 | (PPP). | |
102 | - packet socket receives packets with pulled ll header, | |
103 | so that SOCK_RAW should push it back. | |
104 | ||
105 | On receive: | |
106 | ----------- | |
107 | ||
108 | Incoming, dev->hard_header!=NULL | |
b0e380b1 ACM |
109 | mac_header -> ll header |
110 | data -> data | |
1da177e4 LT |
111 | |
112 | Outgoing, dev->hard_header!=NULL | |
b0e380b1 ACM |
113 | mac_header -> ll header |
114 | data -> ll header | |
1da177e4 LT |
115 | |
116 | Incoming, dev->hard_header==NULL | |
b0e380b1 ACM |
117 | mac_header -> UNKNOWN position. It is very likely, that it points to ll |
118 | header. PPP makes it, that is wrong, because introduce | |
db0c58f9 | 119 | assymetry between rx and tx paths. |
b0e380b1 | 120 | data -> data |
1da177e4 LT |
121 | |
122 | Outgoing, dev->hard_header==NULL | |
b0e380b1 ACM |
123 | mac_header -> data. ll header is still not built! |
124 | data -> data | |
1da177e4 LT |
125 | |
126 | Resume | |
127 | If dev->hard_header==NULL we are unlikely to restore sensible ll header. | |
128 | ||
129 | ||
130 | On transmit: | |
131 | ------------ | |
132 | ||
133 | dev->hard_header != NULL | |
b0e380b1 ACM |
134 | mac_header -> ll header |
135 | data -> ll header | |
1da177e4 LT |
136 | |
137 | dev->hard_header == NULL (ll header is added by device, we cannot control it) | |
b0e380b1 ACM |
138 | mac_header -> data |
139 | data -> data | |
1da177e4 LT |
140 | |
141 | We should set nh.raw on output to correct posistion, | |
142 | packet classifier depends on it. | |
143 | */ | |
144 | ||
1da177e4 LT |
145 | /* Private packet socket structures. */ |
146 | ||
0fb375fb EB |
147 | /* identical to struct packet_mreq except it has |
148 | * a longer address field. | |
149 | */ | |
40d4e3df | 150 | struct packet_mreq_max { |
0fb375fb EB |
151 | int mr_ifindex; |
152 | unsigned short mr_type; | |
153 | unsigned short mr_alen; | |
154 | unsigned char mr_address[MAX_ADDR_LEN]; | |
1da177e4 | 155 | }; |
a2efcfa0 | 156 | |
184f489e DB |
157 | union tpacket_uhdr { |
158 | struct tpacket_hdr *h1; | |
159 | struct tpacket2_hdr *h2; | |
160 | struct tpacket3_hdr *h3; | |
161 | void *raw; | |
162 | }; | |
163 | ||
f6fb8f10 | 164 | static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, |
69e3c75f JB |
165 | int closing, int tx_ring); |
166 | ||
f6fb8f10 | 167 | #define V3_ALIGNMENT (8) |
168 | ||
bc59ba39 | 169 | #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT)) |
f6fb8f10 | 170 | |
171 | #define BLK_PLUS_PRIV(sz_of_priv) \ | |
172 | (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT)) | |
173 | ||
f6fb8f10 | 174 | #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status) |
175 | #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts) | |
176 | #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt) | |
177 | #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len) | |
178 | #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num) | |
179 | #define BLOCK_O2PRIV(x) ((x)->offset_to_priv) | |
180 | #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x))) | |
181 | ||
69e3c75f | 182 | struct packet_sock; |
77f65ebd WB |
183 | static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, |
184 | struct packet_type *pt, struct net_device *orig_dev); | |
1da177e4 | 185 | |
f6fb8f10 | 186 | static void *packet_previous_frame(struct packet_sock *po, |
187 | struct packet_ring_buffer *rb, | |
188 | int status); | |
189 | static void packet_increment_head(struct packet_ring_buffer *buff); | |
878cd3ba | 190 | static int prb_curr_blk_in_use(struct tpacket_block_desc *); |
bc59ba39 | 191 | static void *prb_dispatch_next_block(struct tpacket_kbdq_core *, |
f6fb8f10 | 192 | struct packet_sock *); |
bc59ba39 | 193 | static void prb_retire_current_block(struct tpacket_kbdq_core *, |
f6fb8f10 | 194 | struct packet_sock *, unsigned int status); |
bc59ba39 | 195 | static int prb_queue_frozen(struct tpacket_kbdq_core *); |
196 | static void prb_open_block(struct tpacket_kbdq_core *, | |
197 | struct tpacket_block_desc *); | |
17bfd8c8 | 198 | static void prb_retire_rx_blk_timer_expired(struct timer_list *); |
bc59ba39 | 199 | static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *); |
bc59ba39 | 200 | static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *); |
201 | static void prb_clear_rxhash(struct tpacket_kbdq_core *, | |
202 | struct tpacket3_hdr *); | |
203 | static void prb_fill_vlan_info(struct tpacket_kbdq_core *, | |
204 | struct tpacket3_hdr *); | |
1da177e4 | 205 | static void packet_flush_mclist(struct sock *sk); |
865b03f2 | 206 | static u16 packet_pick_tx_queue(struct sk_buff *skb); |
1da177e4 | 207 | |
ffbc6111 | 208 | struct packet_skb_cb { |
ffbc6111 HX |
209 | union { |
210 | struct sockaddr_pkt pkt; | |
2472d761 EB |
211 | union { |
212 | /* Trick: alias skb original length with | |
213 | * ll.sll_family and ll.protocol in order | |
214 | * to save room. | |
215 | */ | |
216 | unsigned int origlen; | |
217 | struct sockaddr_ll ll; | |
218 | }; | |
ffbc6111 HX |
219 | } sa; |
220 | }; | |
221 | ||
d3869efe DW |
222 | #define vio_le() virtio_legacy_is_little_endian() |
223 | ||
ffbc6111 | 224 | #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb)) |
8dc41944 | 225 | |
bc59ba39 | 226 | #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc)) |
f6fb8f10 | 227 | #define GET_PBLOCK_DESC(x, bid) \ |
bc59ba39 | 228 | ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer)) |
f6fb8f10 | 229 | #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \ |
bc59ba39 | 230 | ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer)) |
f6fb8f10 | 231 | #define GET_NEXT_PRB_BLK_NUM(x) \ |
232 | (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \ | |
233 | ((x)->kactive_blk_num+1) : 0) | |
234 | ||
dc99f600 DM |
235 | static void __fanout_unlink(struct sock *sk, struct packet_sock *po); |
236 | static void __fanout_link(struct sock *sk, struct packet_sock *po); | |
237 | ||
d346a3fa DB |
238 | static int packet_direct_xmit(struct sk_buff *skb) |
239 | { | |
865b03f2 | 240 | return dev_direct_xmit(skb, packet_pick_tx_queue(skb)); |
d346a3fa DB |
241 | } |
242 | ||
66e56cd4 DB |
243 | static struct net_device *packet_cached_dev_get(struct packet_sock *po) |
244 | { | |
245 | struct net_device *dev; | |
246 | ||
247 | rcu_read_lock(); | |
248 | dev = rcu_dereference(po->cached_dev); | |
249 | if (likely(dev)) | |
250 | dev_hold(dev); | |
251 | rcu_read_unlock(); | |
252 | ||
253 | return dev; | |
254 | } | |
255 | ||
256 | static void packet_cached_dev_assign(struct packet_sock *po, | |
257 | struct net_device *dev) | |
258 | { | |
259 | rcu_assign_pointer(po->cached_dev, dev); | |
260 | } | |
261 | ||
262 | static void packet_cached_dev_reset(struct packet_sock *po) | |
263 | { | |
264 | RCU_INIT_POINTER(po->cached_dev, NULL); | |
265 | } | |
266 | ||
d346a3fa DB |
267 | static bool packet_use_direct_xmit(const struct packet_sock *po) |
268 | { | |
269 | return po->xmit == packet_direct_xmit; | |
270 | } | |
271 | ||
865b03f2 | 272 | static u16 packet_pick_tx_queue(struct sk_buff *skb) |
0fd5d57b | 273 | { |
865b03f2 | 274 | struct net_device *dev = skb->dev; |
0fd5d57b | 275 | const struct net_device_ops *ops = dev->netdev_ops; |
b71b5837 | 276 | int cpu = raw_smp_processor_id(); |
0fd5d57b DB |
277 | u16 queue_index; |
278 | ||
b71b5837 PA |
279 | #ifdef CONFIG_XPS |
280 | skb->sender_cpu = cpu + 1; | |
281 | #endif | |
282 | skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues); | |
0fd5d57b | 283 | if (ops->ndo_select_queue) { |
a350ecce | 284 | queue_index = ops->ndo_select_queue(dev, skb, NULL); |
0fd5d57b DB |
285 | queue_index = netdev_cap_txqueue(dev, queue_index); |
286 | } else { | |
b71b5837 | 287 | queue_index = netdev_pick_tx(dev, skb, NULL); |
0fd5d57b DB |
288 | } |
289 | ||
865b03f2 | 290 | return queue_index; |
0fd5d57b DB |
291 | } |
292 | ||
a6361f0c | 293 | /* __register_prot_hook must be invoked through register_prot_hook |
ce06b03e DM |
294 | * or from a context in which asynchronous accesses to the packet |
295 | * socket is not possible (packet_create()). | |
296 | */ | |
a6361f0c | 297 | static void __register_prot_hook(struct sock *sk) |
ce06b03e DM |
298 | { |
299 | struct packet_sock *po = pkt_sk(sk); | |
e40526cb | 300 | |
ce06b03e | 301 | if (!po->running) { |
66e56cd4 | 302 | if (po->fanout) |
dc99f600 | 303 | __fanout_link(sk, po); |
66e56cd4 | 304 | else |
dc99f600 | 305 | dev_add_pack(&po->prot_hook); |
e40526cb | 306 | |
ce06b03e DM |
307 | sock_hold(sk); |
308 | po->running = 1; | |
309 | } | |
310 | } | |
311 | ||
a6361f0c WB |
312 | static void register_prot_hook(struct sock *sk) |
313 | { | |
314 | lockdep_assert_held_once(&pkt_sk(sk)->bind_lock); | |
315 | __register_prot_hook(sk); | |
316 | } | |
317 | ||
318 | /* If the sync parameter is true, we will temporarily drop | |
ce06b03e DM |
319 | * the po->bind_lock and do a synchronize_net to make sure no |
320 | * asynchronous packet processing paths still refer to the elements | |
321 | * of po->prot_hook. If the sync parameter is false, it is the | |
322 | * callers responsibility to take care of this. | |
323 | */ | |
324 | static void __unregister_prot_hook(struct sock *sk, bool sync) | |
325 | { | |
326 | struct packet_sock *po = pkt_sk(sk); | |
327 | ||
a6361f0c WB |
328 | lockdep_assert_held_once(&po->bind_lock); |
329 | ||
ce06b03e | 330 | po->running = 0; |
66e56cd4 DB |
331 | |
332 | if (po->fanout) | |
dc99f600 | 333 | __fanout_unlink(sk, po); |
66e56cd4 | 334 | else |
dc99f600 | 335 | __dev_remove_pack(&po->prot_hook); |
e40526cb | 336 | |
ce06b03e DM |
337 | __sock_put(sk); |
338 | ||
339 | if (sync) { | |
340 | spin_unlock(&po->bind_lock); | |
341 | synchronize_net(); | |
342 | spin_lock(&po->bind_lock); | |
343 | } | |
344 | } | |
345 | ||
346 | static void unregister_prot_hook(struct sock *sk, bool sync) | |
347 | { | |
348 | struct packet_sock *po = pkt_sk(sk); | |
349 | ||
350 | if (po->running) | |
351 | __unregister_prot_hook(sk, sync); | |
352 | } | |
353 | ||
6e58040b | 354 | static inline struct page * __pure pgv_to_page(void *addr) |
0af55bb5 CG |
355 | { |
356 | if (is_vmalloc_addr(addr)) | |
357 | return vmalloc_to_page(addr); | |
358 | return virt_to_page(addr); | |
359 | } | |
360 | ||
69e3c75f | 361 | static void __packet_set_status(struct packet_sock *po, void *frame, int status) |
1da177e4 | 362 | { |
184f489e | 363 | union tpacket_uhdr h; |
1da177e4 | 364 | |
69e3c75f | 365 | h.raw = frame; |
bbd6ef87 PM |
366 | switch (po->tp_version) { |
367 | case TPACKET_V1: | |
69e3c75f | 368 | h.h1->tp_status = status; |
0af55bb5 | 369 | flush_dcache_page(pgv_to_page(&h.h1->tp_status)); |
bbd6ef87 PM |
370 | break; |
371 | case TPACKET_V2: | |
69e3c75f | 372 | h.h2->tp_status = status; |
0af55bb5 | 373 | flush_dcache_page(pgv_to_page(&h.h2->tp_status)); |
bbd6ef87 | 374 | break; |
f6fb8f10 | 375 | case TPACKET_V3: |
7f953ab2 SV |
376 | h.h3->tp_status = status; |
377 | flush_dcache_page(pgv_to_page(&h.h3->tp_status)); | |
378 | break; | |
69e3c75f | 379 | default: |
f6fb8f10 | 380 | WARN(1, "TPACKET version not supported.\n"); |
69e3c75f | 381 | BUG(); |
bbd6ef87 | 382 | } |
69e3c75f JB |
383 | |
384 | smp_wmb(); | |
bbd6ef87 PM |
385 | } |
386 | ||
96f657e6 | 387 | static int __packet_get_status(const struct packet_sock *po, void *frame) |
bbd6ef87 | 388 | { |
184f489e | 389 | union tpacket_uhdr h; |
bbd6ef87 | 390 | |
69e3c75f JB |
391 | smp_rmb(); |
392 | ||
bbd6ef87 PM |
393 | h.raw = frame; |
394 | switch (po->tp_version) { | |
395 | case TPACKET_V1: | |
0af55bb5 | 396 | flush_dcache_page(pgv_to_page(&h.h1->tp_status)); |
69e3c75f | 397 | return h.h1->tp_status; |
bbd6ef87 | 398 | case TPACKET_V2: |
0af55bb5 | 399 | flush_dcache_page(pgv_to_page(&h.h2->tp_status)); |
69e3c75f | 400 | return h.h2->tp_status; |
f6fb8f10 | 401 | case TPACKET_V3: |
7f953ab2 SV |
402 | flush_dcache_page(pgv_to_page(&h.h3->tp_status)); |
403 | return h.h3->tp_status; | |
69e3c75f | 404 | default: |
f6fb8f10 | 405 | WARN(1, "TPACKET version not supported.\n"); |
69e3c75f JB |
406 | BUG(); |
407 | return 0; | |
bbd6ef87 | 408 | } |
1da177e4 | 409 | } |
69e3c75f | 410 | |
b9c32fb2 DB |
411 | static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts, |
412 | unsigned int flags) | |
7a51384c DB |
413 | { |
414 | struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb); | |
415 | ||
68a360e8 WB |
416 | if (shhwtstamps && |
417 | (flags & SOF_TIMESTAMPING_RAW_HARDWARE) && | |
418 | ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts)) | |
419 | return TP_STATUS_TS_RAW_HARDWARE; | |
7a51384c DB |
420 | |
421 | if (ktime_to_timespec_cond(skb->tstamp, ts)) | |
b9c32fb2 | 422 | return TP_STATUS_TS_SOFTWARE; |
7a51384c | 423 | |
b9c32fb2 | 424 | return 0; |
7a51384c DB |
425 | } |
426 | ||
b9c32fb2 DB |
427 | static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame, |
428 | struct sk_buff *skb) | |
2e31396f WB |
429 | { |
430 | union tpacket_uhdr h; | |
431 | struct timespec ts; | |
b9c32fb2 | 432 | __u32 ts_status; |
2e31396f | 433 | |
b9c32fb2 DB |
434 | if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) |
435 | return 0; | |
2e31396f WB |
436 | |
437 | h.raw = frame; | |
438 | switch (po->tp_version) { | |
439 | case TPACKET_V1: | |
440 | h.h1->tp_sec = ts.tv_sec; | |
441 | h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; | |
442 | break; | |
443 | case TPACKET_V2: | |
444 | h.h2->tp_sec = ts.tv_sec; | |
445 | h.h2->tp_nsec = ts.tv_nsec; | |
446 | break; | |
447 | case TPACKET_V3: | |
57ea884b DB |
448 | h.h3->tp_sec = ts.tv_sec; |
449 | h.h3->tp_nsec = ts.tv_nsec; | |
450 | break; | |
2e31396f WB |
451 | default: |
452 | WARN(1, "TPACKET version not supported.\n"); | |
453 | BUG(); | |
454 | } | |
455 | ||
456 | /* one flush is safe, as both fields always lie on the same cacheline */ | |
457 | flush_dcache_page(pgv_to_page(&h.h1->tp_sec)); | |
458 | smp_wmb(); | |
b9c32fb2 DB |
459 | |
460 | return ts_status; | |
2e31396f WB |
461 | } |
462 | ||
d4b5bd98 ED |
463 | static void *packet_lookup_frame(const struct packet_sock *po, |
464 | const struct packet_ring_buffer *rb, | |
465 | unsigned int position, | |
466 | int status) | |
69e3c75f JB |
467 | { |
468 | unsigned int pg_vec_pos, frame_offset; | |
184f489e | 469 | union tpacket_uhdr h; |
69e3c75f JB |
470 | |
471 | pg_vec_pos = position / rb->frames_per_block; | |
472 | frame_offset = position % rb->frames_per_block; | |
473 | ||
0e3125c7 NH |
474 | h.raw = rb->pg_vec[pg_vec_pos].buffer + |
475 | (frame_offset * rb->frame_size); | |
69e3c75f JB |
476 | |
477 | if (status != __packet_get_status(po, h.raw)) | |
478 | return NULL; | |
479 | ||
480 | return h.raw; | |
481 | } | |
482 | ||
eea49cc9 | 483 | static void *packet_current_frame(struct packet_sock *po, |
69e3c75f JB |
484 | struct packet_ring_buffer *rb, |
485 | int status) | |
486 | { | |
487 | return packet_lookup_frame(po, rb, rb->head, status); | |
488 | } | |
489 | ||
bc59ba39 | 490 | static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 491 | { |
492 | del_timer_sync(&pkc->retire_blk_timer); | |
493 | } | |
494 | ||
495 | static void prb_shutdown_retire_blk_timer(struct packet_sock *po, | |
f6fb8f10 | 496 | struct sk_buff_head *rb_queue) |
497 | { | |
bc59ba39 | 498 | struct tpacket_kbdq_core *pkc; |
f6fb8f10 | 499 | |
73d0fcf2 | 500 | pkc = GET_PBDQC_FROM_RB(&po->rx_ring); |
f6fb8f10 | 501 | |
ec6f809f | 502 | spin_lock_bh(&rb_queue->lock); |
f6fb8f10 | 503 | pkc->delete_blk_timer = 1; |
ec6f809f | 504 | spin_unlock_bh(&rb_queue->lock); |
f6fb8f10 | 505 | |
506 | prb_del_retire_blk_timer(pkc); | |
507 | } | |
508 | ||
e8e85cc5 | 509 | static void prb_setup_retire_blk_timer(struct packet_sock *po) |
f6fb8f10 | 510 | { |
bc59ba39 | 511 | struct tpacket_kbdq_core *pkc; |
f6fb8f10 | 512 | |
e8e85cc5 | 513 | pkc = GET_PBDQC_FROM_RB(&po->rx_ring); |
17bfd8c8 KC |
514 | timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired, |
515 | 0); | |
516 | pkc->retire_blk_timer.expires = jiffies; | |
f6fb8f10 | 517 | } |
518 | ||
519 | static int prb_calc_retire_blk_tmo(struct packet_sock *po, | |
520 | int blk_size_in_bytes) | |
521 | { | |
522 | struct net_device *dev; | |
523 | unsigned int mbits = 0, msec = 0, div = 0, tmo = 0; | |
7cad1bac | 524 | struct ethtool_link_ksettings ecmd; |
4bc71cb9 | 525 | int err; |
f6fb8f10 | 526 | |
4bc71cb9 JP |
527 | rtnl_lock(); |
528 | dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex); | |
529 | if (unlikely(!dev)) { | |
530 | rtnl_unlock(); | |
f6fb8f10 | 531 | return DEFAULT_PRB_RETIRE_TOV; |
4bc71cb9 | 532 | } |
7cad1bac | 533 | err = __ethtool_get_link_ksettings(dev, &ecmd); |
4bc71cb9 JP |
534 | rtnl_unlock(); |
535 | if (!err) { | |
4bc71cb9 JP |
536 | /* |
537 | * If the link speed is so slow you don't really | |
538 | * need to worry about perf anyways | |
539 | */ | |
7cad1bac DD |
540 | if (ecmd.base.speed < SPEED_1000 || |
541 | ecmd.base.speed == SPEED_UNKNOWN) { | |
4bc71cb9 | 542 | return DEFAULT_PRB_RETIRE_TOV; |
e440cf2c | 543 | } else { |
544 | msec = 1; | |
7cad1bac | 545 | div = ecmd.base.speed / 1000; |
f6fb8f10 | 546 | } |
fcfcfe0b MW |
547 | } else |
548 | return DEFAULT_PRB_RETIRE_TOV; | |
f6fb8f10 | 549 | |
550 | mbits = (blk_size_in_bytes * 8) / (1024 * 1024); | |
551 | ||
552 | if (div) | |
553 | mbits /= div; | |
554 | ||
555 | tmo = mbits * msec; | |
556 | ||
557 | if (div) | |
558 | return tmo+1; | |
559 | return tmo; | |
560 | } | |
561 | ||
bc59ba39 | 562 | static void prb_init_ft_ops(struct tpacket_kbdq_core *p1, |
f6fb8f10 | 563 | union tpacket_req_u *req_u) |
564 | { | |
565 | p1->feature_req_word = req_u->req3.tp_feature_req_word; | |
566 | } | |
567 | ||
568 | static void init_prb_bdqc(struct packet_sock *po, | |
569 | struct packet_ring_buffer *rb, | |
570 | struct pgv *pg_vec, | |
e8e85cc5 | 571 | union tpacket_req_u *req_u) |
f6fb8f10 | 572 | { |
22781a5b | 573 | struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb); |
bc59ba39 | 574 | struct tpacket_block_desc *pbd; |
f6fb8f10 | 575 | |
576 | memset(p1, 0x0, sizeof(*p1)); | |
577 | ||
578 | p1->knxt_seq_num = 1; | |
579 | p1->pkbdq = pg_vec; | |
bc59ba39 | 580 | pbd = (struct tpacket_block_desc *)pg_vec[0].buffer; |
e3192690 | 581 | p1->pkblk_start = pg_vec[0].buffer; |
f6fb8f10 | 582 | p1->kblk_size = req_u->req3.tp_block_size; |
583 | p1->knum_blocks = req_u->req3.tp_block_nr; | |
584 | p1->hdrlen = po->tp_hdrlen; | |
585 | p1->version = po->tp_version; | |
586 | p1->last_kactive_blk_num = 0; | |
ee80fbf3 | 587 | po->stats.stats3.tp_freeze_q_cnt = 0; |
f6fb8f10 | 588 | if (req_u->req3.tp_retire_blk_tov) |
589 | p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov; | |
590 | else | |
591 | p1->retire_blk_tov = prb_calc_retire_blk_tmo(po, | |
592 | req_u->req3.tp_block_size); | |
593 | p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov); | |
594 | p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv; | |
595 | ||
dc808110 | 596 | p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv); |
f6fb8f10 | 597 | prb_init_ft_ops(p1, req_u); |
e8e85cc5 | 598 | prb_setup_retire_blk_timer(po); |
f6fb8f10 | 599 | prb_open_block(p1, pbd); |
600 | } | |
601 | ||
602 | /* Do NOT update the last_blk_num first. | |
603 | * Assumes sk_buff_head lock is held. | |
604 | */ | |
bc59ba39 | 605 | static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 606 | { |
607 | mod_timer(&pkc->retire_blk_timer, | |
608 | jiffies + pkc->tov_in_jiffies); | |
609 | pkc->last_kactive_blk_num = pkc->kactive_blk_num; | |
610 | } | |
611 | ||
612 | /* | |
613 | * Timer logic: | |
614 | * 1) We refresh the timer only when we open a block. | |
615 | * By doing this we don't waste cycles refreshing the timer | |
616 | * on packet-by-packet basis. | |
617 | * | |
618 | * With a 1MB block-size, on a 1Gbps line, it will take | |
619 | * i) ~8 ms to fill a block + ii) memcpy etc. | |
620 | * In this cut we are not accounting for the memcpy time. | |
621 | * | |
622 | * So, if the user sets the 'tmo' to 10ms then the timer | |
623 | * will never fire while the block is still getting filled | |
624 | * (which is what we want). However, the user could choose | |
625 | * to close a block early and that's fine. | |
626 | * | |
627 | * But when the timer does fire, we check whether or not to refresh it. | |
628 | * Since the tmo granularity is in msecs, it is not too expensive | |
629 | * to refresh the timer, lets say every '8' msecs. | |
630 | * Either the user can set the 'tmo' or we can derive it based on | |
631 | * a) line-speed and b) block-size. | |
632 | * prb_calc_retire_blk_tmo() calculates the tmo. | |
633 | * | |
634 | */ | |
17bfd8c8 | 635 | static void prb_retire_rx_blk_timer_expired(struct timer_list *t) |
f6fb8f10 | 636 | { |
17bfd8c8 KC |
637 | struct packet_sock *po = |
638 | from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer); | |
22781a5b | 639 | struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring); |
f6fb8f10 | 640 | unsigned int frozen; |
bc59ba39 | 641 | struct tpacket_block_desc *pbd; |
f6fb8f10 | 642 | |
643 | spin_lock(&po->sk.sk_receive_queue.lock); | |
644 | ||
645 | frozen = prb_queue_frozen(pkc); | |
646 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); | |
647 | ||
648 | if (unlikely(pkc->delete_blk_timer)) | |
649 | goto out; | |
650 | ||
651 | /* We only need to plug the race when the block is partially filled. | |
652 | * tpacket_rcv: | |
653 | * lock(); increment BLOCK_NUM_PKTS; unlock() | |
654 | * copy_bits() is in progress ... | |
655 | * timer fires on other cpu: | |
656 | * we can't retire the current block because copy_bits | |
657 | * is in progress. | |
658 | * | |
659 | */ | |
660 | if (BLOCK_NUM_PKTS(pbd)) { | |
661 | while (atomic_read(&pkc->blk_fill_in_prog)) { | |
662 | /* Waiting for skb_copy_bits to finish... */ | |
663 | cpu_relax(); | |
664 | } | |
665 | } | |
666 | ||
667 | if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) { | |
668 | if (!frozen) { | |
41a50d62 AD |
669 | if (!BLOCK_NUM_PKTS(pbd)) { |
670 | /* An empty block. Just refresh the timer. */ | |
671 | goto refresh_timer; | |
672 | } | |
f6fb8f10 | 673 | prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO); |
674 | if (!prb_dispatch_next_block(pkc, po)) | |
675 | goto refresh_timer; | |
676 | else | |
677 | goto out; | |
678 | } else { | |
679 | /* Case 1. Queue was frozen because user-space was | |
680 | * lagging behind. | |
681 | */ | |
878cd3ba | 682 | if (prb_curr_blk_in_use(pbd)) { |
f6fb8f10 | 683 | /* |
684 | * Ok, user-space is still behind. | |
685 | * So just refresh the timer. | |
686 | */ | |
687 | goto refresh_timer; | |
688 | } else { | |
689 | /* Case 2. queue was frozen,user-space caught up, | |
690 | * now the link went idle && the timer fired. | |
691 | * We don't have a block to close.So we open this | |
692 | * block and restart the timer. | |
693 | * opening a block thaws the queue,restarts timer | |
694 | * Thawing/timer-refresh is a side effect. | |
695 | */ | |
696 | prb_open_block(pkc, pbd); | |
697 | goto out; | |
698 | } | |
699 | } | |
700 | } | |
701 | ||
702 | refresh_timer: | |
703 | _prb_refresh_rx_retire_blk_timer(pkc); | |
704 | ||
705 | out: | |
706 | spin_unlock(&po->sk.sk_receive_queue.lock); | |
707 | } | |
708 | ||
eea49cc9 | 709 | static void prb_flush_block(struct tpacket_kbdq_core *pkc1, |
bc59ba39 | 710 | struct tpacket_block_desc *pbd1, __u32 status) |
f6fb8f10 | 711 | { |
712 | /* Flush everything minus the block header */ | |
713 | ||
714 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 | |
715 | u8 *start, *end; | |
716 | ||
717 | start = (u8 *)pbd1; | |
718 | ||
719 | /* Skip the block header(we know header WILL fit in 4K) */ | |
720 | start += PAGE_SIZE; | |
721 | ||
722 | end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end); | |
723 | for (; start < end; start += PAGE_SIZE) | |
724 | flush_dcache_page(pgv_to_page(start)); | |
725 | ||
726 | smp_wmb(); | |
727 | #endif | |
728 | ||
729 | /* Now update the block status. */ | |
730 | ||
731 | BLOCK_STATUS(pbd1) = status; | |
732 | ||
733 | /* Flush the block header */ | |
734 | ||
735 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 | |
736 | start = (u8 *)pbd1; | |
737 | flush_dcache_page(pgv_to_page(start)); | |
738 | ||
739 | smp_wmb(); | |
740 | #endif | |
741 | } | |
742 | ||
743 | /* | |
744 | * Side effect: | |
745 | * | |
746 | * 1) flush the block | |
747 | * 2) Increment active_blk_num | |
748 | * | |
749 | * Note:We DONT refresh the timer on purpose. | |
750 | * Because almost always the next block will be opened. | |
751 | */ | |
bc59ba39 | 752 | static void prb_close_block(struct tpacket_kbdq_core *pkc1, |
753 | struct tpacket_block_desc *pbd1, | |
f6fb8f10 | 754 | struct packet_sock *po, unsigned int stat) |
755 | { | |
756 | __u32 status = TP_STATUS_USER | stat; | |
757 | ||
758 | struct tpacket3_hdr *last_pkt; | |
bc59ba39 | 759 | struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; |
da413eec | 760 | struct sock *sk = &po->sk; |
f6fb8f10 | 761 | |
8e8e2951 | 762 | if (atomic_read(&po->tp_drops)) |
f6fb8f10 | 763 | status |= TP_STATUS_LOSING; |
764 | ||
765 | last_pkt = (struct tpacket3_hdr *)pkc1->prev; | |
766 | last_pkt->tp_next_offset = 0; | |
767 | ||
768 | /* Get the ts of the last pkt */ | |
769 | if (BLOCK_NUM_PKTS(pbd1)) { | |
770 | h1->ts_last_pkt.ts_sec = last_pkt->tp_sec; | |
771 | h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec; | |
772 | } else { | |
41a50d62 AD |
773 | /* Ok, we tmo'd - so get the current time. |
774 | * | |
775 | * It shouldn't really happen as we don't close empty | |
776 | * blocks. See prb_retire_rx_blk_timer_expired(). | |
777 | */ | |
f6fb8f10 | 778 | struct timespec ts; |
779 | getnstimeofday(&ts); | |
780 | h1->ts_last_pkt.ts_sec = ts.tv_sec; | |
781 | h1->ts_last_pkt.ts_nsec = ts.tv_nsec; | |
782 | } | |
783 | ||
784 | smp_wmb(); | |
785 | ||
786 | /* Flush the block */ | |
787 | prb_flush_block(pkc1, pbd1, status); | |
788 | ||
da413eec DC |
789 | sk->sk_data_ready(sk); |
790 | ||
f6fb8f10 | 791 | pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1); |
792 | } | |
793 | ||
eea49cc9 | 794 | static void prb_thaw_queue(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 795 | { |
796 | pkc->reset_pending_on_curr_blk = 0; | |
797 | } | |
798 | ||
799 | /* | |
800 | * Side effect of opening a block: | |
801 | * | |
802 | * 1) prb_queue is thawed. | |
803 | * 2) retire_blk_timer is refreshed. | |
804 | * | |
805 | */ | |
bc59ba39 | 806 | static void prb_open_block(struct tpacket_kbdq_core *pkc1, |
807 | struct tpacket_block_desc *pbd1) | |
f6fb8f10 | 808 | { |
809 | struct timespec ts; | |
bc59ba39 | 810 | struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1; |
f6fb8f10 | 811 | |
812 | smp_rmb(); | |
813 | ||
8da3056c DB |
814 | /* We could have just memset this but we will lose the |
815 | * flexibility of making the priv area sticky | |
816 | */ | |
f6fb8f10 | 817 | |
8da3056c DB |
818 | BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++; |
819 | BLOCK_NUM_PKTS(pbd1) = 0; | |
820 | BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); | |
f6fb8f10 | 821 | |
8da3056c DB |
822 | getnstimeofday(&ts); |
823 | ||
824 | h1->ts_first_pkt.ts_sec = ts.tv_sec; | |
825 | h1->ts_first_pkt.ts_nsec = ts.tv_nsec; | |
f6fb8f10 | 826 | |
8da3056c DB |
827 | pkc1->pkblk_start = (char *)pbd1; |
828 | pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); | |
829 | ||
830 | BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv); | |
831 | BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN; | |
832 | ||
833 | pbd1->version = pkc1->version; | |
834 | pkc1->prev = pkc1->nxt_offset; | |
835 | pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size; | |
836 | ||
837 | prb_thaw_queue(pkc1); | |
838 | _prb_refresh_rx_retire_blk_timer(pkc1); | |
839 | ||
840 | smp_wmb(); | |
f6fb8f10 | 841 | } |
842 | ||
843 | /* | |
844 | * Queue freeze logic: | |
845 | * 1) Assume tp_block_nr = 8 blocks. | |
846 | * 2) At time 't0', user opens Rx ring. | |
847 | * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7 | |
848 | * 4) user-space is either sleeping or processing block '0'. | |
849 | * 5) tpacket_rcv is currently filling block '7', since there is no space left, | |
850 | * it will close block-7,loop around and try to fill block '0'. | |
851 | * call-flow: | |
852 | * __packet_lookup_frame_in_block | |
853 | * prb_retire_current_block() | |
854 | * prb_dispatch_next_block() | |
855 | * |->(BLOCK_STATUS == USER) evaluates to true | |
856 | * 5.1) Since block-0 is currently in-use, we just freeze the queue. | |
857 | * 6) Now there are two cases: | |
858 | * 6.1) Link goes idle right after the queue is frozen. | |
859 | * But remember, the last open_block() refreshed the timer. | |
860 | * When this timer expires,it will refresh itself so that we can | |
861 | * re-open block-0 in near future. | |
862 | * 6.2) Link is busy and keeps on receiving packets. This is a simple | |
863 | * case and __packet_lookup_frame_in_block will check if block-0 | |
864 | * is free and can now be re-used. | |
865 | */ | |
eea49cc9 | 866 | static void prb_freeze_queue(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 867 | struct packet_sock *po) |
868 | { | |
869 | pkc->reset_pending_on_curr_blk = 1; | |
ee80fbf3 | 870 | po->stats.stats3.tp_freeze_q_cnt++; |
f6fb8f10 | 871 | } |
872 | ||
873 | #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT)) | |
874 | ||
875 | /* | |
876 | * If the next block is free then we will dispatch it | |
877 | * and return a good offset. | |
878 | * Else, we will freeze the queue. | |
879 | * So, caller must check the return value. | |
880 | */ | |
bc59ba39 | 881 | static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 882 | struct packet_sock *po) |
883 | { | |
bc59ba39 | 884 | struct tpacket_block_desc *pbd; |
f6fb8f10 | 885 | |
886 | smp_rmb(); | |
887 | ||
888 | /* 1. Get current block num */ | |
889 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); | |
890 | ||
891 | /* 2. If this block is currently in_use then freeze the queue */ | |
892 | if (TP_STATUS_USER & BLOCK_STATUS(pbd)) { | |
893 | prb_freeze_queue(pkc, po); | |
894 | return NULL; | |
895 | } | |
896 | ||
897 | /* | |
898 | * 3. | |
899 | * open this block and return the offset where the first packet | |
900 | * needs to get stored. | |
901 | */ | |
902 | prb_open_block(pkc, pbd); | |
903 | return (void *)pkc->nxt_offset; | |
904 | } | |
905 | ||
bc59ba39 | 906 | static void prb_retire_current_block(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 907 | struct packet_sock *po, unsigned int status) |
908 | { | |
bc59ba39 | 909 | struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); |
f6fb8f10 | 910 | |
911 | /* retire/close the current block */ | |
912 | if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) { | |
913 | /* | |
914 | * Plug the case where copy_bits() is in progress on | |
915 | * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't | |
916 | * have space to copy the pkt in the current block and | |
917 | * called prb_retire_current_block() | |
918 | * | |
919 | * We don't need to worry about the TMO case because | |
920 | * the timer-handler already handled this case. | |
921 | */ | |
922 | if (!(status & TP_STATUS_BLK_TMO)) { | |
923 | while (atomic_read(&pkc->blk_fill_in_prog)) { | |
924 | /* Waiting for skb_copy_bits to finish... */ | |
925 | cpu_relax(); | |
926 | } | |
927 | } | |
928 | prb_close_block(pkc, pbd, po, status); | |
929 | return; | |
930 | } | |
f6fb8f10 | 931 | } |
932 | ||
878cd3ba | 933 | static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd) |
f6fb8f10 | 934 | { |
935 | return TP_STATUS_USER & BLOCK_STATUS(pbd); | |
936 | } | |
937 | ||
eea49cc9 | 938 | static int prb_queue_frozen(struct tpacket_kbdq_core *pkc) |
f6fb8f10 | 939 | { |
940 | return pkc->reset_pending_on_curr_blk; | |
941 | } | |
942 | ||
eea49cc9 | 943 | static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb) |
014dc8b9 | 944 | __releases(&pkc->blk_fill_in_prog_lock) |
f6fb8f10 | 945 | { |
bc59ba39 | 946 | struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); |
f6fb8f10 | 947 | atomic_dec(&pkc->blk_fill_in_prog); |
948 | } | |
949 | ||
eea49cc9 | 950 | static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 951 | struct tpacket3_hdr *ppd) |
952 | { | |
3958afa1 | 953 | ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb); |
f6fb8f10 | 954 | } |
955 | ||
eea49cc9 | 956 | static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 957 | struct tpacket3_hdr *ppd) |
958 | { | |
959 | ppd->hv1.tp_rxhash = 0; | |
960 | } | |
961 | ||
eea49cc9 | 962 | static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 963 | struct tpacket3_hdr *ppd) |
964 | { | |
df8a39de JP |
965 | if (skb_vlan_tag_present(pkc->skb)) { |
966 | ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb); | |
a0cdfcf3 AW |
967 | ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto); |
968 | ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; | |
f6fb8f10 | 969 | } else { |
9e67030a | 970 | ppd->hv1.tp_vlan_tci = 0; |
a0cdfcf3 | 971 | ppd->hv1.tp_vlan_tpid = 0; |
9e67030a | 972 | ppd->tp_status = TP_STATUS_AVAILABLE; |
f6fb8f10 | 973 | } |
974 | } | |
975 | ||
bc59ba39 | 976 | static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc, |
f6fb8f10 | 977 | struct tpacket3_hdr *ppd) |
978 | { | |
a0cdfcf3 | 979 | ppd->hv1.tp_padding = 0; |
f6fb8f10 | 980 | prb_fill_vlan_info(pkc, ppd); |
981 | ||
982 | if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH) | |
983 | prb_fill_rxhash(pkc, ppd); | |
984 | else | |
985 | prb_clear_rxhash(pkc, ppd); | |
986 | } | |
987 | ||
eea49cc9 | 988 | static void prb_fill_curr_block(char *curr, |
bc59ba39 | 989 | struct tpacket_kbdq_core *pkc, |
990 | struct tpacket_block_desc *pbd, | |
f6fb8f10 | 991 | unsigned int len) |
014dc8b9 | 992 | __acquires(&pkc->blk_fill_in_prog_lock) |
f6fb8f10 | 993 | { |
994 | struct tpacket3_hdr *ppd; | |
995 | ||
996 | ppd = (struct tpacket3_hdr *)curr; | |
997 | ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len); | |
998 | pkc->prev = curr; | |
999 | pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len); | |
1000 | BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len); | |
1001 | BLOCK_NUM_PKTS(pbd) += 1; | |
1002 | atomic_inc(&pkc->blk_fill_in_prog); | |
1003 | prb_run_all_ft_ops(pkc, ppd); | |
1004 | } | |
1005 | ||
1006 | /* Assumes caller has the sk->rx_queue.lock */ | |
1007 | static void *__packet_lookup_frame_in_block(struct packet_sock *po, | |
1008 | struct sk_buff *skb, | |
f6fb8f10 | 1009 | unsigned int len |
1010 | ) | |
1011 | { | |
bc59ba39 | 1012 | struct tpacket_kbdq_core *pkc; |
1013 | struct tpacket_block_desc *pbd; | |
f6fb8f10 | 1014 | char *curr, *end; |
1015 | ||
e3192690 | 1016 | pkc = GET_PBDQC_FROM_RB(&po->rx_ring); |
f6fb8f10 | 1017 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); |
1018 | ||
1019 | /* Queue is frozen when user space is lagging behind */ | |
1020 | if (prb_queue_frozen(pkc)) { | |
1021 | /* | |
1022 | * Check if that last block which caused the queue to freeze, | |
1023 | * is still in_use by user-space. | |
1024 | */ | |
878cd3ba | 1025 | if (prb_curr_blk_in_use(pbd)) { |
f6fb8f10 | 1026 | /* Can't record this packet */ |
1027 | return NULL; | |
1028 | } else { | |
1029 | /* | |
1030 | * Ok, the block was released by user-space. | |
1031 | * Now let's open that block. | |
1032 | * opening a block also thaws the queue. | |
1033 | * Thawing is a side effect. | |
1034 | */ | |
1035 | prb_open_block(pkc, pbd); | |
1036 | } | |
1037 | } | |
1038 | ||
1039 | smp_mb(); | |
1040 | curr = pkc->nxt_offset; | |
1041 | pkc->skb = skb; | |
e3192690 | 1042 | end = (char *)pbd + pkc->kblk_size; |
f6fb8f10 | 1043 | |
1044 | /* first try the current block */ | |
1045 | if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) { | |
1046 | prb_fill_curr_block(curr, pkc, pbd, len); | |
1047 | return (void *)curr; | |
1048 | } | |
1049 | ||
1050 | /* Ok, close the current block */ | |
1051 | prb_retire_current_block(pkc, po, 0); | |
1052 | ||
1053 | /* Now, try to dispatch the next block */ | |
1054 | curr = (char *)prb_dispatch_next_block(pkc, po); | |
1055 | if (curr) { | |
1056 | pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc); | |
1057 | prb_fill_curr_block(curr, pkc, pbd, len); | |
1058 | return (void *)curr; | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * No free blocks are available.user_space hasn't caught up yet. | |
1063 | * Queue was just frozen and now this packet will get dropped. | |
1064 | */ | |
1065 | return NULL; | |
1066 | } | |
1067 | ||
eea49cc9 | 1068 | static void *packet_current_rx_frame(struct packet_sock *po, |
f6fb8f10 | 1069 | struct sk_buff *skb, |
1070 | int status, unsigned int len) | |
1071 | { | |
1072 | char *curr = NULL; | |
1073 | switch (po->tp_version) { | |
1074 | case TPACKET_V1: | |
1075 | case TPACKET_V2: | |
1076 | curr = packet_lookup_frame(po, &po->rx_ring, | |
1077 | po->rx_ring.head, status); | |
1078 | return curr; | |
1079 | case TPACKET_V3: | |
46088059 | 1080 | return __packet_lookup_frame_in_block(po, skb, len); |
f6fb8f10 | 1081 | default: |
1082 | WARN(1, "TPACKET version not supported\n"); | |
1083 | BUG(); | |
99aa3473 | 1084 | return NULL; |
f6fb8f10 | 1085 | } |
1086 | } | |
1087 | ||
dcf70cef ED |
1088 | static void *prb_lookup_block(const struct packet_sock *po, |
1089 | const struct packet_ring_buffer *rb, | |
1090 | unsigned int idx, | |
1091 | int status) | |
f6fb8f10 | 1092 | { |
bc59ba39 | 1093 | struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb); |
77f65ebd | 1094 | struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx); |
f6fb8f10 | 1095 | |
1096 | if (status != BLOCK_STATUS(pbd)) | |
1097 | return NULL; | |
1098 | return pbd; | |
1099 | } | |
1100 | ||
eea49cc9 | 1101 | static int prb_previous_blk_num(struct packet_ring_buffer *rb) |
f6fb8f10 | 1102 | { |
1103 | unsigned int prev; | |
1104 | if (rb->prb_bdqc.kactive_blk_num) | |
1105 | prev = rb->prb_bdqc.kactive_blk_num-1; | |
1106 | else | |
1107 | prev = rb->prb_bdqc.knum_blocks-1; | |
1108 | return prev; | |
1109 | } | |
1110 | ||
1111 | /* Assumes caller has held the rx_queue.lock */ | |
eea49cc9 | 1112 | static void *__prb_previous_block(struct packet_sock *po, |
f6fb8f10 | 1113 | struct packet_ring_buffer *rb, |
1114 | int status) | |
1115 | { | |
1116 | unsigned int previous = prb_previous_blk_num(rb); | |
1117 | return prb_lookup_block(po, rb, previous, status); | |
1118 | } | |
1119 | ||
eea49cc9 | 1120 | static void *packet_previous_rx_frame(struct packet_sock *po, |
f6fb8f10 | 1121 | struct packet_ring_buffer *rb, |
1122 | int status) | |
1123 | { | |
1124 | if (po->tp_version <= TPACKET_V2) | |
1125 | return packet_previous_frame(po, rb, status); | |
1126 | ||
1127 | return __prb_previous_block(po, rb, status); | |
1128 | } | |
1129 | ||
eea49cc9 | 1130 | static void packet_increment_rx_head(struct packet_sock *po, |
f6fb8f10 | 1131 | struct packet_ring_buffer *rb) |
1132 | { | |
1133 | switch (po->tp_version) { | |
1134 | case TPACKET_V1: | |
1135 | case TPACKET_V2: | |
1136 | return packet_increment_head(rb); | |
1137 | case TPACKET_V3: | |
1138 | default: | |
1139 | WARN(1, "TPACKET version not supported.\n"); | |
1140 | BUG(); | |
1141 | return; | |
1142 | } | |
1143 | } | |
1144 | ||
eea49cc9 | 1145 | static void *packet_previous_frame(struct packet_sock *po, |
69e3c75f JB |
1146 | struct packet_ring_buffer *rb, |
1147 | int status) | |
1148 | { | |
1149 | unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max; | |
1150 | return packet_lookup_frame(po, rb, previous, status); | |
1151 | } | |
1152 | ||
eea49cc9 | 1153 | static void packet_increment_head(struct packet_ring_buffer *buff) |
69e3c75f JB |
1154 | { |
1155 | buff->head = buff->head != buff->frame_max ? buff->head+1 : 0; | |
1156 | } | |
1157 | ||
b0138408 DB |
1158 | static void packet_inc_pending(struct packet_ring_buffer *rb) |
1159 | { | |
1160 | this_cpu_inc(*rb->pending_refcnt); | |
1161 | } | |
1162 | ||
1163 | static void packet_dec_pending(struct packet_ring_buffer *rb) | |
1164 | { | |
1165 | this_cpu_dec(*rb->pending_refcnt); | |
1166 | } | |
1167 | ||
1168 | static unsigned int packet_read_pending(const struct packet_ring_buffer *rb) | |
1169 | { | |
1170 | unsigned int refcnt = 0; | |
1171 | int cpu; | |
1172 | ||
1173 | /* We don't use pending refcount in rx_ring. */ | |
1174 | if (rb->pending_refcnt == NULL) | |
1175 | return 0; | |
1176 | ||
1177 | for_each_possible_cpu(cpu) | |
1178 | refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu); | |
1179 | ||
1180 | return refcnt; | |
1181 | } | |
1182 | ||
1183 | static int packet_alloc_pending(struct packet_sock *po) | |
1184 | { | |
1185 | po->rx_ring.pending_refcnt = NULL; | |
1186 | ||
1187 | po->tx_ring.pending_refcnt = alloc_percpu(unsigned int); | |
1188 | if (unlikely(po->tx_ring.pending_refcnt == NULL)) | |
1189 | return -ENOBUFS; | |
1190 | ||
1191 | return 0; | |
1192 | } | |
1193 | ||
1194 | static void packet_free_pending(struct packet_sock *po) | |
1195 | { | |
1196 | free_percpu(po->tx_ring.pending_refcnt); | |
1197 | } | |
1198 | ||
9954729b WB |
1199 | #define ROOM_POW_OFF 2 |
1200 | #define ROOM_NONE 0x0 | |
1201 | #define ROOM_LOW 0x1 | |
1202 | #define ROOM_NORMAL 0x2 | |
1203 | ||
d4b5bd98 | 1204 | static bool __tpacket_has_room(const struct packet_sock *po, int pow_off) |
77f65ebd | 1205 | { |
9954729b WB |
1206 | int idx, len; |
1207 | ||
d4b5bd98 ED |
1208 | len = READ_ONCE(po->rx_ring.frame_max) + 1; |
1209 | idx = READ_ONCE(po->rx_ring.head); | |
9954729b WB |
1210 | if (pow_off) |
1211 | idx += len >> pow_off; | |
1212 | if (idx >= len) | |
1213 | idx -= len; | |
1214 | return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL); | |
1215 | } | |
1216 | ||
dcf70cef | 1217 | static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off) |
9954729b WB |
1218 | { |
1219 | int idx, len; | |
1220 | ||
dcf70cef ED |
1221 | len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks); |
1222 | idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num); | |
9954729b WB |
1223 | if (pow_off) |
1224 | idx += len >> pow_off; | |
1225 | if (idx >= len) | |
1226 | idx -= len; | |
1227 | return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL); | |
1228 | } | |
77f65ebd | 1229 | |
0338a145 ED |
1230 | static int __packet_rcv_has_room(const struct packet_sock *po, |
1231 | const struct sk_buff *skb) | |
9954729b | 1232 | { |
0338a145 | 1233 | const struct sock *sk = &po->sk; |
9954729b WB |
1234 | int ret = ROOM_NONE; |
1235 | ||
1236 | if (po->prot_hook.func != tpacket_rcv) { | |
0338a145 ED |
1237 | int rcvbuf = READ_ONCE(sk->sk_rcvbuf); |
1238 | int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc) | |
1239 | - (skb ? skb->truesize : 0); | |
1240 | ||
1241 | if (avail > (rcvbuf >> ROOM_POW_OFF)) | |
9954729b WB |
1242 | return ROOM_NORMAL; |
1243 | else if (avail > 0) | |
1244 | return ROOM_LOW; | |
1245 | else | |
1246 | return ROOM_NONE; | |
1247 | } | |
77f65ebd | 1248 | |
9954729b WB |
1249 | if (po->tp_version == TPACKET_V3) { |
1250 | if (__tpacket_v3_has_room(po, ROOM_POW_OFF)) | |
1251 | ret = ROOM_NORMAL; | |
1252 | else if (__tpacket_v3_has_room(po, 0)) | |
1253 | ret = ROOM_LOW; | |
1254 | } else { | |
1255 | if (__tpacket_has_room(po, ROOM_POW_OFF)) | |
1256 | ret = ROOM_NORMAL; | |
1257 | else if (__tpacket_has_room(po, 0)) | |
1258 | ret = ROOM_LOW; | |
1259 | } | |
2ccdbaa6 WB |
1260 | |
1261 | return ret; | |
1262 | } | |
1263 | ||
1264 | static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb) | |
1265 | { | |
3a2bb84e | 1266 | int pressure, ret; |
2ccdbaa6 | 1267 | |
54d7c01d | 1268 | ret = __packet_rcv_has_room(po, skb); |
3a2bb84e ED |
1269 | pressure = ret != ROOM_NORMAL; |
1270 | ||
1271 | if (READ_ONCE(po->pressure) != pressure) | |
1272 | WRITE_ONCE(po->pressure, pressure); | |
77f65ebd | 1273 | |
9954729b | 1274 | return ret; |
77f65ebd WB |
1275 | } |
1276 | ||
9bb6cd65 ED |
1277 | static void packet_rcv_try_clear_pressure(struct packet_sock *po) |
1278 | { | |
1279 | if (READ_ONCE(po->pressure) && | |
1280 | __packet_rcv_has_room(po, NULL) == ROOM_NORMAL) | |
1281 | WRITE_ONCE(po->pressure, 0); | |
1282 | } | |
1283 | ||
1da177e4 LT |
1284 | static void packet_sock_destruct(struct sock *sk) |
1285 | { | |
ed85b565 RC |
1286 | skb_queue_purge(&sk->sk_error_queue); |
1287 | ||
547b792c | 1288 | WARN_ON(atomic_read(&sk->sk_rmem_alloc)); |
14afee4b | 1289 | WARN_ON(refcount_read(&sk->sk_wmem_alloc)); |
1da177e4 LT |
1290 | |
1291 | if (!sock_flag(sk, SOCK_DEAD)) { | |
40d4e3df | 1292 | pr_err("Attempt to release alive packet socket: %p\n", sk); |
1da177e4 LT |
1293 | return; |
1294 | } | |
1295 | ||
17ab56a2 | 1296 | sk_refcnt_debug_dec(sk); |
1da177e4 LT |
1297 | } |
1298 | ||
3b3a5b0a WB |
1299 | static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb) |
1300 | { | |
f6cec329 ED |
1301 | u32 *history = po->rollover->history; |
1302 | u32 victim, rxhash; | |
3b3a5b0a WB |
1303 | int i, count = 0; |
1304 | ||
1305 | rxhash = skb_get_hash(skb); | |
1306 | for (i = 0; i < ROLLOVER_HLEN; i++) | |
f6cec329 | 1307 | if (READ_ONCE(history[i]) == rxhash) |
3b3a5b0a WB |
1308 | count++; |
1309 | ||
f6cec329 ED |
1310 | victim = prandom_u32() % ROLLOVER_HLEN; |
1311 | ||
1312 | /* Avoid dirtying the cache line if possible */ | |
1313 | if (READ_ONCE(history[victim]) != rxhash) | |
1314 | WRITE_ONCE(history[victim], rxhash); | |
1315 | ||
3b3a5b0a WB |
1316 | return count > (ROLLOVER_HLEN >> 1); |
1317 | } | |
1318 | ||
77f65ebd WB |
1319 | static unsigned int fanout_demux_hash(struct packet_fanout *f, |
1320 | struct sk_buff *skb, | |
1321 | unsigned int num) | |
dc99f600 | 1322 | { |
eb70db87 | 1323 | return reciprocal_scale(__skb_get_hash_symmetric(skb), num); |
dc99f600 DM |
1324 | } |
1325 | ||
77f65ebd WB |
1326 | static unsigned int fanout_demux_lb(struct packet_fanout *f, |
1327 | struct sk_buff *skb, | |
1328 | unsigned int num) | |
dc99f600 | 1329 | { |
468479e6 | 1330 | unsigned int val = atomic_inc_return(&f->rr_cur); |
dc99f600 | 1331 | |
468479e6 | 1332 | return val % num; |
77f65ebd WB |
1333 | } |
1334 | ||
1335 | static unsigned int fanout_demux_cpu(struct packet_fanout *f, | |
1336 | struct sk_buff *skb, | |
1337 | unsigned int num) | |
1338 | { | |
1339 | return smp_processor_id() % num; | |
dc99f600 DM |
1340 | } |
1341 | ||
5df0ddfb DB |
1342 | static unsigned int fanout_demux_rnd(struct packet_fanout *f, |
1343 | struct sk_buff *skb, | |
1344 | unsigned int num) | |
1345 | { | |
f337db64 | 1346 | return prandom_u32_max(num); |
5df0ddfb DB |
1347 | } |
1348 | ||
77f65ebd WB |
1349 | static unsigned int fanout_demux_rollover(struct packet_fanout *f, |
1350 | struct sk_buff *skb, | |
ad377cab | 1351 | unsigned int idx, bool try_self, |
77f65ebd | 1352 | unsigned int num) |
95ec3eb4 | 1353 | { |
4633c9e0 | 1354 | struct packet_sock *po, *po_next, *po_skip = NULL; |
a9b63918 | 1355 | unsigned int i, j, room = ROOM_NONE; |
95ec3eb4 | 1356 | |
0648ab70 | 1357 | po = pkt_sk(f->arr[idx]); |
3b3a5b0a WB |
1358 | |
1359 | if (try_self) { | |
1360 | room = packet_rcv_has_room(po, skb); | |
1361 | if (room == ROOM_NORMAL || | |
1362 | (room == ROOM_LOW && !fanout_flow_is_huge(po, skb))) | |
1363 | return idx; | |
4633c9e0 | 1364 | po_skip = po; |
3b3a5b0a | 1365 | } |
ad377cab | 1366 | |
0648ab70 | 1367 | i = j = min_t(int, po->rollover->sock, num - 1); |
77f65ebd | 1368 | do { |
2ccdbaa6 | 1369 | po_next = pkt_sk(f->arr[i]); |
3a2bb84e | 1370 | if (po_next != po_skip && !READ_ONCE(po_next->pressure) && |
2ccdbaa6 | 1371 | packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) { |
77f65ebd | 1372 | if (i != j) |
0648ab70 | 1373 | po->rollover->sock = i; |
a9b63918 WB |
1374 | atomic_long_inc(&po->rollover->num); |
1375 | if (room == ROOM_LOW) | |
1376 | atomic_long_inc(&po->rollover->num_huge); | |
77f65ebd WB |
1377 | return i; |
1378 | } | |
ad377cab | 1379 | |
77f65ebd WB |
1380 | if (++i == num) |
1381 | i = 0; | |
1382 | } while (i != j); | |
1383 | ||
a9b63918 | 1384 | atomic_long_inc(&po->rollover->num_failed); |
77f65ebd WB |
1385 | return idx; |
1386 | } | |
1387 | ||
2d36097d NH |
1388 | static unsigned int fanout_demux_qm(struct packet_fanout *f, |
1389 | struct sk_buff *skb, | |
1390 | unsigned int num) | |
1391 | { | |
1392 | return skb_get_queue_mapping(skb) % num; | |
1393 | } | |
1394 | ||
47dceb8e WB |
1395 | static unsigned int fanout_demux_bpf(struct packet_fanout *f, |
1396 | struct sk_buff *skb, | |
1397 | unsigned int num) | |
1398 | { | |
1399 | struct bpf_prog *prog; | |
1400 | unsigned int ret = 0; | |
1401 | ||
1402 | rcu_read_lock(); | |
1403 | prog = rcu_dereference(f->bpf_prog); | |
1404 | if (prog) | |
ff936a04 | 1405 | ret = bpf_prog_run_clear_cb(prog, skb) % num; |
47dceb8e WB |
1406 | rcu_read_unlock(); |
1407 | ||
1408 | return ret; | |
1409 | } | |
1410 | ||
77f65ebd WB |
1411 | static bool fanout_has_flag(struct packet_fanout *f, u16 flag) |
1412 | { | |
1413 | return f->flags & (flag >> 8); | |
95ec3eb4 DM |
1414 | } |
1415 | ||
95ec3eb4 DM |
1416 | static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev, |
1417 | struct packet_type *pt, struct net_device *orig_dev) | |
dc99f600 DM |
1418 | { |
1419 | struct packet_fanout *f = pt->af_packet_priv; | |
f98f4514 | 1420 | unsigned int num = READ_ONCE(f->num_members); |
19bcf9f2 | 1421 | struct net *net = read_pnet(&f->net); |
dc99f600 | 1422 | struct packet_sock *po; |
77f65ebd | 1423 | unsigned int idx; |
dc99f600 | 1424 | |
19bcf9f2 | 1425 | if (!net_eq(dev_net(dev), net) || !num) { |
dc99f600 DM |
1426 | kfree_skb(skb); |
1427 | return 0; | |
1428 | } | |
1429 | ||
3f34b24a | 1430 | if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) { |
19bcf9f2 | 1431 | skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET); |
3f34b24a AD |
1432 | if (!skb) |
1433 | return 0; | |
1434 | } | |
95ec3eb4 DM |
1435 | switch (f->type) { |
1436 | case PACKET_FANOUT_HASH: | |
1437 | default: | |
77f65ebd | 1438 | idx = fanout_demux_hash(f, skb, num); |
95ec3eb4 DM |
1439 | break; |
1440 | case PACKET_FANOUT_LB: | |
77f65ebd | 1441 | idx = fanout_demux_lb(f, skb, num); |
95ec3eb4 DM |
1442 | break; |
1443 | case PACKET_FANOUT_CPU: | |
77f65ebd WB |
1444 | idx = fanout_demux_cpu(f, skb, num); |
1445 | break; | |
5df0ddfb DB |
1446 | case PACKET_FANOUT_RND: |
1447 | idx = fanout_demux_rnd(f, skb, num); | |
1448 | break; | |
2d36097d NH |
1449 | case PACKET_FANOUT_QM: |
1450 | idx = fanout_demux_qm(f, skb, num); | |
1451 | break; | |
77f65ebd | 1452 | case PACKET_FANOUT_ROLLOVER: |
ad377cab | 1453 | idx = fanout_demux_rollover(f, skb, 0, false, num); |
95ec3eb4 | 1454 | break; |
47dceb8e | 1455 | case PACKET_FANOUT_CBPF: |
f2e52095 | 1456 | case PACKET_FANOUT_EBPF: |
47dceb8e WB |
1457 | idx = fanout_demux_bpf(f, skb, num); |
1458 | break; | |
dc99f600 DM |
1459 | } |
1460 | ||
ad377cab WB |
1461 | if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER)) |
1462 | idx = fanout_demux_rollover(f, skb, idx, true, num); | |
dc99f600 | 1463 | |
ad377cab | 1464 | po = pkt_sk(f->arr[idx]); |
dc99f600 DM |
1465 | return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev); |
1466 | } | |
1467 | ||
fff3321d PE |
1468 | DEFINE_MUTEX(fanout_mutex); |
1469 | EXPORT_SYMBOL_GPL(fanout_mutex); | |
dc99f600 | 1470 | static LIST_HEAD(fanout_list); |
4a69a864 | 1471 | static u16 fanout_next_id; |
dc99f600 DM |
1472 | |
1473 | static void __fanout_link(struct sock *sk, struct packet_sock *po) | |
1474 | { | |
1475 | struct packet_fanout *f = po->fanout; | |
1476 | ||
1477 | spin_lock(&f->lock); | |
1478 | f->arr[f->num_members] = sk; | |
1479 | smp_wmb(); | |
1480 | f->num_members++; | |
2bd624b4 AS |
1481 | if (f->num_members == 1) |
1482 | dev_add_pack(&f->prot_hook); | |
dc99f600 DM |
1483 | spin_unlock(&f->lock); |
1484 | } | |
1485 | ||
1486 | static void __fanout_unlink(struct sock *sk, struct packet_sock *po) | |
1487 | { | |
1488 | struct packet_fanout *f = po->fanout; | |
1489 | int i; | |
1490 | ||
1491 | spin_lock(&f->lock); | |
1492 | for (i = 0; i < f->num_members; i++) { | |
1493 | if (f->arr[i] == sk) | |
1494 | break; | |
1495 | } | |
1496 | BUG_ON(i >= f->num_members); | |
1497 | f->arr[i] = f->arr[f->num_members - 1]; | |
1498 | f->num_members--; | |
2bd624b4 AS |
1499 | if (f->num_members == 0) |
1500 | __dev_remove_pack(&f->prot_hook); | |
dc99f600 DM |
1501 | spin_unlock(&f->lock); |
1502 | } | |
1503 | ||
d4dd8aee | 1504 | static bool match_fanout_group(struct packet_type *ptype, struct sock *sk) |
c0de08d0 | 1505 | { |
161642e2 ED |
1506 | if (sk->sk_family != PF_PACKET) |
1507 | return false; | |
c0de08d0 | 1508 | |
161642e2 | 1509 | return ptype->af_packet_priv == pkt_sk(sk)->fanout; |
c0de08d0 EL |
1510 | } |
1511 | ||
47dceb8e WB |
1512 | static void fanout_init_data(struct packet_fanout *f) |
1513 | { | |
1514 | switch (f->type) { | |
1515 | case PACKET_FANOUT_LB: | |
1516 | atomic_set(&f->rr_cur, 0); | |
1517 | break; | |
1518 | case PACKET_FANOUT_CBPF: | |
f2e52095 | 1519 | case PACKET_FANOUT_EBPF: |
47dceb8e WB |
1520 | RCU_INIT_POINTER(f->bpf_prog, NULL); |
1521 | break; | |
1522 | } | |
1523 | } | |
1524 | ||
1525 | static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new) | |
1526 | { | |
1527 | struct bpf_prog *old; | |
1528 | ||
1529 | spin_lock(&f->lock); | |
1530 | old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock)); | |
1531 | rcu_assign_pointer(f->bpf_prog, new); | |
1532 | spin_unlock(&f->lock); | |
1533 | ||
1534 | if (old) { | |
1535 | synchronize_net(); | |
1536 | bpf_prog_destroy(old); | |
1537 | } | |
1538 | } | |
1539 | ||
1540 | static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data, | |
1541 | unsigned int len) | |
1542 | { | |
1543 | struct bpf_prog *new; | |
1544 | struct sock_fprog fprog; | |
1545 | int ret; | |
1546 | ||
1547 | if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) | |
1548 | return -EPERM; | |
1549 | if (len != sizeof(fprog)) | |
1550 | return -EINVAL; | |
1551 | if (copy_from_user(&fprog, data, len)) | |
1552 | return -EFAULT; | |
1553 | ||
bab18991 | 1554 | ret = bpf_prog_create_from_user(&new, &fprog, NULL, false); |
47dceb8e WB |
1555 | if (ret) |
1556 | return ret; | |
1557 | ||
1558 | __fanout_set_data_bpf(po->fanout, new); | |
1559 | return 0; | |
1560 | } | |
1561 | ||
f2e52095 WB |
1562 | static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data, |
1563 | unsigned int len) | |
1564 | { | |
1565 | struct bpf_prog *new; | |
1566 | u32 fd; | |
1567 | ||
1568 | if (sock_flag(&po->sk, SOCK_FILTER_LOCKED)) | |
1569 | return -EPERM; | |
1570 | if (len != sizeof(fd)) | |
1571 | return -EINVAL; | |
1572 | if (copy_from_user(&fd, data, len)) | |
1573 | return -EFAULT; | |
1574 | ||
113214be | 1575 | new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); |
f2e52095 WB |
1576 | if (IS_ERR(new)) |
1577 | return PTR_ERR(new); | |
f2e52095 WB |
1578 | |
1579 | __fanout_set_data_bpf(po->fanout, new); | |
1580 | return 0; | |
1581 | } | |
1582 | ||
47dceb8e WB |
1583 | static int fanout_set_data(struct packet_sock *po, char __user *data, |
1584 | unsigned int len) | |
1585 | { | |
1586 | switch (po->fanout->type) { | |
1587 | case PACKET_FANOUT_CBPF: | |
1588 | return fanout_set_data_cbpf(po, data, len); | |
f2e52095 WB |
1589 | case PACKET_FANOUT_EBPF: |
1590 | return fanout_set_data_ebpf(po, data, len); | |
47dceb8e WB |
1591 | default: |
1592 | return -EINVAL; | |
07d53ae4 | 1593 | } |
47dceb8e WB |
1594 | } |
1595 | ||
1596 | static void fanout_release_data(struct packet_fanout *f) | |
1597 | { | |
1598 | switch (f->type) { | |
1599 | case PACKET_FANOUT_CBPF: | |
f2e52095 | 1600 | case PACKET_FANOUT_EBPF: |
47dceb8e | 1601 | __fanout_set_data_bpf(f, NULL); |
07d53ae4 | 1602 | } |
47dceb8e WB |
1603 | } |
1604 | ||
4a69a864 MM |
1605 | static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id) |
1606 | { | |
1607 | struct packet_fanout *f; | |
1608 | ||
1609 | list_for_each_entry(f, &fanout_list, list) { | |
1610 | if (f->id == candidate_id && | |
1611 | read_pnet(&f->net) == sock_net(sk)) { | |
1612 | return false; | |
1613 | } | |
1614 | } | |
1615 | return true; | |
1616 | } | |
1617 | ||
1618 | static bool fanout_find_new_id(struct sock *sk, u16 *new_id) | |
1619 | { | |
1620 | u16 id = fanout_next_id; | |
1621 | ||
1622 | do { | |
1623 | if (__fanout_id_is_free(sk, id)) { | |
1624 | *new_id = id; | |
1625 | fanout_next_id = id + 1; | |
1626 | return true; | |
1627 | } | |
1628 | ||
1629 | id++; | |
1630 | } while (id != fanout_next_id); | |
1631 | ||
1632 | return false; | |
1633 | } | |
1634 | ||
7736d33f | 1635 | static int fanout_add(struct sock *sk, u16 id, u16 type_flags) |
dc99f600 | 1636 | { |
d199fab6 | 1637 | struct packet_rollover *rollover = NULL; |
dc99f600 DM |
1638 | struct packet_sock *po = pkt_sk(sk); |
1639 | struct packet_fanout *f, *match; | |
7736d33f | 1640 | u8 type = type_flags & 0xff; |
77f65ebd | 1641 | u8 flags = type_flags >> 8; |
dc99f600 DM |
1642 | int err; |
1643 | ||
1644 | switch (type) { | |
77f65ebd WB |
1645 | case PACKET_FANOUT_ROLLOVER: |
1646 | if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) | |
1647 | return -EINVAL; | |
dc99f600 DM |
1648 | case PACKET_FANOUT_HASH: |
1649 | case PACKET_FANOUT_LB: | |
95ec3eb4 | 1650 | case PACKET_FANOUT_CPU: |
5df0ddfb | 1651 | case PACKET_FANOUT_RND: |
2d36097d | 1652 | case PACKET_FANOUT_QM: |
47dceb8e | 1653 | case PACKET_FANOUT_CBPF: |
f2e52095 | 1654 | case PACKET_FANOUT_EBPF: |
dc99f600 DM |
1655 | break; |
1656 | default: | |
1657 | return -EINVAL; | |
1658 | } | |
1659 | ||
d199fab6 ED |
1660 | mutex_lock(&fanout_mutex); |
1661 | ||
d199fab6 | 1662 | err = -EALREADY; |
dc99f600 | 1663 | if (po->fanout) |
d199fab6 | 1664 | goto out; |
dc99f600 | 1665 | |
4633c9e0 WB |
1666 | if (type == PACKET_FANOUT_ROLLOVER || |
1667 | (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) { | |
d199fab6 ED |
1668 | err = -ENOMEM; |
1669 | rollover = kzalloc(sizeof(*rollover), GFP_KERNEL); | |
1670 | if (!rollover) | |
1671 | goto out; | |
1672 | atomic_long_set(&rollover->num, 0); | |
1673 | atomic_long_set(&rollover->num_huge, 0); | |
1674 | atomic_long_set(&rollover->num_failed, 0); | |
0648ab70 WB |
1675 | } |
1676 | ||
4a69a864 MM |
1677 | if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) { |
1678 | if (id != 0) { | |
1679 | err = -EINVAL; | |
1680 | goto out; | |
1681 | } | |
1682 | if (!fanout_find_new_id(sk, &id)) { | |
1683 | err = -ENOMEM; | |
1684 | goto out; | |
1685 | } | |
1686 | /* ephemeral flag for the first socket in the group: drop it */ | |
1687 | flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8); | |
1688 | } | |
1689 | ||
dc99f600 DM |
1690 | match = NULL; |
1691 | list_for_each_entry(f, &fanout_list, list) { | |
1692 | if (f->id == id && | |
1693 | read_pnet(&f->net) == sock_net(sk)) { | |
1694 | match = f; | |
1695 | break; | |
1696 | } | |
1697 | } | |
afe62c68 | 1698 | err = -EINVAL; |
77f65ebd | 1699 | if (match && match->flags != flags) |
afe62c68 | 1700 | goto out; |
dc99f600 | 1701 | if (!match) { |
afe62c68 | 1702 | err = -ENOMEM; |
dc99f600 | 1703 | match = kzalloc(sizeof(*match), GFP_KERNEL); |
afe62c68 ED |
1704 | if (!match) |
1705 | goto out; | |
1706 | write_pnet(&match->net, sock_net(sk)); | |
1707 | match->id = id; | |
1708 | match->type = type; | |
77f65ebd | 1709 | match->flags = flags; |
afe62c68 ED |
1710 | INIT_LIST_HEAD(&match->list); |
1711 | spin_lock_init(&match->lock); | |
fb5c2c17 | 1712 | refcount_set(&match->sk_ref, 0); |
47dceb8e | 1713 | fanout_init_data(match); |
afe62c68 ED |
1714 | match->prot_hook.type = po->prot_hook.type; |
1715 | match->prot_hook.dev = po->prot_hook.dev; | |
1716 | match->prot_hook.func = packet_rcv_fanout; | |
1717 | match->prot_hook.af_packet_priv = match; | |
c0de08d0 | 1718 | match->prot_hook.id_match = match_fanout_group; |
afe62c68 | 1719 | list_add(&match->list, &fanout_list); |
dc99f600 | 1720 | } |
afe62c68 | 1721 | err = -EINVAL; |
008ba2a1 WB |
1722 | |
1723 | spin_lock(&po->bind_lock); | |
1724 | if (po->running && | |
1725 | match->type == type && | |
afe62c68 ED |
1726 | match->prot_hook.type == po->prot_hook.type && |
1727 | match->prot_hook.dev == po->prot_hook.dev) { | |
1728 | err = -ENOSPC; | |
fb5c2c17 | 1729 | if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) { |
afe62c68 ED |
1730 | __dev_remove_pack(&po->prot_hook); |
1731 | po->fanout = match; | |
57f015f5 MM |
1732 | po->rollover = rollover; |
1733 | rollover = NULL; | |
fb5c2c17 | 1734 | refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1); |
afe62c68 ED |
1735 | __fanout_link(sk, po); |
1736 | err = 0; | |
dc99f600 DM |
1737 | } |
1738 | } | |
008ba2a1 WB |
1739 | spin_unlock(&po->bind_lock); |
1740 | ||
1741 | if (err && !refcount_read(&match->sk_ref)) { | |
1742 | list_del(&match->list); | |
1743 | kfree(match); | |
1744 | } | |
1745 | ||
afe62c68 | 1746 | out: |
57f015f5 | 1747 | kfree(rollover); |
d199fab6 | 1748 | mutex_unlock(&fanout_mutex); |
dc99f600 DM |
1749 | return err; |
1750 | } | |
1751 | ||
2bd624b4 AS |
1752 | /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes |
1753 | * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout. | |
1754 | * It is the responsibility of the caller to call fanout_release_data() and | |
1755 | * free the returned packet_fanout (after synchronize_net()) | |
1756 | */ | |
1757 | static struct packet_fanout *fanout_release(struct sock *sk) | |
dc99f600 DM |
1758 | { |
1759 | struct packet_sock *po = pkt_sk(sk); | |
1760 | struct packet_fanout *f; | |
1761 | ||
fff3321d | 1762 | mutex_lock(&fanout_mutex); |
d199fab6 ED |
1763 | f = po->fanout; |
1764 | if (f) { | |
1765 | po->fanout = NULL; | |
1766 | ||
fb5c2c17 | 1767 | if (refcount_dec_and_test(&f->sk_ref)) |
d199fab6 | 1768 | list_del(&f->list); |
2bd624b4 AS |
1769 | else |
1770 | f = NULL; | |
dc99f600 DM |
1771 | } |
1772 | mutex_unlock(&fanout_mutex); | |
2bd624b4 AS |
1773 | |
1774 | return f; | |
dc99f600 | 1775 | } |
1da177e4 | 1776 | |
3c70c132 DB |
1777 | static bool packet_extra_vlan_len_allowed(const struct net_device *dev, |
1778 | struct sk_buff *skb) | |
1779 | { | |
1780 | /* Earlier code assumed this would be a VLAN pkt, double-check | |
1781 | * this now that we have the actual packet in hand. We can only | |
1782 | * do this check on Ethernet devices. | |
1783 | */ | |
1784 | if (unlikely(dev->type != ARPHRD_ETHER)) | |
1785 | return false; | |
1786 | ||
1787 | skb_reset_mac_header(skb); | |
1788 | return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q)); | |
1789 | } | |
1790 | ||
90ddc4f0 | 1791 | static const struct proto_ops packet_ops; |
1da177e4 | 1792 | |
90ddc4f0 | 1793 | static const struct proto_ops packet_ops_spkt; |
1da177e4 | 1794 | |
40d4e3df ED |
1795 | static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, |
1796 | struct packet_type *pt, struct net_device *orig_dev) | |
1da177e4 LT |
1797 | { |
1798 | struct sock *sk; | |
1799 | struct sockaddr_pkt *spkt; | |
1800 | ||
1801 | /* | |
1802 | * When we registered the protocol we saved the socket in the data | |
1803 | * field for just this event. | |
1804 | */ | |
1805 | ||
1806 | sk = pt->af_packet_priv; | |
1ce4f28b | 1807 | |
1da177e4 LT |
1808 | /* |
1809 | * Yank back the headers [hope the device set this | |
1810 | * right or kerboom...] | |
1811 | * | |
1812 | * Incoming packets have ll header pulled, | |
1813 | * push it back. | |
1814 | * | |
98e399f8 | 1815 | * For outgoing ones skb->data == skb_mac_header(skb) |
1da177e4 LT |
1816 | * so that this procedure is noop. |
1817 | */ | |
1818 | ||
1819 | if (skb->pkt_type == PACKET_LOOPBACK) | |
1820 | goto out; | |
1821 | ||
09ad9bc7 | 1822 | if (!net_eq(dev_net(dev), sock_net(sk))) |
d12d01d6 DL |
1823 | goto out; |
1824 | ||
40d4e3df ED |
1825 | skb = skb_share_check(skb, GFP_ATOMIC); |
1826 | if (skb == NULL) | |
1da177e4 LT |
1827 | goto oom; |
1828 | ||
1829 | /* drop any routing info */ | |
adf30907 | 1830 | skb_dst_drop(skb); |
1da177e4 | 1831 | |
84531c24 | 1832 | /* drop conntrack reference */ |
895b5c9f | 1833 | nf_reset_ct(skb); |
84531c24 | 1834 | |
ffbc6111 | 1835 | spkt = &PACKET_SKB_CB(skb)->sa.pkt; |
1da177e4 | 1836 | |
98e399f8 | 1837 | skb_push(skb, skb->data - skb_mac_header(skb)); |
1da177e4 LT |
1838 | |
1839 | /* | |
1840 | * The SOCK_PACKET socket receives _all_ frames. | |
1841 | */ | |
1842 | ||
1843 | spkt->spkt_family = dev->type; | |
1844 | strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device)); | |
1845 | spkt->spkt_protocol = skb->protocol; | |
1846 | ||
1847 | /* | |
1848 | * Charge the memory to the socket. This is done specifically | |
1849 | * to prevent sockets using all the memory up. | |
1850 | */ | |
1851 | ||
40d4e3df | 1852 | if (sock_queue_rcv_skb(sk, skb) == 0) |
1da177e4 LT |
1853 | return 0; |
1854 | ||
1855 | out: | |
1856 | kfree_skb(skb); | |
1857 | oom: | |
1858 | return 0; | |
1859 | } | |
1860 | ||
75c65772 MM |
1861 | static void packet_parse_headers(struct sk_buff *skb, struct socket *sock) |
1862 | { | |
18bed891 YK |
1863 | if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) && |
1864 | sock->type == SOCK_RAW) { | |
75c65772 MM |
1865 | skb_reset_mac_header(skb); |
1866 | skb->protocol = dev_parse_header_protocol(skb); | |
1867 | } | |
1868 | ||
1869 | skb_probe_transport_header(skb); | |
1870 | } | |
1da177e4 LT |
1871 | |
1872 | /* | |
1873 | * Output a raw packet to a device layer. This bypasses all the other | |
1874 | * protocol layers and you must therefore supply it with a complete frame | |
1875 | */ | |
1ce4f28b | 1876 | |
1b784140 YX |
1877 | static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg, |
1878 | size_t len) | |
1da177e4 LT |
1879 | { |
1880 | struct sock *sk = sock->sk; | |
342dfc30 | 1881 | DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name); |
1a35ca80 | 1882 | struct sk_buff *skb = NULL; |
1da177e4 | 1883 | struct net_device *dev; |
c14ac945 | 1884 | struct sockcm_cookie sockc; |
40d4e3df | 1885 | __be16 proto = 0; |
1da177e4 | 1886 | int err; |
3bdc0eba | 1887 | int extra_len = 0; |
1ce4f28b | 1888 | |
1da177e4 | 1889 | /* |
1ce4f28b | 1890 | * Get and verify the address. |
1da177e4 LT |
1891 | */ |
1892 | ||
40d4e3df | 1893 | if (saddr) { |
1da177e4 | 1894 | if (msg->msg_namelen < sizeof(struct sockaddr)) |
40d4e3df ED |
1895 | return -EINVAL; |
1896 | if (msg->msg_namelen == sizeof(struct sockaddr_pkt)) | |
1897 | proto = saddr->spkt_protocol; | |
1898 | } else | |
1899 | return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */ | |
1da177e4 LT |
1900 | |
1901 | /* | |
1ce4f28b | 1902 | * Find the device first to size check it |
1da177e4 LT |
1903 | */ |
1904 | ||
de74e92a | 1905 | saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0; |
1a35ca80 | 1906 | retry: |
654d1f8a ED |
1907 | rcu_read_lock(); |
1908 | dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device); | |
1da177e4 LT |
1909 | err = -ENODEV; |
1910 | if (dev == NULL) | |
1911 | goto out_unlock; | |
1ce4f28b | 1912 | |
d5e76b0a DM |
1913 | err = -ENETDOWN; |
1914 | if (!(dev->flags & IFF_UP)) | |
1915 | goto out_unlock; | |
1916 | ||
1da177e4 | 1917 | /* |
40d4e3df ED |
1918 | * You may not queue a frame bigger than the mtu. This is the lowest level |
1919 | * raw protocol and you must do your own fragmentation at this level. | |
1da177e4 | 1920 | */ |
1ce4f28b | 1921 | |
3bdc0eba BG |
1922 | if (unlikely(sock_flag(sk, SOCK_NOFCS))) { |
1923 | if (!netif_supports_nofcs(dev)) { | |
1924 | err = -EPROTONOSUPPORT; | |
1925 | goto out_unlock; | |
1926 | } | |
1927 | extra_len = 4; /* We're doing our own CRC */ | |
1928 | } | |
1929 | ||
1da177e4 | 1930 | err = -EMSGSIZE; |
3bdc0eba | 1931 | if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len) |
1da177e4 LT |
1932 | goto out_unlock; |
1933 | ||
1a35ca80 ED |
1934 | if (!skb) { |
1935 | size_t reserved = LL_RESERVED_SPACE(dev); | |
4ce40912 | 1936 | int tlen = dev->needed_tailroom; |
1a35ca80 ED |
1937 | unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0; |
1938 | ||
1939 | rcu_read_unlock(); | |
4ce40912 | 1940 | skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL); |
1a35ca80 ED |
1941 | if (skb == NULL) |
1942 | return -ENOBUFS; | |
1943 | /* FIXME: Save some space for broken drivers that write a hard | |
1944 | * header at transmission time by themselves. PPP is the notable | |
1945 | * one here. This should really be fixed at the driver level. | |
1946 | */ | |
1947 | skb_reserve(skb, reserved); | |
1948 | skb_reset_network_header(skb); | |
1949 | ||
1950 | /* Try to align data part correctly */ | |
1951 | if (hhlen) { | |
1952 | skb->data -= hhlen; | |
1953 | skb->tail -= hhlen; | |
1954 | if (len < hhlen) | |
1955 | skb_reset_network_header(skb); | |
1956 | } | |
6ce8e9ce | 1957 | err = memcpy_from_msg(skb_put(skb, len), msg, len); |
1a35ca80 ED |
1958 | if (err) |
1959 | goto out_free; | |
1960 | goto retry; | |
1da177e4 LT |
1961 | } |
1962 | ||
9ed988cd WB |
1963 | if (!dev_validate_header(dev, skb->data, len)) { |
1964 | err = -EINVAL; | |
1965 | goto out_unlock; | |
1966 | } | |
3c70c132 DB |
1967 | if (len > (dev->mtu + dev->hard_header_len + extra_len) && |
1968 | !packet_extra_vlan_len_allowed(dev, skb)) { | |
1969 | err = -EMSGSIZE; | |
1970 | goto out_unlock; | |
57f89bfa | 1971 | } |
1a35ca80 | 1972 | |
657a0667 | 1973 | sockcm_init(&sockc, sk); |
c14ac945 SHY |
1974 | if (msg->msg_controllen) { |
1975 | err = sock_cmsg_send(sk, msg, &sockc); | |
f8e7718c | 1976 | if (unlikely(err)) |
c14ac945 | 1977 | goto out_unlock; |
c14ac945 SHY |
1978 | } |
1979 | ||
1da177e4 LT |
1980 | skb->protocol = proto; |
1981 | skb->dev = dev; | |
1982 | skb->priority = sk->sk_priority; | |
2d37a186 | 1983 | skb->mark = sk->sk_mark; |
3d0ba8c0 | 1984 | skb->tstamp = sockc.transmit_time; |
bf84a010 | 1985 | |
8f932f76 | 1986 | skb_setup_tx_timestamp(skb, sockc.tsflags); |
1da177e4 | 1987 | |
3bdc0eba BG |
1988 | if (unlikely(extra_len == 4)) |
1989 | skb->no_fcs = 1; | |
1990 | ||
75c65772 | 1991 | packet_parse_headers(skb, sock); |
c1aad275 | 1992 | |
1da177e4 | 1993 | dev_queue_xmit(skb); |
654d1f8a | 1994 | rcu_read_unlock(); |
40d4e3df | 1995 | return len; |
1da177e4 | 1996 | |
1da177e4 | 1997 | out_unlock: |
654d1f8a | 1998 | rcu_read_unlock(); |
1a35ca80 ED |
1999 | out_free: |
2000 | kfree_skb(skb); | |
1da177e4 LT |
2001 | return err; |
2002 | } | |
1da177e4 | 2003 | |
ff936a04 AS |
2004 | static unsigned int run_filter(struct sk_buff *skb, |
2005 | const struct sock *sk, | |
2006 | unsigned int res) | |
1da177e4 LT |
2007 | { |
2008 | struct sk_filter *filter; | |
fda9ef5d | 2009 | |
80f8f102 ED |
2010 | rcu_read_lock(); |
2011 | filter = rcu_dereference(sk->sk_filter); | |
dbcb5855 | 2012 | if (filter != NULL) |
ff936a04 | 2013 | res = bpf_prog_run_clear_cb(filter->prog, skb); |
80f8f102 | 2014 | rcu_read_unlock(); |
1da177e4 | 2015 | |
dbcb5855 | 2016 | return res; |
1da177e4 LT |
2017 | } |
2018 | ||
16cc1400 WB |
2019 | static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb, |
2020 | size_t *len) | |
2021 | { | |
2022 | struct virtio_net_hdr vnet_hdr; | |
2023 | ||
2024 | if (*len < sizeof(vnet_hdr)) | |
2025 | return -EINVAL; | |
2026 | *len -= sizeof(vnet_hdr); | |
2027 | ||
fd3a8862 | 2028 | if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0)) |
16cc1400 WB |
2029 | return -EINVAL; |
2030 | ||
2031 | return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr)); | |
2032 | } | |
2033 | ||
1da177e4 | 2034 | /* |
62ab0812 ED |
2035 | * This function makes lazy skb cloning in hope that most of packets |
2036 | * are discarded by BPF. | |
2037 | * | |
2038 | * Note tricky part: we DO mangle shared skb! skb->data, skb->len | |
2039 | * and skb->cb are mangled. It works because (and until) packets | |
2040 | * falling here are owned by current CPU. Output packets are cloned | |
2041 | * by dev_queue_xmit_nit(), input packets are processed by net_bh | |
2042 | * sequencially, so that if we return skb to original state on exit, | |
2043 | * we will not harm anyone. | |
1da177e4 LT |
2044 | */ |
2045 | ||
40d4e3df ED |
2046 | static int packet_rcv(struct sk_buff *skb, struct net_device *dev, |
2047 | struct packet_type *pt, struct net_device *orig_dev) | |
1da177e4 LT |
2048 | { |
2049 | struct sock *sk; | |
2050 | struct sockaddr_ll *sll; | |
2051 | struct packet_sock *po; | |
40d4e3df | 2052 | u8 *skb_head = skb->data; |
1da177e4 | 2053 | int skb_len = skb->len; |
dbcb5855 | 2054 | unsigned int snaplen, res; |
da37845f | 2055 | bool is_drop_n_account = false; |
1da177e4 LT |
2056 | |
2057 | if (skb->pkt_type == PACKET_LOOPBACK) | |
2058 | goto drop; | |
2059 | ||
2060 | sk = pt->af_packet_priv; | |
2061 | po = pkt_sk(sk); | |
2062 | ||
09ad9bc7 | 2063 | if (!net_eq(dev_net(dev), sock_net(sk))) |
d12d01d6 DL |
2064 | goto drop; |
2065 | ||
1da177e4 LT |
2066 | skb->dev = dev; |
2067 | ||
3b04ddde | 2068 | if (dev->header_ops) { |
1da177e4 | 2069 | /* The device has an explicit notion of ll header, |
62ab0812 ED |
2070 | * exported to higher levels. |
2071 | * | |
2072 | * Otherwise, the device hides details of its frame | |
2073 | * structure, so that corresponding packet head is | |
2074 | * never delivered to user. | |
1da177e4 LT |
2075 | */ |
2076 | if (sk->sk_type != SOCK_DGRAM) | |
98e399f8 | 2077 | skb_push(skb, skb->data - skb_mac_header(skb)); |
1da177e4 LT |
2078 | else if (skb->pkt_type == PACKET_OUTGOING) { |
2079 | /* Special case: outgoing packets have ll header at head */ | |
bbe735e4 | 2080 | skb_pull(skb, skb_network_offset(skb)); |
1da177e4 LT |
2081 | } |
2082 | } | |
2083 | ||
2084 | snaplen = skb->len; | |
2085 | ||
dbcb5855 DM |
2086 | res = run_filter(skb, sk, snaplen); |
2087 | if (!res) | |
fda9ef5d | 2088 | goto drop_n_restore; |
dbcb5855 DM |
2089 | if (snaplen > res) |
2090 | snaplen = res; | |
1da177e4 | 2091 | |
0fd7bac6 | 2092 | if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) |
1da177e4 LT |
2093 | goto drop_n_acct; |
2094 | ||
2095 | if (skb_shared(skb)) { | |
2096 | struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); | |
2097 | if (nskb == NULL) | |
2098 | goto drop_n_acct; | |
2099 | ||
2100 | if (skb_head != skb->data) { | |
2101 | skb->data = skb_head; | |
2102 | skb->len = skb_len; | |
2103 | } | |
abc4e4fa | 2104 | consume_skb(skb); |
1da177e4 LT |
2105 | skb = nskb; |
2106 | } | |
2107 | ||
b4772ef8 | 2108 | sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8); |
ffbc6111 HX |
2109 | |
2110 | sll = &PACKET_SKB_CB(skb)->sa.ll; | |
1da177e4 | 2111 | sll->sll_hatype = dev->type; |
1da177e4 | 2112 | sll->sll_pkttype = skb->pkt_type; |
8032b464 | 2113 | if (unlikely(po->origdev)) |
80feaacb PWJ |
2114 | sll->sll_ifindex = orig_dev->ifindex; |
2115 | else | |
2116 | sll->sll_ifindex = dev->ifindex; | |
1da177e4 | 2117 | |
b95cce35 | 2118 | sll->sll_halen = dev_parse_header(skb, sll->sll_addr); |
1da177e4 | 2119 | |
2472d761 EB |
2120 | /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg(). |
2121 | * Use their space for storing the original skb length. | |
2122 | */ | |
2123 | PACKET_SKB_CB(skb)->sa.origlen = skb->len; | |
8dc41944 | 2124 | |
1da177e4 LT |
2125 | if (pskb_trim(skb, snaplen)) |
2126 | goto drop_n_acct; | |
2127 | ||
2128 | skb_set_owner_r(skb, sk); | |
2129 | skb->dev = NULL; | |
adf30907 | 2130 | skb_dst_drop(skb); |
1da177e4 | 2131 | |
84531c24 | 2132 | /* drop conntrack reference */ |
895b5c9f | 2133 | nf_reset_ct(skb); |
84531c24 | 2134 | |
1da177e4 | 2135 | spin_lock(&sk->sk_receive_queue.lock); |
ee80fbf3 | 2136 | po->stats.stats1.tp_packets++; |
3bc3b96f | 2137 | sock_skb_set_dropcount(sk, skb); |
1da177e4 LT |
2138 | __skb_queue_tail(&sk->sk_receive_queue, skb); |
2139 | spin_unlock(&sk->sk_receive_queue.lock); | |
676d2369 | 2140 | sk->sk_data_ready(sk); |
1da177e4 LT |
2141 | return 0; |
2142 | ||
2143 | drop_n_acct: | |
da37845f | 2144 | is_drop_n_account = true; |
8e8e2951 | 2145 | atomic_inc(&po->tp_drops); |
7091fbd8 | 2146 | atomic_inc(&sk->sk_drops); |
1da177e4 LT |
2147 | |
2148 | drop_n_restore: | |
2149 | if (skb_head != skb->data && skb_shared(skb)) { | |
2150 | skb->data = skb_head; | |
2151 | skb->len = skb_len; | |
2152 | } | |
2153 | drop: | |
da37845f WJ |
2154 | if (!is_drop_n_account) |
2155 | consume_skb(skb); | |
2156 | else | |
2157 | kfree_skb(skb); | |
1da177e4 LT |
2158 | return 0; |
2159 | } | |
2160 | ||
40d4e3df ED |
2161 | static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, |
2162 | struct packet_type *pt, struct net_device *orig_dev) | |
1da177e4 LT |
2163 | { |
2164 | struct sock *sk; | |
2165 | struct packet_sock *po; | |
2166 | struct sockaddr_ll *sll; | |
184f489e | 2167 | union tpacket_uhdr h; |
40d4e3df | 2168 | u8 *skb_head = skb->data; |
1da177e4 | 2169 | int skb_len = skb->len; |
dbcb5855 | 2170 | unsigned int snaplen, res; |
f6fb8f10 | 2171 | unsigned long status = TP_STATUS_USER; |
7b52b013 OC |
2172 | unsigned short macoff, hdrlen; |
2173 | unsigned int netoff; | |
1da177e4 | 2174 | struct sk_buff *copy_skb = NULL; |
bbd6ef87 | 2175 | struct timespec ts; |
b9c32fb2 | 2176 | __u32 ts_status; |
da37845f | 2177 | bool is_drop_n_account = false; |
f897d759 | 2178 | unsigned int slot_id = 0; |
edbd58be | 2179 | bool do_vnet = false; |
1da177e4 | 2180 | |
51846355 AW |
2181 | /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT. |
2182 | * We may add members to them until current aligned size without forcing | |
2183 | * userspace to call getsockopt(..., PACKET_HDRLEN, ...). | |
2184 | */ | |
2185 | BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32); | |
2186 | BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48); | |
2187 | ||
1da177e4 LT |
2188 | if (skb->pkt_type == PACKET_LOOPBACK) |
2189 | goto drop; | |
2190 | ||
2191 | sk = pt->af_packet_priv; | |
2192 | po = pkt_sk(sk); | |
2193 | ||
09ad9bc7 | 2194 | if (!net_eq(dev_net(dev), sock_net(sk))) |
d12d01d6 DL |
2195 | goto drop; |
2196 | ||
3b04ddde | 2197 | if (dev->header_ops) { |
1da177e4 | 2198 | if (sk->sk_type != SOCK_DGRAM) |
98e399f8 | 2199 | skb_push(skb, skb->data - skb_mac_header(skb)); |
1da177e4 LT |
2200 | else if (skb->pkt_type == PACKET_OUTGOING) { |
2201 | /* Special case: outgoing packets have ll header at head */ | |
bbe735e4 | 2202 | skb_pull(skb, skb_network_offset(skb)); |
1da177e4 LT |
2203 | } |
2204 | } | |
2205 | ||
2206 | snaplen = skb->len; | |
2207 | ||
dbcb5855 DM |
2208 | res = run_filter(skb, sk, snaplen); |
2209 | if (!res) | |
fda9ef5d | 2210 | goto drop_n_restore; |
68c2e5de | 2211 | |
2c51c627 ED |
2212 | /* If we are flooded, just give up */ |
2213 | if (__packet_rcv_has_room(po, skb) == ROOM_NONE) { | |
2214 | atomic_inc(&po->tp_drops); | |
2215 | goto drop_n_restore; | |
2216 | } | |
2217 | ||
68c2e5de AD |
2218 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
2219 | status |= TP_STATUS_CSUMNOTREADY; | |
682f048b AD |
2220 | else if (skb->pkt_type != PACKET_OUTGOING && |
2221 | (skb->ip_summed == CHECKSUM_COMPLETE || | |
2222 | skb_csum_unnecessary(skb))) | |
2223 | status |= TP_STATUS_CSUM_VALID; | |
68c2e5de | 2224 | |
dbcb5855 DM |
2225 | if (snaplen > res) |
2226 | snaplen = res; | |
1da177e4 LT |
2227 | |
2228 | if (sk->sk_type == SOCK_DGRAM) { | |
8913336a PM |
2229 | macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 + |
2230 | po->tp_reserve; | |
1da177e4 | 2231 | } else { |
95c96174 | 2232 | unsigned int maclen = skb_network_offset(skb); |
bbd6ef87 | 2233 | netoff = TPACKET_ALIGN(po->tp_hdrlen + |
8913336a | 2234 | (maclen < 16 ? 16 : maclen)) + |
58d19b19 | 2235 | po->tp_reserve; |
edbd58be | 2236 | if (po->has_vnet_hdr) { |
58d19b19 | 2237 | netoff += sizeof(struct virtio_net_hdr); |
edbd58be BP |
2238 | do_vnet = true; |
2239 | } | |
1da177e4 LT |
2240 | macoff = netoff - maclen; |
2241 | } | |
7b52b013 OC |
2242 | if (netoff > USHRT_MAX) { |
2243 | atomic_inc(&po->tp_drops); | |
2244 | goto drop_n_restore; | |
2245 | } | |
f6fb8f10 | 2246 | if (po->tp_version <= TPACKET_V2) { |
2247 | if (macoff + snaplen > po->rx_ring.frame_size) { | |
2248 | if (po->copy_thresh && | |
0fd7bac6 | 2249 | atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { |
f6fb8f10 | 2250 | if (skb_shared(skb)) { |
2251 | copy_skb = skb_clone(skb, GFP_ATOMIC); | |
2252 | } else { | |
2253 | copy_skb = skb_get(skb); | |
2254 | skb_head = skb->data; | |
2255 | } | |
2256 | if (copy_skb) | |
2257 | skb_set_owner_r(copy_skb, sk); | |
1da177e4 | 2258 | } |
f6fb8f10 | 2259 | snaplen = po->rx_ring.frame_size - macoff; |
edbd58be | 2260 | if ((int)snaplen < 0) { |
f6fb8f10 | 2261 | snaplen = 0; |
edbd58be BP |
2262 | do_vnet = false; |
2263 | } | |
1da177e4 | 2264 | } |
dc808110 ED |
2265 | } else if (unlikely(macoff + snaplen > |
2266 | GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) { | |
2267 | u32 nval; | |
2268 | ||
2269 | nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff; | |
2270 | pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n", | |
2271 | snaplen, nval, macoff); | |
2272 | snaplen = nval; | |
2273 | if (unlikely((int)snaplen < 0)) { | |
2274 | snaplen = 0; | |
2275 | macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len; | |
edbd58be | 2276 | do_vnet = false; |
dc808110 | 2277 | } |
1da177e4 | 2278 | } |
1da177e4 | 2279 | spin_lock(&sk->sk_receive_queue.lock); |
f6fb8f10 | 2280 | h.raw = packet_current_rx_frame(po, skb, |
2281 | TP_STATUS_KERNEL, (macoff+snaplen)); | |
bbd6ef87 | 2282 | if (!h.raw) |
58d19b19 | 2283 | goto drop_n_account; |
41442444 | 2284 | |
f897d759 WB |
2285 | if (po->tp_version <= TPACKET_V2) { |
2286 | slot_id = po->rx_ring.head; | |
2287 | if (test_bit(slot_id, po->rx_ring.rx_owner_map)) | |
2288 | goto drop_n_account; | |
2289 | __set_bit(slot_id, po->rx_ring.rx_owner_map); | |
2290 | } | |
2291 | ||
41442444 WB |
2292 | if (do_vnet && |
2293 | virtio_net_hdr_from_skb(skb, h.raw + macoff - | |
2294 | sizeof(struct virtio_net_hdr), | |
014dc8b9 JO |
2295 | vio_le(), true, 0)) { |
2296 | if (po->tp_version == TPACKET_V3) | |
2297 | prb_clear_blk_fill_status(&po->rx_ring); | |
41442444 | 2298 | goto drop_n_account; |
014dc8b9 | 2299 | } |
41442444 | 2300 | |
f6fb8f10 | 2301 | if (po->tp_version <= TPACKET_V2) { |
2302 | packet_increment_rx_head(po, &po->rx_ring); | |
2303 | /* | |
2304 | * LOSING will be reported till you read the stats, | |
2305 | * because it's COR - Clear On Read. | |
2306 | * Anyways, moving it for V1/V2 only as V3 doesn't need this | |
2307 | * at packet level. | |
2308 | */ | |
8e8e2951 | 2309 | if (atomic_read(&po->tp_drops)) |
f6fb8f10 | 2310 | status |= TP_STATUS_LOSING; |
2311 | } | |
945d015e | 2312 | |
ee80fbf3 | 2313 | po->stats.stats1.tp_packets++; |
1da177e4 LT |
2314 | if (copy_skb) { |
2315 | status |= TP_STATUS_COPY; | |
2316 | __skb_queue_tail(&sk->sk_receive_queue, copy_skb); | |
2317 | } | |
1da177e4 LT |
2318 | spin_unlock(&sk->sk_receive_queue.lock); |
2319 | ||
bbd6ef87 | 2320 | skb_copy_bits(skb, 0, h.raw + macoff, snaplen); |
b9c32fb2 DB |
2321 | |
2322 | if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp))) | |
7a51384c | 2323 | getnstimeofday(&ts); |
1da177e4 | 2324 | |
b9c32fb2 DB |
2325 | status |= ts_status; |
2326 | ||
bbd6ef87 PM |
2327 | switch (po->tp_version) { |
2328 | case TPACKET_V1: | |
2329 | h.h1->tp_len = skb->len; | |
2330 | h.h1->tp_snaplen = snaplen; | |
2331 | h.h1->tp_mac = macoff; | |
2332 | h.h1->tp_net = netoff; | |
4b457bdf DB |
2333 | h.h1->tp_sec = ts.tv_sec; |
2334 | h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC; | |
bbd6ef87 PM |
2335 | hdrlen = sizeof(*h.h1); |
2336 | break; | |
2337 | case TPACKET_V2: | |
2338 | h.h2->tp_len = skb->len; | |
2339 | h.h2->tp_snaplen = snaplen; | |
2340 | h.h2->tp_mac = macoff; | |
2341 | h.h2->tp_net = netoff; | |
bbd6ef87 PM |
2342 | h.h2->tp_sec = ts.tv_sec; |
2343 | h.h2->tp_nsec = ts.tv_nsec; | |
df8a39de JP |
2344 | if (skb_vlan_tag_present(skb)) { |
2345 | h.h2->tp_vlan_tci = skb_vlan_tag_get(skb); | |
a0cdfcf3 AW |
2346 | h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto); |
2347 | status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; | |
a3bcc23e BG |
2348 | } else { |
2349 | h.h2->tp_vlan_tci = 0; | |
a0cdfcf3 | 2350 | h.h2->tp_vlan_tpid = 0; |
a3bcc23e | 2351 | } |
e4d26f4b | 2352 | memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding)); |
bbd6ef87 PM |
2353 | hdrlen = sizeof(*h.h2); |
2354 | break; | |
f6fb8f10 | 2355 | case TPACKET_V3: |
2356 | /* tp_nxt_offset,vlan are already populated above. | |
2357 | * So DONT clear those fields here | |
2358 | */ | |
2359 | h.h3->tp_status |= status; | |
2360 | h.h3->tp_len = skb->len; | |
2361 | h.h3->tp_snaplen = snaplen; | |
2362 | h.h3->tp_mac = macoff; | |
2363 | h.h3->tp_net = netoff; | |
f6fb8f10 | 2364 | h.h3->tp_sec = ts.tv_sec; |
2365 | h.h3->tp_nsec = ts.tv_nsec; | |
e4d26f4b | 2366 | memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding)); |
f6fb8f10 | 2367 | hdrlen = sizeof(*h.h3); |
2368 | break; | |
bbd6ef87 PM |
2369 | default: |
2370 | BUG(); | |
2371 | } | |
1da177e4 | 2372 | |
bbd6ef87 | 2373 | sll = h.raw + TPACKET_ALIGN(hdrlen); |
b95cce35 | 2374 | sll->sll_halen = dev_parse_header(skb, sll->sll_addr); |
1da177e4 LT |
2375 | sll->sll_family = AF_PACKET; |
2376 | sll->sll_hatype = dev->type; | |
2377 | sll->sll_protocol = skb->protocol; | |
2378 | sll->sll_pkttype = skb->pkt_type; | |
8032b464 | 2379 | if (unlikely(po->origdev)) |
80feaacb PWJ |
2380 | sll->sll_ifindex = orig_dev->ifindex; |
2381 | else | |
2382 | sll->sll_ifindex = dev->ifindex; | |
1da177e4 | 2383 | |
e16aa207 | 2384 | smp_mb(); |
f0d4eb29 | 2385 | |
f6dafa95 | 2386 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1 |
f0d4eb29 | 2387 | if (po->tp_version <= TPACKET_V2) { |
0af55bb5 CG |
2388 | u8 *start, *end; |
2389 | ||
f0d4eb29 DB |
2390 | end = (u8 *) PAGE_ALIGN((unsigned long) h.raw + |
2391 | macoff + snaplen); | |
2392 | ||
2393 | for (start = h.raw; start < end; start += PAGE_SIZE) | |
2394 | flush_dcache_page(pgv_to_page(start)); | |
1da177e4 | 2395 | } |
f0d4eb29 | 2396 | smp_wmb(); |
f6dafa95 | 2397 | #endif |
f0d4eb29 | 2398 | |
da413eec | 2399 | if (po->tp_version <= TPACKET_V2) { |
f897d759 | 2400 | spin_lock(&sk->sk_receive_queue.lock); |
f6fb8f10 | 2401 | __packet_set_status(po, h.raw, status); |
f897d759 WB |
2402 | __clear_bit(slot_id, po->rx_ring.rx_owner_map); |
2403 | spin_unlock(&sk->sk_receive_queue.lock); | |
da413eec | 2404 | sk->sk_data_ready(sk); |
014dc8b9 | 2405 | } else if (po->tp_version == TPACKET_V3) { |
f6fb8f10 | 2406 | prb_clear_blk_fill_status(&po->rx_ring); |
da413eec | 2407 | } |
1da177e4 LT |
2408 | |
2409 | drop_n_restore: | |
2410 | if (skb_head != skb->data && skb_shared(skb)) { | |
2411 | skb->data = skb_head; | |
2412 | skb->len = skb_len; | |
2413 | } | |
2414 | drop: | |
da37845f WJ |
2415 | if (!is_drop_n_account) |
2416 | consume_skb(skb); | |
2417 | else | |
2418 | kfree_skb(skb); | |
1da177e4 LT |
2419 | return 0; |
2420 | ||
58d19b19 | 2421 | drop_n_account: |
1da177e4 | 2422 | spin_unlock(&sk->sk_receive_queue.lock); |
8e8e2951 ED |
2423 | atomic_inc(&po->tp_drops); |
2424 | is_drop_n_account = true; | |
1da177e4 | 2425 | |
676d2369 | 2426 | sk->sk_data_ready(sk); |
acb5d75b | 2427 | kfree_skb(copy_skb); |
1da177e4 LT |
2428 | goto drop_n_restore; |
2429 | } | |
2430 | ||
69e3c75f JB |
2431 | static void tpacket_destruct_skb(struct sk_buff *skb) |
2432 | { | |
2433 | struct packet_sock *po = pkt_sk(skb->sk); | |
1da177e4 | 2434 | |
69e3c75f | 2435 | if (likely(po->tx_ring.pg_vec)) { |
f0d4eb29 | 2436 | void *ph; |
b9c32fb2 DB |
2437 | __u32 ts; |
2438 | ||
5cd8d46e | 2439 | ph = skb_zcopy_get_nouarg(skb); |
b0138408 | 2440 | packet_dec_pending(&po->tx_ring); |
b9c32fb2 DB |
2441 | |
2442 | ts = __packet_set_timestamp(po, ph, skb); | |
2443 | __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts); | |
89ed5b51 NH |
2444 | |
2445 | if (!packet_read_pending(&po->tx_ring)) | |
2446 | complete(&po->skb_completion); | |
69e3c75f JB |
2447 | } |
2448 | ||
2449 | sock_wfree(skb); | |
2450 | } | |
2451 | ||
16cc1400 WB |
2452 | static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len) |
2453 | { | |
16cc1400 WB |
2454 | if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && |
2455 | (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + | |
2456 | __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 > | |
2457 | __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len))) | |
2458 | vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(), | |
2459 | __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) + | |
2460 | __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2); | |
2461 | ||
2462 | if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len) | |
2463 | return -EINVAL; | |
2464 | ||
16cc1400 WB |
2465 | return 0; |
2466 | } | |
2467 | ||
2468 | static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len, | |
2469 | struct virtio_net_hdr *vnet_hdr) | |
2470 | { | |
16cc1400 WB |
2471 | if (*len < sizeof(*vnet_hdr)) |
2472 | return -EINVAL; | |
2473 | *len -= sizeof(*vnet_hdr); | |
2474 | ||
cbbd26b8 | 2475 | if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter)) |
16cc1400 WB |
2476 | return -EFAULT; |
2477 | ||
2478 | return __packet_snd_vnet_parse(vnet_hdr, *len); | |
2479 | } | |
2480 | ||
40d4e3df | 2481 | static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb, |
8d39b4a6 | 2482 | void *frame, struct net_device *dev, void *data, int tp_len, |
c14ac945 SHY |
2483 | __be16 proto, unsigned char *addr, int hlen, int copylen, |
2484 | const struct sockcm_cookie *sockc) | |
69e3c75f | 2485 | { |
184f489e | 2486 | union tpacket_uhdr ph; |
8d39b4a6 | 2487 | int to_write, offset, len, nr_frags, len_max; |
69e3c75f JB |
2488 | struct socket *sock = po->sk.sk_socket; |
2489 | struct page *page; | |
69e3c75f JB |
2490 | int err; |
2491 | ||
2492 | ph.raw = frame; | |
2493 | ||
2494 | skb->protocol = proto; | |
2495 | skb->dev = dev; | |
2496 | skb->priority = po->sk.sk_priority; | |
2d37a186 | 2497 | skb->mark = po->sk.sk_mark; |
3d0ba8c0 | 2498 | skb->tstamp = sockc->transmit_time; |
8f932f76 | 2499 | skb_setup_tx_timestamp(skb, sockc->tsflags); |
5cd8d46e | 2500 | skb_zcopy_set_nouarg(skb, ph.raw); |
69e3c75f | 2501 | |
ae641949 | 2502 | skb_reserve(skb, hlen); |
69e3c75f | 2503 | skb_reset_network_header(skb); |
c1aad275 | 2504 | |
69e3c75f JB |
2505 | to_write = tp_len; |
2506 | ||
2507 | if (sock->type == SOCK_DGRAM) { | |
2508 | err = dev_hard_header(skb, dev, ntohs(proto), addr, | |
2509 | NULL, tp_len); | |
2510 | if (unlikely(err < 0)) | |
2511 | return -EINVAL; | |
1d036d25 | 2512 | } else if (copylen) { |
9ed988cd WB |
2513 | int hdrlen = min_t(int, copylen, tp_len); |
2514 | ||
69e3c75f | 2515 | skb_push(skb, dev->hard_header_len); |
1d036d25 | 2516 | skb_put(skb, copylen - dev->hard_header_len); |
9ed988cd | 2517 | err = skb_store_bits(skb, 0, data, hdrlen); |
69e3c75f JB |
2518 | if (unlikely(err)) |
2519 | return err; | |
9ed988cd WB |
2520 | if (!dev_validate_header(dev, skb->data, hdrlen)) |
2521 | return -EINVAL; | |
69e3c75f | 2522 | |
9ed988cd WB |
2523 | data += hdrlen; |
2524 | to_write -= hdrlen; | |
69e3c75f JB |
2525 | } |
2526 | ||
69e3c75f JB |
2527 | offset = offset_in_page(data); |
2528 | len_max = PAGE_SIZE - offset; | |
2529 | len = ((to_write > len_max) ? len_max : to_write); | |
2530 | ||
2531 | skb->data_len = to_write; | |
2532 | skb->len += to_write; | |
2533 | skb->truesize += to_write; | |
14afee4b | 2534 | refcount_add(to_write, &po->sk.sk_wmem_alloc); |
69e3c75f JB |
2535 | |
2536 | while (likely(to_write)) { | |
2537 | nr_frags = skb_shinfo(skb)->nr_frags; | |
2538 | ||
2539 | if (unlikely(nr_frags >= MAX_SKB_FRAGS)) { | |
40d4e3df ED |
2540 | pr_err("Packet exceed the number of skb frags(%lu)\n", |
2541 | MAX_SKB_FRAGS); | |
69e3c75f JB |
2542 | return -EFAULT; |
2543 | } | |
2544 | ||
0af55bb5 CG |
2545 | page = pgv_to_page(data); |
2546 | data += len; | |
69e3c75f JB |
2547 | flush_dcache_page(page); |
2548 | get_page(page); | |
0af55bb5 | 2549 | skb_fill_page_desc(skb, nr_frags, page, offset, len); |
69e3c75f JB |
2550 | to_write -= len; |
2551 | offset = 0; | |
2552 | len_max = PAGE_SIZE; | |
2553 | len = ((to_write > len_max) ? len_max : to_write); | |
2554 | } | |
2555 | ||
75c65772 | 2556 | packet_parse_headers(skb, sock); |
efdfa2f7 | 2557 | |
69e3c75f JB |
2558 | return tp_len; |
2559 | } | |
2560 | ||
8d39b4a6 WB |
2561 | static int tpacket_parse_header(struct packet_sock *po, void *frame, |
2562 | int size_max, void **data) | |
2563 | { | |
2564 | union tpacket_uhdr ph; | |
2565 | int tp_len, off; | |
2566 | ||
2567 | ph.raw = frame; | |
2568 | ||
2569 | switch (po->tp_version) { | |
7f953ab2 SV |
2570 | case TPACKET_V3: |
2571 | if (ph.h3->tp_next_offset != 0) { | |
2572 | pr_warn_once("variable sized slot not supported"); | |
2573 | return -EINVAL; | |
2574 | } | |
2575 | tp_len = ph.h3->tp_len; | |
2576 | break; | |
8d39b4a6 WB |
2577 | case TPACKET_V2: |
2578 | tp_len = ph.h2->tp_len; | |
2579 | break; | |
2580 | default: | |
2581 | tp_len = ph.h1->tp_len; | |
2582 | break; | |
2583 | } | |
2584 | if (unlikely(tp_len > size_max)) { | |
2585 | pr_err("packet size is too long (%d > %d)\n", tp_len, size_max); | |
2586 | return -EMSGSIZE; | |
2587 | } | |
2588 | ||
2589 | if (unlikely(po->tp_tx_has_off)) { | |
2590 | int off_min, off_max; | |
2591 | ||
2592 | off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll); | |
2593 | off_max = po->tx_ring.frame_size - tp_len; | |
2594 | if (po->sk.sk_type == SOCK_DGRAM) { | |
2595 | switch (po->tp_version) { | |
7f953ab2 SV |
2596 | case TPACKET_V3: |
2597 | off = ph.h3->tp_net; | |
2598 | break; | |
8d39b4a6 WB |
2599 | case TPACKET_V2: |
2600 | off = ph.h2->tp_net; | |
2601 | break; | |
2602 | default: | |
2603 | off = ph.h1->tp_net; | |
2604 | break; | |
2605 | } | |
2606 | } else { | |
2607 | switch (po->tp_version) { | |
7f953ab2 SV |
2608 | case TPACKET_V3: |
2609 | off = ph.h3->tp_mac; | |
2610 | break; | |
8d39b4a6 WB |
2611 | case TPACKET_V2: |
2612 | off = ph.h2->tp_mac; | |
2613 | break; | |
2614 | default: | |
2615 | off = ph.h1->tp_mac; | |
2616 | break; | |
2617 | } | |
2618 | } | |
2619 | if (unlikely((off < off_min) || (off_max < off))) | |
2620 | return -EINVAL; | |
2621 | } else { | |
2622 | off = po->tp_hdrlen - sizeof(struct sockaddr_ll); | |
2623 | } | |
2624 | ||
2625 | *data = frame + off; | |
2626 | return tp_len; | |
2627 | } | |
2628 | ||
69e3c75f JB |
2629 | static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) |
2630 | { | |
89ed5b51 | 2631 | struct sk_buff *skb = NULL; |
69e3c75f | 2632 | struct net_device *dev; |
1d036d25 | 2633 | struct virtio_net_hdr *vnet_hdr = NULL; |
c14ac945 | 2634 | struct sockcm_cookie sockc; |
69e3c75f | 2635 | __be16 proto; |
09effa67 | 2636 | int err, reserve = 0; |
40d4e3df | 2637 | void *ph; |
342dfc30 | 2638 | DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); |
87a2fd28 | 2639 | bool need_wait = !(msg->msg_flags & MSG_DONTWAIT); |
486efdc8 | 2640 | unsigned char *addr = NULL; |
69e3c75f | 2641 | int tp_len, size_max; |
8d39b4a6 | 2642 | void *data; |
69e3c75f | 2643 | int len_sum = 0; |
9e67030a | 2644 | int status = TP_STATUS_AVAILABLE; |
1d036d25 | 2645 | int hlen, tlen, copylen = 0; |
89ed5b51 | 2646 | long timeo = 0; |
69e3c75f | 2647 | |
69e3c75f JB |
2648 | mutex_lock(&po->pg_vec_lock); |
2649 | ||
32d3182c ED |
2650 | /* packet_sendmsg() check on tx_ring.pg_vec was lockless, |
2651 | * we need to confirm it under protection of pg_vec_lock. | |
2652 | */ | |
2653 | if (unlikely(!po->tx_ring.pg_vec)) { | |
2654 | err = -EBUSY; | |
2655 | goto out; | |
2656 | } | |
66e56cd4 | 2657 | if (likely(saddr == NULL)) { |
e40526cb | 2658 | dev = packet_cached_dev_get(po); |
414b5431 | 2659 | proto = READ_ONCE(po->num); |
69e3c75f JB |
2660 | } else { |
2661 | err = -EINVAL; | |
2662 | if (msg->msg_namelen < sizeof(struct sockaddr_ll)) | |
2663 | goto out; | |
2664 | if (msg->msg_namelen < (saddr->sll_halen | |
2665 | + offsetof(struct sockaddr_ll, | |
2666 | sll_addr))) | |
2667 | goto out; | |
69e3c75f | 2668 | proto = saddr->sll_protocol; |
827d9780 | 2669 | dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex); |
486efdc8 WB |
2670 | if (po->sk.sk_socket->type == SOCK_DGRAM) { |
2671 | if (dev && msg->msg_namelen < dev->addr_len + | |
2672 | offsetof(struct sockaddr_ll, sll_addr)) | |
2673 | goto out_put; | |
2674 | addr = saddr->sll_addr; | |
2675 | } | |
69e3c75f JB |
2676 | } |
2677 | ||
69e3c75f JB |
2678 | err = -ENXIO; |
2679 | if (unlikely(dev == NULL)) | |
2680 | goto out; | |
69e3c75f JB |
2681 | err = -ENETDOWN; |
2682 | if (unlikely(!(dev->flags & IFF_UP))) | |
2683 | goto out_put; | |
2684 | ||
657a0667 | 2685 | sockcm_init(&sockc, &po->sk); |
d19b183c DCS |
2686 | if (msg->msg_controllen) { |
2687 | err = sock_cmsg_send(&po->sk, msg, &sockc); | |
2688 | if (unlikely(err)) | |
2689 | goto out_put; | |
2690 | } | |
2691 | ||
5cfb4c8d DB |
2692 | if (po->sk.sk_socket->type == SOCK_RAW) |
2693 | reserve = dev->hard_header_len; | |
69e3c75f | 2694 | size_max = po->tx_ring.frame_size |
b5dd884e | 2695 | - (po->tp_hdrlen - sizeof(struct sockaddr_ll)); |
69e3c75f | 2696 | |
1d036d25 | 2697 | if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr) |
5cfb4c8d | 2698 | size_max = dev->mtu + reserve + VLAN_HLEN; |
09effa67 | 2699 | |
89ed5b51 NH |
2700 | reinit_completion(&po->skb_completion); |
2701 | ||
69e3c75f JB |
2702 | do { |
2703 | ph = packet_current_frame(po, &po->tx_ring, | |
87a2fd28 | 2704 | TP_STATUS_SEND_REQUEST); |
69e3c75f | 2705 | if (unlikely(ph == NULL)) { |
89ed5b51 NH |
2706 | if (need_wait && skb) { |
2707 | timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT); | |
2708 | timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo); | |
2709 | if (timeo <= 0) { | |
2710 | err = !timeo ? -ETIMEDOUT : -ERESTARTSYS; | |
2711 | goto out_put; | |
2712 | } | |
2713 | } | |
2714 | /* check for additional frames */ | |
69e3c75f JB |
2715 | continue; |
2716 | } | |
2717 | ||
8d39b4a6 WB |
2718 | skb = NULL; |
2719 | tp_len = tpacket_parse_header(po, ph, size_max, &data); | |
2720 | if (tp_len < 0) | |
2721 | goto tpacket_error; | |
2722 | ||
69e3c75f | 2723 | status = TP_STATUS_SEND_REQUEST; |
ae641949 HX |
2724 | hlen = LL_RESERVED_SPACE(dev); |
2725 | tlen = dev->needed_tailroom; | |
1d036d25 WB |
2726 | if (po->has_vnet_hdr) { |
2727 | vnet_hdr = data; | |
2728 | data += sizeof(*vnet_hdr); | |
2729 | tp_len -= sizeof(*vnet_hdr); | |
2730 | if (tp_len < 0 || | |
2731 | __packet_snd_vnet_parse(vnet_hdr, tp_len)) { | |
2732 | tp_len = -EINVAL; | |
2733 | goto tpacket_error; | |
2734 | } | |
2735 | copylen = __virtio16_to_cpu(vio_le(), | |
2736 | vnet_hdr->hdr_len); | |
2737 | } | |
9ed988cd | 2738 | copylen = max_t(int, copylen, dev->hard_header_len); |
69e3c75f | 2739 | skb = sock_alloc_send_skb(&po->sk, |
1d036d25 WB |
2740 | hlen + tlen + sizeof(struct sockaddr_ll) + |
2741 | (copylen - dev->hard_header_len), | |
fbf33a28 | 2742 | !need_wait, &err); |
69e3c75f | 2743 | |
fbf33a28 KM |
2744 | if (unlikely(skb == NULL)) { |
2745 | /* we assume the socket was initially writeable ... */ | |
2746 | if (likely(len_sum > 0)) | |
2747 | err = len_sum; | |
69e3c75f | 2748 | goto out_status; |
fbf33a28 | 2749 | } |
8d39b4a6 | 2750 | tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto, |
c14ac945 | 2751 | addr, hlen, copylen, &sockc); |
dbd46ab4 | 2752 | if (likely(tp_len >= 0) && |
5cfb4c8d | 2753 | tp_len > dev->mtu + reserve && |
1d036d25 | 2754 | !po->has_vnet_hdr && |
3c70c132 DB |
2755 | !packet_extra_vlan_len_allowed(dev, skb)) |
2756 | tp_len = -EMSGSIZE; | |
69e3c75f JB |
2757 | |
2758 | if (unlikely(tp_len < 0)) { | |
8d39b4a6 | 2759 | tpacket_error: |
69e3c75f JB |
2760 | if (po->tp_loss) { |
2761 | __packet_set_status(po, ph, | |
2762 | TP_STATUS_AVAILABLE); | |
2763 | packet_increment_head(&po->tx_ring); | |
2764 | kfree_skb(skb); | |
2765 | continue; | |
2766 | } else { | |
2767 | status = TP_STATUS_WRONG_FORMAT; | |
2768 | err = tp_len; | |
2769 | goto out_status; | |
2770 | } | |
2771 | } | |
2772 | ||
9d2f67e4 JT |
2773 | if (po->has_vnet_hdr) { |
2774 | if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) { | |
2775 | tp_len = -EINVAL; | |
2776 | goto tpacket_error; | |
2777 | } | |
2778 | virtio_net_hdr_set_proto(skb, vnet_hdr); | |
1d036d25 WB |
2779 | } |
2780 | ||
69e3c75f JB |
2781 | skb->destructor = tpacket_destruct_skb; |
2782 | __packet_set_status(po, ph, TP_STATUS_SENDING); | |
b0138408 | 2783 | packet_inc_pending(&po->tx_ring); |
69e3c75f JB |
2784 | |
2785 | status = TP_STATUS_SEND_REQUEST; | |
d346a3fa | 2786 | err = po->xmit(skb); |
eb70df13 JP |
2787 | if (unlikely(err > 0)) { |
2788 | err = net_xmit_errno(err); | |
2789 | if (err && __packet_get_status(po, ph) == | |
2790 | TP_STATUS_AVAILABLE) { | |
2791 | /* skb was destructed already */ | |
2792 | skb = NULL; | |
2793 | goto out_status; | |
2794 | } | |
2795 | /* | |
2796 | * skb was dropped but not destructed yet; | |
2797 | * let's treat it like congestion or err < 0 | |
2798 | */ | |
2799 | err = 0; | |
2800 | } | |
69e3c75f JB |
2801 | packet_increment_head(&po->tx_ring); |
2802 | len_sum += tp_len; | |
b0138408 DB |
2803 | } while (likely((ph != NULL) || |
2804 | /* Note: packet_read_pending() might be slow if we have | |
2805 | * to call it as it's per_cpu variable, but in fast-path | |
2806 | * we already short-circuit the loop with the first | |
2807 | * condition, and luckily don't have to go that path | |
2808 | * anyway. | |
2809 | */ | |
2810 | (need_wait && packet_read_pending(&po->tx_ring)))); | |
69e3c75f JB |
2811 | |
2812 | err = len_sum; | |
2813 | goto out_put; | |
2814 | ||
69e3c75f JB |
2815 | out_status: |
2816 | __packet_set_status(po, ph, status); | |
2817 | kfree_skb(skb); | |
2818 | out_put: | |
e40526cb | 2819 | dev_put(dev); |
69e3c75f JB |
2820 | out: |
2821 | mutex_unlock(&po->pg_vec_lock); | |
2822 | return err; | |
2823 | } | |
69e3c75f | 2824 | |
eea49cc9 OJ |
2825 | static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad, |
2826 | size_t reserve, size_t len, | |
2827 | size_t linear, int noblock, | |
2828 | int *err) | |
bfd5f4a3 SS |
2829 | { |
2830 | struct sk_buff *skb; | |
2831 | ||
2832 | /* Under a page? Don't bother with paged skb. */ | |
2833 | if (prepad + len < PAGE_SIZE || !linear) | |
2834 | linear = len; | |
2835 | ||
2836 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
28d64271 | 2837 | err, 0); |
bfd5f4a3 SS |
2838 | if (!skb) |
2839 | return NULL; | |
2840 | ||
2841 | skb_reserve(skb, reserve); | |
2842 | skb_put(skb, linear); | |
2843 | skb->data_len = len - linear; | |
2844 | skb->len += len - linear; | |
2845 | ||
2846 | return skb; | |
2847 | } | |
2848 | ||
d346a3fa | 2849 | static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len) |
1da177e4 LT |
2850 | { |
2851 | struct sock *sk = sock->sk; | |
342dfc30 | 2852 | DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name); |
1da177e4 LT |
2853 | struct sk_buff *skb; |
2854 | struct net_device *dev; | |
0e11c91e | 2855 | __be16 proto; |
486efdc8 | 2856 | unsigned char *addr = NULL; |
827d9780 | 2857 | int err, reserve = 0; |
c7d39e32 | 2858 | struct sockcm_cookie sockc; |
bfd5f4a3 SS |
2859 | struct virtio_net_hdr vnet_hdr = { 0 }; |
2860 | int offset = 0; | |
bfd5f4a3 | 2861 | struct packet_sock *po = pkt_sk(sk); |
da7c9561 | 2862 | bool has_vnet_hdr = false; |
57031eb7 | 2863 | int hlen, tlen, linear; |
3bdc0eba | 2864 | int extra_len = 0; |
1da177e4 LT |
2865 | |
2866 | /* | |
1ce4f28b | 2867 | * Get and verify the address. |
1da177e4 | 2868 | */ |
1ce4f28b | 2869 | |
66e56cd4 | 2870 | if (likely(saddr == NULL)) { |
e40526cb | 2871 | dev = packet_cached_dev_get(po); |
414b5431 | 2872 | proto = READ_ONCE(po->num); |
1da177e4 LT |
2873 | } else { |
2874 | err = -EINVAL; | |
2875 | if (msg->msg_namelen < sizeof(struct sockaddr_ll)) | |
2876 | goto out; | |
0fb375fb EB |
2877 | if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr))) |
2878 | goto out; | |
1da177e4 | 2879 | proto = saddr->sll_protocol; |
827d9780 | 2880 | dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex); |
486efdc8 WB |
2881 | if (sock->type == SOCK_DGRAM) { |
2882 | if (dev && msg->msg_namelen < dev->addr_len + | |
2883 | offsetof(struct sockaddr_ll, sll_addr)) | |
2884 | goto out_unlock; | |
2885 | addr = saddr->sll_addr; | |
2886 | } | |
1da177e4 LT |
2887 | } |
2888 | ||
1da177e4 | 2889 | err = -ENXIO; |
e40526cb | 2890 | if (unlikely(dev == NULL)) |
1da177e4 | 2891 | goto out_unlock; |
d5e76b0a | 2892 | err = -ENETDOWN; |
e40526cb | 2893 | if (unlikely(!(dev->flags & IFF_UP))) |
d5e76b0a DM |
2894 | goto out_unlock; |
2895 | ||
657a0667 | 2896 | sockcm_init(&sockc, sk); |
c7d39e32 EJ |
2897 | sockc.mark = sk->sk_mark; |
2898 | if (msg->msg_controllen) { | |
2899 | err = sock_cmsg_send(sk, msg, &sockc); | |
2900 | if (unlikely(err)) | |
2901 | goto out_unlock; | |
2902 | } | |
2903 | ||
e40526cb DB |
2904 | if (sock->type == SOCK_RAW) |
2905 | reserve = dev->hard_header_len; | |
bfd5f4a3 | 2906 | if (po->has_vnet_hdr) { |
16cc1400 WB |
2907 | err = packet_snd_vnet_parse(msg, &len, &vnet_hdr); |
2908 | if (err) | |
bfd5f4a3 | 2909 | goto out_unlock; |
da7c9561 | 2910 | has_vnet_hdr = true; |
bfd5f4a3 SS |
2911 | } |
2912 | ||
3bdc0eba BG |
2913 | if (unlikely(sock_flag(sk, SOCK_NOFCS))) { |
2914 | if (!netif_supports_nofcs(dev)) { | |
2915 | err = -EPROTONOSUPPORT; | |
2916 | goto out_unlock; | |
2917 | } | |
2918 | extra_len = 4; /* We're doing our own CRC */ | |
2919 | } | |
2920 | ||
1da177e4 | 2921 | err = -EMSGSIZE; |
16cc1400 WB |
2922 | if (!vnet_hdr.gso_type && |
2923 | (len > dev->mtu + reserve + VLAN_HLEN + extra_len)) | |
1da177e4 LT |
2924 | goto out_unlock; |
2925 | ||
bfd5f4a3 | 2926 | err = -ENOBUFS; |
ae641949 HX |
2927 | hlen = LL_RESERVED_SPACE(dev); |
2928 | tlen = dev->needed_tailroom; | |
57031eb7 WB |
2929 | linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len); |
2930 | linear = max(linear, min_t(int, len, dev->hard_header_len)); | |
2931 | skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear, | |
bfd5f4a3 | 2932 | msg->msg_flags & MSG_DONTWAIT, &err); |
40d4e3df | 2933 | if (skb == NULL) |
1da177e4 LT |
2934 | goto out_unlock; |
2935 | ||
b84bbaf7 | 2936 | skb_reset_network_header(skb); |
1da177e4 | 2937 | |
0c4e8581 | 2938 | err = -EINVAL; |
9c707762 WB |
2939 | if (sock->type == SOCK_DGRAM) { |
2940 | offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len); | |
46d2cfb1 | 2941 | if (unlikely(offset < 0)) |
9c707762 | 2942 | goto out_free; |
b84bbaf7 | 2943 | } else if (reserve) { |
9aad13b0 | 2944 | skb_reserve(skb, -reserve); |
88a8121d ND |
2945 | if (len < reserve + sizeof(struct ipv6hdr) && |
2946 | dev->min_header_len != dev->hard_header_len) | |
993675a3 | 2947 | skb_reset_network_header(skb); |
9c707762 | 2948 | } |
1da177e4 LT |
2949 | |
2950 | /* Returns -EFAULT on error */ | |
c0371da6 | 2951 | err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len); |
1da177e4 LT |
2952 | if (err) |
2953 | goto out_free; | |
bf84a010 | 2954 | |
9ed988cd WB |
2955 | if (sock->type == SOCK_RAW && |
2956 | !dev_validate_header(dev, skb->data, len)) { | |
2957 | err = -EINVAL; | |
2958 | goto out_free; | |
2959 | } | |
2960 | ||
8f932f76 | 2961 | skb_setup_tx_timestamp(skb, sockc.tsflags); |
1da177e4 | 2962 | |
16cc1400 | 2963 | if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) && |
3c70c132 DB |
2964 | !packet_extra_vlan_len_allowed(dev, skb)) { |
2965 | err = -EMSGSIZE; | |
2966 | goto out_free; | |
57f89bfa BG |
2967 | } |
2968 | ||
09effa67 DM |
2969 | skb->protocol = proto; |
2970 | skb->dev = dev; | |
1da177e4 | 2971 | skb->priority = sk->sk_priority; |
c7d39e32 | 2972 | skb->mark = sockc.mark; |
3d0ba8c0 | 2973 | skb->tstamp = sockc.transmit_time; |
0fd5d57b | 2974 | |
da7c9561 | 2975 | if (has_vnet_hdr) { |
db60eb5f | 2976 | err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le()); |
16cc1400 WB |
2977 | if (err) |
2978 | goto out_free; | |
2979 | len += sizeof(vnet_hdr); | |
9d2f67e4 | 2980 | virtio_net_hdr_set_proto(skb, &vnet_hdr); |
bfd5f4a3 SS |
2981 | } |
2982 | ||
75c65772 | 2983 | packet_parse_headers(skb, sock); |
8fd6c80d | 2984 | |
3bdc0eba BG |
2985 | if (unlikely(extra_len == 4)) |
2986 | skb->no_fcs = 1; | |
2987 | ||
d346a3fa | 2988 | err = po->xmit(skb); |
1da177e4 LT |
2989 | if (err > 0 && (err = net_xmit_errno(err)) != 0) |
2990 | goto out_unlock; | |
2991 | ||
e40526cb | 2992 | dev_put(dev); |
1da177e4 | 2993 | |
40d4e3df | 2994 | return len; |
1da177e4 LT |
2995 | |
2996 | out_free: | |
2997 | kfree_skb(skb); | |
2998 | out_unlock: | |
e40526cb | 2999 | if (dev) |
1da177e4 LT |
3000 | dev_put(dev); |
3001 | out: | |
3002 | return err; | |
3003 | } | |
3004 | ||
1b784140 | 3005 | static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) |
69e3c75f | 3006 | { |
69e3c75f JB |
3007 | struct sock *sk = sock->sk; |
3008 | struct packet_sock *po = pkt_sk(sk); | |
d346a3fa | 3009 | |
69e3c75f JB |
3010 | if (po->tx_ring.pg_vec) |
3011 | return tpacket_snd(po, msg); | |
3012 | else | |
69e3c75f JB |
3013 | return packet_snd(sock, msg, len); |
3014 | } | |
3015 | ||
1da177e4 LT |
3016 | /* |
3017 | * Close a PACKET socket. This is fairly simple. We immediately go | |
3018 | * to 'closed' state and remove our protocol entry in the device list. | |
3019 | */ | |
3020 | ||
3021 | static int packet_release(struct socket *sock) | |
3022 | { | |
3023 | struct sock *sk = sock->sk; | |
3024 | struct packet_sock *po; | |
2bd624b4 | 3025 | struct packet_fanout *f; |
d12d01d6 | 3026 | struct net *net; |
f6fb8f10 | 3027 | union tpacket_req_u req_u; |
1da177e4 LT |
3028 | |
3029 | if (!sk) | |
3030 | return 0; | |
3031 | ||
3b1e0a65 | 3032 | net = sock_net(sk); |
1da177e4 LT |
3033 | po = pkt_sk(sk); |
3034 | ||
0fa7fa98 | 3035 | mutex_lock(&net->packet.sklist_lock); |
808f5114 | 3036 | sk_del_node_init_rcu(sk); |
0fa7fa98 PE |
3037 | mutex_unlock(&net->packet.sklist_lock); |
3038 | ||
3039 | preempt_disable(); | |
920de804 | 3040 | sock_prot_inuse_add(net, sk->sk_prot, -1); |
0fa7fa98 | 3041 | preempt_enable(); |
1da177e4 | 3042 | |
808f5114 | 3043 | spin_lock(&po->bind_lock); |
ce06b03e | 3044 | unregister_prot_hook(sk, false); |
66e56cd4 DB |
3045 | packet_cached_dev_reset(po); |
3046 | ||
160ff18a BG |
3047 | if (po->prot_hook.dev) { |
3048 | dev_put(po->prot_hook.dev); | |
3049 | po->prot_hook.dev = NULL; | |
3050 | } | |
808f5114 | 3051 | spin_unlock(&po->bind_lock); |
1da177e4 | 3052 | |
1da177e4 | 3053 | packet_flush_mclist(sk); |
1da177e4 | 3054 | |
5171b37d | 3055 | lock_sock(sk); |
9665d5d6 PS |
3056 | if (po->rx_ring.pg_vec) { |
3057 | memset(&req_u, 0, sizeof(req_u)); | |
f6fb8f10 | 3058 | packet_set_ring(sk, &req_u, 1, 0); |
9665d5d6 | 3059 | } |
69e3c75f | 3060 | |
9665d5d6 PS |
3061 | if (po->tx_ring.pg_vec) { |
3062 | memset(&req_u, 0, sizeof(req_u)); | |
f6fb8f10 | 3063 | packet_set_ring(sk, &req_u, 1, 1); |
9665d5d6 | 3064 | } |
5171b37d | 3065 | release_sock(sk); |
1da177e4 | 3066 | |
2bd624b4 | 3067 | f = fanout_release(sk); |
dc99f600 | 3068 | |
808f5114 | 3069 | synchronize_net(); |
2bd624b4 | 3070 | |
afa0925c | 3071 | kfree(po->rollover); |
2bd624b4 AS |
3072 | if (f) { |
3073 | fanout_release_data(f); | |
3074 | kfree(f); | |
3075 | } | |
1da177e4 LT |
3076 | /* |
3077 | * Now the socket is dead. No more input will appear. | |
3078 | */ | |
1da177e4 LT |
3079 | sock_orphan(sk); |
3080 | sock->sk = NULL; | |
3081 | ||
3082 | /* Purge queues */ | |
3083 | ||
3084 | skb_queue_purge(&sk->sk_receive_queue); | |
b0138408 | 3085 | packet_free_pending(po); |
17ab56a2 | 3086 | sk_refcnt_debug_release(sk); |
1da177e4 LT |
3087 | |
3088 | sock_put(sk); | |
3089 | return 0; | |
3090 | } | |
3091 | ||
3092 | /* | |
3093 | * Attach a packet hook. | |
3094 | */ | |
3095 | ||
30f7ea1c FR |
3096 | static int packet_do_bind(struct sock *sk, const char *name, int ifindex, |
3097 | __be16 proto) | |
1da177e4 LT |
3098 | { |
3099 | struct packet_sock *po = pkt_sk(sk); | |
158cd4af | 3100 | struct net_device *dev_curr; |
902fefb8 DB |
3101 | __be16 proto_curr; |
3102 | bool need_rehook; | |
30f7ea1c FR |
3103 | struct net_device *dev = NULL; |
3104 | int ret = 0; | |
3105 | bool unlisted = false; | |
dc99f600 | 3106 | |
1da177e4 | 3107 | lock_sock(sk); |
1da177e4 | 3108 | spin_lock(&po->bind_lock); |
30f7ea1c FR |
3109 | rcu_read_lock(); |
3110 | ||
4971613c WB |
3111 | if (po->fanout) { |
3112 | ret = -EINVAL; | |
3113 | goto out_unlock; | |
3114 | } | |
3115 | ||
30f7ea1c FR |
3116 | if (name) { |
3117 | dev = dev_get_by_name_rcu(sock_net(sk), name); | |
3118 | if (!dev) { | |
3119 | ret = -ENODEV; | |
3120 | goto out_unlock; | |
3121 | } | |
3122 | } else if (ifindex) { | |
3123 | dev = dev_get_by_index_rcu(sock_net(sk), ifindex); | |
3124 | if (!dev) { | |
3125 | ret = -ENODEV; | |
3126 | goto out_unlock; | |
3127 | } | |
3128 | } | |
3129 | ||
3130 | if (dev) | |
3131 | dev_hold(dev); | |
66e56cd4 | 3132 | |
902fefb8 DB |
3133 | proto_curr = po->prot_hook.type; |
3134 | dev_curr = po->prot_hook.dev; | |
3135 | ||
3136 | need_rehook = proto_curr != proto || dev_curr != dev; | |
3137 | ||
3138 | if (need_rehook) { | |
30f7ea1c FR |
3139 | if (po->running) { |
3140 | rcu_read_unlock(); | |
15fe076e ED |
3141 | /* prevents packet_notifier() from calling |
3142 | * register_prot_hook() | |
3143 | */ | |
414b5431 | 3144 | WRITE_ONCE(po->num, 0); |
30f7ea1c FR |
3145 | __unregister_prot_hook(sk, true); |
3146 | rcu_read_lock(); | |
3147 | dev_curr = po->prot_hook.dev; | |
3148 | if (dev) | |
3149 | unlisted = !dev_get_by_index_rcu(sock_net(sk), | |
3150 | dev->ifindex); | |
3151 | } | |
1da177e4 | 3152 | |
15fe076e | 3153 | BUG_ON(po->running); |
414b5431 | 3154 | WRITE_ONCE(po->num, proto); |
902fefb8 | 3155 | po->prot_hook.type = proto; |
902fefb8 | 3156 | |
30f7ea1c FR |
3157 | if (unlikely(unlisted)) { |
3158 | dev_put(dev); | |
3159 | po->prot_hook.dev = NULL; | |
3160 | po->ifindex = -1; | |
3161 | packet_cached_dev_reset(po); | |
3162 | } else { | |
3163 | po->prot_hook.dev = dev; | |
3164 | po->ifindex = dev ? dev->ifindex : 0; | |
3165 | packet_cached_dev_assign(po, dev); | |
3166 | } | |
902fefb8 | 3167 | } |
158cd4af LW |
3168 | if (dev_curr) |
3169 | dev_put(dev_curr); | |
66e56cd4 | 3170 | |
902fefb8 | 3171 | if (proto == 0 || !need_rehook) |
1da177e4 LT |
3172 | goto out_unlock; |
3173 | ||
30f7ea1c | 3174 | if (!unlisted && (!dev || (dev->flags & IFF_UP))) { |
ce06b03e | 3175 | register_prot_hook(sk); |
be85d4ad UT |
3176 | } else { |
3177 | sk->sk_err = ENETDOWN; | |
3178 | if (!sock_flag(sk, SOCK_DEAD)) | |
3179 | sk->sk_error_report(sk); | |
1da177e4 LT |
3180 | } |
3181 | ||
3182 | out_unlock: | |
30f7ea1c | 3183 | rcu_read_unlock(); |
1da177e4 LT |
3184 | spin_unlock(&po->bind_lock); |
3185 | release_sock(sk); | |
30f7ea1c | 3186 | return ret; |
1da177e4 LT |
3187 | } |
3188 | ||
3189 | /* | |
3190 | * Bind a packet socket to a device | |
3191 | */ | |
3192 | ||
40d4e3df ED |
3193 | static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, |
3194 | int addr_len) | |
1da177e4 | 3195 | { |
40d4e3df | 3196 | struct sock *sk = sock->sk; |
540e2894 | 3197 | char name[sizeof(uaddr->sa_data) + 1]; |
1ce4f28b | 3198 | |
1da177e4 LT |
3199 | /* |
3200 | * Check legality | |
3201 | */ | |
1ce4f28b | 3202 | |
8ae55f04 | 3203 | if (addr_len != sizeof(struct sockaddr)) |
1da177e4 | 3204 | return -EINVAL; |
540e2894 AP |
3205 | /* uaddr->sa_data comes from the userspace, it's not guaranteed to be |
3206 | * zero-terminated. | |
3207 | */ | |
3208 | memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data)); | |
3209 | name[sizeof(uaddr->sa_data)] = 0; | |
1da177e4 | 3210 | |
30f7ea1c | 3211 | return packet_do_bind(sk, name, 0, pkt_sk(sk)->num); |
1da177e4 | 3212 | } |
1da177e4 LT |
3213 | |
3214 | static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) | |
3215 | { | |
40d4e3df ED |
3216 | struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr; |
3217 | struct sock *sk = sock->sk; | |
1da177e4 LT |
3218 | |
3219 | /* | |
3220 | * Check legality | |
3221 | */ | |
1ce4f28b | 3222 | |
1da177e4 LT |
3223 | if (addr_len < sizeof(struct sockaddr_ll)) |
3224 | return -EINVAL; | |
3225 | if (sll->sll_family != AF_PACKET) | |
3226 | return -EINVAL; | |
3227 | ||
30f7ea1c FR |
3228 | return packet_do_bind(sk, NULL, sll->sll_ifindex, |
3229 | sll->sll_protocol ? : pkt_sk(sk)->num); | |
1da177e4 LT |
3230 | } |
3231 | ||
3232 | static struct proto packet_proto = { | |
3233 | .name = "PACKET", | |
3234 | .owner = THIS_MODULE, | |
3235 | .obj_size = sizeof(struct packet_sock), | |
3236 | }; | |
3237 | ||
3238 | /* | |
1ce4f28b | 3239 | * Create a packet of type SOCK_PACKET. |
1da177e4 LT |
3240 | */ |
3241 | ||
3f378b68 EP |
3242 | static int packet_create(struct net *net, struct socket *sock, int protocol, |
3243 | int kern) | |
1da177e4 LT |
3244 | { |
3245 | struct sock *sk; | |
3246 | struct packet_sock *po; | |
0e11c91e | 3247 | __be16 proto = (__force __be16)protocol; /* weird, but documented */ |
1da177e4 LT |
3248 | int err; |
3249 | ||
df008c91 | 3250 | if (!ns_capable(net->user_ns, CAP_NET_RAW)) |
1da177e4 | 3251 | return -EPERM; |
be02097c DM |
3252 | if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW && |
3253 | sock->type != SOCK_PACKET) | |
1da177e4 LT |
3254 | return -ESOCKTNOSUPPORT; |
3255 | ||
3256 | sock->state = SS_UNCONNECTED; | |
3257 | ||
3258 | err = -ENOBUFS; | |
11aa9c28 | 3259 | sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern); |
1da177e4 LT |
3260 | if (sk == NULL) |
3261 | goto out; | |
3262 | ||
3263 | sock->ops = &packet_ops; | |
1da177e4 LT |
3264 | if (sock->type == SOCK_PACKET) |
3265 | sock->ops = &packet_ops_spkt; | |
be02097c | 3266 | |
1da177e4 LT |
3267 | sock_init_data(sock, sk); |
3268 | ||
3269 | po = pkt_sk(sk); | |
89ed5b51 | 3270 | init_completion(&po->skb_completion); |
1da177e4 | 3271 | sk->sk_family = PF_PACKET; |
0e11c91e | 3272 | po->num = proto; |
d346a3fa | 3273 | po->xmit = dev_queue_xmit; |
66e56cd4 | 3274 | |
b0138408 DB |
3275 | err = packet_alloc_pending(po); |
3276 | if (err) | |
3277 | goto out2; | |
3278 | ||
66e56cd4 | 3279 | packet_cached_dev_reset(po); |
1da177e4 LT |
3280 | |
3281 | sk->sk_destruct = packet_sock_destruct; | |
17ab56a2 | 3282 | sk_refcnt_debug_inc(sk); |
1da177e4 LT |
3283 | |
3284 | /* | |
3285 | * Attach a protocol block | |
3286 | */ | |
3287 | ||
3288 | spin_lock_init(&po->bind_lock); | |
905db440 | 3289 | mutex_init(&po->pg_vec_lock); |
0648ab70 | 3290 | po->rollover = NULL; |
1da177e4 | 3291 | po->prot_hook.func = packet_rcv; |
be02097c | 3292 | |
1da177e4 LT |
3293 | if (sock->type == SOCK_PACKET) |
3294 | po->prot_hook.func = packet_rcv_spkt; | |
be02097c | 3295 | |
1da177e4 LT |
3296 | po->prot_hook.af_packet_priv = sk; |
3297 | ||
0e11c91e AV |
3298 | if (proto) { |
3299 | po->prot_hook.type = proto; | |
a6361f0c | 3300 | __register_prot_hook(sk); |
1da177e4 LT |
3301 | } |
3302 | ||
0fa7fa98 | 3303 | mutex_lock(&net->packet.sklist_lock); |
a4dc6a49 | 3304 | sk_add_node_tail_rcu(sk, &net->packet.sklist); |
0fa7fa98 PE |
3305 | mutex_unlock(&net->packet.sklist_lock); |
3306 | ||
3307 | preempt_disable(); | |
3680453c | 3308 | sock_prot_inuse_add(net, &packet_proto, 1); |
0fa7fa98 | 3309 | preempt_enable(); |
808f5114 | 3310 | |
40d4e3df | 3311 | return 0; |
b0138408 DB |
3312 | out2: |
3313 | sk_free(sk); | |
1da177e4 LT |
3314 | out: |
3315 | return err; | |
3316 | } | |
3317 | ||
3318 | /* | |
3319 | * Pull a packet from our receive queue and hand it to the user. | |
3320 | * If necessary we block. | |
3321 | */ | |
3322 | ||
1b784140 YX |
3323 | static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, |
3324 | int flags) | |
1da177e4 LT |
3325 | { |
3326 | struct sock *sk = sock->sk; | |
3327 | struct sk_buff *skb; | |
3328 | int copied, err; | |
bfd5f4a3 | 3329 | int vnet_hdr_len = 0; |
2472d761 | 3330 | unsigned int origlen = 0; |
1da177e4 LT |
3331 | |
3332 | err = -EINVAL; | |
ed85b565 | 3333 | if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE)) |
1da177e4 LT |
3334 | goto out; |
3335 | ||
3336 | #if 0 | |
3337 | /* What error should we return now? EUNATTACH? */ | |
3338 | if (pkt_sk(sk)->ifindex < 0) | |
3339 | return -ENODEV; | |
3340 | #endif | |
3341 | ||
ed85b565 | 3342 | if (flags & MSG_ERRQUEUE) { |
cb820f8e RC |
3343 | err = sock_recv_errqueue(sk, msg, len, |
3344 | SOL_PACKET, PACKET_TX_TIMESTAMP); | |
ed85b565 RC |
3345 | goto out; |
3346 | } | |
3347 | ||
1da177e4 LT |
3348 | /* |
3349 | * Call the generic datagram receiver. This handles all sorts | |
3350 | * of horrible races and re-entrancy so we can forget about it | |
3351 | * in the protocol layers. | |
3352 | * | |
3353 | * Now it will return ENETDOWN, if device have just gone down, | |
3354 | * but then it will block. | |
3355 | */ | |
3356 | ||
40d4e3df | 3357 | skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err); |
1da177e4 LT |
3358 | |
3359 | /* | |
1ce4f28b | 3360 | * An error occurred so return it. Because skb_recv_datagram() |
1da177e4 LT |
3361 | * handles the blocking we don't see and worry about blocking |
3362 | * retries. | |
3363 | */ | |
3364 | ||
8ae55f04 | 3365 | if (skb == NULL) |
1da177e4 LT |
3366 | goto out; |
3367 | ||
9bb6cd65 | 3368 | packet_rcv_try_clear_pressure(pkt_sk(sk)); |
2ccdbaa6 | 3369 | |
bfd5f4a3 | 3370 | if (pkt_sk(sk)->has_vnet_hdr) { |
16cc1400 WB |
3371 | err = packet_rcv_vnet(msg, skb, &len); |
3372 | if (err) | |
bfd5f4a3 | 3373 | goto out_free; |
16cc1400 | 3374 | vnet_hdr_len = sizeof(struct virtio_net_hdr); |
bfd5f4a3 SS |
3375 | } |
3376 | ||
f3d33426 HFS |
3377 | /* You lose any data beyond the buffer you gave. If it worries |
3378 | * a user program they can ask the device for its MTU | |
3379 | * anyway. | |
1da177e4 | 3380 | */ |
1da177e4 | 3381 | copied = skb->len; |
40d4e3df ED |
3382 | if (copied > len) { |
3383 | copied = len; | |
3384 | msg->msg_flags |= MSG_TRUNC; | |
1da177e4 LT |
3385 | } |
3386 | ||
51f3d02b | 3387 | err = skb_copy_datagram_msg(skb, 0, msg, copied); |
1da177e4 LT |
3388 | if (err) |
3389 | goto out_free; | |
3390 | ||
2472d761 EB |
3391 | if (sock->type != SOCK_PACKET) { |
3392 | struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; | |
3393 | ||
3394 | /* Original length was stored in sockaddr_ll fields */ | |
3395 | origlen = PACKET_SKB_CB(skb)->sa.origlen; | |
3396 | sll->sll_family = AF_PACKET; | |
3397 | sll->sll_protocol = skb->protocol; | |
3398 | } | |
3399 | ||
3b885787 | 3400 | sock_recv_ts_and_drops(msg, sk, skb); |
1da177e4 | 3401 | |
f3d33426 | 3402 | if (msg->msg_name) { |
b2cf86e1 WB |
3403 | int copy_len; |
3404 | ||
f3d33426 HFS |
3405 | /* If the address length field is there to be filled |
3406 | * in, we fill it in now. | |
3407 | */ | |
3408 | if (sock->type == SOCK_PACKET) { | |
342dfc30 | 3409 | __sockaddr_check_size(sizeof(struct sockaddr_pkt)); |
f3d33426 | 3410 | msg->msg_namelen = sizeof(struct sockaddr_pkt); |
b2cf86e1 | 3411 | copy_len = msg->msg_namelen; |
f3d33426 HFS |
3412 | } else { |
3413 | struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll; | |
2472d761 | 3414 | |
f3d33426 HFS |
3415 | msg->msg_namelen = sll->sll_halen + |
3416 | offsetof(struct sockaddr_ll, sll_addr); | |
b2cf86e1 WB |
3417 | copy_len = msg->msg_namelen; |
3418 | if (msg->msg_namelen < sizeof(struct sockaddr_ll)) { | |
3419 | memset(msg->msg_name + | |
3420 | offsetof(struct sockaddr_ll, sll_addr), | |
3421 | 0, sizeof(sll->sll_addr)); | |
3422 | msg->msg_namelen = sizeof(struct sockaddr_ll); | |
3423 | } | |
f3d33426 | 3424 | } |
b2cf86e1 | 3425 | memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len); |
f3d33426 | 3426 | } |
1da177e4 | 3427 | |
8dc41944 | 3428 | if (pkt_sk(sk)->auxdata) { |
ffbc6111 HX |
3429 | struct tpacket_auxdata aux; |
3430 | ||
3431 | aux.tp_status = TP_STATUS_USER; | |
3432 | if (skb->ip_summed == CHECKSUM_PARTIAL) | |
3433 | aux.tp_status |= TP_STATUS_CSUMNOTREADY; | |
682f048b AD |
3434 | else if (skb->pkt_type != PACKET_OUTGOING && |
3435 | (skb->ip_summed == CHECKSUM_COMPLETE || | |
3436 | skb_csum_unnecessary(skb))) | |
3437 | aux.tp_status |= TP_STATUS_CSUM_VALID; | |
3438 | ||
2472d761 | 3439 | aux.tp_len = origlen; |
ffbc6111 HX |
3440 | aux.tp_snaplen = skb->len; |
3441 | aux.tp_mac = 0; | |
bbe735e4 | 3442 | aux.tp_net = skb_network_offset(skb); |
df8a39de JP |
3443 | if (skb_vlan_tag_present(skb)) { |
3444 | aux.tp_vlan_tci = skb_vlan_tag_get(skb); | |
a0cdfcf3 AW |
3445 | aux.tp_vlan_tpid = ntohs(skb->vlan_proto); |
3446 | aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID; | |
a3bcc23e BG |
3447 | } else { |
3448 | aux.tp_vlan_tci = 0; | |
a0cdfcf3 | 3449 | aux.tp_vlan_tpid = 0; |
a3bcc23e | 3450 | } |
ffbc6111 | 3451 | put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux); |
8dc41944 HX |
3452 | } |
3453 | ||
1da177e4 LT |
3454 | /* |
3455 | * Free or return the buffer as appropriate. Again this | |
3456 | * hides all the races and re-entrancy issues from us. | |
3457 | */ | |
bfd5f4a3 | 3458 | err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied); |
1da177e4 LT |
3459 | |
3460 | out_free: | |
3461 | skb_free_datagram(sk, skb); | |
3462 | out: | |
3463 | return err; | |
3464 | } | |
3465 | ||
1da177e4 | 3466 | static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr, |
9b2c45d4 | 3467 | int peer) |
1da177e4 LT |
3468 | { |
3469 | struct net_device *dev; | |
3470 | struct sock *sk = sock->sk; | |
3471 | ||
3472 | if (peer) | |
3473 | return -EOPNOTSUPP; | |
3474 | ||
3475 | uaddr->sa_family = AF_PACKET; | |
2dc85bf3 | 3476 | memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data)); |
654d1f8a ED |
3477 | rcu_read_lock(); |
3478 | dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex); | |
3479 | if (dev) | |
2dc85bf3 | 3480 | strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data)); |
654d1f8a | 3481 | rcu_read_unlock(); |
1da177e4 | 3482 | |
9b2c45d4 | 3483 | return sizeof(*uaddr); |
1da177e4 | 3484 | } |
1da177e4 LT |
3485 | |
3486 | static int packet_getname(struct socket *sock, struct sockaddr *uaddr, | |
9b2c45d4 | 3487 | int peer) |
1da177e4 LT |
3488 | { |
3489 | struct net_device *dev; | |
3490 | struct sock *sk = sock->sk; | |
3491 | struct packet_sock *po = pkt_sk(sk); | |
13cfa97b | 3492 | DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr); |
1da177e4 LT |
3493 | |
3494 | if (peer) | |
3495 | return -EOPNOTSUPP; | |
3496 | ||
3497 | sll->sll_family = AF_PACKET; | |
3498 | sll->sll_ifindex = po->ifindex; | |
414b5431 | 3499 | sll->sll_protocol = READ_ONCE(po->num); |
67286640 | 3500 | sll->sll_pkttype = 0; |
654d1f8a ED |
3501 | rcu_read_lock(); |
3502 | dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex); | |
1da177e4 LT |
3503 | if (dev) { |
3504 | sll->sll_hatype = dev->type; | |
3505 | sll->sll_halen = dev->addr_len; | |
3506 | memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len); | |
1da177e4 LT |
3507 | } else { |
3508 | sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */ | |
3509 | sll->sll_halen = 0; | |
3510 | } | |
654d1f8a | 3511 | rcu_read_unlock(); |
1da177e4 | 3512 | |
9b2c45d4 | 3513 | return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen; |
1da177e4 LT |
3514 | } |
3515 | ||
2aeb0b88 WC |
3516 | static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i, |
3517 | int what) | |
1da177e4 LT |
3518 | { |
3519 | switch (i->type) { | |
3520 | case PACKET_MR_MULTICAST: | |
1162563f JP |
3521 | if (i->alen != dev->addr_len) |
3522 | return -EINVAL; | |
1da177e4 | 3523 | if (what > 0) |
22bedad3 | 3524 | return dev_mc_add(dev, i->addr); |
1da177e4 | 3525 | else |
22bedad3 | 3526 | return dev_mc_del(dev, i->addr); |
1da177e4 LT |
3527 | break; |
3528 | case PACKET_MR_PROMISC: | |
2aeb0b88 | 3529 | return dev_set_promiscuity(dev, what); |
1da177e4 | 3530 | case PACKET_MR_ALLMULTI: |
2aeb0b88 | 3531 | return dev_set_allmulti(dev, what); |
d95ed927 | 3532 | case PACKET_MR_UNICAST: |
1162563f JP |
3533 | if (i->alen != dev->addr_len) |
3534 | return -EINVAL; | |
d95ed927 | 3535 | if (what > 0) |
a748ee24 | 3536 | return dev_uc_add(dev, i->addr); |
d95ed927 | 3537 | else |
a748ee24 | 3538 | return dev_uc_del(dev, i->addr); |
d95ed927 | 3539 | break; |
40d4e3df ED |
3540 | default: |
3541 | break; | |
1da177e4 | 3542 | } |
2aeb0b88 | 3543 | return 0; |
1da177e4 LT |
3544 | } |
3545 | ||
82f17091 FR |
3546 | static void packet_dev_mclist_delete(struct net_device *dev, |
3547 | struct packet_mclist **mlp) | |
1da177e4 | 3548 | { |
82f17091 FR |
3549 | struct packet_mclist *ml; |
3550 | ||
3551 | while ((ml = *mlp) != NULL) { | |
3552 | if (ml->ifindex == dev->ifindex) { | |
3553 | packet_dev_mc(dev, ml, -1); | |
3554 | *mlp = ml->next; | |
3555 | kfree(ml); | |
3556 | } else | |
3557 | mlp = &ml->next; | |
1da177e4 LT |
3558 | } |
3559 | } | |
3560 | ||
0fb375fb | 3561 | static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq) |
1da177e4 LT |
3562 | { |
3563 | struct packet_sock *po = pkt_sk(sk); | |
3564 | struct packet_mclist *ml, *i; | |
3565 | struct net_device *dev; | |
3566 | int err; | |
3567 | ||
3568 | rtnl_lock(); | |
3569 | ||
3570 | err = -ENODEV; | |
3b1e0a65 | 3571 | dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex); |
1da177e4 LT |
3572 | if (!dev) |
3573 | goto done; | |
3574 | ||
3575 | err = -EINVAL; | |
1162563f | 3576 | if (mreq->mr_alen > dev->addr_len) |
1da177e4 LT |
3577 | goto done; |
3578 | ||
3579 | err = -ENOBUFS; | |
8b3a7005 | 3580 | i = kmalloc(sizeof(*i), GFP_KERNEL); |
1da177e4 LT |
3581 | if (i == NULL) |
3582 | goto done; | |
3583 | ||
3584 | err = 0; | |
3585 | for (ml = po->mclist; ml; ml = ml->next) { | |
3586 | if (ml->ifindex == mreq->mr_ifindex && | |
3587 | ml->type == mreq->mr_type && | |
3588 | ml->alen == mreq->mr_alen && | |
3589 | memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { | |
3590 | ml->count++; | |
3591 | /* Free the new element ... */ | |
3592 | kfree(i); | |
3593 | goto done; | |
3594 | } | |
3595 | } | |
3596 | ||
3597 | i->type = mreq->mr_type; | |
3598 | i->ifindex = mreq->mr_ifindex; | |
3599 | i->alen = mreq->mr_alen; | |
3600 | memcpy(i->addr, mreq->mr_address, i->alen); | |
309cf37f | 3601 | memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen); |
1da177e4 LT |
3602 | i->count = 1; |
3603 | i->next = po->mclist; | |
3604 | po->mclist = i; | |
2aeb0b88 WC |
3605 | err = packet_dev_mc(dev, i, 1); |
3606 | if (err) { | |
3607 | po->mclist = i->next; | |
3608 | kfree(i); | |
3609 | } | |
1da177e4 LT |
3610 | |
3611 | done: | |
3612 | rtnl_unlock(); | |
3613 | return err; | |
3614 | } | |
3615 | ||
0fb375fb | 3616 | static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq) |
1da177e4 LT |
3617 | { |
3618 | struct packet_mclist *ml, **mlp; | |
3619 | ||
3620 | rtnl_lock(); | |
3621 | ||
3622 | for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) { | |
3623 | if (ml->ifindex == mreq->mr_ifindex && | |
3624 | ml->type == mreq->mr_type && | |
3625 | ml->alen == mreq->mr_alen && | |
3626 | memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) { | |
3627 | if (--ml->count == 0) { | |
3628 | struct net_device *dev; | |
3629 | *mlp = ml->next; | |
ad959e76 ED |
3630 | dev = __dev_get_by_index(sock_net(sk), ml->ifindex); |
3631 | if (dev) | |
1da177e4 | 3632 | packet_dev_mc(dev, ml, -1); |
1da177e4 LT |
3633 | kfree(ml); |
3634 | } | |
82f17091 | 3635 | break; |
1da177e4 LT |
3636 | } |
3637 | } | |
3638 | rtnl_unlock(); | |
82f17091 | 3639 | return 0; |
1da177e4 LT |
3640 | } |
3641 | ||
3642 | static void packet_flush_mclist(struct sock *sk) | |
3643 | { | |
3644 | struct packet_sock *po = pkt_sk(sk); | |
3645 | struct packet_mclist *ml; | |
3646 | ||
3647 | if (!po->mclist) | |
3648 | return; | |
3649 | ||
3650 | rtnl_lock(); | |
3651 | while ((ml = po->mclist) != NULL) { | |
3652 | struct net_device *dev; | |
3653 | ||
3654 | po->mclist = ml->next; | |
ad959e76 ED |
3655 | dev = __dev_get_by_index(sock_net(sk), ml->ifindex); |
3656 | if (dev != NULL) | |
1da177e4 | 3657 | packet_dev_mc(dev, ml, -1); |
1da177e4 LT |
3658 | kfree(ml); |
3659 | } | |
3660 | rtnl_unlock(); | |
3661 | } | |
1da177e4 LT |
3662 | |
3663 | static int | |
b7058842 | 3664 | packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen) |
1da177e4 LT |
3665 | { |
3666 | struct sock *sk = sock->sk; | |
8dc41944 | 3667 | struct packet_sock *po = pkt_sk(sk); |
1da177e4 LT |
3668 | int ret; |
3669 | ||
3670 | if (level != SOL_PACKET) | |
3671 | return -ENOPROTOOPT; | |
3672 | ||
69e3c75f | 3673 | switch (optname) { |
1ce4f28b | 3674 | case PACKET_ADD_MEMBERSHIP: |
1da177e4 LT |
3675 | case PACKET_DROP_MEMBERSHIP: |
3676 | { | |
0fb375fb EB |
3677 | struct packet_mreq_max mreq; |
3678 | int len = optlen; | |
3679 | memset(&mreq, 0, sizeof(mreq)); | |
3680 | if (len < sizeof(struct packet_mreq)) | |
1da177e4 | 3681 | return -EINVAL; |
0fb375fb EB |
3682 | if (len > sizeof(mreq)) |
3683 | len = sizeof(mreq); | |
40d4e3df | 3684 | if (copy_from_user(&mreq, optval, len)) |
1da177e4 | 3685 | return -EFAULT; |
0fb375fb EB |
3686 | if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address))) |
3687 | return -EINVAL; | |
1da177e4 LT |
3688 | if (optname == PACKET_ADD_MEMBERSHIP) |
3689 | ret = packet_mc_add(sk, &mreq); | |
3690 | else | |
3691 | ret = packet_mc_drop(sk, &mreq); | |
3692 | return ret; | |
3693 | } | |
a2efcfa0 | 3694 | |
1da177e4 | 3695 | case PACKET_RX_RING: |
69e3c75f | 3696 | case PACKET_TX_RING: |
1da177e4 | 3697 | { |
f6fb8f10 | 3698 | union tpacket_req_u req_u; |
3699 | int len; | |
1da177e4 | 3700 | |
5171b37d | 3701 | lock_sock(sk); |
f6fb8f10 | 3702 | switch (po->tp_version) { |
3703 | case TPACKET_V1: | |
3704 | case TPACKET_V2: | |
3705 | len = sizeof(req_u.req); | |
3706 | break; | |
3707 | case TPACKET_V3: | |
3708 | default: | |
3709 | len = sizeof(req_u.req3); | |
3710 | break; | |
3711 | } | |
5171b37d ED |
3712 | if (optlen < len) { |
3713 | ret = -EINVAL; | |
3714 | } else { | |
3715 | if (copy_from_user(&req_u.req, optval, len)) | |
3716 | ret = -EFAULT; | |
3717 | else | |
3718 | ret = packet_set_ring(sk, &req_u, 0, | |
3719 | optname == PACKET_TX_RING); | |
3720 | } | |
3721 | release_sock(sk); | |
3722 | return ret; | |
1da177e4 LT |
3723 | } |
3724 | case PACKET_COPY_THRESH: | |
3725 | { | |
3726 | int val; | |
3727 | ||
40d4e3df | 3728 | if (optlen != sizeof(val)) |
1da177e4 | 3729 | return -EINVAL; |
40d4e3df | 3730 | if (copy_from_user(&val, optval, sizeof(val))) |
1da177e4 LT |
3731 | return -EFAULT; |
3732 | ||
3733 | pkt_sk(sk)->copy_thresh = val; | |
3734 | return 0; | |
3735 | } | |
bbd6ef87 PM |
3736 | case PACKET_VERSION: |
3737 | { | |
3738 | int val; | |
3739 | ||
3740 | if (optlen != sizeof(val)) | |
3741 | return -EINVAL; | |
bbd6ef87 PM |
3742 | if (copy_from_user(&val, optval, sizeof(val))) |
3743 | return -EFAULT; | |
3744 | switch (val) { | |
3745 | case TPACKET_V1: | |
3746 | case TPACKET_V2: | |
f6fb8f10 | 3747 | case TPACKET_V3: |
84ac7260 | 3748 | break; |
bbd6ef87 PM |
3749 | default: |
3750 | return -EINVAL; | |
3751 | } | |
84ac7260 PP |
3752 | lock_sock(sk); |
3753 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { | |
3754 | ret = -EBUSY; | |
3755 | } else { | |
3756 | po->tp_version = val; | |
3757 | ret = 0; | |
3758 | } | |
3759 | release_sock(sk); | |
3760 | return ret; | |
bbd6ef87 | 3761 | } |
8913336a PM |
3762 | case PACKET_RESERVE: |
3763 | { | |
3764 | unsigned int val; | |
3765 | ||
3766 | if (optlen != sizeof(val)) | |
3767 | return -EINVAL; | |
8913336a PM |
3768 | if (copy_from_user(&val, optval, sizeof(val))) |
3769 | return -EFAULT; | |
bcc5364b AK |
3770 | if (val > INT_MAX) |
3771 | return -EINVAL; | |
c27927e3 WB |
3772 | lock_sock(sk); |
3773 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { | |
3774 | ret = -EBUSY; | |
3775 | } else { | |
3776 | po->tp_reserve = val; | |
3777 | ret = 0; | |
3778 | } | |
3779 | release_sock(sk); | |
3780 | return ret; | |
8913336a | 3781 | } |
69e3c75f JB |
3782 | case PACKET_LOSS: |
3783 | { | |
3784 | unsigned int val; | |
3785 | ||
3786 | if (optlen != sizeof(val)) | |
3787 | return -EINVAL; | |
69e3c75f JB |
3788 | if (copy_from_user(&val, optval, sizeof(val))) |
3789 | return -EFAULT; | |
a6361f0c WB |
3790 | |
3791 | lock_sock(sk); | |
3792 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { | |
3793 | ret = -EBUSY; | |
3794 | } else { | |
3795 | po->tp_loss = !!val; | |
3796 | ret = 0; | |
3797 | } | |
3798 | release_sock(sk); | |
3799 | return ret; | |
69e3c75f | 3800 | } |
8dc41944 HX |
3801 | case PACKET_AUXDATA: |
3802 | { | |
3803 | int val; | |
3804 | ||
3805 | if (optlen < sizeof(val)) | |
3806 | return -EINVAL; | |
3807 | if (copy_from_user(&val, optval, sizeof(val))) | |
3808 | return -EFAULT; | |
3809 | ||
a6361f0c | 3810 | lock_sock(sk); |
8dc41944 | 3811 | po->auxdata = !!val; |
a6361f0c | 3812 | release_sock(sk); |
8dc41944 HX |
3813 | return 0; |
3814 | } | |
80feaacb PWJ |
3815 | case PACKET_ORIGDEV: |
3816 | { | |
3817 | int val; | |
3818 | ||
3819 | if (optlen < sizeof(val)) | |
3820 | return -EINVAL; | |
3821 | if (copy_from_user(&val, optval, sizeof(val))) | |
3822 | return -EFAULT; | |
3823 | ||
a6361f0c | 3824 | lock_sock(sk); |
80feaacb | 3825 | po->origdev = !!val; |
a6361f0c | 3826 | release_sock(sk); |
80feaacb PWJ |
3827 | return 0; |
3828 | } | |
bfd5f4a3 SS |
3829 | case PACKET_VNET_HDR: |
3830 | { | |
3831 | int val; | |
3832 | ||
3833 | if (sock->type != SOCK_RAW) | |
3834 | return -EINVAL; | |
bfd5f4a3 SS |
3835 | if (optlen < sizeof(val)) |
3836 | return -EINVAL; | |
3837 | if (copy_from_user(&val, optval, sizeof(val))) | |
3838 | return -EFAULT; | |
3839 | ||
a6361f0c WB |
3840 | lock_sock(sk); |
3841 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { | |
3842 | ret = -EBUSY; | |
3843 | } else { | |
3844 | po->has_vnet_hdr = !!val; | |
3845 | ret = 0; | |
3846 | } | |
3847 | release_sock(sk); | |
3848 | return ret; | |
bfd5f4a3 | 3849 | } |
614f60fa SM |
3850 | case PACKET_TIMESTAMP: |
3851 | { | |
3852 | int val; | |
3853 | ||
3854 | if (optlen != sizeof(val)) | |
3855 | return -EINVAL; | |
3856 | if (copy_from_user(&val, optval, sizeof(val))) | |
3857 | return -EFAULT; | |
3858 | ||
3859 | po->tp_tstamp = val; | |
3860 | return 0; | |
3861 | } | |
dc99f600 DM |
3862 | case PACKET_FANOUT: |
3863 | { | |
3864 | int val; | |
3865 | ||
3866 | if (optlen != sizeof(val)) | |
3867 | return -EINVAL; | |
3868 | if (copy_from_user(&val, optval, sizeof(val))) | |
3869 | return -EFAULT; | |
3870 | ||
3871 | return fanout_add(sk, val & 0xffff, val >> 16); | |
3872 | } | |
47dceb8e WB |
3873 | case PACKET_FANOUT_DATA: |
3874 | { | |
3875 | if (!po->fanout) | |
3876 | return -EINVAL; | |
3877 | ||
3878 | return fanout_set_data(po, optval, optlen); | |
3879 | } | |
fa788d98 VW |
3880 | case PACKET_IGNORE_OUTGOING: |
3881 | { | |
3882 | int val; | |
3883 | ||
3884 | if (optlen != sizeof(val)) | |
3885 | return -EINVAL; | |
3886 | if (copy_from_user(&val, optval, sizeof(val))) | |
3887 | return -EFAULT; | |
3888 | if (val < 0 || val > 1) | |
3889 | return -EINVAL; | |
3890 | ||
3891 | po->prot_hook.ignore_outgoing = !!val; | |
3892 | return 0; | |
3893 | } | |
5920cd3a PC |
3894 | case PACKET_TX_HAS_OFF: |
3895 | { | |
3896 | unsigned int val; | |
3897 | ||
3898 | if (optlen != sizeof(val)) | |
3899 | return -EINVAL; | |
5920cd3a PC |
3900 | if (copy_from_user(&val, optval, sizeof(val))) |
3901 | return -EFAULT; | |
a6361f0c WB |
3902 | |
3903 | lock_sock(sk); | |
3904 | if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) { | |
3905 | ret = -EBUSY; | |
3906 | } else { | |
3907 | po->tp_tx_has_off = !!val; | |
3908 | ret = 0; | |
3909 | } | |
3910 | release_sock(sk); | |
5920cd3a PC |
3911 | return 0; |
3912 | } | |
d346a3fa DB |
3913 | case PACKET_QDISC_BYPASS: |
3914 | { | |
3915 | int val; | |
3916 | ||
3917 | if (optlen != sizeof(val)) | |
3918 | return -EINVAL; | |
3919 | if (copy_from_user(&val, optval, sizeof(val))) | |
3920 | return -EFAULT; | |
3921 | ||
3922 | po->xmit = val ? packet_direct_xmit : dev_queue_xmit; | |
3923 | return 0; | |
3924 | } | |
1da177e4 LT |
3925 | default: |
3926 | return -ENOPROTOOPT; | |
3927 | } | |
3928 | } | |
3929 | ||
3930 | static int packet_getsockopt(struct socket *sock, int level, int optname, | |
3931 | char __user *optval, int __user *optlen) | |
3932 | { | |
3933 | int len; | |
c06fff6e | 3934 | int val, lv = sizeof(val); |
1da177e4 LT |
3935 | struct sock *sk = sock->sk; |
3936 | struct packet_sock *po = pkt_sk(sk); | |
c06fff6e | 3937 | void *data = &val; |
ee80fbf3 | 3938 | union tpacket_stats_u st; |
a9b63918 | 3939 | struct tpacket_rollover_stats rstats; |
8e8e2951 | 3940 | int drops; |
1da177e4 LT |
3941 | |
3942 | if (level != SOL_PACKET) | |
3943 | return -ENOPROTOOPT; | |
3944 | ||
8ae55f04 KK |
3945 | if (get_user(len, optlen)) |
3946 | return -EFAULT; | |
1da177e4 LT |
3947 | |
3948 | if (len < 0) | |
3949 | return -EINVAL; | |
1ce4f28b | 3950 | |
69e3c75f | 3951 | switch (optname) { |
1da177e4 | 3952 | case PACKET_STATISTICS: |
1da177e4 | 3953 | spin_lock_bh(&sk->sk_receive_queue.lock); |
ee80fbf3 DB |
3954 | memcpy(&st, &po->stats, sizeof(st)); |
3955 | memset(&po->stats, 0, sizeof(po->stats)); | |
3956 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
8e8e2951 | 3957 | drops = atomic_xchg(&po->tp_drops, 0); |
ee80fbf3 | 3958 | |
f6fb8f10 | 3959 | if (po->tp_version == TPACKET_V3) { |
c06fff6e | 3960 | lv = sizeof(struct tpacket_stats_v3); |
8e8e2951 ED |
3961 | st.stats3.tp_drops = drops; |
3962 | st.stats3.tp_packets += drops; | |
ee80fbf3 | 3963 | data = &st.stats3; |
f6fb8f10 | 3964 | } else { |
c06fff6e | 3965 | lv = sizeof(struct tpacket_stats); |
8e8e2951 ED |
3966 | st.stats1.tp_drops = drops; |
3967 | st.stats1.tp_packets += drops; | |
ee80fbf3 | 3968 | data = &st.stats1; |
f6fb8f10 | 3969 | } |
ee80fbf3 | 3970 | |
8dc41944 HX |
3971 | break; |
3972 | case PACKET_AUXDATA: | |
8dc41944 | 3973 | val = po->auxdata; |
80feaacb PWJ |
3974 | break; |
3975 | case PACKET_ORIGDEV: | |
80feaacb | 3976 | val = po->origdev; |
bfd5f4a3 SS |
3977 | break; |
3978 | case PACKET_VNET_HDR: | |
bfd5f4a3 | 3979 | val = po->has_vnet_hdr; |
1da177e4 | 3980 | break; |
bbd6ef87 | 3981 | case PACKET_VERSION: |
bbd6ef87 | 3982 | val = po->tp_version; |
bbd6ef87 PM |
3983 | break; |
3984 | case PACKET_HDRLEN: | |
3985 | if (len > sizeof(int)) | |
3986 | len = sizeof(int); | |
fd2c83b3 AP |
3987 | if (len < sizeof(int)) |
3988 | return -EINVAL; | |
bbd6ef87 PM |
3989 | if (copy_from_user(&val, optval, len)) |
3990 | return -EFAULT; | |
3991 | switch (val) { | |
3992 | case TPACKET_V1: | |
3993 | val = sizeof(struct tpacket_hdr); | |
3994 | break; | |
3995 | case TPACKET_V2: | |
3996 | val = sizeof(struct tpacket2_hdr); | |
3997 | break; | |
f6fb8f10 | 3998 | case TPACKET_V3: |
3999 | val = sizeof(struct tpacket3_hdr); | |
4000 | break; | |
bbd6ef87 PM |
4001 | default: |
4002 | return -EINVAL; | |
4003 | } | |
bbd6ef87 | 4004 | break; |
8913336a | 4005 | case PACKET_RESERVE: |
8913336a | 4006 | val = po->tp_reserve; |
8913336a | 4007 | break; |
69e3c75f | 4008 | case PACKET_LOSS: |
69e3c75f | 4009 | val = po->tp_loss; |
69e3c75f | 4010 | break; |
614f60fa | 4011 | case PACKET_TIMESTAMP: |
614f60fa | 4012 | val = po->tp_tstamp; |
614f60fa | 4013 | break; |
dc99f600 | 4014 | case PACKET_FANOUT: |
dc99f600 DM |
4015 | val = (po->fanout ? |
4016 | ((u32)po->fanout->id | | |
77f65ebd WB |
4017 | ((u32)po->fanout->type << 16) | |
4018 | ((u32)po->fanout->flags << 24)) : | |
dc99f600 | 4019 | 0); |
dc99f600 | 4020 | break; |
fa788d98 VW |
4021 | case PACKET_IGNORE_OUTGOING: |
4022 | val = po->prot_hook.ignore_outgoing; | |
4023 | break; | |
a9b63918 | 4024 | case PACKET_ROLLOVER_STATS: |
57f015f5 | 4025 | if (!po->rollover) |
a9b63918 | 4026 | return -EINVAL; |
57f015f5 MM |
4027 | rstats.tp_all = atomic_long_read(&po->rollover->num); |
4028 | rstats.tp_huge = atomic_long_read(&po->rollover->num_huge); | |
4029 | rstats.tp_failed = atomic_long_read(&po->rollover->num_failed); | |
4030 | data = &rstats; | |
4031 | lv = sizeof(rstats); | |
a9b63918 | 4032 | break; |
5920cd3a PC |
4033 | case PACKET_TX_HAS_OFF: |
4034 | val = po->tp_tx_has_off; | |
4035 | break; | |
d346a3fa DB |
4036 | case PACKET_QDISC_BYPASS: |
4037 | val = packet_use_direct_xmit(po); | |
4038 | break; | |
1da177e4 LT |
4039 | default: |
4040 | return -ENOPROTOOPT; | |
4041 | } | |
4042 | ||
c06fff6e ED |
4043 | if (len > lv) |
4044 | len = lv; | |
8ae55f04 KK |
4045 | if (put_user(len, optlen)) |
4046 | return -EFAULT; | |
8dc41944 HX |
4047 | if (copy_to_user(optval, data, len)) |
4048 | return -EFAULT; | |
8ae55f04 | 4049 | return 0; |
1da177e4 LT |
4050 | } |
4051 | ||
4052 | ||
719c44d3 WB |
4053 | #ifdef CONFIG_COMPAT |
4054 | static int compat_packet_setsockopt(struct socket *sock, int level, int optname, | |
4055 | char __user *optval, unsigned int optlen) | |
4056 | { | |
4057 | struct packet_sock *po = pkt_sk(sock->sk); | |
4058 | ||
4059 | if (level != SOL_PACKET) | |
4060 | return -ENOPROTOOPT; | |
4061 | ||
4062 | if (optname == PACKET_FANOUT_DATA && | |
4063 | po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) { | |
4064 | optval = (char __user *)get_compat_bpf_fprog(optval); | |
4065 | if (!optval) | |
4066 | return -EFAULT; | |
4067 | optlen = sizeof(struct sock_fprog); | |
4068 | } | |
4069 | ||
4070 | return packet_setsockopt(sock, level, optname, optval, optlen); | |
4071 | } | |
4072 | #endif | |
4073 | ||
351638e7 JP |
4074 | static int packet_notifier(struct notifier_block *this, |
4075 | unsigned long msg, void *ptr) | |
1da177e4 LT |
4076 | { |
4077 | struct sock *sk; | |
351638e7 | 4078 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
c346dca1 | 4079 | struct net *net = dev_net(dev); |
1da177e4 | 4080 | |
808f5114 | 4081 | rcu_read_lock(); |
b67bfe0d | 4082 | sk_for_each_rcu(sk, &net->packet.sklist) { |
1da177e4 LT |
4083 | struct packet_sock *po = pkt_sk(sk); |
4084 | ||
4085 | switch (msg) { | |
4086 | case NETDEV_UNREGISTER: | |
1da177e4 | 4087 | if (po->mclist) |
82f17091 | 4088 | packet_dev_mclist_delete(dev, &po->mclist); |
a2efcfa0 DM |
4089 | /* fallthrough */ |
4090 | ||
1da177e4 LT |
4091 | case NETDEV_DOWN: |
4092 | if (dev->ifindex == po->ifindex) { | |
4093 | spin_lock(&po->bind_lock); | |
4094 | if (po->running) { | |
ce06b03e | 4095 | __unregister_prot_hook(sk, false); |
1da177e4 LT |
4096 | sk->sk_err = ENETDOWN; |
4097 | if (!sock_flag(sk, SOCK_DEAD)) | |
4098 | sk->sk_error_report(sk); | |
4099 | } | |
4100 | if (msg == NETDEV_UNREGISTER) { | |
66e56cd4 | 4101 | packet_cached_dev_reset(po); |
1da177e4 | 4102 | po->ifindex = -1; |
160ff18a BG |
4103 | if (po->prot_hook.dev) |
4104 | dev_put(po->prot_hook.dev); | |
1da177e4 LT |
4105 | po->prot_hook.dev = NULL; |
4106 | } | |
4107 | spin_unlock(&po->bind_lock); | |
4108 | } | |
4109 | break; | |
4110 | case NETDEV_UP: | |
808f5114 | 4111 | if (dev->ifindex == po->ifindex) { |
4112 | spin_lock(&po->bind_lock); | |
ce06b03e DM |
4113 | if (po->num) |
4114 | register_prot_hook(sk); | |
808f5114 | 4115 | spin_unlock(&po->bind_lock); |
1da177e4 | 4116 | } |
1da177e4 LT |
4117 | break; |
4118 | } | |
4119 | } | |
808f5114 | 4120 | rcu_read_unlock(); |
1da177e4 LT |
4121 | return NOTIFY_DONE; |
4122 | } | |
4123 | ||
4124 | ||
4125 | static int packet_ioctl(struct socket *sock, unsigned int cmd, | |
4126 | unsigned long arg) | |
4127 | { | |
4128 | struct sock *sk = sock->sk; | |
4129 | ||
69e3c75f | 4130 | switch (cmd) { |
40d4e3df ED |
4131 | case SIOCOUTQ: |
4132 | { | |
4133 | int amount = sk_wmem_alloc_get(sk); | |
31e6d363 | 4134 | |
40d4e3df ED |
4135 | return put_user(amount, (int __user *)arg); |
4136 | } | |
4137 | case SIOCINQ: | |
4138 | { | |
4139 | struct sk_buff *skb; | |
4140 | int amount = 0; | |
4141 | ||
4142 | spin_lock_bh(&sk->sk_receive_queue.lock); | |
4143 | skb = skb_peek(&sk->sk_receive_queue); | |
4144 | if (skb) | |
4145 | amount = skb->len; | |
4146 | spin_unlock_bh(&sk->sk_receive_queue.lock); | |
4147 | return put_user(amount, (int __user *)arg); | |
4148 | } | |
1da177e4 | 4149 | #ifdef CONFIG_INET |
40d4e3df ED |
4150 | case SIOCADDRT: |
4151 | case SIOCDELRT: | |
4152 | case SIOCDARP: | |
4153 | case SIOCGARP: | |
4154 | case SIOCSARP: | |
4155 | case SIOCGIFADDR: | |
4156 | case SIOCSIFADDR: | |
4157 | case SIOCGIFBRDADDR: | |
4158 | case SIOCSIFBRDADDR: | |
4159 | case SIOCGIFNETMASK: | |
4160 | case SIOCSIFNETMASK: | |
4161 | case SIOCGIFDSTADDR: | |
4162 | case SIOCSIFDSTADDR: | |
4163 | case SIOCSIFFLAGS: | |
40d4e3df | 4164 | return inet_dgram_ops.ioctl(sock, cmd, arg); |
1da177e4 LT |
4165 | #endif |
4166 | ||
40d4e3df ED |
4167 | default: |
4168 | return -ENOIOCTLCMD; | |
1da177e4 LT |
4169 | } |
4170 | return 0; | |
4171 | } | |
4172 | ||
a11e1d43 LT |
4173 | static __poll_t packet_poll(struct file *file, struct socket *sock, |
4174 | poll_table *wait) | |
1da177e4 LT |
4175 | { |
4176 | struct sock *sk = sock->sk; | |
4177 | struct packet_sock *po = pkt_sk(sk); | |
a11e1d43 | 4178 | __poll_t mask = datagram_poll(file, sock, wait); |
1da177e4 LT |
4179 | |
4180 | spin_lock_bh(&sk->sk_receive_queue.lock); | |
69e3c75f | 4181 | if (po->rx_ring.pg_vec) { |
f6fb8f10 | 4182 | if (!packet_previous_rx_frame(po, &po->rx_ring, |
4183 | TP_STATUS_KERNEL)) | |
a9a08845 | 4184 | mask |= EPOLLIN | EPOLLRDNORM; |
1da177e4 | 4185 | } |
9bb6cd65 | 4186 | packet_rcv_try_clear_pressure(po); |
1da177e4 | 4187 | spin_unlock_bh(&sk->sk_receive_queue.lock); |
69e3c75f JB |
4188 | spin_lock_bh(&sk->sk_write_queue.lock); |
4189 | if (po->tx_ring.pg_vec) { | |
4190 | if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE)) | |
a9a08845 | 4191 | mask |= EPOLLOUT | EPOLLWRNORM; |
69e3c75f JB |
4192 | } |
4193 | spin_unlock_bh(&sk->sk_write_queue.lock); | |
1da177e4 LT |
4194 | return mask; |
4195 | } | |
4196 | ||
4197 | ||
4198 | /* Dirty? Well, I still did not learn better way to account | |
4199 | * for user mmaps. | |
4200 | */ | |
4201 | ||
4202 | static void packet_mm_open(struct vm_area_struct *vma) | |
4203 | { | |
4204 | struct file *file = vma->vm_file; | |
40d4e3df | 4205 | struct socket *sock = file->private_data; |
1da177e4 | 4206 | struct sock *sk = sock->sk; |
1ce4f28b | 4207 | |
1da177e4 LT |
4208 | if (sk) |
4209 | atomic_inc(&pkt_sk(sk)->mapped); | |
4210 | } | |
4211 | ||
4212 | static void packet_mm_close(struct vm_area_struct *vma) | |
4213 | { | |
4214 | struct file *file = vma->vm_file; | |
40d4e3df | 4215 | struct socket *sock = file->private_data; |
1da177e4 | 4216 | struct sock *sk = sock->sk; |
1ce4f28b | 4217 | |
1da177e4 LT |
4218 | if (sk) |
4219 | atomic_dec(&pkt_sk(sk)->mapped); | |
4220 | } | |
4221 | ||
f0f37e2f | 4222 | static const struct vm_operations_struct packet_mmap_ops = { |
40d4e3df ED |
4223 | .open = packet_mm_open, |
4224 | .close = packet_mm_close, | |
1da177e4 LT |
4225 | }; |
4226 | ||
3a7ad063 ED |
4227 | static void free_pg_vec(struct pgv *pg_vec, unsigned int order, |
4228 | unsigned int len) | |
1da177e4 LT |
4229 | { |
4230 | int i; | |
4231 | ||
4ebf0ae2 | 4232 | for (i = 0; i < len; i++) { |
0e3125c7 | 4233 | if (likely(pg_vec[i].buffer)) { |
3a7ad063 ED |
4234 | if (is_vmalloc_addr(pg_vec[i].buffer)) |
4235 | vfree(pg_vec[i].buffer); | |
4236 | else | |
4237 | free_pages((unsigned long)pg_vec[i].buffer, | |
4238 | order); | |
0e3125c7 NH |
4239 | pg_vec[i].buffer = NULL; |
4240 | } | |
1da177e4 LT |
4241 | } |
4242 | kfree(pg_vec); | |
4243 | } | |
4244 | ||
3a7ad063 | 4245 | static char *alloc_one_pg_vec_page(unsigned long order) |
4ebf0ae2 | 4246 | { |
f0d4eb29 | 4247 | char *buffer; |
3a7ad063 ED |
4248 | gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | |
4249 | __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY; | |
0e3125c7 | 4250 | |
3a7ad063 | 4251 | buffer = (char *) __get_free_pages(gfp_flags, order); |
0e3125c7 NH |
4252 | if (buffer) |
4253 | return buffer; | |
4254 | ||
3a7ad063 ED |
4255 | /* __get_free_pages failed, fall back to vmalloc */ |
4256 | buffer = vzalloc(array_size((1 << order), PAGE_SIZE)); | |
4257 | if (buffer) | |
4258 | return buffer; | |
0e3125c7 | 4259 | |
3a7ad063 ED |
4260 | /* vmalloc failed, lets dig into swap here */ |
4261 | gfp_flags &= ~__GFP_NORETRY; | |
4262 | buffer = (char *) __get_free_pages(gfp_flags, order); | |
4263 | if (buffer) | |
4264 | return buffer; | |
4265 | ||
4266 | /* complete and utter failure */ | |
4267 | return NULL; | |
4ebf0ae2 DM |
4268 | } |
4269 | ||
3a7ad063 | 4270 | static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order) |
4ebf0ae2 DM |
4271 | { |
4272 | unsigned int block_nr = req->tp_block_nr; | |
0e3125c7 | 4273 | struct pgv *pg_vec; |
4ebf0ae2 DM |
4274 | int i; |
4275 | ||
398f0132 | 4276 | pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN); |
4ebf0ae2 DM |
4277 | if (unlikely(!pg_vec)) |
4278 | goto out; | |
4279 | ||
4280 | for (i = 0; i < block_nr; i++) { | |
3a7ad063 | 4281 | pg_vec[i].buffer = alloc_one_pg_vec_page(order); |
0e3125c7 | 4282 | if (unlikely(!pg_vec[i].buffer)) |
4ebf0ae2 DM |
4283 | goto out_free_pgvec; |
4284 | } | |
4285 | ||
4286 | out: | |
4287 | return pg_vec; | |
4288 | ||
4289 | out_free_pgvec: | |
3a7ad063 | 4290 | free_pg_vec(pg_vec, order, block_nr); |
4ebf0ae2 DM |
4291 | pg_vec = NULL; |
4292 | goto out; | |
4293 | } | |
1da177e4 | 4294 | |
f6fb8f10 | 4295 | static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u, |
69e3c75f | 4296 | int closing, int tx_ring) |
1da177e4 | 4297 | { |
0e3125c7 | 4298 | struct pgv *pg_vec = NULL; |
1da177e4 | 4299 | struct packet_sock *po = pkt_sk(sk); |
f897d759 | 4300 | unsigned long *rx_owner_map = NULL; |
3a7ad063 | 4301 | int was_running, order = 0; |
69e3c75f JB |
4302 | struct packet_ring_buffer *rb; |
4303 | struct sk_buff_head *rb_queue; | |
0e11c91e | 4304 | __be16 num; |
f6fb8f10 | 4305 | int err = -EINVAL; |
4306 | /* Added to avoid minimal code churn */ | |
4307 | struct tpacket_req *req = &req_u->req; | |
4308 | ||
69e3c75f JB |
4309 | rb = tx_ring ? &po->tx_ring : &po->rx_ring; |
4310 | rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue; | |
1da177e4 | 4311 | |
69e3c75f JB |
4312 | err = -EBUSY; |
4313 | if (!closing) { | |
4314 | if (atomic_read(&po->mapped)) | |
4315 | goto out; | |
b0138408 | 4316 | if (packet_read_pending(rb)) |
69e3c75f JB |
4317 | goto out; |
4318 | } | |
1da177e4 | 4319 | |
69e3c75f | 4320 | if (req->tp_block_nr) { |
4576cd46 WB |
4321 | unsigned int min_frame_size; |
4322 | ||
69e3c75f JB |
4323 | /* Sanity tests and some calculations */ |
4324 | err = -EBUSY; | |
4325 | if (unlikely(rb->pg_vec)) | |
4326 | goto out; | |
1da177e4 | 4327 | |
bbd6ef87 PM |
4328 | switch (po->tp_version) { |
4329 | case TPACKET_V1: | |
4330 | po->tp_hdrlen = TPACKET_HDRLEN; | |
4331 | break; | |
4332 | case TPACKET_V2: | |
4333 | po->tp_hdrlen = TPACKET2_HDRLEN; | |
4334 | break; | |
f6fb8f10 | 4335 | case TPACKET_V3: |
4336 | po->tp_hdrlen = TPACKET3_HDRLEN; | |
4337 | break; | |
bbd6ef87 PM |
4338 | } |
4339 | ||
69e3c75f | 4340 | err = -EINVAL; |
4ebf0ae2 | 4341 | if (unlikely((int)req->tp_block_size <= 0)) |
69e3c75f | 4342 | goto out; |
90836b67 | 4343 | if (unlikely(!PAGE_ALIGNED(req->tp_block_size))) |
69e3c75f | 4344 | goto out; |
4576cd46 | 4345 | min_frame_size = po->tp_hdrlen + po->tp_reserve; |
dc808110 | 4346 | if (po->tp_version >= TPACKET_V3 && |
4576cd46 WB |
4347 | req->tp_block_size < |
4348 | BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size) | |
dc808110 | 4349 | goto out; |
4576cd46 | 4350 | if (unlikely(req->tp_frame_size < min_frame_size)) |
69e3c75f | 4351 | goto out; |
4ebf0ae2 | 4352 | if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1))) |
69e3c75f | 4353 | goto out; |
1da177e4 | 4354 | |
4194b491 TK |
4355 | rb->frames_per_block = req->tp_block_size / req->tp_frame_size; |
4356 | if (unlikely(rb->frames_per_block == 0)) | |
69e3c75f | 4357 | goto out; |
fc62814d | 4358 | if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr)) |
8f8d28e4 | 4359 | goto out; |
69e3c75f JB |
4360 | if (unlikely((rb->frames_per_block * req->tp_block_nr) != |
4361 | req->tp_frame_nr)) | |
4362 | goto out; | |
1da177e4 LT |
4363 | |
4364 | err = -ENOMEM; | |
3a7ad063 ED |
4365 | order = get_order(req->tp_block_size); |
4366 | pg_vec = alloc_pg_vec(req, order); | |
4ebf0ae2 | 4367 | if (unlikely(!pg_vec)) |
1da177e4 | 4368 | goto out; |
f6fb8f10 | 4369 | switch (po->tp_version) { |
4370 | case TPACKET_V3: | |
7f953ab2 SV |
4371 | /* Block transmit is not supported yet */ |
4372 | if (!tx_ring) { | |
e8e85cc5 | 4373 | init_prb_bdqc(po, rb, pg_vec, req_u); |
7f953ab2 SV |
4374 | } else { |
4375 | struct tpacket_req3 *req3 = &req_u->req3; | |
4376 | ||
4377 | if (req3->tp_retire_blk_tov || | |
4378 | req3->tp_sizeof_priv || | |
4379 | req3->tp_feature_req_word) { | |
4380 | err = -EINVAL; | |
55655e3d | 4381 | goto out_free_pg_vec; |
7f953ab2 SV |
4382 | } |
4383 | } | |
d7cf0c34 | 4384 | break; |
f6fb8f10 | 4385 | default: |
f897d759 WB |
4386 | if (!tx_ring) { |
4387 | rx_owner_map = bitmap_alloc(req->tp_frame_nr, | |
4388 | GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO); | |
4389 | if (!rx_owner_map) | |
4390 | goto out_free_pg_vec; | |
4391 | } | |
f6fb8f10 | 4392 | break; |
4393 | } | |
69e3c75f JB |
4394 | } |
4395 | /* Done */ | |
4396 | else { | |
4397 | err = -EINVAL; | |
4ebf0ae2 | 4398 | if (unlikely(req->tp_frame_nr)) |
69e3c75f | 4399 | goto out; |
1da177e4 LT |
4400 | } |
4401 | ||
1da177e4 LT |
4402 | |
4403 | /* Detach socket from network */ | |
4404 | spin_lock(&po->bind_lock); | |
4405 | was_running = po->running; | |
4406 | num = po->num; | |
4407 | if (was_running) { | |
414b5431 | 4408 | WRITE_ONCE(po->num, 0); |
ce06b03e | 4409 | __unregister_prot_hook(sk, false); |
1da177e4 LT |
4410 | } |
4411 | spin_unlock(&po->bind_lock); | |
1ce4f28b | 4412 | |
1da177e4 LT |
4413 | synchronize_net(); |
4414 | ||
4415 | err = -EBUSY; | |
905db440 | 4416 | mutex_lock(&po->pg_vec_lock); |
1da177e4 LT |
4417 | if (closing || atomic_read(&po->mapped) == 0) { |
4418 | err = 0; | |
69e3c75f | 4419 | spin_lock_bh(&rb_queue->lock); |
c053fd96 | 4420 | swap(rb->pg_vec, pg_vec); |
f897d759 WB |
4421 | if (po->tp_version <= TPACKET_V2) |
4422 | swap(rb->rx_owner_map, rx_owner_map); | |
69e3c75f JB |
4423 | rb->frame_max = (req->tp_frame_nr - 1); |
4424 | rb->head = 0; | |
4425 | rb->frame_size = req->tp_frame_size; | |
4426 | spin_unlock_bh(&rb_queue->lock); | |
4427 | ||
3a7ad063 | 4428 | swap(rb->pg_vec_order, order); |
c053fd96 | 4429 | swap(rb->pg_vec_len, req->tp_block_nr); |
69e3c75f JB |
4430 | |
4431 | rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE; | |
4432 | po->prot_hook.func = (po->rx_ring.pg_vec) ? | |
4433 | tpacket_rcv : packet_rcv; | |
4434 | skb_queue_purge(rb_queue); | |
1da177e4 | 4435 | if (atomic_read(&po->mapped)) |
40d4e3df ED |
4436 | pr_err("packet_mmap: vma is busy: %d\n", |
4437 | atomic_read(&po->mapped)); | |
1da177e4 | 4438 | } |
905db440 | 4439 | mutex_unlock(&po->pg_vec_lock); |
1da177e4 LT |
4440 | |
4441 | spin_lock(&po->bind_lock); | |
ce06b03e | 4442 | if (was_running) { |
414b5431 | 4443 | WRITE_ONCE(po->num, num); |
ce06b03e | 4444 | register_prot_hook(sk); |
1da177e4 LT |
4445 | } |
4446 | spin_unlock(&po->bind_lock); | |
c800aaf8 | 4447 | if (pg_vec && (po->tp_version > TPACKET_V2)) { |
f6fb8f10 | 4448 | /* Because we don't support block-based V3 on tx-ring */ |
4449 | if (!tx_ring) | |
73d0fcf2 | 4450 | prb_shutdown_retire_blk_timer(po, rb_queue); |
f6fb8f10 | 4451 | } |
1da177e4 | 4452 | |
55655e3d | 4453 | out_free_pg_vec: |
f897d759 | 4454 | bitmap_free(rx_owner_map); |
1da177e4 | 4455 | if (pg_vec) |
3a7ad063 | 4456 | free_pg_vec(pg_vec, order, req->tp_block_nr); |
1da177e4 LT |
4457 | out: |
4458 | return err; | |
4459 | } | |
4460 | ||
69e3c75f JB |
4461 | static int packet_mmap(struct file *file, struct socket *sock, |
4462 | struct vm_area_struct *vma) | |
1da177e4 LT |
4463 | { |
4464 | struct sock *sk = sock->sk; | |
4465 | struct packet_sock *po = pkt_sk(sk); | |
69e3c75f JB |
4466 | unsigned long size, expected_size; |
4467 | struct packet_ring_buffer *rb; | |
1da177e4 LT |
4468 | unsigned long start; |
4469 | int err = -EINVAL; | |
4470 | int i; | |
4471 | ||
4472 | if (vma->vm_pgoff) | |
4473 | return -EINVAL; | |
4474 | ||
905db440 | 4475 | mutex_lock(&po->pg_vec_lock); |
69e3c75f JB |
4476 | |
4477 | expected_size = 0; | |
4478 | for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { | |
4479 | if (rb->pg_vec) { | |
4480 | expected_size += rb->pg_vec_len | |
4481 | * rb->pg_vec_pages | |
4482 | * PAGE_SIZE; | |
4483 | } | |
4484 | } | |
4485 | ||
4486 | if (expected_size == 0) | |
1da177e4 | 4487 | goto out; |
69e3c75f JB |
4488 | |
4489 | size = vma->vm_end - vma->vm_start; | |
4490 | if (size != expected_size) | |
1da177e4 LT |
4491 | goto out; |
4492 | ||
1da177e4 | 4493 | start = vma->vm_start; |
69e3c75f JB |
4494 | for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) { |
4495 | if (rb->pg_vec == NULL) | |
4496 | continue; | |
4497 | ||
4498 | for (i = 0; i < rb->pg_vec_len; i++) { | |
0e3125c7 NH |
4499 | struct page *page; |
4500 | void *kaddr = rb->pg_vec[i].buffer; | |
69e3c75f JB |
4501 | int pg_num; |
4502 | ||
c56b4d90 CG |
4503 | for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) { |
4504 | page = pgv_to_page(kaddr); | |
69e3c75f JB |
4505 | err = vm_insert_page(vma, start, page); |
4506 | if (unlikely(err)) | |
4507 | goto out; | |
4508 | start += PAGE_SIZE; | |
0e3125c7 | 4509 | kaddr += PAGE_SIZE; |
69e3c75f | 4510 | } |
4ebf0ae2 | 4511 | } |
1da177e4 | 4512 | } |
69e3c75f | 4513 | |
4ebf0ae2 | 4514 | atomic_inc(&po->mapped); |
1da177e4 LT |
4515 | vma->vm_ops = &packet_mmap_ops; |
4516 | err = 0; | |
4517 | ||
4518 | out: | |
905db440 | 4519 | mutex_unlock(&po->pg_vec_lock); |
1da177e4 LT |
4520 | return err; |
4521 | } | |
1da177e4 | 4522 | |
90ddc4f0 | 4523 | static const struct proto_ops packet_ops_spkt = { |
1da177e4 LT |
4524 | .family = PF_PACKET, |
4525 | .owner = THIS_MODULE, | |
4526 | .release = packet_release, | |
4527 | .bind = packet_bind_spkt, | |
4528 | .connect = sock_no_connect, | |
4529 | .socketpair = sock_no_socketpair, | |
4530 | .accept = sock_no_accept, | |
4531 | .getname = packet_getname_spkt, | |
a11e1d43 | 4532 | .poll = datagram_poll, |
1da177e4 | 4533 | .ioctl = packet_ioctl, |
c7cbdbf2 | 4534 | .gettstamp = sock_gettstamp, |
1da177e4 LT |
4535 | .listen = sock_no_listen, |
4536 | .shutdown = sock_no_shutdown, | |
4537 | .setsockopt = sock_no_setsockopt, | |
4538 | .getsockopt = sock_no_getsockopt, | |
4539 | .sendmsg = packet_sendmsg_spkt, | |
4540 | .recvmsg = packet_recvmsg, | |
4541 | .mmap = sock_no_mmap, | |
4542 | .sendpage = sock_no_sendpage, | |
4543 | }; | |
1da177e4 | 4544 | |
90ddc4f0 | 4545 | static const struct proto_ops packet_ops = { |
1da177e4 LT |
4546 | .family = PF_PACKET, |
4547 | .owner = THIS_MODULE, | |
4548 | .release = packet_release, | |
4549 | .bind = packet_bind, | |
4550 | .connect = sock_no_connect, | |
4551 | .socketpair = sock_no_socketpair, | |
4552 | .accept = sock_no_accept, | |
1ce4f28b | 4553 | .getname = packet_getname, |
a11e1d43 | 4554 | .poll = packet_poll, |
1da177e4 | 4555 | .ioctl = packet_ioctl, |
c7cbdbf2 | 4556 | .gettstamp = sock_gettstamp, |
1da177e4 LT |
4557 | .listen = sock_no_listen, |
4558 | .shutdown = sock_no_shutdown, | |
4559 | .setsockopt = packet_setsockopt, | |
4560 | .getsockopt = packet_getsockopt, | |
719c44d3 WB |
4561 | #ifdef CONFIG_COMPAT |
4562 | .compat_setsockopt = compat_packet_setsockopt, | |
4563 | #endif | |
1da177e4 LT |
4564 | .sendmsg = packet_sendmsg, |
4565 | .recvmsg = packet_recvmsg, | |
4566 | .mmap = packet_mmap, | |
4567 | .sendpage = sock_no_sendpage, | |
4568 | }; | |
4569 | ||
ec1b4cf7 | 4570 | static const struct net_proto_family packet_family_ops = { |
1da177e4 LT |
4571 | .family = PF_PACKET, |
4572 | .create = packet_create, | |
4573 | .owner = THIS_MODULE, | |
4574 | }; | |
4575 | ||
4576 | static struct notifier_block packet_netdev_notifier = { | |
40d4e3df | 4577 | .notifier_call = packet_notifier, |
1da177e4 LT |
4578 | }; |
4579 | ||
4580 | #ifdef CONFIG_PROC_FS | |
1da177e4 LT |
4581 | |
4582 | static void *packet_seq_start(struct seq_file *seq, loff_t *pos) | |
808f5114 | 4583 | __acquires(RCU) |
1da177e4 | 4584 | { |
e372c414 | 4585 | struct net *net = seq_file_net(seq); |
808f5114 | 4586 | |
4587 | rcu_read_lock(); | |
4588 | return seq_hlist_start_head_rcu(&net->packet.sklist, *pos); | |
1da177e4 LT |
4589 | } |
4590 | ||
4591 | static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
4592 | { | |
1bf40954 | 4593 | struct net *net = seq_file_net(seq); |
808f5114 | 4594 | return seq_hlist_next_rcu(v, &net->packet.sklist, pos); |
1da177e4 LT |
4595 | } |
4596 | ||
4597 | static void packet_seq_stop(struct seq_file *seq, void *v) | |
808f5114 | 4598 | __releases(RCU) |
1da177e4 | 4599 | { |
808f5114 | 4600 | rcu_read_unlock(); |
1da177e4 LT |
4601 | } |
4602 | ||
1ce4f28b | 4603 | static int packet_seq_show(struct seq_file *seq, void *v) |
1da177e4 LT |
4604 | { |
4605 | if (v == SEQ_START_TOKEN) | |
4606 | seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n"); | |
4607 | else { | |
b7ceabd9 | 4608 | struct sock *s = sk_entry(v); |
1da177e4 LT |
4609 | const struct packet_sock *po = pkt_sk(s); |
4610 | ||
4611 | seq_printf(seq, | |
71338aa7 | 4612 | "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n", |
1da177e4 | 4613 | s, |
41c6d650 | 4614 | refcount_read(&s->sk_refcnt), |
1da177e4 | 4615 | s->sk_type, |
414b5431 | 4616 | ntohs(READ_ONCE(po->num)), |
1da177e4 LT |
4617 | po->ifindex, |
4618 | po->running, | |
4619 | atomic_read(&s->sk_rmem_alloc), | |
a7cb5a49 | 4620 | from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)), |
40d4e3df | 4621 | sock_i_ino(s)); |
1da177e4 LT |
4622 | } |
4623 | ||
4624 | return 0; | |
4625 | } | |
4626 | ||
56b3d975 | 4627 | static const struct seq_operations packet_seq_ops = { |
1da177e4 LT |
4628 | .start = packet_seq_start, |
4629 | .next = packet_seq_next, | |
4630 | .stop = packet_seq_stop, | |
4631 | .show = packet_seq_show, | |
4632 | }; | |
1da177e4 LT |
4633 | #endif |
4634 | ||
2c8c1e72 | 4635 | static int __net_init packet_net_init(struct net *net) |
d12d01d6 | 4636 | { |
0fa7fa98 | 4637 | mutex_init(&net->packet.sklist_lock); |
2aaef4e4 | 4638 | INIT_HLIST_HEAD(&net->packet.sklist); |
d12d01d6 | 4639 | |
c3506372 CH |
4640 | if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops, |
4641 | sizeof(struct seq_net_private))) | |
d12d01d6 DL |
4642 | return -ENOMEM; |
4643 | ||
4644 | return 0; | |
4645 | } | |
4646 | ||
2c8c1e72 | 4647 | static void __net_exit packet_net_exit(struct net *net) |
d12d01d6 | 4648 | { |
ece31ffd | 4649 | remove_proc_entry("packet", net->proc_net); |
669f8f1a | 4650 | WARN_ON_ONCE(!hlist_empty(&net->packet.sklist)); |
d12d01d6 DL |
4651 | } |
4652 | ||
4653 | static struct pernet_operations packet_net_ops = { | |
4654 | .init = packet_net_init, | |
4655 | .exit = packet_net_exit, | |
4656 | }; | |
4657 | ||
4658 | ||
1da177e4 LT |
4659 | static void __exit packet_exit(void) |
4660 | { | |
1da177e4 | 4661 | unregister_netdevice_notifier(&packet_netdev_notifier); |
d12d01d6 | 4662 | unregister_pernet_subsys(&packet_net_ops); |
1da177e4 LT |
4663 | sock_unregister(PF_PACKET); |
4664 | proto_unregister(&packet_proto); | |
4665 | } | |
4666 | ||
4667 | static int __init packet_init(void) | |
4668 | { | |
36096f2f | 4669 | int rc; |
1da177e4 | 4670 | |
36096f2f Y |
4671 | rc = proto_register(&packet_proto, 0); |
4672 | if (rc) | |
1da177e4 | 4673 | goto out; |
36096f2f Y |
4674 | rc = sock_register(&packet_family_ops); |
4675 | if (rc) | |
4676 | goto out_proto; | |
4677 | rc = register_pernet_subsys(&packet_net_ops); | |
4678 | if (rc) | |
4679 | goto out_sock; | |
4680 | rc = register_netdevice_notifier(&packet_netdev_notifier); | |
4681 | if (rc) | |
4682 | goto out_pernet; | |
1da177e4 | 4683 | |
36096f2f Y |
4684 | return 0; |
4685 | ||
4686 | out_pernet: | |
4687 | unregister_pernet_subsys(&packet_net_ops); | |
4688 | out_sock: | |
4689 | sock_unregister(PF_PACKET); | |
4690 | out_proto: | |
4691 | proto_unregister(&packet_proto); | |
1da177e4 LT |
4692 | out: |
4693 | return rc; | |
4694 | } | |
4695 | ||
4696 | module_init(packet_init); | |
4697 | module_exit(packet_exit); | |
4698 | MODULE_LICENSE("GPL"); | |
4699 | MODULE_ALIAS_NETPROTO(PF_PACKET); |