]>
Commit | Line | Data |
---|---|---|
9d3616dc AF |
1 | diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig |
2 | index 494b888..851e01f 100644 | |
3 | --- a/drivers/net/Kconfig | |
4 | +++ b/drivers/net/Kconfig | |
5 | @@ -203,6 +203,125 @@ config RIONET_RX_SIZE | |
6 | depends on RIONET | |
7 | default "128" | |
8 | ||
9 | +config IMQ | |
10 | + tristate "IMQ (intermediate queueing device) support" | |
11 | + depends on NETDEVICES && NETFILTER | |
12 | + ---help--- | |
13 | + The IMQ device(s) is used as placeholder for QoS queueing | |
14 | + disciplines. Every packet entering/leaving the IP stack can be | |
15 | + directed through the IMQ device where it's enqueued/dequeued to the | |
16 | + attached qdisc. This allows you to treat network devices as classes | |
17 | + and distribute bandwidth among them. Iptables is used to specify | |
18 | + through which IMQ device, if any, packets travel. | |
19 | + | |
20 | + More information at: http://www.linuximq.net/ | |
21 | + | |
22 | + To compile this driver as a module, choose M here: the module | |
23 | + will be called imq. If unsure, say N. | |
24 | + | |
25 | +choice | |
26 | + prompt "IMQ behavior (PRE/POSTROUTING)" | |
27 | + depends on IMQ | |
28 | + default IMQ_BEHAVIOR_AB | |
29 | + help | |
30 | + This setting defines how IMQ behaves in respect to its | |
31 | + hooking in PREROUTING and POSTROUTING. | |
32 | + | |
33 | + IMQ can work in any of the following ways: | |
34 | + | |
35 | + PREROUTING | POSTROUTING | |
36 | + -----------------|------------------- | |
37 | + #1 After NAT | After NAT | |
38 | + #2 After NAT | Before NAT | |
39 | + #3 Before NAT | After NAT | |
40 | + #4 Before NAT | Before NAT | |
41 | + | |
42 | + The default behavior is to hook before NAT on PREROUTING | |
43 | + and after NAT on POSTROUTING (#3). | |
44 | + | |
45 | + This settings are specially usefull when trying to use IMQ | |
46 | + to shape NATed clients. | |
47 | + | |
48 | + More information can be found at: www.linuximq.net | |
49 | + | |
50 | + If not sure leave the default settings alone. | |
51 | + | |
52 | +config IMQ_BEHAVIOR_AA | |
53 | + bool "IMQ AA" | |
54 | + help | |
55 | + This setting defines how IMQ behaves in respect to its | |
56 | + hooking in PREROUTING and POSTROUTING. | |
57 | + | |
58 | + Choosing this option will make IMQ hook like this: | |
59 | + | |
60 | + PREROUTING: After NAT | |
61 | + POSTROUTING: After NAT | |
62 | + | |
63 | + More information can be found at: www.linuximq.net | |
64 | + | |
65 | + If not sure leave the default settings alone. | |
66 | + | |
67 | +config IMQ_BEHAVIOR_AB | |
68 | + bool "IMQ AB" | |
69 | + help | |
70 | + This setting defines how IMQ behaves in respect to its | |
71 | + hooking in PREROUTING and POSTROUTING. | |
72 | + | |
73 | + Choosing this option will make IMQ hook like this: | |
74 | + | |
75 | + PREROUTING: After NAT | |
76 | + POSTROUTING: Before NAT | |
77 | + | |
78 | + More information can be found at: www.linuximq.net | |
79 | + | |
80 | + If not sure leave the default settings alone. | |
81 | + | |
82 | +config IMQ_BEHAVIOR_BA | |
83 | + bool "IMQ BA" | |
84 | + help | |
85 | + This setting defines how IMQ behaves in respect to its | |
86 | + hooking in PREROUTING and POSTROUTING. | |
87 | + | |
88 | + Choosing this option will make IMQ hook like this: | |
89 | + | |
90 | + PREROUTING: Before NAT | |
91 | + POSTROUTING: After NAT | |
92 | + | |
93 | + More information can be found at: www.linuximq.net | |
94 | + | |
95 | + If not sure leave the default settings alone. | |
96 | + | |
97 | +config IMQ_BEHAVIOR_BB | |
98 | + bool "IMQ BB" | |
99 | + help | |
100 | + This setting defines how IMQ behaves in respect to its | |
101 | + hooking in PREROUTING and POSTROUTING. | |
102 | + | |
103 | + Choosing this option will make IMQ hook like this: | |
104 | + | |
105 | + PREROUTING: Before NAT | |
106 | + POSTROUTING: Before NAT | |
107 | + | |
108 | + More information can be found at: www.linuximq.net | |
109 | + | |
110 | + If not sure leave the default settings alone. | |
111 | + | |
112 | +endchoice | |
113 | + | |
114 | +config IMQ_NUM_DEVS | |
115 | + int "Number of IMQ devices" | |
116 | + range 2 16 | |
117 | + depends on IMQ | |
118 | + default "16" | |
119 | + help | |
120 | + This setting defines how many IMQ devices will be created. | |
121 | + | |
122 | + The default value is 16. | |
123 | + | |
124 | + More information can be found at: www.linuximq.net | |
125 | + | |
126 | + If not sure leave the default settings alone. | |
127 | + | |
128 | config TUN | |
129 | tristate "Universal TUN/TAP device driver support" | |
130 | select CRC32 | |
131 | diff --git a/drivers/net/Makefile b/drivers/net/Makefile | |
132 | index 3fef8a8..12dafc0 100644 | |
133 | --- a/drivers/net/Makefile | |
134 | +++ b/drivers/net/Makefile | |
135 | @@ -9,6 +9,7 @@ obj-$(CONFIG_BONDING) += bonding/ | |
136 | obj-$(CONFIG_DUMMY) += dummy.o | |
137 | obj-$(CONFIG_EQUALIZER) += eql.o | |
138 | obj-$(CONFIG_IFB) += ifb.o | |
139 | +obj-$(CONFIG_IMQ) += imq.o | |
140 | obj-$(CONFIG_MACVLAN) += macvlan.o | |
141 | obj-$(CONFIG_MACVTAP) += macvtap.o | |
142 | obj-$(CONFIG_MII) += mii.o | |
143 | diff --git a/drivers/net/imq.c b/drivers/net/imq.c | |
144 | new file mode 100644 | |
145 | index 0000000..2140535 | |
146 | --- /dev/null | |
147 | +++ b/drivers/net/imq.c | |
148 | @@ -0,0 +1,1011 @@ | |
149 | +/* | |
150 | + * Pseudo-driver for the intermediate queue device. | |
151 | + * | |
152 | + * This program is free software; you can redistribute it and/or | |
153 | + * modify it under the terms of the GNU General Public License | |
154 | + * as published by the Free Software Foundation; either version | |
155 | + * 2 of the License, or (at your option) any later version. | |
156 | + * | |
157 | + * Authors: Patrick McHardy, <kaber@trash.net> | |
158 | + * | |
159 | + * The first version was written by Martin Devera, <devik@cdi.cz> | |
160 | + * | |
161 | + * Credits: Jan Rafaj <imq2t@cedric.vabo.cz> | |
162 | + * - Update patch to 2.4.21 | |
163 | + * Sebastian Strollo <sstrollo@nortelnetworks.com> | |
164 | + * - Fix "Dead-loop on netdevice imq"-issue | |
165 | + * Marcel Sebek <sebek64@post.cz> | |
166 | + * - Update to 2.6.2-rc1 | |
167 | + * | |
168 | + * After some time of inactivity there is a group taking care | |
169 | + * of IMQ again: http://www.linuximq.net | |
170 | + * | |
171 | + * | |
172 | + * 2004/06/30 - New version of IMQ patch to kernels <=2.6.7 | |
173 | + * including the following changes: | |
174 | + * | |
175 | + * - Correction of ipv6 support "+"s issue (Hasso Tepper) | |
176 | + * - Correction of imq_init_devs() issue that resulted in | |
177 | + * kernel OOPS unloading IMQ as module (Norbert Buchmuller) | |
178 | + * - Addition of functionality to choose number of IMQ devices | |
179 | + * during kernel config (Andre Correa) | |
180 | + * - Addition of functionality to choose how IMQ hooks on | |
181 | + * PRE and POSTROUTING (after or before NAT) (Andre Correa) | |
182 | + * - Cosmetic corrections (Norbert Buchmuller) (Andre Correa) | |
183 | + * | |
184 | + * | |
185 | + * 2005/12/16 - IMQ versions between 2.6.7 and 2.6.13 were | |
186 | + * released with almost no problems. 2.6.14-x was released | |
187 | + * with some important changes: nfcache was removed; After | |
188 | + * some weeks of trouble we figured out that some IMQ fields | |
189 | + * in skb were missing in skbuff.c - skb_clone and copy_skb_header. | |
190 | + * These functions are correctly patched by this new patch version. | |
191 | + * | |
192 | + * Thanks for all who helped to figure out all the problems with | |
193 | + * 2.6.14.x: Patrick McHardy, Rune Kock, VeNoMouS, Max CtRiX, | |
194 | + * Kevin Shanahan, Richard Lucassen, Valery Dachev (hopefully | |
195 | + * I didn't forget anybody). I apologize again for my lack of time. | |
196 | + * | |
197 | + * | |
198 | + * 2008/06/17 - 2.6.25 - Changed imq.c to use qdisc_run() instead | |
199 | + * of qdisc_restart() and moved qdisc_run() to tasklet to avoid | |
200 | + * recursive locking. New initialization routines to fix 'rmmod' not | |
201 | + * working anymore. Used code from ifb.c. (Jussi Kivilinna) | |
202 | + * | |
203 | + * 2008/08/06 - 2.6.26 - (JK) | |
204 | + * - Replaced tasklet with 'netif_schedule()'. | |
205 | + * - Cleaned up and added comments for imq_nf_queue(). | |
206 | + * | |
207 | + * 2009/04/12 | |
208 | + * - Add skb_save_cb/skb_restore_cb helper functions for backuping | |
209 | + * control buffer. This is needed because qdisc-layer on kernels | |
210 | + * 2.6.27 and newer overwrite control buffer. (Jussi Kivilinna) | |
211 | + * - Add better locking for IMQ device. Hopefully this will solve | |
212 | + * SMP issues. (Jussi Kivilinna) | |
213 | + * - Port to 2.6.27 | |
214 | + * - Port to 2.6.28 | |
215 | + * - Port to 2.6.29 + fix rmmod not working | |
216 | + * | |
217 | + * 2009/04/20 - (Jussi Kivilinna) | |
218 | + * - Use netdevice feature flags to avoid extra packet handling | |
219 | + * by core networking layer and possibly increase performance. | |
220 | + * | |
221 | + * 2009/09/26 - (Jussi Kivilinna) | |
222 | + * - Add imq_nf_reinject_lockless to fix deadlock with | |
223 | + * imq_nf_queue/imq_nf_reinject. | |
224 | + * | |
225 | + * 2009/12/08 - (Jussi Kivilinna) | |
226 | + * - Port to 2.6.32 | |
227 | + * - Add check for skb->nf_queue_entry==NULL in imq_dev_xmit() | |
228 | + * - Also add better error checking for skb->nf_queue_entry usage | |
229 | + * | |
230 | + * 2010/02/25 - (Jussi Kivilinna) | |
231 | + * - Port to 2.6.33 | |
232 | + * | |
233 | + * 2010/08/15 - (Jussi Kivilinna) | |
234 | + * - Port to 2.6.35 | |
235 | + * - Simplify hook registration by using nf_register_hooks. | |
236 | + * - nf_reinject doesn't need spinlock around it, therefore remove | |
237 | + * imq_nf_reinject function. Other nf_reinject users protect | |
238 | + * their own data with spinlock. With IMQ however all data is | |
239 | + * needed is stored per skbuff, so no locking is needed. | |
240 | + * - Changed IMQ to use 'separate' NF_IMQ_QUEUE instead of | |
241 | + * NF_QUEUE, this allows working coexistance of IMQ and other | |
242 | + * NF_QUEUE users. | |
243 | + * - Make IMQ multi-queue. Number of IMQ device queues can be | |
244 | + * increased with 'numqueues' module parameters. Default number | |
245 | + * of queues is 1, in other words by default IMQ works as | |
246 | + * single-queue device. Multi-queue selection is based on | |
247 | + * IFB multi-queue patch by Changli Gao <xiaosuo@gmail.com>. | |
248 | + * | |
249 | + * 2011/03/18 - (Jussi Kivilinna) | |
250 | + * - Port to 2.6.38 | |
251 | + * | |
252 | + * 2011/07/12 - (syoder89@gmail.com) | |
253 | + * - Crash fix that happens when the receiving interface has more | |
254 | + * than one queue (add missing skb_set_queue_mapping in | |
255 | + * imq_select_queue). | |
256 | + * | |
257 | + * 2011/07/26 - (Jussi Kivilinna) | |
258 | + * - Add queue mapping checks for packets exiting IMQ. | |
259 | + * - Port to 3.0 | |
260 | + * | |
261 | + * 2011/08/16 - (Jussi Kivilinna) | |
262 | + * - Clear IFF_TX_SKB_SHARING flag that was added for linux 3.0.2 | |
263 | + * | |
264 | + * 2011/11/03 - Germano Michel <germanomichel@gmail.com> | |
265 | + * - Fix IMQ for net namespaces | |
266 | + * | |
267 | + * 2011/11/04 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi> | |
268 | + * - Port to 3.1 | |
269 | + * - Clean-up, move 'get imq device pointer by imqX name' to | |
270 | + * separate function from imq_nf_queue(). | |
271 | + * | |
272 | + * 2012/01/05 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi> | |
273 | + * - Port to 3.2 | |
274 | + * | |
275 | + * 2012/03/19 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi> | |
276 | + * - Port to 3.3 | |
277 | + * | |
278 | + * 2012/12/12 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi> | |
279 | + * - Port to 3.7 | |
280 | + * - Fix checkpatch.pl warnings | |
281 | + * | |
282 | + * 2013/09/10 - Jussi Kivilinna <jussi.kivilinna@iki.fi> | |
283 | + * - Fixed GSO handling for 3.10, see imq_nf_queue() for comments. | |
284 | + * - Don't copy skb->cb_next when copying or cloning skbuffs. | |
285 | + * | |
286 | + * 2013/09/16 - Jussi Kivilinna <jussi.kivilinna@iki.fi> | |
287 | + * - Port to 3.11 | |
288 | + * | |
289 | + * 2013/11/12 - Jussi Kivilinna <jussi.kivilinna@iki.fi> | |
290 | + * - Port to 3.12 | |
291 | + * | |
292 | + * 2014/02/07 - Jussi Kivilinna <jussi.kivilinna@iki.fi> | |
293 | + * - Port to 3.13 | |
294 | + * | |
295 | + * Also, many thanks to pablo Sebastian Greco for making the initial | |
296 | + * patch and to those who helped the testing. | |
297 | + * | |
298 | + * More info at: http://www.linuximq.net/ (Andre Correa) | |
299 | + */ | |
300 | + | |
301 | +#include <linux/module.h> | |
302 | +#include <linux/kernel.h> | |
303 | +#include <linux/moduleparam.h> | |
304 | +#include <linux/list.h> | |
305 | +#include <linux/skbuff.h> | |
306 | +#include <linux/netdevice.h> | |
307 | +#include <linux/etherdevice.h> | |
308 | +#include <linux/rtnetlink.h> | |
309 | +#include <linux/if_arp.h> | |
310 | +#include <linux/netfilter.h> | |
311 | +#include <linux/netfilter_ipv4.h> | |
312 | +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
313 | + #include <linux/netfilter_ipv6.h> | |
314 | +#endif | |
315 | +#include <linux/imq.h> | |
316 | +#include <net/pkt_sched.h> | |
317 | +#include <net/netfilter/nf_queue.h> | |
318 | +#include <net/sock.h> | |
319 | +#include <linux/ip.h> | |
320 | +#include <linux/ipv6.h> | |
321 | +#include <linux/if_vlan.h> | |
322 | +#include <linux/if_pppox.h> | |
323 | +#include <net/ip.h> | |
324 | +#include <net/ipv6.h> | |
325 | + | |
326 | +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num); | |
327 | + | |
328 | +static nf_hookfn imq_nf_hook; | |
329 | + | |
330 | +static struct nf_hook_ops imq_ops[] = { | |
331 | + { | |
332 | + /* imq_ingress_ipv4 */ | |
333 | + .hook = imq_nf_hook, | |
334 | + .owner = THIS_MODULE, | |
335 | + .pf = PF_INET, | |
336 | + .hooknum = NF_INET_PRE_ROUTING, | |
337 | +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB) | |
338 | + .priority = NF_IP_PRI_MANGLE + 1, | |
339 | +#else | |
340 | + .priority = NF_IP_PRI_NAT_DST + 1, | |
341 | +#endif | |
342 | + }, | |
343 | + { | |
344 | + /* imq_egress_ipv4 */ | |
345 | + .hook = imq_nf_hook, | |
346 | + .owner = THIS_MODULE, | |
347 | + .pf = PF_INET, | |
348 | + .hooknum = NF_INET_POST_ROUTING, | |
349 | +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA) | |
350 | + .priority = NF_IP_PRI_LAST, | |
351 | +#else | |
352 | + .priority = NF_IP_PRI_NAT_SRC - 1, | |
353 | +#endif | |
354 | + }, | |
355 | +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
356 | + { | |
357 | + /* imq_ingress_ipv6 */ | |
358 | + .hook = imq_nf_hook, | |
359 | + .owner = THIS_MODULE, | |
360 | + .pf = PF_INET6, | |
361 | + .hooknum = NF_INET_PRE_ROUTING, | |
362 | +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB) | |
363 | + .priority = NF_IP6_PRI_MANGLE + 1, | |
364 | +#else | |
365 | + .priority = NF_IP6_PRI_NAT_DST + 1, | |
366 | +#endif | |
367 | + }, | |
368 | + { | |
369 | + /* imq_egress_ipv6 */ | |
370 | + .hook = imq_nf_hook, | |
371 | + .owner = THIS_MODULE, | |
372 | + .pf = PF_INET6, | |
373 | + .hooknum = NF_INET_POST_ROUTING, | |
374 | +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA) | |
375 | + .priority = NF_IP6_PRI_LAST, | |
376 | +#else | |
377 | + .priority = NF_IP6_PRI_NAT_SRC - 1, | |
378 | +#endif | |
379 | + }, | |
380 | +#endif | |
381 | +}; | |
382 | + | |
383 | +#if defined(CONFIG_IMQ_NUM_DEVS) | |
384 | +static int numdevs = CONFIG_IMQ_NUM_DEVS; | |
385 | +#else | |
386 | +static int numdevs = IMQ_MAX_DEVS; | |
387 | +#endif | |
388 | + | |
389 | +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS]; | |
390 | + | |
391 | +#define IMQ_MAX_QUEUES 32 | |
392 | +static int numqueues = 1; | |
393 | +static u32 imq_hashrnd; | |
394 | + | |
395 | +static inline __be16 pppoe_proto(const struct sk_buff *skb) | |
396 | +{ | |
397 | + return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN + | |
398 | + sizeof(struct pppoe_hdr))); | |
399 | +} | |
400 | + | |
401 | +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb) | |
402 | +{ | |
403 | + unsigned int pull_len; | |
404 | + u16 protocol = skb->protocol; | |
405 | + u32 addr1, addr2; | |
406 | + u32 hash, ihl = 0; | |
407 | + union { | |
408 | + u16 in16[2]; | |
409 | + u32 in32; | |
410 | + } ports; | |
411 | + u8 ip_proto; | |
412 | + | |
413 | + pull_len = 0; | |
414 | + | |
415 | +recheck: | |
416 | + switch (protocol) { | |
417 | + case htons(ETH_P_8021Q): { | |
418 | + if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL)) | |
419 | + goto other; | |
420 | + | |
421 | + pull_len += VLAN_HLEN; | |
422 | + skb->network_header += VLAN_HLEN; | |
423 | + | |
424 | + protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto; | |
425 | + goto recheck; | |
426 | + } | |
427 | + | |
428 | + case htons(ETH_P_PPP_SES): { | |
429 | + if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL)) | |
430 | + goto other; | |
431 | + | |
432 | + pull_len += PPPOE_SES_HLEN; | |
433 | + skb->network_header += PPPOE_SES_HLEN; | |
434 | + | |
435 | + protocol = pppoe_proto(skb); | |
436 | + goto recheck; | |
437 | + } | |
438 | + | |
439 | + case htons(ETH_P_IP): { | |
440 | + const struct iphdr *iph = ip_hdr(skb); | |
441 | + | |
442 | + if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr)))) | |
443 | + goto other; | |
444 | + | |
445 | + addr1 = iph->daddr; | |
446 | + addr2 = iph->saddr; | |
447 | + | |
448 | + ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ? | |
449 | + iph->protocol : 0; | |
450 | + ihl = ip_hdrlen(skb); | |
451 | + | |
452 | + break; | |
453 | + } | |
454 | +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
455 | + case htons(ETH_P_IPV6): { | |
456 | + const struct ipv6hdr *iph = ipv6_hdr(skb); | |
457 | + __be16 fo = 0; | |
458 | + | |
459 | + if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr)))) | |
460 | + goto other; | |
461 | + | |
462 | + addr1 = iph->daddr.s6_addr32[3]; | |
463 | + addr2 = iph->saddr.s6_addr32[3]; | |
464 | + ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto, | |
465 | + &fo); | |
466 | + if (unlikely(ihl < 0)) | |
467 | + goto other; | |
468 | + | |
469 | + break; | |
470 | + } | |
471 | +#endif | |
472 | + default: | |
473 | +other: | |
474 | + if (pull_len != 0) { | |
475 | + skb_push(skb, pull_len); | |
476 | + skb->network_header -= pull_len; | |
477 | + } | |
478 | + | |
479 | + return (u16)(ntohs(protocol) % dev->real_num_tx_queues); | |
480 | + } | |
481 | + | |
482 | + if (addr1 > addr2) | |
483 | + swap(addr1, addr2); | |
484 | + | |
485 | + switch (ip_proto) { | |
486 | + case IPPROTO_TCP: | |
487 | + case IPPROTO_UDP: | |
488 | + case IPPROTO_DCCP: | |
489 | + case IPPROTO_ESP: | |
490 | + case IPPROTO_AH: | |
491 | + case IPPROTO_SCTP: | |
492 | + case IPPROTO_UDPLITE: { | |
493 | + if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) { | |
494 | + if (ports.in16[0] > ports.in16[1]) | |
495 | + swap(ports.in16[0], ports.in16[1]); | |
496 | + break; | |
497 | + } | |
498 | + /* fall-through */ | |
499 | + } | |
500 | + default: | |
501 | + ports.in32 = 0; | |
502 | + break; | |
503 | + } | |
504 | + | |
505 | + if (pull_len != 0) { | |
506 | + skb_push(skb, pull_len); | |
507 | + skb->network_header -= pull_len; | |
508 | + } | |
509 | + | |
510 | + hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto); | |
511 | + | |
512 | + return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32); | |
513 | +} | |
514 | + | |
515 | +static inline bool sk_tx_queue_recorded(struct sock *sk) | |
516 | +{ | |
517 | + return (sk_tx_queue_get(sk) >= 0); | |
518 | +} | |
519 | + | |
520 | +static struct netdev_queue *imq_select_queue(struct net_device *dev, | |
521 | + struct sk_buff *skb) | |
522 | +{ | |
523 | + u16 queue_index = 0; | |
524 | + u32 hash; | |
525 | + | |
526 | + if (likely(dev->real_num_tx_queues == 1)) | |
527 | + goto out; | |
528 | + | |
529 | + /* IMQ can be receiving ingress or engress packets. */ | |
530 | + | |
531 | + /* Check first for if rx_queue is set */ | |
532 | + if (skb_rx_queue_recorded(skb)) { | |
533 | + queue_index = skb_get_rx_queue(skb); | |
534 | + goto out; | |
535 | + } | |
536 | + | |
537 | + /* Check if socket has tx_queue set */ | |
538 | + if (sk_tx_queue_recorded(skb->sk)) { | |
539 | + queue_index = sk_tx_queue_get(skb->sk); | |
540 | + goto out; | |
541 | + } | |
542 | + | |
543 | + /* Try use socket hash */ | |
544 | + if (skb->sk && skb->sk->sk_hash) { | |
545 | + hash = skb->sk->sk_hash; | |
546 | + queue_index = | |
547 | + (u16)(((u64)hash * dev->real_num_tx_queues) >> 32); | |
548 | + goto out; | |
549 | + } | |
550 | + | |
551 | + /* Generate hash from packet data */ | |
552 | + queue_index = imq_hash(dev, skb); | |
553 | + | |
554 | +out: | |
555 | + if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
556 | + queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues); | |
557 | + | |
558 | + skb_set_queue_mapping(skb, queue_index); | |
559 | + return netdev_get_tx_queue(dev, queue_index); | |
560 | +} | |
561 | + | |
562 | +static struct net_device_stats *imq_get_stats(struct net_device *dev) | |
563 | +{ | |
564 | + return &dev->stats; | |
565 | +} | |
566 | + | |
567 | +/* called for packets kfree'd in qdiscs at places other than enqueue */ | |
568 | +static void imq_skb_destructor(struct sk_buff *skb) | |
569 | +{ | |
570 | + struct nf_queue_entry *entry = skb->nf_queue_entry; | |
571 | + | |
572 | + skb->nf_queue_entry = NULL; | |
573 | + | |
574 | + if (entry) { | |
575 | + nf_queue_entry_release_refs(entry); | |
576 | + kfree(entry); | |
577 | + } | |
578 | + | |
579 | + skb_restore_cb(skb); /* kfree backup */ | |
580 | +} | |
581 | + | |
582 | +static void imq_done_check_queue_mapping(struct sk_buff *skb, | |
583 | + struct net_device *dev) | |
584 | +{ | |
585 | + unsigned int queue_index; | |
586 | + | |
587 | + /* Don't let queue_mapping be left too large after exiting IMQ */ | |
588 | + if (likely(skb->dev != dev && skb->dev != NULL)) { | |
589 | + queue_index = skb_get_queue_mapping(skb); | |
590 | + if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) { | |
591 | + queue_index = (u16)((u32)queue_index % | |
592 | + skb->dev->real_num_tx_queues); | |
593 | + skb_set_queue_mapping(skb, queue_index); | |
594 | + } | |
595 | + } else { | |
596 | + /* skb->dev was IMQ device itself or NULL, be on safe side and | |
597 | + * just clear queue mapping. | |
598 | + */ | |
599 | + skb_set_queue_mapping(skb, 0); | |
600 | + } | |
601 | +} | |
602 | + | |
603 | +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev) | |
604 | +{ | |
605 | + struct nf_queue_entry *entry = skb->nf_queue_entry; | |
606 | + | |
607 | + skb->nf_queue_entry = NULL; | |
608 | + dev->trans_start = jiffies; | |
609 | + | |
610 | + dev->stats.tx_bytes += skb->len; | |
611 | + dev->stats.tx_packets++; | |
612 | + | |
613 | + if (unlikely(entry == NULL)) { | |
614 | + /* We don't know what is going on here.. packet is queued for | |
615 | + * imq device, but (probably) not by us. | |
616 | + * | |
617 | + * If this packet was not send here by imq_nf_queue(), then | |
618 | + * skb_save_cb() was not used and skb_free() should not show: | |
619 | + * WARNING: IMQ: kfree_skb: skb->cb_next:.. | |
620 | + * and/or | |
621 | + * WARNING: IMQ: kfree_skb: skb->nf_queue_entry... | |
622 | + * | |
623 | + * However if this message is shown, then IMQ is somehow broken | |
624 | + * and you should report this to linuximq.net. | |
625 | + */ | |
626 | + | |
627 | + /* imq_dev_xmit is black hole that eats all packets, report that | |
628 | + * we eat this packet happily and increase dropped counters. | |
629 | + */ | |
630 | + | |
631 | + dev->stats.tx_dropped++; | |
632 | + dev_kfree_skb(skb); | |
633 | + | |
634 | + return NETDEV_TX_OK; | |
635 | + } | |
636 | + | |
637 | + skb_restore_cb(skb); /* restore skb->cb */ | |
638 | + | |
639 | + skb->imq_flags = 0; | |
640 | + skb->destructor = NULL; | |
641 | + | |
642 | + imq_done_check_queue_mapping(skb, dev); | |
643 | + | |
644 | + nf_reinject(entry, NF_ACCEPT); | |
645 | + | |
646 | + return NETDEV_TX_OK; | |
647 | +} | |
648 | + | |
649 | +static struct net_device *get_imq_device_by_index(int index) | |
650 | +{ | |
651 | + struct net_device *dev = NULL; | |
652 | + struct net *net; | |
653 | + char buf[8]; | |
654 | + | |
655 | + /* get device by name and cache result */ | |
656 | + snprintf(buf, sizeof(buf), "imq%d", index); | |
657 | + | |
658 | + /* Search device from all namespaces. */ | |
659 | + for_each_net(net) { | |
660 | + dev = dev_get_by_name(net, buf); | |
661 | + if (dev) | |
662 | + break; | |
663 | + } | |
664 | + | |
665 | + if (WARN_ON_ONCE(dev == NULL)) { | |
666 | + /* IMQ device not found. Exotic config? */ | |
667 | + return ERR_PTR(-ENODEV); | |
668 | + } | |
669 | + | |
670 | + imq_devs_cache[index] = dev; | |
671 | + dev_put(dev); | |
672 | + | |
673 | + return dev; | |
674 | +} | |
675 | + | |
676 | +static struct nf_queue_entry *nf_queue_entry_dup(struct nf_queue_entry *e) | |
677 | +{ | |
678 | + struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC); | |
679 | + if (entry) { | |
680 | + if (nf_queue_entry_get_refs(entry)) | |
681 | + return entry; | |
682 | + kfree(entry); | |
683 | + } | |
684 | + return NULL; | |
685 | +} | |
686 | + | |
687 | +#ifdef CONFIG_BRIDGE_NETFILTER | |
688 | +/* When called from bridge netfilter, skb->data must point to MAC header | |
689 | + * before calling skb_gso_segment(). Else, original MAC header is lost | |
690 | + * and segmented skbs will be sent to wrong destination. | |
691 | + */ | |
692 | +static void nf_bridge_adjust_skb_data(struct sk_buff *skb) | |
693 | +{ | |
694 | + if (skb->nf_bridge) | |
695 | + __skb_push(skb, skb->network_header - skb->mac_header); | |
696 | +} | |
697 | + | |
698 | +static void nf_bridge_adjust_segmented_data(struct sk_buff *skb) | |
699 | +{ | |
700 | + if (skb->nf_bridge) | |
701 | + __skb_pull(skb, skb->network_header - skb->mac_header); | |
702 | +} | |
703 | +#else | |
704 | +#define nf_bridge_adjust_skb_data(s) do {} while (0) | |
705 | +#define nf_bridge_adjust_segmented_data(s) do {} while (0) | |
706 | +#endif | |
707 | + | |
708 | +static void free_entry(struct nf_queue_entry *entry) | |
709 | +{ | |
710 | + nf_queue_entry_release_refs(entry); | |
711 | + kfree(entry); | |
712 | +} | |
713 | + | |
714 | +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev); | |
715 | + | |
716 | +static int __imq_nf_queue_gso(struct nf_queue_entry *entry, | |
717 | + struct net_device *dev, struct sk_buff *skb) | |
718 | +{ | |
719 | + int ret = -ENOMEM; | |
720 | + struct nf_queue_entry *entry_seg; | |
721 | + | |
722 | + nf_bridge_adjust_segmented_data(skb); | |
723 | + | |
724 | + if (skb->next == NULL) { /* last packet, no need to copy entry */ | |
725 | + struct sk_buff *gso_skb = entry->skb; | |
726 | + entry->skb = skb; | |
727 | + ret = __imq_nf_queue(entry, dev); | |
728 | + if (ret) | |
729 | + entry->skb = gso_skb; | |
730 | + return ret; | |
731 | + } | |
732 | + | |
733 | + skb->next = NULL; | |
734 | + | |
735 | + entry_seg = nf_queue_entry_dup(entry); | |
736 | + if (entry_seg) { | |
737 | + entry_seg->skb = skb; | |
738 | + ret = __imq_nf_queue(entry_seg, dev); | |
739 | + if (ret) | |
740 | + free_entry(entry_seg); | |
741 | + } | |
742 | + return ret; | |
743 | +} | |
744 | + | |
745 | +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num) | |
746 | +{ | |
747 | + struct sk_buff *skb, *segs; | |
748 | + struct net_device *dev; | |
749 | + unsigned int queued; | |
750 | + int index, retval, err; | |
751 | + | |
752 | + index = entry->skb->imq_flags & IMQ_F_IFMASK; | |
753 | + if (unlikely(index > numdevs - 1)) { | |
754 | + if (net_ratelimit()) | |
755 | + pr_warn("IMQ: invalid device specified, highest is %u\n", | |
756 | + numdevs - 1); | |
757 | + retval = -EINVAL; | |
758 | + goto out_no_dev; | |
759 | + } | |
760 | + | |
761 | + /* check for imq device by index from cache */ | |
762 | + dev = imq_devs_cache[index]; | |
763 | + if (unlikely(!dev)) { | |
764 | + dev = get_imq_device_by_index(index); | |
765 | + if (IS_ERR(dev)) { | |
766 | + retval = PTR_ERR(dev); | |
767 | + goto out_no_dev; | |
768 | + } | |
769 | + } | |
770 | + | |
771 | + if (unlikely(!(dev->flags & IFF_UP))) { | |
772 | + entry->skb->imq_flags = 0; | |
773 | + retval = -ECANCELED; | |
774 | + goto out_no_dev; | |
775 | + } | |
776 | + | |
777 | + if (!skb_is_gso(entry->skb)) | |
778 | + return __imq_nf_queue(entry, dev); | |
779 | + | |
780 | + /* Since 3.10.x, GSO handling moved here as result of upstream commit | |
781 | + * a5fedd43d5f6c94c71053a66e4c3d2e35f1731a2 (netfilter: move | |
782 | + * skb_gso_segment into nfnetlink_queue module). | |
783 | + * | |
784 | + * Following code replicates the gso handling from | |
785 | + * 'net/netfilter/nfnetlink_queue_core.c':nfqnl_enqueue_packet(). | |
786 | + */ | |
787 | + | |
788 | + skb = entry->skb; | |
789 | + | |
790 | + switch (entry->pf) { | |
791 | + case NFPROTO_IPV4: | |
792 | + skb->protocol = htons(ETH_P_IP); | |
793 | + break; | |
794 | + case NFPROTO_IPV6: | |
795 | + skb->protocol = htons(ETH_P_IPV6); | |
796 | + break; | |
797 | + } | |
798 | + | |
799 | + nf_bridge_adjust_skb_data(skb); | |
800 | + segs = skb_gso_segment(skb, 0); | |
801 | + /* Does not use PTR_ERR to limit the number of error codes that can be | |
802 | + * returned by nf_queue. For instance, callers rely on -ECANCELED to | |
803 | + * mean 'ignore this hook'. | |
804 | + */ | |
805 | + err = -ENOBUFS; | |
806 | + if (IS_ERR(segs)) | |
807 | + goto out_err; | |
808 | + queued = 0; | |
809 | + err = 0; | |
810 | + do { | |
811 | + struct sk_buff *nskb = segs->next; | |
812 | + if (nskb && nskb->next) | |
813 | + nskb->cb_next = NULL; | |
814 | + if (err == 0) | |
815 | + err = __imq_nf_queue_gso(entry, dev, segs); | |
816 | + if (err == 0) | |
817 | + queued++; | |
818 | + else | |
819 | + kfree_skb(segs); | |
820 | + segs = nskb; | |
821 | + } while (segs); | |
822 | + | |
823 | + if (queued) { | |
824 | + if (err) /* some segments are already queued */ | |
825 | + free_entry(entry); | |
826 | + kfree_skb(skb); | |
827 | + return 0; | |
828 | + } | |
829 | + | |
830 | +out_err: | |
831 | + nf_bridge_adjust_segmented_data(skb); | |
832 | + retval = err; | |
833 | +out_no_dev: | |
834 | + return retval; | |
835 | +} | |
836 | + | |
837 | +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev) | |
838 | +{ | |
839 | + struct sk_buff *skb_orig, *skb, *skb_shared; | |
840 | + struct Qdisc *q; | |
841 | + struct netdev_queue *txq; | |
842 | + spinlock_t *root_lock; | |
843 | + int users; | |
844 | + int retval = -EINVAL; | |
845 | + unsigned int orig_queue_index; | |
846 | + | |
847 | + dev->last_rx = jiffies; | |
848 | + | |
849 | + skb = entry->skb; | |
850 | + skb_orig = NULL; | |
851 | + | |
852 | + /* skb has owner? => make clone */ | |
853 | + if (unlikely(skb->destructor)) { | |
854 | + skb_orig = skb; | |
855 | + skb = skb_clone(skb, GFP_ATOMIC); | |
856 | + if (unlikely(!skb)) { | |
857 | + retval = -ENOMEM; | |
858 | + goto out; | |
859 | + } | |
860 | + skb->cb_next = NULL; | |
861 | + entry->skb = skb; | |
862 | + } | |
863 | + | |
864 | + skb->nf_queue_entry = entry; | |
865 | + | |
866 | + dev->stats.rx_bytes += skb->len; | |
867 | + dev->stats.rx_packets++; | |
868 | + | |
869 | + if (!skb->dev) { | |
870 | + /* skb->dev == NULL causes problems, try the find cause. */ | |
871 | + if (net_ratelimit()) { | |
872 | + dev_warn(&dev->dev, | |
873 | + "received packet with skb->dev == NULL\n"); | |
874 | + dump_stack(); | |
875 | + } | |
876 | + | |
877 | + skb->dev = dev; | |
878 | + } | |
879 | + | |
880 | + /* Disables softirqs for lock below */ | |
881 | + rcu_read_lock_bh(); | |
882 | + | |
883 | + /* Multi-queue selection */ | |
884 | + orig_queue_index = skb_get_queue_mapping(skb); | |
885 | + txq = imq_select_queue(dev, skb); | |
886 | + | |
887 | + q = rcu_dereference(txq->qdisc); | |
888 | + if (unlikely(!q->enqueue)) | |
889 | + goto packet_not_eaten_by_imq_dev; | |
890 | + | |
891 | + root_lock = qdisc_lock(q); | |
892 | + spin_lock(root_lock); | |
893 | + | |
894 | + users = atomic_read(&skb->users); | |
895 | + | |
896 | + skb_shared = skb_get(skb); /* increase reference count by one */ | |
897 | + | |
898 | + /* backup skb->cb, as qdisc layer will overwrite it */ | |
899 | + skb_save_cb(skb_shared); | |
900 | + qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */ | |
901 | + | |
902 | + if (likely(atomic_read(&skb_shared->users) == users + 1)) { | |
903 | + kfree_skb(skb_shared); /* decrease reference count by one */ | |
904 | + | |
905 | + skb->destructor = &imq_skb_destructor; | |
906 | + | |
907 | + /* cloned? */ | |
908 | + if (unlikely(skb_orig)) | |
909 | + kfree_skb(skb_orig); /* free original */ | |
910 | + | |
911 | + spin_unlock(root_lock); | |
912 | + rcu_read_unlock_bh(); | |
913 | + | |
914 | + /* schedule qdisc dequeue */ | |
915 | + __netif_schedule(q); | |
916 | + | |
917 | + retval = 0; | |
918 | + goto out; | |
919 | + } else { | |
920 | + skb_restore_cb(skb_shared); /* restore skb->cb */ | |
921 | + skb->nf_queue_entry = NULL; | |
922 | + /* | |
923 | + * qdisc dropped packet and decreased skb reference count of | |
924 | + * skb, so we don't really want to and try refree as that would | |
925 | + * actually destroy the skb. | |
926 | + */ | |
927 | + spin_unlock(root_lock); | |
928 | + goto packet_not_eaten_by_imq_dev; | |
929 | + } | |
930 | + | |
931 | +packet_not_eaten_by_imq_dev: | |
932 | + skb_set_queue_mapping(skb, orig_queue_index); | |
933 | + rcu_read_unlock_bh(); | |
934 | + | |
935 | + /* cloned? restore original */ | |
936 | + if (unlikely(skb_orig)) { | |
937 | + kfree_skb(skb); | |
938 | + entry->skb = skb_orig; | |
939 | + } | |
940 | + retval = -1; | |
941 | +out: | |
942 | + return retval; | |
943 | +} | |
944 | + | |
945 | +static unsigned int imq_nf_hook(const struct nf_hook_ops *hook_ops, | |
946 | + struct sk_buff *pskb, | |
947 | + const struct net_device *indev, | |
948 | + const struct net_device *outdev, | |
949 | + int (*okfn)(struct sk_buff *)) | |
950 | +{ | |
951 | + return (pskb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT; | |
952 | +} | |
953 | + | |
954 | +static int imq_close(struct net_device *dev) | |
955 | +{ | |
956 | + netif_stop_queue(dev); | |
957 | + return 0; | |
958 | +} | |
959 | + | |
960 | +static int imq_open(struct net_device *dev) | |
961 | +{ | |
962 | + netif_start_queue(dev); | |
963 | + return 0; | |
964 | +} | |
965 | + | |
966 | +static const struct net_device_ops imq_netdev_ops = { | |
967 | + .ndo_open = imq_open, | |
968 | + .ndo_stop = imq_close, | |
969 | + .ndo_start_xmit = imq_dev_xmit, | |
970 | + .ndo_get_stats = imq_get_stats, | |
971 | +}; | |
972 | + | |
973 | +static void imq_setup(struct net_device *dev) | |
974 | +{ | |
975 | + dev->netdev_ops = &imq_netdev_ops; | |
976 | + dev->type = ARPHRD_VOID; | |
977 | + dev->mtu = 16000; /* too small? */ | |
978 | + dev->tx_queue_len = 11000; /* too big? */ | |
979 | + dev->flags = IFF_NOARP; | |
980 | + dev->features = NETIF_F_SG | NETIF_F_FRAGLIST | | |
981 | + NETIF_F_GSO | NETIF_F_HW_CSUM | | |
982 | + NETIF_F_HIGHDMA; | |
983 | + dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | | |
984 | + IFF_TX_SKB_SHARING); | |
985 | +} | |
986 | + | |
987 | +static int imq_validate(struct nlattr *tb[], struct nlattr *data[]) | |
988 | +{ | |
989 | + int ret = 0; | |
990 | + | |
991 | + if (tb[IFLA_ADDRESS]) { | |
992 | + if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { | |
993 | + ret = -EINVAL; | |
994 | + goto end; | |
995 | + } | |
996 | + if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { | |
997 | + ret = -EADDRNOTAVAIL; | |
998 | + goto end; | |
999 | + } | |
1000 | + } | |
1001 | + return 0; | |
1002 | +end: | |
1003 | + pr_warn("IMQ: imq_validate failed (%d)\n", ret); | |
1004 | + return ret; | |
1005 | +} | |
1006 | + | |
1007 | +static struct rtnl_link_ops imq_link_ops __read_mostly = { | |
1008 | + .kind = "imq", | |
1009 | + .priv_size = 0, | |
1010 | + .setup = imq_setup, | |
1011 | + .validate = imq_validate, | |
1012 | +}; | |
1013 | + | |
1014 | +static const struct nf_queue_handler imq_nfqh = { | |
1015 | + .outfn = imq_nf_queue, | |
1016 | +}; | |
1017 | + | |
1018 | +static int __init imq_init_hooks(void) | |
1019 | +{ | |
1020 | + int ret; | |
1021 | + | |
1022 | + nf_register_queue_imq_handler(&imq_nfqh); | |
1023 | + | |
1024 | + ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops)); | |
1025 | + if (ret < 0) | |
1026 | + nf_unregister_queue_imq_handler(); | |
1027 | + | |
1028 | + return ret; | |
1029 | +} | |
1030 | + | |
1031 | +static int __init imq_init_one(int index) | |
1032 | +{ | |
1033 | + struct net_device *dev; | |
1034 | + int ret; | |
1035 | + | |
1036 | + dev = alloc_netdev_mq(0, "imq%d", imq_setup, numqueues); | |
1037 | + if (!dev) | |
1038 | + return -ENOMEM; | |
1039 | + | |
1040 | + ret = dev_alloc_name(dev, dev->name); | |
1041 | + if (ret < 0) | |
1042 | + goto fail; | |
1043 | + | |
1044 | + dev->rtnl_link_ops = &imq_link_ops; | |
1045 | + ret = register_netdevice(dev); | |
1046 | + if (ret < 0) | |
1047 | + goto fail; | |
1048 | + | |
1049 | + return 0; | |
1050 | +fail: | |
1051 | + free_netdev(dev); | |
1052 | + return ret; | |
1053 | +} | |
1054 | + | |
1055 | +static int __init imq_init_devs(void) | |
1056 | +{ | |
1057 | + int err, i; | |
1058 | + | |
1059 | + if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) { | |
1060 | + pr_err("IMQ: numdevs has to be betweed 1 and %u\n", | |
1061 | + IMQ_MAX_DEVS); | |
1062 | + return -EINVAL; | |
1063 | + } | |
1064 | + | |
1065 | + if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) { | |
1066 | + pr_err("IMQ: numqueues has to be betweed 1 and %u\n", | |
1067 | + IMQ_MAX_QUEUES); | |
1068 | + return -EINVAL; | |
1069 | + } | |
1070 | + | |
1071 | + get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd)); | |
1072 | + | |
1073 | + rtnl_lock(); | |
1074 | + err = __rtnl_link_register(&imq_link_ops); | |
1075 | + | |
1076 | + for (i = 0; i < numdevs && !err; i++) | |
1077 | + err = imq_init_one(i); | |
1078 | + | |
1079 | + if (err) { | |
1080 | + __rtnl_link_unregister(&imq_link_ops); | |
1081 | + memset(imq_devs_cache, 0, sizeof(imq_devs_cache)); | |
1082 | + } | |
1083 | + rtnl_unlock(); | |
1084 | + | |
1085 | + return err; | |
1086 | +} | |
1087 | + | |
1088 | +static int __init imq_init_module(void) | |
1089 | +{ | |
1090 | + int err; | |
1091 | + | |
1092 | +#if defined(CONFIG_IMQ_NUM_DEVS) | |
1093 | + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16); | |
1094 | + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2); | |
1095 | + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK); | |
1096 | +#endif | |
1097 | + | |
1098 | + err = imq_init_devs(); | |
1099 | + if (err) { | |
1100 | + pr_err("IMQ: Error trying imq_init_devs(net)\n"); | |
1101 | + return err; | |
1102 | + } | |
1103 | + | |
1104 | + err = imq_init_hooks(); | |
1105 | + if (err) { | |
1106 | + pr_err(KERN_ERR "IMQ: Error trying imq_init_hooks()\n"); | |
1107 | + rtnl_link_unregister(&imq_link_ops); | |
1108 | + memset(imq_devs_cache, 0, sizeof(imq_devs_cache)); | |
1109 | + return err; | |
1110 | + } | |
1111 | + | |
1112 | + pr_info("IMQ driver loaded successfully. (numdevs = %d, numqueues = %d)\n", | |
1113 | + numdevs, numqueues); | |
1114 | + | |
1115 | +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB) | |
1116 | + pr_info("\tHooking IMQ before NAT on PREROUTING.\n"); | |
1117 | +#else | |
1118 | + pr_info("\tHooking IMQ after NAT on PREROUTING.\n"); | |
1119 | +#endif | |
1120 | +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB) | |
1121 | + pr_info("\tHooking IMQ before NAT on POSTROUTING.\n"); | |
1122 | +#else | |
1123 | + pr_info("\tHooking IMQ after NAT on POSTROUTING.\n"); | |
1124 | +#endif | |
1125 | + | |
1126 | + return 0; | |
1127 | +} | |
1128 | + | |
1129 | +static void __exit imq_unhook(void) | |
1130 | +{ | |
1131 | + nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops)); | |
1132 | + nf_unregister_queue_imq_handler(); | |
1133 | +} | |
1134 | + | |
1135 | +static void __exit imq_cleanup_devs(void) | |
1136 | +{ | |
1137 | + rtnl_link_unregister(&imq_link_ops); | |
1138 | + memset(imq_devs_cache, 0, sizeof(imq_devs_cache)); | |
1139 | +} | |
1140 | + | |
1141 | +static void __exit imq_exit_module(void) | |
1142 | +{ | |
1143 | + imq_unhook(); | |
1144 | + imq_cleanup_devs(); | |
1145 | + pr_info("IMQ driver unloaded successfully.\n"); | |
1146 | +} | |
1147 | + | |
1148 | +module_init(imq_init_module); | |
1149 | +module_exit(imq_exit_module); | |
1150 | + | |
1151 | +module_param(numdevs, int, 0); | |
1152 | +module_param(numqueues, int, 0); | |
1153 | +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will be created)"); | |
1154 | +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device"); | |
1155 | +MODULE_AUTHOR("http://www.linuximq.net"); | |
1156 | +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information."); | |
1157 | +MODULE_LICENSE("GPL"); | |
1158 | +MODULE_ALIAS_RTNL_LINK("imq"); | |
1159 | + | |
1160 | diff --git a/include/linux/imq.h b/include/linux/imq.h | |
1161 | new file mode 100644 | |
1162 | index 0000000..1babb09 | |
1163 | --- /dev/null | |
1164 | +++ b/include/linux/imq.h | |
1165 | @@ -0,0 +1,13 @@ | |
1166 | +#ifndef _IMQ_H | |
1167 | +#define _IMQ_H | |
1168 | + | |
1169 | +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */ | |
1170 | +#define IMQ_F_BITS 5 | |
1171 | + | |
1172 | +#define IMQ_F_IFMASK 0x0f | |
1173 | +#define IMQ_F_ENQUEUE 0x10 | |
1174 | + | |
1175 | +#define IMQ_MAX_DEVS (IMQ_F_IFMASK + 1) | |
1176 | + | |
1177 | +#endif /* _IMQ_H */ | |
1178 | + | |
1179 | diff --git a/include/linux/netfilter/xt_IMQ.h b/include/linux/netfilter/xt_IMQ.h | |
1180 | new file mode 100644 | |
1181 | index 0000000..9b07230 | |
1182 | --- /dev/null | |
1183 | +++ b/include/linux/netfilter/xt_IMQ.h | |
1184 | @@ -0,0 +1,9 @@ | |
1185 | +#ifndef _XT_IMQ_H | |
1186 | +#define _XT_IMQ_H | |
1187 | + | |
1188 | +struct xt_imq_info { | |
1189 | + unsigned int todev; /* target imq device */ | |
1190 | +}; | |
1191 | + | |
1192 | +#endif /* _XT_IMQ_H */ | |
1193 | + | |
1194 | diff --git a/include/linux/netfilter_ipv4/ipt_IMQ.h b/include/linux/netfilter_ipv4/ipt_IMQ.h | |
1195 | new file mode 100644 | |
1196 | index 0000000..7af320f | |
1197 | --- /dev/null | |
1198 | +++ b/include/linux/netfilter_ipv4/ipt_IMQ.h | |
1199 | @@ -0,0 +1,10 @@ | |
1200 | +#ifndef _IPT_IMQ_H | |
1201 | +#define _IPT_IMQ_H | |
1202 | + | |
1203 | +/* Backwards compatibility for old userspace */ | |
1204 | +#include <linux/netfilter/xt_IMQ.h> | |
1205 | + | |
1206 | +#define ipt_imq_info xt_imq_info | |
1207 | + | |
1208 | +#endif /* _IPT_IMQ_H */ | |
1209 | + | |
1210 | diff --git a/include/linux/netfilter_ipv6/ip6t_IMQ.h b/include/linux/netfilter_ipv6/ip6t_IMQ.h | |
1211 | new file mode 100644 | |
1212 | index 0000000..198ac01 | |
1213 | --- /dev/null | |
1214 | +++ b/include/linux/netfilter_ipv6/ip6t_IMQ.h | |
1215 | @@ -0,0 +1,10 @@ | |
1216 | +#ifndef _IP6T_IMQ_H | |
1217 | +#define _IP6T_IMQ_H | |
1218 | + | |
1219 | +/* Backwards compatibility for old userspace */ | |
1220 | +#include <linux/netfilter/xt_IMQ.h> | |
1221 | + | |
1222 | +#define ip6t_imq_info xt_imq_info | |
1223 | + | |
1224 | +#endif /* _IP6T_IMQ_H */ | |
1225 | + | |
1226 | diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h | |
1227 | index ad8f859..8473090 100644 | |
1228 | --- a/include/linux/skbuff.h | |
1229 | +++ b/include/linux/skbuff.h | |
1230 | @@ -33,6 +33,9 @@ | |
1231 | #include <linux/dma-mapping.h> | |
1232 | #include <linux/netdev_features.h> | |
1233 | #include <net/flow_keys.h> | |
1234 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1235 | +#include <linux/imq.h> | |
1236 | +#endif | |
1237 | ||
1238 | /* A. Checksumming of received packets by device. | |
1239 | * | |
1240 | @@ -441,6 +444,9 @@ struct sk_buff { | |
1241 | * first. This is owned by whoever has the skb queued ATM. | |
1242 | */ | |
1243 | char cb[48] __aligned(8); | |
1244 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1245 | + void *cb_next; | |
1246 | +#endif | |
1247 | ||
1248 | unsigned long _skb_refdst; | |
1249 | #ifdef CONFIG_XFRM | |
1250 | @@ -476,6 +482,9 @@ struct sk_buff { | |
1251 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
1252 | struct nf_conntrack *nfct; | |
1253 | #endif | |
1254 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1255 | + struct nf_queue_entry *nf_queue_entry; | |
1256 | +#endif | |
1257 | #ifdef CONFIG_BRIDGE_NETFILTER | |
1258 | struct nf_bridge_info *nf_bridge; | |
1259 | #endif | |
1260 | @@ -513,6 +522,9 @@ struct sk_buff { | |
1261 | */ | |
1262 | __u8 encapsulation:1; | |
1263 | /* 6/8 bit hole (depending on ndisc_nodetype presence) */ | |
1264 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1265 | + __u8 imq_flags:IMQ_F_BITS; | |
1266 | +#endif | |
1267 | kmemcheck_bitfield_end(flags2); | |
1268 | ||
1269 | #if defined CONFIG_NET_DMA || defined CONFIG_NET_RX_BUSY_POLL | |
1270 | @@ -653,6 +665,12 @@ void kfree_skb_list(struct sk_buff *segs); | |
1271 | void skb_tx_error(struct sk_buff *skb); | |
1272 | void consume_skb(struct sk_buff *skb); | |
1273 | void __kfree_skb(struct sk_buff *skb); | |
1274 | + | |
1275 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1276 | +int skb_save_cb(struct sk_buff *skb); | |
1277 | +int skb_restore_cb(struct sk_buff *skb); | |
1278 | +#endif | |
1279 | + | |
1280 | extern struct kmem_cache *skbuff_head_cache; | |
1281 | ||
1282 | void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); | |
1283 | @@ -2739,6 +2757,10 @@ static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src) | |
1284 | nf_conntrack_get(src->nfct); | |
1285 | dst->nfctinfo = src->nfctinfo; | |
1286 | #endif | |
1287 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1288 | + dst->imq_flags = src->imq_flags; | |
1289 | + dst->nf_queue_entry = src->nf_queue_entry; | |
1290 | +#endif | |
1291 | #ifdef CONFIG_BRIDGE_NETFILTER | |
1292 | dst->nf_bridge = src->nf_bridge; | |
1293 | nf_bridge_get(src->nf_bridge); | |
1294 | diff --git a/include/net/netfilter/nf_queue.h b/include/net/netfilter/nf_queue.h | |
1295 | index 84a53d7..6ffb593 100644 | |
1296 | --- a/include/net/netfilter/nf_queue.h | |
1297 | +++ b/include/net/netfilter/nf_queue.h | |
1298 | @@ -33,6 +33,12 @@ struct nf_queue_handler { | |
1299 | void nf_register_queue_handler(const struct nf_queue_handler *qh); | |
1300 | void nf_unregister_queue_handler(void); | |
1301 | void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict); | |
1302 | +void nf_queue_entry_release_refs(struct nf_queue_entry *entry); | |
1303 | + | |
1304 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1305 | +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh); | |
1306 | +void nf_unregister_queue_imq_handler(void); | |
1307 | +#endif | |
1308 | ||
1309 | bool nf_queue_entry_get_refs(struct nf_queue_entry *entry); | |
1310 | void nf_queue_entry_release_refs(struct nf_queue_entry *entry); | |
1311 | diff --git a/include/uapi/linux/netfilter.h b/include/uapi/linux/netfilter.h | |
1312 | index ef1b1f8..079e5ff 100644 | |
1313 | --- a/include/uapi/linux/netfilter.h | |
1314 | +++ b/include/uapi/linux/netfilter.h | |
1315 | @@ -13,7 +13,8 @@ | |
1316 | #define NF_QUEUE 3 | |
1317 | #define NF_REPEAT 4 | |
1318 | #define NF_STOP 5 | |
1319 | -#define NF_MAX_VERDICT NF_STOP | |
1320 | +#define NF_IMQ_QUEUE 6 | |
1321 | +#define NF_MAX_VERDICT NF_IMQ_QUEUE | |
1322 | ||
1323 | /* we overload the higher bits for encoding auxiliary data such as the queue | |
1324 | * number or errno values. Not nice, but better than additional function | |
1325 | diff --git a/net/core/dev.c b/net/core/dev.c | |
1326 | index 3ed11a5..fd62030 100644 | |
1327 | --- a/net/core/dev.c | |
1328 | +++ b/net/core/dev.c | |
1329 | @@ -132,6 +132,9 @@ | |
1330 | #include <linux/hashtable.h> | |
1331 | #include <linux/vmalloc.h> | |
1332 | #include <linux/if_macvlan.h> | |
1333 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1334 | +#include <linux/imq.h> | |
1335 | +#endif | |
1336 | ||
1337 | #include "net-sysfs.h" | |
1338 | ||
1339 | @@ -2611,7 +2614,12 @@ int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, | |
1340 | } | |
1341 | } | |
1342 | ||
1343 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1344 | + if (!list_empty(&ptype_all) && | |
1345 | + !(skb->imq_flags & IMQ_F_ENQUEUE)) | |
1346 | +#else | |
1347 | if (!list_empty(&ptype_all)) | |
1348 | +#endif | |
1349 | dev_queue_xmit_nit(skb, dev); | |
1350 | ||
1351 | skb_len = skb->len; | |
1352 | diff --git a/net/core/skbuff.c b/net/core/skbuff.c | |
1353 | index baf6fc4..7d30d78 100644 | |
1354 | --- a/net/core/skbuff.c | |
1355 | +++ b/net/core/skbuff.c | |
1356 | @@ -77,6 +77,115 @@ | |
1357 | ||
1358 | struct kmem_cache *skbuff_head_cache __read_mostly; | |
1359 | static struct kmem_cache *skbuff_fclone_cache __read_mostly; | |
1360 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1361 | +static struct kmem_cache *skbuff_cb_store_cache __read_mostly; | |
1362 | +#endif | |
1363 | + | |
1364 | +static void sock_pipe_buf_release(struct pipe_inode_info *pipe, | |
1365 | + struct pipe_buffer *buf) | |
1366 | +{ | |
1367 | + put_page(buf->page); | |
1368 | +} | |
1369 | + | |
1370 | +static void sock_pipe_buf_get(struct pipe_inode_info *pipe, | |
1371 | + struct pipe_buffer *buf) | |
1372 | +{ | |
1373 | + get_page(buf->page); | |
1374 | +} | |
1375 | + | |
1376 | +static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, | |
1377 | + struct pipe_buffer *buf) | |
1378 | +{ | |
1379 | + return 1; | |
1380 | +} | |
1381 | + | |
1382 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1383 | +/* Control buffer save/restore for IMQ devices */ | |
1384 | +struct skb_cb_table { | |
1385 | + char cb[48] __aligned(8); | |
1386 | + void *cb_next; | |
1387 | + atomic_t refcnt; | |
1388 | +}; | |
1389 | + | |
1390 | +static DEFINE_SPINLOCK(skb_cb_store_lock); | |
1391 | + | |
1392 | +int skb_save_cb(struct sk_buff *skb) | |
1393 | +{ | |
1394 | + struct skb_cb_table *next; | |
1395 | + | |
1396 | + next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC); | |
1397 | + if (!next) | |
1398 | + return -ENOMEM; | |
1399 | + | |
1400 | + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb)); | |
1401 | + | |
1402 | + memcpy(next->cb, skb->cb, sizeof(skb->cb)); | |
1403 | + next->cb_next = skb->cb_next; | |
1404 | + | |
1405 | + atomic_set(&next->refcnt, 1); | |
1406 | + | |
1407 | + skb->cb_next = next; | |
1408 | + return 0; | |
1409 | +} | |
1410 | +EXPORT_SYMBOL(skb_save_cb); | |
1411 | + | |
1412 | +int skb_restore_cb(struct sk_buff *skb) | |
1413 | +{ | |
1414 | + struct skb_cb_table *next; | |
1415 | + | |
1416 | + if (!skb->cb_next) | |
1417 | + return 0; | |
1418 | + | |
1419 | + next = skb->cb_next; | |
1420 | + | |
1421 | + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb)); | |
1422 | + | |
1423 | + memcpy(skb->cb, next->cb, sizeof(skb->cb)); | |
1424 | + skb->cb_next = next->cb_next; | |
1425 | + | |
1426 | + spin_lock(&skb_cb_store_lock); | |
1427 | + | |
1428 | + if (atomic_dec_and_test(&next->refcnt)) | |
1429 | + kmem_cache_free(skbuff_cb_store_cache, next); | |
1430 | + | |
1431 | + spin_unlock(&skb_cb_store_lock); | |
1432 | + | |
1433 | + return 0; | |
1434 | +} | |
1435 | +EXPORT_SYMBOL(skb_restore_cb); | |
1436 | + | |
1437 | +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old) | |
1438 | +{ | |
1439 | + struct skb_cb_table *next; | |
1440 | + struct sk_buff *old; | |
1441 | + | |
1442 | + if (!__old->cb_next) { | |
1443 | + new->cb_next = NULL; | |
1444 | + return; | |
1445 | + } | |
1446 | + | |
1447 | + spin_lock(&skb_cb_store_lock); | |
1448 | + | |
1449 | + old = (struct sk_buff *)__old; | |
1450 | + | |
1451 | + next = old->cb_next; | |
1452 | + atomic_inc(&next->refcnt); | |
1453 | + new->cb_next = next; | |
1454 | + | |
1455 | + spin_unlock(&skb_cb_store_lock); | |
1456 | +} | |
1457 | +#endif | |
1458 | + | |
1459 | +/* Pipe buffer operations for a socket. */ | |
1460 | +static const struct pipe_buf_operations sock_pipe_buf_ops = { | |
1461 | + .can_merge = 0, | |
1462 | + .map = generic_pipe_buf_map, | |
1463 | + .unmap = generic_pipe_buf_unmap, | |
1464 | + .confirm = generic_pipe_buf_confirm, | |
1465 | + .release = sock_pipe_buf_release, | |
1466 | + .steal = sock_pipe_buf_steal, | |
1467 | + .get = sock_pipe_buf_get, | |
1468 | +}; | |
1469 | ||
1470 | /** | |
1471 | * skb_panic - private function for out-of-line support | |
1472 | @@ -563,6 +672,28 @@ static void skb_release_head_state(struct sk_buff *skb) | |
1473 | WARN_ON(in_irq()); | |
1474 | skb->destructor(skb); | |
1475 | } | |
1476 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1477 | + /* | |
1478 | + * This should not happen. When it does, avoid memleak by restoring | |
1479 | + * the chain of cb-backups. | |
1480 | + */ | |
1481 | + while (skb->cb_next != NULL) { | |
1482 | + if (net_ratelimit()) | |
1483 | + pr_warn("IMQ: kfree_skb: skb->cb_next: %08x\n", | |
1484 | + (unsigned int)skb->cb_next); | |
1485 | + | |
1486 | + skb_restore_cb(skb); | |
1487 | + } | |
1488 | + /* | |
1489 | + * This should not happen either, nf_queue_entry is nullified in | |
1490 | + * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are | |
1491 | + * leaking entry pointers, maybe memory. We don't know if this is | |
1492 | + * pointer to already freed memory, or should this be freed. | |
1493 | + * If this happens we need to add refcounting, etc for nf_queue_entry. | |
1494 | + */ | |
1495 | + if (skb->nf_queue_entry && net_ratelimit()) | |
1496 | + pr_warn("%s\n", "IMQ: kfree_skb: skb->nf_queue_entry != NULL"); | |
1497 | +#endif | |
1498 | #if IS_ENABLED(CONFIG_NF_CONNTRACK) | |
1499 | nf_conntrack_put(skb->nfct); | |
1500 | #endif | |
1501 | @@ -694,6 +825,10 @@ static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) | |
1502 | new->sp = secpath_get(old->sp); | |
1503 | #endif | |
1504 | memcpy(new->cb, old->cb, sizeof(old->cb)); | |
1505 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1506 | + new->cb_next = NULL; | |
1507 | + /*skb_copy_stored_cb(new, old);*/ | |
1508 | +#endif | |
1509 | new->csum = old->csum; | |
1510 | new->local_df = old->local_df; | |
1511 | new->pkt_type = old->pkt_type; | |
1512 | @@ -3233,6 +3368,13 @@ void __init skb_init(void) | |
1513 | 0, | |
1514 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, | |
1515 | NULL); | |
1516 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1517 | + skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache", | |
1518 | + sizeof(struct skb_cb_table), | |
1519 | + 0, | |
1520 | + SLAB_HWCACHE_ALIGN|SLAB_PANIC, | |
1521 | + NULL); | |
1522 | +#endif | |
1523 | } | |
1524 | ||
1525 | /** | |
1526 | diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c | |
1527 | index 12f7ef0..deb1c9d 100644 | |
1528 | --- a/net/ipv6/ip6_output.c | |
1529 | +++ b/net/ipv6/ip6_output.c | |
1530 | @@ -64,9 +64,6 @@ static int ip6_finish_output2(struct sk_buff *skb) | |
1531 | struct in6_addr *nexthop; | |
1532 | int ret; | |
1533 | ||
1534 | - skb->protocol = htons(ETH_P_IPV6); | |
1535 | - skb->dev = dev; | |
1536 | - | |
1537 | if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) { | |
1538 | struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb)); | |
1539 | ||
1540 | @@ -143,6 +140,13 @@ int ip6_output(struct sk_buff *skb) | |
1541 | return 0; | |
1542 | } | |
1543 | ||
1544 | + /* | |
1545 | + * IMQ-patch: moved setting skb->dev and skb->protocol from | |
1546 | + * ip6_finish_output2 to fix crashing at netif_skb_features(). | |
1547 | + */ | |
1548 | + skb->protocol = htons(ETH_P_IPV6); | |
1549 | + skb->dev = dev; | |
1550 | + | |
1551 | return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev, | |
1552 | ip6_finish_output, | |
1553 | !(IP6CB(skb)->flags & IP6SKB_REROUTED)); | |
1554 | diff --git a/net/netfilter/Kconfig b/net/netfilter/Kconfig | |
1555 | index e9410d1..ba801d5 100644 | |
1556 | --- a/net/netfilter/Kconfig | |
1557 | +++ b/net/netfilter/Kconfig | |
1558 | @@ -751,6 +751,18 @@ config NETFILTER_XT_TARGET_LOG | |
1559 | ||
1560 | To compile it as a module, choose M here. If unsure, say N. | |
1561 | ||
1562 | +config NETFILTER_XT_TARGET_IMQ | |
1563 | + tristate '"IMQ" target support' | |
1564 | + depends on NETFILTER_XTABLES | |
1565 | + depends on IP_NF_MANGLE || IP6_NF_MANGLE | |
1566 | + select IMQ | |
1567 | + default m if NETFILTER_ADVANCED=n | |
1568 | + help | |
1569 | + This option adds a `IMQ' target which is used to specify if and | |
1570 | + to which imq device packets should get enqueued/dequeued. | |
1571 | + | |
1572 | + To compile it as a module, choose M here. If unsure, say N. | |
1573 | + | |
1574 | config NETFILTER_XT_TARGET_MARK | |
1575 | tristate '"MARK" target support' | |
1576 | depends on NETFILTER_ADVANCED | |
1577 | diff --git a/net/netfilter/Makefile b/net/netfilter/Makefile | |
1578 | index bffdad7..050e613 100644 | |
1579 | --- a/net/netfilter/Makefile | |
1580 | +++ b/net/netfilter/Makefile | |
1581 | @@ -103,6 +103,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o | |
1582 | obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o | |
1583 | obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o | |
1584 | obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o | |
1585 | +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o | |
1586 | obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o | |
1587 | obj-$(CONFIG_NETFILTER_XT_TARGET_LOG) += xt_LOG.o | |
1588 | obj-$(CONFIG_NETFILTER_XT_TARGET_NETMAP) += xt_NETMAP.o | |
1589 | diff --git a/net/netfilter/core.c b/net/netfilter/core.c | |
1590 | index 1fbab0c..4493417 100644 | |
1591 | --- a/net/netfilter/core.c | |
1592 | +++ b/net/netfilter/core.c | |
1593 | @@ -191,9 +191,11 @@ next_hook: | |
1594 | ret = NF_DROP_GETERR(verdict); | |
1595 | if (ret == 0) | |
1596 | ret = -EPERM; | |
1597 | - } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) { | |
1598 | + } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE || | |
1599 | + (verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) { | |
1600 | int err = nf_queue(skb, elem, pf, hook, indev, outdev, okfn, | |
1601 | - verdict >> NF_VERDICT_QBITS); | |
1602 | + verdict >> NF_VERDICT_QBITS, | |
1603 | + verdict & NF_VERDICT_MASK); | |
1604 | if (err < 0) { | |
1605 | if (err == -ECANCELED) | |
1606 | goto next_hook; | |
1607 | diff --git a/net/netfilter/nf_internals.h b/net/netfilter/nf_internals.h | |
1608 | index 61a3c92..5388a0e 100644 | |
1609 | --- a/net/netfilter/nf_internals.h | |
1610 | +++ b/net/netfilter/nf_internals.h | |
1611 | @@ -23,7 +23,7 @@ unsigned int nf_iterate(struct list_head *head, struct sk_buff *skb, | |
1612 | int nf_queue(struct sk_buff *skb, struct nf_hook_ops *elem, u_int8_t pf, | |
1613 | unsigned int hook, struct net_device *indev, | |
1614 | struct net_device *outdev, int (*okfn)(struct sk_buff *), | |
1615 | - unsigned int queuenum); | |
1616 | + unsigned int queuenum, unsigned int queuetype); | |
1617 | int __init netfilter_queue_init(void); | |
1618 | ||
1619 | /* nf_log.c */ | |
1620 | diff --git a/net/netfilter/nf_queue.c b/net/netfilter/nf_queue.c | |
1621 | index 5d24b1f..28317dc 100644 | |
1622 | --- a/net/netfilter/nf_queue.c | |
1623 | +++ b/net/netfilter/nf_queue.c | |
1624 | @@ -27,6 +27,23 @@ | |
1625 | */ | |
1626 | static const struct nf_queue_handler __rcu *queue_handler __read_mostly; | |
1627 | ||
1628 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1629 | +static const struct nf_queue_handler __rcu *queue_imq_handler __read_mostly; | |
1630 | + | |
1631 | +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh) | |
1632 | +{ | |
1633 | + rcu_assign_pointer(queue_imq_handler, qh); | |
1634 | +} | |
1635 | +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler); | |
1636 | + | |
1637 | +void nf_unregister_queue_imq_handler(void) | |
1638 | +{ | |
1639 | + RCU_INIT_POINTER(queue_imq_handler, NULL); | |
1640 | + synchronize_rcu(); | |
1641 | +} | |
1642 | +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler); | |
1643 | +#endif | |
1644 | + | |
1645 | /* return EBUSY when somebody else is registered, return EEXIST if the | |
1646 | * same handler is registered, return 0 in case of success. */ | |
1647 | void nf_register_queue_handler(const struct nf_queue_handler *qh) | |
1648 | @@ -105,7 +122,8 @@ int nf_queue(struct sk_buff *skb, | |
1649 | struct net_device *indev, | |
1650 | struct net_device *outdev, | |
1651 | int (*okfn)(struct sk_buff *), | |
1652 | - unsigned int queuenum) | |
1653 | + unsigned int queuenum, | |
1654 | + unsigned int queuetype) | |
1655 | { | |
1656 | int status = -ENOENT; | |
1657 | struct nf_queue_entry *entry = NULL; | |
1658 | @@ -115,7 +133,17 @@ int nf_queue(struct sk_buff *skb, | |
1659 | /* QUEUE == DROP if no one is waiting, to be safe. */ | |
1660 | rcu_read_lock(); | |
1661 | ||
1662 | - qh = rcu_dereference(queue_handler); | |
1663 | + if (queuetype == NF_IMQ_QUEUE) { | |
1664 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) | |
1665 | + qh = rcu_dereference(queue_imq_handler); | |
1666 | +#else | |
1667 | + BUG(); | |
1668 | + goto err_unlock; | |
1669 | +#endif | |
1670 | + } else { | |
1671 | + qh = rcu_dereference(queue_handler); | |
1672 | + } | |
1673 | + | |
1674 | if (!qh) { | |
1675 | status = -ESRCH; | |
1676 | goto err_unlock; | |
1677 | @@ -205,9 +233,11 @@ void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict) | |
1678 | local_bh_enable(); | |
1679 | break; | |
1680 | case NF_QUEUE: | |
1681 | + case NF_IMQ_QUEUE: | |
1682 | err = nf_queue(skb, elem, entry->pf, entry->hook, | |
1683 | entry->indev, entry->outdev, entry->okfn, | |
1684 | - verdict >> NF_VERDICT_QBITS); | |
1685 | + verdict >> NF_VERDICT_QBITS, | |
1686 | + verdict & NF_VERDICT_MASK); | |
1687 | if (err < 0) { | |
1688 | if (err == -ECANCELED) | |
1689 | goto next_hook; | |
1690 | diff --git a/net/netfilter/xt_IMQ.c b/net/netfilter/xt_IMQ.c | |
1691 | new file mode 100644 | |
1692 | index 0000000..1c3cd66 | |
1693 | --- /dev/null | |
1694 | +++ b/net/netfilter/xt_IMQ.c | |
1695 | @@ -0,0 +1,72 @@ | |
1696 | +/* | |
1697 | + * This target marks packets to be enqueued to an imq device | |
1698 | + */ | |
1699 | +#include <linux/module.h> | |
1700 | +#include <linux/skbuff.h> | |
1701 | +#include <linux/netfilter/x_tables.h> | |
1702 | +#include <linux/netfilter/xt_IMQ.h> | |
1703 | +#include <linux/imq.h> | |
1704 | + | |
1705 | +static unsigned int imq_target(struct sk_buff *pskb, | |
1706 | + const struct xt_action_param *par) | |
1707 | +{ | |
1708 | + const struct xt_imq_info *mr = par->targinfo; | |
1709 | + | |
1710 | + pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE; | |
1711 | + | |
1712 | + return XT_CONTINUE; | |
1713 | +} | |
1714 | + | |
1715 | +static int imq_checkentry(const struct xt_tgchk_param *par) | |
1716 | +{ | |
1717 | + struct xt_imq_info *mr = par->targinfo; | |
1718 | + | |
1719 | + if (mr->todev > IMQ_MAX_DEVS - 1) { | |
1720 | + pr_warn("IMQ: invalid device specified, highest is %u\n", | |
1721 | + IMQ_MAX_DEVS - 1); | |
1722 | + return -EINVAL; | |
1723 | + } | |
1724 | + | |
1725 | + return 0; | |
1726 | +} | |
1727 | + | |
1728 | +static struct xt_target xt_imq_reg[] __read_mostly = { | |
1729 | + { | |
1730 | + .name = "IMQ", | |
1731 | + .family = AF_INET, | |
1732 | + .checkentry = imq_checkentry, | |
1733 | + .target = imq_target, | |
1734 | + .targetsize = sizeof(struct xt_imq_info), | |
1735 | + .table = "mangle", | |
1736 | + .me = THIS_MODULE | |
1737 | + }, | |
1738 | + { | |
1739 | + .name = "IMQ", | |
1740 | + .family = AF_INET6, | |
1741 | + .checkentry = imq_checkentry, | |
1742 | + .target = imq_target, | |
1743 | + .targetsize = sizeof(struct xt_imq_info), | |
1744 | + .table = "mangle", | |
1745 | + .me = THIS_MODULE | |
1746 | + }, | |
1747 | +}; | |
1748 | + | |
1749 | +static int __init imq_init(void) | |
1750 | +{ | |
1751 | + return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg)); | |
1752 | +} | |
1753 | + | |
1754 | +static void __exit imq_fini(void) | |
1755 | +{ | |
1756 | + xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg)); | |
1757 | +} | |
1758 | + | |
1759 | +module_init(imq_init); | |
1760 | +module_exit(imq_fini); | |
1761 | + | |
1762 | +MODULE_AUTHOR("http://www.linuximq.net"); | |
1763 | +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information."); | |
1764 | +MODULE_LICENSE("GPL"); | |
1765 | +MODULE_ALIAS("ipt_IMQ"); | |
1766 | +MODULE_ALIAS("ip6t_IMQ"); | |
1767 | + |