YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 1 | /* net/sched/sch_ingress.c - Ingress qdisc |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | * This program is free software; you can redistribute it and/or |
| 3 | * modify it under the terms of the GNU General Public License |
| 4 | * as published by the Free Software Foundation; either version |
| 5 | * 2 of the License, or (at your option) any later version. |
| 6 | * |
| 7 | * Authors: Jamal Hadi Salim 1999 |
| 8 | */ |
| 9 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 10 | #include <linux/module.h> |
| 11 | #include <linux/types.h> |
| 12 | #include <linux/skbuff.h> |
| 13 | #include <linux/netdevice.h> |
| 14 | #include <linux/rtnetlink.h> |
| 15 | #include <linux/netfilter_ipv4.h> |
| 16 | #include <linux/netfilter_ipv6.h> |
| 17 | #include <linux/netfilter.h> |
| 18 | #include <linux/smp.h> |
Arnaldo Carvalho de Melo | dc5fc57 | 2007-03-25 23:06:12 -0700 | [diff] [blame] | 19 | #include <net/netlink.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include <net/pkt_sched.h> |
| 21 | #include <asm/byteorder.h> |
| 22 | #include <asm/uaccess.h> |
| 23 | #include <linux/kmod.h> |
| 24 | #include <linux/stat.h> |
| 25 | #include <linux/interrupt.h> |
| 26 | #include <linux/list.h> |
| 27 | |
| 28 | |
| 29 | #undef DEBUG_INGRESS |
| 30 | |
| 31 | #ifdef DEBUG_INGRESS /* control */ |
| 32 | #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args) |
| 33 | #else |
| 34 | #define DPRINTK(format,args...) |
| 35 | #endif |
| 36 | |
| 37 | #if 0 /* data */ |
| 38 | #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args) |
| 39 | #else |
| 40 | #define D2PRINTK(format,args...) |
| 41 | #endif |
| 42 | |
| 43 | |
| 44 | #define PRIV(sch) qdisc_priv(sch) |
| 45 | |
| 46 | |
| 47 | /* Thanks to Doron Oz for this hack |
| 48 | */ |
| 49 | #ifndef CONFIG_NET_CLS_ACT |
| 50 | #ifdef CONFIG_NETFILTER |
YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 51 | static int nf_registered; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | #endif |
| 53 | #endif |
| 54 | |
| 55 | struct ingress_qdisc_data { |
| 56 | struct Qdisc *q; |
| 57 | struct tcf_proto *filter_list; |
| 58 | }; |
| 59 | |
| 60 | |
| 61 | /* ------------------------- Class/flow operations ------------------------- */ |
| 62 | |
| 63 | |
| 64 | static int ingress_graft(struct Qdisc *sch,unsigned long arg, |
| 65 | struct Qdisc *new,struct Qdisc **old) |
| 66 | { |
| 67 | #ifdef DEBUG_INGRESS |
| 68 | struct ingress_qdisc_data *p = PRIV(sch); |
| 69 | #endif |
| 70 | |
| 71 | DPRINTK("ingress_graft(sch %p,[qdisc %p],new %p,old %p)\n", |
| 72 | sch, p, new, old); |
| 73 | DPRINTK("\n ingress_graft: You cannot add qdiscs to classes"); |
YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 74 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | } |
| 76 | |
| 77 | |
| 78 | static struct Qdisc *ingress_leaf(struct Qdisc *sch, unsigned long arg) |
| 79 | { |
| 80 | return NULL; |
| 81 | } |
| 82 | |
| 83 | |
| 84 | static unsigned long ingress_get(struct Qdisc *sch,u32 classid) |
| 85 | { |
| 86 | #ifdef DEBUG_INGRESS |
| 87 | struct ingress_qdisc_data *p = PRIV(sch); |
| 88 | #endif |
| 89 | DPRINTK("ingress_get(sch %p,[qdisc %p],classid %x)\n", sch, p, classid); |
| 90 | return TC_H_MIN(classid) + 1; |
| 91 | } |
| 92 | |
| 93 | |
| 94 | static unsigned long ingress_bind_filter(struct Qdisc *sch, |
| 95 | unsigned long parent, u32 classid) |
| 96 | { |
| 97 | return ingress_get(sch, classid); |
| 98 | } |
| 99 | |
| 100 | |
| 101 | static void ingress_put(struct Qdisc *sch, unsigned long cl) |
| 102 | { |
| 103 | } |
| 104 | |
| 105 | |
| 106 | static int ingress_change(struct Qdisc *sch, u32 classid, u32 parent, |
| 107 | struct rtattr **tca, unsigned long *arg) |
| 108 | { |
| 109 | #ifdef DEBUG_INGRESS |
| 110 | struct ingress_qdisc_data *p = PRIV(sch); |
| 111 | #endif |
| 112 | DPRINTK("ingress_change(sch %p,[qdisc %p],classid %x,parent %x)," |
| 113 | "arg 0x%lx\n", sch, p, classid, parent, *arg); |
| 114 | DPRINTK("No effect. sch_ingress doesn't maintain classes at the moment"); |
| 115 | return 0; |
| 116 | } |
| 117 | |
| 118 | |
| 119 | |
| 120 | static void ingress_walk(struct Qdisc *sch,struct qdisc_walker *walker) |
| 121 | { |
| 122 | #ifdef DEBUG_INGRESS |
| 123 | struct ingress_qdisc_data *p = PRIV(sch); |
| 124 | #endif |
| 125 | DPRINTK("ingress_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker); |
| 126 | DPRINTK("No effect. sch_ingress doesn't maintain classes at the moment"); |
| 127 | } |
| 128 | |
| 129 | |
| 130 | static struct tcf_proto **ingress_find_tcf(struct Qdisc *sch,unsigned long cl) |
| 131 | { |
| 132 | struct ingress_qdisc_data *p = PRIV(sch); |
| 133 | |
| 134 | return &p->filter_list; |
| 135 | } |
| 136 | |
| 137 | |
| 138 | /* --------------------------- Qdisc operations ---------------------------- */ |
| 139 | |
| 140 | |
| 141 | static int ingress_enqueue(struct sk_buff *skb,struct Qdisc *sch) |
| 142 | { |
| 143 | struct ingress_qdisc_data *p = PRIV(sch); |
| 144 | struct tcf_result res; |
| 145 | int result; |
| 146 | |
| 147 | D2PRINTK("ingress_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p); |
| 148 | result = tc_classify(skb, p->filter_list, &res); |
| 149 | D2PRINTK("result %d class 0x%04x\n", result, res.classid); |
| 150 | /* |
| 151 | * Unlike normal "enqueue" functions, ingress_enqueue returns a |
| 152 | * firewall FW_* code. |
| 153 | */ |
| 154 | #ifdef CONFIG_NET_CLS_ACT |
| 155 | sch->bstats.packets++; |
| 156 | sch->bstats.bytes += skb->len; |
| 157 | switch (result) { |
| 158 | case TC_ACT_SHOT: |
| 159 | result = TC_ACT_SHOT; |
| 160 | sch->qstats.drops++; |
| 161 | break; |
| 162 | case TC_ACT_STOLEN: |
| 163 | case TC_ACT_QUEUED: |
| 164 | result = TC_ACT_STOLEN; |
| 165 | break; |
YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 166 | case TC_ACT_RECLASSIFY: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | case TC_ACT_OK: |
| 168 | case TC_ACT_UNSPEC: |
| 169 | default: |
| 170 | skb->tc_index = TC_H_MIN(res.classid); |
| 171 | result = TC_ACT_OK; |
| 172 | break; |
| 173 | }; |
| 174 | /* backward compat */ |
| 175 | #else |
YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 176 | #ifdef CONFIG_NET_CLS_POLICE |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | switch (result) { |
| 178 | case TC_POLICE_SHOT: |
| 179 | result = NF_DROP; |
| 180 | sch->qstats.drops++; |
| 181 | break; |
| 182 | case TC_POLICE_RECLASSIFY: /* DSCP remarking here ? */ |
| 183 | case TC_POLICE_OK: |
| 184 | case TC_POLICE_UNSPEC: |
| 185 | default: |
| 186 | sch->bstats.packets++; |
| 187 | sch->bstats.bytes += skb->len; |
| 188 | result = NF_ACCEPT; |
| 189 | break; |
| 190 | }; |
| 191 | |
| 192 | #else |
| 193 | D2PRINTK("Overriding result to ACCEPT\n"); |
| 194 | result = NF_ACCEPT; |
| 195 | sch->bstats.packets++; |
| 196 | sch->bstats.bytes += skb->len; |
| 197 | #endif |
| 198 | #endif |
| 199 | |
| 200 | return result; |
| 201 | } |
| 202 | |
| 203 | |
| 204 | static struct sk_buff *ingress_dequeue(struct Qdisc *sch) |
| 205 | { |
| 206 | /* |
| 207 | struct ingress_qdisc_data *p = PRIV(sch); |
| 208 | D2PRINTK("ingress_dequeue(sch %p,[qdisc %p])\n",sch,PRIV(p)); |
| 209 | */ |
| 210 | return NULL; |
| 211 | } |
| 212 | |
| 213 | |
| 214 | static int ingress_requeue(struct sk_buff *skb,struct Qdisc *sch) |
| 215 | { |
| 216 | /* |
| 217 | struct ingress_qdisc_data *p = PRIV(sch); |
| 218 | D2PRINTK("ingress_requeue(skb %p,sch %p,[qdisc %p])\n",skb,sch,PRIV(p)); |
| 219 | */ |
| 220 | return 0; |
| 221 | } |
| 222 | |
| 223 | static unsigned int ingress_drop(struct Qdisc *sch) |
| 224 | { |
| 225 | #ifdef DEBUG_INGRESS |
| 226 | struct ingress_qdisc_data *p = PRIV(sch); |
| 227 | #endif |
| 228 | DPRINTK("ingress_drop(sch %p,[qdisc %p])\n", sch, p); |
| 229 | return 0; |
| 230 | } |
| 231 | |
| 232 | #ifndef CONFIG_NET_CLS_ACT |
| 233 | #ifdef CONFIG_NETFILTER |
| 234 | static unsigned int |
| 235 | ing_hook(unsigned int hook, struct sk_buff **pskb, |
YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 236 | const struct net_device *indev, |
| 237 | const struct net_device *outdev, |
| 238 | int (*okfn)(struct sk_buff *)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | { |
YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 240 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 241 | struct Qdisc *q; |
| 242 | struct sk_buff *skb = *pskb; |
YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 243 | struct net_device *dev = skb->dev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | int fwres=NF_ACCEPT; |
| 245 | |
| 246 | DPRINTK("ing_hook: skb %s dev=%s len=%u\n", |
| 247 | skb->sk ? "(owned)" : "(unowned)", |
| 248 | skb->dev ? (*pskb)->dev->name : "(no dev)", |
| 249 | skb->len); |
| 250 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | if (dev->qdisc_ingress) { |
Patrick McHardy | fd44de7 | 2007-04-16 17:07:08 -0700 | [diff] [blame^] | 252 | spin_lock(&dev->ingress_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 253 | if ((q = dev->qdisc_ingress) != NULL) |
| 254 | fwres = q->enqueue(skb, q); |
Patrick McHardy | fd44de7 | 2007-04-16 17:07:08 -0700 | [diff] [blame^] | 255 | spin_unlock(&dev->ingress_lock); |
YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 256 | } |
| 257 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | return fwres; |
| 259 | } |
| 260 | |
| 261 | /* after ipt_filter */ |
| 262 | static struct nf_hook_ops ing_ops = { |
| 263 | .hook = ing_hook, |
| 264 | .owner = THIS_MODULE, |
| 265 | .pf = PF_INET, |
| 266 | .hooknum = NF_IP_PRE_ROUTING, |
| 267 | .priority = NF_IP_PRI_FILTER + 1, |
| 268 | }; |
| 269 | |
| 270 | static struct nf_hook_ops ing6_ops = { |
| 271 | .hook = ing_hook, |
| 272 | .owner = THIS_MODULE, |
| 273 | .pf = PF_INET6, |
| 274 | .hooknum = NF_IP6_PRE_ROUTING, |
| 275 | .priority = NF_IP6_PRI_FILTER + 1, |
| 276 | }; |
| 277 | |
| 278 | #endif |
| 279 | #endif |
| 280 | |
| 281 | static int ingress_init(struct Qdisc *sch,struct rtattr *opt) |
| 282 | { |
| 283 | struct ingress_qdisc_data *p = PRIV(sch); |
| 284 | |
| 285 | /* Make sure either netfilter or preferably CLS_ACT is |
| 286 | * compiled in */ |
| 287 | #ifndef CONFIG_NET_CLS_ACT |
| 288 | #ifndef CONFIG_NETFILTER |
| 289 | printk("You MUST compile classifier actions into the kernel\n"); |
| 290 | return -EINVAL; |
| 291 | #else |
| 292 | printk("Ingress scheduler: Classifier actions prefered over netfilter\n"); |
| 293 | #endif |
| 294 | #endif |
YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 295 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 296 | #ifndef CONFIG_NET_CLS_ACT |
| 297 | #ifdef CONFIG_NETFILTER |
| 298 | if (!nf_registered) { |
| 299 | if (nf_register_hook(&ing_ops) < 0) { |
| 300 | printk("ingress qdisc registration error \n"); |
| 301 | return -EINVAL; |
| 302 | } |
| 303 | nf_registered++; |
| 304 | |
| 305 | if (nf_register_hook(&ing6_ops) < 0) { |
| 306 | printk("IPv6 ingress qdisc registration error, " \ |
| 307 | "disabling IPv6 support.\n"); |
| 308 | } else |
| 309 | nf_registered++; |
| 310 | } |
| 311 | #endif |
| 312 | #endif |
| 313 | |
| 314 | DPRINTK("ingress_init(sch %p,[qdisc %p],opt %p)\n",sch,p,opt); |
| 315 | p->q = &noop_qdisc; |
| 316 | return 0; |
| 317 | } |
| 318 | |
| 319 | |
| 320 | static void ingress_reset(struct Qdisc *sch) |
| 321 | { |
| 322 | struct ingress_qdisc_data *p = PRIV(sch); |
| 323 | |
| 324 | DPRINTK("ingress_reset(sch %p,[qdisc %p])\n", sch, p); |
| 325 | |
| 326 | /* |
| 327 | #if 0 |
| 328 | */ |
| 329 | /* for future use */ |
| 330 | qdisc_reset(p->q); |
| 331 | /* |
| 332 | #endif |
| 333 | */ |
| 334 | } |
| 335 | |
| 336 | /* ------------------------------------------------------------- */ |
| 337 | |
| 338 | |
| 339 | /* ------------------------------------------------------------- */ |
| 340 | |
| 341 | static void ingress_destroy(struct Qdisc *sch) |
| 342 | { |
| 343 | struct ingress_qdisc_data *p = PRIV(sch); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | |
| 345 | DPRINTK("ingress_destroy(sch %p,[qdisc %p])\n", sch, p); |
Patrick McHardy | a48b5a6 | 2007-03-23 11:29:43 -0700 | [diff] [blame] | 346 | tcf_destroy_chain(p->filter_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 347 | #if 0 |
| 348 | /* for future use */ |
| 349 | qdisc_destroy(p->q); |
| 350 | #endif |
| 351 | } |
| 352 | |
| 353 | |
| 354 | static int ingress_dump(struct Qdisc *sch, struct sk_buff *skb) |
| 355 | { |
Arnaldo Carvalho de Melo | 27a884d | 2007-04-19 20:29:13 -0700 | [diff] [blame] | 356 | unsigned char *b = skb_tail_pointer(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 357 | struct rtattr *rta; |
| 358 | |
| 359 | rta = (struct rtattr *) b; |
| 360 | RTA_PUT(skb, TCA_OPTIONS, 0, NULL); |
Arnaldo Carvalho de Melo | 27a884d | 2007-04-19 20:29:13 -0700 | [diff] [blame] | 361 | rta->rta_len = skb_tail_pointer(skb) - b; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | return skb->len; |
| 363 | |
| 364 | rtattr_failure: |
Arnaldo Carvalho de Melo | dc5fc57 | 2007-03-25 23:06:12 -0700 | [diff] [blame] | 365 | nlmsg_trim(skb, b); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 366 | return -1; |
| 367 | } |
| 368 | |
| 369 | static struct Qdisc_class_ops ingress_class_ops = { |
| 370 | .graft = ingress_graft, |
| 371 | .leaf = ingress_leaf, |
| 372 | .get = ingress_get, |
| 373 | .put = ingress_put, |
| 374 | .change = ingress_change, |
| 375 | .delete = NULL, |
| 376 | .walk = ingress_walk, |
| 377 | .tcf_chain = ingress_find_tcf, |
| 378 | .bind_tcf = ingress_bind_filter, |
| 379 | .unbind_tcf = ingress_put, |
| 380 | .dump = NULL, |
| 381 | }; |
| 382 | |
| 383 | static struct Qdisc_ops ingress_qdisc_ops = { |
| 384 | .next = NULL, |
| 385 | .cl_ops = &ingress_class_ops, |
| 386 | .id = "ingress", |
| 387 | .priv_size = sizeof(struct ingress_qdisc_data), |
| 388 | .enqueue = ingress_enqueue, |
| 389 | .dequeue = ingress_dequeue, |
| 390 | .requeue = ingress_requeue, |
| 391 | .drop = ingress_drop, |
| 392 | .init = ingress_init, |
| 393 | .reset = ingress_reset, |
| 394 | .destroy = ingress_destroy, |
| 395 | .change = NULL, |
| 396 | .dump = ingress_dump, |
| 397 | .owner = THIS_MODULE, |
| 398 | }; |
| 399 | |
| 400 | static int __init ingress_module_init(void) |
| 401 | { |
| 402 | int ret = 0; |
| 403 | |
| 404 | if ((ret = register_qdisc(&ingress_qdisc_ops)) < 0) { |
| 405 | printk("Unable to register Ingress qdisc\n"); |
| 406 | return ret; |
| 407 | } |
| 408 | |
| 409 | return ret; |
| 410 | } |
YOSHIFUJI Hideaki | 10297b9 | 2007-02-09 23:25:16 +0900 | [diff] [blame] | 411 | static void __exit ingress_module_exit(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 412 | { |
| 413 | unregister_qdisc(&ingress_qdisc_ops); |
| 414 | #ifndef CONFIG_NET_CLS_ACT |
| 415 | #ifdef CONFIG_NETFILTER |
| 416 | if (nf_registered) { |
| 417 | nf_unregister_hook(&ing_ops); |
| 418 | if (nf_registered > 1) |
| 419 | nf_unregister_hook(&ing6_ops); |
| 420 | } |
| 421 | #endif |
| 422 | #endif |
| 423 | } |
| 424 | module_init(ingress_module_init) |
| 425 | module_exit(ingress_module_exit) |
| 426 | MODULE_LICENSE("GPL"); |