Linux-2.6.12-rc2
[linux-2.6.git] / net / sched / sch_ingress.c
1 /* net/sched/sch_ingress.c - Ingress qdisc 
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
10 #include <linux/config.h>
11 #include <linux/module.h>
12 #include <linux/types.h>
13 #include <linux/skbuff.h>
14 #include <linux/netdevice.h>
15 #include <linux/rtnetlink.h>
16 #include <linux/netfilter_ipv4.h>
17 #include <linux/netfilter_ipv6.h>
18 #include <linux/netfilter.h>
19 #include <linux/smp.h>
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
51 static int nf_registered; 
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");
74         return 1;
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;
166                 case TC_ACT_RECLASSIFY: 
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
176 #ifdef  CONFIG_NET_CLS_POLICE  
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,
236                              const struct net_device *indev,
237                              const struct net_device *outdev,
238                              int (*okfn)(struct sk_buff *))
239 {
240         
241         struct Qdisc *q;
242         struct sk_buff *skb = *pskb;
243         struct net_device *dev = skb->dev;
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
251 /* 
252 revisit later: Use a private since lock dev->queue_lock is also
253 used on the egress (might slow things for an iota)
254 */
255
256         if (dev->qdisc_ingress) {
257                 spin_lock(&dev->queue_lock);
258                 if ((q = dev->qdisc_ingress) != NULL)
259                         fwres = q->enqueue(skb, q);
260                 spin_unlock(&dev->queue_lock);
261         }
262                         
263         return fwres;
264 }
265
266 /* after ipt_filter */
267 static struct nf_hook_ops ing_ops = {
268         .hook           = ing_hook,
269         .owner          = THIS_MODULE,
270         .pf             = PF_INET,
271         .hooknum        = NF_IP_PRE_ROUTING,
272         .priority       = NF_IP_PRI_FILTER + 1,
273 };
274
275 static struct nf_hook_ops ing6_ops = {
276         .hook           = ing_hook,
277         .owner          = THIS_MODULE,
278         .pf             = PF_INET6,
279         .hooknum        = NF_IP6_PRE_ROUTING,
280         .priority       = NF_IP6_PRI_FILTER + 1,
281 };
282
283 #endif
284 #endif
285
286 static int ingress_init(struct Qdisc *sch,struct rtattr *opt)
287 {
288         struct ingress_qdisc_data *p = PRIV(sch);
289
290 /* Make sure either netfilter or preferably CLS_ACT is
291 * compiled in */
292 #ifndef CONFIG_NET_CLS_ACT
293 #ifndef CONFIG_NETFILTER
294         printk("You MUST compile classifier actions into the kernel\n");
295         return -EINVAL;
296 #else
297         printk("Ingress scheduler: Classifier actions prefered over netfilter\n");
298 #endif
299 #endif
300                                                                                 
301 #ifndef CONFIG_NET_CLS_ACT
302 #ifdef CONFIG_NETFILTER
303         if (!nf_registered) {
304                 if (nf_register_hook(&ing_ops) < 0) {
305                         printk("ingress qdisc registration error \n");
306                         return -EINVAL;
307                 }
308                 nf_registered++;
309
310                 if (nf_register_hook(&ing6_ops) < 0) {
311                         printk("IPv6 ingress qdisc registration error, " \
312                             "disabling IPv6 support.\n");
313                 } else
314                         nf_registered++;
315         }
316 #endif
317 #endif
318
319         DPRINTK("ingress_init(sch %p,[qdisc %p],opt %p)\n",sch,p,opt);
320         p->q = &noop_qdisc;
321         return 0;
322 }
323
324
325 static void ingress_reset(struct Qdisc *sch)
326 {
327         struct ingress_qdisc_data *p = PRIV(sch);
328
329         DPRINTK("ingress_reset(sch %p,[qdisc %p])\n", sch, p);
330
331 /*
332 #if 0
333 */
334 /* for future use */
335         qdisc_reset(p->q);
336 /*
337 #endif
338 */
339 }
340
341 /* ------------------------------------------------------------- */
342
343
344 /* ------------------------------------------------------------- */
345
346 static void ingress_destroy(struct Qdisc *sch)
347 {
348         struct ingress_qdisc_data *p = PRIV(sch);
349         struct tcf_proto *tp;
350
351         DPRINTK("ingress_destroy(sch %p,[qdisc %p])\n", sch, p);
352         while (p->filter_list) {
353                 tp = p->filter_list;
354                 p->filter_list = tp->next;
355                 tcf_destroy(tp);
356         }
357 #if 0
358 /* for future use */
359         qdisc_destroy(p->q);
360 #endif
361 }
362
363
364 static int ingress_dump(struct Qdisc *sch, struct sk_buff *skb)
365 {
366         unsigned char *b = skb->tail;
367         struct rtattr *rta;
368
369         rta = (struct rtattr *) b;
370         RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
371         rta->rta_len = skb->tail - b;
372         return skb->len;
373
374 rtattr_failure:
375         skb_trim(skb, b - skb->data);
376         return -1;
377 }
378
379 static struct Qdisc_class_ops ingress_class_ops = {
380         .graft          =       ingress_graft,
381         .leaf           =       ingress_leaf,
382         .get            =       ingress_get,
383         .put            =       ingress_put,
384         .change         =       ingress_change,
385         .delete         =       NULL,
386         .walk           =       ingress_walk,
387         .tcf_chain      =       ingress_find_tcf,
388         .bind_tcf       =       ingress_bind_filter,
389         .unbind_tcf     =       ingress_put,
390         .dump           =       NULL,
391 };
392
393 static struct Qdisc_ops ingress_qdisc_ops = {
394         .next           =       NULL,
395         .cl_ops         =       &ingress_class_ops,
396         .id             =       "ingress",
397         .priv_size      =       sizeof(struct ingress_qdisc_data),
398         .enqueue        =       ingress_enqueue,
399         .dequeue        =       ingress_dequeue,
400         .requeue        =       ingress_requeue,
401         .drop           =       ingress_drop,
402         .init           =       ingress_init,
403         .reset          =       ingress_reset,
404         .destroy        =       ingress_destroy,
405         .change         =       NULL,
406         .dump           =       ingress_dump,
407         .owner          =       THIS_MODULE,
408 };
409
410 static int __init ingress_module_init(void)
411 {
412         int ret = 0;
413
414         if ((ret = register_qdisc(&ingress_qdisc_ops)) < 0) {
415                 printk("Unable to register Ingress qdisc\n");
416                 return ret;
417         }
418
419         return ret;
420 }
421 static void __exit ingress_module_exit(void) 
422 {
423         unregister_qdisc(&ingress_qdisc_ops);
424 #ifndef CONFIG_NET_CLS_ACT
425 #ifdef CONFIG_NETFILTER
426         if (nf_registered) {
427                 nf_unregister_hook(&ing_ops);
428                 if (nf_registered > 1)
429                         nf_unregister_hook(&ing6_ops);
430         }
431 #endif
432 #endif
433 }
434 module_init(ingress_module_init)
435 module_exit(ingress_module_exit)
436 MODULE_LICENSE("GPL");