]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * net/sched/cls_u32.c Ugly (or Universal) 32bit key Packet Classifier. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> | |
10 | * | |
11 | * The filters are packed to hash tables of key nodes | |
12 | * with a set of 32bit key/mask pairs at every node. | |
13 | * Nodes reference next level hash tables etc. | |
14 | * | |
15 | * This scheme is the best universal classifier I managed to | |
16 | * invent; it is not super-fast, but it is not slow (provided you | |
17 | * program it correctly), and general enough. And its relative | |
18 | * speed grows as the number of rules becomes larger. | |
19 | * | |
20 | * It seems that it represents the best middle point between | |
21 | * speed and manageability both by human and by machine. | |
22 | * | |
23 | * It is especially useful for link sharing combined with QoS; | |
24 | * pure RSVP doesn't need such a general approach and can use | |
25 | * much simpler (and faster) schemes, sort of cls_rsvp.c. | |
26 | * | |
27 | * JHS: We should remove the CONFIG_NET_CLS_IND from here | |
28 | * eventually when the meta match extension is made available | |
29 | * | |
30 | * nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro> | |
31 | */ | |
32 | ||
1da177e4 | 33 | #include <linux/module.h> |
5a0e3ad6 | 34 | #include <linux/slab.h> |
1da177e4 LT |
35 | #include <linux/types.h> |
36 | #include <linux/kernel.h> | |
1da177e4 | 37 | #include <linux/string.h> |
1da177e4 | 38 | #include <linux/errno.h> |
1ce87720 | 39 | #include <linux/percpu.h> |
1da177e4 | 40 | #include <linux/rtnetlink.h> |
1da177e4 | 41 | #include <linux/skbuff.h> |
7801db8a | 42 | #include <linux/bitmap.h> |
0ba48053 | 43 | #include <net/netlink.h> |
1da177e4 LT |
44 | #include <net/act_api.h> |
45 | #include <net/pkt_cls.h> | |
46 | ||
cc7ec456 | 47 | struct tc_u_knode { |
1ce87720 | 48 | struct tc_u_knode __rcu *next; |
1da177e4 | 49 | u32 handle; |
1ce87720 | 50 | struct tc_u_hnode __rcu *ht_up; |
1da177e4 LT |
51 | struct tcf_exts exts; |
52 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 | 53 | int ifindex; |
1da177e4 LT |
54 | #endif |
55 | u8 fshift; | |
56 | struct tcf_result res; | |
1ce87720 | 57 | struct tc_u_hnode __rcu *ht_down; |
1da177e4 | 58 | #ifdef CONFIG_CLS_U32_PERF |
459d5f62 | 59 | struct tc_u32_pcnt __percpu *pf; |
1da177e4 LT |
60 | #endif |
61 | #ifdef CONFIG_CLS_U32_MARK | |
459d5f62 JF |
62 | u32 val; |
63 | u32 mask; | |
64 | u32 __percpu *pcpu_success; | |
1da177e4 | 65 | #endif |
1ce87720 | 66 | struct tcf_proto *tp; |
1ce87720 | 67 | struct rcu_head rcu; |
4e2840ee JF |
68 | /* The 'sel' field MUST be the last field in structure to allow for |
69 | * tc_u32_keys allocated at end of structure. | |
70 | */ | |
71 | struct tc_u32_sel sel; | |
1da177e4 LT |
72 | }; |
73 | ||
cc7ec456 | 74 | struct tc_u_hnode { |
1ce87720 | 75 | struct tc_u_hnode __rcu *next; |
1da177e4 LT |
76 | u32 handle; |
77 | u32 prio; | |
78 | struct tc_u_common *tp_c; | |
79 | int refcnt; | |
cc7ec456 | 80 | unsigned int divisor; |
1ce87720 JF |
81 | struct tc_u_knode __rcu *ht[1]; |
82 | struct rcu_head rcu; | |
1da177e4 LT |
83 | }; |
84 | ||
cc7ec456 | 85 | struct tc_u_common { |
1ce87720 | 86 | struct tc_u_hnode __rcu *hlist; |
1da177e4 LT |
87 | struct Qdisc *q; |
88 | int refcnt; | |
89 | u32 hgenerator; | |
1ce87720 | 90 | struct rcu_head rcu; |
1da177e4 LT |
91 | }; |
92 | ||
cc7ec456 ED |
93 | static inline unsigned int u32_hash_fold(__be32 key, |
94 | const struct tc_u32_sel *sel, | |
95 | u8 fshift) | |
1da177e4 | 96 | { |
cc7ec456 | 97 | unsigned int h = ntohl(key & sel->hmask) >> fshift; |
1da177e4 LT |
98 | |
99 | return h; | |
100 | } | |
101 | ||
dc7f9f6e | 102 | static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, struct tcf_result *res) |
1da177e4 LT |
103 | { |
104 | struct { | |
105 | struct tc_u_knode *knode; | |
fbc2e7d9 | 106 | unsigned int off; |
1da177e4 LT |
107 | } stack[TC_U32_MAXDEPTH]; |
108 | ||
1ce87720 | 109 | struct tc_u_hnode *ht = rcu_dereference_bh(tp->root); |
fbc2e7d9 | 110 | unsigned int off = skb_network_offset(skb); |
1da177e4 LT |
111 | struct tc_u_knode *n; |
112 | int sdepth = 0; | |
113 | int off2 = 0; | |
114 | int sel = 0; | |
115 | #ifdef CONFIG_CLS_U32_PERF | |
116 | int j; | |
117 | #endif | |
118 | int i, r; | |
119 | ||
120 | next_ht: | |
1ce87720 | 121 | n = rcu_dereference_bh(ht->ht[sel]); |
1da177e4 LT |
122 | |
123 | next_knode: | |
124 | if (n) { | |
125 | struct tc_u32_key *key = n->sel.keys; | |
126 | ||
127 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 128 | __this_cpu_inc(n->pf->rcnt); |
1da177e4 LT |
129 | j = 0; |
130 | #endif | |
131 | ||
132 | #ifdef CONFIG_CLS_U32_MARK | |
459d5f62 | 133 | if ((skb->mark & n->mask) != n->val) { |
1ce87720 | 134 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
135 | goto next_knode; |
136 | } else { | |
459d5f62 | 137 | __this_cpu_inc(*n->pcpu_success); |
1da177e4 LT |
138 | } |
139 | #endif | |
140 | ||
cc7ec456 | 141 | for (i = n->sel.nkeys; i > 0; i--, key++) { |
66d50d25 | 142 | int toff = off + key->off + (off2 & key->offmask); |
86fce3ba | 143 | __be32 *data, hdata; |
fbc2e7d9 | 144 | |
4e18b3ed | 145 | if (skb_headroom(skb) + toff > INT_MAX) |
66d50d25 | 146 | goto out; |
147 | ||
86fce3ba | 148 | data = skb_header_pointer(skb, toff, 4, &hdata); |
fbc2e7d9 CG |
149 | if (!data) |
150 | goto out; | |
151 | if ((*data ^ key->val) & key->mask) { | |
1ce87720 | 152 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
153 | goto next_knode; |
154 | } | |
155 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 156 | __this_cpu_inc(n->pf->kcnts[j]); |
1da177e4 LT |
157 | j++; |
158 | #endif | |
159 | } | |
1ce87720 JF |
160 | |
161 | ht = rcu_dereference_bh(n->ht_down); | |
162 | if (!ht) { | |
1da177e4 | 163 | check_terminal: |
cc7ec456 | 164 | if (n->sel.flags & TC_U32_TERMINAL) { |
1da177e4 LT |
165 | |
166 | *res = n->res; | |
167 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 | 168 | if (!tcf_match_indev(skb, n->ifindex)) { |
1ce87720 | 169 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
170 | goto next_knode; |
171 | } | |
172 | #endif | |
173 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 174 | __this_cpu_inc(n->pf->rhit); |
1da177e4 LT |
175 | #endif |
176 | r = tcf_exts_exec(skb, &n->exts, res); | |
177 | if (r < 0) { | |
1ce87720 | 178 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
179 | goto next_knode; |
180 | } | |
181 | ||
182 | return r; | |
183 | } | |
1ce87720 | 184 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
185 | goto next_knode; |
186 | } | |
187 | ||
188 | /* PUSH */ | |
189 | if (sdepth >= TC_U32_MAXDEPTH) | |
190 | goto deadloop; | |
191 | stack[sdepth].knode = n; | |
fbc2e7d9 | 192 | stack[sdepth].off = off; |
1da177e4 LT |
193 | sdepth++; |
194 | ||
1ce87720 | 195 | ht = rcu_dereference_bh(n->ht_down); |
1da177e4 | 196 | sel = 0; |
fbc2e7d9 | 197 | if (ht->divisor) { |
86fce3ba | 198 | __be32 *data, hdata; |
fbc2e7d9 CG |
199 | |
200 | data = skb_header_pointer(skb, off + n->sel.hoff, 4, | |
86fce3ba | 201 | &hdata); |
fbc2e7d9 CG |
202 | if (!data) |
203 | goto out; | |
204 | sel = ht->divisor & u32_hash_fold(*data, &n->sel, | |
205 | n->fshift); | |
206 | } | |
cc7ec456 | 207 | if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT))) |
1da177e4 LT |
208 | goto next_ht; |
209 | ||
cc7ec456 | 210 | if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) { |
1da177e4 | 211 | off2 = n->sel.off + 3; |
fbc2e7d9 | 212 | if (n->sel.flags & TC_U32_VAROFFSET) { |
86fce3ba | 213 | __be16 *data, hdata; |
fbc2e7d9 CG |
214 | |
215 | data = skb_header_pointer(skb, | |
216 | off + n->sel.offoff, | |
86fce3ba | 217 | 2, &hdata); |
fbc2e7d9 CG |
218 | if (!data) |
219 | goto out; | |
220 | off2 += ntohs(n->sel.offmask & *data) >> | |
221 | n->sel.offshift; | |
222 | } | |
1da177e4 LT |
223 | off2 &= ~3; |
224 | } | |
cc7ec456 | 225 | if (n->sel.flags & TC_U32_EAT) { |
fbc2e7d9 | 226 | off += off2; |
1da177e4 LT |
227 | off2 = 0; |
228 | } | |
229 | ||
fbc2e7d9 | 230 | if (off < skb->len) |
1da177e4 LT |
231 | goto next_ht; |
232 | } | |
233 | ||
234 | /* POP */ | |
235 | if (sdepth--) { | |
236 | n = stack[sdepth].knode; | |
1ce87720 | 237 | ht = rcu_dereference_bh(n->ht_up); |
fbc2e7d9 | 238 | off = stack[sdepth].off; |
1da177e4 LT |
239 | goto check_terminal; |
240 | } | |
fbc2e7d9 | 241 | out: |
1da177e4 LT |
242 | return -1; |
243 | ||
244 | deadloop: | |
e87cc472 | 245 | net_warn_ratelimited("cls_u32: dead loop\n"); |
1da177e4 LT |
246 | return -1; |
247 | } | |
248 | ||
cc7ec456 | 249 | static struct tc_u_hnode * |
1da177e4 LT |
250 | u32_lookup_ht(struct tc_u_common *tp_c, u32 handle) |
251 | { | |
252 | struct tc_u_hnode *ht; | |
253 | ||
1ce87720 JF |
254 | for (ht = rtnl_dereference(tp_c->hlist); |
255 | ht; | |
256 | ht = rtnl_dereference(ht->next)) | |
1da177e4 LT |
257 | if (ht->handle == handle) |
258 | break; | |
259 | ||
260 | return ht; | |
261 | } | |
262 | ||
cc7ec456 | 263 | static struct tc_u_knode * |
1da177e4 LT |
264 | u32_lookup_key(struct tc_u_hnode *ht, u32 handle) |
265 | { | |
cc7ec456 | 266 | unsigned int sel; |
1da177e4 LT |
267 | struct tc_u_knode *n = NULL; |
268 | ||
269 | sel = TC_U32_HASH(handle); | |
270 | if (sel > ht->divisor) | |
271 | goto out; | |
272 | ||
1ce87720 JF |
273 | for (n = rtnl_dereference(ht->ht[sel]); |
274 | n; | |
275 | n = rtnl_dereference(n->next)) | |
1da177e4 LT |
276 | if (n->handle == handle) |
277 | break; | |
278 | out: | |
279 | return n; | |
280 | } | |
281 | ||
282 | ||
283 | static unsigned long u32_get(struct tcf_proto *tp, u32 handle) | |
284 | { | |
285 | struct tc_u_hnode *ht; | |
286 | struct tc_u_common *tp_c = tp->data; | |
287 | ||
288 | if (TC_U32_HTID(handle) == TC_U32_ROOT) | |
1ce87720 | 289 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
290 | else |
291 | ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle)); | |
292 | ||
293 | if (!ht) | |
294 | return 0; | |
295 | ||
296 | if (TC_U32_KEY(handle) == 0) | |
297 | return (unsigned long)ht; | |
298 | ||
299 | return (unsigned long)u32_lookup_key(ht, handle); | |
300 | } | |
301 | ||
302 | static void u32_put(struct tcf_proto *tp, unsigned long f) | |
303 | { | |
304 | } | |
305 | ||
306 | static u32 gen_new_htid(struct tc_u_common *tp_c) | |
307 | { | |
308 | int i = 0x800; | |
309 | ||
1ce87720 JF |
310 | /* hgenerator only used inside rtnl lock it is safe to increment |
311 | * without read _copy_ update semantics | |
312 | */ | |
1da177e4 LT |
313 | do { |
314 | if (++tp_c->hgenerator == 0x7FF) | |
315 | tp_c->hgenerator = 1; | |
cc7ec456 | 316 | } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20)); |
1da177e4 LT |
317 | |
318 | return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0; | |
319 | } | |
320 | ||
321 | static int u32_init(struct tcf_proto *tp) | |
322 | { | |
323 | struct tc_u_hnode *root_ht; | |
324 | struct tc_u_common *tp_c; | |
325 | ||
72b25a91 | 326 | tp_c = tp->q->u32_node; |
1da177e4 | 327 | |
0da974f4 | 328 | root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL); |
1da177e4 LT |
329 | if (root_ht == NULL) |
330 | return -ENOBUFS; | |
331 | ||
1da177e4 LT |
332 | root_ht->divisor = 0; |
333 | root_ht->refcnt++; | |
334 | root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000; | |
335 | root_ht->prio = tp->prio; | |
336 | ||
337 | if (tp_c == NULL) { | |
0da974f4 | 338 | tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL); |
1da177e4 LT |
339 | if (tp_c == NULL) { |
340 | kfree(root_ht); | |
341 | return -ENOBUFS; | |
342 | } | |
1da177e4 | 343 | tp_c->q = tp->q; |
72b25a91 | 344 | tp->q->u32_node = tp_c; |
1da177e4 LT |
345 | } |
346 | ||
347 | tp_c->refcnt++; | |
1ce87720 JF |
348 | RCU_INIT_POINTER(root_ht->next, tp_c->hlist); |
349 | rcu_assign_pointer(tp_c->hlist, root_ht); | |
1da177e4 LT |
350 | root_ht->tp_c = tp_c; |
351 | ||
1ce87720 | 352 | rcu_assign_pointer(tp->root, root_ht); |
1da177e4 LT |
353 | tp->data = tp_c; |
354 | return 0; | |
355 | } | |
356 | ||
357 | static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n) | |
358 | { | |
359 | tcf_unbind_filter(tp, &n->res); | |
360 | tcf_exts_destroy(tp, &n->exts); | |
361 | if (n->ht_down) | |
362 | n->ht_down->refcnt--; | |
363 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 364 | free_percpu(n->pf); |
1da177e4 LT |
365 | #endif |
366 | kfree(n); | |
367 | return 0; | |
368 | } | |
369 | ||
1ce87720 JF |
370 | static void u32_delete_key_rcu(struct rcu_head *rcu) |
371 | { | |
372 | struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); | |
373 | ||
374 | u32_destroy_key(key->tp, key); | |
375 | } | |
376 | ||
82d567c2 | 377 | static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key) |
1da177e4 | 378 | { |
1ce87720 JF |
379 | struct tc_u_knode __rcu **kp; |
380 | struct tc_u_knode *pkp; | |
a96366bf | 381 | struct tc_u_hnode *ht = rtnl_dereference(key->ht_up); |
1da177e4 LT |
382 | |
383 | if (ht) { | |
1ce87720 JF |
384 | kp = &ht->ht[TC_U32_HASH(key->handle)]; |
385 | for (pkp = rtnl_dereference(*kp); pkp; | |
386 | kp = &pkp->next, pkp = rtnl_dereference(*kp)) { | |
387 | if (pkp == key) { | |
388 | RCU_INIT_POINTER(*kp, key->next); | |
1da177e4 | 389 | |
1ce87720 | 390 | call_rcu(&key->rcu, u32_delete_key_rcu); |
1da177e4 LT |
391 | return 0; |
392 | } | |
393 | } | |
394 | } | |
547b792c | 395 | WARN_ON(1); |
1da177e4 LT |
396 | return 0; |
397 | } | |
398 | ||
1ce87720 | 399 | static void u32_clear_hnode(struct tc_u_hnode *ht) |
1da177e4 LT |
400 | { |
401 | struct tc_u_knode *n; | |
cc7ec456 | 402 | unsigned int h; |
1da177e4 | 403 | |
cc7ec456 | 404 | for (h = 0; h <= ht->divisor; h++) { |
1ce87720 JF |
405 | while ((n = rtnl_dereference(ht->ht[h])) != NULL) { |
406 | RCU_INIT_POINTER(ht->ht[h], | |
407 | rtnl_dereference(n->next)); | |
408 | call_rcu(&n->rcu, u32_delete_key_rcu); | |
1da177e4 LT |
409 | } |
410 | } | |
411 | } | |
412 | ||
413 | static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) | |
414 | { | |
415 | struct tc_u_common *tp_c = tp->data; | |
1ce87720 JF |
416 | struct tc_u_hnode __rcu **hn; |
417 | struct tc_u_hnode *phn; | |
1da177e4 | 418 | |
547b792c | 419 | WARN_ON(ht->refcnt); |
1da177e4 | 420 | |
1ce87720 | 421 | u32_clear_hnode(ht); |
1da177e4 | 422 | |
1ce87720 JF |
423 | hn = &tp_c->hlist; |
424 | for (phn = rtnl_dereference(*hn); | |
425 | phn; | |
426 | hn = &phn->next, phn = rtnl_dereference(*hn)) { | |
427 | if (phn == ht) { | |
428 | RCU_INIT_POINTER(*hn, ht->next); | |
429 | kfree_rcu(ht, rcu); | |
1da177e4 LT |
430 | return 0; |
431 | } | |
432 | } | |
433 | ||
1da177e4 LT |
434 | return -ENOENT; |
435 | } | |
436 | ||
437 | static void u32_destroy(struct tcf_proto *tp) | |
438 | { | |
439 | struct tc_u_common *tp_c = tp->data; | |
1ce87720 | 440 | struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); |
1da177e4 | 441 | |
547b792c | 442 | WARN_ON(root_ht == NULL); |
1da177e4 LT |
443 | |
444 | if (root_ht && --root_ht->refcnt == 0) | |
445 | u32_destroy_hnode(tp, root_ht); | |
446 | ||
447 | if (--tp_c->refcnt == 0) { | |
448 | struct tc_u_hnode *ht; | |
1da177e4 | 449 | |
72b25a91 | 450 | tp->q->u32_node = NULL; |
1da177e4 | 451 | |
1ce87720 JF |
452 | for (ht = rtnl_dereference(tp_c->hlist); |
453 | ht; | |
454 | ht = rtnl_dereference(ht->next)) { | |
e56cfad1 | 455 | ht->refcnt--; |
1ce87720 | 456 | u32_clear_hnode(ht); |
e56cfad1 | 457 | } |
1da177e4 | 458 | |
1ce87720 JF |
459 | while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) { |
460 | RCU_INIT_POINTER(tp_c->hlist, ht->next); | |
461 | kfree_rcu(ht, rcu); | |
3ff50b79 | 462 | } |
1da177e4 LT |
463 | |
464 | kfree(tp_c); | |
465 | } | |
466 | ||
467 | tp->data = NULL; | |
468 | } | |
469 | ||
470 | static int u32_delete(struct tcf_proto *tp, unsigned long arg) | |
471 | { | |
cc7ec456 | 472 | struct tc_u_hnode *ht = (struct tc_u_hnode *)arg; |
1ce87720 | 473 | struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); |
1da177e4 LT |
474 | |
475 | if (ht == NULL) | |
476 | return 0; | |
477 | ||
478 | if (TC_U32_KEY(ht->handle)) | |
cc7ec456 | 479 | return u32_delete_key(tp, (struct tc_u_knode *)ht); |
1da177e4 | 480 | |
1ce87720 | 481 | if (root_ht == ht) |
1da177e4 LT |
482 | return -EINVAL; |
483 | ||
e56cfad1 JP |
484 | if (ht->refcnt == 1) { |
485 | ht->refcnt--; | |
1da177e4 | 486 | u32_destroy_hnode(tp, ht); |
e56cfad1 JP |
487 | } else { |
488 | return -EBUSY; | |
489 | } | |
1da177e4 LT |
490 | |
491 | return 0; | |
492 | } | |
493 | ||
7801db8a | 494 | #define NR_U32_NODE (1<<12) |
1da177e4 LT |
495 | static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle) |
496 | { | |
497 | struct tc_u_knode *n; | |
7801db8a CW |
498 | unsigned long i; |
499 | unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long), | |
500 | GFP_KERNEL); | |
501 | if (!bitmap) | |
502 | return handle | 0xFFF; | |
1da177e4 | 503 | |
1ce87720 JF |
504 | for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]); |
505 | n; | |
506 | n = rtnl_dereference(n->next)) | |
7801db8a | 507 | set_bit(TC_U32_NODE(n->handle), bitmap); |
1da177e4 | 508 | |
7801db8a CW |
509 | i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800); |
510 | if (i >= NR_U32_NODE) | |
511 | i = find_next_zero_bit(bitmap, NR_U32_NODE, 1); | |
512 | ||
513 | kfree(bitmap); | |
514 | return handle | (i >= NR_U32_NODE ? 0xFFF : i); | |
1da177e4 LT |
515 | } |
516 | ||
6fa8c014 PM |
517 | static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = { |
518 | [TCA_U32_CLASSID] = { .type = NLA_U32 }, | |
519 | [TCA_U32_HASH] = { .type = NLA_U32 }, | |
520 | [TCA_U32_LINK] = { .type = NLA_U32 }, | |
521 | [TCA_U32_DIVISOR] = { .type = NLA_U32 }, | |
522 | [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) }, | |
523 | [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ }, | |
524 | [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) }, | |
525 | }; | |
526 | ||
c1b52739 BL |
527 | static int u32_set_parms(struct net *net, struct tcf_proto *tp, |
528 | unsigned long base, struct tc_u_hnode *ht, | |
add93b61 | 529 | struct tc_u_knode *n, struct nlattr **tb, |
2f7ef2f8 | 530 | struct nlattr *est, bool ovr) |
1da177e4 LT |
531 | { |
532 | int err; | |
533 | struct tcf_exts e; | |
534 | ||
5da57f42 | 535 | tcf_exts_init(&e, TCA_U32_ACT, TCA_U32_POLICE); |
2f7ef2f8 | 536 | err = tcf_exts_validate(net, tp, tb, est, &e, ovr); |
1da177e4 LT |
537 | if (err < 0) |
538 | return err; | |
539 | ||
540 | err = -EINVAL; | |
add93b61 | 541 | if (tb[TCA_U32_LINK]) { |
1587bac4 | 542 | u32 handle = nla_get_u32(tb[TCA_U32_LINK]); |
47a1a1d4 | 543 | struct tc_u_hnode *ht_down = NULL, *ht_old; |
1da177e4 LT |
544 | |
545 | if (TC_U32_KEY(handle)) | |
546 | goto errout; | |
547 | ||
548 | if (handle) { | |
549 | ht_down = u32_lookup_ht(ht->tp_c, handle); | |
550 | ||
551 | if (ht_down == NULL) | |
552 | goto errout; | |
553 | ht_down->refcnt++; | |
554 | } | |
555 | ||
1ce87720 JF |
556 | ht_old = rtnl_dereference(n->ht_down); |
557 | rcu_assign_pointer(n->ht_down, ht_down); | |
1da177e4 | 558 | |
47a1a1d4 PM |
559 | if (ht_old) |
560 | ht_old->refcnt--; | |
1da177e4 | 561 | } |
add93b61 | 562 | if (tb[TCA_U32_CLASSID]) { |
1587bac4 | 563 | n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]); |
1da177e4 LT |
564 | tcf_bind_filter(tp, &n->res, base); |
565 | } | |
566 | ||
567 | #ifdef CONFIG_NET_CLS_IND | |
add93b61 | 568 | if (tb[TCA_U32_INDEV]) { |
2519a602 WC |
569 | int ret; |
570 | ret = tcf_change_indev(net, tb[TCA_U32_INDEV]); | |
571 | if (ret < 0) | |
1da177e4 | 572 | goto errout; |
2519a602 | 573 | n->ifindex = ret; |
1da177e4 LT |
574 | } |
575 | #endif | |
576 | tcf_exts_change(tp, &n->exts, &e); | |
577 | ||
578 | return 0; | |
579 | errout: | |
580 | tcf_exts_destroy(tp, &e); | |
581 | return err; | |
582 | } | |
583 | ||
c1b52739 | 584 | static int u32_change(struct net *net, struct sk_buff *in_skb, |
af4c6641 | 585 | struct tcf_proto *tp, unsigned long base, u32 handle, |
add93b61 | 586 | struct nlattr **tca, |
2f7ef2f8 | 587 | unsigned long *arg, bool ovr) |
1da177e4 LT |
588 | { |
589 | struct tc_u_common *tp_c = tp->data; | |
590 | struct tc_u_hnode *ht; | |
591 | struct tc_u_knode *n; | |
592 | struct tc_u32_sel *s; | |
add93b61 PM |
593 | struct nlattr *opt = tca[TCA_OPTIONS]; |
594 | struct nlattr *tb[TCA_U32_MAX + 1]; | |
1da177e4 LT |
595 | u32 htid; |
596 | int err; | |
459d5f62 JF |
597 | #ifdef CONFIG_CLS_U32_PERF |
598 | size_t size; | |
599 | #endif | |
1da177e4 LT |
600 | |
601 | if (opt == NULL) | |
602 | return handle ? -EINVAL : 0; | |
603 | ||
6fa8c014 | 604 | err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy); |
cee63723 PM |
605 | if (err < 0) |
606 | return err; | |
1da177e4 | 607 | |
cc7ec456 ED |
608 | n = (struct tc_u_knode *)*arg; |
609 | if (n) { | |
1da177e4 LT |
610 | if (TC_U32_KEY(n->handle) == 0) |
611 | return -EINVAL; | |
612 | ||
a96366bf JF |
613 | return u32_set_parms(net, tp, base, |
614 | rtnl_dereference(n->ht_up), n, tb, | |
2f7ef2f8 | 615 | tca[TCA_RATE], ovr); |
1da177e4 LT |
616 | } |
617 | ||
add93b61 | 618 | if (tb[TCA_U32_DIVISOR]) { |
cc7ec456 | 619 | unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]); |
1da177e4 LT |
620 | |
621 | if (--divisor > 0x100) | |
622 | return -EINVAL; | |
623 | if (TC_U32_KEY(handle)) | |
624 | return -EINVAL; | |
625 | if (handle == 0) { | |
626 | handle = gen_new_htid(tp->data); | |
627 | if (handle == 0) | |
628 | return -ENOMEM; | |
629 | } | |
cc7ec456 | 630 | ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL); |
1da177e4 LT |
631 | if (ht == NULL) |
632 | return -ENOBUFS; | |
1da177e4 | 633 | ht->tp_c = tp_c; |
e56cfad1 | 634 | ht->refcnt = 1; |
1da177e4 LT |
635 | ht->divisor = divisor; |
636 | ht->handle = handle; | |
637 | ht->prio = tp->prio; | |
1ce87720 JF |
638 | RCU_INIT_POINTER(ht->next, tp_c->hlist); |
639 | rcu_assign_pointer(tp_c->hlist, ht); | |
1da177e4 LT |
640 | *arg = (unsigned long)ht; |
641 | return 0; | |
642 | } | |
643 | ||
add93b61 | 644 | if (tb[TCA_U32_HASH]) { |
1587bac4 | 645 | htid = nla_get_u32(tb[TCA_U32_HASH]); |
1da177e4 | 646 | if (TC_U32_HTID(htid) == TC_U32_ROOT) { |
1ce87720 | 647 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
648 | htid = ht->handle; |
649 | } else { | |
650 | ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid)); | |
651 | if (ht == NULL) | |
652 | return -EINVAL; | |
653 | } | |
654 | } else { | |
1ce87720 | 655 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
656 | htid = ht->handle; |
657 | } | |
658 | ||
659 | if (ht->divisor < TC_U32_HASH(htid)) | |
660 | return -EINVAL; | |
661 | ||
662 | if (handle) { | |
663 | if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid)) | |
664 | return -EINVAL; | |
665 | handle = htid | TC_U32_NODE(handle); | |
666 | } else | |
667 | handle = gen_new_kid(ht, htid); | |
668 | ||
6fa8c014 | 669 | if (tb[TCA_U32_SEL] == NULL) |
1da177e4 LT |
670 | return -EINVAL; |
671 | ||
add93b61 | 672 | s = nla_data(tb[TCA_U32_SEL]); |
1da177e4 | 673 | |
0da974f4 | 674 | n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL); |
1da177e4 LT |
675 | if (n == NULL) |
676 | return -ENOBUFS; | |
677 | ||
1da177e4 | 678 | #ifdef CONFIG_CLS_U32_PERF |
459d5f62 JF |
679 | size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64); |
680 | n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt)); | |
681 | if (!n->pf) { | |
1da177e4 LT |
682 | kfree(n); |
683 | return -ENOBUFS; | |
684 | } | |
1da177e4 LT |
685 | #endif |
686 | ||
687 | memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); | |
a96366bf | 688 | RCU_INIT_POINTER(n->ht_up, ht); |
1da177e4 | 689 | n->handle = handle; |
b2268016 | 690 | n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0; |
5da57f42 | 691 | tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE); |
1ce87720 | 692 | n->tp = tp; |
1da177e4 LT |
693 | |
694 | #ifdef CONFIG_CLS_U32_MARK | |
459d5f62 JF |
695 | n->pcpu_success = alloc_percpu(u32); |
696 | ||
add93b61 | 697 | if (tb[TCA_U32_MARK]) { |
1da177e4 LT |
698 | struct tc_u32_mark *mark; |
699 | ||
add93b61 | 700 | mark = nla_data(tb[TCA_U32_MARK]); |
459d5f62 JF |
701 | n->val = mark->val; |
702 | n->mask = mark->mask; | |
1da177e4 LT |
703 | } |
704 | #endif | |
705 | ||
2f7ef2f8 | 706 | err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr); |
1da177e4 | 707 | if (err == 0) { |
1ce87720 JF |
708 | struct tc_u_knode __rcu **ins; |
709 | struct tc_u_knode *pins; | |
710 | ||
711 | ins = &ht->ht[TC_U32_HASH(handle)]; | |
712 | for (pins = rtnl_dereference(*ins); pins; | |
713 | ins = &pins->next, pins = rtnl_dereference(*ins)) | |
714 | if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle)) | |
1da177e4 LT |
715 | break; |
716 | ||
1ce87720 JF |
717 | RCU_INIT_POINTER(n->next, pins); |
718 | rcu_assign_pointer(*ins, n); | |
1da177e4 LT |
719 | |
720 | *arg = (unsigned long)n; | |
721 | return 0; | |
722 | } | |
723 | #ifdef CONFIG_CLS_U32_PERF | |
1ce87720 | 724 | free_percpu(n->pf); |
1da177e4 LT |
725 | #endif |
726 | kfree(n); | |
727 | return err; | |
728 | } | |
729 | ||
730 | static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg) | |
731 | { | |
732 | struct tc_u_common *tp_c = tp->data; | |
733 | struct tc_u_hnode *ht; | |
734 | struct tc_u_knode *n; | |
cc7ec456 | 735 | unsigned int h; |
1da177e4 LT |
736 | |
737 | if (arg->stop) | |
738 | return; | |
739 | ||
1ce87720 JF |
740 | for (ht = rtnl_dereference(tp_c->hlist); |
741 | ht; | |
742 | ht = rtnl_dereference(ht->next)) { | |
1da177e4 LT |
743 | if (ht->prio != tp->prio) |
744 | continue; | |
745 | if (arg->count >= arg->skip) { | |
746 | if (arg->fn(tp, (unsigned long)ht, arg) < 0) { | |
747 | arg->stop = 1; | |
748 | return; | |
749 | } | |
750 | } | |
751 | arg->count++; | |
752 | for (h = 0; h <= ht->divisor; h++) { | |
1ce87720 JF |
753 | for (n = rtnl_dereference(ht->ht[h]); |
754 | n; | |
755 | n = rtnl_dereference(n->next)) { | |
1da177e4 LT |
756 | if (arg->count < arg->skip) { |
757 | arg->count++; | |
758 | continue; | |
759 | } | |
760 | if (arg->fn(tp, (unsigned long)n, arg) < 0) { | |
761 | arg->stop = 1; | |
762 | return; | |
763 | } | |
764 | arg->count++; | |
765 | } | |
766 | } | |
767 | } | |
768 | } | |
769 | ||
832d1d5b | 770 | static int u32_dump(struct net *net, struct tcf_proto *tp, unsigned long fh, |
1da177e4 LT |
771 | struct sk_buff *skb, struct tcmsg *t) |
772 | { | |
cc7ec456 | 773 | struct tc_u_knode *n = (struct tc_u_knode *)fh; |
1ce87720 | 774 | struct tc_u_hnode *ht_up, *ht_down; |
4b3550ef | 775 | struct nlattr *nest; |
1da177e4 LT |
776 | |
777 | if (n == NULL) | |
778 | return skb->len; | |
779 | ||
780 | t->tcm_handle = n->handle; | |
781 | ||
4b3550ef PM |
782 | nest = nla_nest_start(skb, TCA_OPTIONS); |
783 | if (nest == NULL) | |
784 | goto nla_put_failure; | |
1da177e4 LT |
785 | |
786 | if (TC_U32_KEY(n->handle) == 0) { | |
cc7ec456 ED |
787 | struct tc_u_hnode *ht = (struct tc_u_hnode *)fh; |
788 | u32 divisor = ht->divisor + 1; | |
789 | ||
1b34ec43 DM |
790 | if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor)) |
791 | goto nla_put_failure; | |
1da177e4 | 792 | } else { |
459d5f62 JF |
793 | #ifdef CONFIG_CLS_U32_PERF |
794 | struct tc_u32_pcnt *gpf; | |
459d5f62 | 795 | int cpu; |
80aab73d | 796 | #endif |
459d5f62 | 797 | |
1b34ec43 DM |
798 | if (nla_put(skb, TCA_U32_SEL, |
799 | sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key), | |
800 | &n->sel)) | |
801 | goto nla_put_failure; | |
1ce87720 JF |
802 | |
803 | ht_up = rtnl_dereference(n->ht_up); | |
804 | if (ht_up) { | |
1da177e4 | 805 | u32 htid = n->handle & 0xFFFFF000; |
1b34ec43 DM |
806 | if (nla_put_u32(skb, TCA_U32_HASH, htid)) |
807 | goto nla_put_failure; | |
1da177e4 | 808 | } |
1b34ec43 DM |
809 | if (n->res.classid && |
810 | nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid)) | |
811 | goto nla_put_failure; | |
1ce87720 JF |
812 | |
813 | ht_down = rtnl_dereference(n->ht_down); | |
814 | if (ht_down && | |
815 | nla_put_u32(skb, TCA_U32_LINK, ht_down->handle)) | |
1b34ec43 | 816 | goto nla_put_failure; |
1da177e4 LT |
817 | |
818 | #ifdef CONFIG_CLS_U32_MARK | |
459d5f62 JF |
819 | if ((n->val || n->mask)) { |
820 | struct tc_u32_mark mark = {.val = n->val, | |
821 | .mask = n->mask, | |
822 | .success = 0}; | |
80aab73d | 823 | int cpum; |
459d5f62 | 824 | |
80aab73d JF |
825 | for_each_possible_cpu(cpum) { |
826 | __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum); | |
459d5f62 JF |
827 | |
828 | mark.success += cnt; | |
829 | } | |
830 | ||
831 | if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark)) | |
832 | goto nla_put_failure; | |
833 | } | |
1da177e4 LT |
834 | #endif |
835 | ||
5da57f42 | 836 | if (tcf_exts_dump(skb, &n->exts) < 0) |
add93b61 | 837 | goto nla_put_failure; |
1da177e4 LT |
838 | |
839 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 WC |
840 | if (n->ifindex) { |
841 | struct net_device *dev; | |
842 | dev = __dev_get_by_index(net, n->ifindex); | |
843 | if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name)) | |
844 | goto nla_put_failure; | |
845 | } | |
1da177e4 LT |
846 | #endif |
847 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 JF |
848 | gpf = kzalloc(sizeof(struct tc_u32_pcnt) + |
849 | n->sel.nkeys * sizeof(u64), | |
850 | GFP_KERNEL); | |
851 | if (!gpf) | |
852 | goto nla_put_failure; | |
853 | ||
854 | for_each_possible_cpu(cpu) { | |
855 | int i; | |
856 | struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu); | |
857 | ||
858 | gpf->rcnt += pf->rcnt; | |
859 | gpf->rhit += pf->rhit; | |
860 | for (i = 0; i < n->sel.nkeys; i++) | |
861 | gpf->kcnts[i] += pf->kcnts[i]; | |
862 | } | |
863 | ||
1b34ec43 DM |
864 | if (nla_put(skb, TCA_U32_PCNT, |
865 | sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64), | |
459d5f62 JF |
866 | gpf)) { |
867 | kfree(gpf); | |
1b34ec43 | 868 | goto nla_put_failure; |
459d5f62 JF |
869 | } |
870 | kfree(gpf); | |
1da177e4 LT |
871 | #endif |
872 | } | |
873 | ||
4b3550ef PM |
874 | nla_nest_end(skb, nest); |
875 | ||
1da177e4 | 876 | if (TC_U32_KEY(n->handle)) |
5da57f42 | 877 | if (tcf_exts_dump_stats(skb, &n->exts) < 0) |
add93b61 | 878 | goto nla_put_failure; |
1da177e4 LT |
879 | return skb->len; |
880 | ||
add93b61 | 881 | nla_put_failure: |
4b3550ef | 882 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
883 | return -1; |
884 | } | |
885 | ||
2eb9d75c | 886 | static struct tcf_proto_ops cls_u32_ops __read_mostly = { |
1da177e4 LT |
887 | .kind = "u32", |
888 | .classify = u32_classify, | |
889 | .init = u32_init, | |
890 | .destroy = u32_destroy, | |
891 | .get = u32_get, | |
892 | .put = u32_put, | |
893 | .change = u32_change, | |
894 | .delete = u32_delete, | |
895 | .walk = u32_walk, | |
896 | .dump = u32_dump, | |
897 | .owner = THIS_MODULE, | |
898 | }; | |
899 | ||
900 | static int __init init_u32(void) | |
901 | { | |
6ff9c364 | 902 | pr_info("u32 classifier\n"); |
1da177e4 | 903 | #ifdef CONFIG_CLS_U32_PERF |
6ff9c364 | 904 | pr_info(" Performance counters on\n"); |
1da177e4 | 905 | #endif |
1da177e4 | 906 | #ifdef CONFIG_NET_CLS_IND |
6ff9c364 | 907 | pr_info(" input device check on\n"); |
1da177e4 LT |
908 | #endif |
909 | #ifdef CONFIG_NET_CLS_ACT | |
6ff9c364 | 910 | pr_info(" Actions configured\n"); |
1da177e4 LT |
911 | #endif |
912 | return register_tcf_proto_ops(&cls_u32_ops); | |
913 | } | |
914 | ||
10297b99 | 915 | static void __exit exit_u32(void) |
1da177e4 LT |
916 | { |
917 | unregister_tcf_proto_ops(&cls_u32_ops); | |
918 | } | |
919 | ||
920 | module_init(init_u32) | |
921 | module_exit(exit_u32) | |
922 | MODULE_LICENSE("GPL"); |