]>
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> | |
a1b7c5fd | 46 | #include <linux/netdevice.h> |
1da177e4 | 47 | |
cc7ec456 | 48 | struct tc_u_knode { |
1ce87720 | 49 | struct tc_u_knode __rcu *next; |
1da177e4 | 50 | u32 handle; |
1ce87720 | 51 | struct tc_u_hnode __rcu *ht_up; |
1da177e4 LT |
52 | struct tcf_exts exts; |
53 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 | 54 | int ifindex; |
1da177e4 LT |
55 | #endif |
56 | u8 fshift; | |
57 | struct tcf_result res; | |
1ce87720 | 58 | struct tc_u_hnode __rcu *ht_down; |
1da177e4 | 59 | #ifdef CONFIG_CLS_U32_PERF |
459d5f62 | 60 | struct tc_u32_pcnt __percpu *pf; |
1da177e4 | 61 | #endif |
9e8ce79c | 62 | u32 flags; |
1da177e4 | 63 | #ifdef CONFIG_CLS_U32_MARK |
459d5f62 JF |
64 | u32 val; |
65 | u32 mask; | |
66 | u32 __percpu *pcpu_success; | |
1da177e4 | 67 | #endif |
1ce87720 | 68 | struct tcf_proto *tp; |
1ce87720 | 69 | struct rcu_head rcu; |
4e2840ee JF |
70 | /* The 'sel' field MUST be the last field in structure to allow for |
71 | * tc_u32_keys allocated at end of structure. | |
72 | */ | |
73 | struct tc_u32_sel sel; | |
1da177e4 LT |
74 | }; |
75 | ||
cc7ec456 | 76 | struct tc_u_hnode { |
1ce87720 | 77 | struct tc_u_hnode __rcu *next; |
1da177e4 LT |
78 | u32 handle; |
79 | u32 prio; | |
80 | struct tc_u_common *tp_c; | |
81 | int refcnt; | |
cc7ec456 | 82 | unsigned int divisor; |
1ce87720 | 83 | struct rcu_head rcu; |
5778d39d WC |
84 | /* The 'ht' field MUST be the last field in structure to allow for |
85 | * more entries allocated at end of structure. | |
86 | */ | |
87 | struct tc_u_knode __rcu *ht[1]; | |
1da177e4 LT |
88 | }; |
89 | ||
cc7ec456 | 90 | struct tc_u_common { |
1ce87720 | 91 | struct tc_u_hnode __rcu *hlist; |
1da177e4 LT |
92 | struct Qdisc *q; |
93 | int refcnt; | |
94 | u32 hgenerator; | |
1ce87720 | 95 | struct rcu_head rcu; |
1da177e4 LT |
96 | }; |
97 | ||
cc7ec456 ED |
98 | static inline unsigned int u32_hash_fold(__be32 key, |
99 | const struct tc_u32_sel *sel, | |
100 | u8 fshift) | |
1da177e4 | 101 | { |
cc7ec456 | 102 | unsigned int h = ntohl(key & sel->hmask) >> fshift; |
1da177e4 LT |
103 | |
104 | return h; | |
105 | } | |
106 | ||
5a7a5555 JHS |
107 | static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, |
108 | struct tcf_result *res) | |
1da177e4 LT |
109 | { |
110 | struct { | |
111 | struct tc_u_knode *knode; | |
fbc2e7d9 | 112 | unsigned int off; |
1da177e4 LT |
113 | } stack[TC_U32_MAXDEPTH]; |
114 | ||
1ce87720 | 115 | struct tc_u_hnode *ht = rcu_dereference_bh(tp->root); |
fbc2e7d9 | 116 | unsigned int off = skb_network_offset(skb); |
1da177e4 LT |
117 | struct tc_u_knode *n; |
118 | int sdepth = 0; | |
119 | int off2 = 0; | |
120 | int sel = 0; | |
121 | #ifdef CONFIG_CLS_U32_PERF | |
122 | int j; | |
123 | #endif | |
124 | int i, r; | |
125 | ||
126 | next_ht: | |
1ce87720 | 127 | n = rcu_dereference_bh(ht->ht[sel]); |
1da177e4 LT |
128 | |
129 | next_knode: | |
130 | if (n) { | |
131 | struct tc_u32_key *key = n->sel.keys; | |
132 | ||
133 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 134 | __this_cpu_inc(n->pf->rcnt); |
1da177e4 LT |
135 | j = 0; |
136 | #endif | |
137 | ||
d34e3e18 SS |
138 | if (tc_skip_sw(n->flags)) { |
139 | n = rcu_dereference_bh(n->next); | |
140 | goto next_knode; | |
141 | } | |
142 | ||
1da177e4 | 143 | #ifdef CONFIG_CLS_U32_MARK |
459d5f62 | 144 | if ((skb->mark & n->mask) != n->val) { |
1ce87720 | 145 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
146 | goto next_knode; |
147 | } else { | |
459d5f62 | 148 | __this_cpu_inc(*n->pcpu_success); |
1da177e4 LT |
149 | } |
150 | #endif | |
151 | ||
cc7ec456 | 152 | for (i = n->sel.nkeys; i > 0; i--, key++) { |
66d50d25 | 153 | int toff = off + key->off + (off2 & key->offmask); |
86fce3ba | 154 | __be32 *data, hdata; |
fbc2e7d9 | 155 | |
4e18b3ed | 156 | if (skb_headroom(skb) + toff > INT_MAX) |
66d50d25 | 157 | goto out; |
158 | ||
86fce3ba | 159 | data = skb_header_pointer(skb, toff, 4, &hdata); |
fbc2e7d9 CG |
160 | if (!data) |
161 | goto out; | |
162 | if ((*data ^ key->val) & key->mask) { | |
1ce87720 | 163 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
164 | goto next_knode; |
165 | } | |
166 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 167 | __this_cpu_inc(n->pf->kcnts[j]); |
1da177e4 LT |
168 | j++; |
169 | #endif | |
170 | } | |
1ce87720 JF |
171 | |
172 | ht = rcu_dereference_bh(n->ht_down); | |
173 | if (!ht) { | |
1da177e4 | 174 | check_terminal: |
cc7ec456 | 175 | if (n->sel.flags & TC_U32_TERMINAL) { |
1da177e4 LT |
176 | |
177 | *res = n->res; | |
178 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 | 179 | if (!tcf_match_indev(skb, n->ifindex)) { |
1ce87720 | 180 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
181 | goto next_knode; |
182 | } | |
183 | #endif | |
184 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 185 | __this_cpu_inc(n->pf->rhit); |
1da177e4 LT |
186 | #endif |
187 | r = tcf_exts_exec(skb, &n->exts, res); | |
188 | if (r < 0) { | |
1ce87720 | 189 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
190 | goto next_knode; |
191 | } | |
192 | ||
193 | return r; | |
194 | } | |
1ce87720 | 195 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
196 | goto next_knode; |
197 | } | |
198 | ||
199 | /* PUSH */ | |
200 | if (sdepth >= TC_U32_MAXDEPTH) | |
201 | goto deadloop; | |
202 | stack[sdepth].knode = n; | |
fbc2e7d9 | 203 | stack[sdepth].off = off; |
1da177e4 LT |
204 | sdepth++; |
205 | ||
1ce87720 | 206 | ht = rcu_dereference_bh(n->ht_down); |
1da177e4 | 207 | sel = 0; |
fbc2e7d9 | 208 | if (ht->divisor) { |
86fce3ba | 209 | __be32 *data, hdata; |
fbc2e7d9 CG |
210 | |
211 | data = skb_header_pointer(skb, off + n->sel.hoff, 4, | |
86fce3ba | 212 | &hdata); |
fbc2e7d9 CG |
213 | if (!data) |
214 | goto out; | |
215 | sel = ht->divisor & u32_hash_fold(*data, &n->sel, | |
216 | n->fshift); | |
217 | } | |
cc7ec456 | 218 | if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT))) |
1da177e4 LT |
219 | goto next_ht; |
220 | ||
cc7ec456 | 221 | if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) { |
1da177e4 | 222 | off2 = n->sel.off + 3; |
fbc2e7d9 | 223 | if (n->sel.flags & TC_U32_VAROFFSET) { |
86fce3ba | 224 | __be16 *data, hdata; |
fbc2e7d9 CG |
225 | |
226 | data = skb_header_pointer(skb, | |
227 | off + n->sel.offoff, | |
86fce3ba | 228 | 2, &hdata); |
fbc2e7d9 CG |
229 | if (!data) |
230 | goto out; | |
231 | off2 += ntohs(n->sel.offmask & *data) >> | |
232 | n->sel.offshift; | |
233 | } | |
1da177e4 LT |
234 | off2 &= ~3; |
235 | } | |
cc7ec456 | 236 | if (n->sel.flags & TC_U32_EAT) { |
fbc2e7d9 | 237 | off += off2; |
1da177e4 LT |
238 | off2 = 0; |
239 | } | |
240 | ||
fbc2e7d9 | 241 | if (off < skb->len) |
1da177e4 LT |
242 | goto next_ht; |
243 | } | |
244 | ||
245 | /* POP */ | |
246 | if (sdepth--) { | |
247 | n = stack[sdepth].knode; | |
1ce87720 | 248 | ht = rcu_dereference_bh(n->ht_up); |
fbc2e7d9 | 249 | off = stack[sdepth].off; |
1da177e4 LT |
250 | goto check_terminal; |
251 | } | |
fbc2e7d9 | 252 | out: |
1da177e4 LT |
253 | return -1; |
254 | ||
255 | deadloop: | |
e87cc472 | 256 | net_warn_ratelimited("cls_u32: dead loop\n"); |
1da177e4 LT |
257 | return -1; |
258 | } | |
259 | ||
5a7a5555 | 260 | static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle) |
1da177e4 LT |
261 | { |
262 | struct tc_u_hnode *ht; | |
263 | ||
1ce87720 JF |
264 | for (ht = rtnl_dereference(tp_c->hlist); |
265 | ht; | |
266 | ht = rtnl_dereference(ht->next)) | |
1da177e4 LT |
267 | if (ht->handle == handle) |
268 | break; | |
269 | ||
270 | return ht; | |
271 | } | |
272 | ||
5a7a5555 | 273 | static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle) |
1da177e4 | 274 | { |
cc7ec456 | 275 | unsigned int sel; |
1da177e4 LT |
276 | struct tc_u_knode *n = NULL; |
277 | ||
278 | sel = TC_U32_HASH(handle); | |
279 | if (sel > ht->divisor) | |
280 | goto out; | |
281 | ||
1ce87720 JF |
282 | for (n = rtnl_dereference(ht->ht[sel]); |
283 | n; | |
284 | n = rtnl_dereference(n->next)) | |
1da177e4 LT |
285 | if (n->handle == handle) |
286 | break; | |
287 | out: | |
288 | return n; | |
289 | } | |
290 | ||
291 | ||
292 | static unsigned long u32_get(struct tcf_proto *tp, u32 handle) | |
293 | { | |
294 | struct tc_u_hnode *ht; | |
295 | struct tc_u_common *tp_c = tp->data; | |
296 | ||
297 | if (TC_U32_HTID(handle) == TC_U32_ROOT) | |
1ce87720 | 298 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
299 | else |
300 | ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle)); | |
301 | ||
302 | if (!ht) | |
303 | return 0; | |
304 | ||
305 | if (TC_U32_KEY(handle) == 0) | |
306 | return (unsigned long)ht; | |
307 | ||
308 | return (unsigned long)u32_lookup_key(ht, handle); | |
309 | } | |
310 | ||
1da177e4 LT |
311 | static u32 gen_new_htid(struct tc_u_common *tp_c) |
312 | { | |
313 | int i = 0x800; | |
314 | ||
1ce87720 JF |
315 | /* hgenerator only used inside rtnl lock it is safe to increment |
316 | * without read _copy_ update semantics | |
317 | */ | |
1da177e4 LT |
318 | do { |
319 | if (++tp_c->hgenerator == 0x7FF) | |
320 | tp_c->hgenerator = 1; | |
cc7ec456 | 321 | } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20)); |
1da177e4 LT |
322 | |
323 | return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0; | |
324 | } | |
325 | ||
326 | static int u32_init(struct tcf_proto *tp) | |
327 | { | |
328 | struct tc_u_hnode *root_ht; | |
329 | struct tc_u_common *tp_c; | |
330 | ||
72b25a91 | 331 | tp_c = tp->q->u32_node; |
1da177e4 | 332 | |
0da974f4 | 333 | root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL); |
1da177e4 LT |
334 | if (root_ht == NULL) |
335 | return -ENOBUFS; | |
336 | ||
1da177e4 LT |
337 | root_ht->divisor = 0; |
338 | root_ht->refcnt++; | |
339 | root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000; | |
340 | root_ht->prio = tp->prio; | |
341 | ||
342 | if (tp_c == NULL) { | |
0da974f4 | 343 | tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL); |
1da177e4 LT |
344 | if (tp_c == NULL) { |
345 | kfree(root_ht); | |
346 | return -ENOBUFS; | |
347 | } | |
1da177e4 | 348 | tp_c->q = tp->q; |
72b25a91 | 349 | tp->q->u32_node = tp_c; |
1da177e4 LT |
350 | } |
351 | ||
352 | tp_c->refcnt++; | |
1ce87720 JF |
353 | RCU_INIT_POINTER(root_ht->next, tp_c->hlist); |
354 | rcu_assign_pointer(tp_c->hlist, root_ht); | |
1da177e4 LT |
355 | root_ht->tp_c = tp_c; |
356 | ||
1ce87720 | 357 | rcu_assign_pointer(tp->root, root_ht); |
1da177e4 LT |
358 | tp->data = tp_c; |
359 | return 0; | |
360 | } | |
361 | ||
5a7a5555 | 362 | static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n, |
de5df632 | 363 | bool free_pf) |
1da177e4 | 364 | { |
18d0264f | 365 | tcf_exts_destroy(&n->exts); |
1da177e4 LT |
366 | if (n->ht_down) |
367 | n->ht_down->refcnt--; | |
368 | #ifdef CONFIG_CLS_U32_PERF | |
de5df632 JF |
369 | if (free_pf) |
370 | free_percpu(n->pf); | |
a1ddcfee JF |
371 | #endif |
372 | #ifdef CONFIG_CLS_U32_MARK | |
de5df632 JF |
373 | if (free_pf) |
374 | free_percpu(n->pcpu_success); | |
1da177e4 LT |
375 | #endif |
376 | kfree(n); | |
377 | return 0; | |
378 | } | |
379 | ||
de5df632 JF |
380 | /* u32_delete_key_rcu should be called when free'ing a copied |
381 | * version of a tc_u_knode obtained from u32_init_knode(). When | |
382 | * copies are obtained from u32_init_knode() the statistics are | |
383 | * shared between the old and new copies to allow readers to | |
384 | * continue to update the statistics during the copy. To support | |
385 | * this the u32_delete_key_rcu variant does not free the percpu | |
386 | * statistics. | |
387 | */ | |
1ce87720 JF |
388 | static void u32_delete_key_rcu(struct rcu_head *rcu) |
389 | { | |
390 | struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); | |
391 | ||
de5df632 JF |
392 | u32_destroy_key(key->tp, key, false); |
393 | } | |
394 | ||
395 | /* u32_delete_key_freepf_rcu is the rcu callback variant | |
396 | * that free's the entire structure including the statistics | |
397 | * percpu variables. Only use this if the key is not a copy | |
398 | * returned by u32_init_knode(). See u32_delete_key_rcu() | |
399 | * for the variant that should be used with keys return from | |
400 | * u32_init_knode() | |
401 | */ | |
402 | static void u32_delete_key_freepf_rcu(struct rcu_head *rcu) | |
403 | { | |
404 | struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); | |
405 | ||
406 | u32_destroy_key(key->tp, key, true); | |
1ce87720 JF |
407 | } |
408 | ||
82d567c2 | 409 | static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key) |
1da177e4 | 410 | { |
1ce87720 JF |
411 | struct tc_u_knode __rcu **kp; |
412 | struct tc_u_knode *pkp; | |
a96366bf | 413 | struct tc_u_hnode *ht = rtnl_dereference(key->ht_up); |
1da177e4 LT |
414 | |
415 | if (ht) { | |
1ce87720 JF |
416 | kp = &ht->ht[TC_U32_HASH(key->handle)]; |
417 | for (pkp = rtnl_dereference(*kp); pkp; | |
418 | kp = &pkp->next, pkp = rtnl_dereference(*kp)) { | |
419 | if (pkp == key) { | |
420 | RCU_INIT_POINTER(*kp, key->next); | |
1da177e4 | 421 | |
a0efb80c | 422 | tcf_unbind_filter(tp, &key->res); |
de5df632 | 423 | call_rcu(&key->rcu, u32_delete_key_freepf_rcu); |
1da177e4 LT |
424 | return 0; |
425 | } | |
426 | } | |
427 | } | |
547b792c | 428 | WARN_ON(1); |
1da177e4 LT |
429 | return 0; |
430 | } | |
431 | ||
a1b7c5fd JF |
432 | static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle) |
433 | { | |
434 | struct net_device *dev = tp->q->dev_queue->dev; | |
435 | struct tc_cls_u32_offload u32_offload = {0}; | |
436 | struct tc_to_netdev offload; | |
437 | ||
438 | offload.type = TC_SETUP_CLSU32; | |
439 | offload.cls_u32 = &u32_offload; | |
440 | ||
92c075db | 441 | if (tc_should_offload(dev, tp, 0)) { |
a1b7c5fd JF |
442 | offload.cls_u32->command = TC_CLSU32_DELETE_KNODE; |
443 | offload.cls_u32->knode.handle = handle; | |
444 | dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, | |
445 | tp->protocol, &offload); | |
446 | } | |
447 | } | |
448 | ||
5a7a5555 JHS |
449 | static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h, |
450 | u32 flags) | |
a1b7c5fd JF |
451 | { |
452 | struct net_device *dev = tp->q->dev_queue->dev; | |
453 | struct tc_cls_u32_offload u32_offload = {0}; | |
454 | struct tc_to_netdev offload; | |
d34e3e18 | 455 | int err; |
a1b7c5fd | 456 | |
92c075db | 457 | if (!tc_should_offload(dev, tp, flags)) |
d47a0f38 JK |
458 | return tc_skip_sw(flags) ? -EINVAL : 0; |
459 | ||
a1b7c5fd JF |
460 | offload.type = TC_SETUP_CLSU32; |
461 | offload.cls_u32 = &u32_offload; | |
462 | ||
d47a0f38 JK |
463 | offload.cls_u32->command = TC_CLSU32_NEW_HNODE; |
464 | offload.cls_u32->hnode.divisor = h->divisor; | |
465 | offload.cls_u32->hnode.handle = h->handle; | |
466 | offload.cls_u32->hnode.prio = h->prio; | |
a1b7c5fd | 467 | |
d47a0f38 JK |
468 | err = dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, |
469 | tp->protocol, &offload); | |
470 | if (tc_skip_sw(flags)) | |
471 | return err; | |
d34e3e18 SS |
472 | |
473 | return 0; | |
a1b7c5fd JF |
474 | } |
475 | ||
476 | static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h) | |
477 | { | |
478 | struct net_device *dev = tp->q->dev_queue->dev; | |
479 | struct tc_cls_u32_offload u32_offload = {0}; | |
480 | struct tc_to_netdev offload; | |
481 | ||
482 | offload.type = TC_SETUP_CLSU32; | |
483 | offload.cls_u32 = &u32_offload; | |
484 | ||
92c075db | 485 | if (tc_should_offload(dev, tp, 0)) { |
a1b7c5fd JF |
486 | offload.cls_u32->command = TC_CLSU32_DELETE_HNODE; |
487 | offload.cls_u32->hnode.divisor = h->divisor; | |
488 | offload.cls_u32->hnode.handle = h->handle; | |
489 | offload.cls_u32->hnode.prio = h->prio; | |
490 | ||
491 | dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, | |
492 | tp->protocol, &offload); | |
493 | } | |
494 | } | |
495 | ||
5a7a5555 JHS |
496 | static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n, |
497 | u32 flags) | |
a1b7c5fd JF |
498 | { |
499 | struct net_device *dev = tp->q->dev_queue->dev; | |
500 | struct tc_cls_u32_offload u32_offload = {0}; | |
501 | struct tc_to_netdev offload; | |
d34e3e18 | 502 | int err; |
a1b7c5fd JF |
503 | |
504 | offload.type = TC_SETUP_CLSU32; | |
505 | offload.cls_u32 = &u32_offload; | |
506 | ||
201c44bd JK |
507 | if (!tc_should_offload(dev, tp, flags)) |
508 | return tc_skip_sw(flags) ? -EINVAL : 0; | |
509 | ||
510 | offload.cls_u32->command = TC_CLSU32_REPLACE_KNODE; | |
511 | offload.cls_u32->knode.handle = n->handle; | |
512 | offload.cls_u32->knode.fshift = n->fshift; | |
a1b7c5fd | 513 | #ifdef CONFIG_CLS_U32_MARK |
201c44bd JK |
514 | offload.cls_u32->knode.val = n->val; |
515 | offload.cls_u32->knode.mask = n->mask; | |
a1b7c5fd | 516 | #else |
201c44bd JK |
517 | offload.cls_u32->knode.val = 0; |
518 | offload.cls_u32->knode.mask = 0; | |
a1b7c5fd | 519 | #endif |
201c44bd JK |
520 | offload.cls_u32->knode.sel = &n->sel; |
521 | offload.cls_u32->knode.exts = &n->exts; | |
522 | if (n->ht_down) | |
523 | offload.cls_u32->knode.link_handle = n->ht_down->handle; | |
524 | ||
525 | err = dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, | |
526 | tp->protocol, &offload); | |
63b029fc OG |
527 | |
528 | if (!err) | |
529 | n->flags |= TCA_CLS_FLAGS_IN_HW; | |
530 | ||
201c44bd JK |
531 | if (tc_skip_sw(flags)) |
532 | return err; | |
d34e3e18 SS |
533 | |
534 | return 0; | |
a1b7c5fd JF |
535 | } |
536 | ||
a0efb80c | 537 | static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) |
1da177e4 LT |
538 | { |
539 | struct tc_u_knode *n; | |
cc7ec456 | 540 | unsigned int h; |
1da177e4 | 541 | |
cc7ec456 | 542 | for (h = 0; h <= ht->divisor; h++) { |
1ce87720 JF |
543 | while ((n = rtnl_dereference(ht->ht[h])) != NULL) { |
544 | RCU_INIT_POINTER(ht->ht[h], | |
545 | rtnl_dereference(n->next)); | |
a0efb80c | 546 | tcf_unbind_filter(tp, &n->res); |
a1b7c5fd | 547 | u32_remove_hw_knode(tp, n->handle); |
de5df632 | 548 | call_rcu(&n->rcu, u32_delete_key_freepf_rcu); |
1da177e4 LT |
549 | } |
550 | } | |
551 | } | |
552 | ||
553 | static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) | |
554 | { | |
555 | struct tc_u_common *tp_c = tp->data; | |
1ce87720 JF |
556 | struct tc_u_hnode __rcu **hn; |
557 | struct tc_u_hnode *phn; | |
1da177e4 | 558 | |
547b792c | 559 | WARN_ON(ht->refcnt); |
1da177e4 | 560 | |
a0efb80c | 561 | u32_clear_hnode(tp, ht); |
1da177e4 | 562 | |
1ce87720 JF |
563 | hn = &tp_c->hlist; |
564 | for (phn = rtnl_dereference(*hn); | |
565 | phn; | |
566 | hn = &phn->next, phn = rtnl_dereference(*hn)) { | |
567 | if (phn == ht) { | |
a1b7c5fd | 568 | u32_clear_hw_hnode(tp, ht); |
1ce87720 JF |
569 | RCU_INIT_POINTER(*hn, ht->next); |
570 | kfree_rcu(ht, rcu); | |
1da177e4 LT |
571 | return 0; |
572 | } | |
573 | } | |
574 | ||
1da177e4 LT |
575 | return -ENOENT; |
576 | } | |
577 | ||
1e052be6 CW |
578 | static bool ht_empty(struct tc_u_hnode *ht) |
579 | { | |
580 | unsigned int h; | |
581 | ||
582 | for (h = 0; h <= ht->divisor; h++) | |
583 | if (rcu_access_pointer(ht->ht[h])) | |
584 | return false; | |
585 | ||
586 | return true; | |
587 | } | |
588 | ||
589 | static bool u32_destroy(struct tcf_proto *tp, bool force) | |
1da177e4 LT |
590 | { |
591 | struct tc_u_common *tp_c = tp->data; | |
1ce87720 | 592 | struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); |
1da177e4 | 593 | |
547b792c | 594 | WARN_ON(root_ht == NULL); |
1da177e4 | 595 | |
1e052be6 CW |
596 | if (!force) { |
597 | if (root_ht) { | |
598 | if (root_ht->refcnt > 1) | |
599 | return false; | |
600 | if (root_ht->refcnt == 1) { | |
601 | if (!ht_empty(root_ht)) | |
602 | return false; | |
603 | } | |
604 | } | |
a6c1aea0 WC |
605 | |
606 | if (tp_c->refcnt > 1) | |
607 | return false; | |
608 | ||
609 | if (tp_c->refcnt == 1) { | |
610 | struct tc_u_hnode *ht; | |
611 | ||
612 | for (ht = rtnl_dereference(tp_c->hlist); | |
613 | ht; | |
614 | ht = rtnl_dereference(ht->next)) | |
615 | if (!ht_empty(ht)) | |
616 | return false; | |
617 | } | |
1e052be6 CW |
618 | } |
619 | ||
1da177e4 LT |
620 | if (root_ht && --root_ht->refcnt == 0) |
621 | u32_destroy_hnode(tp, root_ht); | |
622 | ||
623 | if (--tp_c->refcnt == 0) { | |
624 | struct tc_u_hnode *ht; | |
1da177e4 | 625 | |
72b25a91 | 626 | tp->q->u32_node = NULL; |
1da177e4 | 627 | |
1ce87720 JF |
628 | for (ht = rtnl_dereference(tp_c->hlist); |
629 | ht; | |
630 | ht = rtnl_dereference(ht->next)) { | |
e56cfad1 | 631 | ht->refcnt--; |
a0efb80c | 632 | u32_clear_hnode(tp, ht); |
e56cfad1 | 633 | } |
1da177e4 | 634 | |
1ce87720 JF |
635 | while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) { |
636 | RCU_INIT_POINTER(tp_c->hlist, ht->next); | |
637 | kfree_rcu(ht, rcu); | |
3ff50b79 | 638 | } |
1da177e4 LT |
639 | |
640 | kfree(tp_c); | |
641 | } | |
642 | ||
643 | tp->data = NULL; | |
1e052be6 | 644 | return true; |
1da177e4 LT |
645 | } |
646 | ||
647 | static int u32_delete(struct tcf_proto *tp, unsigned long arg) | |
648 | { | |
cc7ec456 | 649 | struct tc_u_hnode *ht = (struct tc_u_hnode *)arg; |
1ce87720 | 650 | struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); |
1da177e4 LT |
651 | |
652 | if (ht == NULL) | |
653 | return 0; | |
654 | ||
a1b7c5fd JF |
655 | if (TC_U32_KEY(ht->handle)) { |
656 | u32_remove_hw_knode(tp, ht->handle); | |
cc7ec456 | 657 | return u32_delete_key(tp, (struct tc_u_knode *)ht); |
a1b7c5fd | 658 | } |
1da177e4 | 659 | |
1ce87720 | 660 | if (root_ht == ht) |
1da177e4 LT |
661 | return -EINVAL; |
662 | ||
e56cfad1 JP |
663 | if (ht->refcnt == 1) { |
664 | ht->refcnt--; | |
1da177e4 | 665 | u32_destroy_hnode(tp, ht); |
e56cfad1 JP |
666 | } else { |
667 | return -EBUSY; | |
668 | } | |
1da177e4 LT |
669 | |
670 | return 0; | |
671 | } | |
672 | ||
7801db8a | 673 | #define NR_U32_NODE (1<<12) |
1da177e4 LT |
674 | static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle) |
675 | { | |
676 | struct tc_u_knode *n; | |
7801db8a CW |
677 | unsigned long i; |
678 | unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long), | |
679 | GFP_KERNEL); | |
680 | if (!bitmap) | |
681 | return handle | 0xFFF; | |
1da177e4 | 682 | |
1ce87720 JF |
683 | for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]); |
684 | n; | |
685 | n = rtnl_dereference(n->next)) | |
7801db8a | 686 | set_bit(TC_U32_NODE(n->handle), bitmap); |
1da177e4 | 687 | |
7801db8a CW |
688 | i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800); |
689 | if (i >= NR_U32_NODE) | |
690 | i = find_next_zero_bit(bitmap, NR_U32_NODE, 1); | |
691 | ||
692 | kfree(bitmap); | |
693 | return handle | (i >= NR_U32_NODE ? 0xFFF : i); | |
1da177e4 LT |
694 | } |
695 | ||
6fa8c014 PM |
696 | static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = { |
697 | [TCA_U32_CLASSID] = { .type = NLA_U32 }, | |
698 | [TCA_U32_HASH] = { .type = NLA_U32 }, | |
699 | [TCA_U32_LINK] = { .type = NLA_U32 }, | |
700 | [TCA_U32_DIVISOR] = { .type = NLA_U32 }, | |
701 | [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) }, | |
702 | [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ }, | |
703 | [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) }, | |
9e8ce79c | 704 | [TCA_U32_FLAGS] = { .type = NLA_U32 }, |
6fa8c014 PM |
705 | }; |
706 | ||
c1b52739 BL |
707 | static int u32_set_parms(struct net *net, struct tcf_proto *tp, |
708 | unsigned long base, struct tc_u_hnode *ht, | |
add93b61 | 709 | struct tc_u_knode *n, struct nlattr **tb, |
2f7ef2f8 | 710 | struct nlattr *est, bool ovr) |
1da177e4 | 711 | { |
1da177e4 | 712 | struct tcf_exts e; |
b9a24bb7 | 713 | int err; |
1da177e4 | 714 | |
b9a24bb7 | 715 | err = tcf_exts_init(&e, TCA_U32_ACT, TCA_U32_POLICE); |
1da177e4 LT |
716 | if (err < 0) |
717 | return err; | |
b9a24bb7 WC |
718 | err = tcf_exts_validate(net, tp, tb, est, &e, ovr); |
719 | if (err < 0) | |
720 | goto errout; | |
1da177e4 LT |
721 | |
722 | err = -EINVAL; | |
add93b61 | 723 | if (tb[TCA_U32_LINK]) { |
1587bac4 | 724 | u32 handle = nla_get_u32(tb[TCA_U32_LINK]); |
47a1a1d4 | 725 | struct tc_u_hnode *ht_down = NULL, *ht_old; |
1da177e4 LT |
726 | |
727 | if (TC_U32_KEY(handle)) | |
728 | goto errout; | |
729 | ||
730 | if (handle) { | |
731 | ht_down = u32_lookup_ht(ht->tp_c, handle); | |
732 | ||
733 | if (ht_down == NULL) | |
734 | goto errout; | |
735 | ht_down->refcnt++; | |
736 | } | |
737 | ||
1ce87720 JF |
738 | ht_old = rtnl_dereference(n->ht_down); |
739 | rcu_assign_pointer(n->ht_down, ht_down); | |
1da177e4 | 740 | |
47a1a1d4 PM |
741 | if (ht_old) |
742 | ht_old->refcnt--; | |
1da177e4 | 743 | } |
add93b61 | 744 | if (tb[TCA_U32_CLASSID]) { |
1587bac4 | 745 | n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]); |
1da177e4 LT |
746 | tcf_bind_filter(tp, &n->res, base); |
747 | } | |
748 | ||
749 | #ifdef CONFIG_NET_CLS_IND | |
add93b61 | 750 | if (tb[TCA_U32_INDEV]) { |
2519a602 WC |
751 | int ret; |
752 | ret = tcf_change_indev(net, tb[TCA_U32_INDEV]); | |
753 | if (ret < 0) | |
1da177e4 | 754 | goto errout; |
2519a602 | 755 | n->ifindex = ret; |
1da177e4 LT |
756 | } |
757 | #endif | |
758 | tcf_exts_change(tp, &n->exts, &e); | |
759 | ||
760 | return 0; | |
761 | errout: | |
18d0264f | 762 | tcf_exts_destroy(&e); |
1da177e4 LT |
763 | return err; |
764 | } | |
765 | ||
5a7a5555 | 766 | static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c, |
de5df632 JF |
767 | struct tc_u_knode *n) |
768 | { | |
769 | struct tc_u_knode __rcu **ins; | |
770 | struct tc_u_knode *pins; | |
771 | struct tc_u_hnode *ht; | |
772 | ||
773 | if (TC_U32_HTID(n->handle) == TC_U32_ROOT) | |
774 | ht = rtnl_dereference(tp->root); | |
775 | else | |
776 | ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle)); | |
777 | ||
778 | ins = &ht->ht[TC_U32_HASH(n->handle)]; | |
779 | ||
780 | /* The node must always exist for it to be replaced if this is not the | |
781 | * case then something went very wrong elsewhere. | |
782 | */ | |
783 | for (pins = rtnl_dereference(*ins); ; | |
784 | ins = &pins->next, pins = rtnl_dereference(*ins)) | |
785 | if (pins->handle == n->handle) | |
786 | break; | |
787 | ||
788 | RCU_INIT_POINTER(n->next, pins->next); | |
789 | rcu_assign_pointer(*ins, n); | |
790 | } | |
791 | ||
792 | static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp, | |
793 | struct tc_u_knode *n) | |
794 | { | |
795 | struct tc_u_knode *new; | |
796 | struct tc_u32_sel *s = &n->sel; | |
797 | ||
798 | new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), | |
799 | GFP_KERNEL); | |
800 | ||
801 | if (!new) | |
802 | return NULL; | |
803 | ||
804 | RCU_INIT_POINTER(new->next, n->next); | |
805 | new->handle = n->handle; | |
806 | RCU_INIT_POINTER(new->ht_up, n->ht_up); | |
807 | ||
808 | #ifdef CONFIG_NET_CLS_IND | |
809 | new->ifindex = n->ifindex; | |
810 | #endif | |
811 | new->fshift = n->fshift; | |
812 | new->res = n->res; | |
9e8ce79c | 813 | new->flags = n->flags; |
de5df632 JF |
814 | RCU_INIT_POINTER(new->ht_down, n->ht_down); |
815 | ||
816 | /* bump reference count as long as we hold pointer to structure */ | |
817 | if (new->ht_down) | |
818 | new->ht_down->refcnt++; | |
819 | ||
820 | #ifdef CONFIG_CLS_U32_PERF | |
821 | /* Statistics may be incremented by readers during update | |
822 | * so we must keep them in tact. When the node is later destroyed | |
823 | * a special destroy call must be made to not free the pf memory. | |
824 | */ | |
825 | new->pf = n->pf; | |
826 | #endif | |
827 | ||
828 | #ifdef CONFIG_CLS_U32_MARK | |
829 | new->val = n->val; | |
830 | new->mask = n->mask; | |
831 | /* Similarly success statistics must be moved as pointers */ | |
832 | new->pcpu_success = n->pcpu_success; | |
833 | #endif | |
834 | new->tp = tp; | |
835 | memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); | |
836 | ||
b9a24bb7 WC |
837 | if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) { |
838 | kfree(new); | |
839 | return NULL; | |
840 | } | |
de5df632 JF |
841 | |
842 | return new; | |
843 | } | |
844 | ||
c1b52739 | 845 | static int u32_change(struct net *net, struct sk_buff *in_skb, |
af4c6641 | 846 | struct tcf_proto *tp, unsigned long base, u32 handle, |
5a7a5555 | 847 | struct nlattr **tca, unsigned long *arg, bool ovr) |
1da177e4 LT |
848 | { |
849 | struct tc_u_common *tp_c = tp->data; | |
850 | struct tc_u_hnode *ht; | |
851 | struct tc_u_knode *n; | |
852 | struct tc_u32_sel *s; | |
add93b61 PM |
853 | struct nlattr *opt = tca[TCA_OPTIONS]; |
854 | struct nlattr *tb[TCA_U32_MAX + 1]; | |
9e8ce79c | 855 | u32 htid, flags = 0; |
1da177e4 | 856 | int err; |
459d5f62 JF |
857 | #ifdef CONFIG_CLS_U32_PERF |
858 | size_t size; | |
859 | #endif | |
1da177e4 LT |
860 | |
861 | if (opt == NULL) | |
862 | return handle ? -EINVAL : 0; | |
863 | ||
6fa8c014 | 864 | err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy); |
cee63723 PM |
865 | if (err < 0) |
866 | return err; | |
1da177e4 | 867 | |
d34e3e18 | 868 | if (tb[TCA_U32_FLAGS]) { |
9e8ce79c | 869 | flags = nla_get_u32(tb[TCA_U32_FLAGS]); |
d34e3e18 | 870 | if (!tc_flags_valid(flags)) |
1a0f7d29 | 871 | return -EINVAL; |
d34e3e18 | 872 | } |
9e8ce79c | 873 | |
cc7ec456 ED |
874 | n = (struct tc_u_knode *)*arg; |
875 | if (n) { | |
de5df632 JF |
876 | struct tc_u_knode *new; |
877 | ||
1da177e4 LT |
878 | if (TC_U32_KEY(n->handle) == 0) |
879 | return -EINVAL; | |
880 | ||
9e8ce79c JF |
881 | if (n->flags != flags) |
882 | return -EINVAL; | |
883 | ||
de5df632 JF |
884 | new = u32_init_knode(tp, n); |
885 | if (!new) | |
886 | return -ENOMEM; | |
887 | ||
888 | err = u32_set_parms(net, tp, base, | |
889 | rtnl_dereference(n->ht_up), new, tb, | |
890 | tca[TCA_RATE], ovr); | |
891 | ||
892 | if (err) { | |
893 | u32_destroy_key(tp, new, false); | |
894 | return err; | |
895 | } | |
896 | ||
d34e3e18 SS |
897 | err = u32_replace_hw_knode(tp, new, flags); |
898 | if (err) { | |
899 | u32_destroy_key(tp, new, false); | |
900 | return err; | |
901 | } | |
902 | ||
63b029fc OG |
903 | if (!tc_in_hw(new->flags)) |
904 | new->flags |= TCA_CLS_FLAGS_NOT_IN_HW; | |
905 | ||
de5df632 | 906 | u32_replace_knode(tp, tp_c, new); |
a0efb80c | 907 | tcf_unbind_filter(tp, &n->res); |
de5df632 JF |
908 | call_rcu(&n->rcu, u32_delete_key_rcu); |
909 | return 0; | |
1da177e4 LT |
910 | } |
911 | ||
add93b61 | 912 | if (tb[TCA_U32_DIVISOR]) { |
cc7ec456 | 913 | unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]); |
1da177e4 LT |
914 | |
915 | if (--divisor > 0x100) | |
916 | return -EINVAL; | |
917 | if (TC_U32_KEY(handle)) | |
918 | return -EINVAL; | |
919 | if (handle == 0) { | |
920 | handle = gen_new_htid(tp->data); | |
921 | if (handle == 0) | |
922 | return -ENOMEM; | |
923 | } | |
cc7ec456 | 924 | ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL); |
1da177e4 LT |
925 | if (ht == NULL) |
926 | return -ENOBUFS; | |
1da177e4 | 927 | ht->tp_c = tp_c; |
e56cfad1 | 928 | ht->refcnt = 1; |
1da177e4 LT |
929 | ht->divisor = divisor; |
930 | ht->handle = handle; | |
931 | ht->prio = tp->prio; | |
6eef3801 JK |
932 | |
933 | err = u32_replace_hw_hnode(tp, ht, flags); | |
934 | if (err) { | |
935 | kfree(ht); | |
936 | return err; | |
937 | } | |
938 | ||
1ce87720 JF |
939 | RCU_INIT_POINTER(ht->next, tp_c->hlist); |
940 | rcu_assign_pointer(tp_c->hlist, ht); | |
1da177e4 | 941 | *arg = (unsigned long)ht; |
a1b7c5fd | 942 | |
1da177e4 LT |
943 | return 0; |
944 | } | |
945 | ||
add93b61 | 946 | if (tb[TCA_U32_HASH]) { |
1587bac4 | 947 | htid = nla_get_u32(tb[TCA_U32_HASH]); |
1da177e4 | 948 | if (TC_U32_HTID(htid) == TC_U32_ROOT) { |
1ce87720 | 949 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
950 | htid = ht->handle; |
951 | } else { | |
952 | ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid)); | |
953 | if (ht == NULL) | |
954 | return -EINVAL; | |
955 | } | |
956 | } else { | |
1ce87720 | 957 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
958 | htid = ht->handle; |
959 | } | |
960 | ||
961 | if (ht->divisor < TC_U32_HASH(htid)) | |
962 | return -EINVAL; | |
963 | ||
964 | if (handle) { | |
965 | if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid)) | |
966 | return -EINVAL; | |
967 | handle = htid | TC_U32_NODE(handle); | |
968 | } else | |
969 | handle = gen_new_kid(ht, htid); | |
970 | ||
6fa8c014 | 971 | if (tb[TCA_U32_SEL] == NULL) |
1da177e4 LT |
972 | return -EINVAL; |
973 | ||
add93b61 | 974 | s = nla_data(tb[TCA_U32_SEL]); |
1da177e4 | 975 | |
0da974f4 | 976 | n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL); |
1da177e4 LT |
977 | if (n == NULL) |
978 | return -ENOBUFS; | |
979 | ||
1da177e4 | 980 | #ifdef CONFIG_CLS_U32_PERF |
459d5f62 JF |
981 | size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64); |
982 | n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt)); | |
983 | if (!n->pf) { | |
1da177e4 LT |
984 | kfree(n); |
985 | return -ENOBUFS; | |
986 | } | |
1da177e4 LT |
987 | #endif |
988 | ||
989 | memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); | |
a96366bf | 990 | RCU_INIT_POINTER(n->ht_up, ht); |
1da177e4 | 991 | n->handle = handle; |
b2268016 | 992 | n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0; |
9e8ce79c | 993 | n->flags = flags; |
1ce87720 | 994 | n->tp = tp; |
1da177e4 | 995 | |
b9a24bb7 WC |
996 | err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE); |
997 | if (err < 0) | |
998 | goto errout; | |
999 | ||
1da177e4 | 1000 | #ifdef CONFIG_CLS_U32_MARK |
459d5f62 | 1001 | n->pcpu_success = alloc_percpu(u32); |
a1ddcfee JF |
1002 | if (!n->pcpu_success) { |
1003 | err = -ENOMEM; | |
1004 | goto errout; | |
1005 | } | |
459d5f62 | 1006 | |
add93b61 | 1007 | if (tb[TCA_U32_MARK]) { |
1da177e4 LT |
1008 | struct tc_u32_mark *mark; |
1009 | ||
add93b61 | 1010 | mark = nla_data(tb[TCA_U32_MARK]); |
459d5f62 JF |
1011 | n->val = mark->val; |
1012 | n->mask = mark->mask; | |
1da177e4 LT |
1013 | } |
1014 | #endif | |
1015 | ||
2f7ef2f8 | 1016 | err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr); |
1da177e4 | 1017 | if (err == 0) { |
1ce87720 JF |
1018 | struct tc_u_knode __rcu **ins; |
1019 | struct tc_u_knode *pins; | |
1020 | ||
d34e3e18 SS |
1021 | err = u32_replace_hw_knode(tp, n, flags); |
1022 | if (err) | |
1023 | goto errhw; | |
1024 | ||
63b029fc OG |
1025 | if (!tc_in_hw(n->flags)) |
1026 | n->flags |= TCA_CLS_FLAGS_NOT_IN_HW; | |
1027 | ||
1ce87720 JF |
1028 | ins = &ht->ht[TC_U32_HASH(handle)]; |
1029 | for (pins = rtnl_dereference(*ins); pins; | |
1030 | ins = &pins->next, pins = rtnl_dereference(*ins)) | |
1031 | if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle)) | |
1da177e4 LT |
1032 | break; |
1033 | ||
1ce87720 JF |
1034 | RCU_INIT_POINTER(n->next, pins); |
1035 | rcu_assign_pointer(*ins, n); | |
1da177e4 LT |
1036 | *arg = (unsigned long)n; |
1037 | return 0; | |
1038 | } | |
a1ddcfee | 1039 | |
d34e3e18 | 1040 | errhw: |
a1ddcfee JF |
1041 | #ifdef CONFIG_CLS_U32_MARK |
1042 | free_percpu(n->pcpu_success); | |
1043 | #endif | |
1044 | ||
b9a24bb7 WC |
1045 | errout: |
1046 | tcf_exts_destroy(&n->exts); | |
1da177e4 | 1047 | #ifdef CONFIG_CLS_U32_PERF |
1ce87720 | 1048 | free_percpu(n->pf); |
1da177e4 LT |
1049 | #endif |
1050 | kfree(n); | |
1051 | return err; | |
1052 | } | |
1053 | ||
1054 | static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg) | |
1055 | { | |
1056 | struct tc_u_common *tp_c = tp->data; | |
1057 | struct tc_u_hnode *ht; | |
1058 | struct tc_u_knode *n; | |
cc7ec456 | 1059 | unsigned int h; |
1da177e4 LT |
1060 | |
1061 | if (arg->stop) | |
1062 | return; | |
1063 | ||
1ce87720 JF |
1064 | for (ht = rtnl_dereference(tp_c->hlist); |
1065 | ht; | |
1066 | ht = rtnl_dereference(ht->next)) { | |
1da177e4 LT |
1067 | if (ht->prio != tp->prio) |
1068 | continue; | |
1069 | if (arg->count >= arg->skip) { | |
1070 | if (arg->fn(tp, (unsigned long)ht, arg) < 0) { | |
1071 | arg->stop = 1; | |
1072 | return; | |
1073 | } | |
1074 | } | |
1075 | arg->count++; | |
1076 | for (h = 0; h <= ht->divisor; h++) { | |
1ce87720 JF |
1077 | for (n = rtnl_dereference(ht->ht[h]); |
1078 | n; | |
1079 | n = rtnl_dereference(n->next)) { | |
1da177e4 LT |
1080 | if (arg->count < arg->skip) { |
1081 | arg->count++; | |
1082 | continue; | |
1083 | } | |
1084 | if (arg->fn(tp, (unsigned long)n, arg) < 0) { | |
1085 | arg->stop = 1; | |
1086 | return; | |
1087 | } | |
1088 | arg->count++; | |
1089 | } | |
1090 | } | |
1091 | } | |
1092 | } | |
1093 | ||
832d1d5b | 1094 | static int u32_dump(struct net *net, struct tcf_proto *tp, unsigned long fh, |
5a7a5555 | 1095 | struct sk_buff *skb, struct tcmsg *t) |
1da177e4 | 1096 | { |
cc7ec456 | 1097 | struct tc_u_knode *n = (struct tc_u_knode *)fh; |
1ce87720 | 1098 | struct tc_u_hnode *ht_up, *ht_down; |
4b3550ef | 1099 | struct nlattr *nest; |
1da177e4 LT |
1100 | |
1101 | if (n == NULL) | |
1102 | return skb->len; | |
1103 | ||
1104 | t->tcm_handle = n->handle; | |
1105 | ||
4b3550ef PM |
1106 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1107 | if (nest == NULL) | |
1108 | goto nla_put_failure; | |
1da177e4 LT |
1109 | |
1110 | if (TC_U32_KEY(n->handle) == 0) { | |
cc7ec456 ED |
1111 | struct tc_u_hnode *ht = (struct tc_u_hnode *)fh; |
1112 | u32 divisor = ht->divisor + 1; | |
1113 | ||
1b34ec43 DM |
1114 | if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor)) |
1115 | goto nla_put_failure; | |
1da177e4 | 1116 | } else { |
459d5f62 JF |
1117 | #ifdef CONFIG_CLS_U32_PERF |
1118 | struct tc_u32_pcnt *gpf; | |
459d5f62 | 1119 | int cpu; |
80aab73d | 1120 | #endif |
459d5f62 | 1121 | |
1b34ec43 DM |
1122 | if (nla_put(skb, TCA_U32_SEL, |
1123 | sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key), | |
1124 | &n->sel)) | |
1125 | goto nla_put_failure; | |
1ce87720 JF |
1126 | |
1127 | ht_up = rtnl_dereference(n->ht_up); | |
1128 | if (ht_up) { | |
1da177e4 | 1129 | u32 htid = n->handle & 0xFFFFF000; |
1b34ec43 DM |
1130 | if (nla_put_u32(skb, TCA_U32_HASH, htid)) |
1131 | goto nla_put_failure; | |
1da177e4 | 1132 | } |
1b34ec43 DM |
1133 | if (n->res.classid && |
1134 | nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid)) | |
1135 | goto nla_put_failure; | |
1ce87720 JF |
1136 | |
1137 | ht_down = rtnl_dereference(n->ht_down); | |
1138 | if (ht_down && | |
1139 | nla_put_u32(skb, TCA_U32_LINK, ht_down->handle)) | |
1b34ec43 | 1140 | goto nla_put_failure; |
1da177e4 | 1141 | |
9e8ce79c JF |
1142 | if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags)) |
1143 | goto nla_put_failure; | |
1144 | ||
1da177e4 | 1145 | #ifdef CONFIG_CLS_U32_MARK |
459d5f62 JF |
1146 | if ((n->val || n->mask)) { |
1147 | struct tc_u32_mark mark = {.val = n->val, | |
1148 | .mask = n->mask, | |
1149 | .success = 0}; | |
80aab73d | 1150 | int cpum; |
459d5f62 | 1151 | |
80aab73d JF |
1152 | for_each_possible_cpu(cpum) { |
1153 | __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum); | |
459d5f62 JF |
1154 | |
1155 | mark.success += cnt; | |
1156 | } | |
1157 | ||
1158 | if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark)) | |
1159 | goto nla_put_failure; | |
1160 | } | |
1da177e4 LT |
1161 | #endif |
1162 | ||
5da57f42 | 1163 | if (tcf_exts_dump(skb, &n->exts) < 0) |
add93b61 | 1164 | goto nla_put_failure; |
1da177e4 LT |
1165 | |
1166 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 WC |
1167 | if (n->ifindex) { |
1168 | struct net_device *dev; | |
1169 | dev = __dev_get_by_index(net, n->ifindex); | |
1170 | if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name)) | |
1171 | goto nla_put_failure; | |
1172 | } | |
1da177e4 LT |
1173 | #endif |
1174 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 JF |
1175 | gpf = kzalloc(sizeof(struct tc_u32_pcnt) + |
1176 | n->sel.nkeys * sizeof(u64), | |
1177 | GFP_KERNEL); | |
1178 | if (!gpf) | |
1179 | goto nla_put_failure; | |
1180 | ||
1181 | for_each_possible_cpu(cpu) { | |
1182 | int i; | |
1183 | struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu); | |
1184 | ||
1185 | gpf->rcnt += pf->rcnt; | |
1186 | gpf->rhit += pf->rhit; | |
1187 | for (i = 0; i < n->sel.nkeys; i++) | |
1188 | gpf->kcnts[i] += pf->kcnts[i]; | |
1189 | } | |
1190 | ||
9854518e ND |
1191 | if (nla_put_64bit(skb, TCA_U32_PCNT, |
1192 | sizeof(struct tc_u32_pcnt) + | |
1193 | n->sel.nkeys * sizeof(u64), | |
1194 | gpf, TCA_U32_PAD)) { | |
459d5f62 | 1195 | kfree(gpf); |
1b34ec43 | 1196 | goto nla_put_failure; |
459d5f62 JF |
1197 | } |
1198 | kfree(gpf); | |
1da177e4 LT |
1199 | #endif |
1200 | } | |
1201 | ||
4b3550ef PM |
1202 | nla_nest_end(skb, nest); |
1203 | ||
1da177e4 | 1204 | if (TC_U32_KEY(n->handle)) |
5da57f42 | 1205 | if (tcf_exts_dump_stats(skb, &n->exts) < 0) |
add93b61 | 1206 | goto nla_put_failure; |
1da177e4 LT |
1207 | return skb->len; |
1208 | ||
add93b61 | 1209 | nla_put_failure: |
4b3550ef | 1210 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1211 | return -1; |
1212 | } | |
1213 | ||
2eb9d75c | 1214 | static struct tcf_proto_ops cls_u32_ops __read_mostly = { |
1da177e4 LT |
1215 | .kind = "u32", |
1216 | .classify = u32_classify, | |
1217 | .init = u32_init, | |
1218 | .destroy = u32_destroy, | |
1219 | .get = u32_get, | |
1da177e4 LT |
1220 | .change = u32_change, |
1221 | .delete = u32_delete, | |
1222 | .walk = u32_walk, | |
1223 | .dump = u32_dump, | |
1224 | .owner = THIS_MODULE, | |
1225 | }; | |
1226 | ||
1227 | static int __init init_u32(void) | |
1228 | { | |
6ff9c364 | 1229 | pr_info("u32 classifier\n"); |
1da177e4 | 1230 | #ifdef CONFIG_CLS_U32_PERF |
6ff9c364 | 1231 | pr_info(" Performance counters on\n"); |
1da177e4 | 1232 | #endif |
1da177e4 | 1233 | #ifdef CONFIG_NET_CLS_IND |
6ff9c364 | 1234 | pr_info(" input device check on\n"); |
1da177e4 LT |
1235 | #endif |
1236 | #ifdef CONFIG_NET_CLS_ACT | |
6ff9c364 | 1237 | pr_info(" Actions configured\n"); |
1da177e4 LT |
1238 | #endif |
1239 | return register_tcf_proto_ops(&cls_u32_ops); | |
1240 | } | |
1241 | ||
10297b99 | 1242 | static void __exit exit_u32(void) |
1da177e4 LT |
1243 | { |
1244 | unregister_tcf_proto_ops(&cls_u32_ops); | |
1245 | } | |
1246 | ||
1247 | module_init(init_u32) | |
1248 | module_exit(exit_u32) | |
1249 | MODULE_LICENSE("GPL"); |