]>
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 | ||
dc7f9f6e | 107 | static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, struct tcf_result *res) |
1da177e4 LT |
108 | { |
109 | struct { | |
110 | struct tc_u_knode *knode; | |
fbc2e7d9 | 111 | unsigned int off; |
1da177e4 LT |
112 | } stack[TC_U32_MAXDEPTH]; |
113 | ||
1ce87720 | 114 | struct tc_u_hnode *ht = rcu_dereference_bh(tp->root); |
fbc2e7d9 | 115 | unsigned int off = skb_network_offset(skb); |
1da177e4 LT |
116 | struct tc_u_knode *n; |
117 | int sdepth = 0; | |
118 | int off2 = 0; | |
119 | int sel = 0; | |
120 | #ifdef CONFIG_CLS_U32_PERF | |
121 | int j; | |
122 | #endif | |
123 | int i, r; | |
124 | ||
125 | next_ht: | |
1ce87720 | 126 | n = rcu_dereference_bh(ht->ht[sel]); |
1da177e4 LT |
127 | |
128 | next_knode: | |
129 | if (n) { | |
130 | struct tc_u32_key *key = n->sel.keys; | |
131 | ||
132 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 133 | __this_cpu_inc(n->pf->rcnt); |
1da177e4 LT |
134 | j = 0; |
135 | #endif | |
136 | ||
d34e3e18 SS |
137 | if (tc_skip_sw(n->flags)) { |
138 | n = rcu_dereference_bh(n->next); | |
139 | goto next_knode; | |
140 | } | |
141 | ||
1da177e4 | 142 | #ifdef CONFIG_CLS_U32_MARK |
459d5f62 | 143 | if ((skb->mark & n->mask) != n->val) { |
1ce87720 | 144 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
145 | goto next_knode; |
146 | } else { | |
459d5f62 | 147 | __this_cpu_inc(*n->pcpu_success); |
1da177e4 LT |
148 | } |
149 | #endif | |
150 | ||
cc7ec456 | 151 | for (i = n->sel.nkeys; i > 0; i--, key++) { |
66d50d25 | 152 | int toff = off + key->off + (off2 & key->offmask); |
86fce3ba | 153 | __be32 *data, hdata; |
fbc2e7d9 | 154 | |
4e18b3ed | 155 | if (skb_headroom(skb) + toff > INT_MAX) |
66d50d25 | 156 | goto out; |
157 | ||
86fce3ba | 158 | data = skb_header_pointer(skb, toff, 4, &hdata); |
fbc2e7d9 CG |
159 | if (!data) |
160 | goto out; | |
161 | if ((*data ^ key->val) & key->mask) { | |
1ce87720 | 162 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
163 | goto next_knode; |
164 | } | |
165 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 166 | __this_cpu_inc(n->pf->kcnts[j]); |
1da177e4 LT |
167 | j++; |
168 | #endif | |
169 | } | |
1ce87720 JF |
170 | |
171 | ht = rcu_dereference_bh(n->ht_down); | |
172 | if (!ht) { | |
1da177e4 | 173 | check_terminal: |
cc7ec456 | 174 | if (n->sel.flags & TC_U32_TERMINAL) { |
1da177e4 LT |
175 | |
176 | *res = n->res; | |
177 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 | 178 | if (!tcf_match_indev(skb, n->ifindex)) { |
1ce87720 | 179 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
180 | goto next_knode; |
181 | } | |
182 | #endif | |
183 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 184 | __this_cpu_inc(n->pf->rhit); |
1da177e4 LT |
185 | #endif |
186 | r = tcf_exts_exec(skb, &n->exts, res); | |
187 | if (r < 0) { | |
1ce87720 | 188 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
189 | goto next_knode; |
190 | } | |
191 | ||
192 | return r; | |
193 | } | |
1ce87720 | 194 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
195 | goto next_knode; |
196 | } | |
197 | ||
198 | /* PUSH */ | |
199 | if (sdepth >= TC_U32_MAXDEPTH) | |
200 | goto deadloop; | |
201 | stack[sdepth].knode = n; | |
fbc2e7d9 | 202 | stack[sdepth].off = off; |
1da177e4 LT |
203 | sdepth++; |
204 | ||
1ce87720 | 205 | ht = rcu_dereference_bh(n->ht_down); |
1da177e4 | 206 | sel = 0; |
fbc2e7d9 | 207 | if (ht->divisor) { |
86fce3ba | 208 | __be32 *data, hdata; |
fbc2e7d9 CG |
209 | |
210 | data = skb_header_pointer(skb, off + n->sel.hoff, 4, | |
86fce3ba | 211 | &hdata); |
fbc2e7d9 CG |
212 | if (!data) |
213 | goto out; | |
214 | sel = ht->divisor & u32_hash_fold(*data, &n->sel, | |
215 | n->fshift); | |
216 | } | |
cc7ec456 | 217 | if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT))) |
1da177e4 LT |
218 | goto next_ht; |
219 | ||
cc7ec456 | 220 | if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) { |
1da177e4 | 221 | off2 = n->sel.off + 3; |
fbc2e7d9 | 222 | if (n->sel.flags & TC_U32_VAROFFSET) { |
86fce3ba | 223 | __be16 *data, hdata; |
fbc2e7d9 CG |
224 | |
225 | data = skb_header_pointer(skb, | |
226 | off + n->sel.offoff, | |
86fce3ba | 227 | 2, &hdata); |
fbc2e7d9 CG |
228 | if (!data) |
229 | goto out; | |
230 | off2 += ntohs(n->sel.offmask & *data) >> | |
231 | n->sel.offshift; | |
232 | } | |
1da177e4 LT |
233 | off2 &= ~3; |
234 | } | |
cc7ec456 | 235 | if (n->sel.flags & TC_U32_EAT) { |
fbc2e7d9 | 236 | off += off2; |
1da177e4 LT |
237 | off2 = 0; |
238 | } | |
239 | ||
fbc2e7d9 | 240 | if (off < skb->len) |
1da177e4 LT |
241 | goto next_ht; |
242 | } | |
243 | ||
244 | /* POP */ | |
245 | if (sdepth--) { | |
246 | n = stack[sdepth].knode; | |
1ce87720 | 247 | ht = rcu_dereference_bh(n->ht_up); |
fbc2e7d9 | 248 | off = stack[sdepth].off; |
1da177e4 LT |
249 | goto check_terminal; |
250 | } | |
fbc2e7d9 | 251 | out: |
1da177e4 LT |
252 | return -1; |
253 | ||
254 | deadloop: | |
e87cc472 | 255 | net_warn_ratelimited("cls_u32: dead loop\n"); |
1da177e4 LT |
256 | return -1; |
257 | } | |
258 | ||
cc7ec456 | 259 | static struct tc_u_hnode * |
1da177e4 LT |
260 | u32_lookup_ht(struct tc_u_common *tp_c, u32 handle) |
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 | ||
cc7ec456 | 273 | static struct tc_u_knode * |
1da177e4 LT |
274 | u32_lookup_key(struct tc_u_hnode *ht, u32 handle) |
275 | { | |
cc7ec456 | 276 | unsigned int sel; |
1da177e4 LT |
277 | struct tc_u_knode *n = NULL; |
278 | ||
279 | sel = TC_U32_HASH(handle); | |
280 | if (sel > ht->divisor) | |
281 | goto out; | |
282 | ||
1ce87720 JF |
283 | for (n = rtnl_dereference(ht->ht[sel]); |
284 | n; | |
285 | n = rtnl_dereference(n->next)) | |
1da177e4 LT |
286 | if (n->handle == handle) |
287 | break; | |
288 | out: | |
289 | return n; | |
290 | } | |
291 | ||
292 | ||
293 | static unsigned long u32_get(struct tcf_proto *tp, u32 handle) | |
294 | { | |
295 | struct tc_u_hnode *ht; | |
296 | struct tc_u_common *tp_c = tp->data; | |
297 | ||
298 | if (TC_U32_HTID(handle) == TC_U32_ROOT) | |
1ce87720 | 299 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
300 | else |
301 | ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle)); | |
302 | ||
303 | if (!ht) | |
304 | return 0; | |
305 | ||
306 | if (TC_U32_KEY(handle) == 0) | |
307 | return (unsigned long)ht; | |
308 | ||
309 | return (unsigned long)u32_lookup_key(ht, handle); | |
310 | } | |
311 | ||
1da177e4 LT |
312 | static u32 gen_new_htid(struct tc_u_common *tp_c) |
313 | { | |
314 | int i = 0x800; | |
315 | ||
1ce87720 JF |
316 | /* hgenerator only used inside rtnl lock it is safe to increment |
317 | * without read _copy_ update semantics | |
318 | */ | |
1da177e4 LT |
319 | do { |
320 | if (++tp_c->hgenerator == 0x7FF) | |
321 | tp_c->hgenerator = 1; | |
cc7ec456 | 322 | } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20)); |
1da177e4 LT |
323 | |
324 | return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0; | |
325 | } | |
326 | ||
327 | static int u32_init(struct tcf_proto *tp) | |
328 | { | |
329 | struct tc_u_hnode *root_ht; | |
330 | struct tc_u_common *tp_c; | |
331 | ||
72b25a91 | 332 | tp_c = tp->q->u32_node; |
1da177e4 | 333 | |
0da974f4 | 334 | root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL); |
1da177e4 LT |
335 | if (root_ht == NULL) |
336 | return -ENOBUFS; | |
337 | ||
1da177e4 LT |
338 | root_ht->divisor = 0; |
339 | root_ht->refcnt++; | |
340 | root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000; | |
341 | root_ht->prio = tp->prio; | |
342 | ||
343 | if (tp_c == NULL) { | |
0da974f4 | 344 | tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL); |
1da177e4 LT |
345 | if (tp_c == NULL) { |
346 | kfree(root_ht); | |
347 | return -ENOBUFS; | |
348 | } | |
1da177e4 | 349 | tp_c->q = tp->q; |
72b25a91 | 350 | tp->q->u32_node = tp_c; |
1da177e4 LT |
351 | } |
352 | ||
353 | tp_c->refcnt++; | |
1ce87720 JF |
354 | RCU_INIT_POINTER(root_ht->next, tp_c->hlist); |
355 | rcu_assign_pointer(tp_c->hlist, root_ht); | |
1da177e4 LT |
356 | root_ht->tp_c = tp_c; |
357 | ||
1ce87720 | 358 | rcu_assign_pointer(tp->root, root_ht); |
1da177e4 LT |
359 | tp->data = tp_c; |
360 | return 0; | |
361 | } | |
362 | ||
de5df632 JF |
363 | static int u32_destroy_key(struct tcf_proto *tp, |
364 | struct tc_u_knode *n, | |
365 | bool free_pf) | |
1da177e4 | 366 | { |
18d0264f | 367 | tcf_exts_destroy(&n->exts); |
1da177e4 LT |
368 | if (n->ht_down) |
369 | n->ht_down->refcnt--; | |
370 | #ifdef CONFIG_CLS_U32_PERF | |
de5df632 JF |
371 | if (free_pf) |
372 | free_percpu(n->pf); | |
a1ddcfee JF |
373 | #endif |
374 | #ifdef CONFIG_CLS_U32_MARK | |
de5df632 JF |
375 | if (free_pf) |
376 | free_percpu(n->pcpu_success); | |
1da177e4 LT |
377 | #endif |
378 | kfree(n); | |
379 | return 0; | |
380 | } | |
381 | ||
de5df632 JF |
382 | /* u32_delete_key_rcu should be called when free'ing a copied |
383 | * version of a tc_u_knode obtained from u32_init_knode(). When | |
384 | * copies are obtained from u32_init_knode() the statistics are | |
385 | * shared between the old and new copies to allow readers to | |
386 | * continue to update the statistics during the copy. To support | |
387 | * this the u32_delete_key_rcu variant does not free the percpu | |
388 | * statistics. | |
389 | */ | |
1ce87720 JF |
390 | static void u32_delete_key_rcu(struct rcu_head *rcu) |
391 | { | |
392 | struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); | |
393 | ||
de5df632 JF |
394 | u32_destroy_key(key->tp, key, false); |
395 | } | |
396 | ||
397 | /* u32_delete_key_freepf_rcu is the rcu callback variant | |
398 | * that free's the entire structure including the statistics | |
399 | * percpu variables. Only use this if the key is not a copy | |
400 | * returned by u32_init_knode(). See u32_delete_key_rcu() | |
401 | * for the variant that should be used with keys return from | |
402 | * u32_init_knode() | |
403 | */ | |
404 | static void u32_delete_key_freepf_rcu(struct rcu_head *rcu) | |
405 | { | |
406 | struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu); | |
407 | ||
408 | u32_destroy_key(key->tp, key, true); | |
1ce87720 JF |
409 | } |
410 | ||
82d567c2 | 411 | static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key) |
1da177e4 | 412 | { |
1ce87720 JF |
413 | struct tc_u_knode __rcu **kp; |
414 | struct tc_u_knode *pkp; | |
a96366bf | 415 | struct tc_u_hnode *ht = rtnl_dereference(key->ht_up); |
1da177e4 LT |
416 | |
417 | if (ht) { | |
1ce87720 JF |
418 | kp = &ht->ht[TC_U32_HASH(key->handle)]; |
419 | for (pkp = rtnl_dereference(*kp); pkp; | |
420 | kp = &pkp->next, pkp = rtnl_dereference(*kp)) { | |
421 | if (pkp == key) { | |
422 | RCU_INIT_POINTER(*kp, key->next); | |
1da177e4 | 423 | |
a0efb80c | 424 | tcf_unbind_filter(tp, &key->res); |
de5df632 | 425 | call_rcu(&key->rcu, u32_delete_key_freepf_rcu); |
1da177e4 LT |
426 | return 0; |
427 | } | |
428 | } | |
429 | } | |
547b792c | 430 | WARN_ON(1); |
1da177e4 LT |
431 | return 0; |
432 | } | |
433 | ||
a1b7c5fd JF |
434 | static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle) |
435 | { | |
436 | struct net_device *dev = tp->q->dev_queue->dev; | |
437 | struct tc_cls_u32_offload u32_offload = {0}; | |
438 | struct tc_to_netdev offload; | |
439 | ||
440 | offload.type = TC_SETUP_CLSU32; | |
441 | offload.cls_u32 = &u32_offload; | |
442 | ||
9e8ce79c | 443 | if (tc_should_offload(dev, 0)) { |
a1b7c5fd JF |
444 | offload.cls_u32->command = TC_CLSU32_DELETE_KNODE; |
445 | offload.cls_u32->knode.handle = handle; | |
446 | dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, | |
447 | tp->protocol, &offload); | |
448 | } | |
449 | } | |
450 | ||
d34e3e18 | 451 | static int u32_replace_hw_hnode(struct tcf_proto *tp, |
9e8ce79c JF |
452 | struct tc_u_hnode *h, |
453 | u32 flags) | |
a1b7c5fd JF |
454 | { |
455 | struct net_device *dev = tp->q->dev_queue->dev; | |
456 | struct tc_cls_u32_offload u32_offload = {0}; | |
457 | struct tc_to_netdev offload; | |
d34e3e18 | 458 | int err; |
a1b7c5fd | 459 | |
d47a0f38 JK |
460 | if (!tc_should_offload(dev, flags)) |
461 | return tc_skip_sw(flags) ? -EINVAL : 0; | |
462 | ||
a1b7c5fd JF |
463 | offload.type = TC_SETUP_CLSU32; |
464 | offload.cls_u32 = &u32_offload; | |
465 | ||
d47a0f38 JK |
466 | offload.cls_u32->command = TC_CLSU32_NEW_HNODE; |
467 | offload.cls_u32->hnode.divisor = h->divisor; | |
468 | offload.cls_u32->hnode.handle = h->handle; | |
469 | offload.cls_u32->hnode.prio = h->prio; | |
a1b7c5fd | 470 | |
d47a0f38 JK |
471 | err = dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, |
472 | tp->protocol, &offload); | |
473 | if (tc_skip_sw(flags)) | |
474 | return err; | |
d34e3e18 SS |
475 | |
476 | return 0; | |
a1b7c5fd JF |
477 | } |
478 | ||
479 | static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h) | |
480 | { | |
481 | struct net_device *dev = tp->q->dev_queue->dev; | |
482 | struct tc_cls_u32_offload u32_offload = {0}; | |
483 | struct tc_to_netdev offload; | |
484 | ||
485 | offload.type = TC_SETUP_CLSU32; | |
486 | offload.cls_u32 = &u32_offload; | |
487 | ||
9e8ce79c | 488 | if (tc_should_offload(dev, 0)) { |
a1b7c5fd JF |
489 | offload.cls_u32->command = TC_CLSU32_DELETE_HNODE; |
490 | offload.cls_u32->hnode.divisor = h->divisor; | |
491 | offload.cls_u32->hnode.handle = h->handle; | |
492 | offload.cls_u32->hnode.prio = h->prio; | |
493 | ||
494 | dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, | |
495 | tp->protocol, &offload); | |
496 | } | |
497 | } | |
498 | ||
d34e3e18 | 499 | static int u32_replace_hw_knode(struct tcf_proto *tp, |
9e8ce79c JF |
500 | struct tc_u_knode *n, |
501 | u32 flags) | |
a1b7c5fd JF |
502 | { |
503 | struct net_device *dev = tp->q->dev_queue->dev; | |
504 | struct tc_cls_u32_offload u32_offload = {0}; | |
505 | struct tc_to_netdev offload; | |
d34e3e18 | 506 | int err; |
a1b7c5fd JF |
507 | |
508 | offload.type = TC_SETUP_CLSU32; | |
509 | offload.cls_u32 = &u32_offload; | |
510 | ||
9e8ce79c | 511 | if (tc_should_offload(dev, flags)) { |
a1b7c5fd JF |
512 | offload.cls_u32->command = TC_CLSU32_REPLACE_KNODE; |
513 | offload.cls_u32->knode.handle = n->handle; | |
514 | offload.cls_u32->knode.fshift = n->fshift; | |
515 | #ifdef CONFIG_CLS_U32_MARK | |
516 | offload.cls_u32->knode.val = n->val; | |
517 | offload.cls_u32->knode.mask = n->mask; | |
518 | #else | |
519 | offload.cls_u32->knode.val = 0; | |
520 | offload.cls_u32->knode.mask = 0; | |
521 | #endif | |
522 | offload.cls_u32->knode.sel = &n->sel; | |
523 | offload.cls_u32->knode.exts = &n->exts; | |
524 | if (n->ht_down) | |
525 | offload.cls_u32->knode.link_handle = n->ht_down->handle; | |
526 | ||
d34e3e18 SS |
527 | err = dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, |
528 | tp->protocol, &offload); | |
529 | if (tc_skip_sw(flags)) | |
530 | return err; | |
a1b7c5fd | 531 | } |
d34e3e18 SS |
532 | |
533 | return 0; | |
a1b7c5fd JF |
534 | } |
535 | ||
a0efb80c | 536 | static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) |
1da177e4 LT |
537 | { |
538 | struct tc_u_knode *n; | |
cc7ec456 | 539 | unsigned int h; |
1da177e4 | 540 | |
cc7ec456 | 541 | for (h = 0; h <= ht->divisor; h++) { |
1ce87720 JF |
542 | while ((n = rtnl_dereference(ht->ht[h])) != NULL) { |
543 | RCU_INIT_POINTER(ht->ht[h], | |
544 | rtnl_dereference(n->next)); | |
a0efb80c | 545 | tcf_unbind_filter(tp, &n->res); |
a1b7c5fd | 546 | u32_remove_hw_knode(tp, n->handle); |
de5df632 | 547 | call_rcu(&n->rcu, u32_delete_key_freepf_rcu); |
1da177e4 LT |
548 | } |
549 | } | |
550 | } | |
551 | ||
552 | static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) | |
553 | { | |
554 | struct tc_u_common *tp_c = tp->data; | |
1ce87720 JF |
555 | struct tc_u_hnode __rcu **hn; |
556 | struct tc_u_hnode *phn; | |
1da177e4 | 557 | |
547b792c | 558 | WARN_ON(ht->refcnt); |
1da177e4 | 559 | |
a0efb80c | 560 | u32_clear_hnode(tp, ht); |
1da177e4 | 561 | |
1ce87720 JF |
562 | hn = &tp_c->hlist; |
563 | for (phn = rtnl_dereference(*hn); | |
564 | phn; | |
565 | hn = &phn->next, phn = rtnl_dereference(*hn)) { | |
566 | if (phn == ht) { | |
a1b7c5fd | 567 | u32_clear_hw_hnode(tp, ht); |
1ce87720 JF |
568 | RCU_INIT_POINTER(*hn, ht->next); |
569 | kfree_rcu(ht, rcu); | |
1da177e4 LT |
570 | return 0; |
571 | } | |
572 | } | |
573 | ||
1da177e4 LT |
574 | return -ENOENT; |
575 | } | |
576 | ||
1e052be6 CW |
577 | static bool ht_empty(struct tc_u_hnode *ht) |
578 | { | |
579 | unsigned int h; | |
580 | ||
581 | for (h = 0; h <= ht->divisor; h++) | |
582 | if (rcu_access_pointer(ht->ht[h])) | |
583 | return false; | |
584 | ||
585 | return true; | |
586 | } | |
587 | ||
588 | static bool u32_destroy(struct tcf_proto *tp, bool force) | |
1da177e4 LT |
589 | { |
590 | struct tc_u_common *tp_c = tp->data; | |
1ce87720 | 591 | struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); |
1da177e4 | 592 | |
547b792c | 593 | WARN_ON(root_ht == NULL); |
1da177e4 | 594 | |
1e052be6 CW |
595 | if (!force) { |
596 | if (root_ht) { | |
597 | if (root_ht->refcnt > 1) | |
598 | return false; | |
599 | if (root_ht->refcnt == 1) { | |
600 | if (!ht_empty(root_ht)) | |
601 | return false; | |
602 | } | |
603 | } | |
a6c1aea0 WC |
604 | |
605 | if (tp_c->refcnt > 1) | |
606 | return false; | |
607 | ||
608 | if (tp_c->refcnt == 1) { | |
609 | struct tc_u_hnode *ht; | |
610 | ||
611 | for (ht = rtnl_dereference(tp_c->hlist); | |
612 | ht; | |
613 | ht = rtnl_dereference(ht->next)) | |
614 | if (!ht_empty(ht)) | |
615 | return false; | |
616 | } | |
1e052be6 CW |
617 | } |
618 | ||
1da177e4 LT |
619 | if (root_ht && --root_ht->refcnt == 0) |
620 | u32_destroy_hnode(tp, root_ht); | |
621 | ||
622 | if (--tp_c->refcnt == 0) { | |
623 | struct tc_u_hnode *ht; | |
1da177e4 | 624 | |
72b25a91 | 625 | tp->q->u32_node = NULL; |
1da177e4 | 626 | |
1ce87720 JF |
627 | for (ht = rtnl_dereference(tp_c->hlist); |
628 | ht; | |
629 | ht = rtnl_dereference(ht->next)) { | |
e56cfad1 | 630 | ht->refcnt--; |
a0efb80c | 631 | u32_clear_hnode(tp, ht); |
e56cfad1 | 632 | } |
1da177e4 | 633 | |
1ce87720 JF |
634 | while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) { |
635 | RCU_INIT_POINTER(tp_c->hlist, ht->next); | |
636 | kfree_rcu(ht, rcu); | |
3ff50b79 | 637 | } |
1da177e4 LT |
638 | |
639 | kfree(tp_c); | |
640 | } | |
641 | ||
642 | tp->data = NULL; | |
1e052be6 | 643 | return true; |
1da177e4 LT |
644 | } |
645 | ||
646 | static int u32_delete(struct tcf_proto *tp, unsigned long arg) | |
647 | { | |
cc7ec456 | 648 | struct tc_u_hnode *ht = (struct tc_u_hnode *)arg; |
1ce87720 | 649 | struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); |
1da177e4 LT |
650 | |
651 | if (ht == NULL) | |
652 | return 0; | |
653 | ||
a1b7c5fd JF |
654 | if (TC_U32_KEY(ht->handle)) { |
655 | u32_remove_hw_knode(tp, ht->handle); | |
cc7ec456 | 656 | return u32_delete_key(tp, (struct tc_u_knode *)ht); |
a1b7c5fd | 657 | } |
1da177e4 | 658 | |
1ce87720 | 659 | if (root_ht == ht) |
1da177e4 LT |
660 | return -EINVAL; |
661 | ||
e56cfad1 JP |
662 | if (ht->refcnt == 1) { |
663 | ht->refcnt--; | |
1da177e4 | 664 | u32_destroy_hnode(tp, ht); |
e56cfad1 JP |
665 | } else { |
666 | return -EBUSY; | |
667 | } | |
1da177e4 LT |
668 | |
669 | return 0; | |
670 | } | |
671 | ||
7801db8a | 672 | #define NR_U32_NODE (1<<12) |
1da177e4 LT |
673 | static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle) |
674 | { | |
675 | struct tc_u_knode *n; | |
7801db8a CW |
676 | unsigned long i; |
677 | unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long), | |
678 | GFP_KERNEL); | |
679 | if (!bitmap) | |
680 | return handle | 0xFFF; | |
1da177e4 | 681 | |
1ce87720 JF |
682 | for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]); |
683 | n; | |
684 | n = rtnl_dereference(n->next)) | |
7801db8a | 685 | set_bit(TC_U32_NODE(n->handle), bitmap); |
1da177e4 | 686 | |
7801db8a CW |
687 | i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800); |
688 | if (i >= NR_U32_NODE) | |
689 | i = find_next_zero_bit(bitmap, NR_U32_NODE, 1); | |
690 | ||
691 | kfree(bitmap); | |
692 | return handle | (i >= NR_U32_NODE ? 0xFFF : i); | |
1da177e4 LT |
693 | } |
694 | ||
6fa8c014 PM |
695 | static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = { |
696 | [TCA_U32_CLASSID] = { .type = NLA_U32 }, | |
697 | [TCA_U32_HASH] = { .type = NLA_U32 }, | |
698 | [TCA_U32_LINK] = { .type = NLA_U32 }, | |
699 | [TCA_U32_DIVISOR] = { .type = NLA_U32 }, | |
700 | [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) }, | |
701 | [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ }, | |
702 | [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) }, | |
9e8ce79c | 703 | [TCA_U32_FLAGS] = { .type = NLA_U32 }, |
6fa8c014 PM |
704 | }; |
705 | ||
c1b52739 BL |
706 | static int u32_set_parms(struct net *net, struct tcf_proto *tp, |
707 | unsigned long base, struct tc_u_hnode *ht, | |
add93b61 | 708 | struct tc_u_knode *n, struct nlattr **tb, |
2f7ef2f8 | 709 | struct nlattr *est, bool ovr) |
1da177e4 LT |
710 | { |
711 | int err; | |
712 | struct tcf_exts e; | |
713 | ||
5da57f42 | 714 | tcf_exts_init(&e, TCA_U32_ACT, TCA_U32_POLICE); |
2f7ef2f8 | 715 | err = tcf_exts_validate(net, tp, tb, est, &e, ovr); |
1da177e4 LT |
716 | if (err < 0) |
717 | return err; | |
718 | ||
719 | err = -EINVAL; | |
add93b61 | 720 | if (tb[TCA_U32_LINK]) { |
1587bac4 | 721 | u32 handle = nla_get_u32(tb[TCA_U32_LINK]); |
47a1a1d4 | 722 | struct tc_u_hnode *ht_down = NULL, *ht_old; |
1da177e4 LT |
723 | |
724 | if (TC_U32_KEY(handle)) | |
725 | goto errout; | |
726 | ||
727 | if (handle) { | |
728 | ht_down = u32_lookup_ht(ht->tp_c, handle); | |
729 | ||
730 | if (ht_down == NULL) | |
731 | goto errout; | |
732 | ht_down->refcnt++; | |
733 | } | |
734 | ||
1ce87720 JF |
735 | ht_old = rtnl_dereference(n->ht_down); |
736 | rcu_assign_pointer(n->ht_down, ht_down); | |
1da177e4 | 737 | |
47a1a1d4 PM |
738 | if (ht_old) |
739 | ht_old->refcnt--; | |
1da177e4 | 740 | } |
add93b61 | 741 | if (tb[TCA_U32_CLASSID]) { |
1587bac4 | 742 | n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]); |
1da177e4 LT |
743 | tcf_bind_filter(tp, &n->res, base); |
744 | } | |
745 | ||
746 | #ifdef CONFIG_NET_CLS_IND | |
add93b61 | 747 | if (tb[TCA_U32_INDEV]) { |
2519a602 WC |
748 | int ret; |
749 | ret = tcf_change_indev(net, tb[TCA_U32_INDEV]); | |
750 | if (ret < 0) | |
1da177e4 | 751 | goto errout; |
2519a602 | 752 | n->ifindex = ret; |
1da177e4 LT |
753 | } |
754 | #endif | |
755 | tcf_exts_change(tp, &n->exts, &e); | |
756 | ||
757 | return 0; | |
758 | errout: | |
18d0264f | 759 | tcf_exts_destroy(&e); |
1da177e4 LT |
760 | return err; |
761 | } | |
762 | ||
de5df632 JF |
763 | static void u32_replace_knode(struct tcf_proto *tp, |
764 | struct tc_u_common *tp_c, | |
765 | struct tc_u_knode *n) | |
766 | { | |
767 | struct tc_u_knode __rcu **ins; | |
768 | struct tc_u_knode *pins; | |
769 | struct tc_u_hnode *ht; | |
770 | ||
771 | if (TC_U32_HTID(n->handle) == TC_U32_ROOT) | |
772 | ht = rtnl_dereference(tp->root); | |
773 | else | |
774 | ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle)); | |
775 | ||
776 | ins = &ht->ht[TC_U32_HASH(n->handle)]; | |
777 | ||
778 | /* The node must always exist for it to be replaced if this is not the | |
779 | * case then something went very wrong elsewhere. | |
780 | */ | |
781 | for (pins = rtnl_dereference(*ins); ; | |
782 | ins = &pins->next, pins = rtnl_dereference(*ins)) | |
783 | if (pins->handle == n->handle) | |
784 | break; | |
785 | ||
786 | RCU_INIT_POINTER(n->next, pins->next); | |
787 | rcu_assign_pointer(*ins, n); | |
788 | } | |
789 | ||
790 | static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp, | |
791 | struct tc_u_knode *n) | |
792 | { | |
793 | struct tc_u_knode *new; | |
794 | struct tc_u32_sel *s = &n->sel; | |
795 | ||
796 | new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), | |
797 | GFP_KERNEL); | |
798 | ||
799 | if (!new) | |
800 | return NULL; | |
801 | ||
802 | RCU_INIT_POINTER(new->next, n->next); | |
803 | new->handle = n->handle; | |
804 | RCU_INIT_POINTER(new->ht_up, n->ht_up); | |
805 | ||
806 | #ifdef CONFIG_NET_CLS_IND | |
807 | new->ifindex = n->ifindex; | |
808 | #endif | |
809 | new->fshift = n->fshift; | |
810 | new->res = n->res; | |
9e8ce79c | 811 | new->flags = n->flags; |
de5df632 JF |
812 | RCU_INIT_POINTER(new->ht_down, n->ht_down); |
813 | ||
814 | /* bump reference count as long as we hold pointer to structure */ | |
815 | if (new->ht_down) | |
816 | new->ht_down->refcnt++; | |
817 | ||
818 | #ifdef CONFIG_CLS_U32_PERF | |
819 | /* Statistics may be incremented by readers during update | |
820 | * so we must keep them in tact. When the node is later destroyed | |
821 | * a special destroy call must be made to not free the pf memory. | |
822 | */ | |
823 | new->pf = n->pf; | |
824 | #endif | |
825 | ||
826 | #ifdef CONFIG_CLS_U32_MARK | |
827 | new->val = n->val; | |
828 | new->mask = n->mask; | |
829 | /* Similarly success statistics must be moved as pointers */ | |
830 | new->pcpu_success = n->pcpu_success; | |
831 | #endif | |
832 | new->tp = tp; | |
833 | memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); | |
834 | ||
835 | tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE); | |
836 | ||
837 | return new; | |
838 | } | |
839 | ||
c1b52739 | 840 | static int u32_change(struct net *net, struct sk_buff *in_skb, |
af4c6641 | 841 | struct tcf_proto *tp, unsigned long base, u32 handle, |
add93b61 | 842 | struct nlattr **tca, |
2f7ef2f8 | 843 | unsigned long *arg, bool ovr) |
1da177e4 LT |
844 | { |
845 | struct tc_u_common *tp_c = tp->data; | |
846 | struct tc_u_hnode *ht; | |
847 | struct tc_u_knode *n; | |
848 | struct tc_u32_sel *s; | |
add93b61 PM |
849 | struct nlattr *opt = tca[TCA_OPTIONS]; |
850 | struct nlattr *tb[TCA_U32_MAX + 1]; | |
9e8ce79c | 851 | u32 htid, flags = 0; |
1da177e4 | 852 | int err; |
459d5f62 JF |
853 | #ifdef CONFIG_CLS_U32_PERF |
854 | size_t size; | |
855 | #endif | |
1da177e4 LT |
856 | |
857 | if (opt == NULL) | |
858 | return handle ? -EINVAL : 0; | |
859 | ||
6fa8c014 | 860 | err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy); |
cee63723 PM |
861 | if (err < 0) |
862 | return err; | |
1da177e4 | 863 | |
d34e3e18 | 864 | if (tb[TCA_U32_FLAGS]) { |
9e8ce79c | 865 | flags = nla_get_u32(tb[TCA_U32_FLAGS]); |
d34e3e18 | 866 | if (!tc_flags_valid(flags)) |
1a0f7d29 | 867 | return -EINVAL; |
d34e3e18 | 868 | } |
9e8ce79c | 869 | |
cc7ec456 ED |
870 | n = (struct tc_u_knode *)*arg; |
871 | if (n) { | |
de5df632 JF |
872 | struct tc_u_knode *new; |
873 | ||
1da177e4 LT |
874 | if (TC_U32_KEY(n->handle) == 0) |
875 | return -EINVAL; | |
876 | ||
9e8ce79c JF |
877 | if (n->flags != flags) |
878 | return -EINVAL; | |
879 | ||
de5df632 JF |
880 | new = u32_init_knode(tp, n); |
881 | if (!new) | |
882 | return -ENOMEM; | |
883 | ||
884 | err = u32_set_parms(net, tp, base, | |
885 | rtnl_dereference(n->ht_up), new, tb, | |
886 | tca[TCA_RATE], ovr); | |
887 | ||
888 | if (err) { | |
889 | u32_destroy_key(tp, new, false); | |
890 | return err; | |
891 | } | |
892 | ||
d34e3e18 SS |
893 | err = u32_replace_hw_knode(tp, new, flags); |
894 | if (err) { | |
895 | u32_destroy_key(tp, new, false); | |
896 | return err; | |
897 | } | |
898 | ||
de5df632 | 899 | u32_replace_knode(tp, tp_c, new); |
a0efb80c | 900 | tcf_unbind_filter(tp, &n->res); |
de5df632 JF |
901 | call_rcu(&n->rcu, u32_delete_key_rcu); |
902 | return 0; | |
1da177e4 LT |
903 | } |
904 | ||
add93b61 | 905 | if (tb[TCA_U32_DIVISOR]) { |
cc7ec456 | 906 | unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]); |
1da177e4 LT |
907 | |
908 | if (--divisor > 0x100) | |
909 | return -EINVAL; | |
910 | if (TC_U32_KEY(handle)) | |
911 | return -EINVAL; | |
912 | if (handle == 0) { | |
913 | handle = gen_new_htid(tp->data); | |
914 | if (handle == 0) | |
915 | return -ENOMEM; | |
916 | } | |
cc7ec456 | 917 | ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL); |
1da177e4 LT |
918 | if (ht == NULL) |
919 | return -ENOBUFS; | |
1da177e4 | 920 | ht->tp_c = tp_c; |
e56cfad1 | 921 | ht->refcnt = 1; |
1da177e4 LT |
922 | ht->divisor = divisor; |
923 | ht->handle = handle; | |
924 | ht->prio = tp->prio; | |
1ce87720 JF |
925 | RCU_INIT_POINTER(ht->next, tp_c->hlist); |
926 | rcu_assign_pointer(tp_c->hlist, ht); | |
1da177e4 | 927 | *arg = (unsigned long)ht; |
a1b7c5fd | 928 | |
9e8ce79c | 929 | u32_replace_hw_hnode(tp, ht, flags); |
1da177e4 LT |
930 | return 0; |
931 | } | |
932 | ||
add93b61 | 933 | if (tb[TCA_U32_HASH]) { |
1587bac4 | 934 | htid = nla_get_u32(tb[TCA_U32_HASH]); |
1da177e4 | 935 | if (TC_U32_HTID(htid) == TC_U32_ROOT) { |
1ce87720 | 936 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
937 | htid = ht->handle; |
938 | } else { | |
939 | ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid)); | |
940 | if (ht == NULL) | |
941 | return -EINVAL; | |
942 | } | |
943 | } else { | |
1ce87720 | 944 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
945 | htid = ht->handle; |
946 | } | |
947 | ||
948 | if (ht->divisor < TC_U32_HASH(htid)) | |
949 | return -EINVAL; | |
950 | ||
951 | if (handle) { | |
952 | if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid)) | |
953 | return -EINVAL; | |
954 | handle = htid | TC_U32_NODE(handle); | |
955 | } else | |
956 | handle = gen_new_kid(ht, htid); | |
957 | ||
6fa8c014 | 958 | if (tb[TCA_U32_SEL] == NULL) |
1da177e4 LT |
959 | return -EINVAL; |
960 | ||
add93b61 | 961 | s = nla_data(tb[TCA_U32_SEL]); |
1da177e4 | 962 | |
0da974f4 | 963 | n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL); |
1da177e4 LT |
964 | if (n == NULL) |
965 | return -ENOBUFS; | |
966 | ||
1da177e4 | 967 | #ifdef CONFIG_CLS_U32_PERF |
459d5f62 JF |
968 | size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64); |
969 | n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt)); | |
970 | if (!n->pf) { | |
1da177e4 LT |
971 | kfree(n); |
972 | return -ENOBUFS; | |
973 | } | |
1da177e4 LT |
974 | #endif |
975 | ||
976 | memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); | |
a96366bf | 977 | RCU_INIT_POINTER(n->ht_up, ht); |
1da177e4 | 978 | n->handle = handle; |
b2268016 | 979 | n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0; |
9e8ce79c | 980 | n->flags = flags; |
5da57f42 | 981 | tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE); |
1ce87720 | 982 | n->tp = tp; |
1da177e4 LT |
983 | |
984 | #ifdef CONFIG_CLS_U32_MARK | |
459d5f62 | 985 | n->pcpu_success = alloc_percpu(u32); |
a1ddcfee JF |
986 | if (!n->pcpu_success) { |
987 | err = -ENOMEM; | |
988 | goto errout; | |
989 | } | |
459d5f62 | 990 | |
add93b61 | 991 | if (tb[TCA_U32_MARK]) { |
1da177e4 LT |
992 | struct tc_u32_mark *mark; |
993 | ||
add93b61 | 994 | mark = nla_data(tb[TCA_U32_MARK]); |
459d5f62 JF |
995 | n->val = mark->val; |
996 | n->mask = mark->mask; | |
1da177e4 LT |
997 | } |
998 | #endif | |
999 | ||
2f7ef2f8 | 1000 | err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr); |
1da177e4 | 1001 | if (err == 0) { |
1ce87720 JF |
1002 | struct tc_u_knode __rcu **ins; |
1003 | struct tc_u_knode *pins; | |
1004 | ||
d34e3e18 SS |
1005 | err = u32_replace_hw_knode(tp, n, flags); |
1006 | if (err) | |
1007 | goto errhw; | |
1008 | ||
1ce87720 JF |
1009 | ins = &ht->ht[TC_U32_HASH(handle)]; |
1010 | for (pins = rtnl_dereference(*ins); pins; | |
1011 | ins = &pins->next, pins = rtnl_dereference(*ins)) | |
1012 | if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle)) | |
1da177e4 LT |
1013 | break; |
1014 | ||
1ce87720 JF |
1015 | RCU_INIT_POINTER(n->next, pins); |
1016 | rcu_assign_pointer(*ins, n); | |
1da177e4 LT |
1017 | *arg = (unsigned long)n; |
1018 | return 0; | |
1019 | } | |
a1ddcfee | 1020 | |
d34e3e18 | 1021 | errhw: |
a1ddcfee JF |
1022 | #ifdef CONFIG_CLS_U32_MARK |
1023 | free_percpu(n->pcpu_success); | |
a2aeb02a | 1024 | errout: |
a1ddcfee JF |
1025 | #endif |
1026 | ||
1da177e4 | 1027 | #ifdef CONFIG_CLS_U32_PERF |
1ce87720 | 1028 | free_percpu(n->pf); |
1da177e4 LT |
1029 | #endif |
1030 | kfree(n); | |
1031 | return err; | |
1032 | } | |
1033 | ||
1034 | static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg) | |
1035 | { | |
1036 | struct tc_u_common *tp_c = tp->data; | |
1037 | struct tc_u_hnode *ht; | |
1038 | struct tc_u_knode *n; | |
cc7ec456 | 1039 | unsigned int h; |
1da177e4 LT |
1040 | |
1041 | if (arg->stop) | |
1042 | return; | |
1043 | ||
1ce87720 JF |
1044 | for (ht = rtnl_dereference(tp_c->hlist); |
1045 | ht; | |
1046 | ht = rtnl_dereference(ht->next)) { | |
1da177e4 LT |
1047 | if (ht->prio != tp->prio) |
1048 | continue; | |
1049 | if (arg->count >= arg->skip) { | |
1050 | if (arg->fn(tp, (unsigned long)ht, arg) < 0) { | |
1051 | arg->stop = 1; | |
1052 | return; | |
1053 | } | |
1054 | } | |
1055 | arg->count++; | |
1056 | for (h = 0; h <= ht->divisor; h++) { | |
1ce87720 JF |
1057 | for (n = rtnl_dereference(ht->ht[h]); |
1058 | n; | |
1059 | n = rtnl_dereference(n->next)) { | |
1da177e4 LT |
1060 | if (arg->count < arg->skip) { |
1061 | arg->count++; | |
1062 | continue; | |
1063 | } | |
1064 | if (arg->fn(tp, (unsigned long)n, arg) < 0) { | |
1065 | arg->stop = 1; | |
1066 | return; | |
1067 | } | |
1068 | arg->count++; | |
1069 | } | |
1070 | } | |
1071 | } | |
1072 | } | |
1073 | ||
832d1d5b | 1074 | static int u32_dump(struct net *net, struct tcf_proto *tp, unsigned long fh, |
1da177e4 LT |
1075 | struct sk_buff *skb, struct tcmsg *t) |
1076 | { | |
cc7ec456 | 1077 | struct tc_u_knode *n = (struct tc_u_knode *)fh; |
1ce87720 | 1078 | struct tc_u_hnode *ht_up, *ht_down; |
4b3550ef | 1079 | struct nlattr *nest; |
1da177e4 LT |
1080 | |
1081 | if (n == NULL) | |
1082 | return skb->len; | |
1083 | ||
1084 | t->tcm_handle = n->handle; | |
1085 | ||
4b3550ef PM |
1086 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1087 | if (nest == NULL) | |
1088 | goto nla_put_failure; | |
1da177e4 LT |
1089 | |
1090 | if (TC_U32_KEY(n->handle) == 0) { | |
cc7ec456 ED |
1091 | struct tc_u_hnode *ht = (struct tc_u_hnode *)fh; |
1092 | u32 divisor = ht->divisor + 1; | |
1093 | ||
1b34ec43 DM |
1094 | if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor)) |
1095 | goto nla_put_failure; | |
1da177e4 | 1096 | } else { |
459d5f62 JF |
1097 | #ifdef CONFIG_CLS_U32_PERF |
1098 | struct tc_u32_pcnt *gpf; | |
459d5f62 | 1099 | int cpu; |
80aab73d | 1100 | #endif |
459d5f62 | 1101 | |
1b34ec43 DM |
1102 | if (nla_put(skb, TCA_U32_SEL, |
1103 | sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key), | |
1104 | &n->sel)) | |
1105 | goto nla_put_failure; | |
1ce87720 JF |
1106 | |
1107 | ht_up = rtnl_dereference(n->ht_up); | |
1108 | if (ht_up) { | |
1da177e4 | 1109 | u32 htid = n->handle & 0xFFFFF000; |
1b34ec43 DM |
1110 | if (nla_put_u32(skb, TCA_U32_HASH, htid)) |
1111 | goto nla_put_failure; | |
1da177e4 | 1112 | } |
1b34ec43 DM |
1113 | if (n->res.classid && |
1114 | nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid)) | |
1115 | goto nla_put_failure; | |
1ce87720 JF |
1116 | |
1117 | ht_down = rtnl_dereference(n->ht_down); | |
1118 | if (ht_down && | |
1119 | nla_put_u32(skb, TCA_U32_LINK, ht_down->handle)) | |
1b34ec43 | 1120 | goto nla_put_failure; |
1da177e4 | 1121 | |
9e8ce79c JF |
1122 | if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags)) |
1123 | goto nla_put_failure; | |
1124 | ||
1da177e4 | 1125 | #ifdef CONFIG_CLS_U32_MARK |
459d5f62 JF |
1126 | if ((n->val || n->mask)) { |
1127 | struct tc_u32_mark mark = {.val = n->val, | |
1128 | .mask = n->mask, | |
1129 | .success = 0}; | |
80aab73d | 1130 | int cpum; |
459d5f62 | 1131 | |
80aab73d JF |
1132 | for_each_possible_cpu(cpum) { |
1133 | __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum); | |
459d5f62 JF |
1134 | |
1135 | mark.success += cnt; | |
1136 | } | |
1137 | ||
1138 | if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark)) | |
1139 | goto nla_put_failure; | |
1140 | } | |
1da177e4 LT |
1141 | #endif |
1142 | ||
5da57f42 | 1143 | if (tcf_exts_dump(skb, &n->exts) < 0) |
add93b61 | 1144 | goto nla_put_failure; |
1da177e4 LT |
1145 | |
1146 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 WC |
1147 | if (n->ifindex) { |
1148 | struct net_device *dev; | |
1149 | dev = __dev_get_by_index(net, n->ifindex); | |
1150 | if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name)) | |
1151 | goto nla_put_failure; | |
1152 | } | |
1da177e4 LT |
1153 | #endif |
1154 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 JF |
1155 | gpf = kzalloc(sizeof(struct tc_u32_pcnt) + |
1156 | n->sel.nkeys * sizeof(u64), | |
1157 | GFP_KERNEL); | |
1158 | if (!gpf) | |
1159 | goto nla_put_failure; | |
1160 | ||
1161 | for_each_possible_cpu(cpu) { | |
1162 | int i; | |
1163 | struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu); | |
1164 | ||
1165 | gpf->rcnt += pf->rcnt; | |
1166 | gpf->rhit += pf->rhit; | |
1167 | for (i = 0; i < n->sel.nkeys; i++) | |
1168 | gpf->kcnts[i] += pf->kcnts[i]; | |
1169 | } | |
1170 | ||
9854518e ND |
1171 | if (nla_put_64bit(skb, TCA_U32_PCNT, |
1172 | sizeof(struct tc_u32_pcnt) + | |
1173 | n->sel.nkeys * sizeof(u64), | |
1174 | gpf, TCA_U32_PAD)) { | |
459d5f62 | 1175 | kfree(gpf); |
1b34ec43 | 1176 | goto nla_put_failure; |
459d5f62 JF |
1177 | } |
1178 | kfree(gpf); | |
1da177e4 LT |
1179 | #endif |
1180 | } | |
1181 | ||
4b3550ef PM |
1182 | nla_nest_end(skb, nest); |
1183 | ||
1da177e4 | 1184 | if (TC_U32_KEY(n->handle)) |
5da57f42 | 1185 | if (tcf_exts_dump_stats(skb, &n->exts) < 0) |
add93b61 | 1186 | goto nla_put_failure; |
1da177e4 LT |
1187 | return skb->len; |
1188 | ||
add93b61 | 1189 | nla_put_failure: |
4b3550ef | 1190 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1191 | return -1; |
1192 | } | |
1193 | ||
2eb9d75c | 1194 | static struct tcf_proto_ops cls_u32_ops __read_mostly = { |
1da177e4 LT |
1195 | .kind = "u32", |
1196 | .classify = u32_classify, | |
1197 | .init = u32_init, | |
1198 | .destroy = u32_destroy, | |
1199 | .get = u32_get, | |
1da177e4 LT |
1200 | .change = u32_change, |
1201 | .delete = u32_delete, | |
1202 | .walk = u32_walk, | |
1203 | .dump = u32_dump, | |
1204 | .owner = THIS_MODULE, | |
1205 | }; | |
1206 | ||
1207 | static int __init init_u32(void) | |
1208 | { | |
6ff9c364 | 1209 | pr_info("u32 classifier\n"); |
1da177e4 | 1210 | #ifdef CONFIG_CLS_U32_PERF |
6ff9c364 | 1211 | pr_info(" Performance counters on\n"); |
1da177e4 | 1212 | #endif |
1da177e4 | 1213 | #ifdef CONFIG_NET_CLS_IND |
6ff9c364 | 1214 | pr_info(" input device check on\n"); |
1da177e4 LT |
1215 | #endif |
1216 | #ifdef CONFIG_NET_CLS_ACT | |
6ff9c364 | 1217 | pr_info(" Actions configured\n"); |
1da177e4 LT |
1218 | #endif |
1219 | return register_tcf_proto_ops(&cls_u32_ops); | |
1220 | } | |
1221 | ||
10297b99 | 1222 | static void __exit exit_u32(void) |
1da177e4 LT |
1223 | { |
1224 | unregister_tcf_proto_ops(&cls_u32_ops); | |
1225 | } | |
1226 | ||
1227 | module_init(init_u32) | |
1228 | module_exit(exit_u32) | |
1229 | MODULE_LICENSE("GPL"); |