]>
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> |
3cd904ec WC |
43 | #include <linux/netdevice.h> |
44 | #include <linux/hash.h> | |
0ba48053 | 45 | #include <net/netlink.h> |
1da177e4 LT |
46 | #include <net/act_api.h> |
47 | #include <net/pkt_cls.h> | |
e7614370 | 48 | #include <linux/idr.h> |
1da177e4 | 49 | |
cc7ec456 | 50 | struct tc_u_knode { |
1ce87720 | 51 | struct tc_u_knode __rcu *next; |
1da177e4 | 52 | u32 handle; |
1ce87720 | 53 | struct tc_u_hnode __rcu *ht_up; |
1da177e4 LT |
54 | struct tcf_exts exts; |
55 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 | 56 | int ifindex; |
1da177e4 LT |
57 | #endif |
58 | u8 fshift; | |
59 | struct tcf_result res; | |
1ce87720 | 60 | struct tc_u_hnode __rcu *ht_down; |
1da177e4 | 61 | #ifdef CONFIG_CLS_U32_PERF |
459d5f62 | 62 | struct tc_u32_pcnt __percpu *pf; |
1da177e4 | 63 | #endif |
9e8ce79c | 64 | u32 flags; |
1da177e4 | 65 | #ifdef CONFIG_CLS_U32_MARK |
459d5f62 JF |
66 | u32 val; |
67 | u32 mask; | |
68 | u32 __percpu *pcpu_success; | |
1da177e4 | 69 | #endif |
1ce87720 | 70 | struct tcf_proto *tp; |
7afe8e4f | 71 | struct rcu_work rwork; |
4e2840ee JF |
72 | /* The 'sel' field MUST be the last field in structure to allow for |
73 | * tc_u32_keys allocated at end of structure. | |
74 | */ | |
75 | struct tc_u32_sel sel; | |
1da177e4 LT |
76 | }; |
77 | ||
cc7ec456 | 78 | struct tc_u_hnode { |
1ce87720 | 79 | struct tc_u_hnode __rcu *next; |
1da177e4 LT |
80 | u32 handle; |
81 | u32 prio; | |
82 | struct tc_u_common *tp_c; | |
83 | int refcnt; | |
cc7ec456 | 84 | unsigned int divisor; |
e7614370 | 85 | struct idr handle_idr; |
1ce87720 | 86 | struct rcu_head rcu; |
5778d39d WC |
87 | /* The 'ht' field MUST be the last field in structure to allow for |
88 | * more entries allocated at end of structure. | |
89 | */ | |
90 | struct tc_u_knode __rcu *ht[1]; | |
1da177e4 LT |
91 | }; |
92 | ||
cc7ec456 | 93 | struct tc_u_common { |
1ce87720 | 94 | struct tc_u_hnode __rcu *hlist; |
7fa9d974 | 95 | struct tcf_block *block; |
1da177e4 | 96 | int refcnt; |
e7614370 | 97 | struct idr handle_idr; |
3cd904ec | 98 | struct hlist_node hnode; |
1ce87720 | 99 | struct rcu_head rcu; |
1da177e4 LT |
100 | }; |
101 | ||
cc7ec456 ED |
102 | static inline unsigned int u32_hash_fold(__be32 key, |
103 | const struct tc_u32_sel *sel, | |
104 | u8 fshift) | |
1da177e4 | 105 | { |
cc7ec456 | 106 | unsigned int h = ntohl(key & sel->hmask) >> fshift; |
1da177e4 LT |
107 | |
108 | return h; | |
109 | } | |
110 | ||
5a7a5555 JHS |
111 | static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, |
112 | struct tcf_result *res) | |
1da177e4 LT |
113 | { |
114 | struct { | |
115 | struct tc_u_knode *knode; | |
fbc2e7d9 | 116 | unsigned int off; |
1da177e4 LT |
117 | } stack[TC_U32_MAXDEPTH]; |
118 | ||
1ce87720 | 119 | struct tc_u_hnode *ht = rcu_dereference_bh(tp->root); |
fbc2e7d9 | 120 | unsigned int off = skb_network_offset(skb); |
1da177e4 LT |
121 | struct tc_u_knode *n; |
122 | int sdepth = 0; | |
123 | int off2 = 0; | |
124 | int sel = 0; | |
125 | #ifdef CONFIG_CLS_U32_PERF | |
126 | int j; | |
127 | #endif | |
128 | int i, r; | |
129 | ||
130 | next_ht: | |
1ce87720 | 131 | n = rcu_dereference_bh(ht->ht[sel]); |
1da177e4 LT |
132 | |
133 | next_knode: | |
134 | if (n) { | |
135 | struct tc_u32_key *key = n->sel.keys; | |
136 | ||
137 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 138 | __this_cpu_inc(n->pf->rcnt); |
1da177e4 LT |
139 | j = 0; |
140 | #endif | |
141 | ||
d34e3e18 SS |
142 | if (tc_skip_sw(n->flags)) { |
143 | n = rcu_dereference_bh(n->next); | |
144 | goto next_knode; | |
145 | } | |
146 | ||
1da177e4 | 147 | #ifdef CONFIG_CLS_U32_MARK |
459d5f62 | 148 | if ((skb->mark & n->mask) != n->val) { |
1ce87720 | 149 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
150 | goto next_knode; |
151 | } else { | |
459d5f62 | 152 | __this_cpu_inc(*n->pcpu_success); |
1da177e4 LT |
153 | } |
154 | #endif | |
155 | ||
cc7ec456 | 156 | for (i = n->sel.nkeys; i > 0; i--, key++) { |
66d50d25 | 157 | int toff = off + key->off + (off2 & key->offmask); |
86fce3ba | 158 | __be32 *data, hdata; |
fbc2e7d9 | 159 | |
4e18b3ed | 160 | if (skb_headroom(skb) + toff > INT_MAX) |
66d50d25 | 161 | goto out; |
162 | ||
86fce3ba | 163 | data = skb_header_pointer(skb, toff, 4, &hdata); |
fbc2e7d9 CG |
164 | if (!data) |
165 | goto out; | |
166 | if ((*data ^ key->val) & key->mask) { | |
1ce87720 | 167 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
168 | goto next_knode; |
169 | } | |
170 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 171 | __this_cpu_inc(n->pf->kcnts[j]); |
1da177e4 LT |
172 | j++; |
173 | #endif | |
174 | } | |
1ce87720 JF |
175 | |
176 | ht = rcu_dereference_bh(n->ht_down); | |
177 | if (!ht) { | |
1da177e4 | 178 | check_terminal: |
cc7ec456 | 179 | if (n->sel.flags & TC_U32_TERMINAL) { |
1da177e4 LT |
180 | |
181 | *res = n->res; | |
182 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 | 183 | if (!tcf_match_indev(skb, n->ifindex)) { |
1ce87720 | 184 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
185 | goto next_knode; |
186 | } | |
187 | #endif | |
188 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 | 189 | __this_cpu_inc(n->pf->rhit); |
1da177e4 LT |
190 | #endif |
191 | r = tcf_exts_exec(skb, &n->exts, res); | |
192 | if (r < 0) { | |
1ce87720 | 193 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
194 | goto next_knode; |
195 | } | |
196 | ||
197 | return r; | |
198 | } | |
1ce87720 | 199 | n = rcu_dereference_bh(n->next); |
1da177e4 LT |
200 | goto next_knode; |
201 | } | |
202 | ||
203 | /* PUSH */ | |
204 | if (sdepth >= TC_U32_MAXDEPTH) | |
205 | goto deadloop; | |
206 | stack[sdepth].knode = n; | |
fbc2e7d9 | 207 | stack[sdepth].off = off; |
1da177e4 LT |
208 | sdepth++; |
209 | ||
1ce87720 | 210 | ht = rcu_dereference_bh(n->ht_down); |
1da177e4 | 211 | sel = 0; |
fbc2e7d9 | 212 | if (ht->divisor) { |
86fce3ba | 213 | __be32 *data, hdata; |
fbc2e7d9 CG |
214 | |
215 | data = skb_header_pointer(skb, off + n->sel.hoff, 4, | |
86fce3ba | 216 | &hdata); |
fbc2e7d9 CG |
217 | if (!data) |
218 | goto out; | |
219 | sel = ht->divisor & u32_hash_fold(*data, &n->sel, | |
220 | n->fshift); | |
221 | } | |
cc7ec456 | 222 | if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT))) |
1da177e4 LT |
223 | goto next_ht; |
224 | ||
cc7ec456 | 225 | if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) { |
1da177e4 | 226 | off2 = n->sel.off + 3; |
fbc2e7d9 | 227 | if (n->sel.flags & TC_U32_VAROFFSET) { |
86fce3ba | 228 | __be16 *data, hdata; |
fbc2e7d9 CG |
229 | |
230 | data = skb_header_pointer(skb, | |
231 | off + n->sel.offoff, | |
86fce3ba | 232 | 2, &hdata); |
fbc2e7d9 CG |
233 | if (!data) |
234 | goto out; | |
235 | off2 += ntohs(n->sel.offmask & *data) >> | |
236 | n->sel.offshift; | |
237 | } | |
1da177e4 LT |
238 | off2 &= ~3; |
239 | } | |
cc7ec456 | 240 | if (n->sel.flags & TC_U32_EAT) { |
fbc2e7d9 | 241 | off += off2; |
1da177e4 LT |
242 | off2 = 0; |
243 | } | |
244 | ||
fbc2e7d9 | 245 | if (off < skb->len) |
1da177e4 LT |
246 | goto next_ht; |
247 | } | |
248 | ||
249 | /* POP */ | |
250 | if (sdepth--) { | |
251 | n = stack[sdepth].knode; | |
1ce87720 | 252 | ht = rcu_dereference_bh(n->ht_up); |
fbc2e7d9 | 253 | off = stack[sdepth].off; |
1da177e4 LT |
254 | goto check_terminal; |
255 | } | |
fbc2e7d9 | 256 | out: |
1da177e4 LT |
257 | return -1; |
258 | ||
259 | deadloop: | |
e87cc472 | 260 | net_warn_ratelimited("cls_u32: dead loop\n"); |
1da177e4 LT |
261 | return -1; |
262 | } | |
263 | ||
5a7a5555 | 264 | static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle) |
1da177e4 LT |
265 | { |
266 | struct tc_u_hnode *ht; | |
267 | ||
1ce87720 JF |
268 | for (ht = rtnl_dereference(tp_c->hlist); |
269 | ht; | |
270 | ht = rtnl_dereference(ht->next)) | |
1da177e4 LT |
271 | if (ht->handle == handle) |
272 | break; | |
273 | ||
274 | return ht; | |
275 | } | |
276 | ||
5a7a5555 | 277 | static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle) |
1da177e4 | 278 | { |
cc7ec456 | 279 | unsigned int sel; |
1da177e4 LT |
280 | struct tc_u_knode *n = NULL; |
281 | ||
282 | sel = TC_U32_HASH(handle); | |
283 | if (sel > ht->divisor) | |
284 | goto out; | |
285 | ||
1ce87720 JF |
286 | for (n = rtnl_dereference(ht->ht[sel]); |
287 | n; | |
288 | n = rtnl_dereference(n->next)) | |
1da177e4 LT |
289 | if (n->handle == handle) |
290 | break; | |
291 | out: | |
292 | return n; | |
293 | } | |
294 | ||
295 | ||
8113c095 | 296 | static void *u32_get(struct tcf_proto *tp, u32 handle) |
1da177e4 LT |
297 | { |
298 | struct tc_u_hnode *ht; | |
299 | struct tc_u_common *tp_c = tp->data; | |
300 | ||
301 | if (TC_U32_HTID(handle) == TC_U32_ROOT) | |
1ce87720 | 302 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
303 | else |
304 | ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle)); | |
305 | ||
306 | if (!ht) | |
8113c095 | 307 | return NULL; |
1da177e4 LT |
308 | |
309 | if (TC_U32_KEY(handle) == 0) | |
8113c095 | 310 | return ht; |
1da177e4 | 311 | |
8113c095 | 312 | return u32_lookup_key(ht, handle); |
1da177e4 LT |
313 | } |
314 | ||
e7614370 | 315 | static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr) |
1da177e4 | 316 | { |
e7614370 CW |
317 | unsigned long idr_index; |
318 | int err; | |
1da177e4 | 319 | |
e7614370 | 320 | /* This is only used inside rtnl lock it is safe to increment |
1ce87720 JF |
321 | * without read _copy_ update semantics |
322 | */ | |
e7614370 CW |
323 | err = idr_alloc_ext(&tp_c->handle_idr, ptr, &idr_index, |
324 | 1, 0x7FF, GFP_KERNEL); | |
325 | if (err) | |
326 | return 0; | |
327 | return (u32)(idr_index | 0x800) << 20; | |
1da177e4 LT |
328 | } |
329 | ||
3cd904ec WC |
330 | static struct hlist_head *tc_u_common_hash; |
331 | ||
332 | #define U32_HASH_SHIFT 10 | |
333 | #define U32_HASH_SIZE (1 << U32_HASH_SHIFT) | |
334 | ||
335 | static unsigned int tc_u_hash(const struct tcf_proto *tp) | |
336 | { | |
d18b4b35 | 337 | return hash_ptr(tp->chain->block, U32_HASH_SHIFT); |
3cd904ec WC |
338 | } |
339 | ||
340 | static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp) | |
341 | { | |
342 | struct tc_u_common *tc; | |
343 | unsigned int h; | |
344 | ||
345 | h = tc_u_hash(tp); | |
346 | hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) { | |
7fa9d974 | 347 | if (tc->block == tp->chain->block) |
3cd904ec WC |
348 | return tc; |
349 | } | |
350 | return NULL; | |
351 | } | |
352 | ||
1da177e4 LT |
353 | static int u32_init(struct tcf_proto *tp) |
354 | { | |
355 | struct tc_u_hnode *root_ht; | |
356 | struct tc_u_common *tp_c; | |
3cd904ec | 357 | unsigned int h; |
1da177e4 | 358 | |
3cd904ec | 359 | tp_c = tc_u_common_find(tp); |
1da177e4 | 360 | |
0da974f4 | 361 | root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL); |
1da177e4 LT |
362 | if (root_ht == NULL) |
363 | return -ENOBUFS; | |
364 | ||
1da177e4 | 365 | root_ht->refcnt++; |
e7614370 | 366 | root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000; |
1da177e4 | 367 | root_ht->prio = tp->prio; |
e7614370 | 368 | idr_init(&root_ht->handle_idr); |
1da177e4 LT |
369 | |
370 | if (tp_c == NULL) { | |
0da974f4 | 371 | tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL); |
1da177e4 LT |
372 | if (tp_c == NULL) { |
373 | kfree(root_ht); | |
374 | return -ENOBUFS; | |
375 | } | |
7fa9d974 | 376 | tp_c->block = tp->chain->block; |
3cd904ec | 377 | INIT_HLIST_NODE(&tp_c->hnode); |
e7614370 | 378 | idr_init(&tp_c->handle_idr); |
3cd904ec WC |
379 | |
380 | h = tc_u_hash(tp); | |
381 | hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]); | |
1da177e4 LT |
382 | } |
383 | ||
384 | tp_c->refcnt++; | |
1ce87720 JF |
385 | RCU_INIT_POINTER(root_ht->next, tp_c->hlist); |
386 | rcu_assign_pointer(tp_c->hlist, root_ht); | |
1da177e4 LT |
387 | root_ht->tp_c = tp_c; |
388 | ||
1ce87720 | 389 | rcu_assign_pointer(tp->root, root_ht); |
1da177e4 LT |
390 | tp->data = tp_c; |
391 | return 0; | |
392 | } | |
393 | ||
5a7a5555 | 394 | static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n, |
de5df632 | 395 | bool free_pf) |
1da177e4 | 396 | { |
360a6fdd PA |
397 | struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); |
398 | ||
18d0264f | 399 | tcf_exts_destroy(&n->exts); |
35c55fc1 | 400 | tcf_exts_put_net(&n->exts); |
360a6fdd PA |
401 | if (ht && --ht->refcnt == 0) |
402 | kfree(ht); | |
1da177e4 | 403 | #ifdef CONFIG_CLS_U32_PERF |
de5df632 JF |
404 | if (free_pf) |
405 | free_percpu(n->pf); | |
a1ddcfee JF |
406 | #endif |
407 | #ifdef CONFIG_CLS_U32_MARK | |
de5df632 JF |
408 | if (free_pf) |
409 | free_percpu(n->pcpu_success); | |
1da177e4 LT |
410 | #endif |
411 | kfree(n); | |
412 | return 0; | |
413 | } | |
414 | ||
de5df632 JF |
415 | /* u32_delete_key_rcu should be called when free'ing a copied |
416 | * version of a tc_u_knode obtained from u32_init_knode(). When | |
417 | * copies are obtained from u32_init_knode() the statistics are | |
418 | * shared between the old and new copies to allow readers to | |
419 | * continue to update the statistics during the copy. To support | |
420 | * this the u32_delete_key_rcu variant does not free the percpu | |
421 | * statistics. | |
422 | */ | |
c0d378ef CW |
423 | static void u32_delete_key_work(struct work_struct *work) |
424 | { | |
7afe8e4f CW |
425 | struct tc_u_knode *key = container_of(to_rcu_work(work), |
426 | struct tc_u_knode, | |
427 | rwork); | |
c0d378ef CW |
428 | rtnl_lock(); |
429 | u32_destroy_key(key->tp, key, false); | |
430 | rtnl_unlock(); | |
431 | } | |
432 | ||
de5df632 JF |
433 | /* u32_delete_key_freepf_rcu is the rcu callback variant |
434 | * that free's the entire structure including the statistics | |
435 | * percpu variables. Only use this if the key is not a copy | |
436 | * returned by u32_init_knode(). See u32_delete_key_rcu() | |
437 | * for the variant that should be used with keys return from | |
438 | * u32_init_knode() | |
439 | */ | |
c0d378ef CW |
440 | static void u32_delete_key_freepf_work(struct work_struct *work) |
441 | { | |
7afe8e4f CW |
442 | struct tc_u_knode *key = container_of(to_rcu_work(work), |
443 | struct tc_u_knode, | |
444 | rwork); | |
c0d378ef CW |
445 | rtnl_lock(); |
446 | u32_destroy_key(key->tp, key, true); | |
447 | rtnl_unlock(); | |
448 | } | |
449 | ||
82d567c2 | 450 | static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key) |
1da177e4 | 451 | { |
1ce87720 JF |
452 | struct tc_u_knode __rcu **kp; |
453 | struct tc_u_knode *pkp; | |
a96366bf | 454 | struct tc_u_hnode *ht = rtnl_dereference(key->ht_up); |
1da177e4 LT |
455 | |
456 | if (ht) { | |
1ce87720 JF |
457 | kp = &ht->ht[TC_U32_HASH(key->handle)]; |
458 | for (pkp = rtnl_dereference(*kp); pkp; | |
459 | kp = &pkp->next, pkp = rtnl_dereference(*kp)) { | |
460 | if (pkp == key) { | |
461 | RCU_INIT_POINTER(*kp, key->next); | |
1da177e4 | 462 | |
a0efb80c | 463 | tcf_unbind_filter(tp, &key->res); |
cf0a9345 | 464 | idr_remove(&ht->handle_idr, key->handle); |
35c55fc1 | 465 | tcf_exts_get_net(&key->exts); |
7afe8e4f | 466 | tcf_queue_work(&key->rwork, u32_delete_key_freepf_work); |
1da177e4 LT |
467 | return 0; |
468 | } | |
469 | } | |
470 | } | |
547b792c | 471 | WARN_ON(1); |
1da177e4 LT |
472 | return 0; |
473 | } | |
474 | ||
77460411 | 475 | static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h) |
a1b7c5fd | 476 | { |
245dc512 | 477 | struct tcf_block *block = tp->chain->block; |
de4784ca | 478 | struct tc_cls_u32_offload cls_u32 = {}; |
a1b7c5fd | 479 | |
de4784ca | 480 | tc_cls_common_offload_init(&cls_u32.common, tp); |
77460411 JP |
481 | cls_u32.command = TC_CLSU32_DELETE_HNODE; |
482 | cls_u32.hnode.divisor = h->divisor; | |
483 | cls_u32.hnode.handle = h->handle; | |
484 | cls_u32.hnode.prio = h->prio; | |
de4784ca | 485 | |
245dc512 | 486 | tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false); |
a1b7c5fd JF |
487 | } |
488 | ||
5a7a5555 JHS |
489 | static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h, |
490 | u32 flags) | |
a1b7c5fd | 491 | { |
245dc512 | 492 | struct tcf_block *block = tp->chain->block; |
de4784ca | 493 | struct tc_cls_u32_offload cls_u32 = {}; |
245dc512 JP |
494 | bool skip_sw = tc_skip_sw(flags); |
495 | bool offloaded = false; | |
d34e3e18 | 496 | int err; |
a1b7c5fd | 497 | |
de4784ca JP |
498 | tc_cls_common_offload_init(&cls_u32.common, tp); |
499 | cls_u32.command = TC_CLSU32_NEW_HNODE; | |
500 | cls_u32.hnode.divisor = h->divisor; | |
501 | cls_u32.hnode.handle = h->handle; | |
502 | cls_u32.hnode.prio = h->prio; | |
a1b7c5fd | 503 | |
245dc512 JP |
504 | err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw); |
505 | if (err < 0) { | |
506 | u32_clear_hw_hnode(tp, h); | |
d47a0f38 | 507 | return err; |
245dc512 JP |
508 | } else if (err > 0) { |
509 | offloaded = true; | |
510 | } | |
511 | ||
512 | if (skip_sw && !offloaded) | |
513 | return -EINVAL; | |
d34e3e18 SS |
514 | |
515 | return 0; | |
a1b7c5fd JF |
516 | } |
517 | ||
77460411 | 518 | static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle) |
a1b7c5fd | 519 | { |
245dc512 | 520 | struct tcf_block *block = tp->chain->block; |
de4784ca | 521 | struct tc_cls_u32_offload cls_u32 = {}; |
a1b7c5fd | 522 | |
de4784ca | 523 | tc_cls_common_offload_init(&cls_u32.common, tp); |
77460411 JP |
524 | cls_u32.command = TC_CLSU32_DELETE_KNODE; |
525 | cls_u32.knode.handle = handle; | |
a1b7c5fd | 526 | |
245dc512 | 527 | tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false); |
a1b7c5fd JF |
528 | } |
529 | ||
5a7a5555 JHS |
530 | static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n, |
531 | u32 flags) | |
a1b7c5fd | 532 | { |
b49f9c88 | 533 | struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); |
245dc512 | 534 | struct tcf_block *block = tp->chain->block; |
de4784ca | 535 | struct tc_cls_u32_offload cls_u32 = {}; |
245dc512 | 536 | bool skip_sw = tc_skip_sw(flags); |
d34e3e18 | 537 | int err; |
a1b7c5fd | 538 | |
de4784ca JP |
539 | tc_cls_common_offload_init(&cls_u32.common, tp); |
540 | cls_u32.command = TC_CLSU32_REPLACE_KNODE; | |
541 | cls_u32.knode.handle = n->handle; | |
542 | cls_u32.knode.fshift = n->fshift; | |
a1b7c5fd | 543 | #ifdef CONFIG_CLS_U32_MARK |
de4784ca JP |
544 | cls_u32.knode.val = n->val; |
545 | cls_u32.knode.mask = n->mask; | |
a1b7c5fd | 546 | #else |
de4784ca JP |
547 | cls_u32.knode.val = 0; |
548 | cls_u32.knode.mask = 0; | |
a1b7c5fd | 549 | #endif |
de4784ca JP |
550 | cls_u32.knode.sel = &n->sel; |
551 | cls_u32.knode.exts = &n->exts; | |
201c44bd | 552 | if (n->ht_down) |
b49f9c88 | 553 | cls_u32.knode.link_handle = ht->handle; |
201c44bd | 554 | |
245dc512 JP |
555 | err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw); |
556 | if (err < 0) { | |
557 | u32_remove_hw_knode(tp, n->handle); | |
201c44bd | 558 | return err; |
245dc512 JP |
559 | } else if (err > 0) { |
560 | n->flags |= TCA_CLS_FLAGS_IN_HW; | |
561 | } | |
562 | ||
0f04d057 | 563 | if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW)) |
245dc512 | 564 | return -EINVAL; |
d34e3e18 SS |
565 | |
566 | return 0; | |
a1b7c5fd JF |
567 | } |
568 | ||
a0efb80c | 569 | static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) |
1da177e4 LT |
570 | { |
571 | struct tc_u_knode *n; | |
cc7ec456 | 572 | unsigned int h; |
1da177e4 | 573 | |
cc7ec456 | 574 | for (h = 0; h <= ht->divisor; h++) { |
1ce87720 JF |
575 | while ((n = rtnl_dereference(ht->ht[h])) != NULL) { |
576 | RCU_INIT_POINTER(ht->ht[h], | |
577 | rtnl_dereference(n->next)); | |
a0efb80c | 578 | tcf_unbind_filter(tp, &n->res); |
a1b7c5fd | 579 | u32_remove_hw_knode(tp, n->handle); |
e7614370 | 580 | idr_remove_ext(&ht->handle_idr, n->handle); |
35c55fc1 | 581 | if (tcf_exts_get_net(&n->exts)) |
7afe8e4f | 582 | tcf_queue_work(&n->rwork, u32_delete_key_freepf_work); |
35c55fc1 CW |
583 | else |
584 | u32_destroy_key(n->tp, n, true); | |
1da177e4 LT |
585 | } |
586 | } | |
587 | } | |
588 | ||
589 | static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht) | |
590 | { | |
591 | struct tc_u_common *tp_c = tp->data; | |
1ce87720 JF |
592 | struct tc_u_hnode __rcu **hn; |
593 | struct tc_u_hnode *phn; | |
1da177e4 | 594 | |
547b792c | 595 | WARN_ON(ht->refcnt); |
1da177e4 | 596 | |
a0efb80c | 597 | u32_clear_hnode(tp, ht); |
1da177e4 | 598 | |
1ce87720 JF |
599 | hn = &tp_c->hlist; |
600 | for (phn = rtnl_dereference(*hn); | |
601 | phn; | |
602 | hn = &phn->next, phn = rtnl_dereference(*hn)) { | |
603 | if (phn == ht) { | |
a1b7c5fd | 604 | u32_clear_hw_hnode(tp, ht); |
e7614370 CW |
605 | idr_destroy(&ht->handle_idr); |
606 | idr_remove_ext(&tp_c->handle_idr, ht->handle); | |
1ce87720 JF |
607 | RCU_INIT_POINTER(*hn, ht->next); |
608 | kfree_rcu(ht, rcu); | |
1da177e4 LT |
609 | return 0; |
610 | } | |
611 | } | |
612 | ||
1da177e4 LT |
613 | return -ENOENT; |
614 | } | |
615 | ||
1e052be6 CW |
616 | static bool ht_empty(struct tc_u_hnode *ht) |
617 | { | |
618 | unsigned int h; | |
619 | ||
620 | for (h = 0; h <= ht->divisor; h++) | |
621 | if (rcu_access_pointer(ht->ht[h])) | |
622 | return false; | |
623 | ||
624 | return true; | |
625 | } | |
626 | ||
763dbf63 | 627 | static void u32_destroy(struct tcf_proto *tp) |
1da177e4 LT |
628 | { |
629 | struct tc_u_common *tp_c = tp->data; | |
1ce87720 | 630 | struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); |
1da177e4 | 631 | |
547b792c | 632 | WARN_ON(root_ht == NULL); |
1da177e4 LT |
633 | |
634 | if (root_ht && --root_ht->refcnt == 0) | |
635 | u32_destroy_hnode(tp, root_ht); | |
636 | ||
637 | if (--tp_c->refcnt == 0) { | |
638 | struct tc_u_hnode *ht; | |
1da177e4 | 639 | |
3cd904ec | 640 | hlist_del(&tp_c->hnode); |
1da177e4 | 641 | |
1ce87720 | 642 | while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) { |
360a6fdd | 643 | u32_clear_hnode(tp, ht); |
1ce87720 | 644 | RCU_INIT_POINTER(tp_c->hlist, ht->next); |
360a6fdd PA |
645 | |
646 | /* u32_destroy_key() will later free ht for us, if it's | |
647 | * still referenced by some knode | |
648 | */ | |
649 | if (--ht->refcnt == 0) | |
650 | kfree_rcu(ht, rcu); | |
3ff50b79 | 651 | } |
1da177e4 | 652 | |
e7614370 | 653 | idr_destroy(&tp_c->handle_idr); |
1da177e4 LT |
654 | kfree(tp_c); |
655 | } | |
656 | ||
657 | tp->data = NULL; | |
658 | } | |
659 | ||
8113c095 | 660 | static int u32_delete(struct tcf_proto *tp, void *arg, bool *last) |
1da177e4 | 661 | { |
8113c095 | 662 | struct tc_u_hnode *ht = arg; |
1ce87720 | 663 | struct tc_u_hnode *root_ht = rtnl_dereference(tp->root); |
763dbf63 WC |
664 | struct tc_u_common *tp_c = tp->data; |
665 | int ret = 0; | |
1da177e4 LT |
666 | |
667 | if (ht == NULL) | |
763dbf63 | 668 | goto out; |
1da177e4 | 669 | |
a1b7c5fd JF |
670 | if (TC_U32_KEY(ht->handle)) { |
671 | u32_remove_hw_knode(tp, ht->handle); | |
763dbf63 WC |
672 | ret = u32_delete_key(tp, (struct tc_u_knode *)ht); |
673 | goto out; | |
a1b7c5fd | 674 | } |
1da177e4 | 675 | |
1ce87720 | 676 | if (root_ht == ht) |
1da177e4 LT |
677 | return -EINVAL; |
678 | ||
e56cfad1 JP |
679 | if (ht->refcnt == 1) { |
680 | ht->refcnt--; | |
1da177e4 | 681 | u32_destroy_hnode(tp, ht); |
e56cfad1 JP |
682 | } else { |
683 | return -EBUSY; | |
684 | } | |
1da177e4 | 685 | |
763dbf63 WC |
686 | out: |
687 | *last = true; | |
688 | if (root_ht) { | |
689 | if (root_ht->refcnt > 1) { | |
690 | *last = false; | |
691 | goto ret; | |
692 | } | |
693 | if (root_ht->refcnt == 1) { | |
694 | if (!ht_empty(root_ht)) { | |
695 | *last = false; | |
696 | goto ret; | |
697 | } | |
698 | } | |
699 | } | |
700 | ||
701 | if (tp_c->refcnt > 1) { | |
702 | *last = false; | |
703 | goto ret; | |
704 | } | |
705 | ||
706 | if (tp_c->refcnt == 1) { | |
707 | struct tc_u_hnode *ht; | |
708 | ||
709 | for (ht = rtnl_dereference(tp_c->hlist); | |
710 | ht; | |
711 | ht = rtnl_dereference(ht->next)) | |
712 | if (!ht_empty(ht)) { | |
713 | *last = false; | |
714 | break; | |
715 | } | |
716 | } | |
717 | ||
718 | ret: | |
719 | return ret; | |
1da177e4 LT |
720 | } |
721 | ||
e7614370 | 722 | static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid) |
1da177e4 | 723 | { |
e7614370 CW |
724 | unsigned long idr_index; |
725 | u32 start = htid | 0x800; | |
726 | u32 max = htid | 0xFFF; | |
727 | u32 min = htid; | |
728 | ||
729 | if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index, | |
730 | start, max + 1, GFP_KERNEL)) { | |
731 | if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index, | |
732 | min + 1, max + 1, GFP_KERNEL)) | |
733 | return max; | |
734 | } | |
7801db8a | 735 | |
e7614370 | 736 | return (u32)idr_index; |
1da177e4 LT |
737 | } |
738 | ||
6fa8c014 PM |
739 | static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = { |
740 | [TCA_U32_CLASSID] = { .type = NLA_U32 }, | |
741 | [TCA_U32_HASH] = { .type = NLA_U32 }, | |
742 | [TCA_U32_LINK] = { .type = NLA_U32 }, | |
743 | [TCA_U32_DIVISOR] = { .type = NLA_U32 }, | |
744 | [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) }, | |
745 | [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ }, | |
746 | [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) }, | |
9e8ce79c | 747 | [TCA_U32_FLAGS] = { .type = NLA_U32 }, |
6fa8c014 PM |
748 | }; |
749 | ||
c1b52739 BL |
750 | static int u32_set_parms(struct net *net, struct tcf_proto *tp, |
751 | unsigned long base, struct tc_u_hnode *ht, | |
add93b61 | 752 | struct tc_u_knode *n, struct nlattr **tb, |
2f7ef2f8 | 753 | struct nlattr *est, bool ovr) |
1da177e4 | 754 | { |
b9a24bb7 | 755 | int err; |
1da177e4 | 756 | |
705c7091 | 757 | err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr); |
1da177e4 LT |
758 | if (err < 0) |
759 | return err; | |
760 | ||
add93b61 | 761 | if (tb[TCA_U32_LINK]) { |
1587bac4 | 762 | u32 handle = nla_get_u32(tb[TCA_U32_LINK]); |
47a1a1d4 | 763 | struct tc_u_hnode *ht_down = NULL, *ht_old; |
1da177e4 LT |
764 | |
765 | if (TC_U32_KEY(handle)) | |
705c7091 | 766 | return -EINVAL; |
1da177e4 LT |
767 | |
768 | if (handle) { | |
769 | ht_down = u32_lookup_ht(ht->tp_c, handle); | |
770 | ||
771 | if (ht_down == NULL) | |
705c7091 | 772 | return -EINVAL; |
1da177e4 LT |
773 | ht_down->refcnt++; |
774 | } | |
775 | ||
1ce87720 JF |
776 | ht_old = rtnl_dereference(n->ht_down); |
777 | rcu_assign_pointer(n->ht_down, ht_down); | |
1da177e4 | 778 | |
47a1a1d4 PM |
779 | if (ht_old) |
780 | ht_old->refcnt--; | |
1da177e4 | 781 | } |
add93b61 | 782 | if (tb[TCA_U32_CLASSID]) { |
1587bac4 | 783 | n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]); |
1da177e4 LT |
784 | tcf_bind_filter(tp, &n->res, base); |
785 | } | |
786 | ||
787 | #ifdef CONFIG_NET_CLS_IND | |
add93b61 | 788 | if (tb[TCA_U32_INDEV]) { |
2519a602 WC |
789 | int ret; |
790 | ret = tcf_change_indev(net, tb[TCA_U32_INDEV]); | |
791 | if (ret < 0) | |
705c7091 | 792 | return -EINVAL; |
2519a602 | 793 | n->ifindex = ret; |
1da177e4 LT |
794 | } |
795 | #endif | |
1da177e4 | 796 | return 0; |
1da177e4 LT |
797 | } |
798 | ||
5a7a5555 | 799 | static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c, |
de5df632 JF |
800 | struct tc_u_knode *n) |
801 | { | |
802 | struct tc_u_knode __rcu **ins; | |
803 | struct tc_u_knode *pins; | |
804 | struct tc_u_hnode *ht; | |
805 | ||
806 | if (TC_U32_HTID(n->handle) == TC_U32_ROOT) | |
807 | ht = rtnl_dereference(tp->root); | |
808 | else | |
809 | ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle)); | |
810 | ||
811 | ins = &ht->ht[TC_U32_HASH(n->handle)]; | |
812 | ||
813 | /* The node must always exist for it to be replaced if this is not the | |
814 | * case then something went very wrong elsewhere. | |
815 | */ | |
816 | for (pins = rtnl_dereference(*ins); ; | |
817 | ins = &pins->next, pins = rtnl_dereference(*ins)) | |
818 | if (pins->handle == n->handle) | |
819 | break; | |
820 | ||
e7614370 | 821 | idr_replace_ext(&ht->handle_idr, n, n->handle); |
de5df632 JF |
822 | RCU_INIT_POINTER(n->next, pins->next); |
823 | rcu_assign_pointer(*ins, n); | |
824 | } | |
825 | ||
fb6ad63b | 826 | static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp, |
de5df632 JF |
827 | struct tc_u_knode *n) |
828 | { | |
b49f9c88 | 829 | struct tc_u_hnode *ht = rtnl_dereference(n->ht_down); |
de5df632 | 830 | struct tc_u32_sel *s = &n->sel; |
b49f9c88 | 831 | struct tc_u_knode *new; |
de5df632 JF |
832 | |
833 | new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), | |
834 | GFP_KERNEL); | |
835 | ||
836 | if (!new) | |
837 | return NULL; | |
838 | ||
839 | RCU_INIT_POINTER(new->next, n->next); | |
840 | new->handle = n->handle; | |
841 | RCU_INIT_POINTER(new->ht_up, n->ht_up); | |
842 | ||
843 | #ifdef CONFIG_NET_CLS_IND | |
844 | new->ifindex = n->ifindex; | |
845 | #endif | |
846 | new->fshift = n->fshift; | |
847 | new->res = n->res; | |
9e8ce79c | 848 | new->flags = n->flags; |
b49f9c88 | 849 | RCU_INIT_POINTER(new->ht_down, ht); |
de5df632 JF |
850 | |
851 | /* bump reference count as long as we hold pointer to structure */ | |
b49f9c88 PA |
852 | if (ht) |
853 | ht->refcnt++; | |
de5df632 JF |
854 | |
855 | #ifdef CONFIG_CLS_U32_PERF | |
856 | /* Statistics may be incremented by readers during update | |
857 | * so we must keep them in tact. When the node is later destroyed | |
858 | * a special destroy call must be made to not free the pf memory. | |
859 | */ | |
860 | new->pf = n->pf; | |
861 | #endif | |
862 | ||
863 | #ifdef CONFIG_CLS_U32_MARK | |
864 | new->val = n->val; | |
865 | new->mask = n->mask; | |
866 | /* Similarly success statistics must be moved as pointers */ | |
867 | new->pcpu_success = n->pcpu_success; | |
868 | #endif | |
869 | new->tp = tp; | |
870 | memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); | |
871 | ||
fb6ad63b | 872 | if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) { |
b9a24bb7 WC |
873 | kfree(new); |
874 | return NULL; | |
875 | } | |
de5df632 JF |
876 | |
877 | return new; | |
878 | } | |
879 | ||
c1b52739 | 880 | static int u32_change(struct net *net, struct sk_buff *in_skb, |
af4c6641 | 881 | struct tcf_proto *tp, unsigned long base, u32 handle, |
8113c095 | 882 | struct nlattr **tca, void **arg, bool ovr) |
1da177e4 LT |
883 | { |
884 | struct tc_u_common *tp_c = tp->data; | |
885 | struct tc_u_hnode *ht; | |
886 | struct tc_u_knode *n; | |
887 | struct tc_u32_sel *s; | |
add93b61 PM |
888 | struct nlattr *opt = tca[TCA_OPTIONS]; |
889 | struct nlattr *tb[TCA_U32_MAX + 1]; | |
9e8ce79c | 890 | u32 htid, flags = 0; |
1da177e4 | 891 | int err; |
459d5f62 JF |
892 | #ifdef CONFIG_CLS_U32_PERF |
893 | size_t size; | |
894 | #endif | |
1da177e4 LT |
895 | |
896 | if (opt == NULL) | |
897 | return handle ? -EINVAL : 0; | |
898 | ||
fceb6435 | 899 | err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, NULL); |
cee63723 PM |
900 | if (err < 0) |
901 | return err; | |
1da177e4 | 902 | |
d34e3e18 | 903 | if (tb[TCA_U32_FLAGS]) { |
9e8ce79c | 904 | flags = nla_get_u32(tb[TCA_U32_FLAGS]); |
d34e3e18 | 905 | if (!tc_flags_valid(flags)) |
1a0f7d29 | 906 | return -EINVAL; |
d34e3e18 | 907 | } |
9e8ce79c | 908 | |
8113c095 | 909 | n = *arg; |
cc7ec456 | 910 | if (n) { |
de5df632 JF |
911 | struct tc_u_knode *new; |
912 | ||
1da177e4 LT |
913 | if (TC_U32_KEY(n->handle) == 0) |
914 | return -EINVAL; | |
915 | ||
2d965923 IV |
916 | if ((n->flags ^ flags) & |
917 | ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) | |
9e8ce79c JF |
918 | return -EINVAL; |
919 | ||
fb6ad63b | 920 | new = u32_init_knode(net, tp, n); |
de5df632 JF |
921 | if (!new) |
922 | return -ENOMEM; | |
923 | ||
924 | err = u32_set_parms(net, tp, base, | |
925 | rtnl_dereference(n->ht_up), new, tb, | |
926 | tca[TCA_RATE], ovr); | |
927 | ||
928 | if (err) { | |
929 | u32_destroy_key(tp, new, false); | |
930 | return err; | |
931 | } | |
932 | ||
d34e3e18 SS |
933 | err = u32_replace_hw_knode(tp, new, flags); |
934 | if (err) { | |
935 | u32_destroy_key(tp, new, false); | |
936 | return err; | |
937 | } | |
938 | ||
24d3dc6d OG |
939 | if (!tc_in_hw(new->flags)) |
940 | new->flags |= TCA_CLS_FLAGS_NOT_IN_HW; | |
941 | ||
de5df632 | 942 | u32_replace_knode(tp, tp_c, new); |
a0efb80c | 943 | tcf_unbind_filter(tp, &n->res); |
35c55fc1 | 944 | tcf_exts_get_net(&n->exts); |
7afe8e4f | 945 | tcf_queue_work(&n->rwork, u32_delete_key_work); |
de5df632 | 946 | return 0; |
1da177e4 LT |
947 | } |
948 | ||
add93b61 | 949 | if (tb[TCA_U32_DIVISOR]) { |
cc7ec456 | 950 | unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]); |
1da177e4 LT |
951 | |
952 | if (--divisor > 0x100) | |
953 | return -EINVAL; | |
954 | if (TC_U32_KEY(handle)) | |
955 | return -EINVAL; | |
cc7ec456 | 956 | ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL); |
1da177e4 LT |
957 | if (ht == NULL) |
958 | return -ENOBUFS; | |
e7614370 CW |
959 | if (handle == 0) { |
960 | handle = gen_new_htid(tp->data, ht); | |
961 | if (handle == 0) { | |
962 | kfree(ht); | |
963 | return -ENOMEM; | |
964 | } | |
965 | } else { | |
966 | err = idr_alloc_ext(&tp_c->handle_idr, ht, NULL, | |
967 | handle, handle + 1, GFP_KERNEL); | |
968 | if (err) { | |
969 | kfree(ht); | |
970 | return err; | |
971 | } | |
972 | } | |
1da177e4 | 973 | ht->tp_c = tp_c; |
e56cfad1 | 974 | ht->refcnt = 1; |
1da177e4 LT |
975 | ht->divisor = divisor; |
976 | ht->handle = handle; | |
977 | ht->prio = tp->prio; | |
e7614370 | 978 | idr_init(&ht->handle_idr); |
6eef3801 JK |
979 | |
980 | err = u32_replace_hw_hnode(tp, ht, flags); | |
981 | if (err) { | |
e7614370 | 982 | idr_remove_ext(&tp_c->handle_idr, handle); |
6eef3801 JK |
983 | kfree(ht); |
984 | return err; | |
985 | } | |
986 | ||
1ce87720 JF |
987 | RCU_INIT_POINTER(ht->next, tp_c->hlist); |
988 | rcu_assign_pointer(tp_c->hlist, ht); | |
8113c095 | 989 | *arg = ht; |
a1b7c5fd | 990 | |
1da177e4 LT |
991 | return 0; |
992 | } | |
993 | ||
add93b61 | 994 | if (tb[TCA_U32_HASH]) { |
1587bac4 | 995 | htid = nla_get_u32(tb[TCA_U32_HASH]); |
1da177e4 | 996 | if (TC_U32_HTID(htid) == TC_U32_ROOT) { |
1ce87720 | 997 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
998 | htid = ht->handle; |
999 | } else { | |
1000 | ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid)); | |
1001 | if (ht == NULL) | |
1002 | return -EINVAL; | |
1003 | } | |
1004 | } else { | |
1ce87720 | 1005 | ht = rtnl_dereference(tp->root); |
1da177e4 LT |
1006 | htid = ht->handle; |
1007 | } | |
1008 | ||
1009 | if (ht->divisor < TC_U32_HASH(htid)) | |
1010 | return -EINVAL; | |
1011 | ||
1012 | if (handle) { | |
1013 | if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid)) | |
1014 | return -EINVAL; | |
1015 | handle = htid | TC_U32_NODE(handle); | |
e7614370 CW |
1016 | err = idr_alloc_ext(&ht->handle_idr, NULL, NULL, |
1017 | handle, handle + 1, | |
1018 | GFP_KERNEL); | |
1019 | if (err) | |
1020 | return err; | |
1da177e4 LT |
1021 | } else |
1022 | handle = gen_new_kid(ht, htid); | |
1023 | ||
e7614370 CW |
1024 | if (tb[TCA_U32_SEL] == NULL) { |
1025 | err = -EINVAL; | |
1026 | goto erridr; | |
1027 | } | |
1da177e4 | 1028 | |
add93b61 | 1029 | s = nla_data(tb[TCA_U32_SEL]); |
1da177e4 | 1030 | |
0da974f4 | 1031 | n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL); |
e7614370 CW |
1032 | if (n == NULL) { |
1033 | err = -ENOBUFS; | |
1034 | goto erridr; | |
1035 | } | |
1da177e4 | 1036 | |
1da177e4 | 1037 | #ifdef CONFIG_CLS_U32_PERF |
459d5f62 JF |
1038 | size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64); |
1039 | n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt)); | |
1040 | if (!n->pf) { | |
e7614370 CW |
1041 | err = -ENOBUFS; |
1042 | goto errfree; | |
1da177e4 | 1043 | } |
1da177e4 LT |
1044 | #endif |
1045 | ||
1046 | memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key)); | |
a96366bf | 1047 | RCU_INIT_POINTER(n->ht_up, ht); |
1da177e4 | 1048 | n->handle = handle; |
b2268016 | 1049 | n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0; |
9e8ce79c | 1050 | n->flags = flags; |
1ce87720 | 1051 | n->tp = tp; |
1da177e4 | 1052 | |
fb6ad63b | 1053 | err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE); |
b9a24bb7 WC |
1054 | if (err < 0) |
1055 | goto errout; | |
1056 | ||
1da177e4 | 1057 | #ifdef CONFIG_CLS_U32_MARK |
459d5f62 | 1058 | n->pcpu_success = alloc_percpu(u32); |
a1ddcfee JF |
1059 | if (!n->pcpu_success) { |
1060 | err = -ENOMEM; | |
1061 | goto errout; | |
1062 | } | |
459d5f62 | 1063 | |
add93b61 | 1064 | if (tb[TCA_U32_MARK]) { |
1da177e4 LT |
1065 | struct tc_u32_mark *mark; |
1066 | ||
add93b61 | 1067 | mark = nla_data(tb[TCA_U32_MARK]); |
459d5f62 JF |
1068 | n->val = mark->val; |
1069 | n->mask = mark->mask; | |
1da177e4 LT |
1070 | } |
1071 | #endif | |
1072 | ||
2f7ef2f8 | 1073 | err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr); |
1da177e4 | 1074 | if (err == 0) { |
1ce87720 JF |
1075 | struct tc_u_knode __rcu **ins; |
1076 | struct tc_u_knode *pins; | |
1077 | ||
d34e3e18 SS |
1078 | err = u32_replace_hw_knode(tp, n, flags); |
1079 | if (err) | |
1080 | goto errhw; | |
1081 | ||
24d3dc6d OG |
1082 | if (!tc_in_hw(n->flags)) |
1083 | n->flags |= TCA_CLS_FLAGS_NOT_IN_HW; | |
1084 | ||
1ce87720 JF |
1085 | ins = &ht->ht[TC_U32_HASH(handle)]; |
1086 | for (pins = rtnl_dereference(*ins); pins; | |
1087 | ins = &pins->next, pins = rtnl_dereference(*ins)) | |
1088 | if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle)) | |
1da177e4 LT |
1089 | break; |
1090 | ||
1ce87720 JF |
1091 | RCU_INIT_POINTER(n->next, pins); |
1092 | rcu_assign_pointer(*ins, n); | |
8113c095 | 1093 | *arg = n; |
1da177e4 LT |
1094 | return 0; |
1095 | } | |
a1ddcfee | 1096 | |
d34e3e18 | 1097 | errhw: |
a1ddcfee JF |
1098 | #ifdef CONFIG_CLS_U32_MARK |
1099 | free_percpu(n->pcpu_success); | |
1100 | #endif | |
1101 | ||
b9a24bb7 WC |
1102 | errout: |
1103 | tcf_exts_destroy(&n->exts); | |
1da177e4 | 1104 | #ifdef CONFIG_CLS_U32_PERF |
e7614370 | 1105 | errfree: |
1ce87720 | 1106 | free_percpu(n->pf); |
1da177e4 LT |
1107 | #endif |
1108 | kfree(n); | |
e7614370 CW |
1109 | erridr: |
1110 | idr_remove_ext(&ht->handle_idr, handle); | |
1da177e4 LT |
1111 | return err; |
1112 | } | |
1113 | ||
1114 | static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg) | |
1115 | { | |
1116 | struct tc_u_common *tp_c = tp->data; | |
1117 | struct tc_u_hnode *ht; | |
1118 | struct tc_u_knode *n; | |
cc7ec456 | 1119 | unsigned int h; |
1da177e4 LT |
1120 | |
1121 | if (arg->stop) | |
1122 | return; | |
1123 | ||
1ce87720 JF |
1124 | for (ht = rtnl_dereference(tp_c->hlist); |
1125 | ht; | |
1126 | ht = rtnl_dereference(ht->next)) { | |
1da177e4 LT |
1127 | if (ht->prio != tp->prio) |
1128 | continue; | |
1129 | if (arg->count >= arg->skip) { | |
8113c095 | 1130 | if (arg->fn(tp, ht, arg) < 0) { |
1da177e4 LT |
1131 | arg->stop = 1; |
1132 | return; | |
1133 | } | |
1134 | } | |
1135 | arg->count++; | |
1136 | for (h = 0; h <= ht->divisor; h++) { | |
1ce87720 JF |
1137 | for (n = rtnl_dereference(ht->ht[h]); |
1138 | n; | |
1139 | n = rtnl_dereference(n->next)) { | |
1da177e4 LT |
1140 | if (arg->count < arg->skip) { |
1141 | arg->count++; | |
1142 | continue; | |
1143 | } | |
8113c095 | 1144 | if (arg->fn(tp, n, arg) < 0) { |
1da177e4 LT |
1145 | arg->stop = 1; |
1146 | return; | |
1147 | } | |
1148 | arg->count++; | |
1149 | } | |
1150 | } | |
1151 | } | |
1152 | } | |
1153 | ||
07d79fc7 CW |
1154 | static void u32_bind_class(void *fh, u32 classid, unsigned long cl) |
1155 | { | |
1156 | struct tc_u_knode *n = fh; | |
1157 | ||
1158 | if (n && n->res.classid == classid) | |
1159 | n->res.class = cl; | |
1160 | } | |
1161 | ||
8113c095 | 1162 | static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh, |
5a7a5555 | 1163 | struct sk_buff *skb, struct tcmsg *t) |
1da177e4 | 1164 | { |
8113c095 | 1165 | struct tc_u_knode *n = fh; |
1ce87720 | 1166 | struct tc_u_hnode *ht_up, *ht_down; |
4b3550ef | 1167 | struct nlattr *nest; |
1da177e4 LT |
1168 | |
1169 | if (n == NULL) | |
1170 | return skb->len; | |
1171 | ||
1172 | t->tcm_handle = n->handle; | |
1173 | ||
4b3550ef PM |
1174 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1175 | if (nest == NULL) | |
1176 | goto nla_put_failure; | |
1da177e4 LT |
1177 | |
1178 | if (TC_U32_KEY(n->handle) == 0) { | |
8113c095 | 1179 | struct tc_u_hnode *ht = fh; |
cc7ec456 ED |
1180 | u32 divisor = ht->divisor + 1; |
1181 | ||
1b34ec43 DM |
1182 | if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor)) |
1183 | goto nla_put_failure; | |
1da177e4 | 1184 | } else { |
459d5f62 JF |
1185 | #ifdef CONFIG_CLS_U32_PERF |
1186 | struct tc_u32_pcnt *gpf; | |
459d5f62 | 1187 | int cpu; |
80aab73d | 1188 | #endif |
459d5f62 | 1189 | |
1b34ec43 DM |
1190 | if (nla_put(skb, TCA_U32_SEL, |
1191 | sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key), | |
1192 | &n->sel)) | |
1193 | goto nla_put_failure; | |
1ce87720 JF |
1194 | |
1195 | ht_up = rtnl_dereference(n->ht_up); | |
1196 | if (ht_up) { | |
1da177e4 | 1197 | u32 htid = n->handle & 0xFFFFF000; |
1b34ec43 DM |
1198 | if (nla_put_u32(skb, TCA_U32_HASH, htid)) |
1199 | goto nla_put_failure; | |
1da177e4 | 1200 | } |
1b34ec43 DM |
1201 | if (n->res.classid && |
1202 | nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid)) | |
1203 | goto nla_put_failure; | |
1ce87720 JF |
1204 | |
1205 | ht_down = rtnl_dereference(n->ht_down); | |
1206 | if (ht_down && | |
1207 | nla_put_u32(skb, TCA_U32_LINK, ht_down->handle)) | |
1b34ec43 | 1208 | goto nla_put_failure; |
1da177e4 | 1209 | |
9e8ce79c JF |
1210 | if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags)) |
1211 | goto nla_put_failure; | |
1212 | ||
1da177e4 | 1213 | #ifdef CONFIG_CLS_U32_MARK |
459d5f62 JF |
1214 | if ((n->val || n->mask)) { |
1215 | struct tc_u32_mark mark = {.val = n->val, | |
1216 | .mask = n->mask, | |
1217 | .success = 0}; | |
80aab73d | 1218 | int cpum; |
459d5f62 | 1219 | |
80aab73d JF |
1220 | for_each_possible_cpu(cpum) { |
1221 | __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum); | |
459d5f62 JF |
1222 | |
1223 | mark.success += cnt; | |
1224 | } | |
1225 | ||
1226 | if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark)) | |
1227 | goto nla_put_failure; | |
1228 | } | |
1da177e4 LT |
1229 | #endif |
1230 | ||
5da57f42 | 1231 | if (tcf_exts_dump(skb, &n->exts) < 0) |
add93b61 | 1232 | goto nla_put_failure; |
1da177e4 LT |
1233 | |
1234 | #ifdef CONFIG_NET_CLS_IND | |
2519a602 WC |
1235 | if (n->ifindex) { |
1236 | struct net_device *dev; | |
1237 | dev = __dev_get_by_index(net, n->ifindex); | |
1238 | if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name)) | |
1239 | goto nla_put_failure; | |
1240 | } | |
1da177e4 LT |
1241 | #endif |
1242 | #ifdef CONFIG_CLS_U32_PERF | |
459d5f62 JF |
1243 | gpf = kzalloc(sizeof(struct tc_u32_pcnt) + |
1244 | n->sel.nkeys * sizeof(u64), | |
1245 | GFP_KERNEL); | |
1246 | if (!gpf) | |
1247 | goto nla_put_failure; | |
1248 | ||
1249 | for_each_possible_cpu(cpu) { | |
1250 | int i; | |
1251 | struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu); | |
1252 | ||
1253 | gpf->rcnt += pf->rcnt; | |
1254 | gpf->rhit += pf->rhit; | |
1255 | for (i = 0; i < n->sel.nkeys; i++) | |
1256 | gpf->kcnts[i] += pf->kcnts[i]; | |
1257 | } | |
1258 | ||
9854518e ND |
1259 | if (nla_put_64bit(skb, TCA_U32_PCNT, |
1260 | sizeof(struct tc_u32_pcnt) + | |
1261 | n->sel.nkeys * sizeof(u64), | |
1262 | gpf, TCA_U32_PAD)) { | |
459d5f62 | 1263 | kfree(gpf); |
1b34ec43 | 1264 | goto nla_put_failure; |
459d5f62 JF |
1265 | } |
1266 | kfree(gpf); | |
1da177e4 LT |
1267 | #endif |
1268 | } | |
1269 | ||
4b3550ef PM |
1270 | nla_nest_end(skb, nest); |
1271 | ||
1da177e4 | 1272 | if (TC_U32_KEY(n->handle)) |
5da57f42 | 1273 | if (tcf_exts_dump_stats(skb, &n->exts) < 0) |
add93b61 | 1274 | goto nla_put_failure; |
1da177e4 LT |
1275 | return skb->len; |
1276 | ||
add93b61 | 1277 | nla_put_failure: |
4b3550ef | 1278 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1279 | return -1; |
1280 | } | |
1281 | ||
2eb9d75c | 1282 | static struct tcf_proto_ops cls_u32_ops __read_mostly = { |
1da177e4 LT |
1283 | .kind = "u32", |
1284 | .classify = u32_classify, | |
1285 | .init = u32_init, | |
1286 | .destroy = u32_destroy, | |
1287 | .get = u32_get, | |
1da177e4 LT |
1288 | .change = u32_change, |
1289 | .delete = u32_delete, | |
1290 | .walk = u32_walk, | |
1291 | .dump = u32_dump, | |
07d79fc7 | 1292 | .bind_class = u32_bind_class, |
1da177e4 LT |
1293 | .owner = THIS_MODULE, |
1294 | }; | |
1295 | ||
1296 | static int __init init_u32(void) | |
1297 | { | |
3cd904ec WC |
1298 | int i, ret; |
1299 | ||
6ff9c364 | 1300 | pr_info("u32 classifier\n"); |
1da177e4 | 1301 | #ifdef CONFIG_CLS_U32_PERF |
6ff9c364 | 1302 | pr_info(" Performance counters on\n"); |
1da177e4 | 1303 | #endif |
1da177e4 | 1304 | #ifdef CONFIG_NET_CLS_IND |
6ff9c364 | 1305 | pr_info(" input device check on\n"); |
1da177e4 LT |
1306 | #endif |
1307 | #ifdef CONFIG_NET_CLS_ACT | |
6ff9c364 | 1308 | pr_info(" Actions configured\n"); |
1da177e4 | 1309 | #endif |
3cd904ec WC |
1310 | tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE, |
1311 | sizeof(struct hlist_head), | |
1312 | GFP_KERNEL); | |
1313 | if (!tc_u_common_hash) | |
1314 | return -ENOMEM; | |
1315 | ||
1316 | for (i = 0; i < U32_HASH_SIZE; i++) | |
1317 | INIT_HLIST_HEAD(&tc_u_common_hash[i]); | |
1318 | ||
1319 | ret = register_tcf_proto_ops(&cls_u32_ops); | |
1320 | if (ret) | |
1321 | kvfree(tc_u_common_hash); | |
1322 | return ret; | |
1da177e4 LT |
1323 | } |
1324 | ||
10297b99 | 1325 | static void __exit exit_u32(void) |
1da177e4 LT |
1326 | { |
1327 | unregister_tcf_proto_ops(&cls_u32_ops); | |
3cd904ec | 1328 | kvfree(tc_u_common_hash); |
1da177e4 LT |
1329 | } |
1330 | ||
1331 | module_init(init_u32) | |
1332 | module_exit(exit_u32) | |
1333 | MODULE_LICENSE("GPL"); |