Commit | Line | Data |
---|---|---|
77ab9cff MJ |
1 | /* Expectation handling for nf_conntrack. */ |
2 | ||
3 | /* (C) 1999-2001 Paul `Rusty' Russell | |
4 | * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org> | |
5 | * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | */ | |
11 | ||
12 | #include <linux/types.h> | |
13 | #include <linux/netfilter.h> | |
14 | #include <linux/skbuff.h> | |
15 | #include <linux/proc_fs.h> | |
16 | #include <linux/seq_file.h> | |
17 | #include <linux/stddef.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/err.h> | |
20 | #include <linux/percpu.h> | |
21 | #include <linux/kernel.h> | |
22 | ||
23 | #include <net/netfilter/nf_conntrack.h> | |
24 | #include <net/netfilter/nf_conntrack_core.h> | |
25 | #include <net/netfilter/nf_conntrack_expect.h> | |
26 | #include <net/netfilter/nf_conntrack_helper.h> | |
27 | #include <net/netfilter/nf_conntrack_tuple.h> | |
28 | ||
29 | LIST_HEAD(nf_conntrack_expect_list); | |
30 | kmem_cache_t *nf_conntrack_expect_cachep __read_mostly; | |
31 | DECLARE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); | |
32 | static unsigned int nf_conntrack_expect_next_id; | |
33 | ||
34 | /* nf_conntrack_expect helper functions */ | |
35 | void nf_ct_unlink_expect(struct nf_conntrack_expect *exp) | |
36 | { | |
37 | struct nf_conn_help *master_help = nfct_help(exp->master); | |
38 | ||
39 | NF_CT_ASSERT(master_help); | |
40 | NF_CT_ASSERT(!timer_pending(&exp->timeout)); | |
41 | ||
42 | list_del(&exp->list); | |
43 | NF_CT_STAT_INC(expect_delete); | |
44 | master_help->expecting--; | |
45 | nf_conntrack_expect_put(exp); | |
46 | } | |
47 | ||
48 | static void expectation_timed_out(unsigned long ul_expect) | |
49 | { | |
50 | struct nf_conntrack_expect *exp = (void *)ul_expect; | |
51 | ||
52 | write_lock_bh(&nf_conntrack_lock); | |
53 | nf_ct_unlink_expect(exp); | |
54 | write_unlock_bh(&nf_conntrack_lock); | |
55 | nf_conntrack_expect_put(exp); | |
56 | } | |
57 | ||
58 | struct nf_conntrack_expect * | |
59 | __nf_conntrack_expect_find(const struct nf_conntrack_tuple *tuple) | |
60 | { | |
61 | struct nf_conntrack_expect *i; | |
62 | ||
63 | list_for_each_entry(i, &nf_conntrack_expect_list, list) { | |
64 | if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask)) | |
65 | return i; | |
66 | } | |
67 | return NULL; | |
68 | } | |
69 | ||
70 | /* Just find a expectation corresponding to a tuple. */ | |
71 | struct nf_conntrack_expect * | |
72 | nf_conntrack_expect_find(const struct nf_conntrack_tuple *tuple) | |
73 | { | |
74 | struct nf_conntrack_expect *i; | |
75 | ||
76 | read_lock_bh(&nf_conntrack_lock); | |
77 | i = __nf_conntrack_expect_find(tuple); | |
78 | if (i) | |
79 | atomic_inc(&i->use); | |
80 | read_unlock_bh(&nf_conntrack_lock); | |
81 | ||
82 | return i; | |
83 | } | |
84 | ||
85 | /* If an expectation for this connection is found, it gets delete from | |
86 | * global list then returned. */ | |
87 | struct nf_conntrack_expect * | |
88 | find_expectation(const struct nf_conntrack_tuple *tuple) | |
89 | { | |
90 | struct nf_conntrack_expect *i; | |
91 | ||
92 | list_for_each_entry(i, &nf_conntrack_expect_list, list) { | |
93 | /* If master is not in hash table yet (ie. packet hasn't left | |
94 | this machine yet), how can other end know about expected? | |
95 | Hence these are not the droids you are looking for (if | |
96 | master ct never got confirmed, we'd hold a reference to it | |
97 | and weird things would happen to future packets). */ | |
98 | if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) | |
99 | && nf_ct_is_confirmed(i->master)) { | |
100 | if (i->flags & NF_CT_EXPECT_PERMANENT) { | |
101 | atomic_inc(&i->use); | |
102 | return i; | |
103 | } else if (del_timer(&i->timeout)) { | |
104 | nf_ct_unlink_expect(i); | |
105 | return i; | |
106 | } | |
107 | } | |
108 | } | |
109 | return NULL; | |
110 | } | |
111 | ||
112 | /* delete all expectations for this conntrack */ | |
113 | void nf_ct_remove_expectations(struct nf_conn *ct) | |
114 | { | |
115 | struct nf_conntrack_expect *i, *tmp; | |
116 | struct nf_conn_help *help = nfct_help(ct); | |
117 | ||
118 | /* Optimization: most connection never expect any others. */ | |
119 | if (!help || help->expecting == 0) | |
120 | return; | |
121 | ||
122 | list_for_each_entry_safe(i, tmp, &nf_conntrack_expect_list, list) { | |
123 | if (i->master == ct && del_timer(&i->timeout)) { | |
124 | nf_ct_unlink_expect(i); | |
125 | nf_conntrack_expect_put(i); | |
126 | } | |
127 | } | |
128 | } | |
129 | ||
130 | /* Would two expected things clash? */ | |
131 | static inline int expect_clash(const struct nf_conntrack_expect *a, | |
132 | const struct nf_conntrack_expect *b) | |
133 | { | |
134 | /* Part covered by intersection of masks must be unequal, | |
135 | otherwise they clash */ | |
136 | struct nf_conntrack_tuple intersect_mask; | |
137 | int count; | |
138 | ||
139 | intersect_mask.src.l3num = a->mask.src.l3num & b->mask.src.l3num; | |
140 | intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all; | |
141 | intersect_mask.dst.u.all = a->mask.dst.u.all & b->mask.dst.u.all; | |
142 | intersect_mask.dst.protonum = a->mask.dst.protonum | |
143 | & b->mask.dst.protonum; | |
144 | ||
145 | for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){ | |
146 | intersect_mask.src.u3.all[count] = | |
147 | a->mask.src.u3.all[count] & b->mask.src.u3.all[count]; | |
148 | } | |
149 | ||
150 | for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){ | |
151 | intersect_mask.dst.u3.all[count] = | |
152 | a->mask.dst.u3.all[count] & b->mask.dst.u3.all[count]; | |
153 | } | |
154 | ||
155 | return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask); | |
156 | } | |
157 | ||
158 | static inline int expect_matches(const struct nf_conntrack_expect *a, | |
159 | const struct nf_conntrack_expect *b) | |
160 | { | |
161 | return a->master == b->master | |
162 | && nf_ct_tuple_equal(&a->tuple, &b->tuple) | |
163 | && nf_ct_tuple_equal(&a->mask, &b->mask); | |
164 | } | |
165 | ||
166 | /* Generally a bad idea to call this: could have matched already. */ | |
167 | void nf_conntrack_unexpect_related(struct nf_conntrack_expect *exp) | |
168 | { | |
169 | struct nf_conntrack_expect *i; | |
170 | ||
171 | write_lock_bh(&nf_conntrack_lock); | |
172 | /* choose the the oldest expectation to evict */ | |
173 | list_for_each_entry_reverse(i, &nf_conntrack_expect_list, list) { | |
174 | if (expect_matches(i, exp) && del_timer(&i->timeout)) { | |
175 | nf_ct_unlink_expect(i); | |
176 | write_unlock_bh(&nf_conntrack_lock); | |
177 | nf_conntrack_expect_put(i); | |
178 | return; | |
179 | } | |
180 | } | |
181 | write_unlock_bh(&nf_conntrack_lock); | |
182 | } | |
183 | ||
184 | /* We don't increase the master conntrack refcount for non-fulfilled | |
185 | * conntracks. During the conntrack destruction, the expectations are | |
186 | * always killed before the conntrack itself */ | |
187 | struct nf_conntrack_expect *nf_conntrack_expect_alloc(struct nf_conn *me) | |
188 | { | |
189 | struct nf_conntrack_expect *new; | |
190 | ||
191 | new = kmem_cache_alloc(nf_conntrack_expect_cachep, GFP_ATOMIC); | |
192 | if (!new) | |
193 | return NULL; | |
194 | ||
195 | new->master = me; | |
196 | atomic_set(&new->use, 1); | |
197 | return new; | |
198 | } | |
199 | ||
200 | void nf_conntrack_expect_put(struct nf_conntrack_expect *exp) | |
201 | { | |
202 | if (atomic_dec_and_test(&exp->use)) | |
203 | kmem_cache_free(nf_conntrack_expect_cachep, exp); | |
204 | } | |
205 | ||
206 | static void nf_conntrack_expect_insert(struct nf_conntrack_expect *exp) | |
207 | { | |
208 | struct nf_conn_help *master_help = nfct_help(exp->master); | |
209 | ||
210 | atomic_inc(&exp->use); | |
211 | master_help->expecting++; | |
212 | list_add(&exp->list, &nf_conntrack_expect_list); | |
213 | ||
214 | init_timer(&exp->timeout); | |
215 | exp->timeout.data = (unsigned long)exp; | |
216 | exp->timeout.function = expectation_timed_out; | |
217 | exp->timeout.expires = jiffies + master_help->helper->timeout * HZ; | |
218 | add_timer(&exp->timeout); | |
219 | ||
220 | exp->id = ++nf_conntrack_expect_next_id; | |
221 | atomic_inc(&exp->use); | |
222 | NF_CT_STAT_INC(expect_create); | |
223 | } | |
224 | ||
225 | /* Race with expectations being used means we could have none to find; OK. */ | |
226 | static void evict_oldest_expect(struct nf_conn *master) | |
227 | { | |
228 | struct nf_conntrack_expect *i; | |
229 | ||
230 | list_for_each_entry_reverse(i, &nf_conntrack_expect_list, list) { | |
231 | if (i->master == master) { | |
232 | if (del_timer(&i->timeout)) { | |
233 | nf_ct_unlink_expect(i); | |
234 | nf_conntrack_expect_put(i); | |
235 | } | |
236 | break; | |
237 | } | |
238 | } | |
239 | } | |
240 | ||
241 | static inline int refresh_timer(struct nf_conntrack_expect *i) | |
242 | { | |
243 | struct nf_conn_help *master_help = nfct_help(i->master); | |
244 | ||
245 | if (!del_timer(&i->timeout)) | |
246 | return 0; | |
247 | ||
248 | i->timeout.expires = jiffies + master_help->helper->timeout*HZ; | |
249 | add_timer(&i->timeout); | |
250 | return 1; | |
251 | } | |
252 | ||
253 | int nf_conntrack_expect_related(struct nf_conntrack_expect *expect) | |
254 | { | |
255 | struct nf_conntrack_expect *i; | |
256 | struct nf_conn *master = expect->master; | |
257 | struct nf_conn_help *master_help = nfct_help(master); | |
258 | int ret; | |
259 | ||
260 | NF_CT_ASSERT(master_help); | |
261 | ||
262 | write_lock_bh(&nf_conntrack_lock); | |
263 | list_for_each_entry(i, &nf_conntrack_expect_list, list) { | |
264 | if (expect_matches(i, expect)) { | |
265 | /* Refresh timer: if it's dying, ignore.. */ | |
266 | if (refresh_timer(i)) { | |
267 | ret = 0; | |
268 | goto out; | |
269 | } | |
270 | } else if (expect_clash(i, expect)) { | |
271 | ret = -EBUSY; | |
272 | goto out; | |
273 | } | |
274 | } | |
275 | /* Will be over limit? */ | |
276 | if (master_help->helper->max_expected && | |
277 | master_help->expecting >= master_help->helper->max_expected) | |
278 | evict_oldest_expect(master); | |
279 | ||
280 | nf_conntrack_expect_insert(expect); | |
281 | nf_conntrack_expect_event(IPEXP_NEW, expect); | |
282 | ret = 0; | |
283 | out: | |
284 | write_unlock_bh(&nf_conntrack_lock); | |
285 | return ret; | |
286 | } | |
287 | ||
288 | #ifdef CONFIG_PROC_FS | |
289 | static void *exp_seq_start(struct seq_file *s, loff_t *pos) | |
290 | { | |
291 | struct list_head *e = &nf_conntrack_expect_list; | |
292 | loff_t i; | |
293 | ||
294 | /* strange seq_file api calls stop even if we fail, | |
295 | * thus we need to grab lock since stop unlocks */ | |
296 | read_lock_bh(&nf_conntrack_lock); | |
297 | ||
298 | if (list_empty(e)) | |
299 | return NULL; | |
300 | ||
301 | for (i = 0; i <= *pos; i++) { | |
302 | e = e->next; | |
303 | if (e == &nf_conntrack_expect_list) | |
304 | return NULL; | |
305 | } | |
306 | return e; | |
307 | } | |
308 | ||
309 | static void *exp_seq_next(struct seq_file *s, void *v, loff_t *pos) | |
310 | { | |
311 | struct list_head *e = v; | |
312 | ||
313 | ++*pos; | |
314 | e = e->next; | |
315 | ||
316 | if (e == &nf_conntrack_expect_list) | |
317 | return NULL; | |
318 | ||
319 | return e; | |
320 | } | |
321 | ||
322 | static void exp_seq_stop(struct seq_file *s, void *v) | |
323 | { | |
324 | read_unlock_bh(&nf_conntrack_lock); | |
325 | } | |
326 | ||
327 | static int exp_seq_show(struct seq_file *s, void *v) | |
328 | { | |
329 | struct nf_conntrack_expect *expect = v; | |
330 | ||
331 | if (expect->timeout.function) | |
332 | seq_printf(s, "%ld ", timer_pending(&expect->timeout) | |
333 | ? (long)(expect->timeout.expires - jiffies)/HZ : 0); | |
334 | else | |
335 | seq_printf(s, "- "); | |
336 | seq_printf(s, "l3proto = %u proto=%u ", | |
337 | expect->tuple.src.l3num, | |
338 | expect->tuple.dst.protonum); | |
339 | print_tuple(s, &expect->tuple, | |
340 | __nf_ct_l3proto_find(expect->tuple.src.l3num), | |
341 | __nf_ct_proto_find(expect->tuple.src.l3num, | |
342 | expect->tuple.dst.protonum)); | |
343 | return seq_putc(s, '\n'); | |
344 | } | |
345 | ||
346 | static struct seq_operations exp_seq_ops = { | |
347 | .start = exp_seq_start, | |
348 | .next = exp_seq_next, | |
349 | .stop = exp_seq_stop, | |
350 | .show = exp_seq_show | |
351 | }; | |
352 | ||
353 | static int exp_open(struct inode *inode, struct file *file) | |
354 | { | |
355 | return seq_open(file, &exp_seq_ops); | |
356 | } | |
357 | ||
358 | struct file_operations exp_file_ops = { | |
359 | .owner = THIS_MODULE, | |
360 | .open = exp_open, | |
361 | .read = seq_read, | |
362 | .llseek = seq_lseek, | |
363 | .release = seq_release | |
364 | }; | |
365 | #endif /* CONFIG_PROC_FS */ |